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.
28 #include <sys/socket.h>
29 #include <sys/types.h>
34 #include <bin/lttng-consumerd/health-consumerd.h>
35 #include <common/common.h>
36 #include <common/utils.h>
37 #include <common/compat/poll.h>
38 #include <common/compat/endian.h>
39 #include <common/index/index.h>
40 #include <common/kernel-ctl/kernel-ctl.h>
41 #include <common/sessiond-comm/relayd.h>
42 #include <common/sessiond-comm/sessiond-comm.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/relayd/relayd.h>
45 #include <common/ust-consumer/ust-consumer.h>
46 #include <common/consumer-timer.h>
49 #include "consumer-stream.h"
50 #include "consumer-testpoint.h"
53 struct lttng_consumer_global_data consumer_data
= {
56 .type
= LTTNG_CONSUMER_UNKNOWN
,
59 enum consumer_channel_action
{
62 CONSUMER_CHANNEL_QUIT
,
65 struct consumer_channel_msg
{
66 enum consumer_channel_action action
;
67 struct lttng_consumer_channel
*chan
; /* add */
68 uint64_t key
; /* del */
72 * Flag to inform the polling thread to quit when all fd hung up. Updated by
73 * the consumer_thread_receive_fds when it notices that all fds has hung up.
74 * Also updated by the signal handler (consumer_should_exit()). Read by the
77 volatile int consumer_quit
;
80 * Global hash table containing respectively metadata and data streams. The
81 * stream element in this ht should only be updated by the metadata poll thread
82 * for the metadata and the data poll thread for the data.
84 static struct lttng_ht
*metadata_ht
;
85 static struct lttng_ht
*data_ht
;
88 * Notify a thread lttng pipe to poll back again. This usually means that some
89 * global state has changed so we just send back the thread in a poll wait
92 static void notify_thread_lttng_pipe(struct lttng_pipe
*pipe
)
94 struct lttng_consumer_stream
*null_stream
= NULL
;
98 (void) lttng_pipe_write(pipe
, &null_stream
, sizeof(null_stream
));
101 static void notify_health_quit_pipe(int *pipe
)
105 ret
= lttng_write(pipe
[1], "4", 1);
107 PERROR("write consumer health quit");
111 static void notify_channel_pipe(struct lttng_consumer_local_data
*ctx
,
112 struct lttng_consumer_channel
*chan
,
114 enum consumer_channel_action action
)
116 struct consumer_channel_msg msg
;
119 memset(&msg
, 0, sizeof(msg
));
124 ret
= lttng_write(ctx
->consumer_channel_pipe
[1], &msg
, sizeof(msg
));
125 if (ret
< sizeof(msg
)) {
126 PERROR("notify_channel_pipe write error");
130 void notify_thread_del_channel(struct lttng_consumer_local_data
*ctx
,
133 notify_channel_pipe(ctx
, NULL
, key
, CONSUMER_CHANNEL_DEL
);
136 static int read_channel_pipe(struct lttng_consumer_local_data
*ctx
,
137 struct lttng_consumer_channel
**chan
,
139 enum consumer_channel_action
*action
)
141 struct consumer_channel_msg msg
;
144 ret
= lttng_read(ctx
->consumer_channel_pipe
[0], &msg
, sizeof(msg
));
145 if (ret
< sizeof(msg
)) {
149 *action
= msg
.action
;
157 * Cleanup the stream list of a channel. Those streams are not yet globally
160 static void clean_channel_stream_list(struct lttng_consumer_channel
*channel
)
162 struct lttng_consumer_stream
*stream
, *stmp
;
166 /* Delete streams that might have been left in the stream list. */
167 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
169 cds_list_del(&stream
->send_node
);
171 * Once a stream is added to this list, the buffers were created so we
172 * have a guarantee that this call will succeed. Setting the monitor
173 * mode to 0 so we don't lock nor try to delete the stream from the
177 consumer_stream_destroy(stream
, NULL
);
182 * Find a stream. The consumer_data.lock must be locked during this
185 static struct lttng_consumer_stream
*find_stream(uint64_t key
,
188 struct lttng_ht_iter iter
;
189 struct lttng_ht_node_u64
*node
;
190 struct lttng_consumer_stream
*stream
= NULL
;
194 /* -1ULL keys are lookup failures */
195 if (key
== (uint64_t) -1ULL) {
201 lttng_ht_lookup(ht
, &key
, &iter
);
202 node
= lttng_ht_iter_get_node_u64(&iter
);
204 stream
= caa_container_of(node
, struct lttng_consumer_stream
, node
);
212 static void steal_stream_key(uint64_t key
, struct lttng_ht
*ht
)
214 struct lttng_consumer_stream
*stream
;
217 stream
= find_stream(key
, ht
);
219 stream
->key
= (uint64_t) -1ULL;
221 * We don't want the lookup to match, but we still need
222 * to iterate on this stream when iterating over the hash table. Just
223 * change the node key.
225 stream
->node
.key
= (uint64_t) -1ULL;
231 * Return a channel object for the given key.
233 * RCU read side lock MUST be acquired before calling this function and
234 * protects the channel ptr.
236 struct lttng_consumer_channel
*consumer_find_channel(uint64_t key
)
238 struct lttng_ht_iter iter
;
239 struct lttng_ht_node_u64
*node
;
240 struct lttng_consumer_channel
*channel
= NULL
;
242 /* -1ULL keys are lookup failures */
243 if (key
== (uint64_t) -1ULL) {
247 lttng_ht_lookup(consumer_data
.channel_ht
, &key
, &iter
);
248 node
= lttng_ht_iter_get_node_u64(&iter
);
250 channel
= caa_container_of(node
, struct lttng_consumer_channel
, node
);
257 * There is a possibility that the consumer does not have enough time between
258 * the close of the channel on the session daemon and the cleanup in here thus
259 * once we have a channel add with an existing key, we know for sure that this
260 * channel will eventually get cleaned up by all streams being closed.
262 * This function just nullifies the already existing channel key.
264 static void steal_channel_key(uint64_t key
)
266 struct lttng_consumer_channel
*channel
;
269 channel
= consumer_find_channel(key
);
271 channel
->key
= (uint64_t) -1ULL;
273 * We don't want the lookup to match, but we still need to iterate on
274 * this channel when iterating over the hash table. Just change the
277 channel
->node
.key
= (uint64_t) -1ULL;
282 static void free_channel_rcu(struct rcu_head
*head
)
284 struct lttng_ht_node_u64
*node
=
285 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
286 struct lttng_consumer_channel
*channel
=
287 caa_container_of(node
, struct lttng_consumer_channel
, node
);
293 * RCU protected relayd socket pair free.
295 static void free_relayd_rcu(struct rcu_head
*head
)
297 struct lttng_ht_node_u64
*node
=
298 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
299 struct consumer_relayd_sock_pair
*relayd
=
300 caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
303 * Close all sockets. This is done in the call RCU since we don't want the
304 * socket fds to be reassigned thus potentially creating bad state of the
307 * We do not have to lock the control socket mutex here since at this stage
308 * there is no one referencing to this relayd object.
310 (void) relayd_close(&relayd
->control_sock
);
311 (void) relayd_close(&relayd
->data_sock
);
317 * Destroy and free relayd socket pair object.
319 void consumer_destroy_relayd(struct consumer_relayd_sock_pair
*relayd
)
322 struct lttng_ht_iter iter
;
324 if (relayd
== NULL
) {
328 DBG("Consumer destroy and close relayd socket pair");
330 iter
.iter
.node
= &relayd
->node
.node
;
331 ret
= lttng_ht_del(consumer_data
.relayd_ht
, &iter
);
333 /* We assume the relayd is being or is destroyed */
337 /* RCU free() call */
338 call_rcu(&relayd
->node
.head
, free_relayd_rcu
);
342 * Remove a channel from the global list protected by a mutex. This function is
343 * also responsible for freeing its data structures.
345 void consumer_del_channel(struct lttng_consumer_channel
*channel
)
348 struct lttng_ht_iter iter
;
350 DBG("Consumer delete channel key %" PRIu64
, channel
->key
);
352 pthread_mutex_lock(&consumer_data
.lock
);
353 pthread_mutex_lock(&channel
->lock
);
355 /* Destroy streams that might have been left in the stream list. */
356 clean_channel_stream_list(channel
);
358 if (channel
->live_timer_enabled
== 1) {
359 consumer_timer_live_stop(channel
);
362 switch (consumer_data
.type
) {
363 case LTTNG_CONSUMER_KERNEL
:
365 case LTTNG_CONSUMER32_UST
:
366 case LTTNG_CONSUMER64_UST
:
367 lttng_ustconsumer_del_channel(channel
);
370 ERR("Unknown consumer_data type");
376 iter
.iter
.node
= &channel
->node
.node
;
377 ret
= lttng_ht_del(consumer_data
.channel_ht
, &iter
);
381 call_rcu(&channel
->node
.head
, free_channel_rcu
);
383 pthread_mutex_unlock(&channel
->lock
);
384 pthread_mutex_unlock(&consumer_data
.lock
);
388 * Iterate over the relayd hash table and destroy each element. Finally,
389 * destroy the whole hash table.
391 static void cleanup_relayd_ht(void)
393 struct lttng_ht_iter iter
;
394 struct consumer_relayd_sock_pair
*relayd
;
398 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
400 consumer_destroy_relayd(relayd
);
405 lttng_ht_destroy(consumer_data
.relayd_ht
);
409 * Update the end point status of all streams having the given network sequence
410 * index (relayd index).
412 * It's atomically set without having the stream mutex locked which is fine
413 * because we handle the write/read race with a pipe wakeup for each thread.
415 static void update_endpoint_status_by_netidx(uint64_t net_seq_idx
,
416 enum consumer_endpoint_status status
)
418 struct lttng_ht_iter iter
;
419 struct lttng_consumer_stream
*stream
;
421 DBG("Consumer set delete flag on stream by idx %" PRIu64
, net_seq_idx
);
425 /* Let's begin with metadata */
426 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
427 if (stream
->net_seq_idx
== net_seq_idx
) {
428 uatomic_set(&stream
->endpoint_status
, status
);
429 DBG("Delete flag set to metadata stream %d", stream
->wait_fd
);
433 /* Follow up by the data streams */
434 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
435 if (stream
->net_seq_idx
== net_seq_idx
) {
436 uatomic_set(&stream
->endpoint_status
, status
);
437 DBG("Delete flag set to data stream %d", stream
->wait_fd
);
444 * Cleanup a relayd object by flagging every associated streams for deletion,
445 * destroying the object meaning removing it from the relayd hash table,
446 * closing the sockets and freeing the memory in a RCU call.
448 * If a local data context is available, notify the threads that the streams'
449 * state have changed.
451 static void cleanup_relayd(struct consumer_relayd_sock_pair
*relayd
,
452 struct lttng_consumer_local_data
*ctx
)
458 DBG("Cleaning up relayd sockets");
460 /* Save the net sequence index before destroying the object */
461 netidx
= relayd
->net_seq_idx
;
464 * Delete the relayd from the relayd hash table, close the sockets and free
465 * the object in a RCU call.
467 consumer_destroy_relayd(relayd
);
469 /* Set inactive endpoint to all streams */
470 update_endpoint_status_by_netidx(netidx
, CONSUMER_ENDPOINT_INACTIVE
);
473 * With a local data context, notify the threads that the streams' state
474 * have changed. The write() action on the pipe acts as an "implicit"
475 * memory barrier ordering the updates of the end point status from the
476 * read of this status which happens AFTER receiving this notify.
479 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
480 notify_thread_lttng_pipe(ctx
->consumer_metadata_pipe
);
485 * Flag a relayd socket pair for destruction. Destroy it if the refcount
488 * RCU read side lock MUST be aquired before calling this function.
490 void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair
*relayd
)
494 /* Set destroy flag for this object */
495 uatomic_set(&relayd
->destroy_flag
, 1);
497 /* Destroy the relayd if refcount is 0 */
498 if (uatomic_read(&relayd
->refcount
) == 0) {
499 consumer_destroy_relayd(relayd
);
504 * Completly destroy stream from every visiable data structure and the given
507 * One this call returns, the stream object is not longer usable nor visible.
509 void consumer_del_stream(struct lttng_consumer_stream
*stream
,
512 consumer_stream_destroy(stream
, ht
);
516 * XXX naming of del vs destroy is all mixed up.
518 void consumer_del_stream_for_data(struct lttng_consumer_stream
*stream
)
520 consumer_stream_destroy(stream
, data_ht
);
523 void consumer_del_stream_for_metadata(struct lttng_consumer_stream
*stream
)
525 consumer_stream_destroy(stream
, metadata_ht
);
528 struct lttng_consumer_stream
*consumer_allocate_stream(uint64_t channel_key
,
530 enum lttng_consumer_stream_state state
,
531 const char *channel_name
,
538 enum consumer_channel_type type
,
539 unsigned int monitor
)
542 struct lttng_consumer_stream
*stream
;
544 stream
= zmalloc(sizeof(*stream
));
545 if (stream
== NULL
) {
546 PERROR("malloc struct lttng_consumer_stream");
553 stream
->key
= stream_key
;
555 stream
->out_fd_offset
= 0;
556 stream
->output_written
= 0;
557 stream
->state
= state
;
560 stream
->net_seq_idx
= relayd_id
;
561 stream
->session_id
= session_id
;
562 stream
->monitor
= monitor
;
563 stream
->endpoint_status
= CONSUMER_ENDPOINT_ACTIVE
;
564 stream
->index_fd
= -1;
565 pthread_mutex_init(&stream
->lock
, NULL
);
567 /* If channel is the metadata, flag this stream as metadata. */
568 if (type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
569 stream
->metadata_flag
= 1;
570 /* Metadata is flat out. */
571 strncpy(stream
->name
, DEFAULT_METADATA_NAME
, sizeof(stream
->name
));
572 /* Live rendez-vous point. */
573 pthread_cond_init(&stream
->metadata_rdv
, NULL
);
574 pthread_mutex_init(&stream
->metadata_rdv_lock
, NULL
);
576 /* Format stream name to <channel_name>_<cpu_number> */
577 ret
= snprintf(stream
->name
, sizeof(stream
->name
), "%s_%d",
580 PERROR("snprintf stream name");
585 /* Key is always the wait_fd for streams. */
586 lttng_ht_node_init_u64(&stream
->node
, stream
->key
);
588 /* Init node per channel id key */
589 lttng_ht_node_init_u64(&stream
->node_channel_id
, channel_key
);
591 /* Init session id node with the stream session id */
592 lttng_ht_node_init_u64(&stream
->node_session_id
, stream
->session_id
);
594 DBG3("Allocated stream %s (key %" PRIu64
", chan_key %" PRIu64
595 " relayd_id %" PRIu64
", session_id %" PRIu64
,
596 stream
->name
, stream
->key
, channel_key
,
597 stream
->net_seq_idx
, stream
->session_id
);
613 * Add a stream to the global list protected by a mutex.
615 int consumer_add_data_stream(struct lttng_consumer_stream
*stream
)
617 struct lttng_ht
*ht
= data_ht
;
623 DBG3("Adding consumer stream %" PRIu64
, stream
->key
);
625 pthread_mutex_lock(&consumer_data
.lock
);
626 pthread_mutex_lock(&stream
->chan
->lock
);
627 pthread_mutex_lock(&stream
->chan
->timer_lock
);
628 pthread_mutex_lock(&stream
->lock
);
631 /* Steal stream identifier to avoid having streams with the same key */
632 steal_stream_key(stream
->key
, ht
);
634 lttng_ht_add_unique_u64(ht
, &stream
->node
);
636 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
637 &stream
->node_channel_id
);
640 * Add stream to the stream_list_ht of the consumer data. No need to steal
641 * the key since the HT does not use it and we allow to add redundant keys
644 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
647 * When nb_init_stream_left reaches 0, we don't need to trigger any action
648 * in terms of destroying the associated channel, because the action that
649 * causes the count to become 0 also causes a stream to be added. The
650 * channel deletion will thus be triggered by the following removal of this
653 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
654 /* Increment refcount before decrementing nb_init_stream_left */
656 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
659 /* Update consumer data once the node is inserted. */
660 consumer_data
.stream_count
++;
661 consumer_data
.need_update
= 1;
664 pthread_mutex_unlock(&stream
->lock
);
665 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
666 pthread_mutex_unlock(&stream
->chan
->lock
);
667 pthread_mutex_unlock(&consumer_data
.lock
);
672 void consumer_del_data_stream(struct lttng_consumer_stream
*stream
)
674 consumer_del_stream(stream
, data_ht
);
678 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
679 * be acquired before calling this.
681 static int add_relayd(struct consumer_relayd_sock_pair
*relayd
)
684 struct lttng_ht_node_u64
*node
;
685 struct lttng_ht_iter iter
;
689 lttng_ht_lookup(consumer_data
.relayd_ht
,
690 &relayd
->net_seq_idx
, &iter
);
691 node
= lttng_ht_iter_get_node_u64(&iter
);
695 lttng_ht_add_unique_u64(consumer_data
.relayd_ht
, &relayd
->node
);
702 * Allocate and return a consumer relayd socket.
704 struct consumer_relayd_sock_pair
*consumer_allocate_relayd_sock_pair(
705 uint64_t net_seq_idx
)
707 struct consumer_relayd_sock_pair
*obj
= NULL
;
709 /* net sequence index of -1 is a failure */
710 if (net_seq_idx
== (uint64_t) -1ULL) {
714 obj
= zmalloc(sizeof(struct consumer_relayd_sock_pair
));
716 PERROR("zmalloc relayd sock");
720 obj
->net_seq_idx
= net_seq_idx
;
722 obj
->destroy_flag
= 0;
723 obj
->control_sock
.sock
.fd
= -1;
724 obj
->data_sock
.sock
.fd
= -1;
725 lttng_ht_node_init_u64(&obj
->node
, obj
->net_seq_idx
);
726 pthread_mutex_init(&obj
->ctrl_sock_mutex
, NULL
);
733 * Find a relayd socket pair in the global consumer data.
735 * Return the object if found else NULL.
736 * RCU read-side lock must be held across this call and while using the
739 struct consumer_relayd_sock_pair
*consumer_find_relayd(uint64_t key
)
741 struct lttng_ht_iter iter
;
742 struct lttng_ht_node_u64
*node
;
743 struct consumer_relayd_sock_pair
*relayd
= NULL
;
745 /* Negative keys are lookup failures */
746 if (key
== (uint64_t) -1ULL) {
750 lttng_ht_lookup(consumer_data
.relayd_ht
, &key
,
752 node
= lttng_ht_iter_get_node_u64(&iter
);
754 relayd
= caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
762 * Find a relayd and send the stream
764 * Returns 0 on success, < 0 on error
766 int consumer_send_relayd_stream(struct lttng_consumer_stream
*stream
,
770 struct consumer_relayd_sock_pair
*relayd
;
773 assert(stream
->net_seq_idx
!= -1ULL);
776 /* The stream is not metadata. Get relayd reference if exists. */
778 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
779 if (relayd
!= NULL
) {
780 /* Add stream on the relayd */
781 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
782 ret
= relayd_add_stream(&relayd
->control_sock
, stream
->name
,
783 path
, &stream
->relayd_stream_id
,
784 stream
->chan
->tracefile_size
, stream
->chan
->tracefile_count
);
785 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
790 uatomic_inc(&relayd
->refcount
);
791 stream
->sent_to_relayd
= 1;
793 ERR("Stream %" PRIu64
" relayd ID %" PRIu64
" unknown. Can't send it.",
794 stream
->key
, stream
->net_seq_idx
);
799 DBG("Stream %s with key %" PRIu64
" sent to relayd id %" PRIu64
,
800 stream
->name
, stream
->key
, stream
->net_seq_idx
);
808 * Find a relayd and send the streams sent message
810 * Returns 0 on success, < 0 on error
812 int consumer_send_relayd_streams_sent(uint64_t net_seq_idx
)
815 struct consumer_relayd_sock_pair
*relayd
;
817 assert(net_seq_idx
!= -1ULL);
819 /* The stream is not metadata. Get relayd reference if exists. */
821 relayd
= consumer_find_relayd(net_seq_idx
);
822 if (relayd
!= NULL
) {
823 /* Add stream on the relayd */
824 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
825 ret
= relayd_streams_sent(&relayd
->control_sock
);
826 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
831 ERR("Relayd ID %" PRIu64
" unknown. Can't send streams_sent.",
838 DBG("All streams sent relayd id %" PRIu64
, net_seq_idx
);
846 * Find a relayd and close the stream
848 void close_relayd_stream(struct lttng_consumer_stream
*stream
)
850 struct consumer_relayd_sock_pair
*relayd
;
852 /* The stream is not metadata. Get relayd reference if exists. */
854 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
856 consumer_stream_relayd_close(stream
, relayd
);
862 * Handle stream for relayd transmission if the stream applies for network
863 * streaming where the net sequence index is set.
865 * Return destination file descriptor or negative value on error.
867 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
868 size_t data_size
, unsigned long padding
,
869 struct consumer_relayd_sock_pair
*relayd
)
872 struct lttcomm_relayd_data_hdr data_hdr
;
878 /* Reset data header */
879 memset(&data_hdr
, 0, sizeof(data_hdr
));
881 if (stream
->metadata_flag
) {
882 /* Caller MUST acquire the relayd control socket lock */
883 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
888 /* Metadata are always sent on the control socket. */
889 outfd
= relayd
->control_sock
.sock
.fd
;
891 /* Set header with stream information */
892 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
893 data_hdr
.data_size
= htobe32(data_size
);
894 data_hdr
.padding_size
= htobe32(padding
);
896 * Note that net_seq_num below is assigned with the *current* value of
897 * next_net_seq_num and only after that the next_net_seq_num will be
898 * increment. This is why when issuing a command on the relayd using
899 * this next value, 1 should always be substracted in order to compare
900 * the last seen sequence number on the relayd side to the last sent.
902 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
903 /* Other fields are zeroed previously */
905 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
911 ++stream
->next_net_seq_num
;
913 /* Set to go on data socket */
914 outfd
= relayd
->data_sock
.sock
.fd
;
922 * Allocate and return a new lttng_consumer_channel object using the given key
923 * to initialize the hash table node.
925 * On error, return NULL.
927 struct lttng_consumer_channel
*consumer_allocate_channel(uint64_t key
,
929 const char *pathname
,
934 enum lttng_event_output output
,
935 uint64_t tracefile_size
,
936 uint64_t tracefile_count
,
937 uint64_t session_id_per_pid
,
938 unsigned int monitor
,
939 unsigned int live_timer_interval
,
940 const char *shm_path
)
942 struct lttng_consumer_channel
*channel
;
944 channel
= zmalloc(sizeof(*channel
));
945 if (channel
== NULL
) {
946 PERROR("malloc struct lttng_consumer_channel");
951 channel
->refcount
= 0;
952 channel
->session_id
= session_id
;
953 channel
->session_id_per_pid
= session_id_per_pid
;
956 channel
->relayd_id
= relayd_id
;
957 channel
->tracefile_size
= tracefile_size
;
958 channel
->tracefile_count
= tracefile_count
;
959 channel
->monitor
= monitor
;
960 channel
->live_timer_interval
= live_timer_interval
;
961 pthread_mutex_init(&channel
->lock
, NULL
);
962 pthread_mutex_init(&channel
->timer_lock
, NULL
);
965 case LTTNG_EVENT_SPLICE
:
966 channel
->output
= CONSUMER_CHANNEL_SPLICE
;
968 case LTTNG_EVENT_MMAP
:
969 channel
->output
= CONSUMER_CHANNEL_MMAP
;
979 * In monitor mode, the streams associated with the channel will be put in
980 * a special list ONLY owned by this channel. So, the refcount is set to 1
981 * here meaning that the channel itself has streams that are referenced.
983 * On a channel deletion, once the channel is no longer visible, the
984 * refcount is decremented and checked for a zero value to delete it. With
985 * streams in no monitor mode, it will now be safe to destroy the channel.
987 if (!channel
->monitor
) {
988 channel
->refcount
= 1;
991 strncpy(channel
->pathname
, pathname
, sizeof(channel
->pathname
));
992 channel
->pathname
[sizeof(channel
->pathname
) - 1] = '\0';
994 strncpy(channel
->name
, name
, sizeof(channel
->name
));
995 channel
->name
[sizeof(channel
->name
) - 1] = '\0';
998 strncpy(channel
->shm_path
, shm_path
, sizeof(channel
->shm_path
));
999 channel
->shm_path
[sizeof(channel
->shm_path
) - 1] = '\0';
1002 lttng_ht_node_init_u64(&channel
->node
, channel
->key
);
1004 channel
->wait_fd
= -1;
1006 CDS_INIT_LIST_HEAD(&channel
->streams
.head
);
1008 DBG("Allocated channel (key %" PRIu64
")", channel
->key
)
1015 * Add a channel to the global list protected by a mutex.
1017 * Always return 0 indicating success.
1019 int consumer_add_channel(struct lttng_consumer_channel
*channel
,
1020 struct lttng_consumer_local_data
*ctx
)
1022 pthread_mutex_lock(&consumer_data
.lock
);
1023 pthread_mutex_lock(&channel
->lock
);
1024 pthread_mutex_lock(&channel
->timer_lock
);
1027 * This gives us a guarantee that the channel we are about to add to the
1028 * channel hash table will be unique. See this function comment on the why
1029 * we need to steel the channel key at this stage.
1031 steal_channel_key(channel
->key
);
1034 lttng_ht_add_unique_u64(consumer_data
.channel_ht
, &channel
->node
);
1037 pthread_mutex_unlock(&channel
->timer_lock
);
1038 pthread_mutex_unlock(&channel
->lock
);
1039 pthread_mutex_unlock(&consumer_data
.lock
);
1041 if (channel
->wait_fd
!= -1 && channel
->type
== CONSUMER_CHANNEL_TYPE_DATA
) {
1042 notify_channel_pipe(ctx
, channel
, -1, CONSUMER_CHANNEL_ADD
);
1049 * Allocate the pollfd structure and the local view of the out fds to avoid
1050 * doing a lookup in the linked list and concurrency issues when writing is
1051 * needed. Called with consumer_data.lock held.
1053 * Returns the number of fds in the structures.
1055 static int update_poll_array(struct lttng_consumer_local_data
*ctx
,
1056 struct pollfd
**pollfd
, struct lttng_consumer_stream
**local_stream
,
1057 struct lttng_ht
*ht
)
1060 struct lttng_ht_iter iter
;
1061 struct lttng_consumer_stream
*stream
;
1066 assert(local_stream
);
1068 DBG("Updating poll fd array");
1070 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1072 * Only active streams with an active end point can be added to the
1073 * poll set and local stream storage of the thread.
1075 * There is a potential race here for endpoint_status to be updated
1076 * just after the check. However, this is OK since the stream(s) will
1077 * be deleted once the thread is notified that the end point state has
1078 * changed where this function will be called back again.
1080 if (stream
->state
!= LTTNG_CONSUMER_ACTIVE_STREAM
||
1081 stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
1085 * This clobbers way too much the debug output. Uncomment that if you
1086 * need it for debugging purposes.
1088 * DBG("Active FD %d", stream->wait_fd);
1090 (*pollfd
)[i
].fd
= stream
->wait_fd
;
1091 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1092 local_stream
[i
] = stream
;
1098 * Insert the consumer_data_pipe at the end of the array and don't
1099 * increment i so nb_fd is the number of real FD.
1101 (*pollfd
)[i
].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_pipe
);
1102 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1104 (*pollfd
)[i
+ 1].fd
= lttng_pipe_get_readfd(ctx
->consumer_wakeup_pipe
);
1105 (*pollfd
)[i
+ 1].events
= POLLIN
| POLLPRI
;
1110 * Poll on the should_quit pipe and the command socket return -1 on
1111 * error, 1 if should exit, 0 if data is available on the command socket
1113 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
1118 num_rdy
= poll(consumer_sockpoll
, 2, -1);
1119 if (num_rdy
== -1) {
1121 * Restart interrupted system call.
1123 if (errno
== EINTR
) {
1126 PERROR("Poll error");
1129 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
1130 DBG("consumer_should_quit wake up");
1137 * Set the error socket.
1139 void lttng_consumer_set_error_sock(struct lttng_consumer_local_data
*ctx
,
1142 ctx
->consumer_error_socket
= sock
;
1146 * Set the command socket path.
1148 void lttng_consumer_set_command_sock_path(
1149 struct lttng_consumer_local_data
*ctx
, char *sock
)
1151 ctx
->consumer_command_sock_path
= sock
;
1155 * Send return code to the session daemon.
1156 * If the socket is not defined, we return 0, it is not a fatal error
1158 int lttng_consumer_send_error(struct lttng_consumer_local_data
*ctx
, int cmd
)
1160 if (ctx
->consumer_error_socket
> 0) {
1161 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1162 sizeof(enum lttcomm_sessiond_command
));
1169 * Close all the tracefiles and stream fds and MUST be called when all
1170 * instances are destroyed i.e. when all threads were joined and are ended.
1172 void lttng_consumer_cleanup(void)
1174 struct lttng_ht_iter iter
;
1175 struct lttng_consumer_channel
*channel
;
1179 cds_lfht_for_each_entry(consumer_data
.channel_ht
->ht
, &iter
.iter
, channel
,
1181 consumer_del_channel(channel
);
1186 lttng_ht_destroy(consumer_data
.channel_ht
);
1188 cleanup_relayd_ht();
1190 lttng_ht_destroy(consumer_data
.stream_per_chan_id_ht
);
1193 * This HT contains streams that are freed by either the metadata thread or
1194 * the data thread so we do *nothing* on the hash table and simply destroy
1197 lttng_ht_destroy(consumer_data
.stream_list_ht
);
1201 * Called from signal handler.
1203 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1208 ret
= lttng_write(ctx
->consumer_should_quit
[1], "4", 1);
1210 PERROR("write consumer quit");
1213 DBG("Consumer flag that it should quit");
1216 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1219 int outfd
= stream
->out_fd
;
1222 * This does a blocking write-and-wait on any page that belongs to the
1223 * subbuffer prior to the one we just wrote.
1224 * Don't care about error values, as these are just hints and ways to
1225 * limit the amount of page cache used.
1227 if (orig_offset
< stream
->max_sb_size
) {
1230 lttng_sync_file_range(outfd
, orig_offset
- stream
->max_sb_size
,
1231 stream
->max_sb_size
,
1232 SYNC_FILE_RANGE_WAIT_BEFORE
1233 | SYNC_FILE_RANGE_WRITE
1234 | SYNC_FILE_RANGE_WAIT_AFTER
);
1236 * Give hints to the kernel about how we access the file:
1237 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1240 * We need to call fadvise again after the file grows because the
1241 * kernel does not seem to apply fadvise to non-existing parts of the
1244 * Call fadvise _after_ having waited for the page writeback to
1245 * complete because the dirty page writeback semantic is not well
1246 * defined. So it can be expected to lead to lower throughput in
1249 posix_fadvise(outfd
, orig_offset
- stream
->max_sb_size
,
1250 stream
->max_sb_size
, POSIX_FADV_DONTNEED
);
1254 * Initialise the necessary environnement :
1255 * - create a new context
1256 * - create the poll_pipe
1257 * - create the should_quit pipe (for signal handler)
1258 * - create the thread pipe (for splice)
1260 * Takes a function pointer as argument, this function is called when data is
1261 * available on a buffer. This function is responsible to do the
1262 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1263 * buffer configuration and then kernctl_put_next_subbuf at the end.
1265 * Returns a pointer to the new context or NULL on error.
1267 struct lttng_consumer_local_data
*lttng_consumer_create(
1268 enum lttng_consumer_type type
,
1269 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1270 struct lttng_consumer_local_data
*ctx
),
1271 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1272 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1273 int (*update_stream
)(uint64_t stream_key
, uint32_t state
))
1276 struct lttng_consumer_local_data
*ctx
;
1278 assert(consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1279 consumer_data
.type
== type
);
1280 consumer_data
.type
= type
;
1282 ctx
= zmalloc(sizeof(struct lttng_consumer_local_data
));
1284 PERROR("allocating context");
1288 ctx
->consumer_error_socket
= -1;
1289 ctx
->consumer_metadata_socket
= -1;
1290 pthread_mutex_init(&ctx
->metadata_socket_lock
, NULL
);
1291 /* assign the callbacks */
1292 ctx
->on_buffer_ready
= buffer_ready
;
1293 ctx
->on_recv_channel
= recv_channel
;
1294 ctx
->on_recv_stream
= recv_stream
;
1295 ctx
->on_update_stream
= update_stream
;
1297 ctx
->consumer_data_pipe
= lttng_pipe_open(0);
1298 if (!ctx
->consumer_data_pipe
) {
1299 goto error_poll_pipe
;
1302 ctx
->consumer_wakeup_pipe
= lttng_pipe_open(0);
1303 if (!ctx
->consumer_wakeup_pipe
) {
1304 goto error_wakeup_pipe
;
1307 ret
= pipe(ctx
->consumer_should_quit
);
1309 PERROR("Error creating recv pipe");
1310 goto error_quit_pipe
;
1313 ret
= pipe(ctx
->consumer_channel_pipe
);
1315 PERROR("Error creating channel pipe");
1316 goto error_channel_pipe
;
1319 ctx
->consumer_metadata_pipe
= lttng_pipe_open(0);
1320 if (!ctx
->consumer_metadata_pipe
) {
1321 goto error_metadata_pipe
;
1326 error_metadata_pipe
:
1327 utils_close_pipe(ctx
->consumer_channel_pipe
);
1329 utils_close_pipe(ctx
->consumer_should_quit
);
1331 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1333 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1341 * Iterate over all streams of the hashtable and free them properly.
1343 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1345 struct lttng_ht_iter iter
;
1346 struct lttng_consumer_stream
*stream
;
1353 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1355 * Ignore return value since we are currently cleaning up so any error
1358 (void) consumer_del_stream(stream
, ht
);
1362 lttng_ht_destroy(ht
);
1366 * Iterate over all streams of the metadata hashtable and free them
1369 static void destroy_metadata_stream_ht(struct lttng_ht
*ht
)
1371 struct lttng_ht_iter iter
;
1372 struct lttng_consumer_stream
*stream
;
1379 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1381 * Ignore return value since we are currently cleaning up so any error
1384 (void) consumer_del_metadata_stream(stream
, ht
);
1388 lttng_ht_destroy(ht
);
1392 * Close all fds associated with the instance and free the context.
1394 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1398 DBG("Consumer destroying it. Closing everything.");
1404 destroy_data_stream_ht(data_ht
);
1405 destroy_metadata_stream_ht(metadata_ht
);
1407 ret
= close(ctx
->consumer_error_socket
);
1411 ret
= close(ctx
->consumer_metadata_socket
);
1415 utils_close_pipe(ctx
->consumer_channel_pipe
);
1416 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1417 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1418 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1419 utils_close_pipe(ctx
->consumer_should_quit
);
1421 unlink(ctx
->consumer_command_sock_path
);
1426 * Write the metadata stream id on the specified file descriptor.
1428 static int write_relayd_metadata_id(int fd
,
1429 struct lttng_consumer_stream
*stream
,
1430 struct consumer_relayd_sock_pair
*relayd
, unsigned long padding
)
1433 struct lttcomm_relayd_metadata_payload hdr
;
1435 hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
1436 hdr
.padding_size
= htobe32(padding
);
1437 ret
= lttng_write(fd
, (void *) &hdr
, sizeof(hdr
));
1438 if (ret
< sizeof(hdr
)) {
1440 * This error means that the fd's end is closed so ignore the PERROR
1441 * not to clubber the error output since this can happen in a normal
1444 if (errno
!= EPIPE
) {
1445 PERROR("write metadata stream id");
1447 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno
);
1449 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1450 * handle writting the missing part so report that as an error and
1451 * don't lie to the caller.
1456 DBG("Metadata stream id %" PRIu64
" with padding %lu written before data",
1457 stream
->relayd_stream_id
, padding
);
1464 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1465 * core function for writing trace buffers to either the local filesystem or
1468 * It must be called with the stream lock held.
1470 * Careful review MUST be put if any changes occur!
1472 * Returns the number of bytes written
1474 ssize_t
lttng_consumer_on_read_subbuffer_mmap(
1475 struct lttng_consumer_local_data
*ctx
,
1476 struct lttng_consumer_stream
*stream
, unsigned long len
,
1477 unsigned long padding
,
1478 struct ctf_packet_index
*index
)
1480 unsigned long mmap_offset
;
1483 off_t orig_offset
= stream
->out_fd_offset
;
1484 /* Default is on the disk */
1485 int outfd
= stream
->out_fd
;
1486 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1487 unsigned int relayd_hang_up
= 0;
1489 /* RCU lock for the relayd pointer */
1492 /* Flag that the current stream if set for network streaming. */
1493 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1494 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1495 if (relayd
== NULL
) {
1501 /* get the offset inside the fd to mmap */
1502 switch (consumer_data
.type
) {
1503 case LTTNG_CONSUMER_KERNEL
:
1504 mmap_base
= stream
->mmap_base
;
1505 ret
= kernctl_get_mmap_read_offset(stream
->wait_fd
, &mmap_offset
);
1508 PERROR("tracer ctl get_mmap_read_offset");
1512 case LTTNG_CONSUMER32_UST
:
1513 case LTTNG_CONSUMER64_UST
:
1514 mmap_base
= lttng_ustctl_get_mmap_base(stream
);
1516 ERR("read mmap get mmap base for stream %s", stream
->name
);
1520 ret
= lttng_ustctl_get_mmap_read_offset(stream
, &mmap_offset
);
1522 PERROR("tracer ctl get_mmap_read_offset");
1528 ERR("Unknown consumer_data type");
1532 /* Handle stream on the relayd if the output is on the network */
1534 unsigned long netlen
= len
;
1537 * Lock the control socket for the complete duration of the function
1538 * since from this point on we will use the socket.
1540 if (stream
->metadata_flag
) {
1541 /* Metadata requires the control socket. */
1542 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1543 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1546 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1551 /* Use the returned socket. */
1554 /* Write metadata stream id before payload */
1555 if (stream
->metadata_flag
) {
1556 ret
= write_relayd_metadata_id(outfd
, stream
, relayd
, padding
);
1563 /* No streaming, we have to set the len with the full padding */
1567 * Check if we need to change the tracefile before writing the packet.
1569 if (stream
->chan
->tracefile_size
> 0 &&
1570 (stream
->tracefile_size_current
+ len
) >
1571 stream
->chan
->tracefile_size
) {
1572 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1573 stream
->name
, stream
->chan
->tracefile_size
,
1574 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1575 stream
->out_fd
, &(stream
->tracefile_count_current
),
1578 ERR("Rotating output file");
1581 outfd
= stream
->out_fd
;
1583 if (stream
->index_fd
>= 0) {
1584 ret
= index_create_file(stream
->chan
->pathname
,
1585 stream
->name
, stream
->uid
, stream
->gid
,
1586 stream
->chan
->tracefile_size
,
1587 stream
->tracefile_count_current
);
1591 stream
->index_fd
= ret
;
1594 /* Reset current size because we just perform a rotation. */
1595 stream
->tracefile_size_current
= 0;
1596 stream
->out_fd_offset
= 0;
1599 stream
->tracefile_size_current
+= len
;
1601 index
->offset
= htobe64(stream
->out_fd_offset
);
1606 * This call guarantee that len or less is returned. It's impossible to
1607 * receive a ret value that is bigger than len.
1609 ret
= lttng_write(outfd
, mmap_base
+ mmap_offset
, len
);
1610 DBG("Consumer mmap write() ret %zd (len %lu)", ret
, len
);
1611 if (ret
< 0 || ((size_t) ret
!= len
)) {
1613 * Report error to caller if nothing was written else at least send the
1621 /* Socket operation failed. We consider the relayd dead */
1622 if (errno
== EPIPE
|| errno
== EINVAL
|| errno
== EBADF
) {
1624 * This is possible if the fd is closed on the other side
1625 * (outfd) or any write problem. It can be verbose a bit for a
1626 * normal execution if for instance the relayd is stopped
1627 * abruptly. This can happen so set this to a DBG statement.
1629 DBG("Consumer mmap write detected relayd hang up");
1631 /* Unhandled error, print it and stop function right now. */
1632 PERROR("Error in write mmap (ret %zd != len %lu)", ret
, len
);
1636 stream
->output_written
+= ret
;
1638 /* This call is useless on a socket so better save a syscall. */
1640 /* This won't block, but will start writeout asynchronously */
1641 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, len
,
1642 SYNC_FILE_RANGE_WRITE
);
1643 stream
->out_fd_offset
+= len
;
1645 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1649 * This is a special case that the relayd has closed its socket. Let's
1650 * cleanup the relayd object and all associated streams.
1652 if (relayd
&& relayd_hang_up
) {
1653 cleanup_relayd(relayd
, ctx
);
1657 /* Unlock only if ctrl socket used */
1658 if (relayd
&& stream
->metadata_flag
) {
1659 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1667 * Splice the data from the ring buffer to the tracefile.
1669 * It must be called with the stream lock held.
1671 * Returns the number of bytes spliced.
1673 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1674 struct lttng_consumer_local_data
*ctx
,
1675 struct lttng_consumer_stream
*stream
, unsigned long len
,
1676 unsigned long padding
,
1677 struct ctf_packet_index
*index
)
1679 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1681 off_t orig_offset
= stream
->out_fd_offset
;
1682 int fd
= stream
->wait_fd
;
1683 /* Default is on the disk */
1684 int outfd
= stream
->out_fd
;
1685 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1687 unsigned int relayd_hang_up
= 0;
1689 switch (consumer_data
.type
) {
1690 case LTTNG_CONSUMER_KERNEL
:
1692 case LTTNG_CONSUMER32_UST
:
1693 case LTTNG_CONSUMER64_UST
:
1694 /* Not supported for user space tracing */
1697 ERR("Unknown consumer_data type");
1701 /* RCU lock for the relayd pointer */
1704 /* Flag that the current stream if set for network streaming. */
1705 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1706 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1707 if (relayd
== NULL
) {
1712 splice_pipe
= stream
->splice_pipe
;
1714 /* Write metadata stream id before payload */
1716 unsigned long total_len
= len
;
1718 if (stream
->metadata_flag
) {
1720 * Lock the control socket for the complete duration of the function
1721 * since from this point on we will use the socket.
1723 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1725 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
, relayd
,
1733 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1736 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1742 /* Use the returned socket. */
1745 /* No streaming, we have to set the len with the full padding */
1749 * Check if we need to change the tracefile before writing the packet.
1751 if (stream
->chan
->tracefile_size
> 0 &&
1752 (stream
->tracefile_size_current
+ len
) >
1753 stream
->chan
->tracefile_size
) {
1754 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1755 stream
->name
, stream
->chan
->tracefile_size
,
1756 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1757 stream
->out_fd
, &(stream
->tracefile_count_current
),
1761 ERR("Rotating output file");
1764 outfd
= stream
->out_fd
;
1766 if (stream
->index_fd
>= 0) {
1767 ret
= index_create_file(stream
->chan
->pathname
,
1768 stream
->name
, stream
->uid
, stream
->gid
,
1769 stream
->chan
->tracefile_size
,
1770 stream
->tracefile_count_current
);
1775 stream
->index_fd
= ret
;
1778 /* Reset current size because we just perform a rotation. */
1779 stream
->tracefile_size_current
= 0;
1780 stream
->out_fd_offset
= 0;
1783 stream
->tracefile_size_current
+= len
;
1784 index
->offset
= htobe64(stream
->out_fd_offset
);
1788 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1789 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1790 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1791 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1792 DBG("splice chan to pipe, ret %zd", ret_splice
);
1793 if (ret_splice
< 0) {
1796 PERROR("Error in relay splice");
1800 /* Handle stream on the relayd if the output is on the network */
1801 if (relayd
&& stream
->metadata_flag
) {
1802 size_t metadata_payload_size
=
1803 sizeof(struct lttcomm_relayd_metadata_payload
);
1805 /* Update counter to fit the spliced data */
1806 ret_splice
+= metadata_payload_size
;
1807 len
+= metadata_payload_size
;
1809 * We do this so the return value can match the len passed as
1810 * argument to this function.
1812 written
-= metadata_payload_size
;
1815 /* Splice data out */
1816 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1817 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1818 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1820 if (ret_splice
< 0) {
1825 } else if (ret_splice
> len
) {
1827 * We don't expect this code path to be executed but you never know
1828 * so this is an extra protection agains a buggy splice().
1831 written
+= ret_splice
;
1832 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice
,
1836 /* All good, update current len and continue. */
1840 /* This call is useless on a socket so better save a syscall. */
1842 /* This won't block, but will start writeout asynchronously */
1843 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1844 SYNC_FILE_RANGE_WRITE
);
1845 stream
->out_fd_offset
+= ret_splice
;
1847 stream
->output_written
+= ret_splice
;
1848 written
+= ret_splice
;
1850 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1855 * This is a special case that the relayd has closed its socket. Let's
1856 * cleanup the relayd object and all associated streams.
1858 if (relayd
&& relayd_hang_up
) {
1859 cleanup_relayd(relayd
, ctx
);
1860 /* Skip splice error so the consumer does not fail */
1865 /* send the appropriate error description to sessiond */
1868 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1871 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1874 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1879 if (relayd
&& stream
->metadata_flag
) {
1880 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1888 * Take a snapshot for a specific fd
1890 * Returns 0 on success, < 0 on error
1892 int lttng_consumer_take_snapshot(struct lttng_consumer_stream
*stream
)
1894 switch (consumer_data
.type
) {
1895 case LTTNG_CONSUMER_KERNEL
:
1896 return lttng_kconsumer_take_snapshot(stream
);
1897 case LTTNG_CONSUMER32_UST
:
1898 case LTTNG_CONSUMER64_UST
:
1899 return lttng_ustconsumer_take_snapshot(stream
);
1901 ERR("Unknown consumer_data type");
1908 * Get the produced position
1910 * Returns 0 on success, < 0 on error
1912 int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
1915 switch (consumer_data
.type
) {
1916 case LTTNG_CONSUMER_KERNEL
:
1917 return lttng_kconsumer_get_produced_snapshot(stream
, pos
);
1918 case LTTNG_CONSUMER32_UST
:
1919 case LTTNG_CONSUMER64_UST
:
1920 return lttng_ustconsumer_get_produced_snapshot(stream
, pos
);
1922 ERR("Unknown consumer_data type");
1928 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1929 int sock
, struct pollfd
*consumer_sockpoll
)
1931 switch (consumer_data
.type
) {
1932 case LTTNG_CONSUMER_KERNEL
:
1933 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
1934 case LTTNG_CONSUMER32_UST
:
1935 case LTTNG_CONSUMER64_UST
:
1936 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
1938 ERR("Unknown consumer_data type");
1944 void lttng_consumer_close_all_metadata(void)
1946 switch (consumer_data
.type
) {
1947 case LTTNG_CONSUMER_KERNEL
:
1949 * The Kernel consumer has a different metadata scheme so we don't
1950 * close anything because the stream will be closed by the session
1954 case LTTNG_CONSUMER32_UST
:
1955 case LTTNG_CONSUMER64_UST
:
1957 * Close all metadata streams. The metadata hash table is passed and
1958 * this call iterates over it by closing all wakeup fd. This is safe
1959 * because at this point we are sure that the metadata producer is
1960 * either dead or blocked.
1962 lttng_ustconsumer_close_all_metadata(metadata_ht
);
1965 ERR("Unknown consumer_data type");
1971 * Clean up a metadata stream and free its memory.
1973 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
1974 struct lttng_ht
*ht
)
1976 struct lttng_consumer_channel
*free_chan
= NULL
;
1980 * This call should NEVER receive regular stream. It must always be
1981 * metadata stream and this is crucial for data structure synchronization.
1983 assert(stream
->metadata_flag
);
1985 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
1987 pthread_mutex_lock(&consumer_data
.lock
);
1988 pthread_mutex_lock(&stream
->chan
->lock
);
1989 pthread_mutex_lock(&stream
->lock
);
1991 /* Remove any reference to that stream. */
1992 consumer_stream_delete(stream
, ht
);
1994 /* Close down everything including the relayd if one. */
1995 consumer_stream_close(stream
);
1996 /* Destroy tracer buffers of the stream. */
1997 consumer_stream_destroy_buffers(stream
);
1999 /* Atomically decrement channel refcount since other threads can use it. */
2000 if (!uatomic_sub_return(&stream
->chan
->refcount
, 1)
2001 && !uatomic_read(&stream
->chan
->nb_init_stream_left
)) {
2002 /* Go for channel deletion! */
2003 free_chan
= stream
->chan
;
2007 * Nullify the stream reference so it is not used after deletion. The
2008 * channel lock MUST be acquired before being able to check for a NULL
2011 stream
->chan
->metadata_stream
= NULL
;
2013 pthread_mutex_unlock(&stream
->lock
);
2014 pthread_mutex_unlock(&stream
->chan
->lock
);
2015 pthread_mutex_unlock(&consumer_data
.lock
);
2018 consumer_del_channel(free_chan
);
2021 consumer_stream_free(stream
);
2025 * Action done with the metadata stream when adding it to the consumer internal
2026 * data structures to handle it.
2028 int consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
)
2030 struct lttng_ht
*ht
= metadata_ht
;
2032 struct lttng_ht_iter iter
;
2033 struct lttng_ht_node_u64
*node
;
2038 DBG3("Adding metadata stream %" PRIu64
" to hash table", stream
->key
);
2040 pthread_mutex_lock(&consumer_data
.lock
);
2041 pthread_mutex_lock(&stream
->chan
->lock
);
2042 pthread_mutex_lock(&stream
->chan
->timer_lock
);
2043 pthread_mutex_lock(&stream
->lock
);
2046 * From here, refcounts are updated so be _careful_ when returning an error
2053 * Lookup the stream just to make sure it does not exist in our internal
2054 * state. This should NEVER happen.
2056 lttng_ht_lookup(ht
, &stream
->key
, &iter
);
2057 node
= lttng_ht_iter_get_node_u64(&iter
);
2061 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2062 * in terms of destroying the associated channel, because the action that
2063 * causes the count to become 0 also causes a stream to be added. The
2064 * channel deletion will thus be triggered by the following removal of this
2067 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
2068 /* Increment refcount before decrementing nb_init_stream_left */
2070 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
2073 lttng_ht_add_unique_u64(ht
, &stream
->node
);
2075 lttng_ht_add_unique_u64(consumer_data
.stream_per_chan_id_ht
,
2076 &stream
->node_channel_id
);
2079 * Add stream to the stream_list_ht of the consumer data. No need to steal
2080 * the key since the HT does not use it and we allow to add redundant keys
2083 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
2087 pthread_mutex_unlock(&stream
->lock
);
2088 pthread_mutex_unlock(&stream
->chan
->lock
);
2089 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
2090 pthread_mutex_unlock(&consumer_data
.lock
);
2095 * Delete data stream that are flagged for deletion (endpoint_status).
2097 static void validate_endpoint_status_data_stream(void)
2099 struct lttng_ht_iter iter
;
2100 struct lttng_consumer_stream
*stream
;
2102 DBG("Consumer delete flagged data stream");
2105 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2106 /* Validate delete flag of the stream */
2107 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2110 /* Delete it right now */
2111 consumer_del_stream(stream
, data_ht
);
2117 * Delete metadata stream that are flagged for deletion (endpoint_status).
2119 static void validate_endpoint_status_metadata_stream(
2120 struct lttng_poll_event
*pollset
)
2122 struct lttng_ht_iter iter
;
2123 struct lttng_consumer_stream
*stream
;
2125 DBG("Consumer delete flagged metadata stream");
2130 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2131 /* Validate delete flag of the stream */
2132 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2136 * Remove from pollset so the metadata thread can continue without
2137 * blocking on a deleted stream.
2139 lttng_poll_del(pollset
, stream
->wait_fd
);
2141 /* Delete it right now */
2142 consumer_del_metadata_stream(stream
, metadata_ht
);
2148 * Thread polls on metadata file descriptor and write them on disk or on the
2151 void *consumer_thread_metadata_poll(void *data
)
2153 int ret
, i
, pollfd
, err
= -1;
2154 uint32_t revents
, nb_fd
;
2155 struct lttng_consumer_stream
*stream
= NULL
;
2156 struct lttng_ht_iter iter
;
2157 struct lttng_ht_node_u64
*node
;
2158 struct lttng_poll_event events
;
2159 struct lttng_consumer_local_data
*ctx
= data
;
2162 rcu_register_thread();
2164 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_METADATA
);
2166 if (testpoint(consumerd_thread_metadata
)) {
2167 goto error_testpoint
;
2170 health_code_update();
2172 DBG("Thread metadata poll started");
2174 /* Size is set to 1 for the consumer_metadata pipe */
2175 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2177 ERR("Poll set creation failed");
2181 ret
= lttng_poll_add(&events
,
2182 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
), LPOLLIN
);
2188 DBG("Metadata main loop started");
2192 health_code_update();
2193 health_poll_entry();
2194 DBG("Metadata poll wait");
2195 ret
= lttng_poll_wait(&events
, -1);
2196 DBG("Metadata poll return from wait with %d fd(s)",
2197 LTTNG_POLL_GETNB(&events
));
2199 DBG("Metadata event catched in thread");
2201 if (errno
== EINTR
) {
2202 ERR("Poll EINTR catched");
2205 if (LTTNG_POLL_GETNB(&events
) == 0) {
2206 err
= 0; /* All is OK */
2213 /* From here, the event is a metadata wait fd */
2214 for (i
= 0; i
< nb_fd
; i
++) {
2215 health_code_update();
2217 revents
= LTTNG_POLL_GETEV(&events
, i
);
2218 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2221 /* No activity for this FD (poll implementation). */
2225 if (pollfd
== lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
)) {
2226 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2227 DBG("Metadata thread pipe hung up");
2229 * Remove the pipe from the poll set and continue the loop
2230 * since their might be data to consume.
2232 lttng_poll_del(&events
,
2233 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2234 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2236 } else if (revents
& LPOLLIN
) {
2239 pipe_len
= lttng_pipe_read(ctx
->consumer_metadata_pipe
,
2240 &stream
, sizeof(stream
));
2241 if (pipe_len
< sizeof(stream
)) {
2242 PERROR("read metadata stream");
2244 * Continue here to handle the rest of the streams.
2249 /* A NULL stream means that the state has changed. */
2250 if (stream
== NULL
) {
2251 /* Check for deleted streams. */
2252 validate_endpoint_status_metadata_stream(&events
);
2256 DBG("Adding metadata stream %d to poll set",
2259 /* Add metadata stream to the global poll events list */
2260 lttng_poll_add(&events
, stream
->wait_fd
,
2261 LPOLLIN
| LPOLLPRI
| LPOLLHUP
);
2264 /* Handle other stream */
2270 uint64_t tmp_id
= (uint64_t) pollfd
;
2272 lttng_ht_lookup(metadata_ht
, &tmp_id
, &iter
);
2274 node
= lttng_ht_iter_get_node_u64(&iter
);
2277 stream
= caa_container_of(node
, struct lttng_consumer_stream
,
2280 /* Check for error event */
2281 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2282 DBG("Metadata fd %d is hup|err.", pollfd
);
2283 if (!stream
->hangup_flush_done
2284 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2285 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2286 DBG("Attempting to flush and consume the UST buffers");
2287 lttng_ustconsumer_on_stream_hangup(stream
);
2289 /* We just flushed the stream now read it. */
2291 health_code_update();
2293 len
= ctx
->on_buffer_ready(stream
, ctx
);
2295 * We don't check the return value here since if we get
2296 * a negative len, it means an error occured thus we
2297 * simply remove it from the poll set and free the
2303 lttng_poll_del(&events
, stream
->wait_fd
);
2305 * This call update the channel states, closes file descriptors
2306 * and securely free the stream.
2308 consumer_del_metadata_stream(stream
, metadata_ht
);
2309 } else if (revents
& (LPOLLIN
| LPOLLPRI
)) {
2310 /* Get the data out of the metadata file descriptor */
2311 DBG("Metadata available on fd %d", pollfd
);
2312 assert(stream
->wait_fd
== pollfd
);
2315 health_code_update();
2317 len
= ctx
->on_buffer_ready(stream
, ctx
);
2319 * We don't check the return value here since if we get
2320 * a negative len, it means an error occured thus we
2321 * simply remove it from the poll set and free the
2326 /* It's ok to have an unavailable sub-buffer */
2327 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2328 /* Clean up stream from consumer and free it. */
2329 lttng_poll_del(&events
, stream
->wait_fd
);
2330 consumer_del_metadata_stream(stream
, metadata_ht
);
2334 /* Release RCU lock for the stream looked up */
2342 DBG("Metadata poll thread exiting");
2344 lttng_poll_clean(&events
);
2349 ERR("Health error occurred in %s", __func__
);
2351 health_unregister(health_consumerd
);
2352 rcu_unregister_thread();
2357 * This thread polls the fds in the set to consume the data and write
2358 * it to tracefile if necessary.
2360 void *consumer_thread_data_poll(void *data
)
2362 int num_rdy
, num_hup
, high_prio
, ret
, i
, err
= -1;
2363 struct pollfd
*pollfd
= NULL
;
2364 /* local view of the streams */
2365 struct lttng_consumer_stream
**local_stream
= NULL
, *new_stream
= NULL
;
2366 /* local view of consumer_data.fds_count */
2368 struct lttng_consumer_local_data
*ctx
= data
;
2371 rcu_register_thread();
2373 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_DATA
);
2375 if (testpoint(consumerd_thread_data
)) {
2376 goto error_testpoint
;
2379 health_code_update();
2381 local_stream
= zmalloc(sizeof(struct lttng_consumer_stream
*));
2382 if (local_stream
== NULL
) {
2383 PERROR("local_stream malloc");
2388 health_code_update();
2394 * the fds set has been updated, we need to update our
2395 * local array as well
2397 pthread_mutex_lock(&consumer_data
.lock
);
2398 if (consumer_data
.need_update
) {
2403 local_stream
= NULL
;
2406 * Allocate for all fds +1 for the consumer_data_pipe and +1 for
2409 pollfd
= zmalloc((consumer_data
.stream_count
+ 2) * sizeof(struct pollfd
));
2410 if (pollfd
== NULL
) {
2411 PERROR("pollfd malloc");
2412 pthread_mutex_unlock(&consumer_data
.lock
);
2416 local_stream
= zmalloc((consumer_data
.stream_count
+ 2) *
2417 sizeof(struct lttng_consumer_stream
*));
2418 if (local_stream
== NULL
) {
2419 PERROR("local_stream malloc");
2420 pthread_mutex_unlock(&consumer_data
.lock
);
2423 ret
= update_poll_array(ctx
, &pollfd
, local_stream
,
2426 ERR("Error in allocating pollfd or local_outfds");
2427 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2428 pthread_mutex_unlock(&consumer_data
.lock
);
2432 consumer_data
.need_update
= 0;
2434 pthread_mutex_unlock(&consumer_data
.lock
);
2436 /* No FDs and consumer_quit, consumer_cleanup the thread */
2437 if (nb_fd
== 0 && consumer_quit
== 1) {
2438 err
= 0; /* All is OK */
2441 /* poll on the array of fds */
2443 DBG("polling on %d fd", nb_fd
+ 2);
2444 health_poll_entry();
2445 num_rdy
= poll(pollfd
, nb_fd
+ 2, -1);
2447 DBG("poll num_rdy : %d", num_rdy
);
2448 if (num_rdy
== -1) {
2450 * Restart interrupted system call.
2452 if (errno
== EINTR
) {
2455 PERROR("Poll error");
2456 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2458 } else if (num_rdy
== 0) {
2459 DBG("Polling thread timed out");
2464 * If the consumer_data_pipe triggered poll go directly to the
2465 * beginning of the loop to update the array. We want to prioritize
2466 * array update over low-priority reads.
2468 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2469 ssize_t pipe_readlen
;
2471 DBG("consumer_data_pipe wake up");
2472 pipe_readlen
= lttng_pipe_read(ctx
->consumer_data_pipe
,
2473 &new_stream
, sizeof(new_stream
));
2474 if (pipe_readlen
< sizeof(new_stream
)) {
2475 PERROR("Consumer data pipe");
2476 /* Continue so we can at least handle the current stream(s). */
2481 * If the stream is NULL, just ignore it. It's also possible that
2482 * the sessiond poll thread changed the consumer_quit state and is
2483 * waking us up to test it.
2485 if (new_stream
== NULL
) {
2486 validate_endpoint_status_data_stream();
2490 /* Continue to update the local streams and handle prio ones */
2494 /* Handle wakeup pipe. */
2495 if (pollfd
[nb_fd
+ 1].revents
& (POLLIN
| POLLPRI
)) {
2497 ssize_t pipe_readlen
;
2499 pipe_readlen
= lttng_pipe_read(ctx
->consumer_wakeup_pipe
, &dummy
,
2501 if (pipe_readlen
< 0) {
2502 PERROR("Consumer data wakeup pipe");
2504 /* We've been awakened to handle stream(s). */
2505 ctx
->has_wakeup
= 0;
2508 /* Take care of high priority channels first. */
2509 for (i
= 0; i
< nb_fd
; i
++) {
2510 health_code_update();
2512 if (local_stream
[i
] == NULL
) {
2515 if (pollfd
[i
].revents
& POLLPRI
) {
2516 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2518 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2519 /* it's ok to have an unavailable sub-buffer */
2520 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2521 /* Clean the stream and free it. */
2522 consumer_del_stream(local_stream
[i
], data_ht
);
2523 local_stream
[i
] = NULL
;
2524 } else if (len
> 0) {
2525 local_stream
[i
]->data_read
= 1;
2531 * If we read high prio channel in this loop, try again
2532 * for more high prio data.
2538 /* Take care of low priority channels. */
2539 for (i
= 0; i
< nb_fd
; i
++) {
2540 health_code_update();
2542 if (local_stream
[i
] == NULL
) {
2545 if ((pollfd
[i
].revents
& POLLIN
) ||
2546 local_stream
[i
]->hangup_flush_done
||
2547 local_stream
[i
]->has_data
) {
2548 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2549 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2550 /* it's ok to have an unavailable sub-buffer */
2551 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2552 /* Clean the stream and free it. */
2553 consumer_del_stream(local_stream
[i
], data_ht
);
2554 local_stream
[i
] = NULL
;
2555 } else if (len
> 0) {
2556 local_stream
[i
]->data_read
= 1;
2561 /* Handle hangup and errors */
2562 for (i
= 0; i
< nb_fd
; i
++) {
2563 health_code_update();
2565 if (local_stream
[i
] == NULL
) {
2568 if (!local_stream
[i
]->hangup_flush_done
2569 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2570 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2571 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2572 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2574 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2575 /* Attempt read again, for the data we just flushed. */
2576 local_stream
[i
]->data_read
= 1;
2579 * If the poll flag is HUP/ERR/NVAL and we have
2580 * read no data in this pass, we can remove the
2581 * stream from its hash table.
2583 if ((pollfd
[i
].revents
& POLLHUP
)) {
2584 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2585 if (!local_stream
[i
]->data_read
) {
2586 consumer_del_stream(local_stream
[i
], data_ht
);
2587 local_stream
[i
] = NULL
;
2590 } else if (pollfd
[i
].revents
& POLLERR
) {
2591 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2592 if (!local_stream
[i
]->data_read
) {
2593 consumer_del_stream(local_stream
[i
], data_ht
);
2594 local_stream
[i
] = NULL
;
2597 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2598 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2599 if (!local_stream
[i
]->data_read
) {
2600 consumer_del_stream(local_stream
[i
], data_ht
);
2601 local_stream
[i
] = NULL
;
2605 if (local_stream
[i
] != NULL
) {
2606 local_stream
[i
]->data_read
= 0;
2613 DBG("polling thread exiting");
2618 * Close the write side of the pipe so epoll_wait() in
2619 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2620 * read side of the pipe. If we close them both, epoll_wait strangely does
2621 * not return and could create a endless wait period if the pipe is the
2622 * only tracked fd in the poll set. The thread will take care of closing
2625 (void) lttng_pipe_write_close(ctx
->consumer_metadata_pipe
);
2630 ERR("Health error occurred in %s", __func__
);
2632 health_unregister(health_consumerd
);
2634 rcu_unregister_thread();
2639 * Close wake-up end of each stream belonging to the channel. This will
2640 * allow the poll() on the stream read-side to detect when the
2641 * write-side (application) finally closes them.
2644 void consumer_close_channel_streams(struct lttng_consumer_channel
*channel
)
2646 struct lttng_ht
*ht
;
2647 struct lttng_consumer_stream
*stream
;
2648 struct lttng_ht_iter iter
;
2650 ht
= consumer_data
.stream_per_chan_id_ht
;
2653 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2654 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
2655 ht
->match_fct
, &channel
->key
,
2656 &iter
.iter
, stream
, node_channel_id
.node
) {
2658 * Protect against teardown with mutex.
2660 pthread_mutex_lock(&stream
->lock
);
2661 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
2664 switch (consumer_data
.type
) {
2665 case LTTNG_CONSUMER_KERNEL
:
2667 case LTTNG_CONSUMER32_UST
:
2668 case LTTNG_CONSUMER64_UST
:
2669 if (stream
->metadata_flag
) {
2670 /* Safe and protected by the stream lock. */
2671 lttng_ustconsumer_close_metadata(stream
->chan
);
2674 * Note: a mutex is taken internally within
2675 * liblttng-ust-ctl to protect timer wakeup_fd
2676 * use from concurrent close.
2678 lttng_ustconsumer_close_stream_wakeup(stream
);
2682 ERR("Unknown consumer_data type");
2686 pthread_mutex_unlock(&stream
->lock
);
2691 static void destroy_channel_ht(struct lttng_ht
*ht
)
2693 struct lttng_ht_iter iter
;
2694 struct lttng_consumer_channel
*channel
;
2702 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, channel
, wait_fd_node
.node
) {
2703 ret
= lttng_ht_del(ht
, &iter
);
2708 lttng_ht_destroy(ht
);
2712 * This thread polls the channel fds to detect when they are being
2713 * closed. It closes all related streams if the channel is detected as
2714 * closed. It is currently only used as a shim layer for UST because the
2715 * consumerd needs to keep the per-stream wakeup end of pipes open for
2718 void *consumer_thread_channel_poll(void *data
)
2720 int ret
, i
, pollfd
, err
= -1;
2721 uint32_t revents
, nb_fd
;
2722 struct lttng_consumer_channel
*chan
= NULL
;
2723 struct lttng_ht_iter iter
;
2724 struct lttng_ht_node_u64
*node
;
2725 struct lttng_poll_event events
;
2726 struct lttng_consumer_local_data
*ctx
= data
;
2727 struct lttng_ht
*channel_ht
;
2729 rcu_register_thread();
2731 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_CHANNEL
);
2733 if (testpoint(consumerd_thread_channel
)) {
2734 goto error_testpoint
;
2737 health_code_update();
2739 channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2741 /* ENOMEM at this point. Better to bail out. */
2745 DBG("Thread channel poll started");
2747 /* Size is set to 1 for the consumer_channel pipe */
2748 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2750 ERR("Poll set creation failed");
2754 ret
= lttng_poll_add(&events
, ctx
->consumer_channel_pipe
[0], LPOLLIN
);
2760 DBG("Channel main loop started");
2764 health_code_update();
2765 DBG("Channel poll wait");
2766 health_poll_entry();
2767 ret
= lttng_poll_wait(&events
, -1);
2768 DBG("Channel poll return from wait with %d fd(s)",
2769 LTTNG_POLL_GETNB(&events
));
2771 DBG("Channel event catched in thread");
2773 if (errno
== EINTR
) {
2774 ERR("Poll EINTR catched");
2777 if (LTTNG_POLL_GETNB(&events
) == 0) {
2778 err
= 0; /* All is OK */
2785 /* From here, the event is a channel wait fd */
2786 for (i
= 0; i
< nb_fd
; i
++) {
2787 health_code_update();
2789 revents
= LTTNG_POLL_GETEV(&events
, i
);
2790 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2793 /* No activity for this FD (poll implementation). */
2797 if (pollfd
== ctx
->consumer_channel_pipe
[0]) {
2798 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2799 DBG("Channel thread pipe hung up");
2801 * Remove the pipe from the poll set and continue the loop
2802 * since their might be data to consume.
2804 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
2806 } else if (revents
& LPOLLIN
) {
2807 enum consumer_channel_action action
;
2810 ret
= read_channel_pipe(ctx
, &chan
, &key
, &action
);
2812 ERR("Error reading channel pipe");
2817 case CONSUMER_CHANNEL_ADD
:
2818 DBG("Adding channel %d to poll set",
2821 lttng_ht_node_init_u64(&chan
->wait_fd_node
,
2824 lttng_ht_add_unique_u64(channel_ht
,
2825 &chan
->wait_fd_node
);
2827 /* Add channel to the global poll events list */
2828 lttng_poll_add(&events
, chan
->wait_fd
,
2829 LPOLLIN
| LPOLLPRI
);
2831 case CONSUMER_CHANNEL_DEL
:
2834 * This command should never be called if the channel
2835 * has streams monitored by either the data or metadata
2836 * thread. The consumer only notify this thread with a
2837 * channel del. command if it receives a destroy
2838 * channel command from the session daemon that send it
2839 * if a command prior to the GET_CHANNEL failed.
2843 chan
= consumer_find_channel(key
);
2846 ERR("UST consumer get channel key %" PRIu64
" not found for del channel", key
);
2849 lttng_poll_del(&events
, chan
->wait_fd
);
2850 iter
.iter
.node
= &chan
->wait_fd_node
.node
;
2851 ret
= lttng_ht_del(channel_ht
, &iter
);
2854 switch (consumer_data
.type
) {
2855 case LTTNG_CONSUMER_KERNEL
:
2857 case LTTNG_CONSUMER32_UST
:
2858 case LTTNG_CONSUMER64_UST
:
2859 health_code_update();
2860 /* Destroy streams that might have been left in the stream list. */
2861 clean_channel_stream_list(chan
);
2864 ERR("Unknown consumer_data type");
2869 * Release our own refcount. Force channel deletion even if
2870 * streams were not initialized.
2872 if (!uatomic_sub_return(&chan
->refcount
, 1)) {
2873 consumer_del_channel(chan
);
2878 case CONSUMER_CHANNEL_QUIT
:
2880 * Remove the pipe from the poll set and continue the loop
2881 * since their might be data to consume.
2883 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
2886 ERR("Unknown action");
2891 /* Handle other stream */
2897 uint64_t tmp_id
= (uint64_t) pollfd
;
2899 lttng_ht_lookup(channel_ht
, &tmp_id
, &iter
);
2901 node
= lttng_ht_iter_get_node_u64(&iter
);
2904 chan
= caa_container_of(node
, struct lttng_consumer_channel
,
2907 /* Check for error event */
2908 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2909 DBG("Channel fd %d is hup|err.", pollfd
);
2911 lttng_poll_del(&events
, chan
->wait_fd
);
2912 ret
= lttng_ht_del(channel_ht
, &iter
);
2916 * This will close the wait fd for each stream associated to
2917 * this channel AND monitored by the data/metadata thread thus
2918 * will be clean by the right thread.
2920 consumer_close_channel_streams(chan
);
2922 /* Release our own refcount */
2923 if (!uatomic_sub_return(&chan
->refcount
, 1)
2924 && !uatomic_read(&chan
->nb_init_stream_left
)) {
2925 consumer_del_channel(chan
);
2929 /* Release RCU lock for the channel looked up */
2937 lttng_poll_clean(&events
);
2939 destroy_channel_ht(channel_ht
);
2942 DBG("Channel poll thread exiting");
2945 ERR("Health error occurred in %s", __func__
);
2947 health_unregister(health_consumerd
);
2948 rcu_unregister_thread();
2952 static int set_metadata_socket(struct lttng_consumer_local_data
*ctx
,
2953 struct pollfd
*sockpoll
, int client_socket
)
2960 ret
= lttng_consumer_poll_socket(sockpoll
);
2964 DBG("Metadata connection on client_socket");
2966 /* Blocking call, waiting for transmission */
2967 ctx
->consumer_metadata_socket
= lttcomm_accept_unix_sock(client_socket
);
2968 if (ctx
->consumer_metadata_socket
< 0) {
2969 WARN("On accept metadata");
2980 * This thread listens on the consumerd socket and receives the file
2981 * descriptors from the session daemon.
2983 void *consumer_thread_sessiond_poll(void *data
)
2985 int sock
= -1, client_socket
, ret
, err
= -1;
2987 * structure to poll for incoming data on communication socket avoids
2988 * making blocking sockets.
2990 struct pollfd consumer_sockpoll
[2];
2991 struct lttng_consumer_local_data
*ctx
= data
;
2993 rcu_register_thread();
2995 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_SESSIOND
);
2997 if (testpoint(consumerd_thread_sessiond
)) {
2998 goto error_testpoint
;
3001 health_code_update();
3003 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
3004 unlink(ctx
->consumer_command_sock_path
);
3005 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
3006 if (client_socket
< 0) {
3007 ERR("Cannot create command socket");
3011 ret
= lttcomm_listen_unix_sock(client_socket
);
3016 DBG("Sending ready command to lttng-sessiond");
3017 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
3018 /* return < 0 on error, but == 0 is not fatal */
3020 ERR("Error sending ready command to lttng-sessiond");
3024 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3025 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
3026 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
3027 consumer_sockpoll
[1].fd
= client_socket
;
3028 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3030 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3038 DBG("Connection on client_socket");
3040 /* Blocking call, waiting for transmission */
3041 sock
= lttcomm_accept_unix_sock(client_socket
);
3048 * Setup metadata socket which is the second socket connection on the
3049 * command unix socket.
3051 ret
= set_metadata_socket(ctx
, consumer_sockpoll
, client_socket
);
3060 /* This socket is not useful anymore. */
3061 ret
= close(client_socket
);
3063 PERROR("close client_socket");
3067 /* update the polling structure to poll on the established socket */
3068 consumer_sockpoll
[1].fd
= sock
;
3069 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3072 health_code_update();
3074 health_poll_entry();
3075 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3084 DBG("Incoming command on sock");
3085 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
3088 * This could simply be a session daemon quitting. Don't output
3091 DBG("Communication interrupted on command socket");
3095 if (consumer_quit
) {
3096 DBG("consumer_thread_receive_fds received quit from signal");
3097 err
= 0; /* All is OK */
3100 DBG("received command on sock");
3106 DBG("Consumer thread sessiond poll exiting");
3109 * Close metadata streams since the producer is the session daemon which
3112 * NOTE: for now, this only applies to the UST tracer.
3114 lttng_consumer_close_all_metadata();
3117 * when all fds have hung up, the polling thread
3123 * Notify the data poll thread to poll back again and test the
3124 * consumer_quit state that we just set so to quit gracefully.
3126 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
3128 notify_channel_pipe(ctx
, NULL
, -1, CONSUMER_CHANNEL_QUIT
);
3130 notify_health_quit_pipe(health_quit_pipe
);
3132 /* Cleaning up possibly open sockets. */
3136 PERROR("close sock sessiond poll");
3139 if (client_socket
>= 0) {
3140 ret
= close(client_socket
);
3142 PERROR("close client_socket sessiond poll");
3149 ERR("Health error occurred in %s", __func__
);
3151 health_unregister(health_consumerd
);
3153 rcu_unregister_thread();
3157 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
3158 struct lttng_consumer_local_data
*ctx
)
3162 pthread_mutex_lock(&stream
->lock
);
3163 if (stream
->metadata_flag
) {
3164 pthread_mutex_lock(&stream
->metadata_rdv_lock
);
3167 switch (consumer_data
.type
) {
3168 case LTTNG_CONSUMER_KERNEL
:
3169 ret
= lttng_kconsumer_read_subbuffer(stream
, ctx
);
3171 case LTTNG_CONSUMER32_UST
:
3172 case LTTNG_CONSUMER64_UST
:
3173 ret
= lttng_ustconsumer_read_subbuffer(stream
, ctx
);
3176 ERR("Unknown consumer_data type");
3182 if (stream
->metadata_flag
) {
3183 pthread_cond_broadcast(&stream
->metadata_rdv
);
3184 pthread_mutex_unlock(&stream
->metadata_rdv_lock
);
3186 pthread_mutex_unlock(&stream
->lock
);
3190 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3192 switch (consumer_data
.type
) {
3193 case LTTNG_CONSUMER_KERNEL
:
3194 return lttng_kconsumer_on_recv_stream(stream
);
3195 case LTTNG_CONSUMER32_UST
:
3196 case LTTNG_CONSUMER64_UST
:
3197 return lttng_ustconsumer_on_recv_stream(stream
);
3199 ERR("Unknown consumer_data type");
3206 * Allocate and set consumer data hash tables.
3208 int lttng_consumer_init(void)
3210 consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3211 if (!consumer_data
.channel_ht
) {
3215 consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3216 if (!consumer_data
.relayd_ht
) {
3220 consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3221 if (!consumer_data
.stream_list_ht
) {
3225 consumer_data
.stream_per_chan_id_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3226 if (!consumer_data
.stream_per_chan_id_ht
) {
3230 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3235 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3247 * Process the ADD_RELAYD command receive by a consumer.
3249 * This will create a relayd socket pair and add it to the relayd hash table.
3250 * The caller MUST acquire a RCU read side lock before calling it.
3252 int consumer_add_relayd_socket(uint64_t net_seq_idx
, int sock_type
,
3253 struct lttng_consumer_local_data
*ctx
, int sock
,
3254 struct pollfd
*consumer_sockpoll
,
3255 struct lttcomm_relayd_sock
*relayd_sock
, uint64_t sessiond_id
,
3256 uint64_t relayd_session_id
)
3258 int fd
= -1, ret
= -1, relayd_created
= 0;
3259 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3260 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3263 assert(relayd_sock
);
3265 DBG("Consumer adding relayd socket (idx: %" PRIu64
")", net_seq_idx
);
3267 /* Get relayd reference if exists. */
3268 relayd
= consumer_find_relayd(net_seq_idx
);
3269 if (relayd
== NULL
) {
3270 assert(sock_type
== LTTNG_STREAM_CONTROL
);
3271 /* Not found. Allocate one. */
3272 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
3273 if (relayd
== NULL
) {
3275 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3278 relayd
->sessiond_session_id
= sessiond_id
;
3283 * This code path MUST continue to the consumer send status message to
3284 * we can notify the session daemon and continue our work without
3285 * killing everything.
3289 * relayd key should never be found for control socket.
3291 assert(sock_type
!= LTTNG_STREAM_CONTROL
);
3294 /* First send a status message before receiving the fds. */
3295 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
3297 /* Somehow, the session daemon is not responding anymore. */
3298 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3299 goto error_nosignal
;
3302 /* Poll on consumer socket. */
3303 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3305 /* Needing to exit in the middle of a command: error. */
3306 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
3308 goto error_nosignal
;
3311 /* Get relayd socket from session daemon */
3312 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
3313 if (ret
!= sizeof(fd
)) {
3315 fd
= -1; /* Just in case it gets set with an invalid value. */
3318 * Failing to receive FDs might indicate a major problem such as
3319 * reaching a fd limit during the receive where the kernel returns a
3320 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3321 * don't take any chances and stop everything.
3323 * XXX: Feature request #558 will fix that and avoid this possible
3324 * issue when reaching the fd limit.
3326 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
3327 ret_code
= LTTCOMM_CONSUMERD_ERROR_RECV_FD
;
3331 /* Copy socket information and received FD */
3332 switch (sock_type
) {
3333 case LTTNG_STREAM_CONTROL
:
3334 /* Copy received lttcomm socket */
3335 lttcomm_copy_sock(&relayd
->control_sock
.sock
, &relayd_sock
->sock
);
3336 ret
= lttcomm_create_sock(&relayd
->control_sock
.sock
);
3337 /* Handle create_sock error. */
3339 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3343 * Close the socket created internally by
3344 * lttcomm_create_sock, so we can replace it by the one
3345 * received from sessiond.
3347 if (close(relayd
->control_sock
.sock
.fd
)) {
3351 /* Assign new file descriptor */
3352 relayd
->control_sock
.sock
.fd
= fd
;
3353 fd
= -1; /* For error path */
3354 /* Assign version values. */
3355 relayd
->control_sock
.major
= relayd_sock
->major
;
3356 relayd
->control_sock
.minor
= relayd_sock
->minor
;
3358 relayd
->relayd_session_id
= relayd_session_id
;
3361 case LTTNG_STREAM_DATA
:
3362 /* Copy received lttcomm socket */
3363 lttcomm_copy_sock(&relayd
->data_sock
.sock
, &relayd_sock
->sock
);
3364 ret
= lttcomm_create_sock(&relayd
->data_sock
.sock
);
3365 /* Handle create_sock error. */
3367 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3371 * Close the socket created internally by
3372 * lttcomm_create_sock, so we can replace it by the one
3373 * received from sessiond.
3375 if (close(relayd
->data_sock
.sock
.fd
)) {
3379 /* Assign new file descriptor */
3380 relayd
->data_sock
.sock
.fd
= fd
;
3381 fd
= -1; /* for eventual error paths */
3382 /* Assign version values. */
3383 relayd
->data_sock
.major
= relayd_sock
->major
;
3384 relayd
->data_sock
.minor
= relayd_sock
->minor
;
3387 ERR("Unknown relayd socket type (%d)", sock_type
);
3389 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3393 DBG("Consumer %s socket created successfully with net idx %" PRIu64
" (fd: %d)",
3394 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
3395 relayd
->net_seq_idx
, fd
);
3397 /* We successfully added the socket. Send status back. */
3398 ret
= consumer_send_status_msg(sock
, ret_code
);
3400 /* Somehow, the session daemon is not responding anymore. */
3401 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3402 goto error_nosignal
;
3406 * Add relayd socket pair to consumer data hashtable. If object already
3407 * exists or on error, the function gracefully returns.
3415 if (consumer_send_status_msg(sock
, ret_code
) < 0) {
3416 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3420 /* Close received socket if valid. */
3423 PERROR("close received socket");
3427 if (relayd_created
) {
3435 * Try to lock the stream mutex.
3437 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3439 static int stream_try_lock(struct lttng_consumer_stream
*stream
)
3446 * Try to lock the stream mutex. On failure, we know that the stream is
3447 * being used else where hence there is data still being extracted.
3449 ret
= pthread_mutex_trylock(&stream
->lock
);
3451 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3463 * Search for a relayd associated to the session id and return the reference.
3465 * A rcu read side lock MUST be acquire before calling this function and locked
3466 * until the relayd object is no longer necessary.
3468 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
3470 struct lttng_ht_iter iter
;
3471 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3473 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3474 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
3477 * Check by sessiond id which is unique here where the relayd session
3478 * id might not be when having multiple relayd.
3480 if (relayd
->sessiond_session_id
== id
) {
3481 /* Found the relayd. There can be only one per id. */
3493 * Check if for a given session id there is still data needed to be extract
3496 * Return 1 if data is pending or else 0 meaning ready to be read.
3498 int consumer_data_pending(uint64_t id
)
3501 struct lttng_ht_iter iter
;
3502 struct lttng_ht
*ht
;
3503 struct lttng_consumer_stream
*stream
;
3504 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3505 int (*data_pending
)(struct lttng_consumer_stream
*);
3507 DBG("Consumer data pending command on session id %" PRIu64
, id
);
3510 pthread_mutex_lock(&consumer_data
.lock
);
3512 switch (consumer_data
.type
) {
3513 case LTTNG_CONSUMER_KERNEL
:
3514 data_pending
= lttng_kconsumer_data_pending
;
3516 case LTTNG_CONSUMER32_UST
:
3517 case LTTNG_CONSUMER64_UST
:
3518 data_pending
= lttng_ustconsumer_data_pending
;
3521 ERR("Unknown consumer data type");
3525 /* Ease our life a bit */
3526 ht
= consumer_data
.stream_list_ht
;
3528 relayd
= find_relayd_by_session_id(id
);
3530 /* Send init command for data pending. */
3531 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3532 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
3533 relayd
->relayd_session_id
);
3534 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3536 /* Communication error thus the relayd so no data pending. */
3537 goto data_not_pending
;
3541 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3542 ht
->hash_fct(&id
, lttng_ht_seed
),
3544 &iter
.iter
, stream
, node_session_id
.node
) {
3545 /* If this call fails, the stream is being used hence data pending. */
3546 ret
= stream_try_lock(stream
);
3552 * A removed node from the hash table indicates that the stream has
3553 * been deleted thus having a guarantee that the buffers are closed
3554 * on the consumer side. However, data can still be transmitted
3555 * over the network so don't skip the relayd check.
3557 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
3559 /* Check the stream if there is data in the buffers. */
3560 ret
= data_pending(stream
);
3562 pthread_mutex_unlock(&stream
->lock
);
3569 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3570 if (stream
->metadata_flag
) {
3571 ret
= relayd_quiescent_control(&relayd
->control_sock
,
3572 stream
->relayd_stream_id
);
3574 ret
= relayd_data_pending(&relayd
->control_sock
,
3575 stream
->relayd_stream_id
,
3576 stream
->next_net_seq_num
- 1);
3578 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3580 pthread_mutex_unlock(&stream
->lock
);
3584 pthread_mutex_unlock(&stream
->lock
);
3588 unsigned int is_data_inflight
= 0;
3590 /* Send init command for data pending. */
3591 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3592 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3593 relayd
->relayd_session_id
, &is_data_inflight
);
3594 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3596 goto data_not_pending
;
3598 if (is_data_inflight
) {
3604 * Finding _no_ node in the hash table and no inflight data means that the
3605 * stream(s) have been removed thus data is guaranteed to be available for
3606 * analysis from the trace files.
3610 /* Data is available to be read by a viewer. */
3611 pthread_mutex_unlock(&consumer_data
.lock
);
3616 /* Data is still being extracted from buffers. */
3617 pthread_mutex_unlock(&consumer_data
.lock
);
3623 * Send a ret code status message to the sessiond daemon.
3625 * Return the sendmsg() return value.
3627 int consumer_send_status_msg(int sock
, int ret_code
)
3629 struct lttcomm_consumer_status_msg msg
;
3631 memset(&msg
, 0, sizeof(msg
));
3632 msg
.ret_code
= ret_code
;
3634 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3638 * Send a channel status message to the sessiond daemon.
3640 * Return the sendmsg() return value.
3642 int consumer_send_status_channel(int sock
,
3643 struct lttng_consumer_channel
*channel
)
3645 struct lttcomm_consumer_status_channel msg
;
3649 memset(&msg
, 0, sizeof(msg
));
3651 msg
.ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
3653 msg
.ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3654 msg
.key
= channel
->key
;
3655 msg
.stream_count
= channel
->streams
.count
;
3658 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3661 unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos
,
3662 unsigned long produced_pos
, uint64_t nb_packets_per_stream
,
3663 uint64_t max_sb_size
)
3665 unsigned long start_pos
;
3667 if (!nb_packets_per_stream
) {
3668 return consumed_pos
; /* Grab everything */
3670 start_pos
= produced_pos
- offset_align_floor(produced_pos
, max_sb_size
);
3671 start_pos
-= max_sb_size
* nb_packets_per_stream
;
3672 if ((long) (start_pos
- consumed_pos
) < 0) {
3673 return consumed_pos
; /* Grab everything */