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
;
1225 health_code_update();
1227 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1229 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1231 if (ret
!= -EAGAIN
) {
1232 PERROR("ustctl_get_subbuf snapshot");
1233 goto error_close_stream
;
1235 DBG("UST consumer get subbuf failed. Skipping it.");
1236 consumed_pos
+= stream
->max_sb_size
;
1237 stream
->chan
->lost_packets
++;
1241 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1243 ERR("Snapshot ustctl_get_subbuf_size");
1244 goto error_put_subbuf
;
1247 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1249 ERR("Snapshot ustctl_get_padded_subbuf_size");
1250 goto error_put_subbuf
;
1253 ret
= get_current_subbuf_addr(stream
, &subbuf_addr
);
1255 goto error_put_subbuf
;
1258 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
,
1259 stream
, subbuf_addr
, len
,
1260 padded_len
- len
, NULL
);
1262 if (read_len
!= len
) {
1264 goto error_put_subbuf
;
1267 if (read_len
!= padded_len
) {
1269 goto error_put_subbuf
;
1273 ret
= ustctl_put_subbuf(stream
->ustream
);
1275 ERR("Snapshot ustctl_put_subbuf");
1276 goto error_close_stream
;
1278 consumed_pos
+= stream
->max_sb_size
;
1281 /* Simply close the stream so we can use it on the next snapshot. */
1282 consumer_stream_close(stream
);
1283 pthread_mutex_unlock(&stream
->lock
);
1290 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1291 ERR("Snapshot ustctl_put_subbuf");
1294 consumer_stream_close(stream
);
1296 pthread_mutex_unlock(&stream
->lock
);
1302 * Receive the metadata updates from the sessiond. Supports receiving
1303 * overlapping metadata, but is needs to always belong to a contiguous
1304 * range starting from 0.
1305 * Be careful about the locks held when calling this function: it needs
1306 * the metadata cache flush to concurrently progress in order to
1309 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1310 uint64_t len
, uint64_t version
,
1311 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
1313 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1316 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1318 metadata_str
= zmalloc(len
* sizeof(char));
1319 if (!metadata_str
) {
1320 PERROR("zmalloc metadata string");
1321 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1325 health_code_update();
1327 /* Receive metadata string. */
1328 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1330 /* Session daemon is dead so return gracefully. */
1335 health_code_update();
1337 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1338 ret
= consumer_metadata_cache_write(channel
, offset
, len
, version
,
1341 /* Unable to handle metadata. Notify session daemon. */
1342 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1344 * Skip metadata flush on write error since the offset and len might
1345 * not have been updated which could create an infinite loop below when
1346 * waiting for the metadata cache to be flushed.
1348 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1351 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1356 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1357 DBG("Waiting for metadata to be flushed");
1359 health_code_update();
1361 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1371 * Receive command from session daemon and process it.
1373 * Return 1 on success else a negative value or 0.
1375 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1376 int sock
, struct pollfd
*consumer_sockpoll
)
1379 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1380 struct lttcomm_consumer_msg msg
;
1381 struct lttng_consumer_channel
*channel
= NULL
;
1383 health_code_update();
1385 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1386 if (ret
!= sizeof(msg
)) {
1387 DBG("Consumer received unexpected message size %zd (expects %zu)",
1390 * The ret value might 0 meaning an orderly shutdown but this is ok
1391 * since the caller handles this.
1394 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1400 health_code_update();
1403 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1405 health_code_update();
1407 /* relayd needs RCU read-side lock */
1410 switch (msg
.cmd_type
) {
1411 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1413 /* Session daemon status message are handled in the following call. */
1414 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1415 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
1416 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
1417 msg
.u
.relayd_sock
.relayd_session_id
);
1420 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1422 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1423 struct consumer_relayd_sock_pair
*relayd
;
1425 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1427 /* Get relayd reference if exists. */
1428 relayd
= consumer_find_relayd(index
);
1429 if (relayd
== NULL
) {
1430 DBG("Unable to find relayd %" PRIu64
, index
);
1431 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1435 * Each relayd socket pair has a refcount of stream attached to it
1436 * which tells if the relayd is still active or not depending on the
1439 * This will set the destroy flag of the relayd object and destroy it
1440 * if the refcount reaches zero when called.
1442 * The destroy can happen either here or when a stream fd hangs up.
1445 consumer_flag_relayd_for_destroy(relayd
);
1448 goto end_msg_sessiond
;
1450 case LTTNG_CONSUMER_UPDATE_STREAM
:
1455 case LTTNG_CONSUMER_DATA_PENDING
:
1457 int ret
, is_data_pending
;
1458 uint64_t id
= msg
.u
.data_pending
.session_id
;
1460 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1462 is_data_pending
= consumer_data_pending(id
);
1464 /* Send back returned value to session daemon */
1465 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1466 sizeof(is_data_pending
));
1468 DBG("Error when sending the data pending ret code: %d", ret
);
1473 * No need to send back a status message since the data pending
1474 * returned value is the response.
1478 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1481 struct ustctl_consumer_channel_attr attr
;
1482 const uint64_t chunk_id
= msg
.u
.ask_channel
.chunk_id
.value
;
1483 const struct lttng_credentials buffer_credentials
= {
1484 .uid
= msg
.u
.ask_channel
.buffer_credentials
.uid
,
1485 .gid
= msg
.u
.ask_channel
.buffer_credentials
.gid
,
1488 /* Create a plain object and reserve a channel key. */
1489 channel
= allocate_channel(msg
.u
.ask_channel
.session_id
,
1490 msg
.u
.ask_channel
.chunk_id
.is_set
?
1492 msg
.u
.ask_channel
.pathname
,
1493 msg
.u
.ask_channel
.name
,
1494 msg
.u
.ask_channel
.relayd_id
,
1495 msg
.u
.ask_channel
.key
,
1496 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1497 msg
.u
.ask_channel
.tracefile_size
,
1498 msg
.u
.ask_channel
.tracefile_count
,
1499 msg
.u
.ask_channel
.session_id_per_pid
,
1500 msg
.u
.ask_channel
.monitor
,
1501 msg
.u
.ask_channel
.live_timer_interval
,
1502 msg
.u
.ask_channel
.root_shm_path
,
1503 msg
.u
.ask_channel
.shm_path
);
1505 goto end_channel_error
;
1508 LTTNG_OPTIONAL_SET(&channel
->buffer_credentials
,
1509 buffer_credentials
);
1512 * Assign UST application UID to the channel. This value is ignored for
1513 * per PID buffers. This is specific to UST thus setting this after the
1516 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1518 /* Build channel attributes from received message. */
1519 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1520 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1521 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1522 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1523 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1524 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1525 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1526 attr
.blocking_timeout
= msg
.u
.ask_channel
.blocking_timeout
;
1528 /* Match channel buffer type to the UST abi. */
1529 switch (msg
.u
.ask_channel
.output
) {
1530 case LTTNG_EVENT_MMAP
:
1532 attr
.output
= LTTNG_UST_MMAP
;
1536 /* Translate and save channel type. */
1537 switch (msg
.u
.ask_channel
.type
) {
1538 case LTTNG_UST_CHAN_PER_CPU
:
1539 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1540 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1542 * Set refcount to 1 for owner. Below, we will
1543 * pass ownership to the
1544 * consumer_thread_channel_poll() thread.
1546 channel
->refcount
= 1;
1548 case LTTNG_UST_CHAN_METADATA
:
1549 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1550 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1557 health_code_update();
1559 ret
= ask_channel(ctx
, channel
, &attr
);
1561 goto end_channel_error
;
1564 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1565 ret
= consumer_metadata_cache_allocate(channel
);
1567 ERR("Allocating metadata cache");
1568 goto end_channel_error
;
1570 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1571 attr
.switch_timer_interval
= 0;
1573 int monitor_start_ret
;
1575 consumer_timer_live_start(channel
,
1576 msg
.u
.ask_channel
.live_timer_interval
);
1577 monitor_start_ret
= consumer_timer_monitor_start(
1579 msg
.u
.ask_channel
.monitor_timer_interval
);
1580 if (monitor_start_ret
< 0) {
1581 ERR("Starting channel monitoring timer failed");
1582 goto end_channel_error
;
1586 health_code_update();
1589 * Add the channel to the internal state AFTER all streams were created
1590 * and successfully sent to session daemon. This way, all streams must
1591 * be ready before this channel is visible to the threads.
1592 * If add_channel succeeds, ownership of the channel is
1593 * passed to consumer_thread_channel_poll().
1595 ret
= add_channel(channel
, ctx
);
1597 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1598 if (channel
->switch_timer_enabled
== 1) {
1599 consumer_timer_switch_stop(channel
);
1601 consumer_metadata_cache_destroy(channel
);
1603 if (channel
->live_timer_enabled
== 1) {
1604 consumer_timer_live_stop(channel
);
1606 if (channel
->monitor_timer_enabled
== 1) {
1607 consumer_timer_monitor_stop(channel
);
1609 goto end_channel_error
;
1612 health_code_update();
1615 * Channel and streams are now created. Inform the session daemon that
1616 * everything went well and should wait to receive the channel and
1617 * streams with ustctl API.
1619 ret
= consumer_send_status_channel(sock
, channel
);
1622 * There is probably a problem on the socket.
1629 case LTTNG_CONSUMER_GET_CHANNEL
:
1631 int ret
, relayd_err
= 0;
1632 uint64_t key
= msg
.u
.get_channel
.key
;
1633 struct lttng_consumer_channel
*channel
;
1635 channel
= consumer_find_channel(key
);
1637 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1638 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1639 goto end_get_channel
;
1642 health_code_update();
1644 /* Send the channel to sessiond (and relayd, if applicable). */
1645 ret
= send_channel_to_sessiond_and_relayd(sock
, channel
, ctx
,
1650 * We were unable to send to the relayd the stream so avoid
1651 * sending back a fatal error to the thread since this is OK
1652 * and the consumer can continue its work. The above call
1653 * has sent the error status message to the sessiond.
1655 goto end_get_channel_nosignal
;
1658 * The communicaton was broken hence there is a bad state between
1659 * the consumer and sessiond so stop everything.
1661 goto error_get_channel_fatal
;
1664 health_code_update();
1667 * In no monitor mode, the streams ownership is kept inside the channel
1668 * so don't send them to the data thread.
1670 if (!channel
->monitor
) {
1671 goto end_get_channel
;
1674 ret
= send_streams_to_thread(channel
, ctx
);
1677 * If we are unable to send the stream to the thread, there is
1678 * a big problem so just stop everything.
1680 goto error_get_channel_fatal
;
1682 /* List MUST be empty after or else it could be reused. */
1683 assert(cds_list_empty(&channel
->streams
.head
));
1685 goto end_msg_sessiond
;
1686 error_get_channel_fatal
:
1688 end_get_channel_nosignal
:
1691 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1693 uint64_t key
= msg
.u
.destroy_channel
.key
;
1696 * Only called if streams have not been sent to stream
1697 * manager thread. However, channel has been sent to
1698 * channel manager thread.
1700 notify_thread_del_channel(ctx
, key
);
1701 goto end_msg_sessiond
;
1703 case LTTNG_CONSUMER_CLOSE_METADATA
:
1707 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1712 goto end_msg_sessiond
;
1714 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1718 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1723 goto end_msg_sessiond
;
1725 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL
:
1729 ret
= clear_quiescent_channel(
1730 msg
.u
.clear_quiescent_channel
.key
);
1735 goto end_msg_sessiond
;
1737 case LTTNG_CONSUMER_PUSH_METADATA
:
1740 uint64_t len
= msg
.u
.push_metadata
.len
;
1741 uint64_t key
= msg
.u
.push_metadata
.key
;
1742 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1743 uint64_t version
= msg
.u
.push_metadata
.version
;
1744 struct lttng_consumer_channel
*channel
;
1746 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1749 channel
= consumer_find_channel(key
);
1752 * This is possible if the metadata creation on the consumer side
1753 * is in flight vis-a-vis a concurrent push metadata from the
1754 * session daemon. Simply return that the channel failed and the
1755 * session daemon will handle that message correctly considering
1756 * that this race is acceptable thus the DBG() statement here.
1758 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1759 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1760 goto end_push_metadata_msg_sessiond
;
1763 health_code_update();
1767 * There is nothing to receive. We have simply
1768 * checked whether the channel can be found.
1770 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1771 goto end_push_metadata_msg_sessiond
;
1774 /* Tell session daemon we are ready to receive the metadata. */
1775 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1777 /* Somehow, the session daemon is not responding anymore. */
1778 goto error_push_metadata_fatal
;
1781 health_code_update();
1783 /* Wait for more data. */
1784 health_poll_entry();
1785 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1788 goto error_push_metadata_fatal
;
1791 health_code_update();
1793 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1794 len
, version
, channel
, 0, 1);
1796 /* error receiving from sessiond */
1797 goto error_push_metadata_fatal
;
1800 goto end_push_metadata_msg_sessiond
;
1802 end_push_metadata_msg_sessiond
:
1803 goto end_msg_sessiond
;
1804 error_push_metadata_fatal
:
1807 case LTTNG_CONSUMER_SETUP_METADATA
:
1811 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1815 goto end_msg_sessiond
;
1817 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1819 struct lttng_consumer_channel
*channel
;
1820 uint64_t key
= msg
.u
.snapshot_channel
.key
;
1822 channel
= consumer_find_channel(key
);
1824 DBG("UST snapshot channel not found for key %" PRIu64
, key
);
1825 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1827 if (msg
.u
.snapshot_channel
.metadata
) {
1828 ret
= snapshot_metadata(channel
, key
,
1829 msg
.u
.snapshot_channel
.pathname
,
1830 msg
.u
.snapshot_channel
.relayd_id
,
1833 ERR("Snapshot metadata failed");
1834 ret_code
= LTTCOMM_CONSUMERD_SNAPSHOT_FAILED
;
1837 ret
= snapshot_channel(channel
, key
,
1838 msg
.u
.snapshot_channel
.pathname
,
1839 msg
.u
.snapshot_channel
.relayd_id
,
1840 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
1843 ERR("Snapshot channel failed");
1844 ret_code
= LTTCOMM_CONSUMERD_SNAPSHOT_FAILED
;
1848 health_code_update();
1849 ret
= consumer_send_status_msg(sock
, ret_code
);
1851 /* Somehow, the session daemon is not responding anymore. */
1854 health_code_update();
1857 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
1860 uint64_t discarded_events
;
1861 struct lttng_ht_iter iter
;
1862 struct lttng_ht
*ht
;
1863 struct lttng_consumer_stream
*stream
;
1864 uint64_t id
= msg
.u
.discarded_events
.session_id
;
1865 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
1867 DBG("UST consumer discarded events command for session id %"
1870 pthread_mutex_lock(&consumer_data
.lock
);
1872 ht
= consumer_data
.stream_list_ht
;
1875 * We only need a reference to the channel, but they are not
1876 * directly indexed, so we just use the first matching stream
1877 * to extract the information we need, we default to 0 if not
1878 * found (no events are dropped if the channel is not yet in
1881 discarded_events
= 0;
1882 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1883 ht
->hash_fct(&id
, lttng_ht_seed
),
1885 &iter
.iter
, stream
, node_session_id
.node
) {
1886 if (stream
->chan
->key
== key
) {
1887 discarded_events
= stream
->chan
->discarded_events
;
1891 pthread_mutex_unlock(&consumer_data
.lock
);
1894 DBG("UST consumer discarded events command for session id %"
1895 PRIu64
", channel key %" PRIu64
, id
, key
);
1897 health_code_update();
1899 /* Send back returned value to session daemon */
1900 ret
= lttcomm_send_unix_sock(sock
, &discarded_events
, sizeof(discarded_events
));
1902 PERROR("send discarded events");
1908 case LTTNG_CONSUMER_LOST_PACKETS
:
1911 uint64_t lost_packets
;
1912 struct lttng_ht_iter iter
;
1913 struct lttng_ht
*ht
;
1914 struct lttng_consumer_stream
*stream
;
1915 uint64_t id
= msg
.u
.lost_packets
.session_id
;
1916 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
1918 DBG("UST consumer lost packets command for session id %"
1921 pthread_mutex_lock(&consumer_data
.lock
);
1923 ht
= consumer_data
.stream_list_ht
;
1926 * We only need a reference to the channel, but they are not
1927 * directly indexed, so we just use the first matching stream
1928 * to extract the information we need, we default to 0 if not
1929 * found (no packets lost if the channel is not yet in use).
1932 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1933 ht
->hash_fct(&id
, lttng_ht_seed
),
1935 &iter
.iter
, stream
, node_session_id
.node
) {
1936 if (stream
->chan
->key
== key
) {
1937 lost_packets
= stream
->chan
->lost_packets
;
1941 pthread_mutex_unlock(&consumer_data
.lock
);
1944 DBG("UST consumer lost packets command for session id %"
1945 PRIu64
", channel key %" PRIu64
, id
, key
);
1947 health_code_update();
1949 /* Send back returned value to session daemon */
1950 ret
= lttcomm_send_unix_sock(sock
, &lost_packets
,
1951 sizeof(lost_packets
));
1953 PERROR("send lost packets");
1959 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
1961 int channel_monitor_pipe
;
1963 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1964 /* Successfully received the command's type. */
1965 ret
= consumer_send_status_msg(sock
, ret_code
);
1970 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
1972 if (ret
!= sizeof(channel_monitor_pipe
)) {
1973 ERR("Failed to receive channel monitor pipe");
1977 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
1978 ret
= consumer_timer_thread_set_channel_monitor_pipe(
1979 channel_monitor_pipe
);
1983 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1984 /* Set the pipe as non-blocking. */
1985 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
1987 PERROR("fcntl get flags of the channel monitoring pipe");
1992 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
1993 flags
| O_NONBLOCK
);
1995 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1998 DBG("Channel monitor pipe set as non-blocking");
2000 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
2002 goto end_msg_sessiond
;
2004 case LTTNG_CONSUMER_ROTATE_CHANNEL
:
2006 struct lttng_consumer_channel
*channel
;
2007 uint64_t key
= msg
.u
.rotate_channel
.key
;
2009 channel
= consumer_find_channel(key
);
2011 DBG("Channel %" PRIu64
" not found", key
);
2012 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2015 * Sample the rotate position of all the streams in
2018 ret
= lttng_consumer_rotate_channel(channel
, key
,
2019 msg
.u
.rotate_channel
.relayd_id
,
2020 msg
.u
.rotate_channel
.metadata
,
2023 ERR("Rotate channel failed");
2024 ret_code
= LTTCOMM_CONSUMERD_ROTATION_FAIL
;
2027 health_code_update();
2029 ret
= consumer_send_status_msg(sock
, ret_code
);
2031 /* Somehow, the session daemon is not responding anymore. */
2032 goto end_rotate_channel_nosignal
;
2036 * Rotate the streams that are ready right now.
2037 * FIXME: this is a second consecutive iteration over the
2038 * streams in a channel, there is probably a better way to
2039 * handle this, but it needs to be after the
2040 * consumer_send_status_msg() call.
2043 ret
= lttng_consumer_rotate_ready_streams(
2046 ERR("Rotate channel failed");
2050 end_rotate_channel_nosignal
:
2053 case LTTNG_CONSUMER_INIT
:
2055 ret_code
= lttng_consumer_init_command(ctx
,
2056 msg
.u
.init
.sessiond_uuid
);
2057 health_code_update();
2058 ret
= consumer_send_status_msg(sock
, ret_code
);
2060 /* Somehow, the session daemon is not responding anymore. */
2065 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK
:
2067 const struct lttng_credentials credentials
= {
2068 .uid
= msg
.u
.create_trace_chunk
.credentials
.value
.uid
,
2069 .gid
= msg
.u
.create_trace_chunk
.credentials
.value
.gid
,
2071 const bool is_local_trace
=
2072 !msg
.u
.create_trace_chunk
.relayd_id
.is_set
;
2073 const uint64_t relayd_id
=
2074 msg
.u
.create_trace_chunk
.relayd_id
.value
;
2075 const char *chunk_override_name
=
2076 *msg
.u
.create_trace_chunk
.override_name
?
2077 msg
.u
.create_trace_chunk
.override_name
:
2079 LTTNG_OPTIONAL(struct lttng_directory_handle
) chunk_directory_handle
=
2080 LTTNG_OPTIONAL_INIT
;
2083 * The session daemon will only provide a chunk directory file
2084 * descriptor for local traces.
2086 if (is_local_trace
) {
2089 /* Acnowledge the reception of the command. */
2090 ret
= consumer_send_status_msg(sock
,
2091 LTTCOMM_CONSUMERD_SUCCESS
);
2093 /* Somehow, the session daemon is not responding anymore. */
2097 ret
= lttcomm_recv_fds_unix_sock(sock
, &chunk_dirfd
, 1);
2098 if (ret
!= sizeof(chunk_dirfd
)) {
2099 ERR("Failed to receive trace chunk directory file descriptor");
2103 DBG("Received trace chunk directory fd (%d)",
2105 ret
= lttng_directory_handle_init_from_dirfd(
2106 &chunk_directory_handle
.value
,
2109 ERR("Failed to initialize chunk directory handle from directory file descriptor");
2110 if (close(chunk_dirfd
)) {
2111 PERROR("Failed to close chunk directory file descriptor");
2115 chunk_directory_handle
.is_set
= true;
2118 ret_code
= lttng_consumer_create_trace_chunk(
2119 !is_local_trace
? &relayd_id
: NULL
,
2120 msg
.u
.create_trace_chunk
.session_id
,
2121 msg
.u
.create_trace_chunk
.chunk_id
,
2122 (time_t) msg
.u
.create_trace_chunk
2123 .creation_timestamp
,
2124 chunk_override_name
,
2125 msg
.u
.create_trace_chunk
.credentials
.is_set
?
2128 chunk_directory_handle
.is_set
?
2129 &chunk_directory_handle
.value
:
2132 if (chunk_directory_handle
.is_set
) {
2133 lttng_directory_handle_fini(
2134 &chunk_directory_handle
.value
);
2136 goto end_msg_sessiond
;
2138 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK
:
2140 enum lttng_trace_chunk_command_type close_command
=
2141 msg
.u
.close_trace_chunk
.close_command
.value
;
2142 const uint64_t relayd_id
=
2143 msg
.u
.close_trace_chunk
.relayd_id
.value
;
2144 struct lttcomm_consumer_close_trace_chunk_reply reply
;
2145 char closed_trace_chunk_path
[LTTNG_PATH_MAX
];
2148 ret_code
= lttng_consumer_close_trace_chunk(
2149 msg
.u
.close_trace_chunk
.relayd_id
.is_set
?
2152 msg
.u
.close_trace_chunk
.session_id
,
2153 msg
.u
.close_trace_chunk
.chunk_id
,
2154 (time_t) msg
.u
.close_trace_chunk
.close_timestamp
,
2155 msg
.u
.close_trace_chunk
.close_command
.is_set
?
2157 NULL
, closed_trace_chunk_path
);
2158 reply
.ret_code
= ret_code
;
2159 reply
.path_length
= strlen(closed_trace_chunk_path
) + 1;
2160 ret
= lttcomm_send_unix_sock(sock
, &reply
, sizeof(reply
));
2161 if (ret
!= sizeof(reply
)) {
2164 ret
= lttcomm_send_unix_sock(sock
, closed_trace_chunk_path
,
2166 if (ret
!= reply
.path_length
) {
2171 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS
:
2173 const uint64_t relayd_id
=
2174 msg
.u
.trace_chunk_exists
.relayd_id
.value
;
2176 ret_code
= lttng_consumer_trace_chunk_exists(
2177 msg
.u
.trace_chunk_exists
.relayd_id
.is_set
?
2179 msg
.u
.trace_chunk_exists
.session_id
,
2180 msg
.u
.trace_chunk_exists
.chunk_id
);
2181 goto end_msg_sessiond
;
2189 * Return 1 to indicate success since the 0 value can be a socket
2190 * shutdown during the recv() or send() call.
2197 * The returned value here is not useful since either way we'll return 1 to
2198 * the caller because the session daemon socket management is done
2199 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2201 ret
= consumer_send_status_msg(sock
, ret_code
);
2211 * Free channel here since no one has a reference to it. We don't
2212 * free after that because a stream can store this pointer.
2214 destroy_channel(channel
);
2216 /* We have to send a status channel message indicating an error. */
2217 ret
= consumer_send_status_channel(sock
, NULL
);
2219 /* Stop everything if session daemon can not be notified. */
2226 /* This will issue a consumer stop. */
2232 health_code_update();
2236 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream
*stream
,
2237 int producer_active
)
2240 assert(stream
->ustream
);
2242 ustctl_flush_buffer(stream
->ustream
, producer_active
);
2246 * Take a snapshot for a specific stream.
2248 * Returns 0 on success, < 0 on error
2250 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2253 assert(stream
->ustream
);
2255 return ustctl_snapshot(stream
->ustream
);
2259 * Sample consumed and produced positions for a specific stream.
2261 * Returns 0 on success, < 0 on error.
2263 int lttng_ustconsumer_sample_snapshot_positions(
2264 struct lttng_consumer_stream
*stream
)
2267 assert(stream
->ustream
);
2269 return ustctl_snapshot_sample_positions(stream
->ustream
);
2273 * Get the produced position
2275 * Returns 0 on success, < 0 on error
2277 int lttng_ustconsumer_get_produced_snapshot(
2278 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2281 assert(stream
->ustream
);
2284 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
2288 * Get the consumed position
2290 * Returns 0 on success, < 0 on error
2292 int lttng_ustconsumer_get_consumed_snapshot(
2293 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2296 assert(stream
->ustream
);
2299 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
2302 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
2306 assert(stream
->ustream
);
2308 ustctl_flush_buffer(stream
->ustream
, producer
);
2311 int lttng_ustconsumer_get_current_timestamp(
2312 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
2315 assert(stream
->ustream
);
2318 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
2321 int lttng_ustconsumer_get_sequence_number(
2322 struct lttng_consumer_stream
*stream
, uint64_t *seq
)
2325 assert(stream
->ustream
);
2328 return ustctl_get_sequence_number(stream
->ustream
, seq
);
2332 * Called when the stream signals the consumer that it has hung up.
2334 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
2337 assert(stream
->ustream
);
2339 pthread_mutex_lock(&stream
->lock
);
2340 if (!stream
->quiescent
) {
2341 ustctl_flush_buffer(stream
->ustream
, 0);
2342 stream
->quiescent
= true;
2344 pthread_mutex_unlock(&stream
->lock
);
2345 stream
->hangup_flush_done
= 1;
2348 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
2353 assert(chan
->uchan
);
2354 assert(chan
->buffer_credentials
.is_set
);
2356 if (chan
->switch_timer_enabled
== 1) {
2357 consumer_timer_switch_stop(chan
);
2359 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
2362 ret
= close(chan
->stream_fds
[i
]);
2366 if (chan
->shm_path
[0]) {
2367 char shm_path
[PATH_MAX
];
2369 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
2371 ERR("Cannot get stream shm path");
2373 ret
= run_as_unlink(shm_path
,
2374 chan
->buffer_credentials
.value
.uid
,
2375 chan
->buffer_credentials
.value
.gid
);
2377 PERROR("unlink %s", shm_path
);
2383 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
2386 assert(chan
->uchan
);
2387 assert(chan
->buffer_credentials
.is_set
);
2389 consumer_metadata_cache_destroy(chan
);
2390 ustctl_destroy_channel(chan
->uchan
);
2391 /* Try to rmdir all directories under shm_path root. */
2392 if (chan
->root_shm_path
[0]) {
2393 (void) run_as_rmdir_recursive(chan
->root_shm_path
,
2394 chan
->buffer_credentials
.value
.uid
,
2395 chan
->buffer_credentials
.value
.gid
,
2396 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG
);
2398 free(chan
->stream_fds
);
2401 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
2404 assert(stream
->ustream
);
2406 if (stream
->chan
->switch_timer_enabled
== 1) {
2407 consumer_timer_switch_stop(stream
->chan
);
2409 ustctl_destroy_stream(stream
->ustream
);
2412 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
2415 assert(stream
->ustream
);
2417 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
2420 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
2423 assert(stream
->ustream
);
2425 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
2429 * Populate index values of a UST stream. Values are set in big endian order.
2431 * Return 0 on success or else a negative value.
2433 static int get_index_values(struct ctf_packet_index
*index
,
2434 struct ustctl_consumer_stream
*ustream
)
2437 uint64_t packet_size
, content_size
, timestamp_begin
, timestamp_end
,
2438 events_discarded
, stream_id
, stream_instance_id
,
2441 ret
= ustctl_get_timestamp_begin(ustream
, ×tamp_begin
);
2443 PERROR("ustctl_get_timestamp_begin");
2447 ret
= ustctl_get_timestamp_end(ustream
, ×tamp_end
);
2449 PERROR("ustctl_get_timestamp_end");
2453 ret
= ustctl_get_events_discarded(ustream
, &events_discarded
);
2455 PERROR("ustctl_get_events_discarded");
2459 ret
= ustctl_get_content_size(ustream
, &content_size
);
2461 PERROR("ustctl_get_content_size");
2465 ret
= ustctl_get_packet_size(ustream
, &packet_size
);
2467 PERROR("ustctl_get_packet_size");
2471 ret
= ustctl_get_stream_id(ustream
, &stream_id
);
2473 PERROR("ustctl_get_stream_id");
2477 ret
= ustctl_get_instance_id(ustream
, &stream_instance_id
);
2479 PERROR("ustctl_get_instance_id");
2483 ret
= ustctl_get_sequence_number(ustream
, &packet_seq_num
);
2485 PERROR("ustctl_get_sequence_number");
2489 *index
= (typeof(*index
)) {
2490 .offset
= index
->offset
,
2491 .packet_size
= htobe64(packet_size
),
2492 .content_size
= htobe64(content_size
),
2493 .timestamp_begin
= htobe64(timestamp_begin
),
2494 .timestamp_end
= htobe64(timestamp_end
),
2495 .events_discarded
= htobe64(events_discarded
),
2496 .stream_id
= htobe64(stream_id
),
2497 .stream_instance_id
= htobe64(stream_instance_id
),
2498 .packet_seq_num
= htobe64(packet_seq_num
),
2506 void metadata_stream_reset_cache(struct lttng_consumer_stream
*stream
,
2507 struct consumer_metadata_cache
*cache
)
2509 DBG("Metadata stream update to version %" PRIu64
,
2511 stream
->ust_metadata_pushed
= 0;
2512 stream
->metadata_version
= cache
->version
;
2513 stream
->reset_metadata_flag
= 1;
2517 * Check if the version of the metadata stream and metadata cache match.
2518 * If the cache got updated, reset the metadata stream.
2519 * The stream lock and metadata cache lock MUST be held.
2520 * Return 0 on success, a negative value on error.
2523 int metadata_stream_check_version(struct lttng_consumer_stream
*stream
)
2526 struct consumer_metadata_cache
*cache
= stream
->chan
->metadata_cache
;
2528 if (cache
->version
== stream
->metadata_version
) {
2531 metadata_stream_reset_cache(stream
, cache
);
2538 * Write up to one packet from the metadata cache to the channel.
2540 * Returns the number of bytes pushed in the cache, or a negative value
2544 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
2549 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2550 ret
= metadata_stream_check_version(stream
);
2554 if (stream
->chan
->metadata_cache
->max_offset
2555 == stream
->ust_metadata_pushed
) {
2560 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
2561 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
2562 stream
->chan
->metadata_cache
->max_offset
2563 - stream
->ust_metadata_pushed
);
2564 assert(write_len
!= 0);
2565 if (write_len
< 0) {
2566 ERR("Writing one metadata packet");
2570 stream
->ust_metadata_pushed
+= write_len
;
2572 assert(stream
->chan
->metadata_cache
->max_offset
>=
2573 stream
->ust_metadata_pushed
);
2577 * Switch packet (but don't open the next one) on every commit of
2578 * a metadata packet. Since the subbuffer is fully filled (with padding,
2579 * if needed), the stream is "quiescent" after this commit.
2581 ustctl_flush_buffer(stream
->ustream
, 1);
2582 stream
->quiescent
= true;
2584 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2590 * Sync metadata meaning request them to the session daemon and snapshot to the
2591 * metadata thread can consumer them.
2593 * Metadata stream lock is held here, but we need to release it when
2594 * interacting with sessiond, else we cause a deadlock with live
2595 * awaiting on metadata to be pushed out.
2597 * The RCU read side lock must be held by the caller.
2599 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2600 * is empty or a negative value on error.
2602 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data
*ctx
,
2603 struct lttng_consumer_stream
*metadata_stream
)
2607 struct lttng_consumer_channel
*metadata_channel
;
2610 assert(metadata_stream
);
2612 metadata_channel
= metadata_stream
->chan
;
2613 pthread_mutex_unlock(&metadata_stream
->lock
);
2615 * Request metadata from the sessiond, but don't wait for the flush
2616 * because we locked the metadata thread.
2618 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 0);
2619 pthread_mutex_lock(&metadata_stream
->lock
);
2625 * The metadata stream and channel can be deleted while the
2626 * metadata stream lock was released. The streamed is checked
2627 * for deletion before we use it further.
2629 * Note that it is safe to access a logically-deleted stream since its
2630 * existence is still guaranteed by the RCU read side lock. However,
2631 * it should no longer be used. The close/deletion of the metadata
2632 * channel and stream already guarantees that all metadata has been
2633 * consumed. Therefore, there is nothing left to do in this function.
2635 if (consumer_stream_is_deleted(metadata_stream
)) {
2636 DBG("Metadata stream %" PRIu64
" was deleted during the metadata synchronization",
2637 metadata_stream
->key
);
2642 ret
= commit_one_metadata_packet(metadata_stream
);
2645 } else if (ret
> 0) {
2649 ret
= ustctl_snapshot(metadata_stream
->ustream
);
2651 if (errno
!= EAGAIN
) {
2652 ERR("Sync metadata, taking UST snapshot");
2655 DBG("No new metadata when syncing them.");
2656 /* No new metadata, exit. */
2662 * After this flush, we still need to extract metadata.
2673 * Return 0 on success else a negative value.
2675 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2676 struct lttng_consumer_local_data
*ctx
)
2679 struct ustctl_consumer_stream
*ustream
;
2684 ustream
= stream
->ustream
;
2687 * First, we are going to check if there is a new subbuffer available
2688 * before reading the stream wait_fd.
2690 /* Get the next subbuffer */
2691 ret
= ustctl_get_next_subbuf(ustream
);
2693 /* No more data found, flag the stream. */
2694 stream
->has_data
= 0;
2699 ret
= ustctl_put_subbuf(ustream
);
2702 /* This stream still has data. Flag it and wake up the data thread. */
2703 stream
->has_data
= 1;
2705 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2708 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2709 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2714 /* The wake up pipe has been notified. */
2715 ctx
->has_wakeup
= 1;
2724 int update_stream_stats(struct lttng_consumer_stream
*stream
)
2727 uint64_t seq
, discarded
;
2729 ret
= ustctl_get_sequence_number(stream
->ustream
, &seq
);
2731 PERROR("ustctl_get_sequence_number");
2735 * Start the sequence when we extract the first packet in case we don't
2736 * start at 0 (for example if a consumer is not connected to the
2737 * session immediately after the beginning).
2739 if (stream
->last_sequence_number
== -1ULL) {
2740 stream
->last_sequence_number
= seq
;
2741 } else if (seq
> stream
->last_sequence_number
) {
2742 stream
->chan
->lost_packets
+= seq
-
2743 stream
->last_sequence_number
- 1;
2745 /* seq <= last_sequence_number */
2746 ERR("Sequence number inconsistent : prev = %" PRIu64
2747 ", current = %" PRIu64
,
2748 stream
->last_sequence_number
, seq
);
2752 stream
->last_sequence_number
= seq
;
2754 ret
= ustctl_get_events_discarded(stream
->ustream
, &discarded
);
2756 PERROR("kernctl_get_events_discarded");
2759 if (discarded
< stream
->last_discarded_events
) {
2761 * Overflow has occurred. We assume only one wrap-around
2764 stream
->chan
->discarded_events
+=
2765 (1ULL << (CAA_BITS_PER_LONG
- 1)) -
2766 stream
->last_discarded_events
+ discarded
;
2768 stream
->chan
->discarded_events
+= discarded
-
2769 stream
->last_discarded_events
;
2771 stream
->last_discarded_events
= discarded
;
2779 * Read subbuffer from the given stream.
2781 * Stream and channel locks MUST be acquired by the caller.
2783 * Return 0 on success else a negative value.
2785 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
2786 struct lttng_consumer_local_data
*ctx
)
2788 unsigned long len
, subbuf_size
, padding
;
2789 int err
, write_index
= 1, rotation_ret
;
2791 struct ustctl_consumer_stream
*ustream
;
2792 struct ctf_packet_index index
;
2793 const char *subbuf_addr
;
2796 assert(stream
->ustream
);
2799 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream
->wait_fd
,
2802 /* Ease our life for what's next. */
2803 ustream
= stream
->ustream
;
2806 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2807 * error if we cannot read this one byte (read returns 0), or if the error
2808 * is EAGAIN or EWOULDBLOCK.
2810 * This is only done when the stream is monitored by a thread, before the
2811 * flush is done after a hangup and if the stream is not flagged with data
2812 * since there might be nothing to consume in the wait fd but still have
2813 * data available flagged by the consumer wake up pipe.
2815 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2819 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2820 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2827 * If the stream was flagged to be ready for rotation before we extract the
2828 * next packet, rotate it now.
2830 if (stream
->rotate_ready
) {
2831 DBG("Rotate stream before extracting data");
2832 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
);
2833 if (rotation_ret
< 0) {
2834 ERR("Stream rotation error");
2841 /* Get the next subbuffer */
2842 err
= ustctl_get_next_subbuf(ustream
);
2845 * Populate metadata info if the existing info has
2846 * already been read.
2848 if (stream
->metadata_flag
) {
2849 ret
= commit_one_metadata_packet(stream
);
2856 ret
= err
; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2858 * This is a debug message even for single-threaded consumer,
2859 * because poll() have more relaxed criterions than get subbuf,
2860 * so get_subbuf may fail for short race windows where poll()
2861 * would issue wakeups.
2863 DBG("Reserving sub buffer failed (everything is normal, "
2864 "it is due to concurrency) [ret: %d]", err
);
2867 assert(stream
->chan
->output
== CONSUMER_CHANNEL_MMAP
);
2869 if (!stream
->metadata_flag
) {
2870 index
.offset
= htobe64(stream
->out_fd_offset
);
2871 ret
= get_index_values(&index
, ustream
);
2873 err
= ustctl_put_subbuf(ustream
);
2878 /* Update the stream's sequence and discarded events count. */
2879 ret
= update_stream_stats(stream
);
2881 PERROR("kernctl_get_events_discarded");
2882 err
= ustctl_put_subbuf(ustream
);
2890 /* Get the full padded subbuffer size */
2891 err
= ustctl_get_padded_subbuf_size(ustream
, &len
);
2894 /* Get subbuffer data size (without padding) */
2895 err
= ustctl_get_subbuf_size(ustream
, &subbuf_size
);
2898 /* Make sure we don't get a subbuffer size bigger than the padded */
2899 assert(len
>= subbuf_size
);
2901 padding
= len
- subbuf_size
;
2903 ret
= get_current_subbuf_addr(stream
, &subbuf_addr
);
2906 goto error_put_subbuf
;
2909 /* write the subbuffer to the tracefile */
2910 ret
= lttng_consumer_on_read_subbuffer_mmap(
2911 ctx
, stream
, subbuf_addr
, subbuf_size
, padding
, &index
);
2913 * The mmap operation should write subbuf_size amount of data when
2914 * network streaming or the full padding (len) size when we are _not_
2917 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
2918 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
2920 * Display the error but continue processing to try to release the
2921 * subbuffer. This is a DBG statement since any unexpected kill or
2922 * signal, the application gets unregistered, relayd gets closed or
2923 * anything that affects the buffer lifetime will trigger this error.
2924 * So, for the sake of the user, don't print this error since it can
2925 * happen and it is OK with the code flow.
2927 DBG("Error writing to tracefile "
2928 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2929 ret
, len
, subbuf_size
);
2933 err
= ustctl_put_next_subbuf(ustream
);
2937 * This will consumer the byte on the wait_fd if and only if there is not
2938 * next subbuffer to be acquired.
2940 if (!stream
->metadata_flag
) {
2941 ret
= notify_if_more_data(stream
, ctx
);
2947 /* Write index if needed. */
2952 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
2954 * In live, block until all the metadata is sent.
2956 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2957 assert(!stream
->missed_metadata_flush
);
2958 stream
->waiting_on_metadata
= true;
2959 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2961 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
2963 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2964 stream
->waiting_on_metadata
= false;
2965 if (stream
->missed_metadata_flush
) {
2966 stream
->missed_metadata_flush
= false;
2967 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2968 (void) consumer_flush_ust_index(stream
);
2970 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2978 assert(!stream
->metadata_flag
);
2979 err
= consumer_stream_write_index(stream
, &index
);
2986 * After extracting the packet, we check if the stream is now ready to be
2987 * rotated and perform the action immediately.
2989 rotation_ret
= lttng_consumer_stream_is_rotate_ready(stream
);
2990 if (rotation_ret
== 1) {
2991 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
);
2992 if (rotation_ret
< 0) {
2993 ERR("Stream rotation error");
2997 } else if (rotation_ret
< 0) {
2998 ERR("Checking if stream is ready to rotate");
3007 * Called when a stream is created.
3009 * Return 0 on success or else a negative value.
3011 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3018 * Don't create anything if this is set for streaming or if there is
3019 * no current trace chunk on the parent channel.
3021 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
&&
3022 stream
->chan
->trace_chunk
) {
3023 ret
= consumer_stream_create_output_files(stream
, true);
3035 * Check if data is still being extracted from the buffers for a specific
3036 * stream. Consumer data lock MUST be acquired before calling this function
3037 * and the stream lock.
3039 * Return 1 if the traced data are still getting read else 0 meaning that the
3040 * data is available for trace viewer reading.
3042 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
3047 assert(stream
->ustream
);
3049 DBG("UST consumer checking data pending");
3051 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
3056 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
3057 uint64_t contiguous
, pushed
;
3059 /* Ease our life a bit. */
3060 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
3061 pushed
= stream
->ust_metadata_pushed
;
3064 * We can simply check whether all contiguously available data
3065 * has been pushed to the ring buffer, since the push operation
3066 * is performed within get_next_subbuf(), and because both
3067 * get_next_subbuf() and put_next_subbuf() are issued atomically
3068 * thanks to the stream lock within
3069 * lttng_ustconsumer_read_subbuffer(). This basically means that
3070 * whetnever ust_metadata_pushed is incremented, the associated
3071 * metadata has been consumed from the metadata stream.
3073 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
3074 contiguous
, pushed
);
3075 assert(((int64_t) (contiguous
- pushed
)) >= 0);
3076 if ((contiguous
!= pushed
) ||
3077 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
3078 ret
= 1; /* Data is pending */
3082 ret
= ustctl_get_next_subbuf(stream
->ustream
);
3085 * There is still data so let's put back this
3088 ret
= ustctl_put_subbuf(stream
->ustream
);
3090 ret
= 1; /* Data is pending */
3095 /* Data is NOT pending so ready to be read. */
3103 * Stop a given metadata channel timer if enabled and close the wait fd which
3104 * is the poll pipe of the metadata stream.
3106 * This MUST be called with the metadata channel lock acquired.
3108 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
3113 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
3115 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
3117 if (metadata
->switch_timer_enabled
== 1) {
3118 consumer_timer_switch_stop(metadata
);
3121 if (!metadata
->metadata_stream
) {
3126 * Closing write side so the thread monitoring the stream wakes up if any
3127 * and clean the metadata stream.
3129 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
3130 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
3132 PERROR("closing metadata pipe write side");
3134 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
3142 * Close every metadata stream wait fd of the metadata hash table. This
3143 * function MUST be used very carefully so not to run into a race between the
3144 * metadata thread handling streams and this function closing their wait fd.
3146 * For UST, this is used when the session daemon hangs up. Its the metadata
3147 * producer so calling this is safe because we are assured that no state change
3148 * can occur in the metadata thread for the streams in the hash table.
3150 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
3152 struct lttng_ht_iter iter
;
3153 struct lttng_consumer_stream
*stream
;
3155 assert(metadata_ht
);
3156 assert(metadata_ht
->ht
);
3158 DBG("UST consumer closing all metadata streams");
3161 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
3164 health_code_update();
3166 pthread_mutex_lock(&stream
->chan
->lock
);
3167 lttng_ustconsumer_close_metadata(stream
->chan
);
3168 pthread_mutex_unlock(&stream
->chan
->lock
);
3174 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
3178 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
3180 ERR("Unable to close wakeup fd");
3185 * Please refer to consumer-timer.c before adding any lock within this
3186 * function or any of its callees. Timers have a very strict locking
3187 * semantic with respect to teardown. Failure to respect this semantic
3188 * introduces deadlocks.
3190 * DON'T hold the metadata lock when calling this function, else this
3191 * can cause deadlock involving consumer awaiting for metadata to be
3192 * pushed out due to concurrent interaction with the session daemon.
3194 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
3195 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
3197 struct lttcomm_metadata_request_msg request
;
3198 struct lttcomm_consumer_msg msg
;
3199 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3200 uint64_t len
, key
, offset
, version
;
3204 assert(channel
->metadata_cache
);
3206 memset(&request
, 0, sizeof(request
));
3208 /* send the metadata request to sessiond */
3209 switch (consumer_data
.type
) {
3210 case LTTNG_CONSUMER64_UST
:
3211 request
.bits_per_long
= 64;
3213 case LTTNG_CONSUMER32_UST
:
3214 request
.bits_per_long
= 32;
3217 request
.bits_per_long
= 0;
3221 request
.session_id
= channel
->session_id
;
3222 request
.session_id_per_pid
= channel
->session_id_per_pid
;
3224 * Request the application UID here so the metadata of that application can
3225 * be sent back. The channel UID corresponds to the user UID of the session
3226 * used for the rights on the stream file(s).
3228 request
.uid
= channel
->ust_app_uid
;
3229 request
.key
= channel
->key
;
3231 DBG("Sending metadata request to sessiond, session id %" PRIu64
3232 ", per-pid %" PRIu64
", app UID %u and channel key %" PRIu64
,
3233 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
3236 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
3238 health_code_update();
3240 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
3243 ERR("Asking metadata to sessiond");
3247 health_code_update();
3249 /* Receive the metadata from sessiond */
3250 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
3252 if (ret
!= sizeof(msg
)) {
3253 DBG("Consumer received unexpected message size %d (expects %zu)",
3255 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
3257 * The ret value might 0 meaning an orderly shutdown but this is ok
3258 * since the caller handles this.
3263 health_code_update();
3265 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
3266 /* No registry found */
3267 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3271 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
3272 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
3277 len
= msg
.u
.push_metadata
.len
;
3278 key
= msg
.u
.push_metadata
.key
;
3279 offset
= msg
.u
.push_metadata
.target_offset
;
3280 version
= msg
.u
.push_metadata
.version
;
3282 assert(key
== channel
->key
);
3284 DBG("No new metadata to receive for key %" PRIu64
, key
);
3287 health_code_update();
3289 /* Tell session daemon we are ready to receive the metadata. */
3290 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3291 LTTCOMM_CONSUMERD_SUCCESS
);
3292 if (ret
< 0 || len
== 0) {
3294 * Somehow, the session daemon is not responding anymore or there is
3295 * nothing to receive.
3300 health_code_update();
3302 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
3303 key
, offset
, len
, version
, channel
, timer
, wait
);
3306 * Only send the status msg if the sessiond is alive meaning a positive
3309 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
3314 health_code_update();
3316 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
3321 * Return the ustctl call for the get stream id.
3323 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
3324 uint64_t *stream_id
)
3329 return ustctl_get_stream_id(stream
->ustream
, stream_id
);