2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@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,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
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.
23 #include <lttng/ust-ctl.h>
29 #include <sys/socket.h>
31 #include <sys/types.h>
34 #include <urcu/list.h>
38 #include <bin/lttng-consumerd/health-consumerd.h>
39 #include <common/common.h>
40 #include <common/sessiond-comm/sessiond-comm.h>
41 #include <common/relayd/relayd.h>
42 #include <common/compat/fcntl.h>
43 #include <common/compat/endian.h>
44 #include <common/consumer/consumer-metadata-cache.h>
45 #include <common/consumer/consumer-stream.h>
46 #include <common/consumer/consumer-timer.h>
47 #include <common/utils.h>
48 #include <common/index/index.h>
50 #include "ust-consumer.h"
52 #define INT_MAX_STR_LEN 12 /* includes \0 */
54 extern struct lttng_consumer_global_data consumer_data
;
55 extern int consumer_poll_timeout
;
58 * Free channel object and all streams associated with it. This MUST be used
59 * only and only if the channel has _NEVER_ been added to the global channel
62 static void destroy_channel(struct lttng_consumer_channel
*channel
)
64 struct lttng_consumer_stream
*stream
, *stmp
;
68 DBG("UST consumer cleaning stream list");
70 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
75 cds_list_del(&stream
->send_node
);
76 ustctl_destroy_stream(stream
->ustream
);
77 lttng_trace_chunk_put(stream
->trace_chunk
);
82 * If a channel is available meaning that was created before the streams
86 lttng_ustconsumer_del_channel(channel
);
87 lttng_ustconsumer_free_channel(channel
);
93 * Add channel to internal consumer state.
95 * Returns 0 on success or else a negative value.
97 static int add_channel(struct lttng_consumer_channel
*channel
,
98 struct lttng_consumer_local_data
*ctx
)
105 if (ctx
->on_recv_channel
!= NULL
) {
106 ret
= ctx
->on_recv_channel(channel
);
108 ret
= consumer_add_channel(channel
, ctx
);
109 } else if (ret
< 0) {
110 /* Most likely an ENOMEM. */
111 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
115 ret
= consumer_add_channel(channel
, ctx
);
118 DBG("UST consumer channel added (key: %" PRIu64
")", channel
->key
);
125 * Allocate and return a consumer channel object.
127 static struct lttng_consumer_channel
*allocate_channel(uint64_t session_id
,
128 const uint64_t *chunk_id
, const char *pathname
, const char *name
,
129 uint64_t relayd_id
, uint64_t key
, enum lttng_event_output output
,
130 uint64_t tracefile_size
, uint64_t tracefile_count
,
131 uint64_t session_id_per_pid
, unsigned int monitor
,
132 unsigned int live_timer_interval
,
133 const char *root_shm_path
, const char *shm_path
)
138 return consumer_allocate_channel(key
, session_id
, chunk_id
, pathname
,
139 name
, relayd_id
, output
, tracefile_size
,
140 tracefile_count
, session_id_per_pid
, monitor
,
141 live_timer_interval
, root_shm_path
, shm_path
);
145 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
146 * error value if applicable is set in it else it is kept untouched.
148 * Return NULL on error else the newly allocated stream object.
150 static struct lttng_consumer_stream
*allocate_stream(int cpu
, int key
,
151 struct lttng_consumer_channel
*channel
,
152 struct lttng_consumer_local_data
*ctx
, int *_alloc_ret
)
155 struct lttng_consumer_stream
*stream
= NULL
;
160 stream
= consumer_allocate_stream(channel
->key
,
165 channel
->trace_chunk
,
170 if (stream
== NULL
) {
174 * We could not find the channel. Can happen if cpu hotplug
175 * happens while tearing down.
177 DBG3("Could not find channel");
182 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
188 consumer_stream_update_channel_attributes(stream
, channel
);
189 stream
->chan
= channel
;
193 *_alloc_ret
= alloc_ret
;
199 * Send the given stream pointer to the corresponding thread.
201 * Returns 0 on success else a negative value.
203 static int send_stream_to_thread(struct lttng_consumer_stream
*stream
,
204 struct lttng_consumer_local_data
*ctx
)
207 struct lttng_pipe
*stream_pipe
;
209 /* Get the right pipe where the stream will be sent. */
210 if (stream
->metadata_flag
) {
211 consumer_add_metadata_stream(stream
);
212 stream_pipe
= ctx
->consumer_metadata_pipe
;
214 consumer_add_data_stream(stream
);
215 stream_pipe
= ctx
->consumer_data_pipe
;
219 * From this point on, the stream's ownership has been moved away from
220 * the channel and it becomes globally visible. Hence, remove it from
221 * the local stream list to prevent the stream from being both local and
224 stream
->globally_visible
= 1;
225 cds_list_del(&stream
->send_node
);
227 ret
= lttng_pipe_write(stream_pipe
, &stream
, sizeof(stream
));
229 ERR("Consumer write %s stream to pipe %d",
230 stream
->metadata_flag
? "metadata" : "data",
231 lttng_pipe_get_writefd(stream_pipe
));
232 if (stream
->metadata_flag
) {
233 consumer_del_stream_for_metadata(stream
);
235 consumer_del_stream_for_data(stream
);
245 int get_stream_shm_path(char *stream_shm_path
, const char *shm_path
, int cpu
)
247 char cpu_nr
[INT_MAX_STR_LEN
]; /* int max len */
250 strncpy(stream_shm_path
, shm_path
, PATH_MAX
);
251 stream_shm_path
[PATH_MAX
- 1] = '\0';
252 ret
= snprintf(cpu_nr
, INT_MAX_STR_LEN
, "%i", cpu
);
257 strncat(stream_shm_path
, cpu_nr
,
258 PATH_MAX
- strlen(stream_shm_path
) - 1);
265 * Create streams for the given channel using liblttng-ust-ctl.
266 * The channel lock must be acquired by the caller.
268 * Return 0 on success else a negative value.
270 static int create_ust_streams(struct lttng_consumer_channel
*channel
,
271 struct lttng_consumer_local_data
*ctx
)
274 struct ustctl_consumer_stream
*ustream
;
275 struct lttng_consumer_stream
*stream
;
276 pthread_mutex_t
*current_stream_lock
= NULL
;
282 * While a stream is available from ustctl. When NULL is returned, we've
283 * reached the end of the possible stream for the channel.
285 while ((ustream
= ustctl_create_stream(channel
->uchan
, cpu
))) {
287 int ust_metadata_pipe
[2];
289 health_code_update();
291 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& channel
->monitor
) {
292 ret
= utils_create_pipe_cloexec_nonblock(ust_metadata_pipe
);
294 ERR("Create ust metadata poll pipe");
297 wait_fd
= ust_metadata_pipe
[0];
299 wait_fd
= ustctl_stream_get_wait_fd(ustream
);
302 /* Allocate consumer stream object. */
303 stream
= allocate_stream(cpu
, wait_fd
, channel
, ctx
, &ret
);
307 stream
->ustream
= ustream
;
309 * Store it so we can save multiple function calls afterwards since
310 * this value is used heavily in the stream threads. This is UST
311 * specific so this is why it's done after allocation.
313 stream
->wait_fd
= wait_fd
;
316 * Increment channel refcount since the channel reference has now been
317 * assigned in the allocation process above.
319 if (stream
->chan
->monitor
) {
320 uatomic_inc(&stream
->chan
->refcount
);
323 pthread_mutex_lock(&stream
->lock
);
324 current_stream_lock
= &stream
->lock
;
326 * Order is important this is why a list is used. On error, the caller
327 * should clean this list.
329 cds_list_add_tail(&stream
->send_node
, &channel
->streams
.head
);
331 ret
= ustctl_get_max_subbuf_size(stream
->ustream
,
332 &stream
->max_sb_size
);
334 ERR("ustctl_get_max_subbuf_size failed for stream %s",
339 /* Do actions once stream has been received. */
340 if (ctx
->on_recv_stream
) {
341 ret
= ctx
->on_recv_stream(stream
);
347 DBG("UST consumer add stream %s (key: %" PRIu64
") with relayd id %" PRIu64
,
348 stream
->name
, stream
->key
, stream
->relayd_stream_id
);
350 /* Set next CPU stream. */
351 channel
->streams
.count
= ++cpu
;
353 /* Keep stream reference when creating metadata. */
354 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
355 channel
->metadata_stream
= stream
;
356 if (channel
->monitor
) {
357 /* Set metadata poll pipe if we created one */
358 memcpy(stream
->ust_metadata_poll_pipe
,
360 sizeof(ust_metadata_pipe
));
363 pthread_mutex_unlock(&stream
->lock
);
364 current_stream_lock
= NULL
;
371 if (current_stream_lock
) {
372 pthread_mutex_unlock(current_stream_lock
);
378 * create_posix_shm is never called concurrently within a process.
381 int create_posix_shm(void)
383 char tmp_name
[NAME_MAX
];
386 ret
= snprintf(tmp_name
, NAME_MAX
, "/ust-shm-consumer-%d", getpid());
392 * Allocate shm, and immediately unlink its shm oject, keeping
393 * only the file descriptor as a reference to the object.
394 * We specifically do _not_ use the / at the beginning of the
395 * pathname so that some OS implementations can keep it local to
396 * the process (POSIX leaves this implementation-defined).
398 shmfd
= shm_open(tmp_name
, O_CREAT
| O_EXCL
| O_RDWR
, 0700);
403 ret
= shm_unlink(tmp_name
);
404 if (ret
< 0 && errno
!= ENOENT
) {
405 PERROR("shm_unlink");
406 goto error_shm_release
;
419 static int open_ust_stream_fd(struct lttng_consumer_channel
*channel
, int cpu
,
420 const struct lttng_credentials
*session_credentials
)
422 char shm_path
[PATH_MAX
];
425 if (!channel
->shm_path
[0]) {
426 return create_posix_shm();
428 ret
= get_stream_shm_path(shm_path
, channel
->shm_path
, cpu
);
432 return run_as_open(shm_path
,
433 O_RDWR
| O_CREAT
| O_EXCL
, S_IRUSR
| S_IWUSR
,
434 session_credentials
->uid
, session_credentials
->gid
);
441 * Create an UST channel with the given attributes and send it to the session
442 * daemon using the ust ctl API.
444 * Return 0 on success or else a negative value.
446 static int create_ust_channel(struct lttng_consumer_channel
*channel
,
447 struct ustctl_consumer_channel_attr
*attr
,
448 struct ustctl_consumer_channel
**ust_chanp
)
450 int ret
, nr_stream_fds
, i
, j
;
452 struct ustctl_consumer_channel
*ust_channel
;
457 assert(channel
->buffer_credentials
.is_set
);
459 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
460 "subbuf_size: %" PRIu64
", num_subbuf: %" PRIu64
", "
461 "switch_timer_interval: %u, read_timer_interval: %u, "
462 "output: %d, type: %d", attr
->overwrite
, attr
->subbuf_size
,
463 attr
->num_subbuf
, attr
->switch_timer_interval
,
464 attr
->read_timer_interval
, attr
->output
, attr
->type
);
466 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
)
469 nr_stream_fds
= ustctl_get_nr_stream_per_channel();
470 stream_fds
= zmalloc(nr_stream_fds
* sizeof(*stream_fds
));
475 for (i
= 0; i
< nr_stream_fds
; i
++) {
476 stream_fds
[i
] = open_ust_stream_fd(channel
, i
,
477 &channel
->buffer_credentials
.value
);
478 if (stream_fds
[i
] < 0) {
483 ust_channel
= ustctl_create_channel(attr
, stream_fds
, nr_stream_fds
);
488 channel
->nr_stream_fds
= nr_stream_fds
;
489 channel
->stream_fds
= stream_fds
;
490 *ust_chanp
= ust_channel
;
496 for (j
= i
- 1; j
>= 0; j
--) {
499 closeret
= close(stream_fds
[j
]);
503 if (channel
->shm_path
[0]) {
504 char shm_path
[PATH_MAX
];
506 closeret
= get_stream_shm_path(shm_path
,
507 channel
->shm_path
, j
);
509 ERR("Cannot get stream shm path");
511 closeret
= run_as_unlink(shm_path
,
512 channel
->buffer_credentials
.value
.uid
,
513 channel
->buffer_credentials
.value
.gid
);
515 PERROR("unlink %s", shm_path
);
519 /* Try to rmdir all directories under shm_path root. */
520 if (channel
->root_shm_path
[0]) {
521 (void) run_as_rmdir_recursive(channel
->root_shm_path
,
522 channel
->buffer_credentials
.value
.uid
,
523 channel
->buffer_credentials
.value
.gid
,
524 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG
);
532 * Send a single given stream to the session daemon using the sock.
534 * Return 0 on success else a negative value.
536 static int send_sessiond_stream(int sock
, struct lttng_consumer_stream
*stream
)
543 DBG("UST consumer sending stream %" PRIu64
" to sessiond", stream
->key
);
545 /* Send stream to session daemon. */
546 ret
= ustctl_send_stream_to_sessiond(sock
, stream
->ustream
);
556 * Send channel to sessiond and relayd if applicable.
558 * Return 0 on success or else a negative value.
560 static int send_channel_to_sessiond_and_relayd(int sock
,
561 struct lttng_consumer_channel
*channel
,
562 struct lttng_consumer_local_data
*ctx
, int *relayd_error
)
564 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
565 struct lttng_consumer_stream
*stream
;
566 uint64_t net_seq_idx
= -1ULL;
572 DBG("UST consumer sending channel %s to sessiond", channel
->name
);
574 if (channel
->relayd_id
!= (uint64_t) -1ULL) {
575 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
577 health_code_update();
579 /* Try to send the stream to the relayd if one is available. */
580 DBG("Sending stream %" PRIu64
" of channel \"%s\" to relayd",
581 stream
->key
, channel
->name
);
582 ret
= consumer_send_relayd_stream(stream
, stream
->chan
->pathname
);
585 * Flag that the relayd was the problem here probably due to a
586 * communicaton error on the socket.
591 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
593 if (net_seq_idx
== -1ULL) {
594 net_seq_idx
= stream
->net_seq_idx
;
599 /* Inform sessiond that we are about to send channel and streams. */
600 ret
= consumer_send_status_msg(sock
, ret_code
);
601 if (ret
< 0 || ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
603 * Either the session daemon is not responding or the relayd died so we
609 /* Send channel to sessiond. */
610 ret
= ustctl_send_channel_to_sessiond(sock
, channel
->uchan
);
615 ret
= ustctl_channel_close_wakeup_fd(channel
->uchan
);
620 /* The channel was sent successfully to the sessiond at this point. */
621 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
623 health_code_update();
625 /* Send stream to session daemon. */
626 ret
= send_sessiond_stream(sock
, stream
);
632 /* Tell sessiond there is no more stream. */
633 ret
= ustctl_send_stream_to_sessiond(sock
, NULL
);
638 DBG("UST consumer NULL stream sent to sessiond");
643 if (ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
650 * Creates a channel and streams and add the channel it to the channel internal
651 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
654 * Return 0 on success or else, a negative value is returned and the channel
655 * MUST be destroyed by consumer_del_channel().
657 static int ask_channel(struct lttng_consumer_local_data
*ctx
,
658 struct lttng_consumer_channel
*channel
,
659 struct ustctl_consumer_channel_attr
*attr
)
668 * This value is still used by the kernel consumer since for the kernel,
669 * the stream ownership is not IN the consumer so we need to have the
670 * number of left stream that needs to be initialized so we can know when
671 * to delete the channel (see consumer.c).
673 * As for the user space tracer now, the consumer creates and sends the
674 * stream to the session daemon which only sends them to the application
675 * once every stream of a channel is received making this value useless
676 * because we they will be added to the poll thread before the application
677 * receives them. This ensures that a stream can not hang up during
678 * initilization of a channel.
680 channel
->nb_init_stream_left
= 0;
682 /* The reply msg status is handled in the following call. */
683 ret
= create_ust_channel(channel
, attr
, &channel
->uchan
);
688 channel
->wait_fd
= ustctl_channel_get_wait_fd(channel
->uchan
);
691 * For the snapshots (no monitor), we create the metadata streams
692 * on demand, not during the channel creation.
694 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& !channel
->monitor
) {
699 /* Open all streams for this channel. */
700 pthread_mutex_lock(&channel
->lock
);
701 ret
= create_ust_streams(channel
, ctx
);
702 pthread_mutex_unlock(&channel
->lock
);
712 * Send all stream of a channel to the right thread handling it.
714 * On error, return a negative value else 0 on success.
716 static int send_streams_to_thread(struct lttng_consumer_channel
*channel
,
717 struct lttng_consumer_local_data
*ctx
)
720 struct lttng_consumer_stream
*stream
, *stmp
;
725 /* Send streams to the corresponding thread. */
726 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
729 health_code_update();
731 /* Sending the stream to the thread. */
732 ret
= send_stream_to_thread(stream
, ctx
);
735 * If we are unable to send the stream to the thread, there is
736 * a big problem so just stop everything.
747 * Flush channel's streams using the given key to retrieve the channel.
749 * Return 0 on success else an LTTng error code.
751 static int flush_channel(uint64_t chan_key
)
754 struct lttng_consumer_channel
*channel
;
755 struct lttng_consumer_stream
*stream
;
757 struct lttng_ht_iter iter
;
759 DBG("UST consumer flush channel key %" PRIu64
, chan_key
);
762 channel
= consumer_find_channel(chan_key
);
764 ERR("UST consumer flush channel %" PRIu64
" not found", chan_key
);
765 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
769 ht
= consumer_data
.stream_per_chan_id_ht
;
771 /* For each stream of the channel id, flush it. */
772 cds_lfht_for_each_entry_duplicate(ht
->ht
,
773 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
774 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
776 health_code_update();
778 pthread_mutex_lock(&stream
->lock
);
781 * Protect against concurrent teardown of a stream.
783 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
787 if (!stream
->quiescent
) {
788 ustctl_flush_buffer(stream
->ustream
, 0);
789 stream
->quiescent
= true;
792 pthread_mutex_unlock(&stream
->lock
);
800 * Clear quiescent state from channel's streams using the given key to
801 * retrieve the channel.
803 * Return 0 on success else an LTTng error code.
805 static int clear_quiescent_channel(uint64_t chan_key
)
808 struct lttng_consumer_channel
*channel
;
809 struct lttng_consumer_stream
*stream
;
811 struct lttng_ht_iter iter
;
813 DBG("UST consumer clear quiescent channel key %" PRIu64
, chan_key
);
816 channel
= consumer_find_channel(chan_key
);
818 ERR("UST consumer clear quiescent channel %" PRIu64
" not found", chan_key
);
819 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
823 ht
= consumer_data
.stream_per_chan_id_ht
;
825 /* For each stream of the channel id, clear quiescent state. */
826 cds_lfht_for_each_entry_duplicate(ht
->ht
,
827 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
828 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
830 health_code_update();
832 pthread_mutex_lock(&stream
->lock
);
833 stream
->quiescent
= false;
834 pthread_mutex_unlock(&stream
->lock
);
842 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
844 * Return 0 on success else an LTTng error code.
846 static int close_metadata(uint64_t chan_key
)
849 struct lttng_consumer_channel
*channel
;
850 unsigned int channel_monitor
;
852 DBG("UST consumer close metadata key %" PRIu64
, chan_key
);
854 channel
= consumer_find_channel(chan_key
);
857 * This is possible if the metadata thread has issue a delete because
858 * the endpoint point of the stream hung up. There is no way the
859 * session daemon can know about it thus use a DBG instead of an actual
862 DBG("UST consumer close metadata %" PRIu64
" not found", chan_key
);
863 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
867 pthread_mutex_lock(&consumer_data
.lock
);
868 pthread_mutex_lock(&channel
->lock
);
869 channel_monitor
= channel
->monitor
;
870 if (cds_lfht_is_node_deleted(&channel
->node
.node
)) {
874 lttng_ustconsumer_close_metadata(channel
);
875 pthread_mutex_unlock(&channel
->lock
);
876 pthread_mutex_unlock(&consumer_data
.lock
);
879 * The ownership of a metadata channel depends on the type of
880 * session to which it belongs. In effect, the monitor flag is checked
881 * to determine if this metadata channel is in "snapshot" mode or not.
883 * In the non-snapshot case, the metadata channel is created along with
884 * a single stream which will remain present until the metadata channel
885 * is destroyed (on the destruction of its session). In this case, the
886 * metadata stream in "monitored" by the metadata poll thread and holds
887 * the ownership of its channel.
889 * Closing the metadata will cause the metadata stream's "metadata poll
890 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
891 * thread which will teardown the metadata stream which, in return,
892 * deletes the metadata channel.
894 * In the snapshot case, the metadata stream is created and destroyed
895 * on every snapshot record. Since the channel doesn't have an owner
896 * other than the session daemon, it is safe to destroy it immediately
897 * on reception of the CLOSE_METADATA command.
899 if (!channel_monitor
) {
901 * The channel and consumer_data locks must be
902 * released before this call since consumer_del_channel
903 * re-acquires the channel and consumer_data locks to teardown
904 * the channel and queue its reclamation by the "call_rcu"
907 consumer_del_channel(channel
);
912 pthread_mutex_unlock(&channel
->lock
);
913 pthread_mutex_unlock(&consumer_data
.lock
);
919 * RCU read side lock MUST be acquired before calling this function.
921 * Return 0 on success else an LTTng error code.
923 static int setup_metadata(struct lttng_consumer_local_data
*ctx
, uint64_t key
)
926 struct lttng_consumer_channel
*metadata
;
928 DBG("UST consumer setup metadata key %" PRIu64
, key
);
930 metadata
= consumer_find_channel(key
);
932 ERR("UST consumer push metadata %" PRIu64
" not found", key
);
933 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
938 * In no monitor mode, the metadata channel has no stream(s) so skip the
939 * ownership transfer to the metadata thread.
941 if (!metadata
->monitor
) {
942 DBG("Metadata channel in no monitor");
948 * Send metadata stream to relayd if one available. Availability is
949 * known if the stream is still in the list of the channel.
951 if (cds_list_empty(&metadata
->streams
.head
)) {
952 ERR("Metadata channel key %" PRIu64
", no stream available.", key
);
953 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
954 goto error_no_stream
;
957 /* Send metadata stream to relayd if needed. */
958 if (metadata
->metadata_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
959 ret
= consumer_send_relayd_stream(metadata
->metadata_stream
,
962 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
965 ret
= consumer_send_relayd_streams_sent(
966 metadata
->metadata_stream
->net_seq_idx
);
968 ret
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
974 * Ownership of metadata stream is passed along. Freeing is handled by
977 ret
= send_streams_to_thread(metadata
, ctx
);
980 * If we are unable to send the stream to the thread, there is
981 * a big problem so just stop everything.
983 ret
= LTTCOMM_CONSUMERD_FATAL
;
984 goto send_streams_error
;
986 /* List MUST be empty after or else it could be reused. */
987 assert(cds_list_empty(&metadata
->streams
.head
));
994 * Delete metadata channel on error. At this point, the metadata stream can
995 * NOT be monitored by the metadata thread thus having the guarantee that
996 * the stream is still in the local stream list of the channel. This call
997 * will make sure to clean that list.
999 consumer_stream_destroy(metadata
->metadata_stream
, NULL
);
1000 cds_list_del(&metadata
->metadata_stream
->send_node
);
1001 metadata
->metadata_stream
= NULL
;
1009 * Snapshot the whole metadata.
1010 * RCU read-side lock must be held by the caller.
1012 * Returns 0 on success, < 0 on error
1014 static int snapshot_metadata(struct lttng_consumer_channel
*metadata_channel
,
1015 uint64_t key
, char *path
, uint64_t relayd_id
,
1016 struct lttng_consumer_local_data
*ctx
)
1019 struct lttng_consumer_stream
*metadata_stream
;
1024 DBG("UST consumer snapshot metadata with key %" PRIu64
" at path %s",
1029 assert(!metadata_channel
->monitor
);
1031 health_code_update();
1034 * Ask the sessiond if we have new metadata waiting and update the
1035 * consumer metadata cache.
1037 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 1);
1042 health_code_update();
1045 * The metadata stream is NOT created in no monitor mode when the channel
1046 * is created on a sessiond ask channel command.
1048 ret
= create_ust_streams(metadata_channel
, ctx
);
1053 metadata_stream
= metadata_channel
->metadata_stream
;
1054 assert(metadata_stream
);
1056 pthread_mutex_lock(&metadata_stream
->lock
);
1057 if (relayd_id
!= (uint64_t) -1ULL) {
1058 metadata_stream
->net_seq_idx
= relayd_id
;
1059 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
1061 ret
= consumer_stream_create_output_files(metadata_stream
,
1064 pthread_mutex_unlock(&metadata_stream
->lock
);
1070 health_code_update();
1072 ret
= lttng_consumer_read_subbuffer(metadata_stream
, ctx
);
1080 * Clean up the stream completly because the next snapshot will use a new
1083 consumer_stream_destroy(metadata_stream
, NULL
);
1084 cds_list_del(&metadata_stream
->send_node
);
1085 metadata_channel
->metadata_stream
= NULL
;
1093 int get_current_subbuf_addr(struct lttng_consumer_stream
*stream
,
1097 unsigned long mmap_offset
;
1098 const char *mmap_base
;
1100 mmap_base
= ustctl_get_mmap_base(stream
->ustream
);
1102 ERR("Failed to get mmap base for stream `%s`",
1108 ret
= ustctl_get_mmap_read_offset(stream
->ustream
, &mmap_offset
);
1110 ERR("Failed to get mmap offset for stream `%s`", stream
->name
);
1115 *addr
= mmap_base
+ mmap_offset
;
1122 * Take a snapshot of all the stream of a channel.
1123 * RCU read-side lock and the channel lock must be held by the caller.
1125 * Returns 0 on success, < 0 on error
1127 static int snapshot_channel(struct lttng_consumer_channel
*channel
,
1128 uint64_t key
, char *path
, uint64_t relayd_id
,
1129 uint64_t nb_packets_per_stream
,
1130 struct lttng_consumer_local_data
*ctx
)
1133 unsigned use_relayd
= 0;
1134 unsigned long consumed_pos
, produced_pos
;
1135 struct lttng_consumer_stream
*stream
;
1142 if (relayd_id
!= (uint64_t) -1ULL) {
1146 assert(!channel
->monitor
);
1147 DBG("UST consumer snapshot channel %" PRIu64
, key
);
1149 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
1150 health_code_update();
1152 /* Lock stream because we are about to change its state. */
1153 pthread_mutex_lock(&stream
->lock
);
1154 assert(channel
->trace_chunk
);
1155 if (!lttng_trace_chunk_get(channel
->trace_chunk
)) {
1157 * Can't happen barring an internal error as the channel
1158 * holds a reference to the trace chunk.
1160 ERR("Failed to acquire reference to channel's trace chunk");
1164 assert(!stream
->trace_chunk
);
1165 stream
->trace_chunk
= channel
->trace_chunk
;
1167 stream
->net_seq_idx
= relayd_id
;
1170 ret
= consumer_send_relayd_stream(stream
, path
);
1175 ret
= consumer_stream_create_output_files(stream
,
1180 DBG("UST consumer snapshot stream (%" PRIu64
")",
1185 * If tracing is active, we want to perform a "full" buffer flush.
1186 * Else, if quiescent, it has already been done by the prior stop.
1188 if (!stream
->quiescent
) {
1189 ustctl_flush_buffer(stream
->ustream
, 0);
1192 ret
= lttng_ustconsumer_take_snapshot(stream
);
1194 ERR("Taking UST snapshot");
1198 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
1200 ERR("Produced UST snapshot position");
1204 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1206 ERR("Consumerd UST snapshot position");
1211 * The original value is sent back if max stream size is larger than
1212 * the possible size of the snapshot. Also, we assume that the session
1213 * daemon should never send a maximum stream size that is lower than
1216 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
1217 produced_pos
, nb_packets_per_stream
,
1218 stream
->max_sb_size
);
1220 while ((long) (consumed_pos
- produced_pos
) < 0) {
1222 unsigned long len
, padded_len
;
1223 const char *subbuf_addr
;
1224 struct lttng_buffer_view subbuf_view
;
1226 health_code_update();
1228 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1230 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1232 if (ret
!= -EAGAIN
) {
1233 PERROR("ustctl_get_subbuf snapshot");
1234 goto error_close_stream
;
1236 DBG("UST consumer get subbuf failed. Skipping it.");
1237 consumed_pos
+= stream
->max_sb_size
;
1238 stream
->chan
->lost_packets
++;
1242 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1244 ERR("Snapshot ustctl_get_subbuf_size");
1245 goto error_put_subbuf
;
1248 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1250 ERR("Snapshot ustctl_get_padded_subbuf_size");
1251 goto error_put_subbuf
;
1254 ret
= get_current_subbuf_addr(stream
, &subbuf_addr
);
1256 goto error_put_subbuf
;
1259 subbuf_view
= lttng_buffer_view_init(
1260 subbuf_addr
, 0, padded_len
);
1261 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
,
1262 stream
, &subbuf_view
, padded_len
- len
,
1265 if (read_len
!= len
) {
1267 goto error_put_subbuf
;
1270 if (read_len
!= padded_len
) {
1272 goto error_put_subbuf
;
1276 ret
= ustctl_put_subbuf(stream
->ustream
);
1278 ERR("Snapshot ustctl_put_subbuf");
1279 goto error_close_stream
;
1281 consumed_pos
+= stream
->max_sb_size
;
1284 /* Simply close the stream so we can use it on the next snapshot. */
1285 consumer_stream_close(stream
);
1286 pthread_mutex_unlock(&stream
->lock
);
1293 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1294 ERR("Snapshot ustctl_put_subbuf");
1297 consumer_stream_close(stream
);
1299 pthread_mutex_unlock(&stream
->lock
);
1305 * Receive the metadata updates from the sessiond. Supports receiving
1306 * overlapping metadata, but is needs to always belong to a contiguous
1307 * range starting from 0.
1308 * Be careful about the locks held when calling this function: it needs
1309 * the metadata cache flush to concurrently progress in order to
1312 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1313 uint64_t len
, uint64_t version
,
1314 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
1316 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1319 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1321 metadata_str
= zmalloc(len
* sizeof(char));
1322 if (!metadata_str
) {
1323 PERROR("zmalloc metadata string");
1324 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1328 health_code_update();
1330 /* Receive metadata string. */
1331 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1333 /* Session daemon is dead so return gracefully. */
1338 health_code_update();
1340 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1341 ret
= consumer_metadata_cache_write(channel
, offset
, len
, version
,
1344 /* Unable to handle metadata. Notify session daemon. */
1345 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1347 * Skip metadata flush on write error since the offset and len might
1348 * not have been updated which could create an infinite loop below when
1349 * waiting for the metadata cache to be flushed.
1351 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1354 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1359 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1360 DBG("Waiting for metadata to be flushed");
1362 health_code_update();
1364 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1374 * Receive command from session daemon and process it.
1376 * Return 1 on success else a negative value or 0.
1378 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1379 int sock
, struct pollfd
*consumer_sockpoll
)
1382 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1383 struct lttcomm_consumer_msg msg
;
1384 struct lttng_consumer_channel
*channel
= NULL
;
1386 health_code_update();
1388 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1389 if (ret
!= sizeof(msg
)) {
1390 DBG("Consumer received unexpected message size %zd (expects %zu)",
1393 * The ret value might 0 meaning an orderly shutdown but this is ok
1394 * since the caller handles this.
1397 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1403 health_code_update();
1406 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1408 health_code_update();
1410 /* relayd needs RCU read-side lock */
1413 switch (msg
.cmd_type
) {
1414 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1416 /* Session daemon status message are handled in the following call. */
1417 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1418 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
1419 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
1420 msg
.u
.relayd_sock
.relayd_session_id
);
1423 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1425 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1426 struct consumer_relayd_sock_pair
*relayd
;
1428 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1430 /* Get relayd reference if exists. */
1431 relayd
= consumer_find_relayd(index
);
1432 if (relayd
== NULL
) {
1433 DBG("Unable to find relayd %" PRIu64
, index
);
1434 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1438 * Each relayd socket pair has a refcount of stream attached to it
1439 * which tells if the relayd is still active or not depending on the
1442 * This will set the destroy flag of the relayd object and destroy it
1443 * if the refcount reaches zero when called.
1445 * The destroy can happen either here or when a stream fd hangs up.
1448 consumer_flag_relayd_for_destroy(relayd
);
1451 goto end_msg_sessiond
;
1453 case LTTNG_CONSUMER_UPDATE_STREAM
:
1458 case LTTNG_CONSUMER_DATA_PENDING
:
1460 int ret
, is_data_pending
;
1461 uint64_t id
= msg
.u
.data_pending
.session_id
;
1463 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1465 is_data_pending
= consumer_data_pending(id
);
1467 /* Send back returned value to session daemon */
1468 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1469 sizeof(is_data_pending
));
1471 DBG("Error when sending the data pending ret code: %d", ret
);
1476 * No need to send back a status message since the data pending
1477 * returned value is the response.
1481 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1484 struct ustctl_consumer_channel_attr attr
;
1485 const uint64_t chunk_id
= msg
.u
.ask_channel
.chunk_id
.value
;
1486 const struct lttng_credentials buffer_credentials
= {
1487 .uid
= msg
.u
.ask_channel
.buffer_credentials
.uid
,
1488 .gid
= msg
.u
.ask_channel
.buffer_credentials
.gid
,
1491 /* Create a plain object and reserve a channel key. */
1492 channel
= allocate_channel(msg
.u
.ask_channel
.session_id
,
1493 msg
.u
.ask_channel
.chunk_id
.is_set
?
1495 msg
.u
.ask_channel
.pathname
,
1496 msg
.u
.ask_channel
.name
,
1497 msg
.u
.ask_channel
.relayd_id
,
1498 msg
.u
.ask_channel
.key
,
1499 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1500 msg
.u
.ask_channel
.tracefile_size
,
1501 msg
.u
.ask_channel
.tracefile_count
,
1502 msg
.u
.ask_channel
.session_id_per_pid
,
1503 msg
.u
.ask_channel
.monitor
,
1504 msg
.u
.ask_channel
.live_timer_interval
,
1505 msg
.u
.ask_channel
.root_shm_path
,
1506 msg
.u
.ask_channel
.shm_path
);
1508 goto end_channel_error
;
1511 LTTNG_OPTIONAL_SET(&channel
->buffer_credentials
,
1512 buffer_credentials
);
1515 * Assign UST application UID to the channel. This value is ignored for
1516 * per PID buffers. This is specific to UST thus setting this after the
1519 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1521 /* Build channel attributes from received message. */
1522 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1523 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1524 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1525 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1526 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1527 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1528 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1529 attr
.blocking_timeout
= msg
.u
.ask_channel
.blocking_timeout
;
1531 /* Match channel buffer type to the UST abi. */
1532 switch (msg
.u
.ask_channel
.output
) {
1533 case LTTNG_EVENT_MMAP
:
1535 attr
.output
= LTTNG_UST_MMAP
;
1539 /* Translate and save channel type. */
1540 switch (msg
.u
.ask_channel
.type
) {
1541 case LTTNG_UST_CHAN_PER_CPU
:
1542 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1543 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1545 * Set refcount to 1 for owner. Below, we will
1546 * pass ownership to the
1547 * consumer_thread_channel_poll() thread.
1549 channel
->refcount
= 1;
1551 case LTTNG_UST_CHAN_METADATA
:
1552 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1553 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1560 health_code_update();
1562 ret
= ask_channel(ctx
, channel
, &attr
);
1564 goto end_channel_error
;
1567 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1568 ret
= consumer_metadata_cache_allocate(channel
);
1570 ERR("Allocating metadata cache");
1571 goto end_channel_error
;
1573 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1574 attr
.switch_timer_interval
= 0;
1576 int monitor_start_ret
;
1578 consumer_timer_live_start(channel
,
1579 msg
.u
.ask_channel
.live_timer_interval
);
1580 monitor_start_ret
= consumer_timer_monitor_start(
1582 msg
.u
.ask_channel
.monitor_timer_interval
);
1583 if (monitor_start_ret
< 0) {
1584 ERR("Starting channel monitoring timer failed");
1585 goto end_channel_error
;
1589 health_code_update();
1592 * Add the channel to the internal state AFTER all streams were created
1593 * and successfully sent to session daemon. This way, all streams must
1594 * be ready before this channel is visible to the threads.
1595 * If add_channel succeeds, ownership of the channel is
1596 * passed to consumer_thread_channel_poll().
1598 ret
= add_channel(channel
, ctx
);
1600 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1601 if (channel
->switch_timer_enabled
== 1) {
1602 consumer_timer_switch_stop(channel
);
1604 consumer_metadata_cache_destroy(channel
);
1606 if (channel
->live_timer_enabled
== 1) {
1607 consumer_timer_live_stop(channel
);
1609 if (channel
->monitor_timer_enabled
== 1) {
1610 consumer_timer_monitor_stop(channel
);
1612 goto end_channel_error
;
1615 health_code_update();
1618 * Channel and streams are now created. Inform the session daemon that
1619 * everything went well and should wait to receive the channel and
1620 * streams with ustctl API.
1622 ret
= consumer_send_status_channel(sock
, channel
);
1625 * There is probably a problem on the socket.
1632 case LTTNG_CONSUMER_GET_CHANNEL
:
1634 int ret
, relayd_err
= 0;
1635 uint64_t key
= msg
.u
.get_channel
.key
;
1636 struct lttng_consumer_channel
*channel
;
1638 channel
= consumer_find_channel(key
);
1640 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1641 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1642 goto end_get_channel
;
1645 health_code_update();
1647 /* Send the channel to sessiond (and relayd, if applicable). */
1648 ret
= send_channel_to_sessiond_and_relayd(sock
, channel
, ctx
,
1653 * We were unable to send to the relayd the stream so avoid
1654 * sending back a fatal error to the thread since this is OK
1655 * and the consumer can continue its work. The above call
1656 * has sent the error status message to the sessiond.
1658 goto end_get_channel_nosignal
;
1661 * The communicaton was broken hence there is a bad state between
1662 * the consumer and sessiond so stop everything.
1664 goto error_get_channel_fatal
;
1667 health_code_update();
1670 * In no monitor mode, the streams ownership is kept inside the channel
1671 * so don't send them to the data thread.
1673 if (!channel
->monitor
) {
1674 goto end_get_channel
;
1677 ret
= send_streams_to_thread(channel
, ctx
);
1680 * If we are unable to send the stream to the thread, there is
1681 * a big problem so just stop everything.
1683 goto error_get_channel_fatal
;
1685 /* List MUST be empty after or else it could be reused. */
1686 assert(cds_list_empty(&channel
->streams
.head
));
1688 goto end_msg_sessiond
;
1689 error_get_channel_fatal
:
1691 end_get_channel_nosignal
:
1694 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1696 uint64_t key
= msg
.u
.destroy_channel
.key
;
1699 * Only called if streams have not been sent to stream
1700 * manager thread. However, channel has been sent to
1701 * channel manager thread.
1703 notify_thread_del_channel(ctx
, key
);
1704 goto end_msg_sessiond
;
1706 case LTTNG_CONSUMER_CLOSE_METADATA
:
1710 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1715 goto end_msg_sessiond
;
1717 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1721 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1726 goto end_msg_sessiond
;
1728 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL
:
1732 ret
= clear_quiescent_channel(
1733 msg
.u
.clear_quiescent_channel
.key
);
1738 goto end_msg_sessiond
;
1740 case LTTNG_CONSUMER_PUSH_METADATA
:
1743 uint64_t len
= msg
.u
.push_metadata
.len
;
1744 uint64_t key
= msg
.u
.push_metadata
.key
;
1745 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1746 uint64_t version
= msg
.u
.push_metadata
.version
;
1747 struct lttng_consumer_channel
*channel
;
1749 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1752 channel
= consumer_find_channel(key
);
1755 * This is possible if the metadata creation on the consumer side
1756 * is in flight vis-a-vis a concurrent push metadata from the
1757 * session daemon. Simply return that the channel failed and the
1758 * session daemon will handle that message correctly considering
1759 * that this race is acceptable thus the DBG() statement here.
1761 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1762 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1763 goto end_push_metadata_msg_sessiond
;
1766 health_code_update();
1770 * There is nothing to receive. We have simply
1771 * checked whether the channel can be found.
1773 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1774 goto end_push_metadata_msg_sessiond
;
1777 /* Tell session daemon we are ready to receive the metadata. */
1778 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1780 /* Somehow, the session daemon is not responding anymore. */
1781 goto error_push_metadata_fatal
;
1784 health_code_update();
1786 /* Wait for more data. */
1787 health_poll_entry();
1788 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1791 goto error_push_metadata_fatal
;
1794 health_code_update();
1796 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1797 len
, version
, channel
, 0, 1);
1799 /* error receiving from sessiond */
1800 goto error_push_metadata_fatal
;
1803 goto end_push_metadata_msg_sessiond
;
1805 end_push_metadata_msg_sessiond
:
1806 goto end_msg_sessiond
;
1807 error_push_metadata_fatal
:
1810 case LTTNG_CONSUMER_SETUP_METADATA
:
1814 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1818 goto end_msg_sessiond
;
1820 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1822 struct lttng_consumer_channel
*channel
;
1823 uint64_t key
= msg
.u
.snapshot_channel
.key
;
1825 channel
= consumer_find_channel(key
);
1827 DBG("UST snapshot channel not found for key %" PRIu64
, key
);
1828 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1830 if (msg
.u
.snapshot_channel
.metadata
) {
1831 ret
= snapshot_metadata(channel
, key
,
1832 msg
.u
.snapshot_channel
.pathname
,
1833 msg
.u
.snapshot_channel
.relayd_id
,
1836 ERR("Snapshot metadata failed");
1837 ret_code
= LTTCOMM_CONSUMERD_SNAPSHOT_FAILED
;
1840 ret
= snapshot_channel(channel
, key
,
1841 msg
.u
.snapshot_channel
.pathname
,
1842 msg
.u
.snapshot_channel
.relayd_id
,
1843 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
1846 ERR("Snapshot channel failed");
1847 ret_code
= LTTCOMM_CONSUMERD_SNAPSHOT_FAILED
;
1851 health_code_update();
1852 ret
= consumer_send_status_msg(sock
, ret_code
);
1854 /* Somehow, the session daemon is not responding anymore. */
1857 health_code_update();
1860 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
1863 uint64_t discarded_events
;
1864 struct lttng_ht_iter iter
;
1865 struct lttng_ht
*ht
;
1866 struct lttng_consumer_stream
*stream
;
1867 uint64_t id
= msg
.u
.discarded_events
.session_id
;
1868 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
1870 DBG("UST consumer discarded events command for session id %"
1873 pthread_mutex_lock(&consumer_data
.lock
);
1875 ht
= consumer_data
.stream_list_ht
;
1878 * We only need a reference to the channel, but they are not
1879 * directly indexed, so we just use the first matching stream
1880 * to extract the information we need, we default to 0 if not
1881 * found (no events are dropped if the channel is not yet in
1884 discarded_events
= 0;
1885 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1886 ht
->hash_fct(&id
, lttng_ht_seed
),
1888 &iter
.iter
, stream
, node_session_id
.node
) {
1889 if (stream
->chan
->key
== key
) {
1890 discarded_events
= stream
->chan
->discarded_events
;
1894 pthread_mutex_unlock(&consumer_data
.lock
);
1897 DBG("UST consumer discarded events command for session id %"
1898 PRIu64
", channel key %" PRIu64
, id
, key
);
1900 health_code_update();
1902 /* Send back returned value to session daemon */
1903 ret
= lttcomm_send_unix_sock(sock
, &discarded_events
, sizeof(discarded_events
));
1905 PERROR("send discarded events");
1911 case LTTNG_CONSUMER_LOST_PACKETS
:
1914 uint64_t lost_packets
;
1915 struct lttng_ht_iter iter
;
1916 struct lttng_ht
*ht
;
1917 struct lttng_consumer_stream
*stream
;
1918 uint64_t id
= msg
.u
.lost_packets
.session_id
;
1919 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
1921 DBG("UST consumer lost packets command for session id %"
1924 pthread_mutex_lock(&consumer_data
.lock
);
1926 ht
= consumer_data
.stream_list_ht
;
1929 * We only need a reference to the channel, but they are not
1930 * directly indexed, so we just use the first matching stream
1931 * to extract the information we need, we default to 0 if not
1932 * found (no packets lost if the channel is not yet in use).
1935 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1936 ht
->hash_fct(&id
, lttng_ht_seed
),
1938 &iter
.iter
, stream
, node_session_id
.node
) {
1939 if (stream
->chan
->key
== key
) {
1940 lost_packets
= stream
->chan
->lost_packets
;
1944 pthread_mutex_unlock(&consumer_data
.lock
);
1947 DBG("UST consumer lost packets command for session id %"
1948 PRIu64
", channel key %" PRIu64
, id
, key
);
1950 health_code_update();
1952 /* Send back returned value to session daemon */
1953 ret
= lttcomm_send_unix_sock(sock
, &lost_packets
,
1954 sizeof(lost_packets
));
1956 PERROR("send lost packets");
1962 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
1964 int channel_monitor_pipe
;
1966 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1967 /* Successfully received the command's type. */
1968 ret
= consumer_send_status_msg(sock
, ret_code
);
1973 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
1975 if (ret
!= sizeof(channel_monitor_pipe
)) {
1976 ERR("Failed to receive channel monitor pipe");
1980 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
1981 ret
= consumer_timer_thread_set_channel_monitor_pipe(
1982 channel_monitor_pipe
);
1986 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1987 /* Set the pipe as non-blocking. */
1988 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
1990 PERROR("fcntl get flags of the channel monitoring pipe");
1995 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
1996 flags
| O_NONBLOCK
);
1998 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
2001 DBG("Channel monitor pipe set as non-blocking");
2003 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
2005 goto end_msg_sessiond
;
2007 case LTTNG_CONSUMER_ROTATE_CHANNEL
:
2009 struct lttng_consumer_channel
*channel
;
2010 uint64_t key
= msg
.u
.rotate_channel
.key
;
2012 channel
= consumer_find_channel(key
);
2014 DBG("Channel %" PRIu64
" not found", key
);
2015 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2018 * Sample the rotate position of all the streams in
2021 ret
= lttng_consumer_rotate_channel(channel
, key
,
2022 msg
.u
.rotate_channel
.relayd_id
,
2023 msg
.u
.rotate_channel
.metadata
,
2026 ERR("Rotate channel failed");
2027 ret_code
= LTTCOMM_CONSUMERD_ROTATION_FAIL
;
2030 health_code_update();
2032 ret
= consumer_send_status_msg(sock
, ret_code
);
2034 /* Somehow, the session daemon is not responding anymore. */
2035 goto end_rotate_channel_nosignal
;
2039 * Rotate the streams that are ready right now.
2040 * FIXME: this is a second consecutive iteration over the
2041 * streams in a channel, there is probably a better way to
2042 * handle this, but it needs to be after the
2043 * consumer_send_status_msg() call.
2046 ret
= lttng_consumer_rotate_ready_streams(
2049 ERR("Rotate channel failed");
2053 end_rotate_channel_nosignal
:
2056 case LTTNG_CONSUMER_INIT
:
2058 ret_code
= lttng_consumer_init_command(ctx
,
2059 msg
.u
.init
.sessiond_uuid
);
2060 health_code_update();
2061 ret
= consumer_send_status_msg(sock
, ret_code
);
2063 /* Somehow, the session daemon is not responding anymore. */
2068 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK
:
2070 const struct lttng_credentials credentials
= {
2071 .uid
= msg
.u
.create_trace_chunk
.credentials
.value
.uid
,
2072 .gid
= msg
.u
.create_trace_chunk
.credentials
.value
.gid
,
2074 const bool is_local_trace
=
2075 !msg
.u
.create_trace_chunk
.relayd_id
.is_set
;
2076 const uint64_t relayd_id
=
2077 msg
.u
.create_trace_chunk
.relayd_id
.value
;
2078 const char *chunk_override_name
=
2079 *msg
.u
.create_trace_chunk
.override_name
?
2080 msg
.u
.create_trace_chunk
.override_name
:
2082 LTTNG_OPTIONAL(struct lttng_directory_handle
) chunk_directory_handle
=
2083 LTTNG_OPTIONAL_INIT
;
2086 * The session daemon will only provide a chunk directory file
2087 * descriptor for local traces.
2089 if (is_local_trace
) {
2092 /* Acnowledge the reception of the command. */
2093 ret
= consumer_send_status_msg(sock
,
2094 LTTCOMM_CONSUMERD_SUCCESS
);
2096 /* Somehow, the session daemon is not responding anymore. */
2100 ret
= lttcomm_recv_fds_unix_sock(sock
, &chunk_dirfd
, 1);
2101 if (ret
!= sizeof(chunk_dirfd
)) {
2102 ERR("Failed to receive trace chunk directory file descriptor");
2106 DBG("Received trace chunk directory fd (%d)",
2108 ret
= lttng_directory_handle_init_from_dirfd(
2109 &chunk_directory_handle
.value
,
2112 ERR("Failed to initialize chunk directory handle from directory file descriptor");
2113 if (close(chunk_dirfd
)) {
2114 PERROR("Failed to close chunk directory file descriptor");
2118 chunk_directory_handle
.is_set
= true;
2121 ret_code
= lttng_consumer_create_trace_chunk(
2122 !is_local_trace
? &relayd_id
: NULL
,
2123 msg
.u
.create_trace_chunk
.session_id
,
2124 msg
.u
.create_trace_chunk
.chunk_id
,
2125 (time_t) msg
.u
.create_trace_chunk
2126 .creation_timestamp
,
2127 chunk_override_name
,
2128 msg
.u
.create_trace_chunk
.credentials
.is_set
?
2131 chunk_directory_handle
.is_set
?
2132 &chunk_directory_handle
.value
:
2135 if (chunk_directory_handle
.is_set
) {
2136 lttng_directory_handle_fini(
2137 &chunk_directory_handle
.value
);
2139 goto end_msg_sessiond
;
2141 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK
:
2143 enum lttng_trace_chunk_command_type close_command
=
2144 msg
.u
.close_trace_chunk
.close_command
.value
;
2145 const uint64_t relayd_id
=
2146 msg
.u
.close_trace_chunk
.relayd_id
.value
;
2147 struct lttcomm_consumer_close_trace_chunk_reply reply
;
2148 char closed_trace_chunk_path
[LTTNG_PATH_MAX
];
2151 ret_code
= lttng_consumer_close_trace_chunk(
2152 msg
.u
.close_trace_chunk
.relayd_id
.is_set
?
2155 msg
.u
.close_trace_chunk
.session_id
,
2156 msg
.u
.close_trace_chunk
.chunk_id
,
2157 (time_t) msg
.u
.close_trace_chunk
.close_timestamp
,
2158 msg
.u
.close_trace_chunk
.close_command
.is_set
?
2160 NULL
, closed_trace_chunk_path
);
2161 reply
.ret_code
= ret_code
;
2162 reply
.path_length
= strlen(closed_trace_chunk_path
) + 1;
2163 ret
= lttcomm_send_unix_sock(sock
, &reply
, sizeof(reply
));
2164 if (ret
!= sizeof(reply
)) {
2167 ret
= lttcomm_send_unix_sock(sock
, closed_trace_chunk_path
,
2169 if (ret
!= reply
.path_length
) {
2174 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS
:
2176 const uint64_t relayd_id
=
2177 msg
.u
.trace_chunk_exists
.relayd_id
.value
;
2179 ret_code
= lttng_consumer_trace_chunk_exists(
2180 msg
.u
.trace_chunk_exists
.relayd_id
.is_set
?
2182 msg
.u
.trace_chunk_exists
.session_id
,
2183 msg
.u
.trace_chunk_exists
.chunk_id
);
2184 goto end_msg_sessiond
;
2192 * Return 1 to indicate success since the 0 value can be a socket
2193 * shutdown during the recv() or send() call.
2200 * The returned value here is not useful since either way we'll return 1 to
2201 * the caller because the session daemon socket management is done
2202 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2204 ret
= consumer_send_status_msg(sock
, ret_code
);
2214 * Free channel here since no one has a reference to it. We don't
2215 * free after that because a stream can store this pointer.
2217 destroy_channel(channel
);
2219 /* We have to send a status channel message indicating an error. */
2220 ret
= consumer_send_status_channel(sock
, NULL
);
2222 /* Stop everything if session daemon can not be notified. */
2229 /* This will issue a consumer stop. */
2235 health_code_update();
2239 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream
*stream
,
2240 int producer_active
)
2243 assert(stream
->ustream
);
2245 ustctl_flush_buffer(stream
->ustream
, producer_active
);
2249 * Take a snapshot for a specific stream.
2251 * Returns 0 on success, < 0 on error
2253 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2256 assert(stream
->ustream
);
2258 return ustctl_snapshot(stream
->ustream
);
2262 * Sample consumed and produced positions for a specific stream.
2264 * Returns 0 on success, < 0 on error.
2266 int lttng_ustconsumer_sample_snapshot_positions(
2267 struct lttng_consumer_stream
*stream
)
2270 assert(stream
->ustream
);
2272 return ustctl_snapshot_sample_positions(stream
->ustream
);
2276 * Get the produced position
2278 * Returns 0 on success, < 0 on error
2280 int lttng_ustconsumer_get_produced_snapshot(
2281 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2284 assert(stream
->ustream
);
2287 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
2291 * Get the consumed position
2293 * Returns 0 on success, < 0 on error
2295 int lttng_ustconsumer_get_consumed_snapshot(
2296 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2299 assert(stream
->ustream
);
2302 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
2305 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
2309 assert(stream
->ustream
);
2311 ustctl_flush_buffer(stream
->ustream
, producer
);
2314 int lttng_ustconsumer_get_current_timestamp(
2315 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
2318 assert(stream
->ustream
);
2321 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
2324 int lttng_ustconsumer_get_sequence_number(
2325 struct lttng_consumer_stream
*stream
, uint64_t *seq
)
2328 assert(stream
->ustream
);
2331 return ustctl_get_sequence_number(stream
->ustream
, seq
);
2335 * Called when the stream signals the consumer that it has hung up.
2337 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
2340 assert(stream
->ustream
);
2342 pthread_mutex_lock(&stream
->lock
);
2343 if (!stream
->quiescent
) {
2344 ustctl_flush_buffer(stream
->ustream
, 0);
2345 stream
->quiescent
= true;
2347 pthread_mutex_unlock(&stream
->lock
);
2348 stream
->hangup_flush_done
= 1;
2351 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
2356 assert(chan
->uchan
);
2357 assert(chan
->buffer_credentials
.is_set
);
2359 if (chan
->switch_timer_enabled
== 1) {
2360 consumer_timer_switch_stop(chan
);
2362 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
2365 ret
= close(chan
->stream_fds
[i
]);
2369 if (chan
->shm_path
[0]) {
2370 char shm_path
[PATH_MAX
];
2372 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
2374 ERR("Cannot get stream shm path");
2376 ret
= run_as_unlink(shm_path
,
2377 chan
->buffer_credentials
.value
.uid
,
2378 chan
->buffer_credentials
.value
.gid
);
2380 PERROR("unlink %s", shm_path
);
2386 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
2389 assert(chan
->uchan
);
2390 assert(chan
->buffer_credentials
.is_set
);
2392 consumer_metadata_cache_destroy(chan
);
2393 ustctl_destroy_channel(chan
->uchan
);
2394 /* Try to rmdir all directories under shm_path root. */
2395 if (chan
->root_shm_path
[0]) {
2396 (void) run_as_rmdir_recursive(chan
->root_shm_path
,
2397 chan
->buffer_credentials
.value
.uid
,
2398 chan
->buffer_credentials
.value
.gid
,
2399 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG
);
2401 free(chan
->stream_fds
);
2404 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
2407 assert(stream
->ustream
);
2409 if (stream
->chan
->switch_timer_enabled
== 1) {
2410 consumer_timer_switch_stop(stream
->chan
);
2412 ustctl_destroy_stream(stream
->ustream
);
2415 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
2418 assert(stream
->ustream
);
2420 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
2423 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
2426 assert(stream
->ustream
);
2428 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
2432 * Populate index values of a UST stream. Values are set in big endian order.
2434 * Return 0 on success or else a negative value.
2436 static int get_index_values(struct ctf_packet_index
*index
,
2437 struct ustctl_consumer_stream
*ustream
)
2440 uint64_t packet_size
, content_size
, timestamp_begin
, timestamp_end
,
2441 events_discarded
, stream_id
, stream_instance_id
,
2444 ret
= ustctl_get_timestamp_begin(ustream
, ×tamp_begin
);
2446 PERROR("ustctl_get_timestamp_begin");
2450 ret
= ustctl_get_timestamp_end(ustream
, ×tamp_end
);
2452 PERROR("ustctl_get_timestamp_end");
2456 ret
= ustctl_get_events_discarded(ustream
, &events_discarded
);
2458 PERROR("ustctl_get_events_discarded");
2462 ret
= ustctl_get_content_size(ustream
, &content_size
);
2464 PERROR("ustctl_get_content_size");
2468 ret
= ustctl_get_packet_size(ustream
, &packet_size
);
2470 PERROR("ustctl_get_packet_size");
2474 ret
= ustctl_get_stream_id(ustream
, &stream_id
);
2476 PERROR("ustctl_get_stream_id");
2480 ret
= ustctl_get_instance_id(ustream
, &stream_instance_id
);
2482 PERROR("ustctl_get_instance_id");
2486 ret
= ustctl_get_sequence_number(ustream
, &packet_seq_num
);
2488 PERROR("ustctl_get_sequence_number");
2492 *index
= (typeof(*index
)) {
2493 .offset
= index
->offset
,
2494 .packet_size
= htobe64(packet_size
),
2495 .content_size
= htobe64(content_size
),
2496 .timestamp_begin
= htobe64(timestamp_begin
),
2497 .timestamp_end
= htobe64(timestamp_end
),
2498 .events_discarded
= htobe64(events_discarded
),
2499 .stream_id
= htobe64(stream_id
),
2500 .stream_instance_id
= htobe64(stream_instance_id
),
2501 .packet_seq_num
= htobe64(packet_seq_num
),
2509 void metadata_stream_reset_cache(struct lttng_consumer_stream
*stream
,
2510 struct consumer_metadata_cache
*cache
)
2512 DBG("Metadata stream update to version %" PRIu64
,
2514 stream
->ust_metadata_pushed
= 0;
2515 stream
->metadata_version
= cache
->version
;
2516 stream
->reset_metadata_flag
= 1;
2520 * Check if the version of the metadata stream and metadata cache match.
2521 * If the cache got updated, reset the metadata stream.
2522 * The stream lock and metadata cache lock MUST be held.
2523 * Return 0 on success, a negative value on error.
2526 int metadata_stream_check_version(struct lttng_consumer_stream
*stream
)
2529 struct consumer_metadata_cache
*cache
= stream
->chan
->metadata_cache
;
2531 if (cache
->version
== stream
->metadata_version
) {
2534 metadata_stream_reset_cache(stream
, cache
);
2541 * Write up to one packet from the metadata cache to the channel.
2543 * Returns the number of bytes pushed in the cache, or a negative value
2547 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
2552 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2553 ret
= metadata_stream_check_version(stream
);
2557 if (stream
->chan
->metadata_cache
->max_offset
2558 == stream
->ust_metadata_pushed
) {
2563 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
2564 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
2565 stream
->chan
->metadata_cache
->max_offset
2566 - stream
->ust_metadata_pushed
);
2567 assert(write_len
!= 0);
2568 if (write_len
< 0) {
2569 ERR("Writing one metadata packet");
2573 stream
->ust_metadata_pushed
+= write_len
;
2575 assert(stream
->chan
->metadata_cache
->max_offset
>=
2576 stream
->ust_metadata_pushed
);
2580 * Switch packet (but don't open the next one) on every commit of
2581 * a metadata packet. Since the subbuffer is fully filled (with padding,
2582 * if needed), the stream is "quiescent" after this commit.
2584 ustctl_flush_buffer(stream
->ustream
, 1);
2585 stream
->quiescent
= true;
2587 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2593 * Sync metadata meaning request them to the session daemon and snapshot to the
2594 * metadata thread can consumer them.
2596 * Metadata stream lock is held here, but we need to release it when
2597 * interacting with sessiond, else we cause a deadlock with live
2598 * awaiting on metadata to be pushed out.
2600 * The RCU read side lock must be held by the caller.
2602 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2603 * is empty or a negative value on error.
2605 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data
*ctx
,
2606 struct lttng_consumer_stream
*metadata_stream
)
2610 struct lttng_consumer_channel
*metadata_channel
;
2613 assert(metadata_stream
);
2615 metadata_channel
= metadata_stream
->chan
;
2616 pthread_mutex_unlock(&metadata_stream
->lock
);
2618 * Request metadata from the sessiond, but don't wait for the flush
2619 * because we locked the metadata thread.
2621 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 0);
2622 pthread_mutex_lock(&metadata_stream
->lock
);
2628 * The metadata stream and channel can be deleted while the
2629 * metadata stream lock was released. The streamed is checked
2630 * for deletion before we use it further.
2632 * Note that it is safe to access a logically-deleted stream since its
2633 * existence is still guaranteed by the RCU read side lock. However,
2634 * it should no longer be used. The close/deletion of the metadata
2635 * channel and stream already guarantees that all metadata has been
2636 * consumed. Therefore, there is nothing left to do in this function.
2638 if (consumer_stream_is_deleted(metadata_stream
)) {
2639 DBG("Metadata stream %" PRIu64
" was deleted during the metadata synchronization",
2640 metadata_stream
->key
);
2645 ret
= commit_one_metadata_packet(metadata_stream
);
2648 } else if (ret
> 0) {
2652 ret
= ustctl_snapshot(metadata_stream
->ustream
);
2654 if (errno
!= EAGAIN
) {
2655 ERR("Sync metadata, taking UST snapshot");
2658 DBG("No new metadata when syncing them.");
2659 /* No new metadata, exit. */
2665 * After this flush, we still need to extract metadata.
2676 * Return 0 on success else a negative value.
2678 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2679 struct lttng_consumer_local_data
*ctx
)
2682 struct ustctl_consumer_stream
*ustream
;
2687 ustream
= stream
->ustream
;
2690 * First, we are going to check if there is a new subbuffer available
2691 * before reading the stream wait_fd.
2693 /* Get the next subbuffer */
2694 ret
= ustctl_get_next_subbuf(ustream
);
2696 /* No more data found, flag the stream. */
2697 stream
->has_data
= 0;
2702 ret
= ustctl_put_subbuf(ustream
);
2705 /* This stream still has data. Flag it and wake up the data thread. */
2706 stream
->has_data
= 1;
2708 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2711 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2712 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2717 /* The wake up pipe has been notified. */
2718 ctx
->has_wakeup
= 1;
2727 int update_stream_stats(struct lttng_consumer_stream
*stream
)
2730 uint64_t seq
, discarded
;
2732 ret
= ustctl_get_sequence_number(stream
->ustream
, &seq
);
2734 PERROR("ustctl_get_sequence_number");
2738 * Start the sequence when we extract the first packet in case we don't
2739 * start at 0 (for example if a consumer is not connected to the
2740 * session immediately after the beginning).
2742 if (stream
->last_sequence_number
== -1ULL) {
2743 stream
->last_sequence_number
= seq
;
2744 } else if (seq
> stream
->last_sequence_number
) {
2745 stream
->chan
->lost_packets
+= seq
-
2746 stream
->last_sequence_number
- 1;
2748 /* seq <= last_sequence_number */
2749 ERR("Sequence number inconsistent : prev = %" PRIu64
2750 ", current = %" PRIu64
,
2751 stream
->last_sequence_number
, seq
);
2755 stream
->last_sequence_number
= seq
;
2757 ret
= ustctl_get_events_discarded(stream
->ustream
, &discarded
);
2759 PERROR("kernctl_get_events_discarded");
2762 if (discarded
< stream
->last_discarded_events
) {
2764 * Overflow has occurred. We assume only one wrap-around
2767 stream
->chan
->discarded_events
+=
2768 (1ULL << (CAA_BITS_PER_LONG
- 1)) -
2769 stream
->last_discarded_events
+ discarded
;
2771 stream
->chan
->discarded_events
+= discarded
-
2772 stream
->last_discarded_events
;
2774 stream
->last_discarded_events
= discarded
;
2782 * Read subbuffer from the given stream.
2784 * Stream and channel locks MUST be acquired by the caller.
2786 * Return 0 on success else a negative value.
2788 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
2789 struct lttng_consumer_local_data
*ctx
)
2791 unsigned long len
, subbuf_size
, padding
;
2792 int err
, write_index
= 1, rotation_ret
;
2794 struct ustctl_consumer_stream
*ustream
;
2795 struct ctf_packet_index index
;
2796 const char *subbuf_addr
;
2797 struct lttng_buffer_view subbuf_view
;
2800 assert(stream
->ustream
);
2803 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream
->wait_fd
,
2806 /* Ease our life for what's next. */
2807 ustream
= stream
->ustream
;
2810 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2811 * error if we cannot read this one byte (read returns 0), or if the error
2812 * is EAGAIN or EWOULDBLOCK.
2814 * This is only done when the stream is monitored by a thread, before the
2815 * flush is done after a hangup and if the stream is not flagged with data
2816 * since there might be nothing to consume in the wait fd but still have
2817 * data available flagged by the consumer wake up pipe.
2819 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2823 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2824 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2831 * If the stream was flagged to be ready for rotation before we extract the
2832 * next packet, rotate it now.
2834 if (stream
->rotate_ready
) {
2835 DBG("Rotate stream before extracting data");
2836 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
);
2837 if (rotation_ret
< 0) {
2838 ERR("Stream rotation error");
2845 /* Get the next subbuffer */
2846 err
= ustctl_get_next_subbuf(ustream
);
2849 * Populate metadata info if the existing info has
2850 * already been read.
2852 if (stream
->metadata_flag
) {
2853 ret
= commit_one_metadata_packet(stream
);
2860 ret
= err
; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2862 * This is a debug message even for single-threaded consumer,
2863 * because poll() have more relaxed criterions than get subbuf,
2864 * so get_subbuf may fail for short race windows where poll()
2865 * would issue wakeups.
2867 DBG("Reserving sub buffer failed (everything is normal, "
2868 "it is due to concurrency) [ret: %d]", err
);
2871 assert(stream
->chan
->output
== CONSUMER_CHANNEL_MMAP
);
2873 if (!stream
->metadata_flag
) {
2874 index
.offset
= htobe64(stream
->out_fd_offset
);
2875 ret
= get_index_values(&index
, ustream
);
2877 err
= ustctl_put_subbuf(ustream
);
2882 /* Update the stream's sequence and discarded events count. */
2883 ret
= update_stream_stats(stream
);
2885 PERROR("kernctl_get_events_discarded");
2886 err
= ustctl_put_subbuf(ustream
);
2894 /* Get the full padded subbuffer size */
2895 err
= ustctl_get_padded_subbuf_size(ustream
, &len
);
2898 /* Get subbuffer data size (without padding) */
2899 err
= ustctl_get_subbuf_size(ustream
, &subbuf_size
);
2902 /* Make sure we don't get a subbuffer size bigger than the padded */
2903 assert(len
>= subbuf_size
);
2905 padding
= len
- subbuf_size
;
2907 ret
= get_current_subbuf_addr(stream
, &subbuf_addr
);
2910 goto error_put_subbuf
;
2913 subbuf_view
= lttng_buffer_view_init(subbuf_addr
, 0, len
);
2915 /* write the subbuffer to the tracefile */
2916 ret
= lttng_consumer_on_read_subbuffer_mmap(
2917 ctx
, stream
, &subbuf_view
, padding
, &index
);
2919 * The mmap operation should write subbuf_size amount of data when
2920 * network streaming or the full padding (len) size when we are _not_
2923 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
2924 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
2926 * Display the error but continue processing to try to release the
2927 * subbuffer. This is a DBG statement since any unexpected kill or
2928 * signal, the application gets unregistered, relayd gets closed or
2929 * anything that affects the buffer lifetime will trigger this error.
2930 * So, for the sake of the user, don't print this error since it can
2931 * happen and it is OK with the code flow.
2933 DBG("Error writing to tracefile "
2934 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2935 ret
, len
, subbuf_size
);
2939 err
= ustctl_put_next_subbuf(ustream
);
2943 * This will consumer the byte on the wait_fd if and only if there is not
2944 * next subbuffer to be acquired.
2946 if (!stream
->metadata_flag
) {
2947 ret
= notify_if_more_data(stream
, ctx
);
2953 /* Write index if needed. */
2958 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
2960 * In live, block until all the metadata is sent.
2962 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2963 assert(!stream
->missed_metadata_flush
);
2964 stream
->waiting_on_metadata
= true;
2965 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2967 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
2969 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2970 stream
->waiting_on_metadata
= false;
2971 if (stream
->missed_metadata_flush
) {
2972 stream
->missed_metadata_flush
= false;
2973 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2974 (void) consumer_flush_ust_index(stream
);
2976 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2984 assert(!stream
->metadata_flag
);
2985 err
= consumer_stream_write_index(stream
, &index
);
2992 * After extracting the packet, we check if the stream is now ready to be
2993 * rotated and perform the action immediately.
2995 rotation_ret
= lttng_consumer_stream_is_rotate_ready(stream
);
2996 if (rotation_ret
== 1) {
2997 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
);
2998 if (rotation_ret
< 0) {
2999 ERR("Stream rotation error");
3003 } else if (rotation_ret
< 0) {
3004 ERR("Checking if stream is ready to rotate");
3013 * Called when a stream is created.
3015 * Return 0 on success or else a negative value.
3017 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3024 * Don't create anything if this is set for streaming or if there is
3025 * no current trace chunk on the parent channel.
3027 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
&&
3028 stream
->chan
->trace_chunk
) {
3029 ret
= consumer_stream_create_output_files(stream
, true);
3041 * Check if data is still being extracted from the buffers for a specific
3042 * stream. Consumer data lock MUST be acquired before calling this function
3043 * and the stream lock.
3045 * Return 1 if the traced data are still getting read else 0 meaning that the
3046 * data is available for trace viewer reading.
3048 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
3053 assert(stream
->ustream
);
3055 DBG("UST consumer checking data pending");
3057 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
3062 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
3063 uint64_t contiguous
, pushed
;
3065 /* Ease our life a bit. */
3066 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
3067 pushed
= stream
->ust_metadata_pushed
;
3070 * We can simply check whether all contiguously available data
3071 * has been pushed to the ring buffer, since the push operation
3072 * is performed within get_next_subbuf(), and because both
3073 * get_next_subbuf() and put_next_subbuf() are issued atomically
3074 * thanks to the stream lock within
3075 * lttng_ustconsumer_read_subbuffer(). This basically means that
3076 * whetnever ust_metadata_pushed is incremented, the associated
3077 * metadata has been consumed from the metadata stream.
3079 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
3080 contiguous
, pushed
);
3081 assert(((int64_t) (contiguous
- pushed
)) >= 0);
3082 if ((contiguous
!= pushed
) ||
3083 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
3084 ret
= 1; /* Data is pending */
3088 ret
= ustctl_get_next_subbuf(stream
->ustream
);
3091 * There is still data so let's put back this
3094 ret
= ustctl_put_subbuf(stream
->ustream
);
3096 ret
= 1; /* Data is pending */
3101 /* Data is NOT pending so ready to be read. */
3109 * Stop a given metadata channel timer if enabled and close the wait fd which
3110 * is the poll pipe of the metadata stream.
3112 * This MUST be called with the metadata channel lock acquired.
3114 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
3119 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
3121 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
3123 if (metadata
->switch_timer_enabled
== 1) {
3124 consumer_timer_switch_stop(metadata
);
3127 if (!metadata
->metadata_stream
) {
3132 * Closing write side so the thread monitoring the stream wakes up if any
3133 * and clean the metadata stream.
3135 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
3136 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
3138 PERROR("closing metadata pipe write side");
3140 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
3148 * Close every metadata stream wait fd of the metadata hash table. This
3149 * function MUST be used very carefully so not to run into a race between the
3150 * metadata thread handling streams and this function closing their wait fd.
3152 * For UST, this is used when the session daemon hangs up. Its the metadata
3153 * producer so calling this is safe because we are assured that no state change
3154 * can occur in the metadata thread for the streams in the hash table.
3156 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
3158 struct lttng_ht_iter iter
;
3159 struct lttng_consumer_stream
*stream
;
3161 assert(metadata_ht
);
3162 assert(metadata_ht
->ht
);
3164 DBG("UST consumer closing all metadata streams");
3167 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
3170 health_code_update();
3172 pthread_mutex_lock(&stream
->chan
->lock
);
3173 lttng_ustconsumer_close_metadata(stream
->chan
);
3174 pthread_mutex_unlock(&stream
->chan
->lock
);
3180 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
3184 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
3186 ERR("Unable to close wakeup fd");
3191 * Please refer to consumer-timer.c before adding any lock within this
3192 * function or any of its callees. Timers have a very strict locking
3193 * semantic with respect to teardown. Failure to respect this semantic
3194 * introduces deadlocks.
3196 * DON'T hold the metadata lock when calling this function, else this
3197 * can cause deadlock involving consumer awaiting for metadata to be
3198 * pushed out due to concurrent interaction with the session daemon.
3200 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
3201 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
3203 struct lttcomm_metadata_request_msg request
;
3204 struct lttcomm_consumer_msg msg
;
3205 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3206 uint64_t len
, key
, offset
, version
;
3210 assert(channel
->metadata_cache
);
3212 memset(&request
, 0, sizeof(request
));
3214 /* send the metadata request to sessiond */
3215 switch (consumer_data
.type
) {
3216 case LTTNG_CONSUMER64_UST
:
3217 request
.bits_per_long
= 64;
3219 case LTTNG_CONSUMER32_UST
:
3220 request
.bits_per_long
= 32;
3223 request
.bits_per_long
= 0;
3227 request
.session_id
= channel
->session_id
;
3228 request
.session_id_per_pid
= channel
->session_id_per_pid
;
3230 * Request the application UID here so the metadata of that application can
3231 * be sent back. The channel UID corresponds to the user UID of the session
3232 * used for the rights on the stream file(s).
3234 request
.uid
= channel
->ust_app_uid
;
3235 request
.key
= channel
->key
;
3237 DBG("Sending metadata request to sessiond, session id %" PRIu64
3238 ", per-pid %" PRIu64
", app UID %u and channel key %" PRIu64
,
3239 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
3242 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
3244 health_code_update();
3246 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
3249 ERR("Asking metadata to sessiond");
3253 health_code_update();
3255 /* Receive the metadata from sessiond */
3256 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
3258 if (ret
!= sizeof(msg
)) {
3259 DBG("Consumer received unexpected message size %d (expects %zu)",
3261 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
3263 * The ret value might 0 meaning an orderly shutdown but this is ok
3264 * since the caller handles this.
3269 health_code_update();
3271 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
3272 /* No registry found */
3273 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3277 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
3278 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
3283 len
= msg
.u
.push_metadata
.len
;
3284 key
= msg
.u
.push_metadata
.key
;
3285 offset
= msg
.u
.push_metadata
.target_offset
;
3286 version
= msg
.u
.push_metadata
.version
;
3288 assert(key
== channel
->key
);
3290 DBG("No new metadata to receive for key %" PRIu64
, key
);
3293 health_code_update();
3295 /* Tell session daemon we are ready to receive the metadata. */
3296 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3297 LTTCOMM_CONSUMERD_SUCCESS
);
3298 if (ret
< 0 || len
== 0) {
3300 * Somehow, the session daemon is not responding anymore or there is
3301 * nothing to receive.
3306 health_code_update();
3308 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
3309 key
, offset
, len
, version
, channel
, timer
, wait
);
3312 * Only send the status msg if the sessiond is alive meaning a positive
3315 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
3320 health_code_update();
3322 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
3327 * Return the ustctl call for the get stream id.
3329 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
3330 uint64_t *stream_id
)
3335 return ustctl_get_stream_id(stream
->ustream
, stream_id
);