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
);
398 call_rcu(&channel
->node
.head
, free_channel_rcu
);
400 pthread_mutex_unlock(&channel
->lock
);
401 pthread_mutex_unlock(&consumer_data
.lock
);
405 * Iterate over the relayd hash table and destroy each element. Finally,
406 * destroy the whole hash table.
408 static void cleanup_relayd_ht(void)
410 struct lttng_ht_iter iter
;
411 struct consumer_relayd_sock_pair
*relayd
;
415 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
417 consumer_destroy_relayd(relayd
);
422 lttng_ht_destroy(consumer_data
.relayd_ht
);
426 * Update the end point status of all streams having the given network sequence
427 * index (relayd index).
429 * It's atomically set without having the stream mutex locked which is fine
430 * because we handle the write/read race with a pipe wakeup for each thread.
432 static void update_endpoint_status_by_netidx(uint64_t net_seq_idx
,
433 enum consumer_endpoint_status status
)
435 struct lttng_ht_iter iter
;
436 struct lttng_consumer_stream
*stream
;
438 DBG("Consumer set delete flag on stream by idx %" PRIu64
, net_seq_idx
);
442 /* Let's begin with metadata */
443 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
444 if (stream
->net_seq_idx
== net_seq_idx
) {
445 uatomic_set(&stream
->endpoint_status
, status
);
446 DBG("Delete flag set to metadata stream %d", stream
->wait_fd
);
450 /* Follow up by the data streams */
451 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
452 if (stream
->net_seq_idx
== net_seq_idx
) {
453 uatomic_set(&stream
->endpoint_status
, status
);
454 DBG("Delete flag set to data stream %d", stream
->wait_fd
);
461 * Cleanup a relayd object by flagging every associated streams for deletion,
462 * destroying the object meaning removing it from the relayd hash table,
463 * closing the sockets and freeing the memory in a RCU call.
465 * If a local data context is available, notify the threads that the streams'
466 * state have changed.
468 void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair
*relayd
)
474 DBG("Cleaning up relayd object ID %"PRIu64
, relayd
->net_seq_idx
);
476 /* Save the net sequence index before destroying the object */
477 netidx
= relayd
->net_seq_idx
;
480 * Delete the relayd from the relayd hash table, close the sockets and free
481 * the object in a RCU call.
483 consumer_destroy_relayd(relayd
);
485 /* Set inactive endpoint to all streams */
486 update_endpoint_status_by_netidx(netidx
, CONSUMER_ENDPOINT_INACTIVE
);
489 * With a local data context, notify the threads that the streams' state
490 * have changed. The write() action on the pipe acts as an "implicit"
491 * memory barrier ordering the updates of the end point status from the
492 * read of this status which happens AFTER receiving this notify.
494 notify_thread_lttng_pipe(relayd
->ctx
->consumer_data_pipe
);
495 notify_thread_lttng_pipe(relayd
->ctx
->consumer_metadata_pipe
);
499 * Flag a relayd socket pair for destruction. Destroy it if the refcount
502 * RCU read side lock MUST be aquired before calling this function.
504 void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair
*relayd
)
508 /* Set destroy flag for this object */
509 uatomic_set(&relayd
->destroy_flag
, 1);
511 /* Destroy the relayd if refcount is 0 */
512 if (uatomic_read(&relayd
->refcount
) == 0) {
513 consumer_destroy_relayd(relayd
);
518 * Completly destroy stream from every visiable data structure and the given
521 * One this call returns, the stream object is not longer usable nor visible.
523 void consumer_del_stream(struct lttng_consumer_stream
*stream
,
526 consumer_stream_destroy(stream
, ht
);
530 * XXX naming of del vs destroy is all mixed up.
532 void consumer_del_stream_for_data(struct lttng_consumer_stream
*stream
)
534 consumer_stream_destroy(stream
, data_ht
);
537 void consumer_del_stream_for_metadata(struct lttng_consumer_stream
*stream
)
539 consumer_stream_destroy(stream
, metadata_ht
);
542 void consumer_stream_update_channel_attributes(
543 struct lttng_consumer_stream
*stream
,
544 struct lttng_consumer_channel
*channel
)
546 stream
->channel_read_only_attributes
.tracefile_size
=
547 channel
->tracefile_size
;
548 memcpy(stream
->channel_read_only_attributes
.path
, channel
->pathname
,
549 sizeof(stream
->channel_read_only_attributes
.path
));
552 struct lttng_consumer_stream
*consumer_allocate_stream(uint64_t channel_key
,
554 enum lttng_consumer_stream_state state
,
555 const char *channel_name
,
562 enum consumer_channel_type type
,
563 unsigned int monitor
,
564 uint64_t trace_archive_id
)
567 struct lttng_consumer_stream
*stream
;
569 stream
= zmalloc(sizeof(*stream
));
570 if (stream
== NULL
) {
571 PERROR("malloc struct lttng_consumer_stream");
578 stream
->key
= stream_key
;
580 stream
->out_fd_offset
= 0;
581 stream
->output_written
= 0;
582 stream
->state
= state
;
585 stream
->net_seq_idx
= relayd_id
;
586 stream
->session_id
= session_id
;
587 stream
->monitor
= monitor
;
588 stream
->endpoint_status
= CONSUMER_ENDPOINT_ACTIVE
;
589 stream
->index_file
= NULL
;
590 stream
->last_sequence_number
= -1ULL;
591 stream
->trace_archive_id
= trace_archive_id
;
592 pthread_mutex_init(&stream
->lock
, NULL
);
593 pthread_mutex_init(&stream
->metadata_timer_lock
, NULL
);
595 /* If channel is the metadata, flag this stream as metadata. */
596 if (type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
597 stream
->metadata_flag
= 1;
598 /* Metadata is flat out. */
599 strncpy(stream
->name
, DEFAULT_METADATA_NAME
, sizeof(stream
->name
));
600 /* Live rendez-vous point. */
601 pthread_cond_init(&stream
->metadata_rdv
, NULL
);
602 pthread_mutex_init(&stream
->metadata_rdv_lock
, NULL
);
604 /* Format stream name to <channel_name>_<cpu_number> */
605 ret
= snprintf(stream
->name
, sizeof(stream
->name
), "%s_%d",
608 PERROR("snprintf stream name");
613 /* Key is always the wait_fd for streams. */
614 lttng_ht_node_init_u64(&stream
->node
, stream
->key
);
616 /* Init node per channel id key */
617 lttng_ht_node_init_u64(&stream
->node_channel_id
, channel_key
);
619 /* Init session id node with the stream session id */
620 lttng_ht_node_init_u64(&stream
->node_session_id
, stream
->session_id
);
622 DBG3("Allocated stream %s (key %" PRIu64
", chan_key %" PRIu64
623 " relayd_id %" PRIu64
", session_id %" PRIu64
,
624 stream
->name
, stream
->key
, channel_key
,
625 stream
->net_seq_idx
, stream
->session_id
);
641 * Add a stream to the global list protected by a mutex.
643 void consumer_add_data_stream(struct lttng_consumer_stream
*stream
)
645 struct lttng_ht
*ht
= data_ht
;
650 DBG3("Adding consumer stream %" PRIu64
, stream
->key
);
652 pthread_mutex_lock(&consumer_data
.lock
);
653 pthread_mutex_lock(&stream
->chan
->lock
);
654 pthread_mutex_lock(&stream
->chan
->timer_lock
);
655 pthread_mutex_lock(&stream
->lock
);
658 /* Steal stream identifier to avoid having streams with the same key */
659 steal_stream_key(stream
->key
, ht
);
661 lttng_ht_add_unique_u64(ht
, &stream
->node
);
663 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
664 &stream
->node_channel_id
);
667 * Add stream to the stream_list_ht of the consumer data. No need to steal
668 * the key since the HT does not use it and we allow to add redundant keys
671 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
674 * When nb_init_stream_left reaches 0, we don't need to trigger any action
675 * in terms of destroying the associated channel, because the action that
676 * causes the count to become 0 also causes a stream to be added. The
677 * channel deletion will thus be triggered by the following removal of this
680 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
681 /* Increment refcount before decrementing nb_init_stream_left */
683 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
686 /* Update consumer data once the node is inserted. */
687 consumer_data
.stream_count
++;
688 consumer_data
.need_update
= 1;
691 pthread_mutex_unlock(&stream
->lock
);
692 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
693 pthread_mutex_unlock(&stream
->chan
->lock
);
694 pthread_mutex_unlock(&consumer_data
.lock
);
697 void consumer_del_data_stream(struct lttng_consumer_stream
*stream
)
699 consumer_del_stream(stream
, data_ht
);
703 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
704 * be acquired before calling this.
706 static int add_relayd(struct consumer_relayd_sock_pair
*relayd
)
709 struct lttng_ht_node_u64
*node
;
710 struct lttng_ht_iter iter
;
714 lttng_ht_lookup(consumer_data
.relayd_ht
,
715 &relayd
->net_seq_idx
, &iter
);
716 node
= lttng_ht_iter_get_node_u64(&iter
);
720 lttng_ht_add_unique_u64(consumer_data
.relayd_ht
, &relayd
->node
);
727 * Allocate and return a consumer relayd socket.
729 static struct consumer_relayd_sock_pair
*consumer_allocate_relayd_sock_pair(
730 uint64_t net_seq_idx
)
732 struct consumer_relayd_sock_pair
*obj
= NULL
;
734 /* net sequence index of -1 is a failure */
735 if (net_seq_idx
== (uint64_t) -1ULL) {
739 obj
= zmalloc(sizeof(struct consumer_relayd_sock_pair
));
741 PERROR("zmalloc relayd sock");
745 obj
->net_seq_idx
= net_seq_idx
;
747 obj
->destroy_flag
= 0;
748 obj
->control_sock
.sock
.fd
= -1;
749 obj
->data_sock
.sock
.fd
= -1;
750 lttng_ht_node_init_u64(&obj
->node
, obj
->net_seq_idx
);
751 pthread_mutex_init(&obj
->ctrl_sock_mutex
, NULL
);
758 * Find a relayd socket pair in the global consumer data.
760 * Return the object if found else NULL.
761 * RCU read-side lock must be held across this call and while using the
764 struct consumer_relayd_sock_pair
*consumer_find_relayd(uint64_t key
)
766 struct lttng_ht_iter iter
;
767 struct lttng_ht_node_u64
*node
;
768 struct consumer_relayd_sock_pair
*relayd
= NULL
;
770 /* Negative keys are lookup failures */
771 if (key
== (uint64_t) -1ULL) {
775 lttng_ht_lookup(consumer_data
.relayd_ht
, &key
,
777 node
= lttng_ht_iter_get_node_u64(&iter
);
779 relayd
= caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
787 * Find a relayd and send the stream
789 * Returns 0 on success, < 0 on error
791 int consumer_send_relayd_stream(struct lttng_consumer_stream
*stream
,
795 struct consumer_relayd_sock_pair
*relayd
;
798 assert(stream
->net_seq_idx
!= -1ULL);
801 /* The stream is not metadata. Get relayd reference if exists. */
803 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
804 if (relayd
!= NULL
) {
805 /* Add stream on the relayd */
806 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
807 ret
= relayd_add_stream(&relayd
->control_sock
, stream
->name
,
808 path
, &stream
->relayd_stream_id
,
809 stream
->chan
->tracefile_size
, stream
->chan
->tracefile_count
,
810 stream
->trace_archive_id
);
811 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
813 ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
814 lttng_consumer_cleanup_relayd(relayd
);
818 uatomic_inc(&relayd
->refcount
);
819 stream
->sent_to_relayd
= 1;
821 ERR("Stream %" PRIu64
" relayd ID %" PRIu64
" unknown. Can't send it.",
822 stream
->key
, stream
->net_seq_idx
);
827 DBG("Stream %s with key %" PRIu64
" sent to relayd id %" PRIu64
,
828 stream
->name
, stream
->key
, stream
->net_seq_idx
);
836 * Find a relayd and send the streams sent message
838 * Returns 0 on success, < 0 on error
840 int consumer_send_relayd_streams_sent(uint64_t net_seq_idx
)
843 struct consumer_relayd_sock_pair
*relayd
;
845 assert(net_seq_idx
!= -1ULL);
847 /* The stream is not metadata. Get relayd reference if exists. */
849 relayd
= consumer_find_relayd(net_seq_idx
);
850 if (relayd
!= NULL
) {
851 /* Add stream on the relayd */
852 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
853 ret
= relayd_streams_sent(&relayd
->control_sock
);
854 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
856 ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
857 lttng_consumer_cleanup_relayd(relayd
);
861 ERR("Relayd ID %" PRIu64
" unknown. Can't send streams_sent.",
868 DBG("All streams sent relayd id %" PRIu64
, net_seq_idx
);
876 * Find a relayd and close the stream
878 void close_relayd_stream(struct lttng_consumer_stream
*stream
)
880 struct consumer_relayd_sock_pair
*relayd
;
882 /* The stream is not metadata. Get relayd reference if exists. */
884 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
886 consumer_stream_relayd_close(stream
, relayd
);
892 * Handle stream for relayd transmission if the stream applies for network
893 * streaming where the net sequence index is set.
895 * Return destination file descriptor or negative value on error.
897 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
898 size_t data_size
, unsigned long padding
,
899 struct consumer_relayd_sock_pair
*relayd
)
902 struct lttcomm_relayd_data_hdr data_hdr
;
908 /* Reset data header */
909 memset(&data_hdr
, 0, sizeof(data_hdr
));
911 if (stream
->metadata_flag
) {
912 /* Caller MUST acquire the relayd control socket lock */
913 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
918 /* Metadata are always sent on the control socket. */
919 outfd
= relayd
->control_sock
.sock
.fd
;
921 /* Set header with stream information */
922 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
923 data_hdr
.data_size
= htobe32(data_size
);
924 data_hdr
.padding_size
= htobe32(padding
);
926 * Note that net_seq_num below is assigned with the *current* value of
927 * next_net_seq_num and only after that the next_net_seq_num will be
928 * increment. This is why when issuing a command on the relayd using
929 * this next value, 1 should always be substracted in order to compare
930 * the last seen sequence number on the relayd side to the last sent.
932 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
933 /* Other fields are zeroed previously */
935 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
941 ++stream
->next_net_seq_num
;
943 /* Set to go on data socket */
944 outfd
= relayd
->data_sock
.sock
.fd
;
952 * Allocate and return a new lttng_consumer_channel object using the given key
953 * to initialize the hash table node.
955 * On error, return NULL.
957 struct lttng_consumer_channel
*consumer_allocate_channel(uint64_t key
,
959 const char *pathname
,
964 enum lttng_event_output output
,
965 uint64_t tracefile_size
,
966 uint64_t tracefile_count
,
967 uint64_t session_id_per_pid
,
968 unsigned int monitor
,
969 unsigned int live_timer_interval
,
970 const char *root_shm_path
,
971 const char *shm_path
)
973 struct lttng_consumer_channel
*channel
;
975 channel
= zmalloc(sizeof(*channel
));
976 if (channel
== NULL
) {
977 PERROR("malloc struct lttng_consumer_channel");
982 channel
->refcount
= 0;
983 channel
->session_id
= session_id
;
984 channel
->session_id_per_pid
= session_id_per_pid
;
987 channel
->relayd_id
= relayd_id
;
988 channel
->tracefile_size
= tracefile_size
;
989 channel
->tracefile_count
= tracefile_count
;
990 channel
->monitor
= monitor
;
991 channel
->live_timer_interval
= live_timer_interval
;
992 pthread_mutex_init(&channel
->lock
, NULL
);
993 pthread_mutex_init(&channel
->timer_lock
, NULL
);
996 case LTTNG_EVENT_SPLICE
:
997 channel
->output
= CONSUMER_CHANNEL_SPLICE
;
999 case LTTNG_EVENT_MMAP
:
1000 channel
->output
= CONSUMER_CHANNEL_MMAP
;
1010 * In monitor mode, the streams associated with the channel will be put in
1011 * a special list ONLY owned by this channel. So, the refcount is set to 1
1012 * here meaning that the channel itself has streams that are referenced.
1014 * On a channel deletion, once the channel is no longer visible, the
1015 * refcount is decremented and checked for a zero value to delete it. With
1016 * streams in no monitor mode, it will now be safe to destroy the channel.
1018 if (!channel
->monitor
) {
1019 channel
->refcount
= 1;
1022 strncpy(channel
->pathname
, pathname
, sizeof(channel
->pathname
));
1023 channel
->pathname
[sizeof(channel
->pathname
) - 1] = '\0';
1025 strncpy(channel
->name
, name
, sizeof(channel
->name
));
1026 channel
->name
[sizeof(channel
->name
) - 1] = '\0';
1028 if (root_shm_path
) {
1029 strncpy(channel
->root_shm_path
, root_shm_path
, sizeof(channel
->root_shm_path
));
1030 channel
->root_shm_path
[sizeof(channel
->root_shm_path
) - 1] = '\0';
1033 strncpy(channel
->shm_path
, shm_path
, sizeof(channel
->shm_path
));
1034 channel
->shm_path
[sizeof(channel
->shm_path
) - 1] = '\0';
1037 lttng_ht_node_init_u64(&channel
->node
, channel
->key
);
1039 channel
->wait_fd
= -1;
1041 CDS_INIT_LIST_HEAD(&channel
->streams
.head
);
1043 DBG("Allocated channel (key %" PRIu64
")", channel
->key
);
1050 * Add a channel to the global list protected by a mutex.
1052 * Always return 0 indicating success.
1054 int consumer_add_channel(struct lttng_consumer_channel
*channel
,
1055 struct lttng_consumer_local_data
*ctx
)
1057 pthread_mutex_lock(&consumer_data
.lock
);
1058 pthread_mutex_lock(&channel
->lock
);
1059 pthread_mutex_lock(&channel
->timer_lock
);
1062 * This gives us a guarantee that the channel we are about to add to the
1063 * channel hash table will be unique. See this function comment on the why
1064 * we need to steel the channel key at this stage.
1066 steal_channel_key(channel
->key
);
1069 lttng_ht_add_unique_u64(consumer_data
.channel_ht
, &channel
->node
);
1072 pthread_mutex_unlock(&channel
->timer_lock
);
1073 pthread_mutex_unlock(&channel
->lock
);
1074 pthread_mutex_unlock(&consumer_data
.lock
);
1076 if (channel
->wait_fd
!= -1 && channel
->type
== CONSUMER_CHANNEL_TYPE_DATA
) {
1077 notify_channel_pipe(ctx
, channel
, -1, CONSUMER_CHANNEL_ADD
);
1084 * Allocate the pollfd structure and the local view of the out fds to avoid
1085 * doing a lookup in the linked list and concurrency issues when writing is
1086 * needed. Called with consumer_data.lock held.
1088 * Returns the number of fds in the structures.
1090 static int update_poll_array(struct lttng_consumer_local_data
*ctx
,
1091 struct pollfd
**pollfd
, struct lttng_consumer_stream
**local_stream
,
1092 struct lttng_ht
*ht
, int *nb_inactive_fd
)
1095 struct lttng_ht_iter iter
;
1096 struct lttng_consumer_stream
*stream
;
1101 assert(local_stream
);
1103 DBG("Updating poll fd array");
1104 *nb_inactive_fd
= 0;
1106 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1108 * Only active streams with an active end point can be added to the
1109 * poll set and local stream storage of the thread.
1111 * There is a potential race here for endpoint_status to be updated
1112 * just after the check. However, this is OK since the stream(s) will
1113 * be deleted once the thread is notified that the end point state has
1114 * changed where this function will be called back again.
1116 * We track the number of inactive FDs because they still need to be
1117 * closed by the polling thread after a wakeup on the data_pipe or
1120 if (stream
->state
!= LTTNG_CONSUMER_ACTIVE_STREAM
||
1121 stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
1122 (*nb_inactive_fd
)++;
1126 * This clobbers way too much the debug output. Uncomment that if you
1127 * need it for debugging purposes.
1129 * DBG("Active FD %d", stream->wait_fd);
1131 (*pollfd
)[i
].fd
= stream
->wait_fd
;
1132 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1133 local_stream
[i
] = stream
;
1139 * Insert the consumer_data_pipe at the end of the array and don't
1140 * increment i so nb_fd is the number of real FD.
1142 (*pollfd
)[i
].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_pipe
);
1143 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1145 (*pollfd
)[i
+ 1].fd
= lttng_pipe_get_readfd(ctx
->consumer_wakeup_pipe
);
1146 (*pollfd
)[i
+ 1].events
= POLLIN
| POLLPRI
;
1151 * Poll on the should_quit pipe and the command socket return -1 on
1152 * error, 1 if should exit, 0 if data is available on the command socket
1154 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
1159 num_rdy
= poll(consumer_sockpoll
, 2, -1);
1160 if (num_rdy
== -1) {
1162 * Restart interrupted system call.
1164 if (errno
== EINTR
) {
1167 PERROR("Poll error");
1170 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
1171 DBG("consumer_should_quit wake up");
1178 * Set the error socket.
1180 void lttng_consumer_set_error_sock(struct lttng_consumer_local_data
*ctx
,
1183 ctx
->consumer_error_socket
= sock
;
1187 * Set the command socket path.
1189 void lttng_consumer_set_command_sock_path(
1190 struct lttng_consumer_local_data
*ctx
, char *sock
)
1192 ctx
->consumer_command_sock_path
= sock
;
1196 * Send return code to the session daemon.
1197 * If the socket is not defined, we return 0, it is not a fatal error
1199 int lttng_consumer_send_error(struct lttng_consumer_local_data
*ctx
, int cmd
)
1201 if (ctx
->consumer_error_socket
> 0) {
1202 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1203 sizeof(enum lttcomm_sessiond_command
));
1210 * Close all the tracefiles and stream fds and MUST be called when all
1211 * instances are destroyed i.e. when all threads were joined and are ended.
1213 void lttng_consumer_cleanup(void)
1215 struct lttng_ht_iter iter
;
1216 struct lttng_consumer_channel
*channel
;
1220 cds_lfht_for_each_entry(consumer_data
.channel_ht
->ht
, &iter
.iter
, channel
,
1222 consumer_del_channel(channel
);
1227 lttng_ht_destroy(consumer_data
.channel_ht
);
1229 cleanup_relayd_ht();
1231 lttng_ht_destroy(consumer_data
.stream_per_chan_id_ht
);
1234 * This HT contains streams that are freed by either the metadata thread or
1235 * the data thread so we do *nothing* on the hash table and simply destroy
1238 lttng_ht_destroy(consumer_data
.stream_list_ht
);
1242 * Called from signal handler.
1244 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1248 CMM_STORE_SHARED(consumer_quit
, 1);
1249 ret
= lttng_write(ctx
->consumer_should_quit
[1], "4", 1);
1251 PERROR("write consumer quit");
1254 DBG("Consumer flag that it should quit");
1259 * Flush pending writes to trace output disk file.
1262 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1266 int outfd
= stream
->out_fd
;
1269 * This does a blocking write-and-wait on any page that belongs to the
1270 * subbuffer prior to the one we just wrote.
1271 * Don't care about error values, as these are just hints and ways to
1272 * limit the amount of page cache used.
1274 if (orig_offset
< stream
->max_sb_size
) {
1277 lttng_sync_file_range(outfd
, orig_offset
- stream
->max_sb_size
,
1278 stream
->max_sb_size
,
1279 SYNC_FILE_RANGE_WAIT_BEFORE
1280 | SYNC_FILE_RANGE_WRITE
1281 | SYNC_FILE_RANGE_WAIT_AFTER
);
1283 * Give hints to the kernel about how we access the file:
1284 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1287 * We need to call fadvise again after the file grows because the
1288 * kernel does not seem to apply fadvise to non-existing parts of the
1291 * Call fadvise _after_ having waited for the page writeback to
1292 * complete because the dirty page writeback semantic is not well
1293 * defined. So it can be expected to lead to lower throughput in
1296 ret
= posix_fadvise(outfd
, orig_offset
- stream
->max_sb_size
,
1297 stream
->max_sb_size
, POSIX_FADV_DONTNEED
);
1298 if (ret
&& ret
!= -ENOSYS
) {
1300 PERROR("posix_fadvise on fd %i", outfd
);
1305 * Initialise the necessary environnement :
1306 * - create a new context
1307 * - create the poll_pipe
1308 * - create the should_quit pipe (for signal handler)
1309 * - create the thread pipe (for splice)
1311 * Takes a function pointer as argument, this function is called when data is
1312 * available on a buffer. This function is responsible to do the
1313 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1314 * buffer configuration and then kernctl_put_next_subbuf at the end.
1316 * Returns a pointer to the new context or NULL on error.
1318 struct lttng_consumer_local_data
*lttng_consumer_create(
1319 enum lttng_consumer_type type
,
1320 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1321 struct lttng_consumer_local_data
*ctx
),
1322 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1323 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1324 int (*update_stream
)(uint64_t stream_key
, uint32_t state
))
1327 struct lttng_consumer_local_data
*ctx
;
1329 assert(consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1330 consumer_data
.type
== type
);
1331 consumer_data
.type
= type
;
1333 ctx
= zmalloc(sizeof(struct lttng_consumer_local_data
));
1335 PERROR("allocating context");
1339 ctx
->consumer_error_socket
= -1;
1340 ctx
->consumer_metadata_socket
= -1;
1341 pthread_mutex_init(&ctx
->metadata_socket_lock
, NULL
);
1342 /* assign the callbacks */
1343 ctx
->on_buffer_ready
= buffer_ready
;
1344 ctx
->on_recv_channel
= recv_channel
;
1345 ctx
->on_recv_stream
= recv_stream
;
1346 ctx
->on_update_stream
= update_stream
;
1348 ctx
->consumer_data_pipe
= lttng_pipe_open(0);
1349 if (!ctx
->consumer_data_pipe
) {
1350 goto error_poll_pipe
;
1353 ctx
->consumer_wakeup_pipe
= lttng_pipe_open(0);
1354 if (!ctx
->consumer_wakeup_pipe
) {
1355 goto error_wakeup_pipe
;
1358 ret
= pipe(ctx
->consumer_should_quit
);
1360 PERROR("Error creating recv pipe");
1361 goto error_quit_pipe
;
1364 ret
= pipe(ctx
->consumer_channel_pipe
);
1366 PERROR("Error creating channel pipe");
1367 goto error_channel_pipe
;
1370 ctx
->consumer_metadata_pipe
= lttng_pipe_open(0);
1371 if (!ctx
->consumer_metadata_pipe
) {
1372 goto error_metadata_pipe
;
1375 ctx
->channel_monitor_pipe
= -1;
1379 error_metadata_pipe
:
1380 utils_close_pipe(ctx
->consumer_channel_pipe
);
1382 utils_close_pipe(ctx
->consumer_should_quit
);
1384 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1386 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1394 * Iterate over all streams of the hashtable and free them properly.
1396 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1398 struct lttng_ht_iter iter
;
1399 struct lttng_consumer_stream
*stream
;
1406 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1408 * Ignore return value since we are currently cleaning up so any error
1411 (void) consumer_del_stream(stream
, ht
);
1415 lttng_ht_destroy(ht
);
1419 * Iterate over all streams of the metadata hashtable and free them
1422 static void destroy_metadata_stream_ht(struct lttng_ht
*ht
)
1424 struct lttng_ht_iter iter
;
1425 struct lttng_consumer_stream
*stream
;
1432 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1434 * Ignore return value since we are currently cleaning up so any error
1437 (void) consumer_del_metadata_stream(stream
, ht
);
1441 lttng_ht_destroy(ht
);
1445 * Close all fds associated with the instance and free the context.
1447 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1451 DBG("Consumer destroying it. Closing everything.");
1457 destroy_data_stream_ht(data_ht
);
1458 destroy_metadata_stream_ht(metadata_ht
);
1460 ret
= close(ctx
->consumer_error_socket
);
1464 ret
= close(ctx
->consumer_metadata_socket
);
1468 utils_close_pipe(ctx
->consumer_channel_pipe
);
1469 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1470 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1471 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1472 utils_close_pipe(ctx
->consumer_should_quit
);
1474 unlink(ctx
->consumer_command_sock_path
);
1479 * Write the metadata stream id on the specified file descriptor.
1481 static int write_relayd_metadata_id(int fd
,
1482 struct lttng_consumer_stream
*stream
,
1483 unsigned long padding
)
1486 struct lttcomm_relayd_metadata_payload hdr
;
1488 hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
1489 hdr
.padding_size
= htobe32(padding
);
1490 ret
= lttng_write(fd
, (void *) &hdr
, sizeof(hdr
));
1491 if (ret
< sizeof(hdr
)) {
1493 * This error means that the fd's end is closed so ignore the PERROR
1494 * not to clubber the error output since this can happen in a normal
1497 if (errno
!= EPIPE
) {
1498 PERROR("write metadata stream id");
1500 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno
);
1502 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1503 * handle writting the missing part so report that as an error and
1504 * don't lie to the caller.
1509 DBG("Metadata stream id %" PRIu64
" with padding %lu written before data",
1510 stream
->relayd_stream_id
, padding
);
1517 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1518 * core function for writing trace buffers to either the local filesystem or
1521 * It must be called with the stream lock held.
1523 * Careful review MUST be put if any changes occur!
1525 * Returns the number of bytes written
1527 ssize_t
lttng_consumer_on_read_subbuffer_mmap(
1528 struct lttng_consumer_local_data
*ctx
,
1529 struct lttng_consumer_stream
*stream
, unsigned long len
,
1530 unsigned long padding
,
1531 struct ctf_packet_index
*index
)
1533 unsigned long mmap_offset
;
1536 off_t orig_offset
= stream
->out_fd_offset
;
1537 /* Default is on the disk */
1538 int outfd
= stream
->out_fd
;
1539 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1540 unsigned int relayd_hang_up
= 0;
1542 /* RCU lock for the relayd pointer */
1545 /* Flag that the current stream if set for network streaming. */
1546 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1547 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1548 if (relayd
== NULL
) {
1554 /* get the offset inside the fd to mmap */
1555 switch (consumer_data
.type
) {
1556 case LTTNG_CONSUMER_KERNEL
:
1557 mmap_base
= stream
->mmap_base
;
1558 ret
= kernctl_get_mmap_read_offset(stream
->wait_fd
, &mmap_offset
);
1560 PERROR("tracer ctl get_mmap_read_offset");
1564 case LTTNG_CONSUMER32_UST
:
1565 case LTTNG_CONSUMER64_UST
:
1566 mmap_base
= lttng_ustctl_get_mmap_base(stream
);
1568 ERR("read mmap get mmap base for stream %s", stream
->name
);
1572 ret
= lttng_ustctl_get_mmap_read_offset(stream
, &mmap_offset
);
1574 PERROR("tracer ctl get_mmap_read_offset");
1580 ERR("Unknown consumer_data type");
1584 /* Handle stream on the relayd if the output is on the network */
1586 unsigned long netlen
= len
;
1589 * Lock the control socket for the complete duration of the function
1590 * since from this point on we will use the socket.
1592 if (stream
->metadata_flag
) {
1593 /* Metadata requires the control socket. */
1594 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1595 if (stream
->reset_metadata_flag
) {
1596 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1597 stream
->relayd_stream_id
,
1598 stream
->metadata_version
);
1603 stream
->reset_metadata_flag
= 0;
1605 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1608 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1613 /* Use the returned socket. */
1616 /* Write metadata stream id before payload */
1617 if (stream
->metadata_flag
) {
1618 ret
= write_relayd_metadata_id(outfd
, stream
, padding
);
1625 /* No streaming, we have to set the len with the full padding */
1628 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1629 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1631 ERR("Reset metadata file");
1634 stream
->reset_metadata_flag
= 0;
1638 * Check if we need to change the tracefile before writing the packet.
1640 if (stream
->chan
->tracefile_size
> 0 &&
1641 (stream
->tracefile_size_current
+ len
) >
1642 stream
->chan
->tracefile_size
) {
1643 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1644 stream
->name
, stream
->chan
->tracefile_size
,
1645 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1646 stream
->out_fd
, &(stream
->tracefile_count_current
),
1649 ERR("Rotating output file");
1652 outfd
= stream
->out_fd
;
1654 if (stream
->index_file
) {
1655 lttng_index_file_put(stream
->index_file
);
1656 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1657 stream
->name
, stream
->uid
, stream
->gid
,
1658 stream
->chan
->tracefile_size
,
1659 stream
->tracefile_count_current
,
1660 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1661 if (!stream
->index_file
) {
1666 /* Reset current size because we just perform a rotation. */
1667 stream
->tracefile_size_current
= 0;
1668 stream
->out_fd_offset
= 0;
1671 stream
->tracefile_size_current
+= len
;
1673 index
->offset
= htobe64(stream
->out_fd_offset
);
1678 * This call guarantee that len or less is returned. It's impossible to
1679 * receive a ret value that is bigger than len.
1681 ret
= lttng_write(outfd
, mmap_base
+ mmap_offset
, len
);
1682 DBG("Consumer mmap write() ret %zd (len %lu)", ret
, len
);
1683 if (ret
< 0 || ((size_t) ret
!= len
)) {
1685 * Report error to caller if nothing was written else at least send the
1693 /* Socket operation failed. We consider the relayd dead */
1694 if (errno
== EPIPE
|| errno
== EINVAL
|| errno
== EBADF
) {
1696 * This is possible if the fd is closed on the other side
1697 * (outfd) or any write problem. It can be verbose a bit for a
1698 * normal execution if for instance the relayd is stopped
1699 * abruptly. This can happen so set this to a DBG statement.
1701 DBG("Consumer mmap write detected relayd hang up");
1703 /* Unhandled error, print it and stop function right now. */
1704 PERROR("Error in write mmap (ret %zd != len %lu)", ret
, len
);
1708 stream
->output_written
+= ret
;
1710 /* This call is useless on a socket so better save a syscall. */
1712 /* This won't block, but will start writeout asynchronously */
1713 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, len
,
1714 SYNC_FILE_RANGE_WRITE
);
1715 stream
->out_fd_offset
+= len
;
1716 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1721 * This is a special case that the relayd has closed its socket. Let's
1722 * cleanup the relayd object and all associated streams.
1724 if (relayd
&& relayd_hang_up
) {
1725 ERR("Relayd hangup. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
1726 lttng_consumer_cleanup_relayd(relayd
);
1730 /* Unlock only if ctrl socket used */
1731 if (relayd
&& stream
->metadata_flag
) {
1732 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1740 * Splice the data from the ring buffer to the tracefile.
1742 * It must be called with the stream lock held.
1744 * Returns the number of bytes spliced.
1746 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1747 struct lttng_consumer_local_data
*ctx
,
1748 struct lttng_consumer_stream
*stream
, unsigned long len
,
1749 unsigned long padding
,
1750 struct ctf_packet_index
*index
)
1752 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1754 off_t orig_offset
= stream
->out_fd_offset
;
1755 int fd
= stream
->wait_fd
;
1756 /* Default is on the disk */
1757 int outfd
= stream
->out_fd
;
1758 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1760 unsigned int relayd_hang_up
= 0;
1762 switch (consumer_data
.type
) {
1763 case LTTNG_CONSUMER_KERNEL
:
1765 case LTTNG_CONSUMER32_UST
:
1766 case LTTNG_CONSUMER64_UST
:
1767 /* Not supported for user space tracing */
1770 ERR("Unknown consumer_data type");
1774 /* RCU lock for the relayd pointer */
1777 /* Flag that the current stream if set for network streaming. */
1778 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1779 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1780 if (relayd
== NULL
) {
1785 splice_pipe
= stream
->splice_pipe
;
1787 /* Write metadata stream id before payload */
1789 unsigned long total_len
= len
;
1791 if (stream
->metadata_flag
) {
1793 * Lock the control socket for the complete duration of the function
1794 * since from this point on we will use the socket.
1796 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1798 if (stream
->reset_metadata_flag
) {
1799 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1800 stream
->relayd_stream_id
,
1801 stream
->metadata_version
);
1806 stream
->reset_metadata_flag
= 0;
1808 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
,
1816 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1819 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1825 /* Use the returned socket. */
1828 /* No streaming, we have to set the len with the full padding */
1831 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1832 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1834 ERR("Reset metadata file");
1837 stream
->reset_metadata_flag
= 0;
1840 * Check if we need to change the tracefile before writing the packet.
1842 if (stream
->chan
->tracefile_size
> 0 &&
1843 (stream
->tracefile_size_current
+ len
) >
1844 stream
->chan
->tracefile_size
) {
1845 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1846 stream
->name
, stream
->chan
->tracefile_size
,
1847 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1848 stream
->out_fd
, &(stream
->tracefile_count_current
),
1852 ERR("Rotating output file");
1855 outfd
= stream
->out_fd
;
1857 if (stream
->index_file
) {
1858 lttng_index_file_put(stream
->index_file
);
1859 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1860 stream
->name
, stream
->uid
, stream
->gid
,
1861 stream
->chan
->tracefile_size
,
1862 stream
->tracefile_count_current
,
1863 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1864 if (!stream
->index_file
) {
1869 /* Reset current size because we just perform a rotation. */
1870 stream
->tracefile_size_current
= 0;
1871 stream
->out_fd_offset
= 0;
1874 stream
->tracefile_size_current
+= len
;
1875 index
->offset
= htobe64(stream
->out_fd_offset
);
1879 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1880 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1881 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1882 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1883 DBG("splice chan to pipe, ret %zd", ret_splice
);
1884 if (ret_splice
< 0) {
1887 PERROR("Error in relay splice");
1891 /* Handle stream on the relayd if the output is on the network */
1892 if (relayd
&& stream
->metadata_flag
) {
1893 size_t metadata_payload_size
=
1894 sizeof(struct lttcomm_relayd_metadata_payload
);
1896 /* Update counter to fit the spliced data */
1897 ret_splice
+= metadata_payload_size
;
1898 len
+= metadata_payload_size
;
1900 * We do this so the return value can match the len passed as
1901 * argument to this function.
1903 written
-= metadata_payload_size
;
1906 /* Splice data out */
1907 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1908 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1909 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1911 if (ret_splice
< 0) {
1916 } else if (ret_splice
> len
) {
1918 * We don't expect this code path to be executed but you never know
1919 * so this is an extra protection agains a buggy splice().
1922 written
+= ret_splice
;
1923 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice
,
1927 /* All good, update current len and continue. */
1931 /* This call is useless on a socket so better save a syscall. */
1933 /* This won't block, but will start writeout asynchronously */
1934 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1935 SYNC_FILE_RANGE_WRITE
);
1936 stream
->out_fd_offset
+= ret_splice
;
1938 stream
->output_written
+= ret_splice
;
1939 written
+= ret_splice
;
1942 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1948 * This is a special case that the relayd has closed its socket. Let's
1949 * cleanup the relayd object and all associated streams.
1951 if (relayd
&& relayd_hang_up
) {
1952 ERR("Relayd hangup. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
1953 lttng_consumer_cleanup_relayd(relayd
);
1954 /* Skip splice error so the consumer does not fail */
1959 /* send the appropriate error description to sessiond */
1962 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1965 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1968 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1973 if (relayd
&& stream
->metadata_flag
) {
1974 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1982 * Sample the snapshot positions for a specific fd
1984 * Returns 0 on success, < 0 on error
1986 int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream
*stream
)
1988 switch (consumer_data
.type
) {
1989 case LTTNG_CONSUMER_KERNEL
:
1990 return lttng_kconsumer_sample_snapshot_positions(stream
);
1991 case LTTNG_CONSUMER32_UST
:
1992 case LTTNG_CONSUMER64_UST
:
1993 return lttng_ustconsumer_sample_snapshot_positions(stream
);
1995 ERR("Unknown consumer_data type");
2001 * Take a snapshot for a specific fd
2003 * Returns 0 on success, < 0 on error
2005 int lttng_consumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2007 switch (consumer_data
.type
) {
2008 case LTTNG_CONSUMER_KERNEL
:
2009 return lttng_kconsumer_take_snapshot(stream
);
2010 case LTTNG_CONSUMER32_UST
:
2011 case LTTNG_CONSUMER64_UST
:
2012 return lttng_ustconsumer_take_snapshot(stream
);
2014 ERR("Unknown consumer_data type");
2021 * Get the produced position
2023 * Returns 0 on success, < 0 on error
2025 int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
2028 switch (consumer_data
.type
) {
2029 case LTTNG_CONSUMER_KERNEL
:
2030 return lttng_kconsumer_get_produced_snapshot(stream
, pos
);
2031 case LTTNG_CONSUMER32_UST
:
2032 case LTTNG_CONSUMER64_UST
:
2033 return lttng_ustconsumer_get_produced_snapshot(stream
, pos
);
2035 ERR("Unknown consumer_data type");
2042 * Get the consumed position (free-running counter position in bytes).
2044 * Returns 0 on success, < 0 on error
2046 int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
2049 switch (consumer_data
.type
) {
2050 case LTTNG_CONSUMER_KERNEL
:
2051 return lttng_kconsumer_get_consumed_snapshot(stream
, pos
);
2052 case LTTNG_CONSUMER32_UST
:
2053 case LTTNG_CONSUMER64_UST
:
2054 return lttng_ustconsumer_get_consumed_snapshot(stream
, pos
);
2056 ERR("Unknown consumer_data type");
2062 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
2063 int sock
, struct pollfd
*consumer_sockpoll
)
2065 switch (consumer_data
.type
) {
2066 case LTTNG_CONSUMER_KERNEL
:
2067 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2068 case LTTNG_CONSUMER32_UST
:
2069 case LTTNG_CONSUMER64_UST
:
2070 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2072 ERR("Unknown consumer_data type");
2078 void lttng_consumer_close_all_metadata(void)
2080 switch (consumer_data
.type
) {
2081 case LTTNG_CONSUMER_KERNEL
:
2083 * The Kernel consumer has a different metadata scheme so we don't
2084 * close anything because the stream will be closed by the session
2088 case LTTNG_CONSUMER32_UST
:
2089 case LTTNG_CONSUMER64_UST
:
2091 * Close all metadata streams. The metadata hash table is passed and
2092 * this call iterates over it by closing all wakeup fd. This is safe
2093 * because at this point we are sure that the metadata producer is
2094 * either dead or blocked.
2096 lttng_ustconsumer_close_all_metadata(metadata_ht
);
2099 ERR("Unknown consumer_data type");
2105 * Clean up a metadata stream and free its memory.
2107 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
2108 struct lttng_ht
*ht
)
2110 struct lttng_consumer_channel
*free_chan
= NULL
;
2114 * This call should NEVER receive regular stream. It must always be
2115 * metadata stream and this is crucial for data structure synchronization.
2117 assert(stream
->metadata_flag
);
2119 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
2121 pthread_mutex_lock(&consumer_data
.lock
);
2122 pthread_mutex_lock(&stream
->chan
->lock
);
2123 pthread_mutex_lock(&stream
->lock
);
2124 if (stream
->chan
->metadata_cache
) {
2125 /* Only applicable to userspace consumers. */
2126 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2129 /* Remove any reference to that stream. */
2130 consumer_stream_delete(stream
, ht
);
2132 /* Close down everything including the relayd if one. */
2133 consumer_stream_close(stream
);
2134 /* Destroy tracer buffers of the stream. */
2135 consumer_stream_destroy_buffers(stream
);
2137 /* Atomically decrement channel refcount since other threads can use it. */
2138 if (!uatomic_sub_return(&stream
->chan
->refcount
, 1)
2139 && !uatomic_read(&stream
->chan
->nb_init_stream_left
)) {
2140 /* Go for channel deletion! */
2141 free_chan
= stream
->chan
;
2145 * Nullify the stream reference so it is not used after deletion. The
2146 * channel lock MUST be acquired before being able to check for a NULL
2149 stream
->chan
->metadata_stream
= NULL
;
2151 if (stream
->chan
->metadata_cache
) {
2152 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2154 pthread_mutex_unlock(&stream
->lock
);
2155 pthread_mutex_unlock(&stream
->chan
->lock
);
2156 pthread_mutex_unlock(&consumer_data
.lock
);
2159 consumer_del_channel(free_chan
);
2162 consumer_stream_free(stream
);
2166 * Action done with the metadata stream when adding it to the consumer internal
2167 * data structures to handle it.
2169 void consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
)
2171 struct lttng_ht
*ht
= metadata_ht
;
2172 struct lttng_ht_iter iter
;
2173 struct lttng_ht_node_u64
*node
;
2178 DBG3("Adding metadata stream %" PRIu64
" to hash table", stream
->key
);
2180 pthread_mutex_lock(&consumer_data
.lock
);
2181 pthread_mutex_lock(&stream
->chan
->lock
);
2182 pthread_mutex_lock(&stream
->chan
->timer_lock
);
2183 pthread_mutex_lock(&stream
->lock
);
2186 * From here, refcounts are updated so be _careful_ when returning an error
2193 * Lookup the stream just to make sure it does not exist in our internal
2194 * state. This should NEVER happen.
2196 lttng_ht_lookup(ht
, &stream
->key
, &iter
);
2197 node
= lttng_ht_iter_get_node_u64(&iter
);
2201 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2202 * in terms of destroying the associated channel, because the action that
2203 * causes the count to become 0 also causes a stream to be added. The
2204 * channel deletion will thus be triggered by the following removal of this
2207 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
2208 /* Increment refcount before decrementing nb_init_stream_left */
2210 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
2213 lttng_ht_add_unique_u64(ht
, &stream
->node
);
2215 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
2216 &stream
->node_channel_id
);
2219 * Add stream to the stream_list_ht of the consumer data. No need to steal
2220 * the key since the HT does not use it and we allow to add redundant keys
2223 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
2227 pthread_mutex_unlock(&stream
->lock
);
2228 pthread_mutex_unlock(&stream
->chan
->lock
);
2229 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
2230 pthread_mutex_unlock(&consumer_data
.lock
);
2234 * Delete data stream that are flagged for deletion (endpoint_status).
2236 static void validate_endpoint_status_data_stream(void)
2238 struct lttng_ht_iter iter
;
2239 struct lttng_consumer_stream
*stream
;
2241 DBG("Consumer delete flagged data stream");
2244 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2245 /* Validate delete flag of the stream */
2246 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2249 /* Delete it right now */
2250 consumer_del_stream(stream
, data_ht
);
2256 * Delete metadata stream that are flagged for deletion (endpoint_status).
2258 static void validate_endpoint_status_metadata_stream(
2259 struct lttng_poll_event
*pollset
)
2261 struct lttng_ht_iter iter
;
2262 struct lttng_consumer_stream
*stream
;
2264 DBG("Consumer delete flagged metadata stream");
2269 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2270 /* Validate delete flag of the stream */
2271 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2275 * Remove from pollset so the metadata thread can continue without
2276 * blocking on a deleted stream.
2278 lttng_poll_del(pollset
, stream
->wait_fd
);
2280 /* Delete it right now */
2281 consumer_del_metadata_stream(stream
, metadata_ht
);
2287 * Perform operations that need to be done after a stream has
2288 * rotated and released the stream lock.
2290 * Multiple rotations cannot occur simultaneously, so we know the state of the
2291 * "rotated" stream flag cannot change.
2293 * This MUST be called WITHOUT the stream lock held.
2296 int consumer_post_rotation(struct lttng_consumer_stream
*stream
,
2297 struct lttng_consumer_local_data
*ctx
)
2301 pthread_mutex_lock(&stream
->chan
->lock
);
2303 switch (consumer_data
.type
) {
2304 case LTTNG_CONSUMER_KERNEL
:
2306 case LTTNG_CONSUMER32_UST
:
2307 case LTTNG_CONSUMER64_UST
:
2309 * The ust_metadata_pushed counter has been reset to 0, so now
2310 * we can wakeup the metadata thread so it dumps the metadata
2311 * cache to the new file.
2313 if (stream
->metadata_flag
) {
2314 consumer_metadata_wakeup_pipe(stream
->chan
);
2318 ERR("Unknown consumer_data type");
2322 pthread_mutex_unlock(&stream
->chan
->lock
);
2327 * Thread polls on metadata file descriptor and write them on disk or on the
2330 void *consumer_thread_metadata_poll(void *data
)
2332 int ret
, i
, pollfd
, err
= -1;
2333 uint32_t revents
, nb_fd
;
2334 struct lttng_consumer_stream
*stream
= NULL
;
2335 struct lttng_ht_iter iter
;
2336 struct lttng_ht_node_u64
*node
;
2337 struct lttng_poll_event events
;
2338 struct lttng_consumer_local_data
*ctx
= data
;
2341 rcu_register_thread();
2343 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_METADATA
);
2345 if (testpoint(consumerd_thread_metadata
)) {
2346 goto error_testpoint
;
2349 health_code_update();
2351 DBG("Thread metadata poll started");
2353 /* Size is set to 1 for the consumer_metadata pipe */
2354 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2356 ERR("Poll set creation failed");
2360 ret
= lttng_poll_add(&events
,
2361 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
), LPOLLIN
);
2367 DBG("Metadata main loop started");
2371 health_code_update();
2372 health_poll_entry();
2373 DBG("Metadata poll wait");
2374 ret
= lttng_poll_wait(&events
, -1);
2375 DBG("Metadata poll return from wait with %d fd(s)",
2376 LTTNG_POLL_GETNB(&events
));
2378 DBG("Metadata event caught in thread");
2380 if (errno
== EINTR
) {
2381 ERR("Poll EINTR caught");
2384 if (LTTNG_POLL_GETNB(&events
) == 0) {
2385 err
= 0; /* All is OK */
2392 /* From here, the event is a metadata wait fd */
2393 for (i
= 0; i
< nb_fd
; i
++) {
2394 health_code_update();
2396 revents
= LTTNG_POLL_GETEV(&events
, i
);
2397 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2400 /* No activity for this FD (poll implementation). */
2404 if (pollfd
== lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
)) {
2405 if (revents
& LPOLLIN
) {
2408 pipe_len
= lttng_pipe_read(ctx
->consumer_metadata_pipe
,
2409 &stream
, sizeof(stream
));
2410 if (pipe_len
< sizeof(stream
)) {
2412 PERROR("read metadata stream");
2415 * Remove the pipe from the poll set and continue the loop
2416 * since their might be data to consume.
2418 lttng_poll_del(&events
,
2419 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2420 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2424 /* A NULL stream means that the state has changed. */
2425 if (stream
== NULL
) {
2426 /* Check for deleted streams. */
2427 validate_endpoint_status_metadata_stream(&events
);
2431 DBG("Adding metadata stream %d to poll set",
2434 /* Add metadata stream to the global poll events list */
2435 lttng_poll_add(&events
, stream
->wait_fd
,
2436 LPOLLIN
| LPOLLPRI
| LPOLLHUP
);
2437 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2438 DBG("Metadata thread pipe hung up");
2440 * Remove the pipe from the poll set and continue the loop
2441 * since their might be data to consume.
2443 lttng_poll_del(&events
,
2444 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2445 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2448 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2452 /* Handle other stream */
2458 uint64_t tmp_id
= (uint64_t) pollfd
;
2460 lttng_ht_lookup(metadata_ht
, &tmp_id
, &iter
);
2462 node
= lttng_ht_iter_get_node_u64(&iter
);
2465 stream
= caa_container_of(node
, struct lttng_consumer_stream
,
2468 if (revents
& (LPOLLIN
| LPOLLPRI
)) {
2469 /* Get the data out of the metadata file descriptor */
2470 DBG("Metadata available on fd %d", pollfd
);
2471 assert(stream
->wait_fd
== pollfd
);
2474 health_code_update();
2476 len
= ctx
->on_buffer_ready(stream
, ctx
);
2478 * We don't check the return value here since if we get
2479 * a negative len, it means an error occurred thus we
2480 * simply remove it from the poll set and free the
2485 /* It's ok to have an unavailable sub-buffer */
2486 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2487 /* Clean up stream from consumer and free it. */
2488 lttng_poll_del(&events
, stream
->wait_fd
);
2489 consumer_del_metadata_stream(stream
, metadata_ht
);
2491 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2492 DBG("Metadata fd %d is hup|err.", pollfd
);
2493 if (!stream
->hangup_flush_done
2494 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2495 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2496 DBG("Attempting to flush and consume the UST buffers");
2497 lttng_ustconsumer_on_stream_hangup(stream
);
2499 /* We just flushed the stream now read it. */
2501 health_code_update();
2503 len
= ctx
->on_buffer_ready(stream
, ctx
);
2505 * We don't check the return value here since if we get
2506 * a negative len, it means an error occurred thus we
2507 * simply remove it from the poll set and free the
2513 lttng_poll_del(&events
, stream
->wait_fd
);
2515 * This call update the channel states, closes file descriptors
2516 * and securely free the stream.
2518 consumer_del_metadata_stream(stream
, metadata_ht
);
2520 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2524 /* Release RCU lock for the stream looked up */
2532 DBG("Metadata poll thread exiting");
2534 lttng_poll_clean(&events
);
2539 ERR("Health error occurred in %s", __func__
);
2541 health_unregister(health_consumerd
);
2542 rcu_unregister_thread();
2547 * This thread polls the fds in the set to consume the data and write
2548 * it to tracefile if necessary.
2550 void *consumer_thread_data_poll(void *data
)
2552 int num_rdy
, num_hup
, high_prio
, ret
, i
, err
= -1;
2553 struct pollfd
*pollfd
= NULL
;
2554 /* local view of the streams */
2555 struct lttng_consumer_stream
**local_stream
= NULL
, *new_stream
= NULL
;
2556 /* local view of consumer_data.fds_count */
2558 /* 2 for the consumer_data_pipe and wake up pipe */
2559 const int nb_pipes_fd
= 2;
2560 /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
2561 int nb_inactive_fd
= 0;
2562 struct lttng_consumer_local_data
*ctx
= data
;
2565 rcu_register_thread();
2567 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_DATA
);
2569 if (testpoint(consumerd_thread_data
)) {
2570 goto error_testpoint
;
2573 health_code_update();
2575 local_stream
= zmalloc(sizeof(struct lttng_consumer_stream
*));
2576 if (local_stream
== NULL
) {
2577 PERROR("local_stream malloc");
2582 health_code_update();
2588 * the fds set has been updated, we need to update our
2589 * local array as well
2591 pthread_mutex_lock(&consumer_data
.lock
);
2592 if (consumer_data
.need_update
) {
2597 local_stream
= NULL
;
2599 /* Allocate for all fds */
2600 pollfd
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) * sizeof(struct pollfd
));
2601 if (pollfd
== NULL
) {
2602 PERROR("pollfd malloc");
2603 pthread_mutex_unlock(&consumer_data
.lock
);
2607 local_stream
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) *
2608 sizeof(struct lttng_consumer_stream
*));
2609 if (local_stream
== NULL
) {
2610 PERROR("local_stream malloc");
2611 pthread_mutex_unlock(&consumer_data
.lock
);
2614 ret
= update_poll_array(ctx
, &pollfd
, local_stream
,
2615 data_ht
, &nb_inactive_fd
);
2617 ERR("Error in allocating pollfd or local_outfds");
2618 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2619 pthread_mutex_unlock(&consumer_data
.lock
);
2623 consumer_data
.need_update
= 0;
2625 pthread_mutex_unlock(&consumer_data
.lock
);
2627 /* No FDs and consumer_quit, consumer_cleanup the thread */
2628 if (nb_fd
== 0 && nb_inactive_fd
== 0 &&
2629 CMM_LOAD_SHARED(consumer_quit
) == 1) {
2630 err
= 0; /* All is OK */
2633 /* poll on the array of fds */
2635 DBG("polling on %d fd", nb_fd
+ nb_pipes_fd
);
2636 if (testpoint(consumerd_thread_data_poll
)) {
2639 health_poll_entry();
2640 num_rdy
= poll(pollfd
, nb_fd
+ nb_pipes_fd
, -1);
2642 DBG("poll num_rdy : %d", num_rdy
);
2643 if (num_rdy
== -1) {
2645 * Restart interrupted system call.
2647 if (errno
== EINTR
) {
2650 PERROR("Poll error");
2651 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2653 } else if (num_rdy
== 0) {
2654 DBG("Polling thread timed out");
2658 if (caa_unlikely(data_consumption_paused
)) {
2659 DBG("Data consumption paused, sleeping...");
2665 * If the consumer_data_pipe triggered poll go directly to the
2666 * beginning of the loop to update the array. We want to prioritize
2667 * array update over low-priority reads.
2669 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2670 ssize_t pipe_readlen
;
2672 DBG("consumer_data_pipe wake up");
2673 pipe_readlen
= lttng_pipe_read(ctx
->consumer_data_pipe
,
2674 &new_stream
, sizeof(new_stream
));
2675 if (pipe_readlen
< sizeof(new_stream
)) {
2676 PERROR("Consumer data pipe");
2677 /* Continue so we can at least handle the current stream(s). */
2682 * If the stream is NULL, just ignore it. It's also possible that
2683 * the sessiond poll thread changed the consumer_quit state and is
2684 * waking us up to test it.
2686 if (new_stream
== NULL
) {
2687 validate_endpoint_status_data_stream();
2691 /* Continue to update the local streams and handle prio ones */
2695 /* Handle wakeup pipe. */
2696 if (pollfd
[nb_fd
+ 1].revents
& (POLLIN
| POLLPRI
)) {
2698 ssize_t pipe_readlen
;
2700 pipe_readlen
= lttng_pipe_read(ctx
->consumer_wakeup_pipe
, &dummy
,
2702 if (pipe_readlen
< 0) {
2703 PERROR("Consumer data wakeup pipe");
2705 /* We've been awakened to handle stream(s). */
2706 ctx
->has_wakeup
= 0;
2709 /* Take care of high priority channels first. */
2710 for (i
= 0; i
< nb_fd
; i
++) {
2711 health_code_update();
2713 if (local_stream
[i
] == NULL
) {
2716 if (pollfd
[i
].revents
& POLLPRI
) {
2717 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2719 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2720 /* it's ok to have an unavailable sub-buffer */
2721 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2722 /* Clean the stream and free it. */
2723 consumer_del_stream(local_stream
[i
], data_ht
);
2724 local_stream
[i
] = NULL
;
2725 } else if (len
> 0) {
2726 local_stream
[i
]->data_read
= 1;
2732 * If we read high prio channel in this loop, try again
2733 * for more high prio data.
2739 /* Take care of low priority channels. */
2740 for (i
= 0; i
< nb_fd
; i
++) {
2741 health_code_update();
2743 if (local_stream
[i
] == NULL
) {
2746 if ((pollfd
[i
].revents
& POLLIN
) ||
2747 local_stream
[i
]->hangup_flush_done
||
2748 local_stream
[i
]->has_data
) {
2749 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2750 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2751 /* it's ok to have an unavailable sub-buffer */
2752 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2753 /* Clean the stream and free it. */
2754 consumer_del_stream(local_stream
[i
], data_ht
);
2755 local_stream
[i
] = NULL
;
2756 } else if (len
> 0) {
2757 local_stream
[i
]->data_read
= 1;
2762 /* Handle hangup and errors */
2763 for (i
= 0; i
< nb_fd
; i
++) {
2764 health_code_update();
2766 if (local_stream
[i
] == NULL
) {
2769 if (!local_stream
[i
]->hangup_flush_done
2770 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2771 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2772 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2773 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2775 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2776 /* Attempt read again, for the data we just flushed. */
2777 local_stream
[i
]->data_read
= 1;
2780 * If the poll flag is HUP/ERR/NVAL and we have
2781 * read no data in this pass, we can remove the
2782 * stream from its hash table.
2784 if ((pollfd
[i
].revents
& POLLHUP
)) {
2785 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2786 if (!local_stream
[i
]->data_read
) {
2787 consumer_del_stream(local_stream
[i
], data_ht
);
2788 local_stream
[i
] = NULL
;
2791 } else if (pollfd
[i
].revents
& POLLERR
) {
2792 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2793 if (!local_stream
[i
]->data_read
) {
2794 consumer_del_stream(local_stream
[i
], data_ht
);
2795 local_stream
[i
] = NULL
;
2798 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2799 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2800 if (!local_stream
[i
]->data_read
) {
2801 consumer_del_stream(local_stream
[i
], data_ht
);
2802 local_stream
[i
] = NULL
;
2806 if (local_stream
[i
] != NULL
) {
2807 local_stream
[i
]->data_read
= 0;
2814 DBG("polling thread exiting");
2819 * Close the write side of the pipe so epoll_wait() in
2820 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2821 * read side of the pipe. If we close them both, epoll_wait strangely does
2822 * not return and could create a endless wait period if the pipe is the
2823 * only tracked fd in the poll set. The thread will take care of closing
2826 (void) lttng_pipe_write_close(ctx
->consumer_metadata_pipe
);
2831 ERR("Health error occurred in %s", __func__
);
2833 health_unregister(health_consumerd
);
2835 rcu_unregister_thread();
2840 * Close wake-up end of each stream belonging to the channel. This will
2841 * allow the poll() on the stream read-side to detect when the
2842 * write-side (application) finally closes them.
2845 void consumer_close_channel_streams(struct lttng_consumer_channel
*channel
)
2847 struct lttng_ht
*ht
;
2848 struct lttng_consumer_stream
*stream
;
2849 struct lttng_ht_iter iter
;
2851 ht
= consumer_data
.stream_per_chan_id_ht
;
2854 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2855 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
2856 ht
->match_fct
, &channel
->key
,
2857 &iter
.iter
, stream
, node_channel_id
.node
) {
2859 * Protect against teardown with mutex.
2861 pthread_mutex_lock(&stream
->lock
);
2862 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
2865 switch (consumer_data
.type
) {
2866 case LTTNG_CONSUMER_KERNEL
:
2868 case LTTNG_CONSUMER32_UST
:
2869 case LTTNG_CONSUMER64_UST
:
2870 if (stream
->metadata_flag
) {
2871 /* Safe and protected by the stream lock. */
2872 lttng_ustconsumer_close_metadata(stream
->chan
);
2875 * Note: a mutex is taken internally within
2876 * liblttng-ust-ctl to protect timer wakeup_fd
2877 * use from concurrent close.
2879 lttng_ustconsumer_close_stream_wakeup(stream
);
2883 ERR("Unknown consumer_data type");
2887 pthread_mutex_unlock(&stream
->lock
);
2892 static void destroy_channel_ht(struct lttng_ht
*ht
)
2894 struct lttng_ht_iter iter
;
2895 struct lttng_consumer_channel
*channel
;
2903 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, channel
, wait_fd_node
.node
) {
2904 ret
= lttng_ht_del(ht
, &iter
);
2909 lttng_ht_destroy(ht
);
2913 * This thread polls the channel fds to detect when they are being
2914 * closed. It closes all related streams if the channel is detected as
2915 * closed. It is currently only used as a shim layer for UST because the
2916 * consumerd needs to keep the per-stream wakeup end of pipes open for
2919 void *consumer_thread_channel_poll(void *data
)
2921 int ret
, i
, pollfd
, err
= -1;
2922 uint32_t revents
, nb_fd
;
2923 struct lttng_consumer_channel
*chan
= NULL
;
2924 struct lttng_ht_iter iter
;
2925 struct lttng_ht_node_u64
*node
;
2926 struct lttng_poll_event events
;
2927 struct lttng_consumer_local_data
*ctx
= data
;
2928 struct lttng_ht
*channel_ht
;
2930 rcu_register_thread();
2932 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_CHANNEL
);
2934 if (testpoint(consumerd_thread_channel
)) {
2935 goto error_testpoint
;
2938 health_code_update();
2940 channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2942 /* ENOMEM at this point. Better to bail out. */
2946 DBG("Thread channel poll started");
2948 /* Size is set to 1 for the consumer_channel pipe */
2949 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2951 ERR("Poll set creation failed");
2955 ret
= lttng_poll_add(&events
, ctx
->consumer_channel_pipe
[0], LPOLLIN
);
2961 DBG("Channel main loop started");
2965 health_code_update();
2966 DBG("Channel poll wait");
2967 health_poll_entry();
2968 ret
= lttng_poll_wait(&events
, -1);
2969 DBG("Channel poll return from wait with %d fd(s)",
2970 LTTNG_POLL_GETNB(&events
));
2972 DBG("Channel event caught in thread");
2974 if (errno
== EINTR
) {
2975 ERR("Poll EINTR caught");
2978 if (LTTNG_POLL_GETNB(&events
) == 0) {
2979 err
= 0; /* All is OK */
2986 /* From here, the event is a channel wait fd */
2987 for (i
= 0; i
< nb_fd
; i
++) {
2988 health_code_update();
2990 revents
= LTTNG_POLL_GETEV(&events
, i
);
2991 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2994 /* No activity for this FD (poll implementation). */
2998 if (pollfd
== ctx
->consumer_channel_pipe
[0]) {
2999 if (revents
& LPOLLIN
) {
3000 enum consumer_channel_action action
;
3003 ret
= read_channel_pipe(ctx
, &chan
, &key
, &action
);
3006 ERR("Error reading channel pipe");
3008 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3013 case CONSUMER_CHANNEL_ADD
:
3014 DBG("Adding channel %d to poll set",
3017 lttng_ht_node_init_u64(&chan
->wait_fd_node
,
3020 lttng_ht_add_unique_u64(channel_ht
,
3021 &chan
->wait_fd_node
);
3023 /* Add channel to the global poll events list */
3024 lttng_poll_add(&events
, chan
->wait_fd
,
3025 LPOLLERR
| LPOLLHUP
);
3027 case CONSUMER_CHANNEL_DEL
:
3030 * This command should never be called if the channel
3031 * has streams monitored by either the data or metadata
3032 * thread. The consumer only notify this thread with a
3033 * channel del. command if it receives a destroy
3034 * channel command from the session daemon that send it
3035 * if a command prior to the GET_CHANNEL failed.
3039 chan
= consumer_find_channel(key
);
3042 ERR("UST consumer get channel key %" PRIu64
" not found for del channel", key
);
3045 lttng_poll_del(&events
, chan
->wait_fd
);
3046 iter
.iter
.node
= &chan
->wait_fd_node
.node
;
3047 ret
= lttng_ht_del(channel_ht
, &iter
);
3050 switch (consumer_data
.type
) {
3051 case LTTNG_CONSUMER_KERNEL
:
3053 case LTTNG_CONSUMER32_UST
:
3054 case LTTNG_CONSUMER64_UST
:
3055 health_code_update();
3056 /* Destroy streams that might have been left in the stream list. */
3057 clean_channel_stream_list(chan
);
3060 ERR("Unknown consumer_data type");
3065 * Release our own refcount. Force channel deletion even if
3066 * streams were not initialized.
3068 if (!uatomic_sub_return(&chan
->refcount
, 1)) {
3069 consumer_del_channel(chan
);
3074 case CONSUMER_CHANNEL_QUIT
:
3076 * Remove the pipe from the poll set and continue the loop
3077 * since their might be data to consume.
3079 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3082 ERR("Unknown action");
3085 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3086 DBG("Channel thread pipe hung up");
3088 * Remove the pipe from the poll set and continue the loop
3089 * since their might be data to consume.
3091 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3094 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3098 /* Handle other stream */
3104 uint64_t tmp_id
= (uint64_t) pollfd
;
3106 lttng_ht_lookup(channel_ht
, &tmp_id
, &iter
);
3108 node
= lttng_ht_iter_get_node_u64(&iter
);
3111 chan
= caa_container_of(node
, struct lttng_consumer_channel
,
3114 /* Check for error event */
3115 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3116 DBG("Channel fd %d is hup|err.", pollfd
);
3118 lttng_poll_del(&events
, chan
->wait_fd
);
3119 ret
= lttng_ht_del(channel_ht
, &iter
);
3123 * This will close the wait fd for each stream associated to
3124 * this channel AND monitored by the data/metadata thread thus
3125 * will be clean by the right thread.
3127 consumer_close_channel_streams(chan
);
3129 /* Release our own refcount */
3130 if (!uatomic_sub_return(&chan
->refcount
, 1)
3131 && !uatomic_read(&chan
->nb_init_stream_left
)) {
3132 consumer_del_channel(chan
);
3135 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3140 /* Release RCU lock for the channel looked up */
3148 lttng_poll_clean(&events
);
3150 destroy_channel_ht(channel_ht
);
3153 DBG("Channel poll thread exiting");
3156 ERR("Health error occurred in %s", __func__
);
3158 health_unregister(health_consumerd
);
3159 rcu_unregister_thread();
3163 static int set_metadata_socket(struct lttng_consumer_local_data
*ctx
,
3164 struct pollfd
*sockpoll
, int client_socket
)
3171 ret
= lttng_consumer_poll_socket(sockpoll
);
3175 DBG("Metadata connection on client_socket");
3177 /* Blocking call, waiting for transmission */
3178 ctx
->consumer_metadata_socket
= lttcomm_accept_unix_sock(client_socket
);
3179 if (ctx
->consumer_metadata_socket
< 0) {
3180 WARN("On accept metadata");
3191 * This thread listens on the consumerd socket and receives the file
3192 * descriptors from the session daemon.
3194 void *consumer_thread_sessiond_poll(void *data
)
3196 int sock
= -1, client_socket
, ret
, err
= -1;
3198 * structure to poll for incoming data on communication socket avoids
3199 * making blocking sockets.
3201 struct pollfd consumer_sockpoll
[2];
3202 struct lttng_consumer_local_data
*ctx
= data
;
3204 rcu_register_thread();
3206 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_SESSIOND
);
3208 if (testpoint(consumerd_thread_sessiond
)) {
3209 goto error_testpoint
;
3212 health_code_update();
3214 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
3215 unlink(ctx
->consumer_command_sock_path
);
3216 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
3217 if (client_socket
< 0) {
3218 ERR("Cannot create command socket");
3222 ret
= lttcomm_listen_unix_sock(client_socket
);
3227 DBG("Sending ready command to lttng-sessiond");
3228 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
3229 /* return < 0 on error, but == 0 is not fatal */
3231 ERR("Error sending ready command to lttng-sessiond");
3235 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3236 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
3237 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
3238 consumer_sockpoll
[1].fd
= client_socket
;
3239 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3241 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3249 DBG("Connection on client_socket");
3251 /* Blocking call, waiting for transmission */
3252 sock
= lttcomm_accept_unix_sock(client_socket
);
3259 * Setup metadata socket which is the second socket connection on the
3260 * command unix socket.
3262 ret
= set_metadata_socket(ctx
, consumer_sockpoll
, client_socket
);
3271 /* This socket is not useful anymore. */
3272 ret
= close(client_socket
);
3274 PERROR("close client_socket");
3278 /* update the polling structure to poll on the established socket */
3279 consumer_sockpoll
[1].fd
= sock
;
3280 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3283 health_code_update();
3285 health_poll_entry();
3286 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3295 DBG("Incoming command on sock");
3296 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
3299 * This could simply be a session daemon quitting. Don't output
3302 DBG("Communication interrupted on command socket");
3306 if (CMM_LOAD_SHARED(consumer_quit
)) {
3307 DBG("consumer_thread_receive_fds received quit from signal");
3308 err
= 0; /* All is OK */
3311 DBG("received command on sock");
3317 DBG("Consumer thread sessiond poll exiting");
3320 * Close metadata streams since the producer is the session daemon which
3323 * NOTE: for now, this only applies to the UST tracer.
3325 lttng_consumer_close_all_metadata();
3328 * when all fds have hung up, the polling thread
3331 CMM_STORE_SHARED(consumer_quit
, 1);
3334 * Notify the data poll thread to poll back again and test the
3335 * consumer_quit state that we just set so to quit gracefully.
3337 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
3339 notify_channel_pipe(ctx
, NULL
, -1, CONSUMER_CHANNEL_QUIT
);
3341 notify_health_quit_pipe(health_quit_pipe
);
3343 /* Cleaning up possibly open sockets. */
3347 PERROR("close sock sessiond poll");
3350 if (client_socket
>= 0) {
3351 ret
= close(client_socket
);
3353 PERROR("close client_socket sessiond poll");
3360 ERR("Health error occurred in %s", __func__
);
3362 health_unregister(health_consumerd
);
3364 rcu_unregister_thread();
3368 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
3369 struct lttng_consumer_local_data
*ctx
)
3373 bool rotated
= false;
3375 pthread_mutex_lock(&stream
->lock
);
3376 if (stream
->metadata_flag
) {
3377 pthread_mutex_lock(&stream
->metadata_rdv_lock
);
3380 switch (consumer_data
.type
) {
3381 case LTTNG_CONSUMER_KERNEL
:
3382 ret
= lttng_kconsumer_read_subbuffer(stream
, ctx
, &rotated
);
3384 case LTTNG_CONSUMER32_UST
:
3385 case LTTNG_CONSUMER64_UST
:
3386 ret
= lttng_ustconsumer_read_subbuffer(stream
, ctx
, &rotated
);
3389 ERR("Unknown consumer_data type");
3395 if (stream
->metadata_flag
) {
3396 pthread_cond_broadcast(&stream
->metadata_rdv
);
3397 pthread_mutex_unlock(&stream
->metadata_rdv_lock
);
3399 pthread_mutex_unlock(&stream
->lock
);
3401 rotate_ret
= consumer_post_rotation(stream
, ctx
);
3402 if (rotate_ret
< 0) {
3403 ERR("Failed after a rotation");
3411 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3413 switch (consumer_data
.type
) {
3414 case LTTNG_CONSUMER_KERNEL
:
3415 return lttng_kconsumer_on_recv_stream(stream
);
3416 case LTTNG_CONSUMER32_UST
:
3417 case LTTNG_CONSUMER64_UST
:
3418 return lttng_ustconsumer_on_recv_stream(stream
);
3420 ERR("Unknown consumer_data type");
3427 * Allocate and set consumer data hash tables.
3429 int lttng_consumer_init(void)
3431 consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3432 if (!consumer_data
.channel_ht
) {
3436 consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3437 if (!consumer_data
.relayd_ht
) {
3441 consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3442 if (!consumer_data
.stream_list_ht
) {
3446 consumer_data
.stream_per_chan_id_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3447 if (!consumer_data
.stream_per_chan_id_ht
) {
3451 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3456 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3468 * Process the ADD_RELAYD command receive by a consumer.
3470 * This will create a relayd socket pair and add it to the relayd hash table.
3471 * The caller MUST acquire a RCU read side lock before calling it.
3473 void consumer_add_relayd_socket(uint64_t net_seq_idx
, int sock_type
,
3474 struct lttng_consumer_local_data
*ctx
, int sock
,
3475 struct pollfd
*consumer_sockpoll
,
3476 struct lttcomm_relayd_sock
*relayd_sock
, uint64_t sessiond_id
,
3477 uint64_t relayd_session_id
)
3479 int fd
= -1, ret
= -1, relayd_created
= 0;
3480 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3481 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3484 assert(relayd_sock
);
3486 DBG("Consumer adding relayd socket (idx: %" PRIu64
")", net_seq_idx
);
3488 /* Get relayd reference if exists. */
3489 relayd
= consumer_find_relayd(net_seq_idx
);
3490 if (relayd
== NULL
) {
3491 assert(sock_type
== LTTNG_STREAM_CONTROL
);
3492 /* Not found. Allocate one. */
3493 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
3494 if (relayd
== NULL
) {
3495 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3498 relayd
->sessiond_session_id
= sessiond_id
;
3503 * This code path MUST continue to the consumer send status message to
3504 * we can notify the session daemon and continue our work without
3505 * killing everything.
3509 * relayd key should never be found for control socket.
3511 assert(sock_type
!= LTTNG_STREAM_CONTROL
);
3514 /* First send a status message before receiving the fds. */
3515 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
3517 /* Somehow, the session daemon is not responding anymore. */
3518 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3519 goto error_nosignal
;
3522 /* Poll on consumer socket. */
3523 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3525 /* Needing to exit in the middle of a command: error. */
3526 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
3527 goto error_nosignal
;
3530 /* Get relayd socket from session daemon */
3531 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
3532 if (ret
!= sizeof(fd
)) {
3533 fd
= -1; /* Just in case it gets set with an invalid value. */
3536 * Failing to receive FDs might indicate a major problem such as
3537 * reaching a fd limit during the receive where the kernel returns a
3538 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3539 * don't take any chances and stop everything.
3541 * XXX: Feature request #558 will fix that and avoid this possible
3542 * issue when reaching the fd limit.
3544 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
3545 ret_code
= LTTCOMM_CONSUMERD_ERROR_RECV_FD
;
3549 /* Copy socket information and received FD */
3550 switch (sock_type
) {
3551 case LTTNG_STREAM_CONTROL
:
3552 /* Copy received lttcomm socket */
3553 lttcomm_copy_sock(&relayd
->control_sock
.sock
, &relayd_sock
->sock
);
3554 ret
= lttcomm_create_sock(&relayd
->control_sock
.sock
);
3555 /* Handle create_sock error. */
3557 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3561 * Close the socket created internally by
3562 * lttcomm_create_sock, so we can replace it by the one
3563 * received from sessiond.
3565 if (close(relayd
->control_sock
.sock
.fd
)) {
3569 /* Assign new file descriptor */
3570 relayd
->control_sock
.sock
.fd
= fd
;
3571 fd
= -1; /* For error path */
3572 /* Assign version values. */
3573 relayd
->control_sock
.major
= relayd_sock
->major
;
3574 relayd
->control_sock
.minor
= relayd_sock
->minor
;
3576 relayd
->relayd_session_id
= relayd_session_id
;
3579 case LTTNG_STREAM_DATA
:
3580 /* Copy received lttcomm socket */
3581 lttcomm_copy_sock(&relayd
->data_sock
.sock
, &relayd_sock
->sock
);
3582 ret
= lttcomm_create_sock(&relayd
->data_sock
.sock
);
3583 /* Handle create_sock error. */
3585 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3589 * Close the socket created internally by
3590 * lttcomm_create_sock, so we can replace it by the one
3591 * received from sessiond.
3593 if (close(relayd
->data_sock
.sock
.fd
)) {
3597 /* Assign new file descriptor */
3598 relayd
->data_sock
.sock
.fd
= fd
;
3599 fd
= -1; /* for eventual error paths */
3600 /* Assign version values. */
3601 relayd
->data_sock
.major
= relayd_sock
->major
;
3602 relayd
->data_sock
.minor
= relayd_sock
->minor
;
3605 ERR("Unknown relayd socket type (%d)", sock_type
);
3606 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3610 DBG("Consumer %s socket created successfully with net idx %" PRIu64
" (fd: %d)",
3611 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
3612 relayd
->net_seq_idx
, fd
);
3614 /* We successfully added the socket. Send status back. */
3615 ret
= consumer_send_status_msg(sock
, ret_code
);
3617 /* Somehow, the session daemon is not responding anymore. */
3618 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3619 goto error_nosignal
;
3623 * Add relayd socket pair to consumer data hashtable. If object already
3624 * exists or on error, the function gracefully returns.
3633 if (consumer_send_status_msg(sock
, ret_code
) < 0) {
3634 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3638 /* Close received socket if valid. */
3641 PERROR("close received socket");
3645 if (relayd_created
) {
3651 * Search for a relayd associated to the session id and return the reference.
3653 * A rcu read side lock MUST be acquire before calling this function and locked
3654 * until the relayd object is no longer necessary.
3656 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
3658 struct lttng_ht_iter iter
;
3659 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3661 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3662 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
3665 * Check by sessiond id which is unique here where the relayd session
3666 * id might not be when having multiple relayd.
3668 if (relayd
->sessiond_session_id
== id
) {
3669 /* Found the relayd. There can be only one per id. */
3681 * Check if for a given session id there is still data needed to be extract
3684 * Return 1 if data is pending or else 0 meaning ready to be read.
3686 int consumer_data_pending(uint64_t id
)
3689 struct lttng_ht_iter iter
;
3690 struct lttng_ht
*ht
;
3691 struct lttng_consumer_stream
*stream
;
3692 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3693 int (*data_pending
)(struct lttng_consumer_stream
*);
3695 DBG("Consumer data pending command on session id %" PRIu64
, id
);
3698 pthread_mutex_lock(&consumer_data
.lock
);
3700 switch (consumer_data
.type
) {
3701 case LTTNG_CONSUMER_KERNEL
:
3702 data_pending
= lttng_kconsumer_data_pending
;
3704 case LTTNG_CONSUMER32_UST
:
3705 case LTTNG_CONSUMER64_UST
:
3706 data_pending
= lttng_ustconsumer_data_pending
;
3709 ERR("Unknown consumer data type");
3713 /* Ease our life a bit */
3714 ht
= consumer_data
.stream_list_ht
;
3716 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3717 ht
->hash_fct(&id
, lttng_ht_seed
),
3719 &iter
.iter
, stream
, node_session_id
.node
) {
3720 pthread_mutex_lock(&stream
->lock
);
3723 * A removed node from the hash table indicates that the stream has
3724 * been deleted thus having a guarantee that the buffers are closed
3725 * on the consumer side. However, data can still be transmitted
3726 * over the network so don't skip the relayd check.
3728 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
3730 /* Check the stream if there is data in the buffers. */
3731 ret
= data_pending(stream
);
3733 pthread_mutex_unlock(&stream
->lock
);
3738 pthread_mutex_unlock(&stream
->lock
);
3741 relayd
= find_relayd_by_session_id(id
);
3743 unsigned int is_data_inflight
= 0;
3745 /* Send init command for data pending. */
3746 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3747 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
3748 relayd
->relayd_session_id
);
3750 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3751 /* Communication error thus the relayd so no data pending. */
3752 goto data_not_pending
;
3755 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3756 ht
->hash_fct(&id
, lttng_ht_seed
),
3758 &iter
.iter
, stream
, node_session_id
.node
) {
3759 if (stream
->metadata_flag
) {
3760 ret
= relayd_quiescent_control(&relayd
->control_sock
,
3761 stream
->relayd_stream_id
);
3763 ret
= relayd_data_pending(&relayd
->control_sock
,
3764 stream
->relayd_stream_id
,
3765 stream
->next_net_seq_num
- 1);
3769 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3771 } else if (ret
< 0) {
3772 ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
3773 lttng_consumer_cleanup_relayd(relayd
);
3774 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3775 goto data_not_pending
;
3779 /* Send end command for data pending. */
3780 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3781 relayd
->relayd_session_id
, &is_data_inflight
);
3782 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3784 ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
3785 lttng_consumer_cleanup_relayd(relayd
);
3786 goto data_not_pending
;
3788 if (is_data_inflight
) {
3794 * Finding _no_ node in the hash table and no inflight data means that the
3795 * stream(s) have been removed thus data is guaranteed to be available for
3796 * analysis from the trace files.
3800 /* Data is available to be read by a viewer. */
3801 pthread_mutex_unlock(&consumer_data
.lock
);
3806 /* Data is still being extracted from buffers. */
3807 pthread_mutex_unlock(&consumer_data
.lock
);
3813 * Send a ret code status message to the sessiond daemon.
3815 * Return the sendmsg() return value.
3817 int consumer_send_status_msg(int sock
, int ret_code
)
3819 struct lttcomm_consumer_status_msg msg
;
3821 memset(&msg
, 0, sizeof(msg
));
3822 msg
.ret_code
= ret_code
;
3824 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3828 * Send a channel status message to the sessiond daemon.
3830 * Return the sendmsg() return value.
3832 int consumer_send_status_channel(int sock
,
3833 struct lttng_consumer_channel
*channel
)
3835 struct lttcomm_consumer_status_channel msg
;
3839 memset(&msg
, 0, sizeof(msg
));
3841 msg
.ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
3843 msg
.ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3844 msg
.key
= channel
->key
;
3845 msg
.stream_count
= channel
->streams
.count
;
3848 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3851 unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos
,
3852 unsigned long produced_pos
, uint64_t nb_packets_per_stream
,
3853 uint64_t max_sb_size
)
3855 unsigned long start_pos
;
3857 if (!nb_packets_per_stream
) {
3858 return consumed_pos
; /* Grab everything */
3860 start_pos
= produced_pos
- offset_align_floor(produced_pos
, max_sb_size
);
3861 start_pos
-= max_sb_size
* nb_packets_per_stream
;
3862 if ((long) (start_pos
- consumed_pos
) < 0) {
3863 return consumed_pos
; /* Grab everything */
3869 int consumer_flush_buffer(struct lttng_consumer_stream
*stream
, int producer_active
)
3873 switch (consumer_data
.type
) {
3874 case LTTNG_CONSUMER_KERNEL
:
3875 ret
= kernctl_buffer_flush(stream
->wait_fd
);
3877 ERR("Failed to flush kernel stream");
3881 case LTTNG_CONSUMER32_UST
:
3882 case LTTNG_CONSUMER64_UST
:
3883 lttng_ustctl_flush_buffer(stream
, producer_active
);
3886 ERR("Unknown consumer_data type");
3895 * Sample the rotate position for all the streams of a channel. If a stream
3896 * is already at the rotate position (produced == consumed), we flag it as
3897 * ready for rotation. The rotation of ready streams occurs after we have
3898 * replied to the session daemon that we have finished sampling the positions.
3900 * Returns 0 on success, < 0 on error
3902 int lttng_consumer_rotate_channel(uint64_t key
, const char *path
,
3903 uint64_t relayd_id
, uint32_t metadata
, uint64_t new_chunk_id
,
3904 struct lttng_consumer_local_data
*ctx
)
3907 struct lttng_consumer_channel
*channel
;
3908 struct lttng_consumer_stream
*stream
;
3909 struct lttng_ht_iter iter
;
3910 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
3912 DBG("Consumer sample rotate position for channel %" PRIu64
, key
);
3916 channel
= consumer_find_channel(key
);
3918 ERR("No channel found for key %" PRIu64
, key
);
3923 pthread_mutex_lock(&channel
->lock
);
3924 channel
->current_chunk_id
= new_chunk_id
;
3926 ret
= lttng_strncpy(channel
->pathname
, path
, sizeof(channel
->pathname
));
3928 ERR("Failed to copy new path to channel during channel rotation");
3930 goto end_unlock_channel
;
3933 if (relayd_id
== -1ULL) {
3935 * The domain path (/ust or /kernel) has been created before, we
3936 * now need to create the last part of the path: the application/user
3937 * specific section (uid/1000/64-bit).
3939 ret
= utils_mkdir_recursive(channel
->pathname
, S_IRWXU
| S_IRWXG
,
3940 channel
->uid
, channel
->gid
);
3942 ERR("Failed to create trace directory at %s during rotation",
3945 goto end_unlock_channel
;
3949 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3950 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
3951 ht
->match_fct
, &channel
->key
, &iter
.iter
,
3952 stream
, node_channel_id
.node
) {
3953 unsigned long consumed_pos
;
3955 health_code_update();
3958 * Lock stream because we are about to change its state.
3960 pthread_mutex_lock(&stream
->lock
);
3962 ret
= lttng_strncpy(stream
->channel_read_only_attributes
.path
,
3964 sizeof(stream
->channel_read_only_attributes
.path
));
3966 ERR("Failed to sample channel path name during channel rotation");
3967 goto end_unlock_stream
;
3969 ret
= lttng_consumer_sample_snapshot_positions(stream
);
3971 ERR("Failed to sample snapshot position during channel rotation");
3972 goto end_unlock_stream
;
3975 ret
= lttng_consumer_get_produced_snapshot(stream
,
3976 &stream
->rotate_position
);
3978 ERR("Failed to sample produced position during channel rotation");
3979 goto end_unlock_stream
;
3982 lttng_consumer_get_consumed_snapshot(stream
,
3984 if (consumed_pos
== stream
->rotate_position
) {
3985 stream
->rotate_ready
= true;
3988 ret
= consumer_flush_buffer(stream
, 1);
3990 ERR("Failed to flush stream %" PRIu64
" during channel rotation",
3992 goto end_unlock_stream
;
3995 pthread_mutex_unlock(&stream
->lock
);
3997 pthread_mutex_unlock(&channel
->lock
);
4003 pthread_mutex_unlock(&stream
->lock
);
4005 pthread_mutex_unlock(&channel
->lock
);
4012 * Check if a stream is ready to be rotated after extracting it.
4014 * Return 1 if it is ready for rotation, 0 if it is not, a negative value on
4015 * error. Stream lock must be held.
4017 int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream
*stream
)
4020 unsigned long consumed_pos
;
4022 if (!stream
->rotate_position
&& !stream
->rotate_ready
) {
4027 if (stream
->rotate_ready
) {
4033 * If we don't have the rotate_ready flag, check the consumed position
4034 * to determine if we need to rotate.
4036 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4038 ERR("Taking snapshot positions");
4042 ret
= lttng_consumer_get_consumed_snapshot(stream
, &consumed_pos
);
4044 ERR("Consumed snapshot position");
4048 /* Rotate position not reached yet (with check for overflow). */
4049 if ((long) (consumed_pos
- stream
->rotate_position
) < 0) {
4060 * Reset the state for a stream after a rotation occurred.
4062 void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream
*stream
)
4064 stream
->rotate_position
= 0;
4065 stream
->rotate_ready
= false;
4069 * Perform the rotation a local stream file.
4071 int rotate_local_stream(struct lttng_consumer_local_data
*ctx
,
4072 struct lttng_consumer_stream
*stream
)
4076 DBG("Rotate local stream: stream key %" PRIu64
", channel key %" PRIu64
" at path %s",
4079 stream
->channel_read_only_attributes
.path
);
4081 ret
= close(stream
->out_fd
);
4083 PERROR("Closing trace file (fd %d), stream %" PRIu64
,
4084 stream
->out_fd
, stream
->key
);
4089 ret
= utils_create_stream_file(
4090 stream
->channel_read_only_attributes
.path
,
4092 stream
->channel_read_only_attributes
.tracefile_size
,
4093 stream
->tracefile_count_current
,
4094 stream
->uid
, stream
->gid
, NULL
);
4096 ERR("Rotate create stream file");
4099 stream
->out_fd
= ret
;
4100 stream
->tracefile_size_current
= 0;
4102 if (!stream
->metadata_flag
) {
4103 struct lttng_index_file
*index_file
;
4105 lttng_index_file_put(stream
->index_file
);
4107 index_file
= lttng_index_file_create(
4108 stream
->channel_read_only_attributes
.path
,
4109 stream
->name
, stream
->uid
, stream
->gid
,
4110 stream
->channel_read_only_attributes
.tracefile_size
,
4111 stream
->tracefile_count_current
,
4112 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
4114 ERR("Create index file during rotation");
4117 stream
->index_file
= index_file
;
4118 stream
->out_fd_offset
= 0;
4132 * Perform the rotation a stream file on the relay.
4134 int rotate_relay_stream(struct lttng_consumer_local_data
*ctx
,
4135 struct lttng_consumer_stream
*stream
)
4138 struct consumer_relayd_sock_pair
*relayd
;
4140 DBG("Rotate relay stream");
4141 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
4143 ERR("Failed to find relayd");
4148 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4149 ret
= relayd_rotate_stream(&relayd
->control_sock
,
4150 stream
->relayd_stream_id
,
4151 stream
->channel_read_only_attributes
.path
,
4152 stream
->chan
->current_chunk_id
,
4153 stream
->last_sequence_number
);
4154 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4156 ERR("Relayd rotate stream failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
4157 lttng_consumer_cleanup_relayd(relayd
);
4160 ERR("Rotate relay stream");
4168 * Performs the stream rotation for the rotate session feature if needed.
4169 * It must be called with the stream lock held.
4171 * Return 0 on success, a negative number of error.
4173 int lttng_consumer_rotate_stream(struct lttng_consumer_local_data
*ctx
,
4174 struct lttng_consumer_stream
*stream
, bool *rotated
)
4178 DBG("Consumer rotate stream %" PRIu64
, stream
->key
);
4180 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
4181 ret
= rotate_relay_stream(ctx
, stream
);
4183 ret
= rotate_local_stream(ctx
, stream
);
4185 stream
->trace_archive_id
++;
4187 ERR("Failed to rotate stream, ret = %i", ret
);
4191 if (stream
->metadata_flag
) {
4192 switch (consumer_data
.type
) {
4193 case LTTNG_CONSUMER_KERNEL
:
4195 * Reset the position of what has been read from the metadata
4196 * cache to 0 so we can dump it again.
4198 ret
= kernctl_metadata_cache_dump(stream
->wait_fd
);
4200 ERR("Failed to dump the kernel metadata cache after rotation");
4204 case LTTNG_CONSUMER32_UST
:
4205 case LTTNG_CONSUMER64_UST
:
4207 * Reset the position pushed from the metadata cache so it
4208 * will write from the beginning on the next push.
4210 stream
->ust_metadata_pushed
= 0;
4213 ERR("Unknown consumer_data type");
4217 lttng_consumer_reset_stream_rotate_state(stream
);
4230 * Rotate all the ready streams now.
4232 * This is especially important for low throughput streams that have already
4233 * been consumed, we cannot wait for their next packet to perform the
4236 * Returns 0 on success, < 0 on error
4238 int lttng_consumer_rotate_ready_streams(uint64_t key
,
4239 struct lttng_consumer_local_data
*ctx
)
4242 struct lttng_consumer_channel
*channel
;
4243 struct lttng_consumer_stream
*stream
;
4244 struct lttng_ht_iter iter
;
4245 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
4249 DBG("Consumer rotate ready streams in channel %" PRIu64
, key
);
4251 channel
= consumer_find_channel(key
);
4253 ERR("No channel found for key %" PRIu64
, key
);
4258 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4259 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4260 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4261 stream
, node_channel_id
.node
) {
4262 health_code_update();
4264 pthread_mutex_lock(&stream
->lock
);
4266 if (!stream
->rotate_ready
) {
4267 pthread_mutex_unlock(&stream
->lock
);
4270 DBG("Consumer rotate ready stream %" PRIu64
, stream
->key
);
4272 ret
= lttng_consumer_rotate_stream(ctx
, stream
, NULL
);
4273 pthread_mutex_unlock(&stream
->lock
);
4278 ret
= consumer_post_rotation(stream
, ctx
);
4292 int rotate_rename_local(const char *old_path
, const char *new_path
,
4293 uid_t uid
, gid_t gid
)
4300 ret
= utils_mkdir_recursive(new_path
, S_IRWXU
| S_IRWXG
, uid
, gid
);
4302 ERR("Create directory on rotate");
4306 ret
= rename(old_path
, new_path
);
4307 if (ret
< 0 && errno
!= ENOENT
) {
4308 PERROR("Rename completed rotation chunk");
4318 int rotate_rename_relay(const char *old_path
, const char *new_path
,
4322 struct consumer_relayd_sock_pair
*relayd
;
4324 relayd
= consumer_find_relayd(relayd_id
);
4326 ERR("Failed to find relayd while running rotate_rename_relay command");
4331 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4332 ret
= relayd_rotate_rename(&relayd
->control_sock
, old_path
, new_path
);
4334 ERR("Relayd rotate rename failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
4335 lttng_consumer_cleanup_relayd(relayd
);
4337 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4342 int lttng_consumer_rotate_rename(const char *old_path
, const char *new_path
,
4343 uid_t uid
, gid_t gid
, uint64_t relayd_id
)
4345 if (relayd_id
!= -1ULL) {
4346 return rotate_rename_relay(old_path
, new_path
, relayd_id
);
4348 return rotate_rename_local(old_path
, new_path
, uid
, gid
);
4352 /* Stream lock must be acquired by the caller. */
4354 bool check_stream_rotation_pending(const struct lttng_consumer_stream
*stream
,
4355 uint64_t session_id
, uint64_t chunk_id
)
4357 bool pending
= false;
4359 if (stream
->session_id
!= session_id
) {
4365 * If the stream's archive_id belongs to the chunk being rotated (or an
4366 * even older one), it means that the consumer has not consumed all the
4367 * buffers that belong to the chunk being rotated. Therefore, the
4368 * rotation is considered as ongoing/pending.
4370 pending
= stream
->trace_archive_id
<= chunk_id
;
4375 /* RCU read lock must be acquired by the caller. */
4376 int lttng_consumer_check_rotation_pending_local(uint64_t session_id
,
4379 struct lttng_ht_iter iter
;
4380 struct lttng_consumer_stream
*stream
;
4381 bool rotation_pending
= false;
4383 /* Start with the metadata streams... */
4384 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
4385 pthread_mutex_lock(&stream
->lock
);
4386 rotation_pending
= check_stream_rotation_pending(stream
,
4387 session_id
, chunk_id
);
4388 pthread_mutex_unlock(&stream
->lock
);
4389 if (rotation_pending
) {
4394 /* ... followed by the data streams. */
4395 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
4396 pthread_mutex_lock(&stream
->lock
);
4397 rotation_pending
= check_stream_rotation_pending(stream
,
4398 session_id
, chunk_id
);
4399 pthread_mutex_unlock(&stream
->lock
);
4400 if (rotation_pending
) {
4406 return !!rotation_pending
;
4409 int lttng_consumer_check_rotation_pending_relay(uint64_t session_id
,
4410 uint64_t relayd_id
, uint64_t chunk_id
)
4413 struct consumer_relayd_sock_pair
*relayd
;
4415 relayd
= consumer_find_relayd(relayd_id
);
4417 ERR("Failed to find relayd id %" PRIu64
, relayd_id
);
4422 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4423 ret
= relayd_rotate_pending(&relayd
->control_sock
, chunk_id
);
4425 ERR("Relayd rotate pending failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
4426 lttng_consumer_cleanup_relayd(relayd
);
4428 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4435 int mkdir_local(const char *path
, uid_t uid
, gid_t gid
)
4439 ret
= utils_mkdir_recursive(path
, S_IRWXU
| S_IRWXG
, uid
, gid
);
4441 /* utils_mkdir_recursive logs an error. */
4451 int mkdir_relay(const char *path
, uint64_t relayd_id
)
4454 struct consumer_relayd_sock_pair
*relayd
;
4456 relayd
= consumer_find_relayd(relayd_id
);
4458 ERR("Failed to find relayd");
4463 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4464 ret
= relayd_mkdir(&relayd
->control_sock
, path
);
4466 ERR("Relayd mkdir failed. Cleaning up relayd %" PRIu64
".", relayd
->net_seq_idx
);
4467 lttng_consumer_cleanup_relayd(relayd
);
4469 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4476 int lttng_consumer_mkdir(const char *path
, uid_t uid
, gid_t gid
,
4479 if (relayd_id
!= -1ULL) {
4480 return mkdir_relay(path
, relayd_id
);
4482 return mkdir_local(path
, uid
, gid
);