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
);
330 * Destroy and free relayd socket pair object.
332 void consumer_destroy_relayd(struct consumer_relayd_sock_pair
*relayd
)
335 struct lttng_ht_iter iter
;
337 if (relayd
== NULL
) {
341 DBG("Consumer destroy and close relayd socket pair");
343 iter
.iter
.node
= &relayd
->node
.node
;
344 ret
= lttng_ht_del(consumer_data
.relayd_ht
, &iter
);
346 /* We assume the relayd is being or is destroyed */
350 /* RCU free() call */
351 call_rcu(&relayd
->node
.head
, free_relayd_rcu
);
355 * Remove a channel from the global list protected by a mutex. This function is
356 * also responsible for freeing its data structures.
358 void consumer_del_channel(struct lttng_consumer_channel
*channel
)
361 struct lttng_ht_iter iter
;
363 DBG("Consumer delete channel key %" PRIu64
, channel
->key
);
365 pthread_mutex_lock(&consumer_data
.lock
);
366 pthread_mutex_lock(&channel
->lock
);
368 /* Destroy streams that might have been left in the stream list. */
369 clean_channel_stream_list(channel
);
371 if (channel
->live_timer_enabled
== 1) {
372 consumer_timer_live_stop(channel
);
374 if (channel
->monitor_timer_enabled
== 1) {
375 consumer_timer_monitor_stop(channel
);
378 switch (consumer_data
.type
) {
379 case LTTNG_CONSUMER_KERNEL
:
381 case LTTNG_CONSUMER32_UST
:
382 case LTTNG_CONSUMER64_UST
:
383 lttng_ustconsumer_del_channel(channel
);
386 ERR("Unknown consumer_data type");
392 iter
.iter
.node
= &channel
->node
.node
;
393 ret
= lttng_ht_del(consumer_data
.channel_ht
, &iter
);
397 call_rcu(&channel
->node
.head
, free_channel_rcu
);
399 pthread_mutex_unlock(&channel
->lock
);
400 pthread_mutex_unlock(&consumer_data
.lock
);
404 * Iterate over the relayd hash table and destroy each element. Finally,
405 * destroy the whole hash table.
407 static void cleanup_relayd_ht(void)
409 struct lttng_ht_iter iter
;
410 struct consumer_relayd_sock_pair
*relayd
;
414 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
416 consumer_destroy_relayd(relayd
);
421 lttng_ht_destroy(consumer_data
.relayd_ht
);
425 * Update the end point status of all streams having the given network sequence
426 * index (relayd index).
428 * It's atomically set without having the stream mutex locked which is fine
429 * because we handle the write/read race with a pipe wakeup for each thread.
431 static void update_endpoint_status_by_netidx(uint64_t net_seq_idx
,
432 enum consumer_endpoint_status status
)
434 struct lttng_ht_iter iter
;
435 struct lttng_consumer_stream
*stream
;
437 DBG("Consumer set delete flag on stream by idx %" PRIu64
, net_seq_idx
);
441 /* Let's begin with metadata */
442 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
443 if (stream
->net_seq_idx
== net_seq_idx
) {
444 uatomic_set(&stream
->endpoint_status
, status
);
445 DBG("Delete flag set to metadata stream %d", stream
->wait_fd
);
449 /* Follow up by the data streams */
450 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
451 if (stream
->net_seq_idx
== net_seq_idx
) {
452 uatomic_set(&stream
->endpoint_status
, status
);
453 DBG("Delete flag set to data stream %d", stream
->wait_fd
);
460 * Cleanup a relayd object by flagging every associated streams for deletion,
461 * destroying the object meaning removing it from the relayd hash table,
462 * closing the sockets and freeing the memory in a RCU call.
464 * If a local data context is available, notify the threads that the streams'
465 * state have changed.
467 static void cleanup_relayd(struct consumer_relayd_sock_pair
*relayd
,
468 struct lttng_consumer_local_data
*ctx
)
474 DBG("Cleaning up relayd sockets");
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.
495 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
496 notify_thread_lttng_pipe(ctx
->consumer_metadata_pipe
);
501 * Flag a relayd socket pair for destruction. Destroy it if the refcount
504 * RCU read side lock MUST be aquired before calling this function.
506 void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair
*relayd
)
510 /* Set destroy flag for this object */
511 uatomic_set(&relayd
->destroy_flag
, 1);
513 /* Destroy the relayd if refcount is 0 */
514 if (uatomic_read(&relayd
->refcount
) == 0) {
515 consumer_destroy_relayd(relayd
);
520 * Completly destroy stream from every visiable data structure and the given
523 * One this call returns, the stream object is not longer usable nor visible.
525 void consumer_del_stream(struct lttng_consumer_stream
*stream
,
528 consumer_stream_destroy(stream
, ht
);
532 * XXX naming of del vs destroy is all mixed up.
534 void consumer_del_stream_for_data(struct lttng_consumer_stream
*stream
)
536 consumer_stream_destroy(stream
, data_ht
);
539 void consumer_del_stream_for_metadata(struct lttng_consumer_stream
*stream
)
541 consumer_stream_destroy(stream
, metadata_ht
);
544 void consumer_stream_copy_ro_channel_values(struct lttng_consumer_stream
*stream
,
545 struct lttng_consumer_channel
*channel
)
547 stream
->channel_ro_tracefile_size
= channel
->tracefile_size
;
548 memcpy(stream
->channel_ro_pathname
, channel
->pathname
, PATH_MAX
);
551 struct lttng_consumer_stream
*consumer_allocate_stream(uint64_t channel_key
,
553 enum lttng_consumer_stream_state state
,
554 const char *channel_name
,
561 enum consumer_channel_type type
,
562 unsigned int monitor
)
565 struct lttng_consumer_stream
*stream
;
567 stream
= zmalloc(sizeof(*stream
));
568 if (stream
== NULL
) {
569 PERROR("malloc struct lttng_consumer_stream");
576 stream
->key
= stream_key
;
578 stream
->out_fd_offset
= 0;
579 stream
->output_written
= 0;
580 stream
->state
= state
;
583 stream
->net_seq_idx
= relayd_id
;
584 stream
->session_id
= session_id
;
585 stream
->monitor
= monitor
;
586 stream
->endpoint_status
= CONSUMER_ENDPOINT_ACTIVE
;
587 stream
->index_file
= NULL
;
588 stream
->last_sequence_number
= -1ULL;
589 pthread_mutex_init(&stream
->lock
, NULL
);
590 pthread_mutex_init(&stream
->metadata_timer_lock
, NULL
);
592 /* If channel is the metadata, flag this stream as metadata. */
593 if (type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
594 stream
->metadata_flag
= 1;
595 /* Metadata is flat out. */
596 strncpy(stream
->name
, DEFAULT_METADATA_NAME
, sizeof(stream
->name
));
597 /* Live rendez-vous point. */
598 pthread_cond_init(&stream
->metadata_rdv
, NULL
);
599 pthread_mutex_init(&stream
->metadata_rdv_lock
, NULL
);
601 /* Format stream name to <channel_name>_<cpu_number> */
602 ret
= snprintf(stream
->name
, sizeof(stream
->name
), "%s_%d",
605 PERROR("snprintf stream name");
610 /* Key is always the wait_fd for streams. */
611 lttng_ht_node_init_u64(&stream
->node
, stream
->key
);
613 /* Init node per channel id key */
614 lttng_ht_node_init_u64(&stream
->node_channel_id
, channel_key
);
616 /* Init session id node with the stream session id */
617 lttng_ht_node_init_u64(&stream
->node_session_id
, stream
->session_id
);
619 DBG3("Allocated stream %s (key %" PRIu64
", chan_key %" PRIu64
620 " relayd_id %" PRIu64
", session_id %" PRIu64
,
621 stream
->name
, stream
->key
, channel_key
,
622 stream
->net_seq_idx
, stream
->session_id
);
638 * Add a stream to the global list protected by a mutex.
640 int consumer_add_data_stream(struct lttng_consumer_stream
*stream
)
642 struct lttng_ht
*ht
= data_ht
;
648 DBG3("Adding consumer stream %" PRIu64
, stream
->key
);
650 pthread_mutex_lock(&consumer_data
.lock
);
651 pthread_mutex_lock(&stream
->chan
->lock
);
652 pthread_mutex_lock(&stream
->chan
->timer_lock
);
653 pthread_mutex_lock(&stream
->lock
);
656 /* Steal stream identifier to avoid having streams with the same key */
657 steal_stream_key(stream
->key
, ht
);
659 lttng_ht_add_unique_u64(ht
, &stream
->node
);
661 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
662 &stream
->node_channel_id
);
665 * Add stream to the stream_list_ht of the consumer data. No need to steal
666 * the key since the HT does not use it and we allow to add redundant keys
669 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
672 * When nb_init_stream_left reaches 0, we don't need to trigger any action
673 * in terms of destroying the associated channel, because the action that
674 * causes the count to become 0 also causes a stream to be added. The
675 * channel deletion will thus be triggered by the following removal of this
678 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
679 /* Increment refcount before decrementing nb_init_stream_left */
681 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
684 /* Update consumer data once the node is inserted. */
685 consumer_data
.stream_count
++;
686 consumer_data
.need_update
= 1;
689 pthread_mutex_unlock(&stream
->lock
);
690 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
691 pthread_mutex_unlock(&stream
->chan
->lock
);
692 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
,
792 char *path
, enum lttng_domain_type domain
)
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
,
811 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
816 uatomic_inc(&relayd
->refcount
);
817 stream
->sent_to_relayd
= 1;
819 ERR("Stream %" PRIu64
" relayd ID %" PRIu64
" unknown. Can't send it.",
820 stream
->key
, stream
->net_seq_idx
);
825 DBG("Stream %s with key %" PRIu64
" sent to relayd id %" PRIu64
,
826 stream
->name
, stream
->key
, stream
->net_seq_idx
);
834 * Find a relayd and send the streams sent message
836 * Returns 0 on success, < 0 on error
838 int consumer_send_relayd_streams_sent(uint64_t net_seq_idx
)
841 struct consumer_relayd_sock_pair
*relayd
;
843 assert(net_seq_idx
!= -1ULL);
845 /* The stream is not metadata. Get relayd reference if exists. */
847 relayd
= consumer_find_relayd(net_seq_idx
);
848 if (relayd
!= NULL
) {
849 /* Add stream on the relayd */
850 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
851 ret
= relayd_streams_sent(&relayd
->control_sock
);
852 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
857 ERR("Relayd ID %" PRIu64
" unknown. Can't send streams_sent.",
864 DBG("All streams sent relayd id %" PRIu64
, net_seq_idx
);
872 * Find a relayd and close the stream
874 void close_relayd_stream(struct lttng_consumer_stream
*stream
)
876 struct consumer_relayd_sock_pair
*relayd
;
878 /* The stream is not metadata. Get relayd reference if exists. */
880 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
882 consumer_stream_relayd_close(stream
, relayd
);
888 * Handle stream for relayd transmission if the stream applies for network
889 * streaming where the net sequence index is set.
891 * Return destination file descriptor or negative value on error.
893 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
894 size_t data_size
, unsigned long padding
,
895 struct consumer_relayd_sock_pair
*relayd
)
898 struct lttcomm_relayd_data_hdr data_hdr
;
904 /* Reset data header */
905 memset(&data_hdr
, 0, sizeof(data_hdr
));
907 if (stream
->metadata_flag
) {
908 /* Caller MUST acquire the relayd control socket lock */
909 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
914 /* Metadata are always sent on the control socket. */
915 outfd
= relayd
->control_sock
.sock
.fd
;
917 /* Set header with stream information */
918 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
919 data_hdr
.data_size
= htobe32(data_size
);
920 data_hdr
.padding_size
= htobe32(padding
);
922 * Note that net_seq_num below is assigned with the *current* value of
923 * next_net_seq_num and only after that the next_net_seq_num will be
924 * increment. This is why when issuing a command on the relayd using
925 * this next value, 1 should always be substracted in order to compare
926 * the last seen sequence number on the relayd side to the last sent.
928 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
929 /* Other fields are zeroed previously */
931 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
937 ++stream
->next_net_seq_num
;
939 /* Set to go on data socket */
940 outfd
= relayd
->data_sock
.sock
.fd
;
948 * Allocate and return a new lttng_consumer_channel object using the given key
949 * to initialize the hash table node.
951 * On error, return NULL.
953 struct lttng_consumer_channel
*consumer_allocate_channel(uint64_t key
,
955 const char *pathname
,
960 enum lttng_event_output output
,
961 uint64_t tracefile_size
,
962 uint64_t tracefile_count
,
963 uint64_t session_id_per_pid
,
964 unsigned int monitor
,
965 unsigned int live_timer_interval
,
966 const char *root_shm_path
,
967 const char *shm_path
)
969 struct lttng_consumer_channel
*channel
;
971 channel
= zmalloc(sizeof(*channel
));
972 if (channel
== NULL
) {
973 PERROR("malloc struct lttng_consumer_channel");
978 channel
->refcount
= 0;
979 channel
->session_id
= session_id
;
980 channel
->session_id_per_pid
= session_id_per_pid
;
983 channel
->relayd_id
= relayd_id
;
984 channel
->tracefile_size
= tracefile_size
;
985 channel
->tracefile_count
= tracefile_count
;
986 channel
->monitor
= monitor
;
987 channel
->live_timer_interval
= live_timer_interval
;
988 pthread_mutex_init(&channel
->lock
, NULL
);
989 pthread_mutex_init(&channel
->timer_lock
, NULL
);
992 case LTTNG_EVENT_SPLICE
:
993 channel
->output
= CONSUMER_CHANNEL_SPLICE
;
995 case LTTNG_EVENT_MMAP
:
996 channel
->output
= CONSUMER_CHANNEL_MMAP
;
1006 * In monitor mode, the streams associated with the channel will be put in
1007 * a special list ONLY owned by this channel. So, the refcount is set to 1
1008 * here meaning that the channel itself has streams that are referenced.
1010 * On a channel deletion, once the channel is no longer visible, the
1011 * refcount is decremented and checked for a zero value to delete it. With
1012 * streams in no monitor mode, it will now be safe to destroy the channel.
1014 if (!channel
->monitor
) {
1015 channel
->refcount
= 1;
1018 strncpy(channel
->pathname
, pathname
, sizeof(channel
->pathname
));
1019 channel
->pathname
[sizeof(channel
->pathname
) - 1] = '\0';
1021 strncpy(channel
->name
, name
, sizeof(channel
->name
));
1022 channel
->name
[sizeof(channel
->name
) - 1] = '\0';
1024 if (root_shm_path
) {
1025 strncpy(channel
->root_shm_path
, root_shm_path
, sizeof(channel
->root_shm_path
));
1026 channel
->root_shm_path
[sizeof(channel
->root_shm_path
) - 1] = '\0';
1029 strncpy(channel
->shm_path
, shm_path
, sizeof(channel
->shm_path
));
1030 channel
->shm_path
[sizeof(channel
->shm_path
) - 1] = '\0';
1033 lttng_ht_node_init_u64(&channel
->node
, channel
->key
);
1035 channel
->wait_fd
= -1;
1037 CDS_INIT_LIST_HEAD(&channel
->streams
.head
);
1039 DBG("Allocated channel (key %" PRIu64
")", channel
->key
);
1046 * Add a channel to the global list protected by a mutex.
1048 * Always return 0 indicating success.
1050 int consumer_add_channel(struct lttng_consumer_channel
*channel
,
1051 struct lttng_consumer_local_data
*ctx
)
1053 pthread_mutex_lock(&consumer_data
.lock
);
1054 pthread_mutex_lock(&channel
->lock
);
1055 pthread_mutex_lock(&channel
->timer_lock
);
1058 * This gives us a guarantee that the channel we are about to add to the
1059 * channel hash table will be unique. See this function comment on the why
1060 * we need to steel the channel key at this stage.
1062 steal_channel_key(channel
->key
);
1065 lttng_ht_add_unique_u64(consumer_data
.channel_ht
, &channel
->node
);
1068 pthread_mutex_unlock(&channel
->timer_lock
);
1069 pthread_mutex_unlock(&channel
->lock
);
1070 pthread_mutex_unlock(&consumer_data
.lock
);
1072 if (channel
->wait_fd
!= -1 && channel
->type
== CONSUMER_CHANNEL_TYPE_DATA
) {
1073 notify_channel_pipe(ctx
, channel
, -1, CONSUMER_CHANNEL_ADD
);
1080 * Allocate the pollfd structure and the local view of the out fds to avoid
1081 * doing a lookup in the linked list and concurrency issues when writing is
1082 * needed. Called with consumer_data.lock held.
1084 * Returns the number of fds in the structures.
1086 static int update_poll_array(struct lttng_consumer_local_data
*ctx
,
1087 struct pollfd
**pollfd
, struct lttng_consumer_stream
**local_stream
,
1088 struct lttng_ht
*ht
)
1091 struct lttng_ht_iter iter
;
1092 struct lttng_consumer_stream
*stream
;
1097 assert(local_stream
);
1099 DBG("Updating poll fd array");
1101 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1103 * Only active streams with an active end point can be added to the
1104 * poll set and local stream storage of the thread.
1106 * There is a potential race here for endpoint_status to be updated
1107 * just after the check. However, this is OK since the stream(s) will
1108 * be deleted once the thread is notified that the end point state has
1109 * changed where this function will be called back again.
1111 if (stream
->state
!= LTTNG_CONSUMER_ACTIVE_STREAM
||
1112 stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
1116 * This clobbers way too much the debug output. Uncomment that if you
1117 * need it for debugging purposes.
1119 * DBG("Active FD %d", stream->wait_fd);
1121 (*pollfd
)[i
].fd
= stream
->wait_fd
;
1122 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1123 local_stream
[i
] = stream
;
1129 * Insert the consumer_data_pipe at the end of the array and don't
1130 * increment i so nb_fd is the number of real FD.
1132 (*pollfd
)[i
].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_pipe
);
1133 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1135 (*pollfd
)[i
+ 1].fd
= lttng_pipe_get_readfd(ctx
->consumer_wakeup_pipe
);
1136 (*pollfd
)[i
+ 1].events
= POLLIN
| POLLPRI
;
1138 (*pollfd
)[i
+ 2].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_rotate_pipe
);
1139 (*pollfd
)[i
+ 2].events
= POLLIN
| POLLPRI
;
1144 * Poll on the should_quit pipe and the command socket return -1 on
1145 * error, 1 if should exit, 0 if data is available on the command socket
1147 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
1152 num_rdy
= poll(consumer_sockpoll
, 2, -1);
1153 if (num_rdy
== -1) {
1155 * Restart interrupted system call.
1157 if (errno
== EINTR
) {
1160 PERROR("Poll error");
1163 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
1164 DBG("consumer_should_quit wake up");
1171 * Set the error socket.
1173 void lttng_consumer_set_error_sock(struct lttng_consumer_local_data
*ctx
,
1176 ctx
->consumer_error_socket
= sock
;
1180 * Set the command socket path.
1182 void lttng_consumer_set_command_sock_path(
1183 struct lttng_consumer_local_data
*ctx
, char *sock
)
1185 ctx
->consumer_command_sock_path
= sock
;
1189 * Send return code to the session daemon.
1190 * If the socket is not defined, we return 0, it is not a fatal error
1192 int lttng_consumer_send_error(struct lttng_consumer_local_data
*ctx
, int cmd
)
1194 if (ctx
->consumer_error_socket
> 0) {
1195 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1196 sizeof(enum lttcomm_sessiond_command
));
1203 * Close all the tracefiles and stream fds and MUST be called when all
1204 * instances are destroyed i.e. when all threads were joined and are ended.
1206 void lttng_consumer_cleanup(void)
1208 struct lttng_ht_iter iter
;
1209 struct lttng_consumer_channel
*channel
;
1213 cds_lfht_for_each_entry(consumer_data
.channel_ht
->ht
, &iter
.iter
, channel
,
1215 consumer_del_channel(channel
);
1220 lttng_ht_destroy(consumer_data
.channel_ht
);
1222 cleanup_relayd_ht();
1224 lttng_ht_destroy(consumer_data
.stream_per_chan_id_ht
);
1227 * This HT contains streams that are freed by either the metadata thread or
1228 * the data thread so we do *nothing* on the hash table and simply destroy
1231 lttng_ht_destroy(consumer_data
.stream_list_ht
);
1235 * Called from signal handler.
1237 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1241 CMM_STORE_SHARED(consumer_quit
, 1);
1242 ret
= lttng_write(ctx
->consumer_should_quit
[1], "4", 1);
1244 PERROR("write consumer quit");
1247 DBG("Consumer flag that it should quit");
1252 * Flush pending writes to trace output disk file.
1255 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1259 int outfd
= stream
->out_fd
;
1262 * This does a blocking write-and-wait on any page that belongs to the
1263 * subbuffer prior to the one we just wrote.
1264 * Don't care about error values, as these are just hints and ways to
1265 * limit the amount of page cache used.
1267 if (orig_offset
< stream
->max_sb_size
) {
1270 lttng_sync_file_range(outfd
, orig_offset
- stream
->max_sb_size
,
1271 stream
->max_sb_size
,
1272 SYNC_FILE_RANGE_WAIT_BEFORE
1273 | SYNC_FILE_RANGE_WRITE
1274 | SYNC_FILE_RANGE_WAIT_AFTER
);
1276 * Give hints to the kernel about how we access the file:
1277 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1280 * We need to call fadvise again after the file grows because the
1281 * kernel does not seem to apply fadvise to non-existing parts of the
1284 * Call fadvise _after_ having waited for the page writeback to
1285 * complete because the dirty page writeback semantic is not well
1286 * defined. So it can be expected to lead to lower throughput in
1289 ret
= posix_fadvise(outfd
, orig_offset
- stream
->max_sb_size
,
1290 stream
->max_sb_size
, POSIX_FADV_DONTNEED
);
1291 if (ret
&& ret
!= -ENOSYS
) {
1293 PERROR("posix_fadvise on fd %i", outfd
);
1298 * Initialise the necessary environnement :
1299 * - create a new context
1300 * - create the poll_pipe
1301 * - create the should_quit pipe (for signal handler)
1302 * - create the thread pipe (for splice)
1304 * Takes a function pointer as argument, this function is called when data is
1305 * available on a buffer. This function is responsible to do the
1306 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1307 * buffer configuration and then kernctl_put_next_subbuf at the end.
1309 * Returns a pointer to the new context or NULL on error.
1311 struct lttng_consumer_local_data
*lttng_consumer_create(
1312 enum lttng_consumer_type type
,
1313 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1314 struct lttng_consumer_local_data
*ctx
),
1315 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1316 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1317 int (*update_stream
)(uint64_t stream_key
, uint32_t state
))
1320 struct lttng_consumer_local_data
*ctx
;
1322 assert(consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1323 consumer_data
.type
== type
);
1324 consumer_data
.type
= type
;
1326 ctx
= zmalloc(sizeof(struct lttng_consumer_local_data
));
1328 PERROR("allocating context");
1332 ctx
->consumer_error_socket
= -1;
1333 ctx
->consumer_metadata_socket
= -1;
1334 pthread_mutex_init(&ctx
->metadata_socket_lock
, NULL
);
1335 /* assign the callbacks */
1336 ctx
->on_buffer_ready
= buffer_ready
;
1337 ctx
->on_recv_channel
= recv_channel
;
1338 ctx
->on_recv_stream
= recv_stream
;
1339 ctx
->on_update_stream
= update_stream
;
1341 ctx
->consumer_data_pipe
= lttng_pipe_open(0);
1342 if (!ctx
->consumer_data_pipe
) {
1343 goto error_poll_pipe
;
1346 ctx
->consumer_wakeup_pipe
= lttng_pipe_open(0);
1347 if (!ctx
->consumer_wakeup_pipe
) {
1348 goto error_wakeup_pipe
;
1351 ctx
->consumer_data_rotate_pipe
= lttng_pipe_open(0);
1352 if (!ctx
->consumer_data_rotate_pipe
) {
1353 goto error_data_rotate_pipe
;
1356 ctx
->consumer_metadata_rotate_pipe
= lttng_pipe_open(0);
1357 if (!ctx
->consumer_metadata_rotate_pipe
) {
1358 goto error_metadata_rotate_pipe
;
1361 ret
= pipe(ctx
->consumer_should_quit
);
1363 PERROR("Error creating recv pipe");
1364 goto error_quit_pipe
;
1367 ret
= pipe(ctx
->consumer_channel_pipe
);
1369 PERROR("Error creating channel pipe");
1370 goto error_channel_pipe
;
1373 ctx
->consumer_metadata_pipe
= lttng_pipe_open(0);
1374 if (!ctx
->consumer_metadata_pipe
) {
1375 goto error_metadata_pipe
;
1378 ctx
->channel_monitor_pipe
= -1;
1382 error_metadata_pipe
:
1383 utils_close_pipe(ctx
->consumer_channel_pipe
);
1385 utils_close_pipe(ctx
->consumer_should_quit
);
1387 lttng_pipe_destroy(ctx
->consumer_metadata_rotate_pipe
);
1388 error_metadata_rotate_pipe
:
1389 lttng_pipe_destroy(ctx
->consumer_data_rotate_pipe
);
1390 error_data_rotate_pipe
:
1391 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1393 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1401 * Iterate over all streams of the hashtable and free them properly.
1403 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1405 struct lttng_ht_iter iter
;
1406 struct lttng_consumer_stream
*stream
;
1413 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1415 * Ignore return value since we are currently cleaning up so any error
1418 (void) consumer_del_stream(stream
, ht
);
1422 lttng_ht_destroy(ht
);
1426 * Iterate over all streams of the metadata hashtable and free them
1429 static void destroy_metadata_stream_ht(struct lttng_ht
*ht
)
1431 struct lttng_ht_iter iter
;
1432 struct lttng_consumer_stream
*stream
;
1439 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1441 * Ignore return value since we are currently cleaning up so any error
1444 (void) consumer_del_metadata_stream(stream
, ht
);
1448 lttng_ht_destroy(ht
);
1452 * Close all fds associated with the instance and free the context.
1454 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1458 DBG("Consumer destroying it. Closing everything.");
1464 destroy_data_stream_ht(data_ht
);
1465 destroy_metadata_stream_ht(metadata_ht
);
1467 ret
= close(ctx
->consumer_error_socket
);
1471 ret
= close(ctx
->consumer_metadata_socket
);
1475 utils_close_pipe(ctx
->consumer_channel_pipe
);
1476 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1477 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1478 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1479 lttng_pipe_destroy(ctx
->consumer_data_rotate_pipe
);
1480 lttng_pipe_destroy(ctx
->consumer_metadata_rotate_pipe
);
1481 utils_close_pipe(ctx
->consumer_should_quit
);
1483 unlink(ctx
->consumer_command_sock_path
);
1488 * Write the metadata stream id on the specified file descriptor.
1490 static int write_relayd_metadata_id(int fd
,
1491 struct lttng_consumer_stream
*stream
,
1492 struct consumer_relayd_sock_pair
*relayd
, unsigned long padding
)
1495 struct lttcomm_relayd_metadata_payload hdr
;
1497 hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
1498 hdr
.padding_size
= htobe32(padding
);
1499 ret
= lttng_write(fd
, (void *) &hdr
, sizeof(hdr
));
1500 if (ret
< sizeof(hdr
)) {
1502 * This error means that the fd's end is closed so ignore the PERROR
1503 * not to clubber the error output since this can happen in a normal
1506 if (errno
!= EPIPE
) {
1507 PERROR("write metadata stream id");
1509 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno
);
1511 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1512 * handle writting the missing part so report that as an error and
1513 * don't lie to the caller.
1518 DBG("Metadata stream id %" PRIu64
" with padding %lu written before data",
1519 stream
->relayd_stream_id
, padding
);
1526 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1527 * core function for writing trace buffers to either the local filesystem or
1530 * It must be called with the stream lock held.
1532 * Careful review MUST be put if any changes occur!
1534 * Returns the number of bytes written
1536 ssize_t
lttng_consumer_on_read_subbuffer_mmap(
1537 struct lttng_consumer_local_data
*ctx
,
1538 struct lttng_consumer_stream
*stream
, unsigned long len
,
1539 unsigned long padding
,
1540 struct ctf_packet_index
*index
)
1542 unsigned long mmap_offset
;
1545 off_t orig_offset
= stream
->out_fd_offset
;
1546 /* Default is on the disk */
1547 int outfd
= stream
->out_fd
;
1548 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1549 unsigned int relayd_hang_up
= 0;
1551 /* RCU lock for the relayd pointer */
1554 /* Flag that the current stream if set for network streaming. */
1555 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1556 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1557 if (relayd
== NULL
) {
1563 /* get the offset inside the fd to mmap */
1564 switch (consumer_data
.type
) {
1565 case LTTNG_CONSUMER_KERNEL
:
1566 mmap_base
= stream
->mmap_base
;
1567 ret
= kernctl_get_mmap_read_offset(stream
->wait_fd
, &mmap_offset
);
1569 PERROR("tracer ctl get_mmap_read_offset");
1573 case LTTNG_CONSUMER32_UST
:
1574 case LTTNG_CONSUMER64_UST
:
1575 mmap_base
= lttng_ustctl_get_mmap_base(stream
);
1577 ERR("read mmap get mmap base for stream %s", stream
->name
);
1581 ret
= lttng_ustctl_get_mmap_read_offset(stream
, &mmap_offset
);
1583 PERROR("tracer ctl get_mmap_read_offset");
1589 ERR("Unknown consumer_data type");
1593 /* Handle stream on the relayd if the output is on the network */
1595 unsigned long netlen
= len
;
1598 * Lock the control socket for the complete duration of the function
1599 * since from this point on we will use the socket.
1601 if (stream
->metadata_flag
) {
1602 /* Metadata requires the control socket. */
1603 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1604 if (stream
->reset_metadata_flag
) {
1605 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1606 stream
->relayd_stream_id
,
1607 stream
->metadata_version
);
1612 stream
->reset_metadata_flag
= 0;
1614 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1617 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1622 /* Use the returned socket. */
1625 /* Write metadata stream id before payload */
1626 if (stream
->metadata_flag
) {
1627 ret
= write_relayd_metadata_id(outfd
, stream
, relayd
, padding
);
1634 /* No streaming, we have to set the len with the full padding */
1637 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1638 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1640 ERR("Reset metadata file");
1643 stream
->reset_metadata_flag
= 0;
1647 * Check if we need to change the tracefile before writing the packet.
1649 if (stream
->chan
->tracefile_size
> 0 &&
1650 (stream
->tracefile_size_current
+ len
) >
1651 stream
->chan
->tracefile_size
) {
1652 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1653 stream
->name
, stream
->chan
->tracefile_size
,
1654 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1655 stream
->out_fd
, &(stream
->tracefile_count_current
),
1658 ERR("Rotating output file");
1661 outfd
= stream
->out_fd
;
1663 if (stream
->index_file
) {
1664 lttng_index_file_put(stream
->index_file
);
1665 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1666 stream
->name
, stream
->uid
, stream
->gid
,
1667 stream
->chan
->tracefile_size
,
1668 stream
->tracefile_count_current
,
1669 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1670 if (!stream
->index_file
) {
1675 /* Reset current size because we just perform a rotation. */
1676 stream
->tracefile_size_current
= 0;
1677 stream
->out_fd_offset
= 0;
1680 stream
->tracefile_size_current
+= len
;
1682 index
->offset
= htobe64(stream
->out_fd_offset
);
1687 * This call guarantee that len or less is returned. It's impossible to
1688 * receive a ret value that is bigger than len.
1690 ret
= lttng_write(outfd
, mmap_base
+ mmap_offset
, len
);
1691 DBG("Consumer mmap write() ret %zd (len %lu)", ret
, len
);
1692 if (ret
< 0 || ((size_t) ret
!= len
)) {
1694 * Report error to caller if nothing was written else at least send the
1702 /* Socket operation failed. We consider the relayd dead */
1703 if (errno
== EPIPE
|| errno
== EINVAL
|| errno
== EBADF
) {
1705 * This is possible if the fd is closed on the other side
1706 * (outfd) or any write problem. It can be verbose a bit for a
1707 * normal execution if for instance the relayd is stopped
1708 * abruptly. This can happen so set this to a DBG statement.
1710 DBG("Consumer mmap write detected relayd hang up");
1712 /* Unhandled error, print it and stop function right now. */
1713 PERROR("Error in write mmap (ret %zd != len %lu)", ret
, len
);
1717 stream
->output_written
+= ret
;
1719 /* This call is useless on a socket so better save a syscall. */
1721 /* This won't block, but will start writeout asynchronously */
1722 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, len
,
1723 SYNC_FILE_RANGE_WRITE
);
1724 stream
->out_fd_offset
+= len
;
1725 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1730 * This is a special case that the relayd has closed its socket. Let's
1731 * cleanup the relayd object and all associated streams.
1733 if (relayd
&& relayd_hang_up
) {
1734 cleanup_relayd(relayd
, ctx
);
1738 /* Unlock only if ctrl socket used */
1739 if (relayd
&& stream
->metadata_flag
) {
1740 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1748 * Splice the data from the ring buffer to the tracefile.
1750 * It must be called with the stream lock held.
1752 * Returns the number of bytes spliced.
1754 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1755 struct lttng_consumer_local_data
*ctx
,
1756 struct lttng_consumer_stream
*stream
, unsigned long len
,
1757 unsigned long padding
,
1758 struct ctf_packet_index
*index
)
1760 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1762 off_t orig_offset
= stream
->out_fd_offset
;
1763 int fd
= stream
->wait_fd
;
1764 /* Default is on the disk */
1765 int outfd
= stream
->out_fd
;
1766 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1768 unsigned int relayd_hang_up
= 0;
1770 switch (consumer_data
.type
) {
1771 case LTTNG_CONSUMER_KERNEL
:
1773 case LTTNG_CONSUMER32_UST
:
1774 case LTTNG_CONSUMER64_UST
:
1775 /* Not supported for user space tracing */
1778 ERR("Unknown consumer_data type");
1782 /* RCU lock for the relayd pointer */
1785 /* Flag that the current stream if set for network streaming. */
1786 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1787 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1788 if (relayd
== NULL
) {
1793 splice_pipe
= stream
->splice_pipe
;
1795 /* Write metadata stream id before payload */
1797 unsigned long total_len
= len
;
1799 if (stream
->metadata_flag
) {
1801 * Lock the control socket for the complete duration of the function
1802 * since from this point on we will use the socket.
1804 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1806 if (stream
->reset_metadata_flag
) {
1807 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1808 stream
->relayd_stream_id
,
1809 stream
->metadata_version
);
1814 stream
->reset_metadata_flag
= 0;
1816 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
, relayd
,
1824 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1827 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1833 /* Use the returned socket. */
1836 /* No streaming, we have to set the len with the full padding */
1839 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1840 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1842 ERR("Reset metadata file");
1845 stream
->reset_metadata_flag
= 0;
1848 * Check if we need to change the tracefile before writing the packet.
1850 if (stream
->chan
->tracefile_size
> 0 &&
1851 (stream
->tracefile_size_current
+ len
) >
1852 stream
->chan
->tracefile_size
) {
1853 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1854 stream
->name
, stream
->chan
->tracefile_size
,
1855 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1856 stream
->out_fd
, &(stream
->tracefile_count_current
),
1860 ERR("Rotating output file");
1863 outfd
= stream
->out_fd
;
1865 if (stream
->index_file
) {
1866 lttng_index_file_put(stream
->index_file
);
1867 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1868 stream
->name
, stream
->uid
, stream
->gid
,
1869 stream
->chan
->tracefile_size
,
1870 stream
->tracefile_count_current
,
1871 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1872 if (!stream
->index_file
) {
1877 /* Reset current size because we just perform a rotation. */
1878 stream
->tracefile_size_current
= 0;
1879 stream
->out_fd_offset
= 0;
1882 stream
->tracefile_size_current
+= len
;
1883 index
->offset
= htobe64(stream
->out_fd_offset
);
1887 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1888 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1889 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1890 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1891 DBG("splice chan to pipe, ret %zd", ret_splice
);
1892 if (ret_splice
< 0) {
1895 PERROR("Error in relay splice");
1899 /* Handle stream on the relayd if the output is on the network */
1900 if (relayd
&& stream
->metadata_flag
) {
1901 size_t metadata_payload_size
=
1902 sizeof(struct lttcomm_relayd_metadata_payload
);
1904 /* Update counter to fit the spliced data */
1905 ret_splice
+= metadata_payload_size
;
1906 len
+= metadata_payload_size
;
1908 * We do this so the return value can match the len passed as
1909 * argument to this function.
1911 written
-= metadata_payload_size
;
1914 /* Splice data out */
1915 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1916 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1917 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1919 if (ret_splice
< 0) {
1924 } else if (ret_splice
> len
) {
1926 * We don't expect this code path to be executed but you never know
1927 * so this is an extra protection agains a buggy splice().
1930 written
+= ret_splice
;
1931 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice
,
1935 /* All good, update current len and continue. */
1939 /* This call is useless on a socket so better save a syscall. */
1941 /* This won't block, but will start writeout asynchronously */
1942 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1943 SYNC_FILE_RANGE_WRITE
);
1944 stream
->out_fd_offset
+= ret_splice
;
1946 stream
->output_written
+= ret_splice
;
1947 written
+= ret_splice
;
1950 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1956 * This is a special case that the relayd has closed its socket. Let's
1957 * cleanup the relayd object and all associated streams.
1959 if (relayd
&& relayd_hang_up
) {
1960 cleanup_relayd(relayd
, ctx
);
1961 /* Skip splice error so the consumer does not fail */
1966 /* send the appropriate error description to sessiond */
1969 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1972 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1975 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1980 if (relayd
&& stream
->metadata_flag
) {
1981 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1989 * Sample the snapshot positions for a specific fd
1991 * Returns 0 on success, < 0 on error
1993 int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream
*stream
)
1995 switch (consumer_data
.type
) {
1996 case LTTNG_CONSUMER_KERNEL
:
1997 return lttng_kconsumer_sample_snapshot_positions(stream
);
1998 case LTTNG_CONSUMER32_UST
:
1999 case LTTNG_CONSUMER64_UST
:
2000 return lttng_ustconsumer_sample_snapshot_positions(stream
);
2002 ERR("Unknown consumer_data type");
2008 * Take a snapshot for a specific fd
2010 * Returns 0 on success, < 0 on error
2012 int lttng_consumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2014 switch (consumer_data
.type
) {
2015 case LTTNG_CONSUMER_KERNEL
:
2016 return lttng_kconsumer_take_snapshot(stream
);
2017 case LTTNG_CONSUMER32_UST
:
2018 case LTTNG_CONSUMER64_UST
:
2019 return lttng_ustconsumer_take_snapshot(stream
);
2021 ERR("Unknown consumer_data type");
2028 * Get the produced position
2030 * Returns 0 on success, < 0 on error
2032 int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
2035 switch (consumer_data
.type
) {
2036 case LTTNG_CONSUMER_KERNEL
:
2037 return lttng_kconsumer_get_produced_snapshot(stream
, pos
);
2038 case LTTNG_CONSUMER32_UST
:
2039 case LTTNG_CONSUMER64_UST
:
2040 return lttng_ustconsumer_get_produced_snapshot(stream
, pos
);
2042 ERR("Unknown consumer_data type");
2049 * Get the consumed position
2051 * Returns 0 on success, < 0 on error
2053 int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
2056 switch (consumer_data
.type
) {
2057 case LTTNG_CONSUMER_KERNEL
:
2058 return lttng_kconsumer_get_consumed_snapshot(stream
, pos
);
2059 case LTTNG_CONSUMER32_UST
:
2060 case LTTNG_CONSUMER64_UST
:
2061 return lttng_ustconsumer_get_consumed_snapshot(stream
, pos
);
2063 ERR("Unknown consumer_data type");
2069 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
2070 int sock
, struct pollfd
*consumer_sockpoll
)
2072 switch (consumer_data
.type
) {
2073 case LTTNG_CONSUMER_KERNEL
:
2074 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2075 case LTTNG_CONSUMER32_UST
:
2076 case LTTNG_CONSUMER64_UST
:
2077 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2079 ERR("Unknown consumer_data type");
2085 void lttng_consumer_close_all_metadata(void)
2087 switch (consumer_data
.type
) {
2088 case LTTNG_CONSUMER_KERNEL
:
2090 * The Kernel consumer has a different metadata scheme so we don't
2091 * close anything because the stream will be closed by the session
2095 case LTTNG_CONSUMER32_UST
:
2096 case LTTNG_CONSUMER64_UST
:
2098 * Close all metadata streams. The metadata hash table is passed and
2099 * this call iterates over it by closing all wakeup fd. This is safe
2100 * because at this point we are sure that the metadata producer is
2101 * either dead or blocked.
2103 lttng_ustconsumer_close_all_metadata(metadata_ht
);
2106 ERR("Unknown consumer_data type");
2112 * Clean up a metadata stream and free its memory.
2114 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
2115 struct lttng_ht
*ht
)
2117 struct lttng_consumer_channel
*free_chan
= NULL
;
2121 * This call should NEVER receive regular stream. It must always be
2122 * metadata stream and this is crucial for data structure synchronization.
2124 assert(stream
->metadata_flag
);
2126 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
2128 pthread_mutex_lock(&consumer_data
.lock
);
2129 pthread_mutex_lock(&stream
->chan
->lock
);
2130 pthread_mutex_lock(&stream
->lock
);
2131 if (stream
->chan
->metadata_cache
) {
2132 /* Only applicable to userspace consumers. */
2133 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2136 /* Remove any reference to that stream. */
2137 consumer_stream_delete(stream
, ht
);
2139 /* Close down everything including the relayd if one. */
2140 consumer_stream_close(stream
);
2141 /* Destroy tracer buffers of the stream. */
2142 consumer_stream_destroy_buffers(stream
);
2144 /* Atomically decrement channel refcount since other threads can use it. */
2145 if (!uatomic_sub_return(&stream
->chan
->refcount
, 1)
2146 && !uatomic_read(&stream
->chan
->nb_init_stream_left
)) {
2147 /* Go for channel deletion! */
2148 free_chan
= stream
->chan
;
2152 * Nullify the stream reference so it is not used after deletion. The
2153 * channel lock MUST be acquired before being able to check for a NULL
2156 stream
->chan
->metadata_stream
= NULL
;
2158 if (stream
->chan
->metadata_cache
) {
2159 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2161 pthread_mutex_unlock(&stream
->lock
);
2162 pthread_mutex_unlock(&stream
->chan
->lock
);
2163 pthread_mutex_unlock(&consumer_data
.lock
);
2166 consumer_del_channel(free_chan
);
2169 consumer_stream_free(stream
);
2173 * Action done with the metadata stream when adding it to the consumer internal
2174 * data structures to handle it.
2176 int consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
)
2178 struct lttng_ht
*ht
= metadata_ht
;
2180 struct lttng_ht_iter iter
;
2181 struct lttng_ht_node_u64
*node
;
2186 DBG3("Adding metadata stream %" PRIu64
" to hash table", stream
->key
);
2188 pthread_mutex_lock(&consumer_data
.lock
);
2189 pthread_mutex_lock(&stream
->chan
->lock
);
2190 pthread_mutex_lock(&stream
->chan
->timer_lock
);
2191 pthread_mutex_lock(&stream
->lock
);
2194 * From here, refcounts are updated so be _careful_ when returning an error
2201 * Lookup the stream just to make sure it does not exist in our internal
2202 * state. This should NEVER happen.
2204 lttng_ht_lookup(ht
, &stream
->key
, &iter
);
2205 node
= lttng_ht_iter_get_node_u64(&iter
);
2209 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2210 * in terms of destroying the associated channel, because the action that
2211 * causes the count to become 0 also causes a stream to be added. The
2212 * channel deletion will thus be triggered by the following removal of this
2215 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
2216 /* Increment refcount before decrementing nb_init_stream_left */
2218 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
2221 lttng_ht_add_unique_u64(ht
, &stream
->node
);
2223 lttng_ht_add_unique_u64(consumer_data
.stream_per_chan_id_ht
,
2224 &stream
->node_channel_id
);
2227 * Add stream to the stream_list_ht of the consumer data. No need to steal
2228 * the key since the HT does not use it and we allow to add redundant keys
2231 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
2235 pthread_mutex_unlock(&stream
->lock
);
2236 pthread_mutex_unlock(&stream
->chan
->lock
);
2237 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
2238 pthread_mutex_unlock(&consumer_data
.lock
);
2243 * Delete data stream that are flagged for deletion (endpoint_status).
2245 static void validate_endpoint_status_data_stream(void)
2247 struct lttng_ht_iter iter
;
2248 struct lttng_consumer_stream
*stream
;
2250 DBG("Consumer delete flagged data stream");
2253 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2254 /* Validate delete flag of the stream */
2255 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2258 /* Delete it right now */
2259 consumer_del_stream(stream
, data_ht
);
2265 * Delete metadata stream that are flagged for deletion (endpoint_status).
2267 static void validate_endpoint_status_metadata_stream(
2268 struct lttng_poll_event
*pollset
)
2270 struct lttng_ht_iter iter
;
2271 struct lttng_consumer_stream
*stream
;
2273 DBG("Consumer delete flagged metadata stream");
2278 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2279 /* Validate delete flag of the stream */
2280 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2284 * Remove from pollset so the metadata thread can continue without
2285 * blocking on a deleted stream.
2287 lttng_poll_del(pollset
, stream
->wait_fd
);
2289 /* Delete it right now */
2290 consumer_del_metadata_stream(stream
, metadata_ht
);
2296 int rotate_notify_sessiond(struct lttng_consumer_local_data
*ctx
,
2301 fprintf(stderr
, "Notif send\n");
2303 ret
= write(ctx
->channel_rotate_pipe
, &key
, sizeof(key
));
2304 } while (ret
== -1 && errno
== EINTR
);
2306 PERROR("write to the channel rotate pipe");
2308 DBG("Sent channel rotation notification for channel key %"
2311 fprintf(stderr
, "Notif done\n");
2317 * Perform operations that need to be done after a stream has
2318 * rotated and released the stream lock.
2320 * Multiple rotations cannot occur simultaneously, so we know the state of the
2321 * "rotated" stream flag cannot change.
2323 * This MUST be called WITHOUT the stream lock held.
2326 int consumer_post_rotation(struct lttng_consumer_stream
*stream
,
2327 struct lttng_consumer_local_data
*ctx
)
2331 if (!stream
->rotated
) {
2335 pthread_mutex_lock(&stream
->chan
->lock
);
2336 switch (consumer_data
.type
) {
2337 case LTTNG_CONSUMER_KERNEL
:
2339 case LTTNG_CONSUMER32_UST
:
2340 case LTTNG_CONSUMER64_UST
:
2342 * The ust_metadata_pushed counter has been reset to 0, so now
2343 * we can wakeup the metadata thread so it dumps the metadata
2344 * cache to the new file.
2346 if (stream
->metadata_flag
) {
2347 consumer_metadata_wakeup_pipe(stream
->chan
);
2351 ERR("Unknown consumer_data type");
2355 fprintf(stderr
, "nr_pending: %lu\n", stream
->chan
->nr_stream_rotate_pending
);
2356 if (--stream
->chan
->nr_stream_rotate_pending
== 0) {
2357 ret
= rotate_notify_sessiond(ctx
, stream
->chan
->key
);
2359 pthread_mutex_unlock(&stream
->chan
->lock
);
2360 stream
->rotated
= 0;
2367 int handle_rotate_wakeup_pipe(struct lttng_consumer_local_data
*ctx
,
2368 struct lttng_pipe
*stream_pipe
)
2372 struct lttng_consumer_stream
*stream
;
2374 pipe_len
= lttng_pipe_read(stream_pipe
, &stream
, sizeof(stream
));
2375 if (pipe_len
< sizeof(stream
)) {
2377 PERROR("read metadata stream");
2379 ERR("Failed to read stream on metadata rotate pipe");
2384 pthread_mutex_lock(&stream
->lock
);
2385 fprintf(stderr
, "Rotate wakeup pipe, stream %lu\n", stream
->key
);
2386 ret
= lttng_consumer_rotate_stream(ctx
, stream
);
2387 pthread_mutex_unlock(&stream
->lock
);
2389 ERR("Failed to rotate metadata stream");
2392 ret
= consumer_post_rotation(stream
, ctx
);
2394 ERR("Failed after a rotation");
2405 * Thread polls on metadata file descriptor and write them on disk or on the
2408 void *consumer_thread_metadata_poll(void *data
)
2410 int ret
, i
, pollfd
, err
= -1;
2411 uint32_t revents
, nb_fd
;
2412 struct lttng_consumer_stream
*stream
= NULL
;
2413 struct lttng_ht_iter iter
;
2414 struct lttng_ht_node_u64
*node
;
2415 struct lttng_poll_event events
;
2416 struct lttng_consumer_local_data
*ctx
= data
;
2419 rcu_register_thread();
2421 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_METADATA
);
2423 if (testpoint(consumerd_thread_metadata
)) {
2424 goto error_testpoint
;
2427 health_code_update();
2429 DBG("Thread metadata poll started");
2431 /* Size is set to 1 for the consumer_metadata pipe */
2432 ret
= lttng_poll_create(&events
, 3, LTTNG_CLOEXEC
);
2434 ERR("Poll set creation failed");
2438 ret
= lttng_poll_add(&events
,
2439 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
), LPOLLIN
);
2444 ret
= lttng_poll_add(&events
,
2445 lttng_pipe_get_readfd(ctx
->consumer_metadata_rotate_pipe
), LPOLLIN
);
2451 DBG("Metadata main loop started");
2455 health_code_update();
2456 health_poll_entry();
2457 DBG("Metadata poll wait");
2458 ret
= lttng_poll_wait(&events
, -1);
2459 DBG("Metadata poll return from wait with %d fd(s)",
2460 LTTNG_POLL_GETNB(&events
));
2462 DBG("Metadata event caught in thread");
2464 if (errno
== EINTR
) {
2465 ERR("Poll EINTR caught");
2468 if (LTTNG_POLL_GETNB(&events
) == 0) {
2469 err
= 0; /* All is OK */
2476 /* From here, the event is a metadata wait fd */
2477 for (i
= 0; i
< nb_fd
; i
++) {
2478 health_code_update();
2480 revents
= LTTNG_POLL_GETEV(&events
, i
);
2481 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2484 /* No activity for this FD (poll implementation). */
2488 if (pollfd
== lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
)) {
2489 if (revents
& LPOLLIN
) {
2492 pipe_len
= lttng_pipe_read(ctx
->consumer_metadata_pipe
,
2493 &stream
, sizeof(stream
));
2494 if (pipe_len
< sizeof(stream
)) {
2496 PERROR("read metadata stream");
2499 * Remove the pipe from the poll set and continue the loop
2500 * since their might be data to consume.
2502 lttng_poll_del(&events
,
2503 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2504 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2508 /* A NULL stream means that the state has changed. */
2509 if (stream
== NULL
) {
2510 /* Check for deleted streams. */
2511 validate_endpoint_status_metadata_stream(&events
);
2515 DBG("Adding metadata stream %d to poll set",
2518 /* Add metadata stream to the global poll events list */
2519 lttng_poll_add(&events
, stream
->wait_fd
,
2520 LPOLLIN
| LPOLLPRI
| LPOLLHUP
);
2521 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2522 DBG("Metadata thread pipe hung up");
2524 * Remove the pipe from the poll set and continue the loop
2525 * since their might be data to consume.
2527 lttng_poll_del(&events
,
2528 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2529 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2532 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2536 /* Handle other stream */
2538 } else if (pollfd
== lttng_pipe_get_readfd(ctx
->consumer_metadata_rotate_pipe
)) {
2539 if (revents
& LPOLLIN
) {
2540 ret
= handle_rotate_wakeup_pipe(ctx
,
2541 ctx
->consumer_metadata_rotate_pipe
);
2543 ERR("Failed to rotate metadata stream");
2544 lttng_poll_del(&events
,
2545 lttng_pipe_get_readfd(
2546 ctx
->consumer_metadata_rotate_pipe
));
2547 lttng_pipe_read_close(
2548 ctx
->consumer_metadata_rotate_pipe
);
2551 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2552 DBG("Metadata rotate pipe hung up");
2553 fprintf(stderr
, "Metadata rotate pipe hung up");
2555 * Remove the pipe from the poll set and continue the loop
2556 * since their might be data to consume.
2558 lttng_poll_del(&events
,
2559 lttng_pipe_get_readfd(ctx
->consumer_metadata_rotate_pipe
));
2560 lttng_pipe_read_close(ctx
->consumer_metadata_rotate_pipe
);
2563 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2571 uint64_t tmp_id
= (uint64_t) pollfd
;
2573 lttng_ht_lookup(metadata_ht
, &tmp_id
, &iter
);
2575 node
= lttng_ht_iter_get_node_u64(&iter
);
2578 stream
= caa_container_of(node
, struct lttng_consumer_stream
,
2581 if (revents
& (LPOLLIN
| LPOLLPRI
)) {
2582 /* Get the data out of the metadata file descriptor */
2583 DBG("Metadata available on fd %d", pollfd
);
2584 assert(stream
->wait_fd
== pollfd
);
2587 health_code_update();
2589 len
= ctx
->on_buffer_ready(stream
, ctx
);
2591 * We don't check the return value here since if we get
2592 * a negative len, it means an error occurred thus we
2593 * simply remove it from the poll set and free the
2598 /* It's ok to have an unavailable sub-buffer */
2599 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2600 /* Clean up stream from consumer and free it. */
2601 lttng_poll_del(&events
, stream
->wait_fd
);
2602 consumer_del_metadata_stream(stream
, metadata_ht
);
2604 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2605 DBG("Metadata fd %d is hup|err.", pollfd
);
2606 if (!stream
->hangup_flush_done
2607 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2608 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2609 DBG("Attempting to flush and consume the UST buffers");
2610 lttng_ustconsumer_on_stream_hangup(stream
);
2612 /* We just flushed the stream now read it. */
2614 health_code_update();
2616 len
= ctx
->on_buffer_ready(stream
, ctx
);
2618 * We don't check the return value here since if we get
2619 * a negative len, it means an error occurred thus we
2620 * simply remove it from the poll set and free the
2626 lttng_poll_del(&events
, stream
->wait_fd
);
2628 * This call update the channel states, closes file descriptors
2629 * and securely free the stream.
2631 consumer_del_metadata_stream(stream
, metadata_ht
);
2633 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2637 /* Release RCU lock for the stream looked up */
2645 DBG("Metadata poll thread exiting");
2647 lttng_poll_clean(&events
);
2652 ERR("Health error occurred in %s", __func__
);
2654 health_unregister(health_consumerd
);
2655 rcu_unregister_thread();
2660 * This thread polls the fds in the set to consume the data and write
2661 * it to tracefile if necessary.
2663 void *consumer_thread_data_poll(void *data
)
2665 int num_rdy
, num_hup
, high_prio
, ret
, i
, err
= -1;
2666 struct pollfd
*pollfd
= NULL
;
2667 /* local view of the streams */
2668 struct lttng_consumer_stream
**local_stream
= NULL
, *new_stream
= NULL
;
2669 /* local view of consumer_data.fds_count */
2670 int nb_fd
= 0, nb_pipes_fd
;
2671 struct lttng_consumer_local_data
*ctx
= data
;
2674 rcu_register_thread();
2676 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_DATA
);
2678 if (testpoint(consumerd_thread_data
)) {
2679 goto error_testpoint
;
2682 health_code_update();
2684 local_stream
= zmalloc(sizeof(struct lttng_consumer_stream
*));
2685 if (local_stream
== NULL
) {
2686 PERROR("local_stream malloc");
2691 health_code_update();
2697 * the fds set has been updated, we need to update our
2698 * local array as well
2700 pthread_mutex_lock(&consumer_data
.lock
);
2701 if (consumer_data
.need_update
) {
2706 local_stream
= NULL
;
2709 * Allocate for all fds + 3:
2710 * +1 for the consumer_data_pipe
2711 * +1 for wake up pipe
2712 * +1 for consumer_data_rotate_pipe.
2715 pollfd
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) * sizeof(struct pollfd
));
2716 if (pollfd
== NULL
) {
2717 PERROR("pollfd malloc");
2718 pthread_mutex_unlock(&consumer_data
.lock
);
2722 local_stream
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) *
2723 sizeof(struct lttng_consumer_stream
*));
2724 if (local_stream
== NULL
) {
2725 PERROR("local_stream malloc");
2726 pthread_mutex_unlock(&consumer_data
.lock
);
2729 ret
= update_poll_array(ctx
, &pollfd
, local_stream
,
2732 ERR("Error in allocating pollfd or local_outfds");
2733 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2734 pthread_mutex_unlock(&consumer_data
.lock
);
2738 consumer_data
.need_update
= 0;
2740 pthread_mutex_unlock(&consumer_data
.lock
);
2742 /* No FDs and consumer_quit, consumer_cleanup the thread */
2743 if (nb_fd
== 0 && CMM_LOAD_SHARED(consumer_quit
) == 1) {
2744 err
= 0; /* All is OK */
2747 /* poll on the array of fds */
2749 DBG("polling on %d fd", nb_fd
+ nb_pipes_fd
);
2750 if (testpoint(consumerd_thread_data_poll
)) {
2753 health_poll_entry();
2754 num_rdy
= poll(pollfd
, nb_fd
+ nb_pipes_fd
, -1);
2756 DBG("poll num_rdy : %d", num_rdy
);
2757 if (num_rdy
== -1) {
2759 * Restart interrupted system call.
2761 if (errno
== EINTR
) {
2764 PERROR("Poll error");
2765 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2767 } else if (num_rdy
== 0) {
2768 DBG("Polling thread timed out");
2772 if (caa_unlikely(data_consumption_paused
)) {
2773 DBG("Data consumption paused, sleeping...");
2779 * If the consumer_data_pipe triggered poll go directly to the
2780 * beginning of the loop to update the array. We want to prioritize
2781 * array update over low-priority reads.
2783 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2784 ssize_t pipe_readlen
;
2786 DBG("consumer_data_pipe wake up");
2787 pipe_readlen
= lttng_pipe_read(ctx
->consumer_data_pipe
,
2788 &new_stream
, sizeof(new_stream
));
2789 if (pipe_readlen
< sizeof(new_stream
)) {
2790 PERROR("Consumer data pipe");
2791 /* Continue so we can at least handle the current stream(s). */
2796 * If the stream is NULL, just ignore it. It's also possible that
2797 * the sessiond poll thread changed the consumer_quit state and is
2798 * waking us up to test it.
2800 if (new_stream
== NULL
) {
2801 validate_endpoint_status_data_stream();
2805 /* Continue to update the local streams and handle prio ones */
2809 /* Handle wakeup pipe. */
2810 if (pollfd
[nb_fd
+ 1].revents
& (POLLIN
| POLLPRI
)) {
2812 ssize_t pipe_readlen
;
2814 pipe_readlen
= lttng_pipe_read(ctx
->consumer_wakeup_pipe
, &dummy
,
2816 if (pipe_readlen
< 0) {
2817 PERROR("Consumer data wakeup pipe");
2819 /* We've been awakened to handle stream(s). */
2820 ctx
->has_wakeup
= 0;
2823 /* Handle consumer_data_rotate_pipe. */
2824 if (pollfd
[nb_fd
+ 2].revents
& (POLLIN
| POLLPRI
)) {
2825 fprintf(stderr
, "data wakeup pipe\n");
2826 ret
= handle_rotate_wakeup_pipe(ctx
,
2827 ctx
->consumer_data_rotate_pipe
);
2829 ERR("Failed to rotate metadata stream");
2834 /* Take care of high priority channels first. */
2835 for (i
= 0; i
< nb_fd
; i
++) {
2836 health_code_update();
2838 if (local_stream
[i
] == NULL
) {
2841 if (pollfd
[i
].revents
& POLLPRI
) {
2842 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2844 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2845 /* it's ok to have an unavailable sub-buffer */
2846 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2847 /* Clean the stream and free it. */
2848 consumer_del_stream(local_stream
[i
], data_ht
);
2849 local_stream
[i
] = NULL
;
2850 } else if (len
> 0) {
2851 local_stream
[i
]->data_read
= 1;
2857 * If we read high prio channel in this loop, try again
2858 * for more high prio data.
2864 /* Take care of low priority channels. */
2865 for (i
= 0; i
< nb_fd
; i
++) {
2866 health_code_update();
2868 if (local_stream
[i
] == NULL
) {
2871 if ((pollfd
[i
].revents
& POLLIN
) ||
2872 local_stream
[i
]->hangup_flush_done
||
2873 local_stream
[i
]->has_data
) {
2874 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2875 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2876 /* it's ok to have an unavailable sub-buffer */
2877 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2878 /* Clean the stream and free it. */
2879 consumer_del_stream(local_stream
[i
], data_ht
);
2880 local_stream
[i
] = NULL
;
2881 } else if (len
> 0) {
2882 local_stream
[i
]->data_read
= 1;
2887 /* Handle hangup and errors */
2888 for (i
= 0; i
< nb_fd
; i
++) {
2889 health_code_update();
2891 if (local_stream
[i
] == NULL
) {
2894 if (!local_stream
[i
]->hangup_flush_done
2895 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2896 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2897 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2898 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2900 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2901 /* Attempt read again, for the data we just flushed. */
2902 local_stream
[i
]->data_read
= 1;
2905 * If the poll flag is HUP/ERR/NVAL and we have
2906 * read no data in this pass, we can remove the
2907 * stream from its hash table.
2909 if ((pollfd
[i
].revents
& POLLHUP
)) {
2910 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2911 if (!local_stream
[i
]->data_read
) {
2912 consumer_del_stream(local_stream
[i
], data_ht
);
2913 local_stream
[i
] = NULL
;
2916 } else if (pollfd
[i
].revents
& POLLERR
) {
2917 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2918 if (!local_stream
[i
]->data_read
) {
2919 consumer_del_stream(local_stream
[i
], data_ht
);
2920 local_stream
[i
] = NULL
;
2923 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2924 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2925 if (!local_stream
[i
]->data_read
) {
2926 consumer_del_stream(local_stream
[i
], data_ht
);
2927 local_stream
[i
] = NULL
;
2931 if (local_stream
[i
] != NULL
) {
2932 local_stream
[i
]->data_read
= 0;
2939 DBG("polling thread exiting");
2944 * Close the write side of the pipe so epoll_wait() in
2945 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2946 * read side of the pipe. If we close them both, epoll_wait strangely does
2947 * not return and could create a endless wait period if the pipe is the
2948 * only tracked fd in the poll set. The thread will take care of closing
2951 (void) lttng_pipe_write_close(ctx
->consumer_metadata_pipe
);
2952 (void) lttng_pipe_write_close(ctx
->consumer_metadata_rotate_pipe
);
2957 ERR("Health error occurred in %s", __func__
);
2959 health_unregister(health_consumerd
);
2961 rcu_unregister_thread();
2966 * Close wake-up end of each stream belonging to the channel. This will
2967 * allow the poll() on the stream read-side to detect when the
2968 * write-side (application) finally closes them.
2971 void consumer_close_channel_streams(struct lttng_consumer_channel
*channel
)
2973 struct lttng_ht
*ht
;
2974 struct lttng_consumer_stream
*stream
;
2975 struct lttng_ht_iter iter
;
2977 ht
= consumer_data
.stream_per_chan_id_ht
;
2980 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2981 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
2982 ht
->match_fct
, &channel
->key
,
2983 &iter
.iter
, stream
, node_channel_id
.node
) {
2985 * Protect against teardown with mutex.
2987 pthread_mutex_lock(&stream
->lock
);
2988 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
2991 switch (consumer_data
.type
) {
2992 case LTTNG_CONSUMER_KERNEL
:
2994 case LTTNG_CONSUMER32_UST
:
2995 case LTTNG_CONSUMER64_UST
:
2996 if (stream
->metadata_flag
) {
2997 /* Safe and protected by the stream lock. */
2998 lttng_ustconsumer_close_metadata(stream
->chan
);
3001 * Note: a mutex is taken internally within
3002 * liblttng-ust-ctl to protect timer wakeup_fd
3003 * use from concurrent close.
3005 lttng_ustconsumer_close_stream_wakeup(stream
);
3009 ERR("Unknown consumer_data type");
3013 pthread_mutex_unlock(&stream
->lock
);
3018 static void destroy_channel_ht(struct lttng_ht
*ht
)
3020 struct lttng_ht_iter iter
;
3021 struct lttng_consumer_channel
*channel
;
3029 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, channel
, wait_fd_node
.node
) {
3030 ret
= lttng_ht_del(ht
, &iter
);
3035 lttng_ht_destroy(ht
);
3039 * This thread polls the channel fds to detect when they are being
3040 * closed. It closes all related streams if the channel is detected as
3041 * closed. It is currently only used as a shim layer for UST because the
3042 * consumerd needs to keep the per-stream wakeup end of pipes open for
3045 void *consumer_thread_channel_poll(void *data
)
3047 int ret
, i
, pollfd
, err
= -1;
3048 uint32_t revents
, nb_fd
;
3049 struct lttng_consumer_channel
*chan
= NULL
;
3050 struct lttng_ht_iter iter
;
3051 struct lttng_ht_node_u64
*node
;
3052 struct lttng_poll_event events
;
3053 struct lttng_consumer_local_data
*ctx
= data
;
3054 struct lttng_ht
*channel_ht
;
3056 rcu_register_thread();
3058 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_CHANNEL
);
3060 if (testpoint(consumerd_thread_channel
)) {
3061 goto error_testpoint
;
3064 health_code_update();
3066 channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3068 /* ENOMEM at this point. Better to bail out. */
3072 DBG("Thread channel poll started");
3074 /* Size is set to 1 for the consumer_channel pipe */
3075 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
3077 ERR("Poll set creation failed");
3081 ret
= lttng_poll_add(&events
, ctx
->consumer_channel_pipe
[0], LPOLLIN
);
3087 DBG("Channel main loop started");
3091 health_code_update();
3092 DBG("Channel poll wait");
3093 health_poll_entry();
3094 ret
= lttng_poll_wait(&events
, -1);
3095 DBG("Channel poll return from wait with %d fd(s)",
3096 LTTNG_POLL_GETNB(&events
));
3098 DBG("Channel event caught in thread");
3100 if (errno
== EINTR
) {
3101 ERR("Poll EINTR caught");
3104 if (LTTNG_POLL_GETNB(&events
) == 0) {
3105 err
= 0; /* All is OK */
3112 /* From here, the event is a channel wait fd */
3113 for (i
= 0; i
< nb_fd
; i
++) {
3114 health_code_update();
3116 revents
= LTTNG_POLL_GETEV(&events
, i
);
3117 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3120 /* No activity for this FD (poll implementation). */
3124 if (pollfd
== ctx
->consumer_channel_pipe
[0]) {
3125 if (revents
& LPOLLIN
) {
3126 enum consumer_channel_action action
;
3129 ret
= read_channel_pipe(ctx
, &chan
, &key
, &action
);
3132 ERR("Error reading channel pipe");
3134 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3139 case CONSUMER_CHANNEL_ADD
:
3140 DBG("Adding channel %d to poll set",
3143 lttng_ht_node_init_u64(&chan
->wait_fd_node
,
3146 lttng_ht_add_unique_u64(channel_ht
,
3147 &chan
->wait_fd_node
);
3149 /* Add channel to the global poll events list */
3150 lttng_poll_add(&events
, chan
->wait_fd
,
3151 LPOLLERR
| LPOLLHUP
);
3153 case CONSUMER_CHANNEL_DEL
:
3156 * This command should never be called if the channel
3157 * has streams monitored by either the data or metadata
3158 * thread. The consumer only notify this thread with a
3159 * channel del. command if it receives a destroy
3160 * channel command from the session daemon that send it
3161 * if a command prior to the GET_CHANNEL failed.
3165 chan
= consumer_find_channel(key
);
3168 ERR("UST consumer get channel key %" PRIu64
" not found for del channel", key
);
3171 lttng_poll_del(&events
, chan
->wait_fd
);
3172 iter
.iter
.node
= &chan
->wait_fd_node
.node
;
3173 ret
= lttng_ht_del(channel_ht
, &iter
);
3176 switch (consumer_data
.type
) {
3177 case LTTNG_CONSUMER_KERNEL
:
3179 case LTTNG_CONSUMER32_UST
:
3180 case LTTNG_CONSUMER64_UST
:
3181 health_code_update();
3182 /* Destroy streams that might have been left in the stream list. */
3183 clean_channel_stream_list(chan
);
3186 ERR("Unknown consumer_data type");
3191 * Release our own refcount. Force channel deletion even if
3192 * streams were not initialized.
3194 if (!uatomic_sub_return(&chan
->refcount
, 1)) {
3195 consumer_del_channel(chan
);
3200 case CONSUMER_CHANNEL_QUIT
:
3202 * Remove the pipe from the poll set and continue the loop
3203 * since their might be data to consume.
3205 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3208 ERR("Unknown action");
3211 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3212 DBG("Channel thread pipe hung up");
3214 * Remove the pipe from the poll set and continue the loop
3215 * since their might be data to consume.
3217 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3220 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3224 /* Handle other stream */
3230 uint64_t tmp_id
= (uint64_t) pollfd
;
3232 lttng_ht_lookup(channel_ht
, &tmp_id
, &iter
);
3234 node
= lttng_ht_iter_get_node_u64(&iter
);
3237 chan
= caa_container_of(node
, struct lttng_consumer_channel
,
3240 /* Check for error event */
3241 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3242 DBG("Channel fd %d is hup|err.", pollfd
);
3244 lttng_poll_del(&events
, chan
->wait_fd
);
3245 ret
= lttng_ht_del(channel_ht
, &iter
);
3249 * This will close the wait fd for each stream associated to
3250 * this channel AND monitored by the data/metadata thread thus
3251 * will be clean by the right thread.
3253 consumer_close_channel_streams(chan
);
3255 /* Release our own refcount */
3256 if (!uatomic_sub_return(&chan
->refcount
, 1)
3257 && !uatomic_read(&chan
->nb_init_stream_left
)) {
3258 consumer_del_channel(chan
);
3261 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3266 /* Release RCU lock for the channel looked up */
3274 lttng_poll_clean(&events
);
3276 destroy_channel_ht(channel_ht
);
3279 DBG("Channel poll thread exiting");
3282 ERR("Health error occurred in %s", __func__
);
3284 health_unregister(health_consumerd
);
3285 rcu_unregister_thread();
3289 static int set_metadata_socket(struct lttng_consumer_local_data
*ctx
,
3290 struct pollfd
*sockpoll
, int client_socket
)
3297 ret
= lttng_consumer_poll_socket(sockpoll
);
3301 DBG("Metadata connection on client_socket");
3303 /* Blocking call, waiting for transmission */
3304 ctx
->consumer_metadata_socket
= lttcomm_accept_unix_sock(client_socket
);
3305 if (ctx
->consumer_metadata_socket
< 0) {
3306 WARN("On accept metadata");
3317 * This thread listens on the consumerd socket and receives the file
3318 * descriptors from the session daemon.
3320 void *consumer_thread_sessiond_poll(void *data
)
3322 int sock
= -1, client_socket
, ret
, err
= -1;
3324 * structure to poll for incoming data on communication socket avoids
3325 * making blocking sockets.
3327 struct pollfd consumer_sockpoll
[2];
3328 struct lttng_consumer_local_data
*ctx
= data
;
3330 rcu_register_thread();
3332 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_SESSIOND
);
3334 if (testpoint(consumerd_thread_sessiond
)) {
3335 goto error_testpoint
;
3338 health_code_update();
3340 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
3341 unlink(ctx
->consumer_command_sock_path
);
3342 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
3343 if (client_socket
< 0) {
3344 ERR("Cannot create command socket");
3348 ret
= lttcomm_listen_unix_sock(client_socket
);
3353 DBG("Sending ready command to lttng-sessiond");
3354 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
3355 /* return < 0 on error, but == 0 is not fatal */
3357 ERR("Error sending ready command to lttng-sessiond");
3361 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3362 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
3363 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
3364 consumer_sockpoll
[1].fd
= client_socket
;
3365 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3367 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3375 DBG("Connection on client_socket");
3377 /* Blocking call, waiting for transmission */
3378 sock
= lttcomm_accept_unix_sock(client_socket
);
3385 * Setup metadata socket which is the second socket connection on the
3386 * command unix socket.
3388 ret
= set_metadata_socket(ctx
, consumer_sockpoll
, client_socket
);
3397 /* This socket is not useful anymore. */
3398 ret
= close(client_socket
);
3400 PERROR("close client_socket");
3404 /* update the polling structure to poll on the established socket */
3405 consumer_sockpoll
[1].fd
= sock
;
3406 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3409 health_code_update();
3411 health_poll_entry();
3412 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3421 DBG("Incoming command on sock");
3422 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
3425 * This could simply be a session daemon quitting. Don't output
3428 DBG("Communication interrupted on command socket");
3432 if (CMM_LOAD_SHARED(consumer_quit
)) {
3433 DBG("consumer_thread_receive_fds received quit from signal");
3434 err
= 0; /* All is OK */
3437 DBG("received command on sock");
3443 DBG("Consumer thread sessiond poll exiting");
3446 * Close metadata streams since the producer is the session daemon which
3449 * NOTE: for now, this only applies to the UST tracer.
3451 lttng_consumer_close_all_metadata();
3454 * when all fds have hung up, the polling thread
3457 CMM_STORE_SHARED(consumer_quit
, 1);
3460 * Notify the data poll thread to poll back again and test the
3461 * consumer_quit state that we just set so to quit gracefully.
3463 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
3465 notify_channel_pipe(ctx
, NULL
, -1, CONSUMER_CHANNEL_QUIT
);
3467 notify_health_quit_pipe(health_quit_pipe
);
3469 /* Cleaning up possibly open sockets. */
3473 PERROR("close sock sessiond poll");
3476 if (client_socket
>= 0) {
3477 ret
= close(client_socket
);
3479 PERROR("close client_socket sessiond poll");
3486 ERR("Health error occurred in %s", __func__
);
3488 health_unregister(health_consumerd
);
3490 rcu_unregister_thread();
3494 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
3495 struct lttng_consumer_local_data
*ctx
)
3500 pthread_mutex_lock(&stream
->lock
);
3501 if (stream
->metadata_flag
) {
3502 pthread_mutex_lock(&stream
->metadata_rdv_lock
);
3505 switch (consumer_data
.type
) {
3506 case LTTNG_CONSUMER_KERNEL
:
3507 ret
= lttng_kconsumer_read_subbuffer(stream
, ctx
);
3509 case LTTNG_CONSUMER32_UST
:
3510 case LTTNG_CONSUMER64_UST
:
3511 ret
= lttng_ustconsumer_read_subbuffer(stream
, ctx
);
3514 ERR("Unknown consumer_data type");
3520 if (stream
->metadata_flag
) {
3521 pthread_cond_broadcast(&stream
->metadata_rdv
);
3522 pthread_mutex_unlock(&stream
->metadata_rdv_lock
);
3524 fprintf(stderr
, "rotated: %d\n", stream
->rotated
);
3525 pthread_mutex_unlock(&stream
->lock
);
3527 rotate_ret
= consumer_post_rotation(stream
, ctx
);
3528 if (rotate_ret
< 0) {
3529 ERR("Failed after a rotation");
3536 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3538 switch (consumer_data
.type
) {
3539 case LTTNG_CONSUMER_KERNEL
:
3540 return lttng_kconsumer_on_recv_stream(stream
);
3541 case LTTNG_CONSUMER32_UST
:
3542 case LTTNG_CONSUMER64_UST
:
3543 return lttng_ustconsumer_on_recv_stream(stream
);
3545 ERR("Unknown consumer_data type");
3552 * Allocate and set consumer data hash tables.
3554 int lttng_consumer_init(void)
3556 consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3557 if (!consumer_data
.channel_ht
) {
3561 consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3562 if (!consumer_data
.relayd_ht
) {
3566 consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3567 if (!consumer_data
.stream_list_ht
) {
3571 consumer_data
.stream_per_chan_id_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3572 if (!consumer_data
.stream_per_chan_id_ht
) {
3576 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3581 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3593 * Process the ADD_RELAYD command receive by a consumer.
3595 * This will create a relayd socket pair and add it to the relayd hash table.
3596 * The caller MUST acquire a RCU read side lock before calling it.
3598 void consumer_add_relayd_socket(uint64_t net_seq_idx
, int sock_type
,
3599 struct lttng_consumer_local_data
*ctx
, int sock
,
3600 struct pollfd
*consumer_sockpoll
,
3601 struct lttcomm_relayd_sock
*relayd_sock
, uint64_t sessiond_id
,
3602 uint64_t relayd_session_id
)
3604 int fd
= -1, ret
= -1, relayd_created
= 0;
3605 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3606 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3609 assert(relayd_sock
);
3611 DBG("Consumer adding relayd socket (idx: %" PRIu64
")", net_seq_idx
);
3613 /* Get relayd reference if exists. */
3614 relayd
= consumer_find_relayd(net_seq_idx
);
3615 if (relayd
== NULL
) {
3616 assert(sock_type
== LTTNG_STREAM_CONTROL
);
3617 /* Not found. Allocate one. */
3618 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
3619 if (relayd
== NULL
) {
3620 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3623 relayd
->sessiond_session_id
= sessiond_id
;
3628 * This code path MUST continue to the consumer send status message to
3629 * we can notify the session daemon and continue our work without
3630 * killing everything.
3634 * relayd key should never be found for control socket.
3636 assert(sock_type
!= LTTNG_STREAM_CONTROL
);
3639 /* First send a status message before receiving the fds. */
3640 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
3642 /* Somehow, the session daemon is not responding anymore. */
3643 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3644 goto error_nosignal
;
3647 /* Poll on consumer socket. */
3648 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3650 /* Needing to exit in the middle of a command: error. */
3651 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
3652 goto error_nosignal
;
3655 /* Get relayd socket from session daemon */
3656 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
3657 if (ret
!= sizeof(fd
)) {
3658 fd
= -1; /* Just in case it gets set with an invalid value. */
3661 * Failing to receive FDs might indicate a major problem such as
3662 * reaching a fd limit during the receive where the kernel returns a
3663 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3664 * don't take any chances and stop everything.
3666 * XXX: Feature request #558 will fix that and avoid this possible
3667 * issue when reaching the fd limit.
3669 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
3670 ret_code
= LTTCOMM_CONSUMERD_ERROR_RECV_FD
;
3674 /* Copy socket information and received FD */
3675 switch (sock_type
) {
3676 case LTTNG_STREAM_CONTROL
:
3677 /* Copy received lttcomm socket */
3678 lttcomm_copy_sock(&relayd
->control_sock
.sock
, &relayd_sock
->sock
);
3679 ret
= lttcomm_create_sock(&relayd
->control_sock
.sock
);
3680 /* Handle create_sock error. */
3682 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3686 * Close the socket created internally by
3687 * lttcomm_create_sock, so we can replace it by the one
3688 * received from sessiond.
3690 if (close(relayd
->control_sock
.sock
.fd
)) {
3694 /* Assign new file descriptor */
3695 relayd
->control_sock
.sock
.fd
= fd
;
3696 fd
= -1; /* For error path */
3697 /* Assign version values. */
3698 relayd
->control_sock
.major
= relayd_sock
->major
;
3699 relayd
->control_sock
.minor
= relayd_sock
->minor
;
3701 relayd
->relayd_session_id
= relayd_session_id
;
3704 case LTTNG_STREAM_DATA
:
3705 /* Copy received lttcomm socket */
3706 lttcomm_copy_sock(&relayd
->data_sock
.sock
, &relayd_sock
->sock
);
3707 ret
= lttcomm_create_sock(&relayd
->data_sock
.sock
);
3708 /* Handle create_sock error. */
3710 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3714 * Close the socket created internally by
3715 * lttcomm_create_sock, so we can replace it by the one
3716 * received from sessiond.
3718 if (close(relayd
->data_sock
.sock
.fd
)) {
3722 /* Assign new file descriptor */
3723 relayd
->data_sock
.sock
.fd
= fd
;
3724 fd
= -1; /* for eventual error paths */
3725 /* Assign version values. */
3726 relayd
->data_sock
.major
= relayd_sock
->major
;
3727 relayd
->data_sock
.minor
= relayd_sock
->minor
;
3730 ERR("Unknown relayd socket type (%d)", sock_type
);
3731 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3735 DBG("Consumer %s socket created successfully with net idx %" PRIu64
" (fd: %d)",
3736 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
3737 relayd
->net_seq_idx
, fd
);
3739 /* We successfully added the socket. Send status back. */
3740 ret
= consumer_send_status_msg(sock
, ret_code
);
3742 /* Somehow, the session daemon is not responding anymore. */
3743 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3744 goto error_nosignal
;
3748 * Add relayd socket pair to consumer data hashtable. If object already
3749 * exists or on error, the function gracefully returns.
3757 if (consumer_send_status_msg(sock
, ret_code
) < 0) {
3758 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3762 /* Close received socket if valid. */
3765 PERROR("close received socket");
3769 if (relayd_created
) {
3775 * Try to lock the stream mutex.
3777 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3779 static int stream_try_lock(struct lttng_consumer_stream
*stream
)
3786 * Try to lock the stream mutex. On failure, we know that the stream is
3787 * being used else where hence there is data still being extracted.
3789 ret
= pthread_mutex_trylock(&stream
->lock
);
3791 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3803 * Search for a relayd associated to the session id and return the reference.
3805 * A rcu read side lock MUST be acquire before calling this function and locked
3806 * until the relayd object is no longer necessary.
3808 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
3810 struct lttng_ht_iter iter
;
3811 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3813 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3814 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
3817 * Check by sessiond id which is unique here where the relayd session
3818 * id might not be when having multiple relayd.
3820 if (relayd
->sessiond_session_id
== id
) {
3821 /* Found the relayd. There can be only one per id. */
3833 * Check if for a given session id there is still data needed to be extract
3836 * Return 1 if data is pending or else 0 meaning ready to be read.
3838 int consumer_data_pending(uint64_t id
)
3841 struct lttng_ht_iter iter
;
3842 struct lttng_ht
*ht
;
3843 struct lttng_consumer_stream
*stream
;
3844 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3845 int (*data_pending
)(struct lttng_consumer_stream
*);
3847 DBG("Consumer data pending command on session id %" PRIu64
, id
);
3850 pthread_mutex_lock(&consumer_data
.lock
);
3852 switch (consumer_data
.type
) {
3853 case LTTNG_CONSUMER_KERNEL
:
3854 data_pending
= lttng_kconsumer_data_pending
;
3856 case LTTNG_CONSUMER32_UST
:
3857 case LTTNG_CONSUMER64_UST
:
3858 data_pending
= lttng_ustconsumer_data_pending
;
3861 ERR("Unknown consumer data type");
3865 /* Ease our life a bit */
3866 ht
= consumer_data
.stream_list_ht
;
3868 relayd
= find_relayd_by_session_id(id
);
3870 /* Send init command for data pending. */
3871 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3872 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
3873 relayd
->relayd_session_id
);
3874 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3876 /* Communication error thus the relayd so no data pending. */
3877 goto data_not_pending
;
3881 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3882 ht
->hash_fct(&id
, lttng_ht_seed
),
3884 &iter
.iter
, stream
, node_session_id
.node
) {
3885 /* If this call fails, the stream is being used hence data pending. */
3886 ret
= stream_try_lock(stream
);
3892 * A removed node from the hash table indicates that the stream has
3893 * been deleted thus having a guarantee that the buffers are closed
3894 * on the consumer side. However, data can still be transmitted
3895 * over the network so don't skip the relayd check.
3897 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
3899 /* Check the stream if there is data in the buffers. */
3900 ret
= data_pending(stream
);
3902 pthread_mutex_unlock(&stream
->lock
);
3909 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3910 if (stream
->metadata_flag
) {
3911 ret
= relayd_quiescent_control(&relayd
->control_sock
,
3912 stream
->relayd_stream_id
);
3914 ret
= relayd_data_pending(&relayd
->control_sock
,
3915 stream
->relayd_stream_id
,
3916 stream
->next_net_seq_num
- 1);
3918 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3920 pthread_mutex_unlock(&stream
->lock
);
3924 pthread_mutex_unlock(&stream
->lock
);
3928 unsigned int is_data_inflight
= 0;
3930 /* Send init command for data pending. */
3931 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3932 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3933 relayd
->relayd_session_id
, &is_data_inflight
);
3934 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3936 goto data_not_pending
;
3938 if (is_data_inflight
) {
3944 * Finding _no_ node in the hash table and no inflight data means that the
3945 * stream(s) have been removed thus data is guaranteed to be available for
3946 * analysis from the trace files.
3950 /* Data is available to be read by a viewer. */
3951 pthread_mutex_unlock(&consumer_data
.lock
);
3956 /* Data is still being extracted from buffers. */
3957 pthread_mutex_unlock(&consumer_data
.lock
);
3963 * Send a ret code status message to the sessiond daemon.
3965 * Return the sendmsg() return value.
3967 int consumer_send_status_msg(int sock
, int ret_code
)
3969 struct lttcomm_consumer_status_msg msg
;
3971 memset(&msg
, 0, sizeof(msg
));
3972 msg
.ret_code
= ret_code
;
3974 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3978 * Send a channel status message to the sessiond daemon.
3980 * Return the sendmsg() return value.
3982 int consumer_send_status_channel(int sock
,
3983 struct lttng_consumer_channel
*channel
)
3985 struct lttcomm_consumer_status_channel msg
;
3989 memset(&msg
, 0, sizeof(msg
));
3991 msg
.ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
3993 msg
.ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3994 msg
.key
= channel
->key
;
3995 msg
.stream_count
= channel
->streams
.count
;
3998 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
4001 unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos
,
4002 unsigned long produced_pos
, uint64_t nb_packets_per_stream
,
4003 uint64_t max_sb_size
)
4005 unsigned long start_pos
;
4007 if (!nb_packets_per_stream
) {
4008 return consumed_pos
; /* Grab everything */
4010 start_pos
= produced_pos
- offset_align_floor(produced_pos
, max_sb_size
);
4011 start_pos
-= max_sb_size
* nb_packets_per_stream
;
4012 if ((long) (start_pos
- consumed_pos
) < 0) {
4013 return consumed_pos
; /* Grab everything */
4019 int consumer_flush_buffer(struct lttng_consumer_stream
*stream
, int producer_active
)
4023 switch (consumer_data
.type
) {
4024 case LTTNG_CONSUMER_KERNEL
:
4025 ret
= kernctl_buffer_flush(stream
->wait_fd
);
4027 ERR("Failed to flush kernel stream");
4031 case LTTNG_CONSUMER32_UST
:
4032 case LTTNG_CONSUMER64_UST
:
4033 lttng_ustctl_flush_buffer(stream
, producer_active
);
4036 ERR("Unknown consumer_data type");
4045 * Sample the rotate position for all the streams of a channel.
4047 * Returns 0 on success, < 0 on error
4049 int lttng_consumer_rotate_channel(uint64_t key
, char *path
,
4050 uint64_t relayd_id
, uint32_t metadata
, uint64_t new_chunk_id
,
4051 struct lttng_consumer_local_data
*ctx
)
4054 struct lttng_consumer_channel
*channel
;
4055 struct lttng_consumer_stream
*stream
;
4056 struct lttng_ht_iter iter
;
4057 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
4059 DBG("Consumer sample rotate position for channel %" PRIu64
, key
);
4063 channel
= consumer_find_channel(key
);
4065 ERR("No channel found for key %" PRIu64
, key
);
4069 pthread_mutex_lock(&channel
->lock
);
4070 channel
->current_chunk_id
= new_chunk_id
;
4071 snprintf(channel
->pathname
, PATH_MAX
, "%s", path
);
4072 ret
= utils_mkdir_recursive(channel
->pathname
, S_IRWXU
| S_IRWXG
,
4073 channel
->uid
, channel
->gid
);
4075 ERR("Trace directory creation error");
4077 pthread_mutex_unlock(&channel
->lock
);
4080 pthread_mutex_unlock(&channel
->lock
);
4082 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4083 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4084 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4085 stream
, node_channel_id
.node
) {
4086 uint64_t consumed_pos
;
4088 health_code_update();
4091 * Lock stream because we are about to change its state.
4093 pthread_mutex_lock(&stream
->lock
);
4095 memcpy(stream
->channel_ro_pathname
, channel
->pathname
, PATH_MAX
);
4096 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4098 ERR("Taking snapshot positions");
4102 ret
= lttng_consumer_get_produced_snapshot(stream
,
4103 &stream
->rotate_position
);
4105 ERR("Produced snapshot position");
4108 fprintf(stderr
, "Stream %lu should rotate after %lu to %s\n",
4109 stream
->key
, stream
->rotate_position
,
4111 lttng_consumer_get_consumed_snapshot(stream
,
4113 fprintf(stderr
, "consumed %lu\n", consumed_pos
);
4114 if (consumed_pos
== stream
->rotate_position
) {
4115 stream
->rotate_ready
= 1;
4116 fprintf(stderr
, "Stream %lu ready to rotate to %s\n",
4117 stream
->key
, channel
->pathname
);
4119 fprintf(stderr
, "before increasinc nr_pending: %lu\n", channel
->nr_stream_rotate_pending
);
4120 channel
->nr_stream_rotate_pending
++;
4122 ret
= consumer_flush_buffer(stream
, 1);
4124 ERR("Failed to flush stream");
4128 pthread_mutex_unlock(&stream
->lock
);
4135 pthread_mutex_unlock(&stream
->lock
);
4142 * Check if a stream is ready to be rotated after extracting it.
4144 * Return 1 if it is ready for rotation, 0 if it is not, a negative value on
4145 * error. Stream lock must be held.
4147 int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream
*stream
)
4150 unsigned long consumed_pos
;
4152 if (!stream
->rotate_position
&& !stream
->rotate_ready
) {
4157 if (stream
->rotate_ready
) {
4158 fprintf(stderr
, "Rotate position reached for stream %lu\n",
4165 * If we don't have the rotate_ready flag, check the consumed position
4166 * to determine if we need to rotate.
4168 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4170 ERR("Taking kernel snapshot positions");
4174 ret
= lttng_consumer_get_consumed_snapshot(stream
, &consumed_pos
);
4176 ERR("Consumed kernel snapshot position");
4180 fprintf(stderr
, "packet %lu, pos %lu\n", stream
->key
, consumed_pos
);
4181 /* Rotate position not reached yet. */
4182 if (consumed_pos
< stream
->rotate_position
) {
4186 fprintf(stderr
, "Rotate position %lu (expected %lu) reached for stream %lu\n",
4187 consumed_pos
, stream
->rotate_position
,
4196 * Reset the state for a stream after a rotation occurred.
4198 void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream
*stream
)
4200 stream
->rotate_position
= 0;
4201 stream
->rotate_ready
= 0;
4202 stream
->rotated
= 1;
4206 * Perform the rotation a local stream file.
4208 int rotate_local_stream(struct lttng_consumer_local_data
*ctx
,
4209 struct lttng_consumer_stream
*stream
)
4213 ret
= close(stream
->out_fd
);
4215 PERROR("Closing tracefile");
4219 fprintf(stderr
, "Rotating stream %lu to %s/%s\n", stream
->key
,
4220 stream
->channel_ro_pathname
, stream
->name
);
4221 ret
= utils_create_stream_file(stream
->channel_ro_pathname
, stream
->name
,
4222 stream
->channel_ro_tracefile_size
, stream
->tracefile_count_current
,
4223 stream
->uid
, stream
->gid
, NULL
);
4227 stream
->out_fd
= ret
;
4228 stream
->tracefile_size_current
= 0;
4230 if (!stream
->metadata_flag
) {
4231 struct lttng_index_file
*index_file
;
4233 lttng_index_file_put(stream
->index_file
);
4235 index_file
= lttng_index_file_create(stream
->channel_ro_pathname
,
4236 stream
->name
, stream
->uid
, stream
->gid
,
4237 stream
->channel_ro_tracefile_size
,
4238 stream
->tracefile_count_current
,
4239 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
4243 stream
->index_file
= index_file
;
4244 stream
->out_fd_offset
= 0;
4258 * Perform the rotation a stream file on the relay.
4260 int rotate_relay_stream(struct lttng_consumer_local_data
*ctx
,
4261 struct lttng_consumer_stream
*stream
)
4264 struct consumer_relayd_sock_pair
*relayd
;
4266 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
4268 ERR("Failed to find relayd");
4273 /* FIXME: chan_ro ? */
4274 ret
= relayd_rotate_stream(&relayd
->control_sock
,
4275 stream
->relayd_stream_id
, stream
->channel_ro_pathname
,
4276 stream
->chan
->current_chunk_id
,
4277 stream
->last_sequence_number
);
4284 * Performs the stream rotation for the rotate session feature if needed.
4285 * It must be called with the stream lock held.
4287 * Return 0 on success, a negative number of error.
4289 int lttng_consumer_rotate_stream(struct lttng_consumer_local_data
*ctx
,
4290 struct lttng_consumer_stream
*stream
)
4294 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
4295 ret
= rotate_relay_stream(ctx
, stream
);
4297 ret
= rotate_local_stream(ctx
, stream
);
4303 if (stream
->metadata_flag
) {
4304 switch (consumer_data
.type
) {
4305 case LTTNG_CONSUMER_KERNEL
:
4307 * Reset the position of what has been read from the metadata
4308 * cache to 0 so we can dump it again.
4310 ret
= kernctl_metadata_cache_dump(stream
->wait_fd
);
4312 ERR("Failed to dump the kernel metadata cache after rotation");
4316 case LTTNG_CONSUMER32_UST
:
4317 case LTTNG_CONSUMER64_UST
:
4319 * Reset the position pushed from the metadata cache so it
4320 * will write from the beginning on the next push.
4322 stream
->ust_metadata_pushed
= 0;
4325 ERR("Unknown consumer_data type");
4329 lttng_consumer_reset_stream_rotate_state(stream
);
4338 * Rotate all the ready streams.
4340 * This is especially important for low throughput streams that have already
4341 * been consumed, we cannot wait for their next packet to perform the
4344 * Returns 0 on success, < 0 on error
4346 int lttng_consumer_rotate_ready_streams(uint64_t key
,
4347 struct lttng_consumer_local_data
*ctx
)
4350 struct lttng_consumer_channel
*channel
;
4351 struct lttng_consumer_stream
*stream
;
4352 struct lttng_pipe
*stream_pipe
;
4353 struct lttng_ht_iter iter
;
4354 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
4358 channel
= consumer_find_channel(key
);
4360 ERR("No channel found for key %" PRIu64
, key
);
4365 if (channel
->metadata_stream
) {
4366 fprintf(stderr
, "M\n");
4367 stream_pipe
= ctx
->consumer_metadata_rotate_pipe
;
4369 fprintf(stderr
, "D\n");
4370 stream_pipe
= ctx
->consumer_data_rotate_pipe
;
4373 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4374 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4375 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4376 stream
, node_channel_id
.node
) {
4377 health_code_update();
4379 if (stream
->rotate_ready
== 0) {
4382 fprintf(stderr
, "send stream %lu on wakeup pipe\n", stream
->key
);
4383 ret
= lttng_pipe_write(stream_pipe
, &stream
, sizeof(stream
));
4385 ERR("Failed to wakeup consumer rotate pipe");
4388 fprintf(stderr
, "done sending stream %lu on wakeup pipe\n", stream
->key
);
4399 int rotate_rename_local(char *current_path
, char *new_path
,
4400 uid_t uid
, gid_t gid
)
4404 ret
= utils_mkdir_recursive(new_path
, S_IRWXU
| S_IRWXG
,
4407 ERR("Create directory on rotate");
4411 ret
= rename(current_path
, new_path
);
4413 * If a domain has not yet created its channel, the domain-specific
4414 * folder might not exist, but this is not an error.
4416 if (ret
< 0 && errno
!= ENOENT
) {
4417 PERROR("Rename completed rotation chunk");
4428 int rotate_rename_relay(char *current_path
, char *new_path
, uint64_t relayd_id
)
4431 struct consumer_relayd_sock_pair
*relayd
;
4433 relayd
= consumer_find_relayd(relayd_id
);
4435 ERR("Failed to find relayd");
4440 ret
= relayd_rotate_rename(&relayd
->control_sock
, current_path
, new_path
);
4447 int lttng_consumer_rotate_rename(char *current_path
, char *new_path
,
4448 uid_t uid
, gid_t gid
, uint64_t relayd_id
)
4450 if (relayd_id
!= (uint64_t) -1ULL) {
4451 return rotate_rename_relay(current_path
, new_path
, relayd_id
);
4453 return rotate_rename_local(current_path
, new_path
, uid
, gid
);
4457 int lttng_consumer_rotate_pending_relay(uint64_t session_id
,
4458 uint64_t relayd_id
, uint64_t chunk_id
)
4461 struct consumer_relayd_sock_pair
*relayd
;
4463 relayd
= consumer_find_relayd(relayd_id
);
4465 ERR("Failed to find relayd");
4470 ret
= relayd_rotate_pending(&relayd
->control_sock
, chunk_id
);