2 * Copyright (C) 2011 EfficiOS Inc.
3 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
6 * SPDX-License-Identifier: GPL-2.0-only
18 #include <sys/socket.h>
19 #include <sys/types.h>
22 #include <bin/lttng-consumerd/health-consumerd.hpp>
23 #include <common/align.hpp>
24 #include <common/common.hpp>
25 #include <common/compat/endian.hpp>
26 #include <common/compat/poll.hpp>
27 #include <common/consumer/consumer-metadata-cache.hpp>
28 #include <common/consumer/consumer-stream.hpp>
29 #include <common/consumer/consumer-testpoint.hpp>
30 #include <common/consumer/consumer-timer.hpp>
31 #include <common/consumer/consumer.hpp>
32 #include <common/dynamic-array.hpp>
33 #include <common/index/ctf-index.hpp>
34 #include <common/index/index.hpp>
35 #include <common/kernel-consumer/kernel-consumer.hpp>
36 #include <common/kernel-ctl/kernel-ctl.hpp>
37 #include <common/relayd/relayd.hpp>
38 #include <common/sessiond-comm/relayd.hpp>
39 #include <common/sessiond-comm/sessiond-comm.hpp>
40 #include <common/string-utils/format.hpp>
41 #include <common/time.hpp>
42 #include <common/trace-chunk-registry.hpp>
43 #include <common/trace-chunk.hpp>
44 #include <common/ust-consumer/ust-consumer.hpp>
45 #include <common/utils.hpp>
47 lttng_consumer_global_data the_consumer_data
;
49 enum consumer_channel_action
{
52 CONSUMER_CHANNEL_QUIT
,
55 struct consumer_channel_msg
{
56 enum consumer_channel_action action
;
57 struct lttng_consumer_channel
*chan
; /* add */
58 uint64_t key
; /* del */
61 /* Flag used to temporarily pause data consumption from testpoints. */
62 int data_consumption_paused
;
65 * Flag to inform the polling thread to quit when all fd hung up. Updated by
66 * the consumer_thread_receive_fds when it notices that all fds has hung up.
67 * Also updated by the signal handler (consumer_should_exit()). Read by the
73 * Global hash table containing respectively metadata and data streams. The
74 * stream element in this ht should only be updated by the metadata poll thread
75 * for the metadata and the data poll thread for the data.
77 static struct lttng_ht
*metadata_ht
;
78 static struct lttng_ht
*data_ht
;
80 static const char *get_consumer_domain(void)
82 switch (the_consumer_data
.type
) {
83 case LTTNG_CONSUMER_KERNEL
:
84 return DEFAULT_KERNEL_TRACE_DIR
;
85 case LTTNG_CONSUMER64_UST
:
87 case LTTNG_CONSUMER32_UST
:
88 return DEFAULT_UST_TRACE_DIR
;
95 * Notify a thread lttng pipe to poll back again. This usually means that some
96 * global state has changed so we just send back the thread in a poll wait
99 static void notify_thread_lttng_pipe(struct lttng_pipe
*pipe
)
101 struct lttng_consumer_stream
*null_stream
= NULL
;
105 (void) lttng_pipe_write(pipe
, &null_stream
, sizeof(null_stream
));
108 static void notify_health_quit_pipe(int *pipe
)
112 ret
= lttng_write(pipe
[1], "4", 1);
114 PERROR("write consumer health quit");
118 static void notify_channel_pipe(struct lttng_consumer_local_data
*ctx
,
119 struct lttng_consumer_channel
*chan
,
121 enum consumer_channel_action action
)
123 struct consumer_channel_msg msg
;
126 memset(&msg
, 0, sizeof(msg
));
131 ret
= lttng_write(ctx
->consumer_channel_pipe
[1], &msg
, sizeof(msg
));
132 if (ret
< sizeof(msg
)) {
133 PERROR("notify_channel_pipe write error");
137 void notify_thread_del_channel(struct lttng_consumer_local_data
*ctx
,
140 notify_channel_pipe(ctx
, NULL
, key
, CONSUMER_CHANNEL_DEL
);
143 static int read_channel_pipe(struct lttng_consumer_local_data
*ctx
,
144 struct lttng_consumer_channel
**chan
,
146 enum consumer_channel_action
*action
)
148 struct consumer_channel_msg msg
;
151 ret
= lttng_read(ctx
->consumer_channel_pipe
[0], &msg
, sizeof(msg
));
152 if (ret
< sizeof(msg
)) {
156 *action
= msg
.action
;
164 * Cleanup the stream list of a channel. Those streams are not yet globally
167 static void clean_channel_stream_list(struct lttng_consumer_channel
*channel
)
169 struct lttng_consumer_stream
*stream
, *stmp
;
171 LTTNG_ASSERT(channel
);
173 /* Delete streams that might have been left in the stream list. */
174 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
177 * Once a stream is added to this list, the buffers were created so we
178 * have a guarantee that this call will succeed. Setting the monitor
179 * mode to 0 so we don't lock nor try to delete the stream from the
183 consumer_stream_destroy(stream
, NULL
);
188 * Find a stream. The consumer_data.lock must be locked during this
191 static struct lttng_consumer_stream
*find_stream(uint64_t key
,
194 struct lttng_ht_iter iter
;
195 struct lttng_ht_node_u64
*node
;
196 struct lttng_consumer_stream
*stream
= NULL
;
200 /* -1ULL keys are lookup failures */
201 if (key
== (uint64_t) -1ULL) {
207 lttng_ht_lookup(ht
, &key
, &iter
);
208 node
= lttng_ht_iter_get_node_u64(&iter
);
210 stream
= caa_container_of(node
, struct lttng_consumer_stream
, node
);
218 static void steal_stream_key(uint64_t key
, struct lttng_ht
*ht
)
220 struct lttng_consumer_stream
*stream
;
223 stream
= find_stream(key
, ht
);
225 stream
->key
= (uint64_t) -1ULL;
227 * We don't want the lookup to match, but we still need
228 * to iterate on this stream when iterating over the hash table. Just
229 * change the node key.
231 stream
->node
.key
= (uint64_t) -1ULL;
237 * Return a channel object for the given key.
239 * RCU read side lock MUST be acquired before calling this function and
240 * protects the channel ptr.
242 struct lttng_consumer_channel
*consumer_find_channel(uint64_t key
)
244 struct lttng_ht_iter iter
;
245 struct lttng_ht_node_u64
*node
;
246 struct lttng_consumer_channel
*channel
= NULL
;
248 ASSERT_RCU_READ_LOCKED();
250 /* -1ULL keys are lookup failures */
251 if (key
== (uint64_t) -1ULL) {
255 lttng_ht_lookup(the_consumer_data
.channel_ht
, &key
, &iter
);
256 node
= lttng_ht_iter_get_node_u64(&iter
);
258 channel
= caa_container_of(node
, struct lttng_consumer_channel
, node
);
265 * There is a possibility that the consumer does not have enough time between
266 * the close of the channel on the session daemon and the cleanup in here thus
267 * once we have a channel add with an existing key, we know for sure that this
268 * channel will eventually get cleaned up by all streams being closed.
270 * This function just nullifies the already existing channel key.
272 static void steal_channel_key(uint64_t key
)
274 struct lttng_consumer_channel
*channel
;
277 channel
= consumer_find_channel(key
);
279 channel
->key
= (uint64_t) -1ULL;
281 * We don't want the lookup to match, but we still need to iterate on
282 * this channel when iterating over the hash table. Just change the
285 channel
->node
.key
= (uint64_t) -1ULL;
290 static void free_channel_rcu(struct rcu_head
*head
)
292 struct lttng_ht_node_u64
*node
=
293 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
294 struct lttng_consumer_channel
*channel
=
295 caa_container_of(node
, struct lttng_consumer_channel
, node
);
297 switch (the_consumer_data
.type
) {
298 case LTTNG_CONSUMER_KERNEL
:
300 case LTTNG_CONSUMER32_UST
:
301 case LTTNG_CONSUMER64_UST
:
302 lttng_ustconsumer_free_channel(channel
);
305 ERR("Unknown consumer_data type");
312 * RCU protected relayd socket pair free.
314 static void free_relayd_rcu(struct rcu_head
*head
)
316 struct lttng_ht_node_u64
*node
=
317 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
318 struct consumer_relayd_sock_pair
*relayd
=
319 caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
322 * Close all sockets. This is done in the call RCU since we don't want the
323 * socket fds to be reassigned thus potentially creating bad state of the
326 * We do not have to lock the control socket mutex here since at this stage
327 * there is no one referencing to this relayd object.
329 (void) relayd_close(&relayd
->control_sock
);
330 (void) relayd_close(&relayd
->data_sock
);
332 pthread_mutex_destroy(&relayd
->ctrl_sock_mutex
);
337 * Destroy and free relayd socket pair object.
339 void consumer_destroy_relayd(struct consumer_relayd_sock_pair
*relayd
)
342 struct lttng_ht_iter iter
;
344 if (relayd
== NULL
) {
348 DBG("Consumer destroy and close relayd socket pair");
350 iter
.iter
.node
= &relayd
->node
.node
;
351 ret
= lttng_ht_del(the_consumer_data
.relayd_ht
, &iter
);
353 /* We assume the relayd is being or is destroyed */
357 /* RCU free() call */
358 call_rcu(&relayd
->node
.head
, free_relayd_rcu
);
362 * Remove a channel from the global list protected by a mutex. This function is
363 * also responsible for freeing its data structures.
365 void consumer_del_channel(struct lttng_consumer_channel
*channel
)
367 struct lttng_ht_iter iter
;
369 DBG("Consumer delete channel key %" PRIu64
, channel
->key
);
371 pthread_mutex_lock(&the_consumer_data
.lock
);
372 pthread_mutex_lock(&channel
->lock
);
374 /* Destroy streams that might have been left in the stream list. */
375 clean_channel_stream_list(channel
);
377 if (channel
->live_timer_enabled
== 1) {
378 consumer_timer_live_stop(channel
);
380 if (channel
->monitor_timer_enabled
== 1) {
381 consumer_timer_monitor_stop(channel
);
384 switch (the_consumer_data
.type
) {
385 case LTTNG_CONSUMER_KERNEL
:
387 case LTTNG_CONSUMER32_UST
:
388 case LTTNG_CONSUMER64_UST
:
389 lttng_ustconsumer_del_channel(channel
);
392 ERR("Unknown consumer_data type");
397 lttng_trace_chunk_put(channel
->trace_chunk
);
398 channel
->trace_chunk
= NULL
;
400 if (channel
->is_published
) {
404 iter
.iter
.node
= &channel
->node
.node
;
405 ret
= lttng_ht_del(the_consumer_data
.channel_ht
, &iter
);
408 iter
.iter
.node
= &channel
->channels_by_session_id_ht_node
.node
;
409 ret
= lttng_ht_del(the_consumer_data
.channels_by_session_id_ht
,
415 channel
->is_deleted
= true;
416 call_rcu(&channel
->node
.head
, free_channel_rcu
);
418 pthread_mutex_unlock(&channel
->lock
);
419 pthread_mutex_unlock(&the_consumer_data
.lock
);
423 * Iterate over the relayd hash table and destroy each element. Finally,
424 * destroy the whole hash table.
426 static void cleanup_relayd_ht(void)
428 struct lttng_ht_iter iter
;
429 struct consumer_relayd_sock_pair
*relayd
;
433 cds_lfht_for_each_entry(the_consumer_data
.relayd_ht
->ht
, &iter
.iter
,
435 consumer_destroy_relayd(relayd
);
440 lttng_ht_destroy(the_consumer_data
.relayd_ht
);
444 * Update the end point status of all streams having the given network sequence
445 * index (relayd index).
447 * It's atomically set without having the stream mutex locked which is fine
448 * because we handle the write/read race with a pipe wakeup for each thread.
450 static void update_endpoint_status_by_netidx(uint64_t net_seq_idx
,
451 enum consumer_endpoint_status status
)
453 struct lttng_ht_iter iter
;
454 struct lttng_consumer_stream
*stream
;
456 DBG("Consumer set delete flag on stream by idx %" PRIu64
, net_seq_idx
);
460 /* Let's begin with metadata */
461 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
462 if (stream
->net_seq_idx
== net_seq_idx
) {
463 uatomic_set(&stream
->endpoint_status
, status
);
464 DBG("Delete flag set to metadata stream %d", stream
->wait_fd
);
468 /* Follow up by the data streams */
469 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
470 if (stream
->net_seq_idx
== net_seq_idx
) {
471 uatomic_set(&stream
->endpoint_status
, status
);
472 DBG("Delete flag set to data stream %d", stream
->wait_fd
);
479 * Cleanup a relayd object by flagging every associated streams for deletion,
480 * destroying the object meaning removing it from the relayd hash table,
481 * closing the sockets and freeing the memory in a RCU call.
483 * If a local data context is available, notify the threads that the streams'
484 * state have changed.
486 void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair
*relayd
)
490 LTTNG_ASSERT(relayd
);
492 DBG("Cleaning up relayd object ID %" PRIu64
, relayd
->net_seq_idx
);
494 /* Save the net sequence index before destroying the object */
495 netidx
= relayd
->net_seq_idx
;
498 * Delete the relayd from the relayd hash table, close the sockets and free
499 * the object in a RCU call.
501 consumer_destroy_relayd(relayd
);
503 /* Set inactive endpoint to all streams */
504 update_endpoint_status_by_netidx(netidx
, CONSUMER_ENDPOINT_INACTIVE
);
507 * With a local data context, notify the threads that the streams' state
508 * have changed. The write() action on the pipe acts as an "implicit"
509 * memory barrier ordering the updates of the end point status from the
510 * read of this status which happens AFTER receiving this notify.
512 notify_thread_lttng_pipe(relayd
->ctx
->consumer_data_pipe
);
513 notify_thread_lttng_pipe(relayd
->ctx
->consumer_metadata_pipe
);
517 * Flag a relayd socket pair for destruction. Destroy it if the refcount
520 * RCU read side lock MUST be aquired before calling this function.
522 void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair
*relayd
)
524 LTTNG_ASSERT(relayd
);
525 ASSERT_RCU_READ_LOCKED();
527 /* Set destroy flag for this object */
528 uatomic_set(&relayd
->destroy_flag
, 1);
530 /* Destroy the relayd if refcount is 0 */
531 if (uatomic_read(&relayd
->refcount
) == 0) {
532 consumer_destroy_relayd(relayd
);
537 * Completly destroy stream from every visiable data structure and the given
540 * One this call returns, the stream object is not longer usable nor visible.
542 void consumer_del_stream(struct lttng_consumer_stream
*stream
,
545 consumer_stream_destroy(stream
, ht
);
549 * XXX naming of del vs destroy is all mixed up.
551 void consumer_del_stream_for_data(struct lttng_consumer_stream
*stream
)
553 consumer_stream_destroy(stream
, data_ht
);
556 void consumer_del_stream_for_metadata(struct lttng_consumer_stream
*stream
)
558 consumer_stream_destroy(stream
, metadata_ht
);
561 void consumer_stream_update_channel_attributes(
562 struct lttng_consumer_stream
*stream
,
563 struct lttng_consumer_channel
*channel
)
565 stream
->channel_read_only_attributes
.tracefile_size
=
566 channel
->tracefile_size
;
570 * Add a stream to the global list protected by a mutex.
572 void consumer_add_data_stream(struct lttng_consumer_stream
*stream
)
574 struct lttng_ht
*ht
= data_ht
;
576 LTTNG_ASSERT(stream
);
579 DBG3("Adding consumer stream %" PRIu64
, stream
->key
);
581 pthread_mutex_lock(&the_consumer_data
.lock
);
582 pthread_mutex_lock(&stream
->chan
->lock
);
583 pthread_mutex_lock(&stream
->chan
->timer_lock
);
584 pthread_mutex_lock(&stream
->lock
);
587 /* Steal stream identifier to avoid having streams with the same key */
588 steal_stream_key(stream
->key
, ht
);
590 lttng_ht_add_unique_u64(ht
, &stream
->node
);
592 lttng_ht_add_u64(the_consumer_data
.stream_per_chan_id_ht
,
593 &stream
->node_channel_id
);
596 * Add stream to the stream_list_ht of the consumer data. No need to steal
597 * the key since the HT does not use it and we allow to add redundant keys
600 lttng_ht_add_u64(the_consumer_data
.stream_list_ht
,
601 &stream
->node_session_id
);
604 * When nb_init_stream_left reaches 0, we don't need to trigger any action
605 * in terms of destroying the associated channel, because the action that
606 * causes the count to become 0 also causes a stream to be added. The
607 * channel deletion will thus be triggered by the following removal of this
610 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
611 /* Increment refcount before decrementing nb_init_stream_left */
613 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
616 /* Update consumer data once the node is inserted. */
617 the_consumer_data
.stream_count
++;
618 the_consumer_data
.need_update
= 1;
621 pthread_mutex_unlock(&stream
->lock
);
622 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
623 pthread_mutex_unlock(&stream
->chan
->lock
);
624 pthread_mutex_unlock(&the_consumer_data
.lock
);
628 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
629 * be acquired before calling this.
631 static int add_relayd(struct consumer_relayd_sock_pair
*relayd
)
634 struct lttng_ht_node_u64
*node
;
635 struct lttng_ht_iter iter
;
637 LTTNG_ASSERT(relayd
);
638 ASSERT_RCU_READ_LOCKED();
640 lttng_ht_lookup(the_consumer_data
.relayd_ht
, &relayd
->net_seq_idx
,
642 node
= lttng_ht_iter_get_node_u64(&iter
);
646 lttng_ht_add_unique_u64(the_consumer_data
.relayd_ht
, &relayd
->node
);
653 * Allocate and return a consumer relayd socket.
655 static struct consumer_relayd_sock_pair
*consumer_allocate_relayd_sock_pair(
656 uint64_t net_seq_idx
)
658 struct consumer_relayd_sock_pair
*obj
= NULL
;
660 /* net sequence index of -1 is a failure */
661 if (net_seq_idx
== (uint64_t) -1ULL) {
665 obj
= zmalloc
<consumer_relayd_sock_pair
>();
667 PERROR("zmalloc relayd sock");
671 obj
->net_seq_idx
= net_seq_idx
;
673 obj
->destroy_flag
= 0;
674 obj
->control_sock
.sock
.fd
= -1;
675 obj
->data_sock
.sock
.fd
= -1;
676 lttng_ht_node_init_u64(&obj
->node
, obj
->net_seq_idx
);
677 pthread_mutex_init(&obj
->ctrl_sock_mutex
, NULL
);
684 * Find a relayd socket pair in the global consumer data.
686 * Return the object if found else NULL.
687 * RCU read-side lock must be held across this call and while using the
690 struct consumer_relayd_sock_pair
*consumer_find_relayd(uint64_t key
)
692 struct lttng_ht_iter iter
;
693 struct lttng_ht_node_u64
*node
;
694 struct consumer_relayd_sock_pair
*relayd
= NULL
;
696 ASSERT_RCU_READ_LOCKED();
698 /* Negative keys are lookup failures */
699 if (key
== (uint64_t) -1ULL) {
703 lttng_ht_lookup(the_consumer_data
.relayd_ht
, &key
, &iter
);
704 node
= lttng_ht_iter_get_node_u64(&iter
);
706 relayd
= caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
714 * Find a relayd and send the stream
716 * Returns 0 on success, < 0 on error
718 int consumer_send_relayd_stream(struct lttng_consumer_stream
*stream
,
722 struct consumer_relayd_sock_pair
*relayd
;
724 LTTNG_ASSERT(stream
);
725 LTTNG_ASSERT(stream
->net_seq_idx
!= -1ULL);
728 /* The stream is not metadata. Get relayd reference if exists. */
730 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
731 if (relayd
!= NULL
) {
732 /* Add stream on the relayd */
733 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
734 ret
= relayd_add_stream(&relayd
->control_sock
, stream
->name
,
735 get_consumer_domain(), path
, &stream
->relayd_stream_id
,
736 stream
->chan
->tracefile_size
,
737 stream
->chan
->tracefile_count
,
738 stream
->trace_chunk
);
739 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
741 ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
742 lttng_consumer_cleanup_relayd(relayd
);
746 uatomic_inc(&relayd
->refcount
);
747 stream
->sent_to_relayd
= 1;
749 ERR("Stream %" PRIu64
" relayd ID %" PRIu64
" unknown. Can't send it.",
750 stream
->key
, stream
->net_seq_idx
);
755 DBG("Stream %s with key %" PRIu64
" sent to relayd id %" PRIu64
,
756 stream
->name
, stream
->key
, stream
->net_seq_idx
);
764 * Find a relayd and send the streams sent message
766 * Returns 0 on success, < 0 on error
768 int consumer_send_relayd_streams_sent(uint64_t net_seq_idx
)
771 struct consumer_relayd_sock_pair
*relayd
;
773 LTTNG_ASSERT(net_seq_idx
!= -1ULL);
775 /* The stream is not metadata. Get relayd reference if exists. */
777 relayd
= consumer_find_relayd(net_seq_idx
);
778 if (relayd
!= NULL
) {
779 /* Add stream on the relayd */
780 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
781 ret
= relayd_streams_sent(&relayd
->control_sock
);
782 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
784 ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
785 lttng_consumer_cleanup_relayd(relayd
);
789 ERR("Relayd ID %" PRIu64
" unknown. Can't send streams_sent.",
796 DBG("All streams sent relayd id %" PRIu64
, net_seq_idx
);
804 * Find a relayd and close the stream
806 void close_relayd_stream(struct lttng_consumer_stream
*stream
)
808 struct consumer_relayd_sock_pair
*relayd
;
810 /* The stream is not metadata. Get relayd reference if exists. */
812 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
814 consumer_stream_relayd_close(stream
, relayd
);
820 * Handle stream for relayd transmission if the stream applies for network
821 * streaming where the net sequence index is set.
823 * Return destination file descriptor or negative value on error.
825 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
826 size_t data_size
, unsigned long padding
,
827 struct consumer_relayd_sock_pair
*relayd
)
830 struct lttcomm_relayd_data_hdr data_hdr
;
833 LTTNG_ASSERT(stream
);
834 LTTNG_ASSERT(relayd
);
836 /* Reset data header */
837 memset(&data_hdr
, 0, sizeof(data_hdr
));
839 if (stream
->metadata_flag
) {
840 /* Caller MUST acquire the relayd control socket lock */
841 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
846 /* Metadata are always sent on the control socket. */
847 outfd
= relayd
->control_sock
.sock
.fd
;
849 /* Set header with stream information */
850 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
851 data_hdr
.data_size
= htobe32(data_size
);
852 data_hdr
.padding_size
= htobe32(padding
);
855 * Note that net_seq_num below is assigned with the *current* value of
856 * next_net_seq_num and only after that the next_net_seq_num will be
857 * increment. This is why when issuing a command on the relayd using
858 * this next value, 1 should always be substracted in order to compare
859 * the last seen sequence number on the relayd side to the last sent.
861 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
862 /* Other fields are zeroed previously */
864 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
870 ++stream
->next_net_seq_num
;
872 /* Set to go on data socket */
873 outfd
= relayd
->data_sock
.sock
.fd
;
881 * Write a character on the metadata poll pipe to wake the metadata thread.
882 * Returns 0 on success, -1 on error.
884 int consumer_metadata_wakeup_pipe(const struct lttng_consumer_channel
*channel
)
888 DBG("Waking up metadata poll thread (writing to pipe): channel name = '%s'",
890 if (channel
->monitor
&& channel
->metadata_stream
) {
891 const char dummy
= 'c';
892 const ssize_t write_ret
= lttng_write(
893 channel
->metadata_stream
->ust_metadata_poll_pipe
[1],
897 if (errno
== EWOULDBLOCK
) {
899 * This is fine, the metadata poll thread
900 * is having a hard time keeping-up, but
901 * it will eventually wake-up and consume
902 * the available data.
906 PERROR("Failed to write to UST metadata pipe while attempting to wake-up the metadata poll thread");
918 * Trigger a dump of the metadata content. Following/during the succesful
919 * completion of this call, the metadata poll thread will start receiving
920 * metadata packets to consume.
922 * The caller must hold the channel and stream locks.
925 int consumer_metadata_stream_dump(struct lttng_consumer_stream
*stream
)
929 ASSERT_LOCKED(stream
->chan
->lock
);
930 ASSERT_LOCKED(stream
->lock
);
931 LTTNG_ASSERT(stream
->metadata_flag
);
932 LTTNG_ASSERT(stream
->chan
->trace_chunk
);
934 switch (the_consumer_data
.type
) {
935 case LTTNG_CONSUMER_KERNEL
:
937 * Reset the position of what has been read from the
938 * metadata cache to 0 so we can dump it again.
940 ret
= kernctl_metadata_cache_dump(stream
->wait_fd
);
942 case LTTNG_CONSUMER32_UST
:
943 case LTTNG_CONSUMER64_UST
:
945 * Reset the position pushed from the metadata cache so it
946 * will write from the beginning on the next push.
948 stream
->ust_metadata_pushed
= 0;
949 ret
= consumer_metadata_wakeup_pipe(stream
->chan
);
952 ERR("Unknown consumer_data type");
956 ERR("Failed to dump the metadata cache");
962 int lttng_consumer_channel_set_trace_chunk(
963 struct lttng_consumer_channel
*channel
,
964 struct lttng_trace_chunk
*new_trace_chunk
)
966 pthread_mutex_lock(&channel
->lock
);
967 if (channel
->is_deleted
) {
969 * The channel has been logically deleted and should no longer
970 * be used. It has released its reference to its current trace
971 * chunk and should not acquire a new one.
973 * Return success as there is nothing for the caller to do.
979 * The acquisition of the reference cannot fail (barring
980 * a severe internal error) since a reference to the published
981 * chunk is already held by the caller.
983 if (new_trace_chunk
) {
984 const bool acquired_reference
= lttng_trace_chunk_get(
987 LTTNG_ASSERT(acquired_reference
);
990 lttng_trace_chunk_put(channel
->trace_chunk
);
991 channel
->trace_chunk
= new_trace_chunk
;
993 pthread_mutex_unlock(&channel
->lock
);
998 * Allocate and return a new lttng_consumer_channel object using the given key
999 * to initialize the hash table node.
1001 * On error, return NULL.
1003 struct lttng_consumer_channel
*consumer_allocate_channel(uint64_t key
,
1004 uint64_t session_id
,
1005 const uint64_t *chunk_id
,
1006 const char *pathname
,
1009 enum lttng_event_output output
,
1010 uint64_t tracefile_size
,
1011 uint64_t tracefile_count
,
1012 uint64_t session_id_per_pid
,
1013 unsigned int monitor
,
1014 unsigned int live_timer_interval
,
1015 bool is_in_live_session
,
1016 const char *root_shm_path
,
1017 const char *shm_path
)
1019 struct lttng_consumer_channel
*channel
= NULL
;
1020 struct lttng_trace_chunk
*trace_chunk
= NULL
;
1023 trace_chunk
= lttng_trace_chunk_registry_find_chunk(
1024 the_consumer_data
.chunk_registry
, session_id
,
1027 ERR("Failed to find trace chunk reference during creation of channel");
1032 channel
= zmalloc
<lttng_consumer_channel
>();
1033 if (channel
== NULL
) {
1034 PERROR("malloc struct lttng_consumer_channel");
1039 channel
->refcount
= 0;
1040 channel
->session_id
= session_id
;
1041 channel
->session_id_per_pid
= session_id_per_pid
;
1042 channel
->relayd_id
= relayd_id
;
1043 channel
->tracefile_size
= tracefile_size
;
1044 channel
->tracefile_count
= tracefile_count
;
1045 channel
->monitor
= monitor
;
1046 channel
->live_timer_interval
= live_timer_interval
;
1047 channel
->is_live
= is_in_live_session
;
1048 pthread_mutex_init(&channel
->lock
, NULL
);
1049 pthread_mutex_init(&channel
->timer_lock
, NULL
);
1052 case LTTNG_EVENT_SPLICE
:
1053 channel
->output
= CONSUMER_CHANNEL_SPLICE
;
1055 case LTTNG_EVENT_MMAP
:
1056 channel
->output
= CONSUMER_CHANNEL_MMAP
;
1066 * In monitor mode, the streams associated with the channel will be put in
1067 * a special list ONLY owned by this channel. So, the refcount is set to 1
1068 * here meaning that the channel itself has streams that are referenced.
1070 * On a channel deletion, once the channel is no longer visible, the
1071 * refcount is decremented and checked for a zero value to delete it. With
1072 * streams in no monitor mode, it will now be safe to destroy the channel.
1074 if (!channel
->monitor
) {
1075 channel
->refcount
= 1;
1078 strncpy(channel
->pathname
, pathname
, sizeof(channel
->pathname
));
1079 channel
->pathname
[sizeof(channel
->pathname
) - 1] = '\0';
1081 strncpy(channel
->name
, name
, sizeof(channel
->name
));
1082 channel
->name
[sizeof(channel
->name
) - 1] = '\0';
1084 if (root_shm_path
) {
1085 strncpy(channel
->root_shm_path
, root_shm_path
, sizeof(channel
->root_shm_path
));
1086 channel
->root_shm_path
[sizeof(channel
->root_shm_path
) - 1] = '\0';
1089 strncpy(channel
->shm_path
, shm_path
, sizeof(channel
->shm_path
));
1090 channel
->shm_path
[sizeof(channel
->shm_path
) - 1] = '\0';
1093 lttng_ht_node_init_u64(&channel
->node
, channel
->key
);
1094 lttng_ht_node_init_u64(&channel
->channels_by_session_id_ht_node
,
1095 channel
->session_id
);
1097 channel
->wait_fd
= -1;
1098 CDS_INIT_LIST_HEAD(&channel
->streams
.head
);
1101 int ret
= lttng_consumer_channel_set_trace_chunk(channel
,
1108 DBG("Allocated channel (key %" PRIu64
")", channel
->key
);
1111 lttng_trace_chunk_put(trace_chunk
);
1114 consumer_del_channel(channel
);
1120 * Add a channel to the global list protected by a mutex.
1122 * Always return 0 indicating success.
1124 int consumer_add_channel(struct lttng_consumer_channel
*channel
,
1125 struct lttng_consumer_local_data
*ctx
)
1127 pthread_mutex_lock(&the_consumer_data
.lock
);
1128 pthread_mutex_lock(&channel
->lock
);
1129 pthread_mutex_lock(&channel
->timer_lock
);
1132 * This gives us a guarantee that the channel we are about to add to the
1133 * channel hash table will be unique. See this function comment on the why
1134 * we need to steel the channel key at this stage.
1136 steal_channel_key(channel
->key
);
1139 lttng_ht_add_unique_u64(the_consumer_data
.channel_ht
, &channel
->node
);
1140 lttng_ht_add_u64(the_consumer_data
.channels_by_session_id_ht
,
1141 &channel
->channels_by_session_id_ht_node
);
1143 channel
->is_published
= true;
1145 pthread_mutex_unlock(&channel
->timer_lock
);
1146 pthread_mutex_unlock(&channel
->lock
);
1147 pthread_mutex_unlock(&the_consumer_data
.lock
);
1149 if (channel
->wait_fd
!= -1 && channel
->type
== CONSUMER_CHANNEL_TYPE_DATA
) {
1150 notify_channel_pipe(ctx
, channel
, -1, CONSUMER_CHANNEL_ADD
);
1157 * Allocate the pollfd structure and the local view of the out fds to avoid
1158 * doing a lookup in the linked list and concurrency issues when writing is
1159 * needed. Called with consumer_data.lock held.
1161 * Returns the number of fds in the structures.
1163 static int update_poll_array(struct lttng_consumer_local_data
*ctx
,
1164 struct pollfd
**pollfd
, struct lttng_consumer_stream
**local_stream
,
1165 struct lttng_ht
*ht
, int *nb_inactive_fd
)
1168 struct lttng_ht_iter iter
;
1169 struct lttng_consumer_stream
*stream
;
1173 LTTNG_ASSERT(pollfd
);
1174 LTTNG_ASSERT(local_stream
);
1176 DBG("Updating poll fd array");
1177 *nb_inactive_fd
= 0;
1179 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1181 * Only active streams with an active end point can be added to the
1182 * poll set and local stream storage of the thread.
1184 * There is a potential race here for endpoint_status to be updated
1185 * just after the check. However, this is OK since the stream(s) will
1186 * be deleted once the thread is notified that the end point state has
1187 * changed where this function will be called back again.
1189 * We track the number of inactive FDs because they still need to be
1190 * closed by the polling thread after a wakeup on the data_pipe or
1193 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
1194 (*nb_inactive_fd
)++;
1198 * This clobbers way too much the debug output. Uncomment that if you
1199 * need it for debugging purposes.
1201 (*pollfd
)[i
].fd
= stream
->wait_fd
;
1202 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1203 local_stream
[i
] = stream
;
1209 * Insert the consumer_data_pipe at the end of the array and don't
1210 * increment i so nb_fd is the number of real FD.
1212 (*pollfd
)[i
].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_pipe
);
1213 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1215 (*pollfd
)[i
+ 1].fd
= lttng_pipe_get_readfd(ctx
->consumer_wakeup_pipe
);
1216 (*pollfd
)[i
+ 1].events
= POLLIN
| POLLPRI
;
1221 * Poll on the should_quit pipe and the command socket return -1 on
1222 * error, 1 if should exit, 0 if data is available on the command socket
1224 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
1229 num_rdy
= poll(consumer_sockpoll
, 2, -1);
1230 if (num_rdy
== -1) {
1232 * Restart interrupted system call.
1234 if (errno
== EINTR
) {
1237 PERROR("Poll error");
1240 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
1241 DBG("consumer_should_quit wake up");
1248 * Set the error socket.
1250 void lttng_consumer_set_error_sock(struct lttng_consumer_local_data
*ctx
,
1253 ctx
->consumer_error_socket
= sock
;
1257 * Set the command socket path.
1259 void lttng_consumer_set_command_sock_path(
1260 struct lttng_consumer_local_data
*ctx
, char *sock
)
1262 ctx
->consumer_command_sock_path
= sock
;
1266 * Send return code to the session daemon.
1267 * If the socket is not defined, we return 0, it is not a fatal error
1269 int lttng_consumer_send_error(struct lttng_consumer_local_data
*ctx
, int cmd
)
1271 if (ctx
->consumer_error_socket
> 0) {
1272 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1273 sizeof(enum lttcomm_sessiond_command
));
1280 * Close all the tracefiles and stream fds and MUST be called when all
1281 * instances are destroyed i.e. when all threads were joined and are ended.
1283 void lttng_consumer_cleanup(void)
1285 struct lttng_ht_iter iter
;
1286 struct lttng_consumer_channel
*channel
;
1287 unsigned int trace_chunks_left
;
1291 cds_lfht_for_each_entry(the_consumer_data
.channel_ht
->ht
, &iter
.iter
,
1292 channel
, node
.node
) {
1293 consumer_del_channel(channel
);
1298 lttng_ht_destroy(the_consumer_data
.channel_ht
);
1299 lttng_ht_destroy(the_consumer_data
.channels_by_session_id_ht
);
1301 cleanup_relayd_ht();
1303 lttng_ht_destroy(the_consumer_data
.stream_per_chan_id_ht
);
1306 * This HT contains streams that are freed by either the metadata thread or
1307 * the data thread so we do *nothing* on the hash table and simply destroy
1310 lttng_ht_destroy(the_consumer_data
.stream_list_ht
);
1313 * Trace chunks in the registry may still exist if the session
1314 * daemon has encountered an internal error and could not
1315 * tear down its sessions and/or trace chunks properly.
1317 * Release the session daemon's implicit reference to any remaining
1318 * trace chunk and print an error if any trace chunk was found. Note
1319 * that there are _no_ legitimate cases for trace chunks to be left,
1320 * it is a leak. However, it can happen following a crash of the
1321 * session daemon and not emptying the registry would cause an assertion
1324 trace_chunks_left
= lttng_trace_chunk_registry_put_each_chunk(
1325 the_consumer_data
.chunk_registry
);
1326 if (trace_chunks_left
) {
1327 ERR("%u trace chunks are leaked by lttng-consumerd. "
1328 "This can be caused by an internal error of the session daemon.",
1331 /* Run all callbacks freeing each chunk. */
1333 lttng_trace_chunk_registry_destroy(the_consumer_data
.chunk_registry
);
1337 * Called from signal handler.
1339 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1343 CMM_STORE_SHARED(consumer_quit
, 1);
1344 ret
= lttng_write(ctx
->consumer_should_quit
[1], "4", 1);
1346 PERROR("write consumer quit");
1349 DBG("Consumer flag that it should quit");
1354 * Flush pending writes to trace output disk file.
1357 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1361 int outfd
= stream
->out_fd
;
1364 * This does a blocking write-and-wait on any page that belongs to the
1365 * subbuffer prior to the one we just wrote.
1366 * Don't care about error values, as these are just hints and ways to
1367 * limit the amount of page cache used.
1369 if (orig_offset
< stream
->max_sb_size
) {
1372 lttng_sync_file_range(outfd
, orig_offset
- stream
->max_sb_size
,
1373 stream
->max_sb_size
,
1374 SYNC_FILE_RANGE_WAIT_BEFORE
1375 | SYNC_FILE_RANGE_WRITE
1376 | SYNC_FILE_RANGE_WAIT_AFTER
);
1378 * Give hints to the kernel about how we access the file:
1379 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1382 * We need to call fadvise again after the file grows because the
1383 * kernel does not seem to apply fadvise to non-existing parts of the
1386 * Call fadvise _after_ having waited for the page writeback to
1387 * complete because the dirty page writeback semantic is not well
1388 * defined. So it can be expected to lead to lower throughput in
1391 ret
= posix_fadvise(outfd
, orig_offset
- stream
->max_sb_size
,
1392 stream
->max_sb_size
, POSIX_FADV_DONTNEED
);
1393 if (ret
&& ret
!= -ENOSYS
) {
1395 PERROR("posix_fadvise on fd %i", outfd
);
1400 * Initialise the necessary environnement :
1401 * - create a new context
1402 * - create the poll_pipe
1403 * - create the should_quit pipe (for signal handler)
1404 * - create the thread pipe (for splice)
1406 * Takes a function pointer as argument, this function is called when data is
1407 * available on a buffer. This function is responsible to do the
1408 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1409 * buffer configuration and then kernctl_put_next_subbuf at the end.
1411 * Returns a pointer to the new context or NULL on error.
1413 struct lttng_consumer_local_data
*lttng_consumer_create(
1414 enum lttng_consumer_type type
,
1415 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1416 struct lttng_consumer_local_data
*ctx
, bool locked_by_caller
),
1417 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1418 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1419 int (*update_stream
)(uint64_t stream_key
, uint32_t state
))
1422 struct lttng_consumer_local_data
*ctx
;
1424 LTTNG_ASSERT(the_consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1425 the_consumer_data
.type
== type
);
1426 the_consumer_data
.type
= type
;
1428 ctx
= zmalloc
<lttng_consumer_local_data
>();
1430 PERROR("allocating context");
1434 ctx
->consumer_error_socket
= -1;
1435 ctx
->consumer_metadata_socket
= -1;
1436 pthread_mutex_init(&ctx
->metadata_socket_lock
, NULL
);
1437 /* assign the callbacks */
1438 ctx
->on_buffer_ready
= buffer_ready
;
1439 ctx
->on_recv_channel
= recv_channel
;
1440 ctx
->on_recv_stream
= recv_stream
;
1441 ctx
->on_update_stream
= update_stream
;
1443 ctx
->consumer_data_pipe
= lttng_pipe_open(0);
1444 if (!ctx
->consumer_data_pipe
) {
1445 goto error_poll_pipe
;
1448 ctx
->consumer_wakeup_pipe
= lttng_pipe_open(0);
1449 if (!ctx
->consumer_wakeup_pipe
) {
1450 goto error_wakeup_pipe
;
1453 ret
= pipe(ctx
->consumer_should_quit
);
1455 PERROR("Error creating recv pipe");
1456 goto error_quit_pipe
;
1459 ret
= pipe(ctx
->consumer_channel_pipe
);
1461 PERROR("Error creating channel pipe");
1462 goto error_channel_pipe
;
1465 ctx
->consumer_metadata_pipe
= lttng_pipe_open(0);
1466 if (!ctx
->consumer_metadata_pipe
) {
1467 goto error_metadata_pipe
;
1470 ctx
->channel_monitor_pipe
= -1;
1474 error_metadata_pipe
:
1475 utils_close_pipe(ctx
->consumer_channel_pipe
);
1477 utils_close_pipe(ctx
->consumer_should_quit
);
1479 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1481 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1489 * Iterate over all streams of the hashtable and free them properly.
1491 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1493 struct lttng_ht_iter iter
;
1494 struct lttng_consumer_stream
*stream
;
1501 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1503 * Ignore return value since we are currently cleaning up so any error
1506 (void) consumer_del_stream(stream
, ht
);
1510 lttng_ht_destroy(ht
);
1514 * Iterate over all streams of the metadata hashtable and free them
1517 static void destroy_metadata_stream_ht(struct lttng_ht
*ht
)
1519 struct lttng_ht_iter iter
;
1520 struct lttng_consumer_stream
*stream
;
1527 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1529 * Ignore return value since we are currently cleaning up so any error
1532 (void) consumer_del_metadata_stream(stream
, ht
);
1536 lttng_ht_destroy(ht
);
1540 * Close all fds associated with the instance and free the context.
1542 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1546 DBG("Consumer destroying it. Closing everything.");
1552 destroy_data_stream_ht(data_ht
);
1553 destroy_metadata_stream_ht(metadata_ht
);
1555 ret
= close(ctx
->consumer_error_socket
);
1559 ret
= close(ctx
->consumer_metadata_socket
);
1563 utils_close_pipe(ctx
->consumer_channel_pipe
);
1564 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1565 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1566 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1567 utils_close_pipe(ctx
->consumer_should_quit
);
1569 unlink(ctx
->consumer_command_sock_path
);
1574 * Write the metadata stream id on the specified file descriptor.
1576 static int write_relayd_metadata_id(int fd
,
1577 struct lttng_consumer_stream
*stream
,
1578 unsigned long padding
)
1581 struct lttcomm_relayd_metadata_payload hdr
;
1583 hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
1584 hdr
.padding_size
= htobe32(padding
);
1585 ret
= lttng_write(fd
, (void *) &hdr
, sizeof(hdr
));
1586 if (ret
< sizeof(hdr
)) {
1588 * This error means that the fd's end is closed so ignore the PERROR
1589 * not to clubber the error output since this can happen in a normal
1592 if (errno
!= EPIPE
) {
1593 PERROR("write metadata stream id");
1595 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno
);
1597 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1598 * handle writting the missing part so report that as an error and
1599 * don't lie to the caller.
1604 DBG("Metadata stream id %" PRIu64
" with padding %lu written before data",
1605 stream
->relayd_stream_id
, padding
);
1612 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1613 * core function for writing trace buffers to either the local filesystem or
1616 * It must be called with the stream and the channel lock held.
1618 * Careful review MUST be put if any changes occur!
1620 * Returns the number of bytes written
1622 ssize_t
lttng_consumer_on_read_subbuffer_mmap(
1623 struct lttng_consumer_stream
*stream
,
1624 const struct lttng_buffer_view
*buffer
,
1625 unsigned long padding
)
1628 off_t orig_offset
= stream
->out_fd_offset
;
1629 /* Default is on the disk */
1630 int outfd
= stream
->out_fd
;
1631 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1632 unsigned int relayd_hang_up
= 0;
1633 const size_t subbuf_content_size
= buffer
->size
- padding
;
1636 /* RCU lock for the relayd pointer */
1638 LTTNG_ASSERT(stream
->net_seq_idx
!= (uint64_t) -1ULL ||
1639 stream
->trace_chunk
);
1641 /* Flag that the current stream if set for network streaming. */
1642 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1643 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1644 if (relayd
== NULL
) {
1650 /* Handle stream on the relayd if the output is on the network */
1652 unsigned long netlen
= subbuf_content_size
;
1655 * Lock the control socket for the complete duration of the function
1656 * since from this point on we will use the socket.
1658 if (stream
->metadata_flag
) {
1659 /* Metadata requires the control socket. */
1660 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1661 if (stream
->reset_metadata_flag
) {
1662 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1663 stream
->relayd_stream_id
,
1664 stream
->metadata_version
);
1669 stream
->reset_metadata_flag
= 0;
1671 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1674 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1679 /* Use the returned socket. */
1682 /* Write metadata stream id before payload */
1683 if (stream
->metadata_flag
) {
1684 ret
= write_relayd_metadata_id(outfd
, stream
, padding
);
1691 write_len
= subbuf_content_size
;
1693 /* No streaming; we have to write the full padding. */
1694 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1695 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1697 ERR("Reset metadata file");
1700 stream
->reset_metadata_flag
= 0;
1704 * Check if we need to change the tracefile before writing the packet.
1706 if (stream
->chan
->tracefile_size
> 0 &&
1707 (stream
->tracefile_size_current
+ buffer
->size
) >
1708 stream
->chan
->tracefile_size
) {
1709 ret
= consumer_stream_rotate_output_files(stream
);
1713 outfd
= stream
->out_fd
;
1716 stream
->tracefile_size_current
+= buffer
->size
;
1717 write_len
= buffer
->size
;
1721 * This call guarantee that len or less is returned. It's impossible to
1722 * receive a ret value that is bigger than len.
1724 ret
= lttng_write(outfd
, buffer
->data
, write_len
);
1725 DBG("Consumer mmap write() ret %zd (len %zu)", ret
, write_len
);
1726 if (ret
< 0 || ((size_t) ret
!= write_len
)) {
1728 * Report error to caller if nothing was written else at least send the
1736 /* Socket operation failed. We consider the relayd dead */
1737 if (errno
== EPIPE
) {
1739 * This is possible if the fd is closed on the other side
1740 * (outfd) or any write problem. It can be verbose a bit for a
1741 * normal execution if for instance the relayd is stopped
1742 * abruptly. This can happen so set this to a DBG statement.
1744 DBG("Consumer mmap write detected relayd hang up");
1746 /* Unhandled error, print it and stop function right now. */
1747 PERROR("Error in write mmap (ret %zd != write_len %zu)", ret
,
1752 stream
->output_written
+= ret
;
1754 /* This call is useless on a socket so better save a syscall. */
1756 /* This won't block, but will start writeout asynchronously */
1757 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, write_len
,
1758 SYNC_FILE_RANGE_WRITE
);
1759 stream
->out_fd_offset
+= write_len
;
1760 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1765 * This is a special case that the relayd has closed its socket. Let's
1766 * cleanup the relayd object and all associated streams.
1768 if (relayd
&& relayd_hang_up
) {
1769 ERR("Relayd hangup. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
1770 lttng_consumer_cleanup_relayd(relayd
);
1774 /* Unlock only if ctrl socket used */
1775 if (relayd
&& stream
->metadata_flag
) {
1776 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1784 * Splice the data from the ring buffer to the tracefile.
1786 * It must be called with the stream lock held.
1788 * Returns the number of bytes spliced.
1790 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1791 struct lttng_consumer_local_data
*ctx
,
1792 struct lttng_consumer_stream
*stream
, unsigned long len
,
1793 unsigned long padding
)
1795 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1797 off_t orig_offset
= stream
->out_fd_offset
;
1798 int fd
= stream
->wait_fd
;
1799 /* Default is on the disk */
1800 int outfd
= stream
->out_fd
;
1801 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1803 unsigned int relayd_hang_up
= 0;
1805 switch (the_consumer_data
.type
) {
1806 case LTTNG_CONSUMER_KERNEL
:
1808 case LTTNG_CONSUMER32_UST
:
1809 case LTTNG_CONSUMER64_UST
:
1810 /* Not supported for user space tracing */
1813 ERR("Unknown consumer_data type");
1817 /* RCU lock for the relayd pointer */
1820 /* Flag that the current stream if set for network streaming. */
1821 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1822 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1823 if (relayd
== NULL
) {
1828 splice_pipe
= stream
->splice_pipe
;
1830 /* Write metadata stream id before payload */
1832 unsigned long total_len
= len
;
1834 if (stream
->metadata_flag
) {
1836 * Lock the control socket for the complete duration of the function
1837 * since from this point on we will use the socket.
1839 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1841 if (stream
->reset_metadata_flag
) {
1842 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1843 stream
->relayd_stream_id
,
1844 stream
->metadata_version
);
1849 stream
->reset_metadata_flag
= 0;
1851 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
,
1859 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1862 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1868 /* Use the returned socket. */
1871 /* No streaming, we have to set the len with the full padding */
1874 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1875 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1877 ERR("Reset metadata file");
1880 stream
->reset_metadata_flag
= 0;
1883 * Check if we need to change the tracefile before writing the packet.
1885 if (stream
->chan
->tracefile_size
> 0 &&
1886 (stream
->tracefile_size_current
+ len
) >
1887 stream
->chan
->tracefile_size
) {
1888 ret
= consumer_stream_rotate_output_files(stream
);
1893 outfd
= stream
->out_fd
;
1896 stream
->tracefile_size_current
+= len
;
1900 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1901 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1902 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1903 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1904 DBG("splice chan to pipe, ret %zd", ret_splice
);
1905 if (ret_splice
< 0) {
1908 PERROR("Error in relay splice");
1912 /* Handle stream on the relayd if the output is on the network */
1913 if (relayd
&& stream
->metadata_flag
) {
1914 size_t metadata_payload_size
=
1915 sizeof(struct lttcomm_relayd_metadata_payload
);
1917 /* Update counter to fit the spliced data */
1918 ret_splice
+= metadata_payload_size
;
1919 len
+= metadata_payload_size
;
1921 * We do this so the return value can match the len passed as
1922 * argument to this function.
1924 written
-= metadata_payload_size
;
1927 /* Splice data out */
1928 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1929 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1930 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1932 if (ret_splice
< 0) {
1937 } else if (ret_splice
> len
) {
1939 * We don't expect this code path to be executed but you never know
1940 * so this is an extra protection agains a buggy splice().
1943 written
+= ret_splice
;
1944 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice
,
1948 /* All good, update current len and continue. */
1952 /* This call is useless on a socket so better save a syscall. */
1954 /* This won't block, but will start writeout asynchronously */
1955 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1956 SYNC_FILE_RANGE_WRITE
);
1957 stream
->out_fd_offset
+= ret_splice
;
1959 stream
->output_written
+= ret_splice
;
1960 written
+= ret_splice
;
1963 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1969 * This is a special case that the relayd has closed its socket. Let's
1970 * cleanup the relayd object and all associated streams.
1972 if (relayd
&& relayd_hang_up
) {
1973 ERR("Relayd hangup. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
1974 lttng_consumer_cleanup_relayd(relayd
);
1975 /* Skip splice error so the consumer does not fail */
1980 /* send the appropriate error description to sessiond */
1983 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1986 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1989 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1994 if (relayd
&& stream
->metadata_flag
) {
1995 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
2003 * Sample the snapshot positions for a specific fd
2005 * Returns 0 on success, < 0 on error
2007 int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream
*stream
)
2009 switch (the_consumer_data
.type
) {
2010 case LTTNG_CONSUMER_KERNEL
:
2011 return lttng_kconsumer_sample_snapshot_positions(stream
);
2012 case LTTNG_CONSUMER32_UST
:
2013 case LTTNG_CONSUMER64_UST
:
2014 return lttng_ustconsumer_sample_snapshot_positions(stream
);
2016 ERR("Unknown consumer_data type");
2022 * Take a snapshot for a specific fd
2024 * Returns 0 on success, < 0 on error
2026 int lttng_consumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2028 switch (the_consumer_data
.type
) {
2029 case LTTNG_CONSUMER_KERNEL
:
2030 return lttng_kconsumer_take_snapshot(stream
);
2031 case LTTNG_CONSUMER32_UST
:
2032 case LTTNG_CONSUMER64_UST
:
2033 return lttng_ustconsumer_take_snapshot(stream
);
2035 ERR("Unknown consumer_data type");
2042 * Get the produced position
2044 * Returns 0 on success, < 0 on error
2046 int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
2049 switch (the_consumer_data
.type
) {
2050 case LTTNG_CONSUMER_KERNEL
:
2051 return lttng_kconsumer_get_produced_snapshot(stream
, pos
);
2052 case LTTNG_CONSUMER32_UST
:
2053 case LTTNG_CONSUMER64_UST
:
2054 return lttng_ustconsumer_get_produced_snapshot(stream
, pos
);
2056 ERR("Unknown consumer_data type");
2063 * Get the consumed position (free-running counter position in bytes).
2065 * Returns 0 on success, < 0 on error
2067 int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
2070 switch (the_consumer_data
.type
) {
2071 case LTTNG_CONSUMER_KERNEL
:
2072 return lttng_kconsumer_get_consumed_snapshot(stream
, pos
);
2073 case LTTNG_CONSUMER32_UST
:
2074 case LTTNG_CONSUMER64_UST
:
2075 return lttng_ustconsumer_get_consumed_snapshot(stream
, pos
);
2077 ERR("Unknown consumer_data type");
2083 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
2084 int sock
, struct pollfd
*consumer_sockpoll
)
2086 switch (the_consumer_data
.type
) {
2087 case LTTNG_CONSUMER_KERNEL
:
2088 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2089 case LTTNG_CONSUMER32_UST
:
2090 case LTTNG_CONSUMER64_UST
:
2091 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2093 ERR("Unknown consumer_data type");
2100 void lttng_consumer_close_all_metadata(void)
2102 switch (the_consumer_data
.type
) {
2103 case LTTNG_CONSUMER_KERNEL
:
2105 * The Kernel consumer has a different metadata scheme so we don't
2106 * close anything because the stream will be closed by the session
2110 case LTTNG_CONSUMER32_UST
:
2111 case LTTNG_CONSUMER64_UST
:
2113 * Close all metadata streams. The metadata hash table is passed and
2114 * this call iterates over it by closing all wakeup fd. This is safe
2115 * because at this point we are sure that the metadata producer is
2116 * either dead or blocked.
2118 lttng_ustconsumer_close_all_metadata(metadata_ht
);
2121 ERR("Unknown consumer_data type");
2127 * Clean up a metadata stream and free its memory.
2129 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
2130 struct lttng_ht
*ht
)
2132 struct lttng_consumer_channel
*channel
= NULL
;
2133 bool free_channel
= false;
2135 LTTNG_ASSERT(stream
);
2137 * This call should NEVER receive regular stream. It must always be
2138 * metadata stream and this is crucial for data structure synchronization.
2140 LTTNG_ASSERT(stream
->metadata_flag
);
2142 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
2144 pthread_mutex_lock(&the_consumer_data
.lock
);
2146 * Note that this assumes that a stream's channel is never changed and
2147 * that the stream's lock doesn't need to be taken to sample its
2150 channel
= stream
->chan
;
2151 pthread_mutex_lock(&channel
->lock
);
2152 pthread_mutex_lock(&stream
->lock
);
2153 if (channel
->metadata_cache
) {
2154 /* Only applicable to userspace consumers. */
2155 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
2158 /* Remove any reference to that stream. */
2159 consumer_stream_delete(stream
, ht
);
2161 /* Close down everything including the relayd if one. */
2162 consumer_stream_close(stream
);
2163 /* Destroy tracer buffers of the stream. */
2164 consumer_stream_destroy_buffers(stream
);
2166 /* Atomically decrement channel refcount since other threads can use it. */
2167 if (!uatomic_sub_return(&channel
->refcount
, 1)
2168 && !uatomic_read(&channel
->nb_init_stream_left
)) {
2169 /* Go for channel deletion! */
2170 free_channel
= true;
2172 stream
->chan
= NULL
;
2175 * Nullify the stream reference so it is not used after deletion. The
2176 * channel lock MUST be acquired before being able to check for a NULL
2179 channel
->metadata_stream
= NULL
;
2181 if (channel
->metadata_cache
) {
2182 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
2184 pthread_mutex_unlock(&stream
->lock
);
2185 pthread_mutex_unlock(&channel
->lock
);
2186 pthread_mutex_unlock(&the_consumer_data
.lock
);
2189 consumer_del_channel(channel
);
2192 lttng_trace_chunk_put(stream
->trace_chunk
);
2193 stream
->trace_chunk
= NULL
;
2194 consumer_stream_free(stream
);
2198 * Action done with the metadata stream when adding it to the consumer internal
2199 * data structures to handle it.
2201 void consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
)
2203 struct lttng_ht
*ht
= metadata_ht
;
2204 struct lttng_ht_iter iter
;
2205 struct lttng_ht_node_u64
*node
;
2207 LTTNG_ASSERT(stream
);
2210 DBG3("Adding metadata stream %" PRIu64
" to hash table", stream
->key
);
2212 pthread_mutex_lock(&the_consumer_data
.lock
);
2213 pthread_mutex_lock(&stream
->chan
->lock
);
2214 pthread_mutex_lock(&stream
->chan
->timer_lock
);
2215 pthread_mutex_lock(&stream
->lock
);
2218 * From here, refcounts are updated so be _careful_ when returning an error
2225 * Lookup the stream just to make sure it does not exist in our internal
2226 * state. This should NEVER happen.
2228 lttng_ht_lookup(ht
, &stream
->key
, &iter
);
2229 node
= lttng_ht_iter_get_node_u64(&iter
);
2230 LTTNG_ASSERT(!node
);
2233 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2234 * in terms of destroying the associated channel, because the action that
2235 * causes the count to become 0 also causes a stream to be added. The
2236 * channel deletion will thus be triggered by the following removal of this
2239 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
2240 /* Increment refcount before decrementing nb_init_stream_left */
2242 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
2245 lttng_ht_add_unique_u64(ht
, &stream
->node
);
2247 lttng_ht_add_u64(the_consumer_data
.stream_per_chan_id_ht
,
2248 &stream
->node_channel_id
);
2251 * Add stream to the stream_list_ht of the consumer data. No need to steal
2252 * the key since the HT does not use it and we allow to add redundant keys
2255 lttng_ht_add_u64(the_consumer_data
.stream_list_ht
,
2256 &stream
->node_session_id
);
2260 pthread_mutex_unlock(&stream
->lock
);
2261 pthread_mutex_unlock(&stream
->chan
->lock
);
2262 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
2263 pthread_mutex_unlock(&the_consumer_data
.lock
);
2267 * Delete data stream that are flagged for deletion (endpoint_status).
2269 static void validate_endpoint_status_data_stream(void)
2271 struct lttng_ht_iter iter
;
2272 struct lttng_consumer_stream
*stream
;
2274 DBG("Consumer delete flagged data stream");
2277 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2278 /* Validate delete flag of the stream */
2279 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2282 /* Delete it right now */
2283 consumer_del_stream(stream
, data_ht
);
2289 * Delete metadata stream that are flagged for deletion (endpoint_status).
2291 static void validate_endpoint_status_metadata_stream(
2292 struct lttng_poll_event
*pollset
)
2294 struct lttng_ht_iter iter
;
2295 struct lttng_consumer_stream
*stream
;
2297 DBG("Consumer delete flagged metadata stream");
2299 LTTNG_ASSERT(pollset
);
2302 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2303 /* Validate delete flag of the stream */
2304 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2308 * Remove from pollset so the metadata thread can continue without
2309 * blocking on a deleted stream.
2311 lttng_poll_del(pollset
, stream
->wait_fd
);
2313 /* Delete it right now */
2314 consumer_del_metadata_stream(stream
, metadata_ht
);
2320 * Thread polls on metadata file descriptor and write them on disk or on the
2323 void *consumer_thread_metadata_poll(void *data
)
2325 int ret
, i
, pollfd
, err
= -1;
2326 uint32_t revents
, nb_fd
;
2327 struct lttng_consumer_stream
*stream
= NULL
;
2328 struct lttng_ht_iter iter
;
2329 struct lttng_ht_node_u64
*node
;
2330 struct lttng_poll_event events
;
2331 struct lttng_consumer_local_data
*ctx
= (lttng_consumer_local_data
*) data
;
2334 rcu_register_thread();
2336 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_METADATA
);
2338 if (testpoint(consumerd_thread_metadata
)) {
2339 goto error_testpoint
;
2342 health_code_update();
2344 DBG("Thread metadata poll started");
2346 /* Size is set to 1 for the consumer_metadata pipe */
2347 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2349 ERR("Poll set creation failed");
2353 ret
= lttng_poll_add(&events
,
2354 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
), LPOLLIN
);
2360 DBG("Metadata main loop started");
2364 health_code_update();
2365 health_poll_entry();
2366 DBG("Metadata poll wait");
2367 ret
= lttng_poll_wait(&events
, -1);
2368 DBG("Metadata poll return from wait with %d fd(s)",
2369 LTTNG_POLL_GETNB(&events
));
2371 DBG("Metadata event caught in thread");
2373 if (errno
== EINTR
) {
2374 ERR("Poll EINTR caught");
2377 if (LTTNG_POLL_GETNB(&events
) == 0) {
2378 err
= 0; /* All is OK */
2385 /* From here, the event is a metadata wait fd */
2386 for (i
= 0; i
< nb_fd
; i
++) {
2387 health_code_update();
2389 revents
= LTTNG_POLL_GETEV(&events
, i
);
2390 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2392 if (pollfd
== lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
)) {
2393 if (revents
& LPOLLIN
) {
2396 pipe_len
= lttng_pipe_read(ctx
->consumer_metadata_pipe
,
2397 &stream
, sizeof(stream
));
2398 if (pipe_len
< sizeof(stream
)) {
2400 PERROR("read metadata stream");
2403 * Remove the pipe from the poll set and continue the loop
2404 * since their might be data to consume.
2406 lttng_poll_del(&events
,
2407 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2408 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2412 /* A NULL stream means that the state has changed. */
2413 if (stream
== NULL
) {
2414 /* Check for deleted streams. */
2415 validate_endpoint_status_metadata_stream(&events
);
2419 DBG("Adding metadata stream %d to poll set",
2422 /* Add metadata stream to the global poll events list */
2423 lttng_poll_add(&events
, stream
->wait_fd
,
2424 LPOLLIN
| LPOLLPRI
| LPOLLHUP
);
2425 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2426 DBG("Metadata thread pipe hung up");
2428 * Remove the pipe from the poll set and continue the loop
2429 * since their might be data to consume.
2431 lttng_poll_del(&events
,
2432 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2433 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2436 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2440 /* Handle other stream */
2446 uint64_t tmp_id
= (uint64_t) pollfd
;
2448 lttng_ht_lookup(metadata_ht
, &tmp_id
, &iter
);
2450 node
= lttng_ht_iter_get_node_u64(&iter
);
2453 stream
= caa_container_of(node
, struct lttng_consumer_stream
,
2456 if (revents
& (LPOLLIN
| LPOLLPRI
)) {
2457 /* Get the data out of the metadata file descriptor */
2458 DBG("Metadata available on fd %d", pollfd
);
2459 LTTNG_ASSERT(stream
->wait_fd
== pollfd
);
2462 health_code_update();
2464 len
= ctx
->on_buffer_ready(stream
, ctx
, false);
2466 * We don't check the return value here since if we get
2467 * a negative len, it means an error occurred thus we
2468 * simply remove it from the poll set and free the
2473 /* It's ok to have an unavailable sub-buffer */
2474 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2475 /* Clean up stream from consumer and free it. */
2476 lttng_poll_del(&events
, stream
->wait_fd
);
2477 consumer_del_metadata_stream(stream
, metadata_ht
);
2479 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2480 DBG("Metadata fd %d is hup|err.", pollfd
);
2481 if (!stream
->hangup_flush_done
&&
2482 (the_consumer_data
.type
== LTTNG_CONSUMER32_UST
||
2483 the_consumer_data
.type
==
2484 LTTNG_CONSUMER64_UST
)) {
2485 DBG("Attempting to flush and consume the UST buffers");
2486 lttng_ustconsumer_on_stream_hangup(stream
);
2488 /* We just flushed the stream now read it. */
2490 health_code_update();
2492 len
= ctx
->on_buffer_ready(stream
, ctx
, false);
2494 * We don't check the return value here since if we get
2495 * a negative len, it means an error occurred thus we
2496 * simply remove it from the poll set and free the
2502 lttng_poll_del(&events
, stream
->wait_fd
);
2504 * This call update the channel states, closes file descriptors
2505 * and securely free the stream.
2507 consumer_del_metadata_stream(stream
, metadata_ht
);
2509 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2513 /* Release RCU lock for the stream looked up */
2521 DBG("Metadata poll thread exiting");
2523 lttng_poll_clean(&events
);
2528 ERR("Health error occurred in %s", __func__
);
2530 health_unregister(health_consumerd
);
2531 rcu_unregister_thread();
2536 * This thread polls the fds in the set to consume the data and write
2537 * it to tracefile if necessary.
2539 void *consumer_thread_data_poll(void *data
)
2541 int num_rdy
, num_hup
, high_prio
, ret
, i
, err
= -1;
2542 struct pollfd
*pollfd
= NULL
;
2543 /* local view of the streams */
2544 struct lttng_consumer_stream
**local_stream
= NULL
, *new_stream
= NULL
;
2545 /* local view of consumer_data.fds_count */
2547 /* 2 for the consumer_data_pipe and wake up pipe */
2548 const int nb_pipes_fd
= 2;
2549 /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
2550 int nb_inactive_fd
= 0;
2551 struct lttng_consumer_local_data
*ctx
= (lttng_consumer_local_data
*) data
;
2554 rcu_register_thread();
2556 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_DATA
);
2558 if (testpoint(consumerd_thread_data
)) {
2559 goto error_testpoint
;
2562 health_code_update();
2564 local_stream
= zmalloc
<lttng_consumer_stream
*>();
2565 if (local_stream
== NULL
) {
2566 PERROR("local_stream malloc");
2571 health_code_update();
2577 * the fds set has been updated, we need to update our
2578 * local array as well
2580 pthread_mutex_lock(&the_consumer_data
.lock
);
2581 if (the_consumer_data
.need_update
) {
2586 local_stream
= NULL
;
2588 /* Allocate for all fds */
2589 pollfd
= calloc
<struct pollfd
>(the_consumer_data
.stream_count
+ nb_pipes_fd
);
2590 if (pollfd
== NULL
) {
2591 PERROR("pollfd malloc");
2592 pthread_mutex_unlock(&the_consumer_data
.lock
);
2596 local_stream
= calloc
<lttng_consumer_stream
*>(the_consumer_data
.stream_count
+ nb_pipes_fd
);
2597 if (local_stream
== NULL
) {
2598 PERROR("local_stream malloc");
2599 pthread_mutex_unlock(&the_consumer_data
.lock
);
2602 ret
= update_poll_array(ctx
, &pollfd
, local_stream
,
2603 data_ht
, &nb_inactive_fd
);
2605 ERR("Error in allocating pollfd or local_outfds");
2606 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2607 pthread_mutex_unlock(&the_consumer_data
.lock
);
2611 the_consumer_data
.need_update
= 0;
2613 pthread_mutex_unlock(&the_consumer_data
.lock
);
2615 /* No FDs and consumer_quit, consumer_cleanup the thread */
2616 if (nb_fd
== 0 && nb_inactive_fd
== 0 &&
2617 CMM_LOAD_SHARED(consumer_quit
) == 1) {
2618 err
= 0; /* All is OK */
2621 /* poll on the array of fds */
2623 DBG("polling on %d fd", nb_fd
+ nb_pipes_fd
);
2624 if (testpoint(consumerd_thread_data_poll
)) {
2627 health_poll_entry();
2628 num_rdy
= poll(pollfd
, nb_fd
+ nb_pipes_fd
, -1);
2630 DBG("poll num_rdy : %d", num_rdy
);
2631 if (num_rdy
== -1) {
2633 * Restart interrupted system call.
2635 if (errno
== EINTR
) {
2638 PERROR("Poll error");
2639 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2641 } else if (num_rdy
== 0) {
2642 DBG("Polling thread timed out");
2646 if (caa_unlikely(data_consumption_paused
)) {
2647 DBG("Data consumption paused, sleeping...");
2653 * If the consumer_data_pipe triggered poll go directly to the
2654 * beginning of the loop to update the array. We want to prioritize
2655 * array update over low-priority reads.
2657 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2658 ssize_t pipe_readlen
;
2660 DBG("consumer_data_pipe wake up");
2661 pipe_readlen
= lttng_pipe_read(ctx
->consumer_data_pipe
,
2662 &new_stream
, sizeof(new_stream
));
2663 if (pipe_readlen
< sizeof(new_stream
)) {
2664 PERROR("Consumer data pipe");
2665 /* Continue so we can at least handle the current stream(s). */
2670 * If the stream is NULL, just ignore it. It's also possible that
2671 * the sessiond poll thread changed the consumer_quit state and is
2672 * waking us up to test it.
2674 if (new_stream
== NULL
) {
2675 validate_endpoint_status_data_stream();
2679 /* Continue to update the local streams and handle prio ones */
2683 /* Handle wakeup pipe. */
2684 if (pollfd
[nb_fd
+ 1].revents
& (POLLIN
| POLLPRI
)) {
2686 ssize_t pipe_readlen
;
2688 pipe_readlen
= lttng_pipe_read(ctx
->consumer_wakeup_pipe
, &dummy
,
2690 if (pipe_readlen
< 0) {
2691 PERROR("Consumer data wakeup pipe");
2693 /* We've been awakened to handle stream(s). */
2694 ctx
->has_wakeup
= 0;
2697 /* Take care of high priority channels first. */
2698 for (i
= 0; i
< nb_fd
; i
++) {
2699 health_code_update();
2701 if (local_stream
[i
] == NULL
) {
2704 if (pollfd
[i
].revents
& POLLPRI
) {
2705 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2707 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
, false);
2708 /* it's ok to have an unavailable sub-buffer */
2709 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2710 /* Clean the stream and free it. */
2711 consumer_del_stream(local_stream
[i
], data_ht
);
2712 local_stream
[i
] = NULL
;
2713 } else if (len
> 0) {
2714 local_stream
[i
]->data_read
= 1;
2720 * If we read high prio channel in this loop, try again
2721 * for more high prio data.
2727 /* Take care of low priority channels. */
2728 for (i
= 0; i
< nb_fd
; i
++) {
2729 health_code_update();
2731 if (local_stream
[i
] == NULL
) {
2734 if ((pollfd
[i
].revents
& POLLIN
) ||
2735 local_stream
[i
]->hangup_flush_done
||
2736 local_stream
[i
]->has_data
) {
2737 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2738 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
, false);
2739 /* it's ok to have an unavailable sub-buffer */
2740 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2741 /* Clean the stream and free it. */
2742 consumer_del_stream(local_stream
[i
], data_ht
);
2743 local_stream
[i
] = NULL
;
2744 } else if (len
> 0) {
2745 local_stream
[i
]->data_read
= 1;
2750 /* Handle hangup and errors */
2751 for (i
= 0; i
< nb_fd
; i
++) {
2752 health_code_update();
2754 if (local_stream
[i
] == NULL
) {
2757 if (!local_stream
[i
]->hangup_flush_done
2758 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2759 && (the_consumer_data
.type
== LTTNG_CONSUMER32_UST
2760 || the_consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2761 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2763 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2764 /* Attempt read again, for the data we just flushed. */
2765 local_stream
[i
]->data_read
= 1;
2768 * If the poll flag is HUP/ERR/NVAL and we have
2769 * read no data in this pass, we can remove the
2770 * stream from its hash table.
2772 if ((pollfd
[i
].revents
& POLLHUP
)) {
2773 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2774 if (!local_stream
[i
]->data_read
) {
2775 consumer_del_stream(local_stream
[i
], data_ht
);
2776 local_stream
[i
] = NULL
;
2779 } else if (pollfd
[i
].revents
& POLLERR
) {
2780 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2781 if (!local_stream
[i
]->data_read
) {
2782 consumer_del_stream(local_stream
[i
], data_ht
);
2783 local_stream
[i
] = NULL
;
2786 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2787 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2788 if (!local_stream
[i
]->data_read
) {
2789 consumer_del_stream(local_stream
[i
], data_ht
);
2790 local_stream
[i
] = NULL
;
2794 if (local_stream
[i
] != NULL
) {
2795 local_stream
[i
]->data_read
= 0;
2802 DBG("polling thread exiting");
2807 * Close the write side of the pipe so epoll_wait() in
2808 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2809 * read side of the pipe. If we close them both, epoll_wait strangely does
2810 * not return and could create a endless wait period if the pipe is the
2811 * only tracked fd in the poll set. The thread will take care of closing
2814 (void) lttng_pipe_write_close(ctx
->consumer_metadata_pipe
);
2819 ERR("Health error occurred in %s", __func__
);
2821 health_unregister(health_consumerd
);
2823 rcu_unregister_thread();
2828 * Close wake-up end of each stream belonging to the channel. This will
2829 * allow the poll() on the stream read-side to detect when the
2830 * write-side (application) finally closes them.
2833 void consumer_close_channel_streams(struct lttng_consumer_channel
*channel
)
2835 struct lttng_ht
*ht
;
2836 struct lttng_consumer_stream
*stream
;
2837 struct lttng_ht_iter iter
;
2839 ht
= the_consumer_data
.stream_per_chan_id_ht
;
2842 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2843 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
2844 ht
->match_fct
, &channel
->key
,
2845 &iter
.iter
, stream
, node_channel_id
.node
) {
2847 * Protect against teardown with mutex.
2849 pthread_mutex_lock(&stream
->lock
);
2850 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
2853 switch (the_consumer_data
.type
) {
2854 case LTTNG_CONSUMER_KERNEL
:
2856 case LTTNG_CONSUMER32_UST
:
2857 case LTTNG_CONSUMER64_UST
:
2858 if (stream
->metadata_flag
) {
2859 /* Safe and protected by the stream lock. */
2860 lttng_ustconsumer_close_metadata(stream
->chan
);
2863 * Note: a mutex is taken internally within
2864 * liblttng-ust-ctl to protect timer wakeup_fd
2865 * use from concurrent close.
2867 lttng_ustconsumer_close_stream_wakeup(stream
);
2871 ERR("Unknown consumer_data type");
2875 pthread_mutex_unlock(&stream
->lock
);
2880 static void destroy_channel_ht(struct lttng_ht
*ht
)
2882 struct lttng_ht_iter iter
;
2883 struct lttng_consumer_channel
*channel
;
2891 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, channel
, wait_fd_node
.node
) {
2892 ret
= lttng_ht_del(ht
, &iter
);
2893 LTTNG_ASSERT(ret
!= 0);
2897 lttng_ht_destroy(ht
);
2901 * This thread polls the channel fds to detect when they are being
2902 * closed. It closes all related streams if the channel is detected as
2903 * closed. It is currently only used as a shim layer for UST because the
2904 * consumerd needs to keep the per-stream wakeup end of pipes open for
2907 void *consumer_thread_channel_poll(void *data
)
2909 int ret
, i
, pollfd
, err
= -1;
2910 uint32_t revents
, nb_fd
;
2911 struct lttng_consumer_channel
*chan
= NULL
;
2912 struct lttng_ht_iter iter
;
2913 struct lttng_ht_node_u64
*node
;
2914 struct lttng_poll_event events
;
2915 struct lttng_consumer_local_data
*ctx
= (lttng_consumer_local_data
*) data
;
2916 struct lttng_ht
*channel_ht
;
2918 rcu_register_thread();
2920 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_CHANNEL
);
2922 if (testpoint(consumerd_thread_channel
)) {
2923 goto error_testpoint
;
2926 health_code_update();
2928 channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2930 /* ENOMEM at this point. Better to bail out. */
2934 DBG("Thread channel poll started");
2936 /* Size is set to 1 for the consumer_channel pipe */
2937 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2939 ERR("Poll set creation failed");
2943 ret
= lttng_poll_add(&events
, ctx
->consumer_channel_pipe
[0], LPOLLIN
);
2949 DBG("Channel main loop started");
2953 health_code_update();
2954 DBG("Channel poll wait");
2955 health_poll_entry();
2956 ret
= lttng_poll_wait(&events
, -1);
2957 DBG("Channel poll return from wait with %d fd(s)",
2958 LTTNG_POLL_GETNB(&events
));
2960 DBG("Channel event caught in thread");
2962 if (errno
== EINTR
) {
2963 ERR("Poll EINTR caught");
2966 if (LTTNG_POLL_GETNB(&events
) == 0) {
2967 err
= 0; /* All is OK */
2974 /* From here, the event is a channel wait fd */
2975 for (i
= 0; i
< nb_fd
; i
++) {
2976 health_code_update();
2978 revents
= LTTNG_POLL_GETEV(&events
, i
);
2979 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2981 if (pollfd
== ctx
->consumer_channel_pipe
[0]) {
2982 if (revents
& LPOLLIN
) {
2983 enum consumer_channel_action action
;
2986 ret
= read_channel_pipe(ctx
, &chan
, &key
, &action
);
2989 ERR("Error reading channel pipe");
2991 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
2996 case CONSUMER_CHANNEL_ADD
:
2997 DBG("Adding channel %d to poll set",
3000 lttng_ht_node_init_u64(&chan
->wait_fd_node
,
3003 lttng_ht_add_unique_u64(channel_ht
,
3004 &chan
->wait_fd_node
);
3006 /* Add channel to the global poll events list */
3007 lttng_poll_add(&events
, chan
->wait_fd
,
3008 LPOLLERR
| LPOLLHUP
);
3010 case CONSUMER_CHANNEL_DEL
:
3013 * This command should never be called if the channel
3014 * has streams monitored by either the data or metadata
3015 * thread. The consumer only notify this thread with a
3016 * channel del. command if it receives a destroy
3017 * channel command from the session daemon that send it
3018 * if a command prior to the GET_CHANNEL failed.
3022 chan
= consumer_find_channel(key
);
3025 ERR("UST consumer get channel key %" PRIu64
" not found for del channel", key
);
3028 lttng_poll_del(&events
, chan
->wait_fd
);
3029 iter
.iter
.node
= &chan
->wait_fd_node
.node
;
3030 ret
= lttng_ht_del(channel_ht
, &iter
);
3031 LTTNG_ASSERT(ret
== 0);
3033 switch (the_consumer_data
.type
) {
3034 case LTTNG_CONSUMER_KERNEL
:
3036 case LTTNG_CONSUMER32_UST
:
3037 case LTTNG_CONSUMER64_UST
:
3038 health_code_update();
3039 /* Destroy streams that might have been left in the stream list. */
3040 clean_channel_stream_list(chan
);
3043 ERR("Unknown consumer_data type");
3048 * Release our own refcount. Force channel deletion even if
3049 * streams were not initialized.
3051 if (!uatomic_sub_return(&chan
->refcount
, 1)) {
3052 consumer_del_channel(chan
);
3057 case CONSUMER_CHANNEL_QUIT
:
3059 * Remove the pipe from the poll set and continue the loop
3060 * since their might be data to consume.
3062 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3065 ERR("Unknown action");
3068 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3069 DBG("Channel thread pipe hung up");
3071 * Remove the pipe from the poll set and continue the loop
3072 * since their might be data to consume.
3074 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3077 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3081 /* Handle other stream */
3087 uint64_t tmp_id
= (uint64_t) pollfd
;
3089 lttng_ht_lookup(channel_ht
, &tmp_id
, &iter
);
3091 node
= lttng_ht_iter_get_node_u64(&iter
);
3094 chan
= caa_container_of(node
, struct lttng_consumer_channel
,
3097 /* Check for error event */
3098 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3099 DBG("Channel fd %d is hup|err.", pollfd
);
3101 lttng_poll_del(&events
, chan
->wait_fd
);
3102 ret
= lttng_ht_del(channel_ht
, &iter
);
3103 LTTNG_ASSERT(ret
== 0);
3106 * This will close the wait fd for each stream associated to
3107 * this channel AND monitored by the data/metadata thread thus
3108 * will be clean by the right thread.
3110 consumer_close_channel_streams(chan
);
3112 /* Release our own refcount */
3113 if (!uatomic_sub_return(&chan
->refcount
, 1)
3114 && !uatomic_read(&chan
->nb_init_stream_left
)) {
3115 consumer_del_channel(chan
);
3118 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3123 /* Release RCU lock for the channel looked up */
3131 lttng_poll_clean(&events
);
3133 destroy_channel_ht(channel_ht
);
3136 DBG("Channel poll thread exiting");
3139 ERR("Health error occurred in %s", __func__
);
3141 health_unregister(health_consumerd
);
3142 rcu_unregister_thread();
3146 static int set_metadata_socket(struct lttng_consumer_local_data
*ctx
,
3147 struct pollfd
*sockpoll
, int client_socket
)
3152 LTTNG_ASSERT(sockpoll
);
3154 ret
= lttng_consumer_poll_socket(sockpoll
);
3158 DBG("Metadata connection on client_socket");
3160 /* Blocking call, waiting for transmission */
3161 ctx
->consumer_metadata_socket
= lttcomm_accept_unix_sock(client_socket
);
3162 if (ctx
->consumer_metadata_socket
< 0) {
3163 WARN("On accept metadata");
3174 * This thread listens on the consumerd socket and receives the file
3175 * descriptors from the session daemon.
3177 void *consumer_thread_sessiond_poll(void *data
)
3179 int sock
= -1, client_socket
, ret
, err
= -1;
3181 * structure to poll for incoming data on communication socket avoids
3182 * making blocking sockets.
3184 struct pollfd consumer_sockpoll
[2];
3185 struct lttng_consumer_local_data
*ctx
= (lttng_consumer_local_data
*) data
;
3187 rcu_register_thread();
3189 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_SESSIOND
);
3191 if (testpoint(consumerd_thread_sessiond
)) {
3192 goto error_testpoint
;
3195 health_code_update();
3197 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
3198 unlink(ctx
->consumer_command_sock_path
);
3199 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
3200 if (client_socket
< 0) {
3201 ERR("Cannot create command socket");
3205 ret
= lttcomm_listen_unix_sock(client_socket
);
3210 DBG("Sending ready command to lttng-sessiond");
3211 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
3212 /* return < 0 on error, but == 0 is not fatal */
3214 ERR("Error sending ready command to lttng-sessiond");
3218 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3219 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
3220 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
3221 consumer_sockpoll
[1].fd
= client_socket
;
3222 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3224 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3232 DBG("Connection on client_socket");
3234 /* Blocking call, waiting for transmission */
3235 sock
= lttcomm_accept_unix_sock(client_socket
);
3242 * Setup metadata socket which is the second socket connection on the
3243 * command unix socket.
3245 ret
= set_metadata_socket(ctx
, consumer_sockpoll
, client_socket
);
3254 /* This socket is not useful anymore. */
3255 ret
= close(client_socket
);
3257 PERROR("close client_socket");
3261 /* update the polling structure to poll on the established socket */
3262 consumer_sockpoll
[1].fd
= sock
;
3263 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3266 health_code_update();
3268 health_poll_entry();
3269 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3278 DBG("Incoming command on sock");
3279 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
3282 * This could simply be a session daemon quitting. Don't output
3285 DBG("Communication interrupted on command socket");
3289 if (CMM_LOAD_SHARED(consumer_quit
)) {
3290 DBG("consumer_thread_receive_fds received quit from signal");
3291 err
= 0; /* All is OK */
3294 DBG("Received command on sock");
3300 DBG("Consumer thread sessiond poll exiting");
3303 * Close metadata streams since the producer is the session daemon which
3306 * NOTE: for now, this only applies to the UST tracer.
3308 lttng_consumer_close_all_metadata();
3311 * when all fds have hung up, the polling thread
3314 CMM_STORE_SHARED(consumer_quit
, 1);
3317 * Notify the data poll thread to poll back again and test the
3318 * consumer_quit state that we just set so to quit gracefully.
3320 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
3322 notify_channel_pipe(ctx
, NULL
, -1, CONSUMER_CHANNEL_QUIT
);
3324 notify_health_quit_pipe(health_quit_pipe
);
3326 /* Cleaning up possibly open sockets. */
3330 PERROR("close sock sessiond poll");
3333 if (client_socket
>= 0) {
3334 ret
= close(client_socket
);
3336 PERROR("close client_socket sessiond poll");
3343 ERR("Health error occurred in %s", __func__
);
3345 health_unregister(health_consumerd
);
3347 rcu_unregister_thread();
3351 static int post_consume(struct lttng_consumer_stream
*stream
,
3352 const struct stream_subbuffer
*subbuffer
,
3353 struct lttng_consumer_local_data
*ctx
)
3357 const size_t count
= lttng_dynamic_array_get_count(
3358 &stream
->read_subbuffer_ops
.post_consume_cbs
);
3360 for (i
= 0; i
< count
; i
++) {
3361 const post_consume_cb op
= *(post_consume_cb
*) lttng_dynamic_array_get_element(
3362 &stream
->read_subbuffer_ops
.post_consume_cbs
,
3365 ret
= op(stream
, subbuffer
, ctx
);
3374 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
3375 struct lttng_consumer_local_data
*ctx
,
3376 bool locked_by_caller
)
3378 ssize_t ret
, written_bytes
= 0;
3380 struct stream_subbuffer subbuffer
= {};
3381 enum get_next_subbuffer_status get_next_status
;
3383 if (!locked_by_caller
) {
3384 stream
->read_subbuffer_ops
.lock(stream
);
3386 stream
->read_subbuffer_ops
.assert_locked(stream
);
3389 if (stream
->read_subbuffer_ops
.on_wake_up
) {
3390 ret
= stream
->read_subbuffer_ops
.on_wake_up(stream
);
3397 * If the stream was flagged to be ready for rotation before we extract
3398 * the next packet, rotate it now.
3400 if (stream
->rotate_ready
) {
3401 DBG("Rotate stream before consuming data");
3402 ret
= lttng_consumer_rotate_stream(stream
);
3404 ERR("Stream rotation error before consuming data");
3409 get_next_status
= stream
->read_subbuffer_ops
.get_next_subbuffer(
3410 stream
, &subbuffer
);
3411 switch (get_next_status
) {
3412 case GET_NEXT_SUBBUFFER_STATUS_OK
:
3414 case GET_NEXT_SUBBUFFER_STATUS_NO_DATA
:
3418 case GET_NEXT_SUBBUFFER_STATUS_ERROR
:
3425 ret
= stream
->read_subbuffer_ops
.pre_consume_subbuffer(
3426 stream
, &subbuffer
);
3428 goto error_put_subbuf
;
3431 written_bytes
= stream
->read_subbuffer_ops
.consume_subbuffer(
3432 ctx
, stream
, &subbuffer
);
3433 if (written_bytes
<= 0) {
3434 ERR("Error consuming subbuffer: (%zd)", written_bytes
);
3435 ret
= (int) written_bytes
;
3436 goto error_put_subbuf
;
3439 ret
= stream
->read_subbuffer_ops
.put_next_subbuffer(stream
, &subbuffer
);
3444 ret
= post_consume(stream
, &subbuffer
, ctx
);
3450 * After extracting the packet, we check if the stream is now ready to
3451 * be rotated and perform the action immediately.
3453 * Don't overwrite `ret` as callers expect the number of bytes
3454 * consumed to be returned on success.
3456 rotation_ret
= lttng_consumer_stream_is_rotate_ready(stream
);
3457 if (rotation_ret
== 1) {
3458 rotation_ret
= lttng_consumer_rotate_stream(stream
);
3459 if (rotation_ret
< 0) {
3461 ERR("Stream rotation error after consuming data");
3465 } else if (rotation_ret
< 0) {
3467 ERR("Failed to check if stream was ready to rotate after consuming data");
3472 if (stream
->read_subbuffer_ops
.on_sleep
) {
3473 stream
->read_subbuffer_ops
.on_sleep(stream
, ctx
);
3476 ret
= written_bytes
;
3478 if (!locked_by_caller
) {
3479 stream
->read_subbuffer_ops
.unlock(stream
);
3484 (void) stream
->read_subbuffer_ops
.put_next_subbuffer(stream
, &subbuffer
);
3488 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3490 switch (the_consumer_data
.type
) {
3491 case LTTNG_CONSUMER_KERNEL
:
3492 return lttng_kconsumer_on_recv_stream(stream
);
3493 case LTTNG_CONSUMER32_UST
:
3494 case LTTNG_CONSUMER64_UST
:
3495 return lttng_ustconsumer_on_recv_stream(stream
);
3497 ERR("Unknown consumer_data type");
3504 * Allocate and set consumer data hash tables.
3506 int lttng_consumer_init(void)
3508 the_consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3509 if (!the_consumer_data
.channel_ht
) {
3513 the_consumer_data
.channels_by_session_id_ht
=
3514 lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3515 if (!the_consumer_data
.channels_by_session_id_ht
) {
3519 the_consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3520 if (!the_consumer_data
.relayd_ht
) {
3524 the_consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3525 if (!the_consumer_data
.stream_list_ht
) {
3529 the_consumer_data
.stream_per_chan_id_ht
=
3530 lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3531 if (!the_consumer_data
.stream_per_chan_id_ht
) {
3535 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3540 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3545 the_consumer_data
.chunk_registry
= lttng_trace_chunk_registry_create();
3546 if (!the_consumer_data
.chunk_registry
) {
3557 * Process the ADD_RELAYD command receive by a consumer.
3559 * This will create a relayd socket pair and add it to the relayd hash table.
3560 * The caller MUST acquire a RCU read side lock before calling it.
3562 void consumer_add_relayd_socket(uint64_t net_seq_idx
,
3564 struct lttng_consumer_local_data
*ctx
,
3566 struct pollfd
*consumer_sockpoll
,
3567 uint64_t sessiond_id
,
3568 uint64_t relayd_session_id
,
3569 uint32_t relayd_version_major
,
3570 uint32_t relayd_version_minor
,
3571 enum lttcomm_sock_proto relayd_socket_protocol
)
3573 int fd
= -1, ret
= -1, relayd_created
= 0;
3574 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3575 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3578 LTTNG_ASSERT(sock
>= 0);
3579 ASSERT_RCU_READ_LOCKED();
3581 DBG("Consumer adding relayd socket (idx: %" PRIu64
")", net_seq_idx
);
3583 /* Get relayd reference if exists. */
3584 relayd
= consumer_find_relayd(net_seq_idx
);
3585 if (relayd
== NULL
) {
3586 LTTNG_ASSERT(sock_type
== LTTNG_STREAM_CONTROL
);
3587 /* Not found. Allocate one. */
3588 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
3589 if (relayd
== NULL
) {
3590 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3593 relayd
->sessiond_session_id
= sessiond_id
;
3598 * This code path MUST continue to the consumer send status message to
3599 * we can notify the session daemon and continue our work without
3600 * killing everything.
3604 * relayd key should never be found for control socket.
3606 LTTNG_ASSERT(sock_type
!= LTTNG_STREAM_CONTROL
);
3609 /* First send a status message before receiving the fds. */
3610 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
3612 /* Somehow, the session daemon is not responding anymore. */
3613 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3614 goto error_nosignal
;
3617 /* Poll on consumer socket. */
3618 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3620 /* Needing to exit in the middle of a command: error. */
3621 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
3622 goto error_nosignal
;
3625 /* Get relayd socket from session daemon */
3626 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
3627 if (ret
!= sizeof(fd
)) {
3628 fd
= -1; /* Just in case it gets set with an invalid value. */
3631 * Failing to receive FDs might indicate a major problem such as
3632 * reaching a fd limit during the receive where the kernel returns a
3633 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3634 * don't take any chances and stop everything.
3636 * XXX: Feature request #558 will fix that and avoid this possible
3637 * issue when reaching the fd limit.
3639 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
3640 ret_code
= LTTCOMM_CONSUMERD_ERROR_RECV_FD
;
3644 /* Copy socket information and received FD */
3645 switch (sock_type
) {
3646 case LTTNG_STREAM_CONTROL
:
3647 /* Copy received lttcomm socket */
3648 ret
= lttcomm_populate_sock_from_open_socket(
3649 &relayd
->control_sock
.sock
, fd
,
3650 relayd_socket_protocol
);
3652 /* Assign version values. */
3653 relayd
->control_sock
.major
= relayd_version_major
;
3654 relayd
->control_sock
.minor
= relayd_version_minor
;
3656 relayd
->relayd_session_id
= relayd_session_id
;
3659 case LTTNG_STREAM_DATA
:
3660 /* Copy received lttcomm socket */
3661 ret
= lttcomm_populate_sock_from_open_socket(
3662 &relayd
->data_sock
.sock
, fd
,
3663 relayd_socket_protocol
);
3664 /* Assign version values. */
3665 relayd
->data_sock
.major
= relayd_version_major
;
3666 relayd
->data_sock
.minor
= relayd_version_minor
;
3669 ERR("Unknown relayd socket type (%d)", sock_type
);
3670 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3675 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3679 DBG("Consumer %s socket created successfully with net idx %" PRIu64
" (fd: %d)",
3680 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
3681 relayd
->net_seq_idx
, fd
);
3683 * We gave the ownership of the fd to the relayd structure. Set the
3684 * fd to -1 so we don't call close() on it in the error path below.
3688 /* We successfully added the socket. Send status back. */
3689 ret
= consumer_send_status_msg(sock
, ret_code
);
3691 /* Somehow, the session daemon is not responding anymore. */
3692 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3693 goto error_nosignal
;
3697 * Add relayd socket pair to consumer data hashtable. If object already
3698 * exists or on error, the function gracefully returns.
3707 if (consumer_send_status_msg(sock
, ret_code
) < 0) {
3708 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3712 /* Close received socket if valid. */
3715 PERROR("close received socket");
3719 if (relayd_created
) {
3725 * Search for a relayd associated to the session id and return the reference.
3727 * A rcu read side lock MUST be acquire before calling this function and locked
3728 * until the relayd object is no longer necessary.
3730 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
3732 struct lttng_ht_iter iter
;
3733 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3735 ASSERT_RCU_READ_LOCKED();
3737 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3738 cds_lfht_for_each_entry(the_consumer_data
.relayd_ht
->ht
, &iter
.iter
,
3739 relayd
, node
.node
) {
3741 * Check by sessiond id which is unique here where the relayd session
3742 * id might not be when having multiple relayd.
3744 if (relayd
->sessiond_session_id
== id
) {
3745 /* Found the relayd. There can be only one per id. */
3757 * Check if for a given session id there is still data needed to be extract
3760 * Return 1 if data is pending or else 0 meaning ready to be read.
3762 int consumer_data_pending(uint64_t id
)
3765 struct lttng_ht_iter iter
;
3766 struct lttng_ht
*ht
;
3767 struct lttng_consumer_stream
*stream
;
3768 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3769 int (*data_pending
)(struct lttng_consumer_stream
*);
3771 DBG("Consumer data pending command on session id %" PRIu64
, id
);
3774 pthread_mutex_lock(&the_consumer_data
.lock
);
3776 switch (the_consumer_data
.type
) {
3777 case LTTNG_CONSUMER_KERNEL
:
3778 data_pending
= lttng_kconsumer_data_pending
;
3780 case LTTNG_CONSUMER32_UST
:
3781 case LTTNG_CONSUMER64_UST
:
3782 data_pending
= lttng_ustconsumer_data_pending
;
3785 ERR("Unknown consumer data type");
3789 /* Ease our life a bit */
3790 ht
= the_consumer_data
.stream_list_ht
;
3792 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3793 ht
->hash_fct(&id
, lttng_ht_seed
),
3795 &iter
.iter
, stream
, node_session_id
.node
) {
3796 pthread_mutex_lock(&stream
->lock
);
3799 * A removed node from the hash table indicates that the stream has
3800 * been deleted thus having a guarantee that the buffers are closed
3801 * on the consumer side. However, data can still be transmitted
3802 * over the network so don't skip the relayd check.
3804 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
3806 /* Check the stream if there is data in the buffers. */
3807 ret
= data_pending(stream
);
3809 pthread_mutex_unlock(&stream
->lock
);
3814 pthread_mutex_unlock(&stream
->lock
);
3817 relayd
= find_relayd_by_session_id(id
);
3819 unsigned int is_data_inflight
= 0;
3821 /* Send init command for data pending. */
3822 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3823 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
3824 relayd
->relayd_session_id
);
3826 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3827 /* Communication error thus the relayd so no data pending. */
3828 goto data_not_pending
;
3831 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3832 ht
->hash_fct(&id
, lttng_ht_seed
),
3834 &iter
.iter
, stream
, node_session_id
.node
) {
3835 if (stream
->metadata_flag
) {
3836 ret
= relayd_quiescent_control(&relayd
->control_sock
,
3837 stream
->relayd_stream_id
);
3839 ret
= relayd_data_pending(&relayd
->control_sock
,
3840 stream
->relayd_stream_id
,
3841 stream
->next_net_seq_num
- 1);
3845 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3847 } else if (ret
< 0) {
3848 ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
3849 lttng_consumer_cleanup_relayd(relayd
);
3850 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3851 goto data_not_pending
;
3855 /* Send end command for data pending. */
3856 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3857 relayd
->relayd_session_id
, &is_data_inflight
);
3858 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3860 ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
3861 lttng_consumer_cleanup_relayd(relayd
);
3862 goto data_not_pending
;
3864 if (is_data_inflight
) {
3870 * Finding _no_ node in the hash table and no inflight data means that the
3871 * stream(s) have been removed thus data is guaranteed to be available for
3872 * analysis from the trace files.
3876 /* Data is available to be read by a viewer. */
3877 pthread_mutex_unlock(&the_consumer_data
.lock
);
3882 /* Data is still being extracted from buffers. */
3883 pthread_mutex_unlock(&the_consumer_data
.lock
);
3889 * Send a ret code status message to the sessiond daemon.
3891 * Return the sendmsg() return value.
3893 int consumer_send_status_msg(int sock
, int ret_code
)
3895 struct lttcomm_consumer_status_msg msg
;
3897 memset(&msg
, 0, sizeof(msg
));
3898 msg
.ret_code
= (lttcomm_return_code
) ret_code
;
3900 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3904 * Send a channel status message to the sessiond daemon.
3906 * Return the sendmsg() return value.
3908 int consumer_send_status_channel(int sock
,
3909 struct lttng_consumer_channel
*channel
)
3911 struct lttcomm_consumer_status_channel msg
;
3913 LTTNG_ASSERT(sock
>= 0);
3915 memset(&msg
, 0, sizeof(msg
));
3917 msg
.ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
3919 msg
.ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3920 msg
.key
= channel
->key
;
3921 msg
.stream_count
= channel
->streams
.count
;
3924 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3927 unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos
,
3928 unsigned long produced_pos
, uint64_t nb_packets_per_stream
,
3929 uint64_t max_sb_size
)
3931 unsigned long start_pos
;
3933 if (!nb_packets_per_stream
) {
3934 return consumed_pos
; /* Grab everything */
3936 start_pos
= produced_pos
- lttng_offset_align_floor(produced_pos
, max_sb_size
);
3937 start_pos
-= max_sb_size
* nb_packets_per_stream
;
3938 if ((long) (start_pos
- consumed_pos
) < 0) {
3939 return consumed_pos
; /* Grab everything */
3944 /* Stream lock must be held by the caller. */
3945 static int sample_stream_positions(struct lttng_consumer_stream
*stream
,
3946 unsigned long *produced
, unsigned long *consumed
)
3950 ASSERT_LOCKED(stream
->lock
);
3952 ret
= lttng_consumer_sample_snapshot_positions(stream
);
3954 ERR("Failed to sample snapshot positions");
3958 ret
= lttng_consumer_get_produced_snapshot(stream
, produced
);
3960 ERR("Failed to sample produced position");
3964 ret
= lttng_consumer_get_consumed_snapshot(stream
, consumed
);
3966 ERR("Failed to sample consumed position");
3975 * Sample the rotate position for all the streams of a channel. If a stream
3976 * is already at the rotate position (produced == consumed), we flag it as
3977 * ready for rotation. The rotation of ready streams occurs after we have
3978 * replied to the session daemon that we have finished sampling the positions.
3979 * Must be called with RCU read-side lock held to ensure existence of channel.
3981 * Returns 0 on success, < 0 on error
3983 int lttng_consumer_rotate_channel(struct lttng_consumer_channel
*channel
,
3984 uint64_t key
, uint64_t relayd_id
)
3987 struct lttng_consumer_stream
*stream
;
3988 struct lttng_ht_iter iter
;
3989 struct lttng_ht
*ht
= the_consumer_data
.stream_per_chan_id_ht
;
3990 struct lttng_dynamic_array stream_rotation_positions
;
3991 uint64_t next_chunk_id
, stream_count
= 0;
3992 enum lttng_trace_chunk_status chunk_status
;
3993 const bool is_local_trace
= relayd_id
== -1ULL;
3994 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3995 bool rotating_to_new_chunk
= true;
3996 /* Array of `struct lttng_consumer_stream *` */
3997 struct lttng_dynamic_pointer_array streams_packet_to_open
;
4000 ASSERT_RCU_READ_LOCKED();
4002 DBG("Consumer sample rotate position for channel %" PRIu64
, key
);
4004 lttng_dynamic_array_init(&stream_rotation_positions
,
4005 sizeof(struct relayd_stream_rotation_position
), NULL
);
4006 lttng_dynamic_pointer_array_init(&streams_packet_to_open
, NULL
);
4010 pthread_mutex_lock(&channel
->lock
);
4011 LTTNG_ASSERT(channel
->trace_chunk
);
4012 chunk_status
= lttng_trace_chunk_get_id(channel
->trace_chunk
,
4014 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
4016 goto end_unlock_channel
;
4019 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4020 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4021 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4022 stream
, node_channel_id
.node
) {
4023 unsigned long produced_pos
= 0, consumed_pos
= 0;
4025 health_code_update();
4028 * Lock stream because we are about to change its state.
4030 pthread_mutex_lock(&stream
->lock
);
4032 if (stream
->trace_chunk
== stream
->chan
->trace_chunk
) {
4033 rotating_to_new_chunk
= false;
4037 * Do not flush a packet when rotating from a NULL trace
4038 * chunk. The stream has no means to output data, and the prior
4039 * rotation which rotated to NULL performed that side-effect
4040 * already. No new data can be produced when a stream has no
4041 * associated trace chunk (e.g. a stop followed by a rotate).
4043 if (stream
->trace_chunk
) {
4046 if (stream
->metadata_flag
) {
4048 * Don't produce an empty metadata packet,
4049 * simply close the current one.
4051 * Metadata is regenerated on every trace chunk
4052 * switch; there is no concern that no data was
4055 flush_active
= true;
4058 * Only flush an empty packet if the "packet
4059 * open" could not be performed on transition
4060 * to a new trace chunk and no packets were
4061 * consumed within the chunk's lifetime.
4063 if (stream
->opened_packet_in_current_trace_chunk
) {
4064 flush_active
= true;
4067 * Stream could have been full at the
4068 * time of rotation, but then have had
4069 * no activity at all.
4071 * It is important to flush a packet
4072 * to prevent 0-length files from being
4073 * produced as most viewers choke on
4076 * Unfortunately viewers will not be
4077 * able to know that tracing was active
4078 * for this stream during this trace
4081 ret
= sample_stream_positions(stream
, &produced_pos
, &consumed_pos
);
4083 goto end_unlock_stream
;
4087 * Don't flush an empty packet if data
4088 * was produced; it will be consumed
4089 * before the rotation completes.
4091 flush_active
= produced_pos
!= consumed_pos
;
4092 if (!flush_active
) {
4093 const char *trace_chunk_name
;
4094 uint64_t trace_chunk_id
;
4096 chunk_status
= lttng_trace_chunk_get_name(
4097 stream
->trace_chunk
,
4100 if (chunk_status
== LTTNG_TRACE_CHUNK_STATUS_NONE
) {
4101 trace_chunk_name
= "none";
4105 * Consumer trace chunks are
4108 chunk_status
= lttng_trace_chunk_get_id(
4109 stream
->trace_chunk
,
4111 LTTNG_ASSERT(chunk_status
==
4112 LTTNG_TRACE_CHUNK_STATUS_OK
);
4114 DBG("Unable to open packet for stream during trace chunk's lifetime. "
4115 "Flushing an empty packet to prevent an empty file from being created: "
4116 "stream id = %" PRIu64
", trace chunk name = `%s`, trace chunk id = %" PRIu64
,
4117 stream
->key
, trace_chunk_name
, trace_chunk_id
);
4123 * Close the current packet before sampling the
4124 * ring buffer positions.
4126 ret
= consumer_stream_flush_buffer(stream
, flush_active
);
4128 ERR("Failed to flush stream %" PRIu64
" during channel rotation",
4130 goto end_unlock_stream
;
4134 ret
= lttng_consumer_take_snapshot(stream
);
4135 if (ret
< 0 && ret
!= -ENODATA
&& ret
!= -EAGAIN
) {
4136 ERR("Failed to sample snapshot position during channel rotation");
4137 goto end_unlock_stream
;
4140 ret
= lttng_consumer_get_produced_snapshot(stream
,
4143 ERR("Failed to sample produced position during channel rotation");
4144 goto end_unlock_stream
;
4147 ret
= lttng_consumer_get_consumed_snapshot(stream
,
4150 ERR("Failed to sample consumed position during channel rotation");
4151 goto end_unlock_stream
;
4155 * Align produced position on the start-of-packet boundary of the first
4156 * packet going into the next trace chunk.
4158 produced_pos
= lttng_align_floor(produced_pos
, stream
->max_sb_size
);
4159 if (consumed_pos
== produced_pos
) {
4160 DBG("Set rotate ready for stream %" PRIu64
" produced = %lu consumed = %lu",
4161 stream
->key
, produced_pos
, consumed_pos
);
4162 stream
->rotate_ready
= true;
4164 DBG("Different consumed and produced positions "
4165 "for stream %" PRIu64
" produced = %lu consumed = %lu",
4166 stream
->key
, produced_pos
, consumed_pos
);
4169 * The rotation position is based on the packet_seq_num of the
4170 * packet following the last packet that was consumed for this
4171 * stream, incremented by the offset between produced and
4172 * consumed positions. This rotation position is a lower bound
4173 * (inclusive) at which the next trace chunk starts. Since it
4174 * is a lower bound, it is OK if the packet_seq_num does not
4175 * correspond exactly to the same packet identified by the
4176 * consumed_pos, which can happen in overwrite mode.
4178 if (stream
->sequence_number_unavailable
) {
4180 * Rotation should never be performed on a session which
4181 * interacts with a pre-2.8 lttng-modules, which does
4182 * not implement packet sequence number.
4184 ERR("Failure to rotate stream %" PRIu64
": sequence number unavailable",
4187 goto end_unlock_stream
;
4189 stream
->rotate_position
= stream
->last_sequence_number
+ 1 +
4190 ((produced_pos
- consumed_pos
) / stream
->max_sb_size
);
4191 DBG("Set rotation position for stream %" PRIu64
" at position %" PRIu64
,
4192 stream
->key
, stream
->rotate_position
);
4194 if (!is_local_trace
) {
4196 * The relay daemon control protocol expects a rotation
4197 * position as "the sequence number of the first packet
4198 * _after_ the current trace chunk".
4200 const struct relayd_stream_rotation_position position
= {
4201 .stream_id
= stream
->relayd_stream_id
,
4202 .rotate_at_seq_num
= stream
->rotate_position
,
4205 ret
= lttng_dynamic_array_add_element(
4206 &stream_rotation_positions
,
4209 ERR("Failed to allocate stream rotation position");
4210 goto end_unlock_stream
;
4215 stream
->opened_packet_in_current_trace_chunk
= false;
4217 if (rotating_to_new_chunk
&& !stream
->metadata_flag
) {
4219 * Attempt to flush an empty packet as close to the
4220 * rotation point as possible. In the event where a
4221 * stream remains inactive after the rotation point,
4222 * this ensures that the new trace chunk has a
4223 * beginning timestamp set at the begining of the
4224 * trace chunk instead of only creating an empty
4225 * packet when the trace chunk is stopped.
4227 * This indicates to the viewers that the stream
4228 * was being recorded, but more importantly it
4229 * allows viewers to determine a useable trace
4232 * This presents a problem in the case where the
4233 * ring-buffer is completely full.
4235 * Consider the following scenario:
4236 * - The consumption of data is slow (slow network,
4238 * - The ring buffer is full,
4239 * - A rotation is initiated,
4240 * - The flush below does nothing (no space left to
4241 * open a new packet),
4242 * - The other streams rotate very soon, and new
4243 * data is produced in the new chunk,
4244 * - This stream completes its rotation long after the
4245 * rotation was initiated
4246 * - The session is stopped before any event can be
4247 * produced in this stream's buffers.
4249 * The resulting trace chunk will have a single packet
4250 * temporaly at the end of the trace chunk for this
4251 * stream making the stream intersection more narrow
4252 * than it should be.
4254 * To work-around this, an empty flush is performed
4255 * after the first consumption of a packet during a
4256 * rotation if open_packet fails. The idea is that
4257 * consuming a packet frees enough space to switch
4258 * packets in this scenario and allows the tracer to
4259 * "stamp" the beginning of the new trace chunk at the
4260 * earliest possible point.
4262 * The packet open is performed after the channel
4263 * rotation to ensure that no attempt to open a packet
4264 * is performed in a stream that has no active trace
4267 ret
= lttng_dynamic_pointer_array_add_pointer(
4268 &streams_packet_to_open
, stream
);
4270 PERROR("Failed to add a stream pointer to array of streams in which to open a packet");
4272 goto end_unlock_stream
;
4276 pthread_mutex_unlock(&stream
->lock
);
4280 if (!is_local_trace
) {
4281 relayd
= consumer_find_relayd(relayd_id
);
4283 ERR("Failed to find relayd %" PRIu64
, relayd_id
);
4285 goto end_unlock_channel
;
4288 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4289 ret
= relayd_rotate_streams(&relayd
->control_sock
, stream_count
,
4290 rotating_to_new_chunk
? &next_chunk_id
: NULL
,
4291 (const struct relayd_stream_rotation_position
*)
4292 stream_rotation_positions
.buffer
4294 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4296 ERR("Relayd rotate stream failed. Cleaning up relayd %" PRIu64
,
4297 relayd
->net_seq_idx
);
4298 lttng_consumer_cleanup_relayd(relayd
);
4299 goto end_unlock_channel
;
4303 for (stream_idx
= 0;
4304 stream_idx
< lttng_dynamic_pointer_array_get_count(
4305 &streams_packet_to_open
);
4307 enum consumer_stream_open_packet_status status
;
4309 stream
= (lttng_consumer_stream
*) lttng_dynamic_pointer_array_get_pointer(
4310 &streams_packet_to_open
, stream_idx
);
4312 pthread_mutex_lock(&stream
->lock
);
4313 status
= consumer_stream_open_packet(stream
);
4314 pthread_mutex_unlock(&stream
->lock
);
4316 case CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED
:
4317 DBG("Opened a packet after a rotation: stream id = %" PRIu64
4318 ", channel name = %s, session id = %" PRIu64
,
4319 stream
->key
, stream
->chan
->name
,
4320 stream
->chan
->session_id
);
4322 case CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE
:
4324 * Can't open a packet as there is no space left
4325 * in the buffer. A new packet will be opened
4326 * once one has been consumed.
4328 DBG("No space left to open a packet after a rotation: stream id = %" PRIu64
4329 ", channel name = %s, session id = %" PRIu64
,
4330 stream
->key
, stream
->chan
->name
,
4331 stream
->chan
->session_id
);
4333 case CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR
:
4334 /* Logged by callee. */
4336 goto end_unlock_channel
;
4342 pthread_mutex_unlock(&channel
->lock
);
4347 pthread_mutex_unlock(&stream
->lock
);
4349 pthread_mutex_unlock(&channel
->lock
);
4352 lttng_dynamic_array_reset(&stream_rotation_positions
);
4353 lttng_dynamic_pointer_array_reset(&streams_packet_to_open
);
4358 int consumer_clear_buffer(struct lttng_consumer_stream
*stream
)
4361 unsigned long consumed_pos_before
, consumed_pos_after
;
4363 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4365 ERR("Taking snapshot positions");
4369 ret
= lttng_consumer_get_consumed_snapshot(stream
, &consumed_pos_before
);
4371 ERR("Consumed snapshot position");
4375 switch (the_consumer_data
.type
) {
4376 case LTTNG_CONSUMER_KERNEL
:
4377 ret
= kernctl_buffer_clear(stream
->wait_fd
);
4379 ERR("Failed to clear kernel stream (ret = %d)", ret
);
4383 case LTTNG_CONSUMER32_UST
:
4384 case LTTNG_CONSUMER64_UST
:
4385 ret
= lttng_ustconsumer_clear_buffer(stream
);
4387 ERR("Failed to clear ust stream (ret = %d)", ret
);
4392 ERR("Unknown consumer_data type");
4396 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4398 ERR("Taking snapshot positions");
4401 ret
= lttng_consumer_get_consumed_snapshot(stream
, &consumed_pos_after
);
4403 ERR("Consumed snapshot position");
4406 DBG("clear: before: %lu after: %lu", consumed_pos_before
, consumed_pos_after
);
4412 int consumer_clear_stream(struct lttng_consumer_stream
*stream
)
4416 ret
= consumer_stream_flush_buffer(stream
, 1);
4418 ERR("Failed to flush stream %" PRIu64
" during channel clear",
4420 ret
= LTTCOMM_CONSUMERD_FATAL
;
4424 ret
= consumer_clear_buffer(stream
);
4426 ERR("Failed to clear stream %" PRIu64
" during channel clear",
4428 ret
= LTTCOMM_CONSUMERD_FATAL
;
4432 ret
= LTTCOMM_CONSUMERD_SUCCESS
;
4438 int consumer_clear_unmonitored_channel(struct lttng_consumer_channel
*channel
)
4441 struct lttng_consumer_stream
*stream
;
4444 pthread_mutex_lock(&channel
->lock
);
4445 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
4446 health_code_update();
4447 pthread_mutex_lock(&stream
->lock
);
4448 ret
= consumer_clear_stream(stream
);
4452 pthread_mutex_unlock(&stream
->lock
);
4454 pthread_mutex_unlock(&channel
->lock
);
4459 pthread_mutex_unlock(&stream
->lock
);
4460 pthread_mutex_unlock(&channel
->lock
);
4466 * Check if a stream is ready to be rotated after extracting it.
4468 * Return 1 if it is ready for rotation, 0 if it is not, a negative value on
4469 * error. Stream lock must be held.
4471 int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream
*stream
)
4473 DBG("Check is rotate ready for stream %" PRIu64
4474 " ready %u rotate_position %" PRIu64
4475 " last_sequence_number %" PRIu64
,
4476 stream
->key
, stream
->rotate_ready
,
4477 stream
->rotate_position
, stream
->last_sequence_number
);
4478 if (stream
->rotate_ready
) {
4483 * If packet seq num is unavailable, it means we are interacting
4484 * with a pre-2.8 lttng-modules which does not implement the
4485 * sequence number. Rotation should never be used by sessiond in this
4488 if (stream
->sequence_number_unavailable
) {
4489 ERR("Internal error: rotation used on stream %" PRIu64
4490 " with unavailable sequence number",
4495 if (stream
->rotate_position
== -1ULL ||
4496 stream
->last_sequence_number
== -1ULL) {
4501 * Rotate position not reached yet. The stream rotate position is
4502 * the position of the next packet belonging to the next trace chunk,
4503 * but consumerd considers rotation ready when reaching the last
4504 * packet of the current chunk, hence the "rotate_position - 1".
4507 DBG("Check is rotate ready for stream %" PRIu64
4508 " last_sequence_number %" PRIu64
4509 " rotate_position %" PRIu64
,
4510 stream
->key
, stream
->last_sequence_number
,
4511 stream
->rotate_position
);
4512 if (stream
->last_sequence_number
>= stream
->rotate_position
- 1) {
4520 * Reset the state for a stream after a rotation occurred.
4522 void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream
*stream
)
4524 DBG("lttng_consumer_reset_stream_rotate_state for stream %" PRIu64
,
4526 stream
->rotate_position
= -1ULL;
4527 stream
->rotate_ready
= false;
4531 * Perform the rotation a local stream file.
4534 int rotate_local_stream(struct lttng_consumer_stream
*stream
)
4538 DBG("Rotate local stream: stream key %" PRIu64
", channel key %" PRIu64
,
4541 stream
->tracefile_size_current
= 0;
4542 stream
->tracefile_count_current
= 0;
4544 if (stream
->out_fd
>= 0) {
4545 ret
= close(stream
->out_fd
);
4547 PERROR("Failed to close stream out_fd of channel \"%s\"",
4548 stream
->chan
->name
);
4550 stream
->out_fd
= -1;
4553 if (stream
->index_file
) {
4554 lttng_index_file_put(stream
->index_file
);
4555 stream
->index_file
= NULL
;
4558 if (!stream
->trace_chunk
) {
4562 ret
= consumer_stream_create_output_files(stream
, true);
4568 * Performs the stream rotation for the rotate session feature if needed.
4569 * It must be called with the channel and stream locks held.
4571 * Return 0 on success, a negative number of error.
4573 int lttng_consumer_rotate_stream(struct lttng_consumer_stream
*stream
)
4577 DBG("Consumer rotate stream %" PRIu64
, stream
->key
);
4580 * Update the stream's 'current' chunk to the session's (channel)
4581 * now-current chunk.
4583 lttng_trace_chunk_put(stream
->trace_chunk
);
4584 if (stream
->chan
->trace_chunk
== stream
->trace_chunk
) {
4586 * A channel can be rotated and not have a "next" chunk
4587 * to transition to. In that case, the channel's "current chunk"
4588 * has not been closed yet, but it has not been updated to
4589 * a "next" trace chunk either. Hence, the stream, like its
4590 * parent channel, becomes part of no chunk and can't output
4591 * anything until a new trace chunk is created.
4593 stream
->trace_chunk
= NULL
;
4594 } else if (stream
->chan
->trace_chunk
&&
4595 !lttng_trace_chunk_get(stream
->chan
->trace_chunk
)) {
4596 ERR("Failed to acquire a reference to channel's trace chunk during stream rotation");
4601 * Update the stream's trace chunk to its parent channel's
4602 * current trace chunk.
4604 stream
->trace_chunk
= stream
->chan
->trace_chunk
;
4607 if (stream
->net_seq_idx
== (uint64_t) -1ULL) {
4608 ret
= rotate_local_stream(stream
);
4610 ERR("Failed to rotate stream, ret = %i", ret
);
4615 if (stream
->metadata_flag
&& stream
->trace_chunk
) {
4617 * If the stream has transitioned to a new trace
4618 * chunk, the metadata should be re-dumped to the
4621 * However, it is possible for a stream to transition to
4622 * a "no-chunk" state. This can happen if a rotation
4623 * occurs on an inactive session. In such cases, the metadata
4624 * regeneration will happen when the next trace chunk is
4627 ret
= consumer_metadata_stream_dump(stream
);
4632 lttng_consumer_reset_stream_rotate_state(stream
);
4641 * Rotate all the ready streams now.
4643 * This is especially important for low throughput streams that have already
4644 * been consumed, we cannot wait for their next packet to perform the
4646 * Need to be called with RCU read-side lock held to ensure existence of
4649 * Returns 0 on success, < 0 on error
4651 int lttng_consumer_rotate_ready_streams(struct lttng_consumer_channel
*channel
,
4655 struct lttng_consumer_stream
*stream
;
4656 struct lttng_ht_iter iter
;
4657 struct lttng_ht
*ht
= the_consumer_data
.stream_per_chan_id_ht
;
4659 ASSERT_RCU_READ_LOCKED();
4663 DBG("Consumer rotate ready streams in channel %" PRIu64
, key
);
4665 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4666 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4667 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4668 stream
, node_channel_id
.node
) {
4669 health_code_update();
4671 pthread_mutex_lock(&stream
->chan
->lock
);
4672 pthread_mutex_lock(&stream
->lock
);
4674 if (!stream
->rotate_ready
) {
4675 pthread_mutex_unlock(&stream
->lock
);
4676 pthread_mutex_unlock(&stream
->chan
->lock
);
4679 DBG("Consumer rotate ready stream %" PRIu64
, stream
->key
);
4681 ret
= lttng_consumer_rotate_stream(stream
);
4682 pthread_mutex_unlock(&stream
->lock
);
4683 pthread_mutex_unlock(&stream
->chan
->lock
);
4696 enum lttcomm_return_code
lttng_consumer_init_command(
4697 struct lttng_consumer_local_data
*ctx
,
4698 const lttng_uuid sessiond_uuid
)
4700 enum lttcomm_return_code ret
;
4701 char uuid_str
[LTTNG_UUID_STR_LEN
];
4703 if (ctx
->sessiond_uuid
.is_set
) {
4704 ret
= LTTCOMM_CONSUMERD_ALREADY_SET
;
4708 ctx
->sessiond_uuid
.is_set
= true;
4709 memcpy(ctx
->sessiond_uuid
.value
, sessiond_uuid
, sizeof(lttng_uuid
));
4710 ret
= LTTCOMM_CONSUMERD_SUCCESS
;
4711 lttng_uuid_to_str(sessiond_uuid
, uuid_str
);
4712 DBG("Received session daemon UUID: %s", uuid_str
);
4717 enum lttcomm_return_code
lttng_consumer_create_trace_chunk(
4718 const uint64_t *relayd_id
, uint64_t session_id
,
4720 time_t chunk_creation_timestamp
,
4721 const char *chunk_override_name
,
4722 const struct lttng_credentials
*credentials
,
4723 struct lttng_directory_handle
*chunk_directory_handle
)
4726 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
4727 struct lttng_trace_chunk
*created_chunk
= NULL
, *published_chunk
= NULL
;
4728 enum lttng_trace_chunk_status chunk_status
;
4729 char relayd_id_buffer
[MAX_INT_DEC_LEN(*relayd_id
)];
4730 char creation_timestamp_buffer
[ISO8601_STR_LEN
];
4731 const char *relayd_id_str
= "(none)";
4732 const char *creation_timestamp_str
;
4733 struct lttng_ht_iter iter
;
4734 struct lttng_consumer_channel
*channel
;
4737 /* Only used for logging purposes. */
4738 ret
= snprintf(relayd_id_buffer
, sizeof(relayd_id_buffer
),
4739 "%" PRIu64
, *relayd_id
);
4740 if (ret
> 0 && ret
< sizeof(relayd_id_buffer
)) {
4741 relayd_id_str
= relayd_id_buffer
;
4743 relayd_id_str
= "(formatting error)";
4747 /* Local protocol error. */
4748 LTTNG_ASSERT(chunk_creation_timestamp
);
4749 ret
= time_to_iso8601_str(chunk_creation_timestamp
,
4750 creation_timestamp_buffer
,
4751 sizeof(creation_timestamp_buffer
));
4752 creation_timestamp_str
= !ret
? creation_timestamp_buffer
:
4753 "(formatting error)";
4755 DBG("Consumer create trace chunk command: relay_id = %s"
4756 ", session_id = %" PRIu64
", chunk_id = %" PRIu64
4757 ", chunk_override_name = %s"
4758 ", chunk_creation_timestamp = %s",
4759 relayd_id_str
, session_id
, chunk_id
,
4760 chunk_override_name
? : "(none)",
4761 creation_timestamp_str
);
4764 * The trace chunk registry, as used by the consumer daemon, implicitly
4765 * owns the trace chunks. This is only needed in the consumer since
4766 * the consumer has no notion of a session beyond session IDs being
4767 * used to identify other objects.
4769 * The lttng_trace_chunk_registry_publish() call below provides a
4770 * reference which is not released; it implicitly becomes the session
4771 * daemon's reference to the chunk in the consumer daemon.
4773 * The lifetime of trace chunks in the consumer daemon is managed by
4774 * the session daemon through the LTTNG_CONSUMER_CREATE_TRACE_CHUNK
4775 * and LTTNG_CONSUMER_DESTROY_TRACE_CHUNK commands.
4777 created_chunk
= lttng_trace_chunk_create(chunk_id
,
4778 chunk_creation_timestamp
, NULL
);
4779 if (!created_chunk
) {
4780 ERR("Failed to create trace chunk");
4781 ret_code
= LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED
;
4785 if (chunk_override_name
) {
4786 chunk_status
= lttng_trace_chunk_override_name(created_chunk
,
4787 chunk_override_name
);
4788 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
4789 ret_code
= LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED
;
4794 if (chunk_directory_handle
) {
4795 chunk_status
= lttng_trace_chunk_set_credentials(created_chunk
,
4797 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
4798 ERR("Failed to set trace chunk credentials");
4799 ret_code
= LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED
;
4803 * The consumer daemon has no ownership of the chunk output
4806 chunk_status
= lttng_trace_chunk_set_as_user(created_chunk
,
4807 chunk_directory_handle
);
4808 chunk_directory_handle
= NULL
;
4809 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
4810 ERR("Failed to set trace chunk's directory handle");
4811 ret_code
= LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED
;
4816 published_chunk
= lttng_trace_chunk_registry_publish_chunk(
4817 the_consumer_data
.chunk_registry
, session_id
,
4819 lttng_trace_chunk_put(created_chunk
);
4820 created_chunk
= NULL
;
4821 if (!published_chunk
) {
4822 ERR("Failed to publish trace chunk");
4823 ret_code
= LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED
;
4828 cds_lfht_for_each_entry_duplicate(
4829 the_consumer_data
.channels_by_session_id_ht
->ht
,
4830 the_consumer_data
.channels_by_session_id_ht
->hash_fct(
4831 &session_id
, lttng_ht_seed
),
4832 the_consumer_data
.channels_by_session_id_ht
->match_fct
,
4833 &session_id
, &iter
.iter
, channel
,
4834 channels_by_session_id_ht_node
.node
) {
4835 ret
= lttng_consumer_channel_set_trace_chunk(channel
,
4839 * Roll-back the creation of this chunk.
4841 * This is important since the session daemon will
4842 * assume that the creation of this chunk failed and
4843 * will never ask for it to be closed, resulting
4844 * in a leak and an inconsistent state for some
4847 enum lttcomm_return_code close_ret
;
4848 char path
[LTTNG_PATH_MAX
];
4850 DBG("Failed to set new trace chunk on existing channels, rolling back");
4851 close_ret
= lttng_consumer_close_trace_chunk(relayd_id
,
4852 session_id
, chunk_id
,
4853 chunk_creation_timestamp
, NULL
,
4855 if (close_ret
!= LTTCOMM_CONSUMERD_SUCCESS
) {
4856 ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64
", chunk_id = %" PRIu64
,
4857 session_id
, chunk_id
);
4860 ret_code
= LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED
;
4866 struct consumer_relayd_sock_pair
*relayd
;
4868 relayd
= consumer_find_relayd(*relayd_id
);
4870 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4871 ret
= relayd_create_trace_chunk(
4872 &relayd
->control_sock
, published_chunk
);
4873 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4875 ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64
, *relayd_id
);
4878 if (!relayd
|| ret
) {
4879 enum lttcomm_return_code close_ret
;
4880 char path
[LTTNG_PATH_MAX
];
4882 close_ret
= lttng_consumer_close_trace_chunk(relayd_id
,
4885 chunk_creation_timestamp
,
4887 if (close_ret
!= LTTCOMM_CONSUMERD_SUCCESS
) {
4888 ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64
", chunk_id = %" PRIu64
,
4893 ret_code
= LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED
;
4900 /* Release the reference returned by the "publish" operation. */
4901 lttng_trace_chunk_put(published_chunk
);
4902 lttng_trace_chunk_put(created_chunk
);
4906 enum lttcomm_return_code
lttng_consumer_close_trace_chunk(
4907 const uint64_t *relayd_id
, uint64_t session_id
,
4908 uint64_t chunk_id
, time_t chunk_close_timestamp
,
4909 const enum lttng_trace_chunk_command_type
*close_command
,
4912 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
4913 struct lttng_trace_chunk
*chunk
;
4914 char relayd_id_buffer
[MAX_INT_DEC_LEN(*relayd_id
)];
4915 const char *relayd_id_str
= "(none)";
4916 const char *close_command_name
= "none";
4917 struct lttng_ht_iter iter
;
4918 struct lttng_consumer_channel
*channel
;
4919 enum lttng_trace_chunk_status chunk_status
;
4924 /* Only used for logging purposes. */
4925 ret
= snprintf(relayd_id_buffer
, sizeof(relayd_id_buffer
),
4926 "%" PRIu64
, *relayd_id
);
4927 if (ret
> 0 && ret
< sizeof(relayd_id_buffer
)) {
4928 relayd_id_str
= relayd_id_buffer
;
4930 relayd_id_str
= "(formatting error)";
4933 if (close_command
) {
4934 close_command_name
= lttng_trace_chunk_command_type_get_name(
4938 DBG("Consumer close trace chunk command: relayd_id = %s"
4939 ", session_id = %" PRIu64
", chunk_id = %" PRIu64
4940 ", close command = %s",
4941 relayd_id_str
, session_id
, chunk_id
,
4942 close_command_name
);
4944 chunk
= lttng_trace_chunk_registry_find_chunk(
4945 the_consumer_data
.chunk_registry
, session_id
, chunk_id
);
4947 ERR("Failed to find chunk: session_id = %" PRIu64
4948 ", chunk_id = %" PRIu64
,
4949 session_id
, chunk_id
);
4950 ret_code
= LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK
;
4954 chunk_status
= lttng_trace_chunk_set_close_timestamp(chunk
,
4955 chunk_close_timestamp
);
4956 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
4957 ret_code
= LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED
;
4961 if (close_command
) {
4962 chunk_status
= lttng_trace_chunk_set_close_command(
4963 chunk
, *close_command
);
4964 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
4965 ret_code
= LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED
;
4971 * chunk is now invalid to access as we no longer hold a reference to
4972 * it; it is only kept around to compare it (by address) to the
4973 * current chunk found in the session's channels.
4976 cds_lfht_for_each_entry(the_consumer_data
.channel_ht
->ht
, &iter
.iter
,
4977 channel
, node
.node
) {
4981 * Only change the channel's chunk to NULL if it still
4982 * references the chunk being closed. The channel may
4983 * reference a newer channel in the case of a session
4984 * rotation. When a session rotation occurs, the "next"
4985 * chunk is created before the "current" chunk is closed.
4987 if (channel
->trace_chunk
!= chunk
) {
4990 ret
= lttng_consumer_channel_set_trace_chunk(channel
, NULL
);
4993 * Attempt to close the chunk on as many channels as
4996 ret_code
= LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED
;
5002 struct consumer_relayd_sock_pair
*relayd
;
5004 relayd
= consumer_find_relayd(*relayd_id
);
5006 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
5007 ret
= relayd_close_trace_chunk(
5008 &relayd
->control_sock
, chunk
,
5010 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
5012 ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64
,
5016 if (!relayd
|| ret
) {
5017 ret_code
= LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED
;
5025 * Release the reference returned by the "find" operation and
5026 * the session daemon's implicit reference to the chunk.
5028 lttng_trace_chunk_put(chunk
);
5029 lttng_trace_chunk_put(chunk
);
5034 enum lttcomm_return_code
lttng_consumer_trace_chunk_exists(
5035 const uint64_t *relayd_id
, uint64_t session_id
,
5039 enum lttcomm_return_code ret_code
;
5040 char relayd_id_buffer
[MAX_INT_DEC_LEN(*relayd_id
)];
5041 const char *relayd_id_str
= "(none)";
5042 const bool is_local_trace
= !relayd_id
;
5043 struct consumer_relayd_sock_pair
*relayd
= NULL
;
5044 bool chunk_exists_local
, chunk_exists_remote
;
5047 /* Only used for logging purposes. */
5048 ret
= snprintf(relayd_id_buffer
, sizeof(relayd_id_buffer
),
5049 "%" PRIu64
, *relayd_id
);
5050 if (ret
> 0 && ret
< sizeof(relayd_id_buffer
)) {
5051 relayd_id_str
= relayd_id_buffer
;
5053 relayd_id_str
= "(formatting error)";
5057 DBG("Consumer trace chunk exists command: relayd_id = %s"
5058 ", chunk_id = %" PRIu64
, relayd_id_str
,
5060 ret
= lttng_trace_chunk_registry_chunk_exists(
5061 the_consumer_data
.chunk_registry
, session_id
, chunk_id
,
5062 &chunk_exists_local
);
5064 /* Internal error. */
5065 ERR("Failed to query the existence of a trace chunk");
5066 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
5069 DBG("Trace chunk %s locally",
5070 chunk_exists_local
? "exists" : "does not exist");
5071 if (chunk_exists_local
) {
5072 ret_code
= LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL
;
5074 } else if (is_local_trace
) {
5075 ret_code
= LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK
;
5080 relayd
= consumer_find_relayd(*relayd_id
);
5082 ERR("Failed to find relayd %" PRIu64
, *relayd_id
);
5083 ret_code
= LTTCOMM_CONSUMERD_INVALID_PARAMETERS
;
5084 goto end_rcu_unlock
;
5086 DBG("Looking up existence of trace chunk on relay daemon");
5087 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
5088 ret
= relayd_trace_chunk_exists(&relayd
->control_sock
, chunk_id
,
5089 &chunk_exists_remote
);
5090 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
5092 ERR("Failed to look-up the existence of trace chunk on relay daemon");
5093 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
5094 goto end_rcu_unlock
;
5097 ret_code
= chunk_exists_remote
?
5098 LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_REMOTE
:
5099 LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK
;
5100 DBG("Trace chunk %s on relay daemon",
5101 chunk_exists_remote
? "exists" : "does not exist");
5110 int consumer_clear_monitored_channel(struct lttng_consumer_channel
*channel
)
5112 struct lttng_ht
*ht
;
5113 struct lttng_consumer_stream
*stream
;
5114 struct lttng_ht_iter iter
;
5117 ht
= the_consumer_data
.stream_per_chan_id_ht
;
5120 cds_lfht_for_each_entry_duplicate(ht
->ht
,
5121 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
5122 ht
->match_fct
, &channel
->key
,
5123 &iter
.iter
, stream
, node_channel_id
.node
) {
5125 * Protect against teardown with mutex.
5127 pthread_mutex_lock(&stream
->lock
);
5128 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
5131 ret
= consumer_clear_stream(stream
);
5136 pthread_mutex_unlock(&stream
->lock
);
5139 return LTTCOMM_CONSUMERD_SUCCESS
;
5142 pthread_mutex_unlock(&stream
->lock
);
5147 int lttng_consumer_clear_channel(struct lttng_consumer_channel
*channel
)
5151 DBG("Consumer clear channel %" PRIu64
, channel
->key
);
5153 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
5155 * Nothing to do for the metadata channel/stream.
5156 * Snapshot mechanism already take care of the metadata
5157 * handling/generation, and monitored channels only need to
5158 * have their data stream cleared..
5160 ret
= LTTCOMM_CONSUMERD_SUCCESS
;
5164 if (!channel
->monitor
) {
5165 ret
= consumer_clear_unmonitored_channel(channel
);
5167 ret
= consumer_clear_monitored_channel(channel
);
5173 enum lttcomm_return_code
lttng_consumer_open_channel_packets(
5174 struct lttng_consumer_channel
*channel
)
5176 struct lttng_consumer_stream
*stream
;
5177 enum lttcomm_return_code ret
= LTTCOMM_CONSUMERD_SUCCESS
;
5179 if (channel
->metadata_stream
) {
5180 ERR("Open channel packets command attempted on a metadata channel");
5181 ret
= LTTCOMM_CONSUMERD_INVALID_PARAMETERS
;
5186 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
5187 enum consumer_stream_open_packet_status status
;
5189 pthread_mutex_lock(&stream
->lock
);
5190 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
5194 status
= consumer_stream_open_packet(stream
);
5196 case CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED
:
5197 DBG("Opened a packet in \"open channel packets\" command: stream id = %" PRIu64
5198 ", channel name = %s, session id = %" PRIu64
,
5199 stream
->key
, stream
->chan
->name
,
5200 stream
->chan
->session_id
);
5201 stream
->opened_packet_in_current_trace_chunk
= true;
5203 case CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE
:
5204 DBG("No space left to open a packet in \"open channel packets\" command: stream id = %" PRIu64
5205 ", channel name = %s, session id = %" PRIu64
,
5206 stream
->key
, stream
->chan
->name
,
5207 stream
->chan
->session_id
);
5209 case CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR
:
5211 * Only unexpected internal errors can lead to this
5212 * failing. Report an unknown error.
5214 ERR("Failed to flush empty buffer in \"open channel packets\" command: stream id = %" PRIu64
5215 ", channel id = %" PRIu64
5216 ", channel name = %s"
5217 ", session id = %" PRIu64
,
5218 stream
->key
, channel
->key
,
5219 channel
->name
, channel
->session_id
);
5220 ret
= LTTCOMM_CONSUMERD_UNKNOWN_ERROR
;
5227 pthread_mutex_unlock(&stream
->lock
);
5236 pthread_mutex_unlock(&stream
->lock
);
5237 goto end_rcu_unlock
;
5240 void lttng_consumer_sigbus_handle(void *addr
)
5242 lttng_ustconsumer_sigbus_handle(addr
);