2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2012 - David Goulet <dgoulet@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
27 #include <sys/socket.h>
28 #include <sys/types.h>
33 #include <bin/lttng-consumerd/health-consumerd.h>
34 #include <common/common.h>
35 #include <common/utils.h>
36 #include <common/compat/poll.h>
37 #include <common/compat/endian.h>
38 #include <common/index/index.h>
39 #include <common/kernel-ctl/kernel-ctl.h>
40 #include <common/sessiond-comm/relayd.h>
41 #include <common/sessiond-comm/sessiond-comm.h>
42 #include <common/kernel-consumer/kernel-consumer.h>
43 #include <common/relayd/relayd.h>
44 #include <common/ust-consumer/ust-consumer.h>
45 #include <common/consumer/consumer-timer.h>
46 #include <common/consumer/consumer.h>
47 #include <common/consumer/consumer-stream.h>
48 #include <common/consumer/consumer-testpoint.h>
49 #include <common/align.h>
50 #include <common/consumer/consumer-metadata-cache.h>
52 struct lttng_consumer_global_data consumer_data
= {
55 .type
= LTTNG_CONSUMER_UNKNOWN
,
58 enum consumer_channel_action
{
61 CONSUMER_CHANNEL_QUIT
,
64 struct consumer_channel_msg
{
65 enum consumer_channel_action action
;
66 struct lttng_consumer_channel
*chan
; /* add */
67 uint64_t key
; /* del */
70 /* Flag used to temporarily pause data consumption from testpoints. */
71 int data_consumption_paused
;
74 * Flag to inform the polling thread to quit when all fd hung up. Updated by
75 * the consumer_thread_receive_fds when it notices that all fds has hung up.
76 * Also updated by the signal handler (consumer_should_exit()). Read by the
82 * Global hash table containing respectively metadata and data streams. The
83 * stream element in this ht should only be updated by the metadata poll thread
84 * for the metadata and the data poll thread for the data.
86 static struct lttng_ht
*metadata_ht
;
87 static struct lttng_ht
*data_ht
;
90 * Notify a thread lttng pipe to poll back again. This usually means that some
91 * global state has changed so we just send back the thread in a poll wait
94 static void notify_thread_lttng_pipe(struct lttng_pipe
*pipe
)
96 struct lttng_consumer_stream
*null_stream
= NULL
;
100 (void) lttng_pipe_write(pipe
, &null_stream
, sizeof(null_stream
));
103 static void notify_health_quit_pipe(int *pipe
)
107 ret
= lttng_write(pipe
[1], "4", 1);
109 PERROR("write consumer health quit");
113 static void notify_channel_pipe(struct lttng_consumer_local_data
*ctx
,
114 struct lttng_consumer_channel
*chan
,
116 enum consumer_channel_action action
)
118 struct consumer_channel_msg msg
;
121 memset(&msg
, 0, sizeof(msg
));
126 ret
= lttng_write(ctx
->consumer_channel_pipe
[1], &msg
, sizeof(msg
));
127 if (ret
< sizeof(msg
)) {
128 PERROR("notify_channel_pipe write error");
132 void notify_thread_del_channel(struct lttng_consumer_local_data
*ctx
,
135 notify_channel_pipe(ctx
, NULL
, key
, CONSUMER_CHANNEL_DEL
);
138 static int read_channel_pipe(struct lttng_consumer_local_data
*ctx
,
139 struct lttng_consumer_channel
**chan
,
141 enum consumer_channel_action
*action
)
143 struct consumer_channel_msg msg
;
146 ret
= lttng_read(ctx
->consumer_channel_pipe
[0], &msg
, sizeof(msg
));
147 if (ret
< sizeof(msg
)) {
151 *action
= msg
.action
;
159 * Cleanup the stream list of a channel. Those streams are not yet globally
162 static void clean_channel_stream_list(struct lttng_consumer_channel
*channel
)
164 struct lttng_consumer_stream
*stream
, *stmp
;
168 /* Delete streams that might have been left in the stream list. */
169 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
171 cds_list_del(&stream
->send_node
);
173 * Once a stream is added to this list, the buffers were created so we
174 * have a guarantee that this call will succeed. Setting the monitor
175 * mode to 0 so we don't lock nor try to delete the stream from the
179 consumer_stream_destroy(stream
, NULL
);
184 * Find a stream. The consumer_data.lock must be locked during this
187 static struct lttng_consumer_stream
*find_stream(uint64_t key
,
190 struct lttng_ht_iter iter
;
191 struct lttng_ht_node_u64
*node
;
192 struct lttng_consumer_stream
*stream
= NULL
;
196 /* -1ULL keys are lookup failures */
197 if (key
== (uint64_t) -1ULL) {
203 lttng_ht_lookup(ht
, &key
, &iter
);
204 node
= lttng_ht_iter_get_node_u64(&iter
);
206 stream
= caa_container_of(node
, struct lttng_consumer_stream
, node
);
214 static void steal_stream_key(uint64_t key
, struct lttng_ht
*ht
)
216 struct lttng_consumer_stream
*stream
;
219 stream
= find_stream(key
, ht
);
221 stream
->key
= (uint64_t) -1ULL;
223 * We don't want the lookup to match, but we still need
224 * to iterate on this stream when iterating over the hash table. Just
225 * change the node key.
227 stream
->node
.key
= (uint64_t) -1ULL;
233 * Return a channel object for the given key.
235 * RCU read side lock MUST be acquired before calling this function and
236 * protects the channel ptr.
238 struct lttng_consumer_channel
*consumer_find_channel(uint64_t key
)
240 struct lttng_ht_iter iter
;
241 struct lttng_ht_node_u64
*node
;
242 struct lttng_consumer_channel
*channel
= NULL
;
244 /* -1ULL keys are lookup failures */
245 if (key
== (uint64_t) -1ULL) {
249 lttng_ht_lookup(consumer_data
.channel_ht
, &key
, &iter
);
250 node
= lttng_ht_iter_get_node_u64(&iter
);
252 channel
= caa_container_of(node
, struct lttng_consumer_channel
, node
);
259 * There is a possibility that the consumer does not have enough time between
260 * the close of the channel on the session daemon and the cleanup in here thus
261 * once we have a channel add with an existing key, we know for sure that this
262 * channel will eventually get cleaned up by all streams being closed.
264 * This function just nullifies the already existing channel key.
266 static void steal_channel_key(uint64_t key
)
268 struct lttng_consumer_channel
*channel
;
271 channel
= consumer_find_channel(key
);
273 channel
->key
= (uint64_t) -1ULL;
275 * We don't want the lookup to match, but we still need to iterate on
276 * this channel when iterating over the hash table. Just change the
279 channel
->node
.key
= (uint64_t) -1ULL;
284 static void free_channel_rcu(struct rcu_head
*head
)
286 struct lttng_ht_node_u64
*node
=
287 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
288 struct lttng_consumer_channel
*channel
=
289 caa_container_of(node
, struct lttng_consumer_channel
, node
);
291 switch (consumer_data
.type
) {
292 case LTTNG_CONSUMER_KERNEL
:
294 case LTTNG_CONSUMER32_UST
:
295 case LTTNG_CONSUMER64_UST
:
296 lttng_ustconsumer_free_channel(channel
);
299 ERR("Unknown consumer_data type");
306 * RCU protected relayd socket pair free.
308 static void free_relayd_rcu(struct rcu_head
*head
)
310 struct lttng_ht_node_u64
*node
=
311 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
312 struct consumer_relayd_sock_pair
*relayd
=
313 caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
316 * Close all sockets. This is done in the call RCU since we don't want the
317 * socket fds to be reassigned thus potentially creating bad state of the
320 * We do not have to lock the control socket mutex here since at this stage
321 * there is no one referencing to this relayd object.
323 (void) relayd_close(&relayd
->control_sock
);
324 (void) relayd_close(&relayd
->data_sock
);
326 pthread_mutex_destroy(&relayd
->ctrl_sock_mutex
);
331 * Destroy and free relayd socket pair object.
333 void consumer_destroy_relayd(struct consumer_relayd_sock_pair
*relayd
)
336 struct lttng_ht_iter iter
;
338 if (relayd
== NULL
) {
342 DBG("Consumer destroy and close relayd socket pair");
344 iter
.iter
.node
= &relayd
->node
.node
;
345 ret
= lttng_ht_del(consumer_data
.relayd_ht
, &iter
);
347 /* We assume the relayd is being or is destroyed */
351 /* RCU free() call */
352 call_rcu(&relayd
->node
.head
, free_relayd_rcu
);
356 * Remove a channel from the global list protected by a mutex. This function is
357 * also responsible for freeing its data structures.
359 void consumer_del_channel(struct lttng_consumer_channel
*channel
)
362 struct lttng_ht_iter iter
;
364 DBG("Consumer delete channel key %" PRIu64
, channel
->key
);
366 pthread_mutex_lock(&consumer_data
.lock
);
367 pthread_mutex_lock(&channel
->lock
);
369 /* Destroy streams that might have been left in the stream list. */
370 clean_channel_stream_list(channel
);
372 if (channel
->live_timer_enabled
== 1) {
373 consumer_timer_live_stop(channel
);
375 if (channel
->monitor_timer_enabled
== 1) {
376 consumer_timer_monitor_stop(channel
);
379 switch (consumer_data
.type
) {
380 case LTTNG_CONSUMER_KERNEL
:
382 case LTTNG_CONSUMER32_UST
:
383 case LTTNG_CONSUMER64_UST
:
384 lttng_ustconsumer_del_channel(channel
);
387 ERR("Unknown consumer_data type");
393 iter
.iter
.node
= &channel
->node
.node
;
394 ret
= lttng_ht_del(consumer_data
.channel_ht
, &iter
);
397 iter
.iter
.node
= &channel
->channels_by_session_id_ht_node
.node
;
398 ret
= lttng_ht_del(consumer_data
.channels_by_session_id_ht
, &iter
);
402 call_rcu(&channel
->node
.head
, free_channel_rcu
);
404 pthread_mutex_unlock(&channel
->lock
);
405 pthread_mutex_unlock(&consumer_data
.lock
);
409 * Iterate over the relayd hash table and destroy each element. Finally,
410 * destroy the whole hash table.
412 static void cleanup_relayd_ht(void)
414 struct lttng_ht_iter iter
;
415 struct consumer_relayd_sock_pair
*relayd
;
419 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
421 consumer_destroy_relayd(relayd
);
426 lttng_ht_destroy(consumer_data
.relayd_ht
);
430 * Update the end point status of all streams having the given network sequence
431 * index (relayd index).
433 * It's atomically set without having the stream mutex locked which is fine
434 * because we handle the write/read race with a pipe wakeup for each thread.
436 static void update_endpoint_status_by_netidx(uint64_t net_seq_idx
,
437 enum consumer_endpoint_status status
)
439 struct lttng_ht_iter iter
;
440 struct lttng_consumer_stream
*stream
;
442 DBG("Consumer set delete flag on stream by idx %" PRIu64
, net_seq_idx
);
446 /* Let's begin with metadata */
447 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
448 if (stream
->net_seq_idx
== net_seq_idx
) {
449 uatomic_set(&stream
->endpoint_status
, status
);
450 DBG("Delete flag set to metadata stream %d", stream
->wait_fd
);
454 /* Follow up by the data streams */
455 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
456 if (stream
->net_seq_idx
== net_seq_idx
) {
457 uatomic_set(&stream
->endpoint_status
, status
);
458 DBG("Delete flag set to data stream %d", stream
->wait_fd
);
465 * Cleanup a relayd object by flagging every associated streams for deletion,
466 * destroying the object meaning removing it from the relayd hash table,
467 * closing the sockets and freeing the memory in a RCU call.
469 * If a local data context is available, notify the threads that the streams'
470 * state have changed.
472 void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair
*relayd
)
478 DBG("Cleaning up relayd object ID %"PRIu64
, relayd
->net_seq_idx
);
480 /* Save the net sequence index before destroying the object */
481 netidx
= relayd
->net_seq_idx
;
484 * Delete the relayd from the relayd hash table, close the sockets and free
485 * the object in a RCU call.
487 consumer_destroy_relayd(relayd
);
489 /* Set inactive endpoint to all streams */
490 update_endpoint_status_by_netidx(netidx
, CONSUMER_ENDPOINT_INACTIVE
);
493 * With a local data context, notify the threads that the streams' state
494 * have changed. The write() action on the pipe acts as an "implicit"
495 * memory barrier ordering the updates of the end point status from the
496 * read of this status which happens AFTER receiving this notify.
498 notify_thread_lttng_pipe(relayd
->ctx
->consumer_data_pipe
);
499 notify_thread_lttng_pipe(relayd
->ctx
->consumer_metadata_pipe
);
503 * Flag a relayd socket pair for destruction. Destroy it if the refcount
506 * RCU read side lock MUST be aquired before calling this function.
508 void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair
*relayd
)
512 /* Set destroy flag for this object */
513 uatomic_set(&relayd
->destroy_flag
, 1);
515 /* Destroy the relayd if refcount is 0 */
516 if (uatomic_read(&relayd
->refcount
) == 0) {
517 consumer_destroy_relayd(relayd
);
522 * Completly destroy stream from every visiable data structure and the given
525 * One this call returns, the stream object is not longer usable nor visible.
527 void consumer_del_stream(struct lttng_consumer_stream
*stream
,
530 consumer_stream_destroy(stream
, ht
);
534 * XXX naming of del vs destroy is all mixed up.
536 void consumer_del_stream_for_data(struct lttng_consumer_stream
*stream
)
538 consumer_stream_destroy(stream
, data_ht
);
541 void consumer_del_stream_for_metadata(struct lttng_consumer_stream
*stream
)
543 consumer_stream_destroy(stream
, metadata_ht
);
546 void consumer_stream_update_channel_attributes(
547 struct lttng_consumer_stream
*stream
,
548 struct lttng_consumer_channel
*channel
)
550 stream
->channel_read_only_attributes
.tracefile_size
=
551 channel
->tracefile_size
;
552 memcpy(stream
->channel_read_only_attributes
.path
, channel
->pathname
,
553 sizeof(stream
->channel_read_only_attributes
.path
));
556 struct lttng_consumer_stream
*consumer_allocate_stream(uint64_t channel_key
,
558 enum lttng_consumer_stream_state state
,
559 const char *channel_name
,
566 enum consumer_channel_type type
,
567 unsigned int monitor
,
568 uint64_t trace_archive_id
)
571 struct lttng_consumer_stream
*stream
;
573 stream
= zmalloc(sizeof(*stream
));
574 if (stream
== NULL
) {
575 PERROR("malloc struct lttng_consumer_stream");
582 stream
->key
= stream_key
;
584 stream
->out_fd_offset
= 0;
585 stream
->output_written
= 0;
586 stream
->state
= state
;
589 stream
->net_seq_idx
= relayd_id
;
590 stream
->session_id
= session_id
;
591 stream
->monitor
= monitor
;
592 stream
->endpoint_status
= CONSUMER_ENDPOINT_ACTIVE
;
593 stream
->index_file
= NULL
;
594 stream
->last_sequence_number
= -1ULL;
595 stream
->trace_archive_id
= trace_archive_id
;
596 pthread_mutex_init(&stream
->lock
, NULL
);
597 pthread_mutex_init(&stream
->metadata_timer_lock
, NULL
);
599 /* If channel is the metadata, flag this stream as metadata. */
600 if (type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
601 stream
->metadata_flag
= 1;
602 /* Metadata is flat out. */
603 strncpy(stream
->name
, DEFAULT_METADATA_NAME
, sizeof(stream
->name
));
604 /* Live rendez-vous point. */
605 pthread_cond_init(&stream
->metadata_rdv
, NULL
);
606 pthread_mutex_init(&stream
->metadata_rdv_lock
, NULL
);
608 /* Format stream name to <channel_name>_<cpu_number> */
609 ret
= snprintf(stream
->name
, sizeof(stream
->name
), "%s_%d",
612 PERROR("snprintf stream name");
617 /* Key is always the wait_fd for streams. */
618 lttng_ht_node_init_u64(&stream
->node
, stream
->key
);
620 /* Init node per channel id key */
621 lttng_ht_node_init_u64(&stream
->node_channel_id
, channel_key
);
623 /* Init session id node with the stream session id */
624 lttng_ht_node_init_u64(&stream
->node_session_id
, stream
->session_id
);
626 DBG3("Allocated stream %s (key %" PRIu64
", chan_key %" PRIu64
627 " relayd_id %" PRIu64
", session_id %" PRIu64
,
628 stream
->name
, stream
->key
, channel_key
,
629 stream
->net_seq_idx
, stream
->session_id
);
645 * Add a stream to the global list protected by a mutex.
647 void consumer_add_data_stream(struct lttng_consumer_stream
*stream
)
649 struct lttng_ht
*ht
= data_ht
;
654 DBG3("Adding consumer stream %" PRIu64
, stream
->key
);
656 pthread_mutex_lock(&consumer_data
.lock
);
657 pthread_mutex_lock(&stream
->chan
->lock
);
658 pthread_mutex_lock(&stream
->chan
->timer_lock
);
659 pthread_mutex_lock(&stream
->lock
);
662 /* Steal stream identifier to avoid having streams with the same key */
663 steal_stream_key(stream
->key
, ht
);
665 lttng_ht_add_unique_u64(ht
, &stream
->node
);
667 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
668 &stream
->node_channel_id
);
671 * Add stream to the stream_list_ht of the consumer data. No need to steal
672 * the key since the HT does not use it and we allow to add redundant keys
675 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
678 * When nb_init_stream_left reaches 0, we don't need to trigger any action
679 * in terms of destroying the associated channel, because the action that
680 * causes the count to become 0 also causes a stream to be added. The
681 * channel deletion will thus be triggered by the following removal of this
684 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
685 /* Increment refcount before decrementing nb_init_stream_left */
687 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
690 /* Update consumer data once the node is inserted. */
691 consumer_data
.stream_count
++;
692 consumer_data
.need_update
= 1;
695 pthread_mutex_unlock(&stream
->lock
);
696 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
697 pthread_mutex_unlock(&stream
->chan
->lock
);
698 pthread_mutex_unlock(&consumer_data
.lock
);
701 void consumer_del_data_stream(struct lttng_consumer_stream
*stream
)
703 consumer_del_stream(stream
, data_ht
);
707 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
708 * be acquired before calling this.
710 static int add_relayd(struct consumer_relayd_sock_pair
*relayd
)
713 struct lttng_ht_node_u64
*node
;
714 struct lttng_ht_iter iter
;
718 lttng_ht_lookup(consumer_data
.relayd_ht
,
719 &relayd
->net_seq_idx
, &iter
);
720 node
= lttng_ht_iter_get_node_u64(&iter
);
724 lttng_ht_add_unique_u64(consumer_data
.relayd_ht
, &relayd
->node
);
731 * Allocate and return a consumer relayd socket.
733 static struct consumer_relayd_sock_pair
*consumer_allocate_relayd_sock_pair(
734 uint64_t net_seq_idx
)
736 struct consumer_relayd_sock_pair
*obj
= NULL
;
738 /* net sequence index of -1 is a failure */
739 if (net_seq_idx
== (uint64_t) -1ULL) {
743 obj
= zmalloc(sizeof(struct consumer_relayd_sock_pair
));
745 PERROR("zmalloc relayd sock");
749 obj
->net_seq_idx
= net_seq_idx
;
751 obj
->destroy_flag
= 0;
752 obj
->control_sock
.sock
.fd
= -1;
753 obj
->data_sock
.sock
.fd
= -1;
754 lttng_ht_node_init_u64(&obj
->node
, obj
->net_seq_idx
);
755 pthread_mutex_init(&obj
->ctrl_sock_mutex
, NULL
);
762 * Find a relayd socket pair in the global consumer data.
764 * Return the object if found else NULL.
765 * RCU read-side lock must be held across this call and while using the
768 struct consumer_relayd_sock_pair
*consumer_find_relayd(uint64_t key
)
770 struct lttng_ht_iter iter
;
771 struct lttng_ht_node_u64
*node
;
772 struct consumer_relayd_sock_pair
*relayd
= NULL
;
774 /* Negative keys are lookup failures */
775 if (key
== (uint64_t) -1ULL) {
779 lttng_ht_lookup(consumer_data
.relayd_ht
, &key
,
781 node
= lttng_ht_iter_get_node_u64(&iter
);
783 relayd
= caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
791 * Find a relayd and send the stream
793 * Returns 0 on success, < 0 on error
795 int consumer_send_relayd_stream(struct lttng_consumer_stream
*stream
,
799 struct consumer_relayd_sock_pair
*relayd
;
802 assert(stream
->net_seq_idx
!= -1ULL);
805 /* The stream is not metadata. Get relayd reference if exists. */
807 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
808 if (relayd
!= NULL
) {
809 /* Add stream on the relayd */
810 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
811 ret
= relayd_add_stream(&relayd
->control_sock
, stream
->name
,
812 path
, &stream
->relayd_stream_id
,
813 stream
->chan
->tracefile_size
, stream
->chan
->tracefile_count
,
814 stream
->trace_archive_id
);
815 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
817 ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
818 lttng_consumer_cleanup_relayd(relayd
);
822 uatomic_inc(&relayd
->refcount
);
823 stream
->sent_to_relayd
= 1;
825 ERR("Stream %" PRIu64
" relayd ID %" PRIu64
" unknown. Can't send it.",
826 stream
->key
, stream
->net_seq_idx
);
831 DBG("Stream %s with key %" PRIu64
" sent to relayd id %" PRIu64
,
832 stream
->name
, stream
->key
, stream
->net_seq_idx
);
840 * Find a relayd and send the streams sent message
842 * Returns 0 on success, < 0 on error
844 int consumer_send_relayd_streams_sent(uint64_t net_seq_idx
)
847 struct consumer_relayd_sock_pair
*relayd
;
849 assert(net_seq_idx
!= -1ULL);
851 /* The stream is not metadata. Get relayd reference if exists. */
853 relayd
= consumer_find_relayd(net_seq_idx
);
854 if (relayd
!= NULL
) {
855 /* Add stream on the relayd */
856 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
857 ret
= relayd_streams_sent(&relayd
->control_sock
);
858 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
860 ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
861 lttng_consumer_cleanup_relayd(relayd
);
865 ERR("Relayd ID %" PRIu64
" unknown. Can't send streams_sent.",
872 DBG("All streams sent relayd id %" PRIu64
, net_seq_idx
);
880 * Find a relayd and close the stream
882 void close_relayd_stream(struct lttng_consumer_stream
*stream
)
884 struct consumer_relayd_sock_pair
*relayd
;
886 /* The stream is not metadata. Get relayd reference if exists. */
888 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
890 consumer_stream_relayd_close(stream
, relayd
);
896 * Handle stream for relayd transmission if the stream applies for network
897 * streaming where the net sequence index is set.
899 * Return destination file descriptor or negative value on error.
901 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
902 size_t data_size
, unsigned long padding
,
903 struct consumer_relayd_sock_pair
*relayd
)
906 struct lttcomm_relayd_data_hdr data_hdr
;
912 /* Reset data header */
913 memset(&data_hdr
, 0, sizeof(data_hdr
));
915 if (stream
->metadata_flag
) {
916 /* Caller MUST acquire the relayd control socket lock */
917 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
922 /* Metadata are always sent on the control socket. */
923 outfd
= relayd
->control_sock
.sock
.fd
;
925 /* Set header with stream information */
926 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
927 data_hdr
.data_size
= htobe32(data_size
);
928 data_hdr
.padding_size
= htobe32(padding
);
930 * Note that net_seq_num below is assigned with the *current* value of
931 * next_net_seq_num and only after that the next_net_seq_num will be
932 * increment. This is why when issuing a command on the relayd using
933 * this next value, 1 should always be substracted in order to compare
934 * the last seen sequence number on the relayd side to the last sent.
936 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
937 /* Other fields are zeroed previously */
939 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
945 ++stream
->next_net_seq_num
;
947 /* Set to go on data socket */
948 outfd
= relayd
->data_sock
.sock
.fd
;
956 * Allocate and return a new lttng_consumer_channel object using the given key
957 * to initialize the hash table node.
959 * On error, return NULL.
961 struct lttng_consumer_channel
*consumer_allocate_channel(uint64_t key
,
963 const char *pathname
,
968 enum lttng_event_output output
,
969 uint64_t tracefile_size
,
970 uint64_t tracefile_count
,
971 uint64_t session_id_per_pid
,
972 unsigned int monitor
,
973 unsigned int live_timer_interval
,
974 const char *root_shm_path
,
975 const char *shm_path
)
977 struct lttng_consumer_channel
*channel
;
979 channel
= zmalloc(sizeof(*channel
));
980 if (channel
== NULL
) {
981 PERROR("malloc struct lttng_consumer_channel");
986 channel
->refcount
= 0;
987 channel
->session_id
= session_id
;
988 channel
->session_id_per_pid
= session_id_per_pid
;
991 channel
->relayd_id
= relayd_id
;
992 channel
->tracefile_size
= tracefile_size
;
993 channel
->tracefile_count
= tracefile_count
;
994 channel
->monitor
= monitor
;
995 channel
->live_timer_interval
= live_timer_interval
;
996 pthread_mutex_init(&channel
->lock
, NULL
);
997 pthread_mutex_init(&channel
->timer_lock
, NULL
);
1000 case LTTNG_EVENT_SPLICE
:
1001 channel
->output
= CONSUMER_CHANNEL_SPLICE
;
1003 case LTTNG_EVENT_MMAP
:
1004 channel
->output
= CONSUMER_CHANNEL_MMAP
;
1014 * In monitor mode, the streams associated with the channel will be put in
1015 * a special list ONLY owned by this channel. So, the refcount is set to 1
1016 * here meaning that the channel itself has streams that are referenced.
1018 * On a channel deletion, once the channel is no longer visible, the
1019 * refcount is decremented and checked for a zero value to delete it. With
1020 * streams in no monitor mode, it will now be safe to destroy the channel.
1022 if (!channel
->monitor
) {
1023 channel
->refcount
= 1;
1026 strncpy(channel
->pathname
, pathname
, sizeof(channel
->pathname
));
1027 channel
->pathname
[sizeof(channel
->pathname
) - 1] = '\0';
1029 strncpy(channel
->name
, name
, sizeof(channel
->name
));
1030 channel
->name
[sizeof(channel
->name
) - 1] = '\0';
1032 if (root_shm_path
) {
1033 strncpy(channel
->root_shm_path
, root_shm_path
, sizeof(channel
->root_shm_path
));
1034 channel
->root_shm_path
[sizeof(channel
->root_shm_path
) - 1] = '\0';
1037 strncpy(channel
->shm_path
, shm_path
, sizeof(channel
->shm_path
));
1038 channel
->shm_path
[sizeof(channel
->shm_path
) - 1] = '\0';
1041 lttng_ht_node_init_u64(&channel
->node
, channel
->key
);
1042 lttng_ht_node_init_u64(&channel
->channels_by_session_id_ht_node
,
1043 channel
->session_id
);
1045 channel
->wait_fd
= -1;
1047 CDS_INIT_LIST_HEAD(&channel
->streams
.head
);
1049 DBG("Allocated channel (key %" PRIu64
")", channel
->key
);
1056 * Add a channel to the global list protected by a mutex.
1058 * Always return 0 indicating success.
1060 int consumer_add_channel(struct lttng_consumer_channel
*channel
,
1061 struct lttng_consumer_local_data
*ctx
)
1063 pthread_mutex_lock(&consumer_data
.lock
);
1064 pthread_mutex_lock(&channel
->lock
);
1065 pthread_mutex_lock(&channel
->timer_lock
);
1068 * This gives us a guarantee that the channel we are about to add to the
1069 * channel hash table will be unique. See this function comment on the why
1070 * we need to steel the channel key at this stage.
1072 steal_channel_key(channel
->key
);
1075 lttng_ht_add_unique_u64(consumer_data
.channel_ht
, &channel
->node
);
1076 lttng_ht_add_u64(consumer_data
.channels_by_session_id_ht
,
1077 &channel
->channels_by_session_id_ht_node
);
1080 pthread_mutex_unlock(&channel
->timer_lock
);
1081 pthread_mutex_unlock(&channel
->lock
);
1082 pthread_mutex_unlock(&consumer_data
.lock
);
1084 if (channel
->wait_fd
!= -1 && channel
->type
== CONSUMER_CHANNEL_TYPE_DATA
) {
1085 notify_channel_pipe(ctx
, channel
, -1, CONSUMER_CHANNEL_ADD
);
1092 * Allocate the pollfd structure and the local view of the out fds to avoid
1093 * doing a lookup in the linked list and concurrency issues when writing is
1094 * needed. Called with consumer_data.lock held.
1096 * Returns the number of fds in the structures.
1098 static int update_poll_array(struct lttng_consumer_local_data
*ctx
,
1099 struct pollfd
**pollfd
, struct lttng_consumer_stream
**local_stream
,
1100 struct lttng_ht
*ht
, int *nb_inactive_fd
)
1103 struct lttng_ht_iter iter
;
1104 struct lttng_consumer_stream
*stream
;
1109 assert(local_stream
);
1111 DBG("Updating poll fd array");
1112 *nb_inactive_fd
= 0;
1114 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1116 * Only active streams with an active end point can be added to the
1117 * poll set and local stream storage of the thread.
1119 * There is a potential race here for endpoint_status to be updated
1120 * just after the check. However, this is OK since the stream(s) will
1121 * be deleted once the thread is notified that the end point state has
1122 * changed where this function will be called back again.
1124 * We track the number of inactive FDs because they still need to be
1125 * closed by the polling thread after a wakeup on the data_pipe or
1128 if (stream
->state
!= LTTNG_CONSUMER_ACTIVE_STREAM
||
1129 stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
1130 (*nb_inactive_fd
)++;
1134 * This clobbers way too much the debug output. Uncomment that if you
1135 * need it for debugging purposes.
1137 * DBG("Active FD %d", stream->wait_fd);
1139 (*pollfd
)[i
].fd
= stream
->wait_fd
;
1140 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1141 local_stream
[i
] = stream
;
1147 * Insert the consumer_data_pipe at the end of the array and don't
1148 * increment i so nb_fd is the number of real FD.
1150 (*pollfd
)[i
].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_pipe
);
1151 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1153 (*pollfd
)[i
+ 1].fd
= lttng_pipe_get_readfd(ctx
->consumer_wakeup_pipe
);
1154 (*pollfd
)[i
+ 1].events
= POLLIN
| POLLPRI
;
1159 * Poll on the should_quit pipe and the command socket return -1 on
1160 * error, 1 if should exit, 0 if data is available on the command socket
1162 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
1167 num_rdy
= poll(consumer_sockpoll
, 2, -1);
1168 if (num_rdy
== -1) {
1170 * Restart interrupted system call.
1172 if (errno
== EINTR
) {
1175 PERROR("Poll error");
1178 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
1179 DBG("consumer_should_quit wake up");
1186 * Set the error socket.
1188 void lttng_consumer_set_error_sock(struct lttng_consumer_local_data
*ctx
,
1191 ctx
->consumer_error_socket
= sock
;
1195 * Set the command socket path.
1197 void lttng_consumer_set_command_sock_path(
1198 struct lttng_consumer_local_data
*ctx
, char *sock
)
1200 ctx
->consumer_command_sock_path
= sock
;
1204 * Send return code to the session daemon.
1205 * If the socket is not defined, we return 0, it is not a fatal error
1207 int lttng_consumer_send_error(struct lttng_consumer_local_data
*ctx
, int cmd
)
1209 if (ctx
->consumer_error_socket
> 0) {
1210 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1211 sizeof(enum lttcomm_sessiond_command
));
1218 * Close all the tracefiles and stream fds and MUST be called when all
1219 * instances are destroyed i.e. when all threads were joined and are ended.
1221 void lttng_consumer_cleanup(void)
1223 struct lttng_ht_iter iter
;
1224 struct lttng_consumer_channel
*channel
;
1228 cds_lfht_for_each_entry(consumer_data
.channel_ht
->ht
, &iter
.iter
, channel
,
1230 consumer_del_channel(channel
);
1235 lttng_ht_destroy(consumer_data
.channel_ht
);
1236 lttng_ht_destroy(consumer_data
.channels_by_session_id_ht
);
1238 cleanup_relayd_ht();
1240 lttng_ht_destroy(consumer_data
.stream_per_chan_id_ht
);
1243 * This HT contains streams that are freed by either the metadata thread or
1244 * the data thread so we do *nothing* on the hash table and simply destroy
1247 lttng_ht_destroy(consumer_data
.stream_list_ht
);
1251 * Called from signal handler.
1253 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1257 CMM_STORE_SHARED(consumer_quit
, 1);
1258 ret
= lttng_write(ctx
->consumer_should_quit
[1], "4", 1);
1260 PERROR("write consumer quit");
1263 DBG("Consumer flag that it should quit");
1268 * Flush pending writes to trace output disk file.
1271 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1275 int outfd
= stream
->out_fd
;
1278 * This does a blocking write-and-wait on any page that belongs to the
1279 * subbuffer prior to the one we just wrote.
1280 * Don't care about error values, as these are just hints and ways to
1281 * limit the amount of page cache used.
1283 if (orig_offset
< stream
->max_sb_size
) {
1286 lttng_sync_file_range(outfd
, orig_offset
- stream
->max_sb_size
,
1287 stream
->max_sb_size
,
1288 SYNC_FILE_RANGE_WAIT_BEFORE
1289 | SYNC_FILE_RANGE_WRITE
1290 | SYNC_FILE_RANGE_WAIT_AFTER
);
1292 * Give hints to the kernel about how we access the file:
1293 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1296 * We need to call fadvise again after the file grows because the
1297 * kernel does not seem to apply fadvise to non-existing parts of the
1300 * Call fadvise _after_ having waited for the page writeback to
1301 * complete because the dirty page writeback semantic is not well
1302 * defined. So it can be expected to lead to lower throughput in
1305 ret
= posix_fadvise(outfd
, orig_offset
- stream
->max_sb_size
,
1306 stream
->max_sb_size
, POSIX_FADV_DONTNEED
);
1307 if (ret
&& ret
!= -ENOSYS
) {
1309 PERROR("posix_fadvise on fd %i", outfd
);
1314 * Initialise the necessary environnement :
1315 * - create a new context
1316 * - create the poll_pipe
1317 * - create the should_quit pipe (for signal handler)
1318 * - create the thread pipe (for splice)
1320 * Takes a function pointer as argument, this function is called when data is
1321 * available on a buffer. This function is responsible to do the
1322 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1323 * buffer configuration and then kernctl_put_next_subbuf at the end.
1325 * Returns a pointer to the new context or NULL on error.
1327 struct lttng_consumer_local_data
*lttng_consumer_create(
1328 enum lttng_consumer_type type
,
1329 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1330 struct lttng_consumer_local_data
*ctx
),
1331 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1332 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1333 int (*update_stream
)(uint64_t stream_key
, uint32_t state
))
1336 struct lttng_consumer_local_data
*ctx
;
1338 assert(consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1339 consumer_data
.type
== type
);
1340 consumer_data
.type
= type
;
1342 ctx
= zmalloc(sizeof(struct lttng_consumer_local_data
));
1344 PERROR("allocating context");
1348 ctx
->consumer_error_socket
= -1;
1349 ctx
->consumer_metadata_socket
= -1;
1350 pthread_mutex_init(&ctx
->metadata_socket_lock
, NULL
);
1351 /* assign the callbacks */
1352 ctx
->on_buffer_ready
= buffer_ready
;
1353 ctx
->on_recv_channel
= recv_channel
;
1354 ctx
->on_recv_stream
= recv_stream
;
1355 ctx
->on_update_stream
= update_stream
;
1357 ctx
->consumer_data_pipe
= lttng_pipe_open(0);
1358 if (!ctx
->consumer_data_pipe
) {
1359 goto error_poll_pipe
;
1362 ctx
->consumer_wakeup_pipe
= lttng_pipe_open(0);
1363 if (!ctx
->consumer_wakeup_pipe
) {
1364 goto error_wakeup_pipe
;
1367 ret
= pipe(ctx
->consumer_should_quit
);
1369 PERROR("Error creating recv pipe");
1370 goto error_quit_pipe
;
1373 ret
= pipe(ctx
->consumer_channel_pipe
);
1375 PERROR("Error creating channel pipe");
1376 goto error_channel_pipe
;
1379 ctx
->consumer_metadata_pipe
= lttng_pipe_open(0);
1380 if (!ctx
->consumer_metadata_pipe
) {
1381 goto error_metadata_pipe
;
1384 ctx
->channel_monitor_pipe
= -1;
1388 error_metadata_pipe
:
1389 utils_close_pipe(ctx
->consumer_channel_pipe
);
1391 utils_close_pipe(ctx
->consumer_should_quit
);
1393 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1395 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1403 * Iterate over all streams of the hashtable and free them properly.
1405 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1407 struct lttng_ht_iter iter
;
1408 struct lttng_consumer_stream
*stream
;
1415 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1417 * Ignore return value since we are currently cleaning up so any error
1420 (void) consumer_del_stream(stream
, ht
);
1424 lttng_ht_destroy(ht
);
1428 * Iterate over all streams of the metadata hashtable and free them
1431 static void destroy_metadata_stream_ht(struct lttng_ht
*ht
)
1433 struct lttng_ht_iter iter
;
1434 struct lttng_consumer_stream
*stream
;
1441 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1443 * Ignore return value since we are currently cleaning up so any error
1446 (void) consumer_del_metadata_stream(stream
, ht
);
1450 lttng_ht_destroy(ht
);
1454 * Close all fds associated with the instance and free the context.
1456 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1460 DBG("Consumer destroying it. Closing everything.");
1466 destroy_data_stream_ht(data_ht
);
1467 destroy_metadata_stream_ht(metadata_ht
);
1469 ret
= close(ctx
->consumer_error_socket
);
1473 ret
= close(ctx
->consumer_metadata_socket
);
1477 utils_close_pipe(ctx
->consumer_channel_pipe
);
1478 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1479 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1480 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1481 utils_close_pipe(ctx
->consumer_should_quit
);
1483 unlink(ctx
->consumer_command_sock_path
);
1488 * Write the metadata stream id on the specified file descriptor.
1490 static int write_relayd_metadata_id(int fd
,
1491 struct lttng_consumer_stream
*stream
,
1492 unsigned long padding
)
1495 struct lttcomm_relayd_metadata_payload hdr
;
1497 hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
1498 hdr
.padding_size
= htobe32(padding
);
1499 ret
= lttng_write(fd
, (void *) &hdr
, sizeof(hdr
));
1500 if (ret
< sizeof(hdr
)) {
1502 * This error means that the fd's end is closed so ignore the PERROR
1503 * not to clubber the error output since this can happen in a normal
1506 if (errno
!= EPIPE
) {
1507 PERROR("write metadata stream id");
1509 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno
);
1511 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1512 * handle writting the missing part so report that as an error and
1513 * don't lie to the caller.
1518 DBG("Metadata stream id %" PRIu64
" with padding %lu written before data",
1519 stream
->relayd_stream_id
, padding
);
1526 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1527 * core function for writing trace buffers to either the local filesystem or
1530 * It must be called with the stream lock held.
1532 * Careful review MUST be put if any changes occur!
1534 * Returns the number of bytes written
1536 ssize_t
lttng_consumer_on_read_subbuffer_mmap(
1537 struct lttng_consumer_local_data
*ctx
,
1538 struct lttng_consumer_stream
*stream
, unsigned long len
,
1539 unsigned long padding
,
1540 struct ctf_packet_index
*index
)
1542 unsigned long mmap_offset
;
1545 off_t orig_offset
= stream
->out_fd_offset
;
1546 /* Default is on the disk */
1547 int outfd
= stream
->out_fd
;
1548 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1549 unsigned int relayd_hang_up
= 0;
1551 /* RCU lock for the relayd pointer */
1554 /* Flag that the current stream if set for network streaming. */
1555 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1556 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1557 if (relayd
== NULL
) {
1563 /* get the offset inside the fd to mmap */
1564 switch (consumer_data
.type
) {
1565 case LTTNG_CONSUMER_KERNEL
:
1566 mmap_base
= stream
->mmap_base
;
1567 ret
= kernctl_get_mmap_read_offset(stream
->wait_fd
, &mmap_offset
);
1569 PERROR("tracer ctl get_mmap_read_offset");
1573 case LTTNG_CONSUMER32_UST
:
1574 case LTTNG_CONSUMER64_UST
:
1575 mmap_base
= lttng_ustctl_get_mmap_base(stream
);
1577 ERR("read mmap get mmap base for stream %s", stream
->name
);
1581 ret
= lttng_ustctl_get_mmap_read_offset(stream
, &mmap_offset
);
1583 PERROR("tracer ctl get_mmap_read_offset");
1589 ERR("Unknown consumer_data type");
1593 /* Handle stream on the relayd if the output is on the network */
1595 unsigned long netlen
= len
;
1598 * Lock the control socket for the complete duration of the function
1599 * since from this point on we will use the socket.
1601 if (stream
->metadata_flag
) {
1602 /* Metadata requires the control socket. */
1603 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1604 if (stream
->reset_metadata_flag
) {
1605 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1606 stream
->relayd_stream_id
,
1607 stream
->metadata_version
);
1612 stream
->reset_metadata_flag
= 0;
1614 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1617 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1622 /* Use the returned socket. */
1625 /* Write metadata stream id before payload */
1626 if (stream
->metadata_flag
) {
1627 ret
= write_relayd_metadata_id(outfd
, stream
, padding
);
1634 /* No streaming, we have to set the len with the full padding */
1637 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1638 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1640 ERR("Reset metadata file");
1643 stream
->reset_metadata_flag
= 0;
1647 * Check if we need to change the tracefile before writing the packet.
1649 if (stream
->chan
->tracefile_size
> 0 &&
1650 (stream
->tracefile_size_current
+ len
) >
1651 stream
->chan
->tracefile_size
) {
1652 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1653 stream
->name
, stream
->chan
->tracefile_size
,
1654 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1655 stream
->out_fd
, &(stream
->tracefile_count_current
),
1658 ERR("Rotating output file");
1661 outfd
= stream
->out_fd
;
1663 if (stream
->index_file
) {
1664 lttng_index_file_put(stream
->index_file
);
1665 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1666 stream
->name
, stream
->uid
, stream
->gid
,
1667 stream
->chan
->tracefile_size
,
1668 stream
->tracefile_count_current
,
1669 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1670 if (!stream
->index_file
) {
1675 /* Reset current size because we just perform a rotation. */
1676 stream
->tracefile_size_current
= 0;
1677 stream
->out_fd_offset
= 0;
1680 stream
->tracefile_size_current
+= len
;
1682 index
->offset
= htobe64(stream
->out_fd_offset
);
1687 * This call guarantee that len or less is returned. It's impossible to
1688 * receive a ret value that is bigger than len.
1690 ret
= lttng_write(outfd
, mmap_base
+ mmap_offset
, len
);
1691 DBG("Consumer mmap write() ret %zd (len %lu)", ret
, len
);
1692 if (ret
< 0 || ((size_t) ret
!= len
)) {
1694 * Report error to caller if nothing was written else at least send the
1702 /* Socket operation failed. We consider the relayd dead */
1703 if (errno
== EPIPE
|| errno
== EINVAL
|| errno
== EBADF
) {
1705 * This is possible if the fd is closed on the other side
1706 * (outfd) or any write problem. It can be verbose a bit for a
1707 * normal execution if for instance the relayd is stopped
1708 * abruptly. This can happen so set this to a DBG statement.
1710 DBG("Consumer mmap write detected relayd hang up");
1712 /* Unhandled error, print it and stop function right now. */
1713 PERROR("Error in write mmap (ret %zd != len %lu)", ret
, len
);
1717 stream
->output_written
+= ret
;
1719 /* This call is useless on a socket so better save a syscall. */
1721 /* This won't block, but will start writeout asynchronously */
1722 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, len
,
1723 SYNC_FILE_RANGE_WRITE
);
1724 stream
->out_fd_offset
+= len
;
1725 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1730 * This is a special case that the relayd has closed its socket. Let's
1731 * cleanup the relayd object and all associated streams.
1733 if (relayd
&& relayd_hang_up
) {
1734 ERR("Relayd hangup. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
1735 lttng_consumer_cleanup_relayd(relayd
);
1739 /* Unlock only if ctrl socket used */
1740 if (relayd
&& stream
->metadata_flag
) {
1741 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1749 * Splice the data from the ring buffer to the tracefile.
1751 * It must be called with the stream lock held.
1753 * Returns the number of bytes spliced.
1755 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1756 struct lttng_consumer_local_data
*ctx
,
1757 struct lttng_consumer_stream
*stream
, unsigned long len
,
1758 unsigned long padding
,
1759 struct ctf_packet_index
*index
)
1761 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1763 off_t orig_offset
= stream
->out_fd_offset
;
1764 int fd
= stream
->wait_fd
;
1765 /* Default is on the disk */
1766 int outfd
= stream
->out_fd
;
1767 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1769 unsigned int relayd_hang_up
= 0;
1771 switch (consumer_data
.type
) {
1772 case LTTNG_CONSUMER_KERNEL
:
1774 case LTTNG_CONSUMER32_UST
:
1775 case LTTNG_CONSUMER64_UST
:
1776 /* Not supported for user space tracing */
1779 ERR("Unknown consumer_data type");
1783 /* RCU lock for the relayd pointer */
1786 /* Flag that the current stream if set for network streaming. */
1787 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1788 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1789 if (relayd
== NULL
) {
1794 splice_pipe
= stream
->splice_pipe
;
1796 /* Write metadata stream id before payload */
1798 unsigned long total_len
= len
;
1800 if (stream
->metadata_flag
) {
1802 * Lock the control socket for the complete duration of the function
1803 * since from this point on we will use the socket.
1805 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1807 if (stream
->reset_metadata_flag
) {
1808 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1809 stream
->relayd_stream_id
,
1810 stream
->metadata_version
);
1815 stream
->reset_metadata_flag
= 0;
1817 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
,
1825 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1828 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1834 /* Use the returned socket. */
1837 /* No streaming, we have to set the len with the full padding */
1840 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1841 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1843 ERR("Reset metadata file");
1846 stream
->reset_metadata_flag
= 0;
1849 * Check if we need to change the tracefile before writing the packet.
1851 if (stream
->chan
->tracefile_size
> 0 &&
1852 (stream
->tracefile_size_current
+ len
) >
1853 stream
->chan
->tracefile_size
) {
1854 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1855 stream
->name
, stream
->chan
->tracefile_size
,
1856 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1857 stream
->out_fd
, &(stream
->tracefile_count_current
),
1861 ERR("Rotating output file");
1864 outfd
= stream
->out_fd
;
1866 if (stream
->index_file
) {
1867 lttng_index_file_put(stream
->index_file
);
1868 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1869 stream
->name
, stream
->uid
, stream
->gid
,
1870 stream
->chan
->tracefile_size
,
1871 stream
->tracefile_count_current
,
1872 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1873 if (!stream
->index_file
) {
1878 /* Reset current size because we just perform a rotation. */
1879 stream
->tracefile_size_current
= 0;
1880 stream
->out_fd_offset
= 0;
1883 stream
->tracefile_size_current
+= len
;
1884 index
->offset
= htobe64(stream
->out_fd_offset
);
1888 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1889 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1890 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1891 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1892 DBG("splice chan to pipe, ret %zd", ret_splice
);
1893 if (ret_splice
< 0) {
1896 PERROR("Error in relay splice");
1900 /* Handle stream on the relayd if the output is on the network */
1901 if (relayd
&& stream
->metadata_flag
) {
1902 size_t metadata_payload_size
=
1903 sizeof(struct lttcomm_relayd_metadata_payload
);
1905 /* Update counter to fit the spliced data */
1906 ret_splice
+= metadata_payload_size
;
1907 len
+= metadata_payload_size
;
1909 * We do this so the return value can match the len passed as
1910 * argument to this function.
1912 written
-= metadata_payload_size
;
1915 /* Splice data out */
1916 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1917 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1918 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1920 if (ret_splice
< 0) {
1925 } else if (ret_splice
> len
) {
1927 * We don't expect this code path to be executed but you never know
1928 * so this is an extra protection agains a buggy splice().
1931 written
+= ret_splice
;
1932 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice
,
1936 /* All good, update current len and continue. */
1940 /* This call is useless on a socket so better save a syscall. */
1942 /* This won't block, but will start writeout asynchronously */
1943 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1944 SYNC_FILE_RANGE_WRITE
);
1945 stream
->out_fd_offset
+= ret_splice
;
1947 stream
->output_written
+= ret_splice
;
1948 written
+= ret_splice
;
1951 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1957 * This is a special case that the relayd has closed its socket. Let's
1958 * cleanup the relayd object and all associated streams.
1960 if (relayd
&& relayd_hang_up
) {
1961 ERR("Relayd hangup. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
1962 lttng_consumer_cleanup_relayd(relayd
);
1963 /* Skip splice error so the consumer does not fail */
1968 /* send the appropriate error description to sessiond */
1971 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1974 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1977 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1982 if (relayd
&& stream
->metadata_flag
) {
1983 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1991 * Sample the snapshot positions for a specific fd
1993 * Returns 0 on success, < 0 on error
1995 int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream
*stream
)
1997 switch (consumer_data
.type
) {
1998 case LTTNG_CONSUMER_KERNEL
:
1999 return lttng_kconsumer_sample_snapshot_positions(stream
);
2000 case LTTNG_CONSUMER32_UST
:
2001 case LTTNG_CONSUMER64_UST
:
2002 return lttng_ustconsumer_sample_snapshot_positions(stream
);
2004 ERR("Unknown consumer_data type");
2010 * Take a snapshot for a specific fd
2012 * Returns 0 on success, < 0 on error
2014 int lttng_consumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2016 switch (consumer_data
.type
) {
2017 case LTTNG_CONSUMER_KERNEL
:
2018 return lttng_kconsumer_take_snapshot(stream
);
2019 case LTTNG_CONSUMER32_UST
:
2020 case LTTNG_CONSUMER64_UST
:
2021 return lttng_ustconsumer_take_snapshot(stream
);
2023 ERR("Unknown consumer_data type");
2030 * Get the produced position
2032 * Returns 0 on success, < 0 on error
2034 int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
2037 switch (consumer_data
.type
) {
2038 case LTTNG_CONSUMER_KERNEL
:
2039 return lttng_kconsumer_get_produced_snapshot(stream
, pos
);
2040 case LTTNG_CONSUMER32_UST
:
2041 case LTTNG_CONSUMER64_UST
:
2042 return lttng_ustconsumer_get_produced_snapshot(stream
, pos
);
2044 ERR("Unknown consumer_data type");
2051 * Get the consumed position (free-running counter position in bytes).
2053 * Returns 0 on success, < 0 on error
2055 int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
2058 switch (consumer_data
.type
) {
2059 case LTTNG_CONSUMER_KERNEL
:
2060 return lttng_kconsumer_get_consumed_snapshot(stream
, pos
);
2061 case LTTNG_CONSUMER32_UST
:
2062 case LTTNG_CONSUMER64_UST
:
2063 return lttng_ustconsumer_get_consumed_snapshot(stream
, pos
);
2065 ERR("Unknown consumer_data type");
2071 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
2072 int sock
, struct pollfd
*consumer_sockpoll
)
2074 switch (consumer_data
.type
) {
2075 case LTTNG_CONSUMER_KERNEL
:
2076 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2077 case LTTNG_CONSUMER32_UST
:
2078 case LTTNG_CONSUMER64_UST
:
2079 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2081 ERR("Unknown consumer_data type");
2087 void lttng_consumer_close_all_metadata(void)
2089 switch (consumer_data
.type
) {
2090 case LTTNG_CONSUMER_KERNEL
:
2092 * The Kernel consumer has a different metadata scheme so we don't
2093 * close anything because the stream will be closed by the session
2097 case LTTNG_CONSUMER32_UST
:
2098 case LTTNG_CONSUMER64_UST
:
2100 * Close all metadata streams. The metadata hash table is passed and
2101 * this call iterates over it by closing all wakeup fd. This is safe
2102 * because at this point we are sure that the metadata producer is
2103 * either dead or blocked.
2105 lttng_ustconsumer_close_all_metadata(metadata_ht
);
2108 ERR("Unknown consumer_data type");
2114 * Clean up a metadata stream and free its memory.
2116 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
2117 struct lttng_ht
*ht
)
2119 struct lttng_consumer_channel
*free_chan
= NULL
;
2123 * This call should NEVER receive regular stream. It must always be
2124 * metadata stream and this is crucial for data structure synchronization.
2126 assert(stream
->metadata_flag
);
2128 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
2130 pthread_mutex_lock(&consumer_data
.lock
);
2131 pthread_mutex_lock(&stream
->chan
->lock
);
2132 pthread_mutex_lock(&stream
->lock
);
2133 if (stream
->chan
->metadata_cache
) {
2134 /* Only applicable to userspace consumers. */
2135 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2138 /* Remove any reference to that stream. */
2139 consumer_stream_delete(stream
, ht
);
2141 /* Close down everything including the relayd if one. */
2142 consumer_stream_close(stream
);
2143 /* Destroy tracer buffers of the stream. */
2144 consumer_stream_destroy_buffers(stream
);
2146 /* Atomically decrement channel refcount since other threads can use it. */
2147 if (!uatomic_sub_return(&stream
->chan
->refcount
, 1)
2148 && !uatomic_read(&stream
->chan
->nb_init_stream_left
)) {
2149 /* Go for channel deletion! */
2150 free_chan
= stream
->chan
;
2154 * Nullify the stream reference so it is not used after deletion. The
2155 * channel lock MUST be acquired before being able to check for a NULL
2158 stream
->chan
->metadata_stream
= NULL
;
2160 if (stream
->chan
->metadata_cache
) {
2161 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2163 pthread_mutex_unlock(&stream
->lock
);
2164 pthread_mutex_unlock(&stream
->chan
->lock
);
2165 pthread_mutex_unlock(&consumer_data
.lock
);
2168 consumer_del_channel(free_chan
);
2171 consumer_stream_free(stream
);
2175 * Action done with the metadata stream when adding it to the consumer internal
2176 * data structures to handle it.
2178 void consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
)
2180 struct lttng_ht
*ht
= metadata_ht
;
2181 struct lttng_ht_iter iter
;
2182 struct lttng_ht_node_u64
*node
;
2187 DBG3("Adding metadata stream %" PRIu64
" to hash table", stream
->key
);
2189 pthread_mutex_lock(&consumer_data
.lock
);
2190 pthread_mutex_lock(&stream
->chan
->lock
);
2191 pthread_mutex_lock(&stream
->chan
->timer_lock
);
2192 pthread_mutex_lock(&stream
->lock
);
2195 * From here, refcounts are updated so be _careful_ when returning an error
2202 * Lookup the stream just to make sure it does not exist in our internal
2203 * state. This should NEVER happen.
2205 lttng_ht_lookup(ht
, &stream
->key
, &iter
);
2206 node
= lttng_ht_iter_get_node_u64(&iter
);
2210 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2211 * in terms of destroying the associated channel, because the action that
2212 * causes the count to become 0 also causes a stream to be added. The
2213 * channel deletion will thus be triggered by the following removal of this
2216 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
2217 /* Increment refcount before decrementing nb_init_stream_left */
2219 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
2222 lttng_ht_add_unique_u64(ht
, &stream
->node
);
2224 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
2225 &stream
->node_channel_id
);
2228 * Add stream to the stream_list_ht of the consumer data. No need to steal
2229 * the key since the HT does not use it and we allow to add redundant keys
2232 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
2236 pthread_mutex_unlock(&stream
->lock
);
2237 pthread_mutex_unlock(&stream
->chan
->lock
);
2238 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
2239 pthread_mutex_unlock(&consumer_data
.lock
);
2243 * Delete data stream that are flagged for deletion (endpoint_status).
2245 static void validate_endpoint_status_data_stream(void)
2247 struct lttng_ht_iter iter
;
2248 struct lttng_consumer_stream
*stream
;
2250 DBG("Consumer delete flagged data stream");
2253 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2254 /* Validate delete flag of the stream */
2255 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2258 /* Delete it right now */
2259 consumer_del_stream(stream
, data_ht
);
2265 * Delete metadata stream that are flagged for deletion (endpoint_status).
2267 static void validate_endpoint_status_metadata_stream(
2268 struct lttng_poll_event
*pollset
)
2270 struct lttng_ht_iter iter
;
2271 struct lttng_consumer_stream
*stream
;
2273 DBG("Consumer delete flagged metadata stream");
2278 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2279 /* Validate delete flag of the stream */
2280 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2284 * Remove from pollset so the metadata thread can continue without
2285 * blocking on a deleted stream.
2287 lttng_poll_del(pollset
, stream
->wait_fd
);
2289 /* Delete it right now */
2290 consumer_del_metadata_stream(stream
, metadata_ht
);
2296 * Perform operations that need to be done after a stream has
2297 * rotated and released the stream lock.
2299 * Multiple rotations cannot occur simultaneously, so we know the state of the
2300 * "rotated" stream flag cannot change.
2302 * This MUST be called WITHOUT the stream lock held.
2305 int consumer_post_rotation(struct lttng_consumer_stream
*stream
,
2306 struct lttng_consumer_local_data
*ctx
)
2310 pthread_mutex_lock(&stream
->chan
->lock
);
2312 switch (consumer_data
.type
) {
2313 case LTTNG_CONSUMER_KERNEL
:
2315 case LTTNG_CONSUMER32_UST
:
2316 case LTTNG_CONSUMER64_UST
:
2318 * The ust_metadata_pushed counter has been reset to 0, so now
2319 * we can wakeup the metadata thread so it dumps the metadata
2320 * cache to the new file.
2322 if (stream
->metadata_flag
) {
2323 consumer_metadata_wakeup_pipe(stream
->chan
);
2327 ERR("Unknown consumer_data type");
2331 pthread_mutex_unlock(&stream
->chan
->lock
);
2336 * Thread polls on metadata file descriptor and write them on disk or on the
2339 void *consumer_thread_metadata_poll(void *data
)
2341 int ret
, i
, pollfd
, err
= -1;
2342 uint32_t revents
, nb_fd
;
2343 struct lttng_consumer_stream
*stream
= NULL
;
2344 struct lttng_ht_iter iter
;
2345 struct lttng_ht_node_u64
*node
;
2346 struct lttng_poll_event events
;
2347 struct lttng_consumer_local_data
*ctx
= data
;
2350 rcu_register_thread();
2352 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_METADATA
);
2354 if (testpoint(consumerd_thread_metadata
)) {
2355 goto error_testpoint
;
2358 health_code_update();
2360 DBG("Thread metadata poll started");
2362 /* Size is set to 1 for the consumer_metadata pipe */
2363 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2365 ERR("Poll set creation failed");
2369 ret
= lttng_poll_add(&events
,
2370 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
), LPOLLIN
);
2376 DBG("Metadata main loop started");
2380 health_code_update();
2381 health_poll_entry();
2382 DBG("Metadata poll wait");
2383 ret
= lttng_poll_wait(&events
, -1);
2384 DBG("Metadata poll return from wait with %d fd(s)",
2385 LTTNG_POLL_GETNB(&events
));
2387 DBG("Metadata event caught in thread");
2389 if (errno
== EINTR
) {
2390 ERR("Poll EINTR caught");
2393 if (LTTNG_POLL_GETNB(&events
) == 0) {
2394 err
= 0; /* All is OK */
2401 /* From here, the event is a metadata wait fd */
2402 for (i
= 0; i
< nb_fd
; i
++) {
2403 health_code_update();
2405 revents
= LTTNG_POLL_GETEV(&events
, i
);
2406 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2408 if (pollfd
== lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
)) {
2409 if (revents
& LPOLLIN
) {
2412 pipe_len
= lttng_pipe_read(ctx
->consumer_metadata_pipe
,
2413 &stream
, sizeof(stream
));
2414 if (pipe_len
< sizeof(stream
)) {
2416 PERROR("read metadata stream");
2419 * Remove the pipe from the poll set and continue the loop
2420 * since their might be data to consume.
2422 lttng_poll_del(&events
,
2423 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2424 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2428 /* A NULL stream means that the state has changed. */
2429 if (stream
== NULL
) {
2430 /* Check for deleted streams. */
2431 validate_endpoint_status_metadata_stream(&events
);
2435 DBG("Adding metadata stream %d to poll set",
2438 /* Add metadata stream to the global poll events list */
2439 lttng_poll_add(&events
, stream
->wait_fd
,
2440 LPOLLIN
| LPOLLPRI
| LPOLLHUP
);
2441 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2442 DBG("Metadata thread pipe hung up");
2444 * Remove the pipe from the poll set and continue the loop
2445 * since their might be data to consume.
2447 lttng_poll_del(&events
,
2448 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2449 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2452 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2456 /* Handle other stream */
2462 uint64_t tmp_id
= (uint64_t) pollfd
;
2464 lttng_ht_lookup(metadata_ht
, &tmp_id
, &iter
);
2466 node
= lttng_ht_iter_get_node_u64(&iter
);
2469 stream
= caa_container_of(node
, struct lttng_consumer_stream
,
2472 if (revents
& (LPOLLIN
| LPOLLPRI
)) {
2473 /* Get the data out of the metadata file descriptor */
2474 DBG("Metadata available on fd %d", pollfd
);
2475 assert(stream
->wait_fd
== pollfd
);
2478 health_code_update();
2480 len
= ctx
->on_buffer_ready(stream
, ctx
);
2482 * We don't check the return value here since if we get
2483 * a negative len, it means an error occurred thus we
2484 * simply remove it from the poll set and free the
2489 /* It's ok to have an unavailable sub-buffer */
2490 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2491 /* Clean up stream from consumer and free it. */
2492 lttng_poll_del(&events
, stream
->wait_fd
);
2493 consumer_del_metadata_stream(stream
, metadata_ht
);
2495 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2496 DBG("Metadata fd %d is hup|err.", pollfd
);
2497 if (!stream
->hangup_flush_done
2498 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2499 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2500 DBG("Attempting to flush and consume the UST buffers");
2501 lttng_ustconsumer_on_stream_hangup(stream
);
2503 /* We just flushed the stream now read it. */
2505 health_code_update();
2507 len
= ctx
->on_buffer_ready(stream
, ctx
);
2509 * We don't check the return value here since if we get
2510 * a negative len, it means an error occurred thus we
2511 * simply remove it from the poll set and free the
2517 lttng_poll_del(&events
, stream
->wait_fd
);
2519 * This call update the channel states, closes file descriptors
2520 * and securely free the stream.
2522 consumer_del_metadata_stream(stream
, metadata_ht
);
2524 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2528 /* Release RCU lock for the stream looked up */
2536 DBG("Metadata poll thread exiting");
2538 lttng_poll_clean(&events
);
2543 ERR("Health error occurred in %s", __func__
);
2545 health_unregister(health_consumerd
);
2546 rcu_unregister_thread();
2551 * This thread polls the fds in the set to consume the data and write
2552 * it to tracefile if necessary.
2554 void *consumer_thread_data_poll(void *data
)
2556 int num_rdy
, num_hup
, high_prio
, ret
, i
, err
= -1;
2557 struct pollfd
*pollfd
= NULL
;
2558 /* local view of the streams */
2559 struct lttng_consumer_stream
**local_stream
= NULL
, *new_stream
= NULL
;
2560 /* local view of consumer_data.fds_count */
2562 /* 2 for the consumer_data_pipe and wake up pipe */
2563 const int nb_pipes_fd
= 2;
2564 /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
2565 int nb_inactive_fd
= 0;
2566 struct lttng_consumer_local_data
*ctx
= data
;
2569 rcu_register_thread();
2571 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_DATA
);
2573 if (testpoint(consumerd_thread_data
)) {
2574 goto error_testpoint
;
2577 health_code_update();
2579 local_stream
= zmalloc(sizeof(struct lttng_consumer_stream
*));
2580 if (local_stream
== NULL
) {
2581 PERROR("local_stream malloc");
2586 health_code_update();
2592 * the fds set has been updated, we need to update our
2593 * local array as well
2595 pthread_mutex_lock(&consumer_data
.lock
);
2596 if (consumer_data
.need_update
) {
2601 local_stream
= NULL
;
2603 /* Allocate for all fds */
2604 pollfd
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) * sizeof(struct pollfd
));
2605 if (pollfd
== NULL
) {
2606 PERROR("pollfd malloc");
2607 pthread_mutex_unlock(&consumer_data
.lock
);
2611 local_stream
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) *
2612 sizeof(struct lttng_consumer_stream
*));
2613 if (local_stream
== NULL
) {
2614 PERROR("local_stream malloc");
2615 pthread_mutex_unlock(&consumer_data
.lock
);
2618 ret
= update_poll_array(ctx
, &pollfd
, local_stream
,
2619 data_ht
, &nb_inactive_fd
);
2621 ERR("Error in allocating pollfd or local_outfds");
2622 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2623 pthread_mutex_unlock(&consumer_data
.lock
);
2627 consumer_data
.need_update
= 0;
2629 pthread_mutex_unlock(&consumer_data
.lock
);
2631 /* No FDs and consumer_quit, consumer_cleanup the thread */
2632 if (nb_fd
== 0 && nb_inactive_fd
== 0 &&
2633 CMM_LOAD_SHARED(consumer_quit
) == 1) {
2634 err
= 0; /* All is OK */
2637 /* poll on the array of fds */
2639 DBG("polling on %d fd", nb_fd
+ nb_pipes_fd
);
2640 if (testpoint(consumerd_thread_data_poll
)) {
2643 health_poll_entry();
2644 num_rdy
= poll(pollfd
, nb_fd
+ nb_pipes_fd
, -1);
2646 DBG("poll num_rdy : %d", num_rdy
);
2647 if (num_rdy
== -1) {
2649 * Restart interrupted system call.
2651 if (errno
== EINTR
) {
2654 PERROR("Poll error");
2655 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2657 } else if (num_rdy
== 0) {
2658 DBG("Polling thread timed out");
2662 if (caa_unlikely(data_consumption_paused
)) {
2663 DBG("Data consumption paused, sleeping...");
2669 * If the consumer_data_pipe triggered poll go directly to the
2670 * beginning of the loop to update the array. We want to prioritize
2671 * array update over low-priority reads.
2673 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2674 ssize_t pipe_readlen
;
2676 DBG("consumer_data_pipe wake up");
2677 pipe_readlen
= lttng_pipe_read(ctx
->consumer_data_pipe
,
2678 &new_stream
, sizeof(new_stream
));
2679 if (pipe_readlen
< sizeof(new_stream
)) {
2680 PERROR("Consumer data pipe");
2681 /* Continue so we can at least handle the current stream(s). */
2686 * If the stream is NULL, just ignore it. It's also possible that
2687 * the sessiond poll thread changed the consumer_quit state and is
2688 * waking us up to test it.
2690 if (new_stream
== NULL
) {
2691 validate_endpoint_status_data_stream();
2695 /* Continue to update the local streams and handle prio ones */
2699 /* Handle wakeup pipe. */
2700 if (pollfd
[nb_fd
+ 1].revents
& (POLLIN
| POLLPRI
)) {
2702 ssize_t pipe_readlen
;
2704 pipe_readlen
= lttng_pipe_read(ctx
->consumer_wakeup_pipe
, &dummy
,
2706 if (pipe_readlen
< 0) {
2707 PERROR("Consumer data wakeup pipe");
2709 /* We've been awakened to handle stream(s). */
2710 ctx
->has_wakeup
= 0;
2713 /* Take care of high priority channels first. */
2714 for (i
= 0; i
< nb_fd
; i
++) {
2715 health_code_update();
2717 if (local_stream
[i
] == NULL
) {
2720 if (pollfd
[i
].revents
& POLLPRI
) {
2721 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2723 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2724 /* it's ok to have an unavailable sub-buffer */
2725 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2726 /* Clean the stream and free it. */
2727 consumer_del_stream(local_stream
[i
], data_ht
);
2728 local_stream
[i
] = NULL
;
2729 } else if (len
> 0) {
2730 local_stream
[i
]->data_read
= 1;
2736 * If we read high prio channel in this loop, try again
2737 * for more high prio data.
2743 /* Take care of low priority channels. */
2744 for (i
= 0; i
< nb_fd
; i
++) {
2745 health_code_update();
2747 if (local_stream
[i
] == NULL
) {
2750 if ((pollfd
[i
].revents
& POLLIN
) ||
2751 local_stream
[i
]->hangup_flush_done
||
2752 local_stream
[i
]->has_data
) {
2753 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2754 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2755 /* it's ok to have an unavailable sub-buffer */
2756 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2757 /* Clean the stream and free it. */
2758 consumer_del_stream(local_stream
[i
], data_ht
);
2759 local_stream
[i
] = NULL
;
2760 } else if (len
> 0) {
2761 local_stream
[i
]->data_read
= 1;
2766 /* Handle hangup and errors */
2767 for (i
= 0; i
< nb_fd
; i
++) {
2768 health_code_update();
2770 if (local_stream
[i
] == NULL
) {
2773 if (!local_stream
[i
]->hangup_flush_done
2774 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2775 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2776 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2777 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2779 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2780 /* Attempt read again, for the data we just flushed. */
2781 local_stream
[i
]->data_read
= 1;
2784 * If the poll flag is HUP/ERR/NVAL and we have
2785 * read no data in this pass, we can remove the
2786 * stream from its hash table.
2788 if ((pollfd
[i
].revents
& POLLHUP
)) {
2789 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2790 if (!local_stream
[i
]->data_read
) {
2791 consumer_del_stream(local_stream
[i
], data_ht
);
2792 local_stream
[i
] = NULL
;
2795 } else if (pollfd
[i
].revents
& POLLERR
) {
2796 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2797 if (!local_stream
[i
]->data_read
) {
2798 consumer_del_stream(local_stream
[i
], data_ht
);
2799 local_stream
[i
] = NULL
;
2802 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2803 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2804 if (!local_stream
[i
]->data_read
) {
2805 consumer_del_stream(local_stream
[i
], data_ht
);
2806 local_stream
[i
] = NULL
;
2810 if (local_stream
[i
] != NULL
) {
2811 local_stream
[i
]->data_read
= 0;
2818 DBG("polling thread exiting");
2823 * Close the write side of the pipe so epoll_wait() in
2824 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2825 * read side of the pipe. If we close them both, epoll_wait strangely does
2826 * not return and could create a endless wait period if the pipe is the
2827 * only tracked fd in the poll set. The thread will take care of closing
2830 (void) lttng_pipe_write_close(ctx
->consumer_metadata_pipe
);
2835 ERR("Health error occurred in %s", __func__
);
2837 health_unregister(health_consumerd
);
2839 rcu_unregister_thread();
2844 * Close wake-up end of each stream belonging to the channel. This will
2845 * allow the poll() on the stream read-side to detect when the
2846 * write-side (application) finally closes them.
2849 void consumer_close_channel_streams(struct lttng_consumer_channel
*channel
)
2851 struct lttng_ht
*ht
;
2852 struct lttng_consumer_stream
*stream
;
2853 struct lttng_ht_iter iter
;
2855 ht
= consumer_data
.stream_per_chan_id_ht
;
2858 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2859 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
2860 ht
->match_fct
, &channel
->key
,
2861 &iter
.iter
, stream
, node_channel_id
.node
) {
2863 * Protect against teardown with mutex.
2865 pthread_mutex_lock(&stream
->lock
);
2866 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
2869 switch (consumer_data
.type
) {
2870 case LTTNG_CONSUMER_KERNEL
:
2872 case LTTNG_CONSUMER32_UST
:
2873 case LTTNG_CONSUMER64_UST
:
2874 if (stream
->metadata_flag
) {
2875 /* Safe and protected by the stream lock. */
2876 lttng_ustconsumer_close_metadata(stream
->chan
);
2879 * Note: a mutex is taken internally within
2880 * liblttng-ust-ctl to protect timer wakeup_fd
2881 * use from concurrent close.
2883 lttng_ustconsumer_close_stream_wakeup(stream
);
2887 ERR("Unknown consumer_data type");
2891 pthread_mutex_unlock(&stream
->lock
);
2896 static void destroy_channel_ht(struct lttng_ht
*ht
)
2898 struct lttng_ht_iter iter
;
2899 struct lttng_consumer_channel
*channel
;
2907 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, channel
, wait_fd_node
.node
) {
2908 ret
= lttng_ht_del(ht
, &iter
);
2913 lttng_ht_destroy(ht
);
2917 * This thread polls the channel fds to detect when they are being
2918 * closed. It closes all related streams if the channel is detected as
2919 * closed. It is currently only used as a shim layer for UST because the
2920 * consumerd needs to keep the per-stream wakeup end of pipes open for
2923 void *consumer_thread_channel_poll(void *data
)
2925 int ret
, i
, pollfd
, err
= -1;
2926 uint32_t revents
, nb_fd
;
2927 struct lttng_consumer_channel
*chan
= NULL
;
2928 struct lttng_ht_iter iter
;
2929 struct lttng_ht_node_u64
*node
;
2930 struct lttng_poll_event events
;
2931 struct lttng_consumer_local_data
*ctx
= data
;
2932 struct lttng_ht
*channel_ht
;
2934 rcu_register_thread();
2936 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_CHANNEL
);
2938 if (testpoint(consumerd_thread_channel
)) {
2939 goto error_testpoint
;
2942 health_code_update();
2944 channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2946 /* ENOMEM at this point. Better to bail out. */
2950 DBG("Thread channel poll started");
2952 /* Size is set to 1 for the consumer_channel pipe */
2953 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2955 ERR("Poll set creation failed");
2959 ret
= lttng_poll_add(&events
, ctx
->consumer_channel_pipe
[0], LPOLLIN
);
2965 DBG("Channel main loop started");
2969 health_code_update();
2970 DBG("Channel poll wait");
2971 health_poll_entry();
2972 ret
= lttng_poll_wait(&events
, -1);
2973 DBG("Channel poll return from wait with %d fd(s)",
2974 LTTNG_POLL_GETNB(&events
));
2976 DBG("Channel event caught in thread");
2978 if (errno
== EINTR
) {
2979 ERR("Poll EINTR caught");
2982 if (LTTNG_POLL_GETNB(&events
) == 0) {
2983 err
= 0; /* All is OK */
2990 /* From here, the event is a channel wait fd */
2991 for (i
= 0; i
< nb_fd
; i
++) {
2992 health_code_update();
2994 revents
= LTTNG_POLL_GETEV(&events
, i
);
2995 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2997 if (pollfd
== ctx
->consumer_channel_pipe
[0]) {
2998 if (revents
& LPOLLIN
) {
2999 enum consumer_channel_action action
;
3002 ret
= read_channel_pipe(ctx
, &chan
, &key
, &action
);
3005 ERR("Error reading channel pipe");
3007 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3012 case CONSUMER_CHANNEL_ADD
:
3013 DBG("Adding channel %d to poll set",
3016 lttng_ht_node_init_u64(&chan
->wait_fd_node
,
3019 lttng_ht_add_unique_u64(channel_ht
,
3020 &chan
->wait_fd_node
);
3022 /* Add channel to the global poll events list */
3023 lttng_poll_add(&events
, chan
->wait_fd
,
3024 LPOLLERR
| LPOLLHUP
);
3026 case CONSUMER_CHANNEL_DEL
:
3029 * This command should never be called if the channel
3030 * has streams monitored by either the data or metadata
3031 * thread. The consumer only notify this thread with a
3032 * channel del. command if it receives a destroy
3033 * channel command from the session daemon that send it
3034 * if a command prior to the GET_CHANNEL failed.
3038 chan
= consumer_find_channel(key
);
3041 ERR("UST consumer get channel key %" PRIu64
" not found for del channel", key
);
3044 lttng_poll_del(&events
, chan
->wait_fd
);
3045 iter
.iter
.node
= &chan
->wait_fd_node
.node
;
3046 ret
= lttng_ht_del(channel_ht
, &iter
);
3049 switch (consumer_data
.type
) {
3050 case LTTNG_CONSUMER_KERNEL
:
3052 case LTTNG_CONSUMER32_UST
:
3053 case LTTNG_CONSUMER64_UST
:
3054 health_code_update();
3055 /* Destroy streams that might have been left in the stream list. */
3056 clean_channel_stream_list(chan
);
3059 ERR("Unknown consumer_data type");
3064 * Release our own refcount. Force channel deletion even if
3065 * streams were not initialized.
3067 if (!uatomic_sub_return(&chan
->refcount
, 1)) {
3068 consumer_del_channel(chan
);
3073 case CONSUMER_CHANNEL_QUIT
:
3075 * Remove the pipe from the poll set and continue the loop
3076 * since their might be data to consume.
3078 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3081 ERR("Unknown action");
3084 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3085 DBG("Channel thread pipe hung up");
3087 * Remove the pipe from the poll set and continue the loop
3088 * since their might be data to consume.
3090 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3093 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3097 /* Handle other stream */
3103 uint64_t tmp_id
= (uint64_t) pollfd
;
3105 lttng_ht_lookup(channel_ht
, &tmp_id
, &iter
);
3107 node
= lttng_ht_iter_get_node_u64(&iter
);
3110 chan
= caa_container_of(node
, struct lttng_consumer_channel
,
3113 /* Check for error event */
3114 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3115 DBG("Channel fd %d is hup|err.", pollfd
);
3117 lttng_poll_del(&events
, chan
->wait_fd
);
3118 ret
= lttng_ht_del(channel_ht
, &iter
);
3122 * This will close the wait fd for each stream associated to
3123 * this channel AND monitored by the data/metadata thread thus
3124 * will be clean by the right thread.
3126 consumer_close_channel_streams(chan
);
3128 /* Release our own refcount */
3129 if (!uatomic_sub_return(&chan
->refcount
, 1)
3130 && !uatomic_read(&chan
->nb_init_stream_left
)) {
3131 consumer_del_channel(chan
);
3134 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3139 /* Release RCU lock for the channel looked up */
3147 lttng_poll_clean(&events
);
3149 destroy_channel_ht(channel_ht
);
3152 DBG("Channel poll thread exiting");
3155 ERR("Health error occurred in %s", __func__
);
3157 health_unregister(health_consumerd
);
3158 rcu_unregister_thread();
3162 static int set_metadata_socket(struct lttng_consumer_local_data
*ctx
,
3163 struct pollfd
*sockpoll
, int client_socket
)
3170 ret
= lttng_consumer_poll_socket(sockpoll
);
3174 DBG("Metadata connection on client_socket");
3176 /* Blocking call, waiting for transmission */
3177 ctx
->consumer_metadata_socket
= lttcomm_accept_unix_sock(client_socket
);
3178 if (ctx
->consumer_metadata_socket
< 0) {
3179 WARN("On accept metadata");
3190 * This thread listens on the consumerd socket and receives the file
3191 * descriptors from the session daemon.
3193 void *consumer_thread_sessiond_poll(void *data
)
3195 int sock
= -1, client_socket
, ret
, err
= -1;
3197 * structure to poll for incoming data on communication socket avoids
3198 * making blocking sockets.
3200 struct pollfd consumer_sockpoll
[2];
3201 struct lttng_consumer_local_data
*ctx
= data
;
3203 rcu_register_thread();
3205 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_SESSIOND
);
3207 if (testpoint(consumerd_thread_sessiond
)) {
3208 goto error_testpoint
;
3211 health_code_update();
3213 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
3214 unlink(ctx
->consumer_command_sock_path
);
3215 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
3216 if (client_socket
< 0) {
3217 ERR("Cannot create command socket");
3221 ret
= lttcomm_listen_unix_sock(client_socket
);
3226 DBG("Sending ready command to lttng-sessiond");
3227 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
3228 /* return < 0 on error, but == 0 is not fatal */
3230 ERR("Error sending ready command to lttng-sessiond");
3234 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3235 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
3236 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
3237 consumer_sockpoll
[1].fd
= client_socket
;
3238 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3240 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3248 DBG("Connection on client_socket");
3250 /* Blocking call, waiting for transmission */
3251 sock
= lttcomm_accept_unix_sock(client_socket
);
3258 * Setup metadata socket which is the second socket connection on the
3259 * command unix socket.
3261 ret
= set_metadata_socket(ctx
, consumer_sockpoll
, client_socket
);
3270 /* This socket is not useful anymore. */
3271 ret
= close(client_socket
);
3273 PERROR("close client_socket");
3277 /* update the polling structure to poll on the established socket */
3278 consumer_sockpoll
[1].fd
= sock
;
3279 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3282 health_code_update();
3284 health_poll_entry();
3285 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3294 DBG("Incoming command on sock");
3295 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
3298 * This could simply be a session daemon quitting. Don't output
3301 DBG("Communication interrupted on command socket");
3305 if (CMM_LOAD_SHARED(consumer_quit
)) {
3306 DBG("consumer_thread_receive_fds received quit from signal");
3307 err
= 0; /* All is OK */
3310 DBG("received command on sock");
3316 DBG("Consumer thread sessiond poll exiting");
3319 * Close metadata streams since the producer is the session daemon which
3322 * NOTE: for now, this only applies to the UST tracer.
3324 lttng_consumer_close_all_metadata();
3327 * when all fds have hung up, the polling thread
3330 CMM_STORE_SHARED(consumer_quit
, 1);
3333 * Notify the data poll thread to poll back again and test the
3334 * consumer_quit state that we just set so to quit gracefully.
3336 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
3338 notify_channel_pipe(ctx
, NULL
, -1, CONSUMER_CHANNEL_QUIT
);
3340 notify_health_quit_pipe(health_quit_pipe
);
3342 /* Cleaning up possibly open sockets. */
3346 PERROR("close sock sessiond poll");
3349 if (client_socket
>= 0) {
3350 ret
= close(client_socket
);
3352 PERROR("close client_socket sessiond poll");
3359 ERR("Health error occurred in %s", __func__
);
3361 health_unregister(health_consumerd
);
3363 rcu_unregister_thread();
3367 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
3368 struct lttng_consumer_local_data
*ctx
)
3372 bool rotated
= false;
3374 pthread_mutex_lock(&stream
->lock
);
3375 if (stream
->metadata_flag
) {
3376 pthread_mutex_lock(&stream
->metadata_rdv_lock
);
3379 switch (consumer_data
.type
) {
3380 case LTTNG_CONSUMER_KERNEL
:
3381 ret
= lttng_kconsumer_read_subbuffer(stream
, ctx
, &rotated
);
3383 case LTTNG_CONSUMER32_UST
:
3384 case LTTNG_CONSUMER64_UST
:
3385 ret
= lttng_ustconsumer_read_subbuffer(stream
, ctx
, &rotated
);
3388 ERR("Unknown consumer_data type");
3394 if (stream
->metadata_flag
) {
3395 pthread_cond_broadcast(&stream
->metadata_rdv
);
3396 pthread_mutex_unlock(&stream
->metadata_rdv_lock
);
3398 pthread_mutex_unlock(&stream
->lock
);
3400 rotate_ret
= consumer_post_rotation(stream
, ctx
);
3401 if (rotate_ret
< 0) {
3402 ERR("Failed after a rotation");
3410 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3412 switch (consumer_data
.type
) {
3413 case LTTNG_CONSUMER_KERNEL
:
3414 return lttng_kconsumer_on_recv_stream(stream
);
3415 case LTTNG_CONSUMER32_UST
:
3416 case LTTNG_CONSUMER64_UST
:
3417 return lttng_ustconsumer_on_recv_stream(stream
);
3419 ERR("Unknown consumer_data type");
3426 * Allocate and set consumer data hash tables.
3428 int lttng_consumer_init(void)
3430 consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3431 if (!consumer_data
.channel_ht
) {
3435 consumer_data
.channels_by_session_id_ht
=
3436 lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3437 if (!consumer_data
.channels_by_session_id_ht
) {
3441 consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3442 if (!consumer_data
.relayd_ht
) {
3446 consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3447 if (!consumer_data
.stream_list_ht
) {
3451 consumer_data
.stream_per_chan_id_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3452 if (!consumer_data
.stream_per_chan_id_ht
) {
3456 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3461 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3473 * Process the ADD_RELAYD command receive by a consumer.
3475 * This will create a relayd socket pair and add it to the relayd hash table.
3476 * The caller MUST acquire a RCU read side lock before calling it.
3478 void consumer_add_relayd_socket(uint64_t net_seq_idx
, int sock_type
,
3479 struct lttng_consumer_local_data
*ctx
, int sock
,
3480 struct pollfd
*consumer_sockpoll
,
3481 struct lttcomm_relayd_sock
*relayd_sock
, uint64_t sessiond_id
,
3482 uint64_t relayd_session_id
)
3484 int fd
= -1, ret
= -1, relayd_created
= 0;
3485 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3486 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3489 assert(relayd_sock
);
3491 DBG("Consumer adding relayd socket (idx: %" PRIu64
")", net_seq_idx
);
3493 /* Get relayd reference if exists. */
3494 relayd
= consumer_find_relayd(net_seq_idx
);
3495 if (relayd
== NULL
) {
3496 assert(sock_type
== LTTNG_STREAM_CONTROL
);
3497 /* Not found. Allocate one. */
3498 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
3499 if (relayd
== NULL
) {
3500 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3503 relayd
->sessiond_session_id
= sessiond_id
;
3508 * This code path MUST continue to the consumer send status message to
3509 * we can notify the session daemon and continue our work without
3510 * killing everything.
3514 * relayd key should never be found for control socket.
3516 assert(sock_type
!= LTTNG_STREAM_CONTROL
);
3519 /* First send a status message before receiving the fds. */
3520 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
3522 /* Somehow, the session daemon is not responding anymore. */
3523 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3524 goto error_nosignal
;
3527 /* Poll on consumer socket. */
3528 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3530 /* Needing to exit in the middle of a command: error. */
3531 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
3532 goto error_nosignal
;
3535 /* Get relayd socket from session daemon */
3536 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
3537 if (ret
!= sizeof(fd
)) {
3538 fd
= -1; /* Just in case it gets set with an invalid value. */
3541 * Failing to receive FDs might indicate a major problem such as
3542 * reaching a fd limit during the receive where the kernel returns a
3543 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3544 * don't take any chances and stop everything.
3546 * XXX: Feature request #558 will fix that and avoid this possible
3547 * issue when reaching the fd limit.
3549 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
3550 ret_code
= LTTCOMM_CONSUMERD_ERROR_RECV_FD
;
3554 /* Copy socket information and received FD */
3555 switch (sock_type
) {
3556 case LTTNG_STREAM_CONTROL
:
3557 /* Copy received lttcomm socket */
3558 lttcomm_copy_sock(&relayd
->control_sock
.sock
, &relayd_sock
->sock
);
3559 ret
= lttcomm_create_sock(&relayd
->control_sock
.sock
);
3560 /* Handle create_sock error. */
3562 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3566 * Close the socket created internally by
3567 * lttcomm_create_sock, so we can replace it by the one
3568 * received from sessiond.
3570 if (close(relayd
->control_sock
.sock
.fd
)) {
3574 /* Assign new file descriptor */
3575 relayd
->control_sock
.sock
.fd
= fd
;
3576 /* Assign version values. */
3577 relayd
->control_sock
.major
= relayd_sock
->major
;
3578 relayd
->control_sock
.minor
= relayd_sock
->minor
;
3580 relayd
->relayd_session_id
= relayd_session_id
;
3583 case LTTNG_STREAM_DATA
:
3584 /* Copy received lttcomm socket */
3585 lttcomm_copy_sock(&relayd
->data_sock
.sock
, &relayd_sock
->sock
);
3586 ret
= lttcomm_create_sock(&relayd
->data_sock
.sock
);
3587 /* Handle create_sock error. */
3589 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3593 * Close the socket created internally by
3594 * lttcomm_create_sock, so we can replace it by the one
3595 * received from sessiond.
3597 if (close(relayd
->data_sock
.sock
.fd
)) {
3601 /* Assign new file descriptor */
3602 relayd
->data_sock
.sock
.fd
= fd
;
3603 /* Assign version values. */
3604 relayd
->data_sock
.major
= relayd_sock
->major
;
3605 relayd
->data_sock
.minor
= relayd_sock
->minor
;
3608 ERR("Unknown relayd socket type (%d)", sock_type
);
3609 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3613 DBG("Consumer %s socket created successfully with net idx %" PRIu64
" (fd: %d)",
3614 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
3615 relayd
->net_seq_idx
, fd
);
3617 * We gave the ownership of the fd to the relayd structure. Set the
3618 * fd to -1 so we don't call close() on it in the error path below.
3622 /* We successfully added the socket. Send status back. */
3623 ret
= consumer_send_status_msg(sock
, ret_code
);
3625 /* Somehow, the session daemon is not responding anymore. */
3626 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3627 goto error_nosignal
;
3631 * Add relayd socket pair to consumer data hashtable. If object already
3632 * exists or on error, the function gracefully returns.
3641 if (consumer_send_status_msg(sock
, ret_code
) < 0) {
3642 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3646 /* Close received socket if valid. */
3649 PERROR("close received socket");
3653 if (relayd_created
) {
3659 * Search for a relayd associated to the session id and return the reference.
3661 * A rcu read side lock MUST be acquire before calling this function and locked
3662 * until the relayd object is no longer necessary.
3664 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
3666 struct lttng_ht_iter iter
;
3667 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3669 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3670 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
3673 * Check by sessiond id which is unique here where the relayd session
3674 * id might not be when having multiple relayd.
3676 if (relayd
->sessiond_session_id
== id
) {
3677 /* Found the relayd. There can be only one per id. */
3689 * Check if for a given session id there is still data needed to be extract
3692 * Return 1 if data is pending or else 0 meaning ready to be read.
3694 int consumer_data_pending(uint64_t id
)
3697 struct lttng_ht_iter iter
;
3698 struct lttng_ht
*ht
;
3699 struct lttng_consumer_stream
*stream
;
3700 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3701 int (*data_pending
)(struct lttng_consumer_stream
*);
3703 DBG("Consumer data pending command on session id %" PRIu64
, id
);
3706 pthread_mutex_lock(&consumer_data
.lock
);
3708 switch (consumer_data
.type
) {
3709 case LTTNG_CONSUMER_KERNEL
:
3710 data_pending
= lttng_kconsumer_data_pending
;
3712 case LTTNG_CONSUMER32_UST
:
3713 case LTTNG_CONSUMER64_UST
:
3714 data_pending
= lttng_ustconsumer_data_pending
;
3717 ERR("Unknown consumer data type");
3721 /* Ease our life a bit */
3722 ht
= consumer_data
.stream_list_ht
;
3724 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3725 ht
->hash_fct(&id
, lttng_ht_seed
),
3727 &iter
.iter
, stream
, node_session_id
.node
) {
3728 pthread_mutex_lock(&stream
->lock
);
3731 * A removed node from the hash table indicates that the stream has
3732 * been deleted thus having a guarantee that the buffers are closed
3733 * on the consumer side. However, data can still be transmitted
3734 * over the network so don't skip the relayd check.
3736 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
3738 /* Check the stream if there is data in the buffers. */
3739 ret
= data_pending(stream
);
3741 pthread_mutex_unlock(&stream
->lock
);
3746 pthread_mutex_unlock(&stream
->lock
);
3749 relayd
= find_relayd_by_session_id(id
);
3751 unsigned int is_data_inflight
= 0;
3753 /* Send init command for data pending. */
3754 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3755 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
3756 relayd
->relayd_session_id
);
3758 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3759 /* Communication error thus the relayd so no data pending. */
3760 goto data_not_pending
;
3763 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3764 ht
->hash_fct(&id
, lttng_ht_seed
),
3766 &iter
.iter
, stream
, node_session_id
.node
) {
3767 if (stream
->metadata_flag
) {
3768 ret
= relayd_quiescent_control(&relayd
->control_sock
,
3769 stream
->relayd_stream_id
);
3771 ret
= relayd_data_pending(&relayd
->control_sock
,
3772 stream
->relayd_stream_id
,
3773 stream
->next_net_seq_num
- 1);
3777 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3779 } else if (ret
< 0) {
3780 ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
3781 lttng_consumer_cleanup_relayd(relayd
);
3782 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3783 goto data_not_pending
;
3787 /* Send end command for data pending. */
3788 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3789 relayd
->relayd_session_id
, &is_data_inflight
);
3790 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3792 ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
3793 lttng_consumer_cleanup_relayd(relayd
);
3794 goto data_not_pending
;
3796 if (is_data_inflight
) {
3802 * Finding _no_ node in the hash table and no inflight data means that the
3803 * stream(s) have been removed thus data is guaranteed to be available for
3804 * analysis from the trace files.
3808 /* Data is available to be read by a viewer. */
3809 pthread_mutex_unlock(&consumer_data
.lock
);
3814 /* Data is still being extracted from buffers. */
3815 pthread_mutex_unlock(&consumer_data
.lock
);
3821 * Send a ret code status message to the sessiond daemon.
3823 * Return the sendmsg() return value.
3825 int consumer_send_status_msg(int sock
, int ret_code
)
3827 struct lttcomm_consumer_status_msg msg
;
3829 memset(&msg
, 0, sizeof(msg
));
3830 msg
.ret_code
= ret_code
;
3832 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3836 * Send a channel status message to the sessiond daemon.
3838 * Return the sendmsg() return value.
3840 int consumer_send_status_channel(int sock
,
3841 struct lttng_consumer_channel
*channel
)
3843 struct lttcomm_consumer_status_channel msg
;
3847 memset(&msg
, 0, sizeof(msg
));
3849 msg
.ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
3851 msg
.ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3852 msg
.key
= channel
->key
;
3853 msg
.stream_count
= channel
->streams
.count
;
3856 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3859 unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos
,
3860 unsigned long produced_pos
, uint64_t nb_packets_per_stream
,
3861 uint64_t max_sb_size
)
3863 unsigned long start_pos
;
3865 if (!nb_packets_per_stream
) {
3866 return consumed_pos
; /* Grab everything */
3868 start_pos
= produced_pos
- offset_align_floor(produced_pos
, max_sb_size
);
3869 start_pos
-= max_sb_size
* nb_packets_per_stream
;
3870 if ((long) (start_pos
- consumed_pos
) < 0) {
3871 return consumed_pos
; /* Grab everything */
3877 int consumer_flush_buffer(struct lttng_consumer_stream
*stream
, int producer_active
)
3881 switch (consumer_data
.type
) {
3882 case LTTNG_CONSUMER_KERNEL
:
3883 ret
= kernctl_buffer_flush(stream
->wait_fd
);
3885 ERR("Failed to flush kernel stream");
3889 case LTTNG_CONSUMER32_UST
:
3890 case LTTNG_CONSUMER64_UST
:
3891 lttng_ustctl_flush_buffer(stream
, producer_active
);
3894 ERR("Unknown consumer_data type");
3903 * Sample the rotate position for all the streams of a channel. If a stream
3904 * is already at the rotate position (produced == consumed), we flag it as
3905 * ready for rotation. The rotation of ready streams occurs after we have
3906 * replied to the session daemon that we have finished sampling the positions.
3907 * Must be called with RCU read-side lock held to ensure existence of channel.
3909 * Returns 0 on success, < 0 on error
3911 int lttng_consumer_rotate_channel(struct lttng_consumer_channel
*channel
,
3912 uint64_t key
, const char *path
, uint64_t relayd_id
,
3913 uint32_t metadata
, uint64_t new_chunk_id
,
3914 struct lttng_consumer_local_data
*ctx
)
3917 struct lttng_consumer_stream
*stream
;
3918 struct lttng_ht_iter iter
;
3919 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
3921 DBG("Consumer sample rotate position for channel %" PRIu64
, key
);
3925 pthread_mutex_lock(&channel
->lock
);
3926 channel
->current_chunk_id
= new_chunk_id
;
3928 ret
= lttng_strncpy(channel
->pathname
, path
, sizeof(channel
->pathname
));
3930 ERR("Failed to copy new path to channel during channel rotation");
3932 goto end_unlock_channel
;
3935 if (relayd_id
== -1ULL) {
3937 * The domain path (/ust or /kernel) has been created before, we
3938 * now need to create the last part of the path: the application/user
3939 * specific section (uid/1000/64-bit).
3941 ret
= utils_mkdir_recursive(channel
->pathname
, S_IRWXU
| S_IRWXG
,
3942 channel
->uid
, channel
->gid
);
3944 ERR("Failed to create trace directory at %s during rotation",
3947 goto end_unlock_channel
;
3951 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3952 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
3953 ht
->match_fct
, &channel
->key
, &iter
.iter
,
3954 stream
, node_channel_id
.node
) {
3955 unsigned long consumed_pos
;
3957 health_code_update();
3960 * Lock stream because we are about to change its state.
3962 pthread_mutex_lock(&stream
->lock
);
3964 ret
= lttng_strncpy(stream
->channel_read_only_attributes
.path
,
3966 sizeof(stream
->channel_read_only_attributes
.path
));
3968 ERR("Failed to sample channel path name during channel rotation");
3969 goto end_unlock_stream
;
3971 ret
= lttng_consumer_sample_snapshot_positions(stream
);
3973 ERR("Failed to sample snapshot position during channel rotation");
3974 goto end_unlock_stream
;
3977 ret
= lttng_consumer_get_produced_snapshot(stream
,
3978 &stream
->rotate_position
);
3980 ERR("Failed to sample produced position during channel rotation");
3981 goto end_unlock_stream
;
3984 lttng_consumer_get_consumed_snapshot(stream
,
3986 if (consumed_pos
== stream
->rotate_position
) {
3987 stream
->rotate_ready
= true;
3990 ret
= consumer_flush_buffer(stream
, 1);
3992 ERR("Failed to flush stream %" PRIu64
" during channel rotation",
3994 goto end_unlock_stream
;
3997 pthread_mutex_unlock(&stream
->lock
);
3999 pthread_mutex_unlock(&channel
->lock
);
4005 pthread_mutex_unlock(&stream
->lock
);
4007 pthread_mutex_unlock(&channel
->lock
);
4014 * Check if a stream is ready to be rotated after extracting it.
4016 * Return 1 if it is ready for rotation, 0 if it is not, a negative value on
4017 * error. Stream lock must be held.
4019 int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream
*stream
)
4022 unsigned long consumed_pos
;
4024 if (!stream
->rotate_position
&& !stream
->rotate_ready
) {
4029 if (stream
->rotate_ready
) {
4035 * If we don't have the rotate_ready flag, check the consumed position
4036 * to determine if we need to rotate.
4038 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4040 ERR("Taking snapshot positions");
4044 ret
= lttng_consumer_get_consumed_snapshot(stream
, &consumed_pos
);
4046 ERR("Consumed snapshot position");
4050 /* Rotate position not reached yet (with check for overflow). */
4051 if ((long) (consumed_pos
- stream
->rotate_position
) < 0) {
4062 * Reset the state for a stream after a rotation occurred.
4064 void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream
*stream
)
4066 stream
->rotate_position
= 0;
4067 stream
->rotate_ready
= false;
4071 * Perform the rotation a local stream file.
4073 int rotate_local_stream(struct lttng_consumer_local_data
*ctx
,
4074 struct lttng_consumer_stream
*stream
)
4078 DBG("Rotate local stream: stream key %" PRIu64
", channel key %" PRIu64
" at path %s",
4081 stream
->channel_read_only_attributes
.path
);
4083 ret
= close(stream
->out_fd
);
4085 PERROR("Closing trace file (fd %d), stream %" PRIu64
,
4086 stream
->out_fd
, stream
->key
);
4091 ret
= utils_create_stream_file(
4092 stream
->channel_read_only_attributes
.path
,
4094 stream
->channel_read_only_attributes
.tracefile_size
,
4095 stream
->tracefile_count_current
,
4096 stream
->uid
, stream
->gid
, NULL
);
4098 ERR("Rotate create stream file");
4101 stream
->out_fd
= ret
;
4102 stream
->tracefile_size_current
= 0;
4104 if (!stream
->metadata_flag
) {
4105 struct lttng_index_file
*index_file
;
4107 lttng_index_file_put(stream
->index_file
);
4109 index_file
= lttng_index_file_create(
4110 stream
->channel_read_only_attributes
.path
,
4111 stream
->name
, stream
->uid
, stream
->gid
,
4112 stream
->channel_read_only_attributes
.tracefile_size
,
4113 stream
->tracefile_count_current
,
4114 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
4116 ERR("Create index file during rotation");
4119 stream
->index_file
= index_file
;
4120 stream
->out_fd_offset
= 0;
4134 * Perform the rotation a stream file on the relay.
4136 int rotate_relay_stream(struct lttng_consumer_local_data
*ctx
,
4137 struct lttng_consumer_stream
*stream
)
4140 struct consumer_relayd_sock_pair
*relayd
;
4142 DBG("Rotate relay stream");
4143 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
4145 ERR("Failed to find relayd");
4150 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4151 ret
= relayd_rotate_stream(&relayd
->control_sock
,
4152 stream
->relayd_stream_id
,
4153 stream
->channel_read_only_attributes
.path
,
4154 stream
->chan
->current_chunk_id
,
4155 stream
->last_sequence_number
);
4156 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4158 ERR("Relayd rotate stream failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
4159 lttng_consumer_cleanup_relayd(relayd
);
4162 ERR("Rotate relay stream");
4170 * Performs the stream rotation for the rotate session feature if needed.
4171 * It must be called with the stream lock held.
4173 * Return 0 on success, a negative number of error.
4175 int lttng_consumer_rotate_stream(struct lttng_consumer_local_data
*ctx
,
4176 struct lttng_consumer_stream
*stream
, bool *rotated
)
4180 DBG("Consumer rotate stream %" PRIu64
, stream
->key
);
4182 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
4183 ret
= rotate_relay_stream(ctx
, stream
);
4185 ret
= rotate_local_stream(ctx
, stream
);
4187 stream
->trace_archive_id
++;
4189 ERR("Failed to rotate stream, ret = %i", ret
);
4193 if (stream
->metadata_flag
) {
4194 switch (consumer_data
.type
) {
4195 case LTTNG_CONSUMER_KERNEL
:
4197 * Reset the position of what has been read from the metadata
4198 * cache to 0 so we can dump it again.
4200 ret
= kernctl_metadata_cache_dump(stream
->wait_fd
);
4202 ERR("Failed to dump the kernel metadata cache after rotation");
4206 case LTTNG_CONSUMER32_UST
:
4207 case LTTNG_CONSUMER64_UST
:
4209 * Reset the position pushed from the metadata cache so it
4210 * will write from the beginning on the next push.
4212 stream
->ust_metadata_pushed
= 0;
4215 ERR("Unknown consumer_data type");
4219 lttng_consumer_reset_stream_rotate_state(stream
);
4232 * Rotate all the ready streams now.
4234 * This is especially important for low throughput streams that have already
4235 * been consumed, we cannot wait for their next packet to perform the
4237 * Need to be called with RCU read-side lock held to ensure existence of
4240 * Returns 0 on success, < 0 on error
4242 int lttng_consumer_rotate_ready_streams(struct lttng_consumer_channel
*channel
,
4243 uint64_t key
, struct lttng_consumer_local_data
*ctx
)
4246 struct lttng_consumer_stream
*stream
;
4247 struct lttng_ht_iter iter
;
4248 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
4252 DBG("Consumer rotate ready streams in channel %" PRIu64
, key
);
4254 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4255 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4256 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4257 stream
, node_channel_id
.node
) {
4258 health_code_update();
4260 pthread_mutex_lock(&stream
->lock
);
4262 if (!stream
->rotate_ready
) {
4263 pthread_mutex_unlock(&stream
->lock
);
4266 DBG("Consumer rotate ready stream %" PRIu64
, stream
->key
);
4268 ret
= lttng_consumer_rotate_stream(ctx
, stream
, NULL
);
4269 pthread_mutex_unlock(&stream
->lock
);
4274 ret
= consumer_post_rotation(stream
, ctx
);
4288 int rotate_rename_local(const char *old_path
, const char *new_path
,
4289 uid_t uid
, gid_t gid
)
4296 ret
= utils_mkdir_recursive(new_path
, S_IRWXU
| S_IRWXG
, uid
, gid
);
4298 ERR("Create directory on rotate");
4302 ret
= rename(old_path
, new_path
);
4303 if (ret
< 0 && errno
!= ENOENT
) {
4304 PERROR("Rename completed rotation chunk");
4314 int rotate_rename_relay(const char *old_path
, const char *new_path
,
4318 struct consumer_relayd_sock_pair
*relayd
;
4320 relayd
= consumer_find_relayd(relayd_id
);
4322 ERR("Failed to find relayd while running rotate_rename_relay command");
4327 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4328 ret
= relayd_rotate_rename(&relayd
->control_sock
, old_path
, new_path
);
4330 ERR("Relayd rotate rename failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
4331 lttng_consumer_cleanup_relayd(relayd
);
4333 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4338 int lttng_consumer_rotate_rename(const char *old_path
, const char *new_path
,
4339 uid_t uid
, gid_t gid
, uint64_t relayd_id
)
4341 if (relayd_id
!= -1ULL) {
4342 return rotate_rename_relay(old_path
, new_path
, relayd_id
);
4344 return rotate_rename_local(old_path
, new_path
, uid
, gid
);
4348 /* Stream lock must be acquired by the caller. */
4350 bool check_stream_rotation_pending(const struct lttng_consumer_stream
*stream
,
4351 uint64_t session_id
, uint64_t chunk_id
)
4353 bool pending
= false;
4355 if (stream
->session_id
!= session_id
) {
4361 * If the stream's archive_id belongs to the chunk being rotated (or an
4362 * even older one), it means that the consumer has not consumed all the
4363 * buffers that belong to the chunk being rotated. Therefore, the
4364 * rotation is considered as ongoing/pending.
4366 pending
= stream
->trace_archive_id
<= chunk_id
;
4371 /* RCU read lock must be acquired by the caller. */
4372 int lttng_consumer_check_rotation_pending_local(uint64_t session_id
,
4375 struct lttng_ht_iter iter
;
4376 struct lttng_consumer_stream
*stream
;
4377 bool rotation_pending
= false;
4379 /* Start with the metadata streams... */
4380 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
4381 pthread_mutex_lock(&stream
->lock
);
4382 rotation_pending
= check_stream_rotation_pending(stream
,
4383 session_id
, chunk_id
);
4384 pthread_mutex_unlock(&stream
->lock
);
4385 if (rotation_pending
) {
4390 /* ... followed by the data streams. */
4391 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
4392 pthread_mutex_lock(&stream
->lock
);
4393 rotation_pending
= check_stream_rotation_pending(stream
,
4394 session_id
, chunk_id
);
4395 pthread_mutex_unlock(&stream
->lock
);
4396 if (rotation_pending
) {
4402 return !!rotation_pending
;
4405 int lttng_consumer_check_rotation_pending_relay(uint64_t session_id
,
4406 uint64_t relayd_id
, uint64_t chunk_id
)
4409 struct consumer_relayd_sock_pair
*relayd
;
4411 relayd
= consumer_find_relayd(relayd_id
);
4413 ERR("Failed to find relayd id %" PRIu64
, relayd_id
);
4418 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4419 ret
= relayd_rotate_pending(&relayd
->control_sock
, chunk_id
);
4421 ERR("Relayd rotate pending failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
4422 lttng_consumer_cleanup_relayd(relayd
);
4424 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4431 int mkdir_local(const char *path
, uid_t uid
, gid_t gid
)
4435 ret
= utils_mkdir_recursive(path
, S_IRWXU
| S_IRWXG
, uid
, gid
);
4437 /* utils_mkdir_recursive logs an error. */
4447 int mkdir_relay(const char *path
, uint64_t relayd_id
)
4450 struct consumer_relayd_sock_pair
*relayd
;
4452 relayd
= consumer_find_relayd(relayd_id
);
4454 ERR("Failed to find relayd");
4459 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4460 ret
= relayd_mkdir(&relayd
->control_sock
, path
);
4462 ERR("Relayd mkdir failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
4463 lttng_consumer_cleanup_relayd(relayd
);
4465 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4472 int lttng_consumer_mkdir(const char *path
, uid_t uid
, gid_t gid
,
4475 if (relayd_id
!= -1ULL) {
4476 return mkdir_relay(path
, relayd_id
);
4478 return mkdir_local(path
, uid
, gid
);