2 * Copyright (C) 2011 Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * SPDX-License-Identifier: GPL-2.0-only
12 #include <lttng/ust-ctl.h>
18 #include <sys/socket.h>
20 #include <sys/types.h>
23 #include <urcu/list.h>
28 #include <bin/lttng-consumerd/health-consumerd.h>
29 #include <common/common.h>
30 #include <common/sessiond-comm/sessiond-comm.h>
31 #include <common/relayd/relayd.h>
32 #include <common/compat/fcntl.h>
33 #include <common/compat/endian.h>
34 #include <common/consumer/consumer-metadata-cache.h>
35 #include <common/consumer/consumer-stream.h>
36 #include <common/consumer/consumer-timer.h>
37 #include <common/utils.h>
38 #include <common/index/index.h>
39 #include <common/consumer/consumer.h>
40 #include <common/optional.h>
42 #include "ust-consumer.h"
44 #define INT_MAX_STR_LEN 12 /* includes \0 */
46 extern struct lttng_consumer_global_data consumer_data
;
47 extern int consumer_poll_timeout
;
50 * Free channel object and all streams associated with it. This MUST be used
51 * only and only if the channel has _NEVER_ been added to the global channel
54 static void destroy_channel(struct lttng_consumer_channel
*channel
)
56 struct lttng_consumer_stream
*stream
, *stmp
;
60 DBG("UST consumer cleaning stream list");
62 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
67 cds_list_del(&stream
->send_node
);
68 ustctl_destroy_stream(stream
->ustream
);
69 lttng_trace_chunk_put(stream
->trace_chunk
);
74 * If a channel is available meaning that was created before the streams
78 lttng_ustconsumer_del_channel(channel
);
79 lttng_ustconsumer_free_channel(channel
);
82 if (channel
->trace_chunk
) {
83 lttng_trace_chunk_put(channel
->trace_chunk
);
90 * Add channel to internal consumer state.
92 * Returns 0 on success or else a negative value.
94 static int add_channel(struct lttng_consumer_channel
*channel
,
95 struct lttng_consumer_local_data
*ctx
)
102 if (ctx
->on_recv_channel
!= NULL
) {
103 ret
= ctx
->on_recv_channel(channel
);
105 ret
= consumer_add_channel(channel
, ctx
);
106 } else if (ret
< 0) {
107 /* Most likely an ENOMEM. */
108 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
112 ret
= consumer_add_channel(channel
, ctx
);
115 DBG("UST consumer channel added (key: %" PRIu64
")", channel
->key
);
122 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
123 * error value if applicable is set in it else it is kept untouched.
125 * Return NULL on error else the newly allocated stream object.
127 static struct lttng_consumer_stream
*allocate_stream(int cpu
, int key
,
128 struct lttng_consumer_channel
*channel
,
129 struct lttng_consumer_local_data
*ctx
, int *_alloc_ret
)
132 struct lttng_consumer_stream
*stream
= NULL
;
137 stream
= consumer_stream_create(
144 channel
->trace_chunk
,
149 if (stream
== NULL
) {
153 * We could not find the channel. Can happen if cpu hotplug
154 * happens while tearing down.
156 DBG3("Could not find channel");
161 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
167 consumer_stream_update_channel_attributes(stream
, channel
);
171 *_alloc_ret
= alloc_ret
;
177 * Send the given stream pointer to the corresponding thread.
179 * Returns 0 on success else a negative value.
181 static int send_stream_to_thread(struct lttng_consumer_stream
*stream
,
182 struct lttng_consumer_local_data
*ctx
)
185 struct lttng_pipe
*stream_pipe
;
187 /* Get the right pipe where the stream will be sent. */
188 if (stream
->metadata_flag
) {
189 consumer_add_metadata_stream(stream
);
190 stream_pipe
= ctx
->consumer_metadata_pipe
;
192 consumer_add_data_stream(stream
);
193 stream_pipe
= ctx
->consumer_data_pipe
;
197 * From this point on, the stream's ownership has been moved away from
198 * the channel and it becomes globally visible. Hence, remove it from
199 * the local stream list to prevent the stream from being both local and
202 stream
->globally_visible
= 1;
203 cds_list_del(&stream
->send_node
);
205 ret
= lttng_pipe_write(stream_pipe
, &stream
, sizeof(stream
));
207 ERR("Consumer write %s stream to pipe %d",
208 stream
->metadata_flag
? "metadata" : "data",
209 lttng_pipe_get_writefd(stream_pipe
));
210 if (stream
->metadata_flag
) {
211 consumer_del_stream_for_metadata(stream
);
213 consumer_del_stream_for_data(stream
);
223 int get_stream_shm_path(char *stream_shm_path
, const char *shm_path
, int cpu
)
225 char cpu_nr
[INT_MAX_STR_LEN
]; /* int max len */
228 strncpy(stream_shm_path
, shm_path
, PATH_MAX
);
229 stream_shm_path
[PATH_MAX
- 1] = '\0';
230 ret
= snprintf(cpu_nr
, INT_MAX_STR_LEN
, "%i", cpu
);
235 strncat(stream_shm_path
, cpu_nr
,
236 PATH_MAX
- strlen(stream_shm_path
) - 1);
243 * Create streams for the given channel using liblttng-ust-ctl.
244 * The channel lock must be acquired by the caller.
246 * Return 0 on success else a negative value.
248 static int create_ust_streams(struct lttng_consumer_channel
*channel
,
249 struct lttng_consumer_local_data
*ctx
)
252 struct ustctl_consumer_stream
*ustream
;
253 struct lttng_consumer_stream
*stream
;
254 pthread_mutex_t
*current_stream_lock
= NULL
;
260 * While a stream is available from ustctl. When NULL is returned, we've
261 * reached the end of the possible stream for the channel.
263 while ((ustream
= ustctl_create_stream(channel
->uchan
, cpu
))) {
265 int ust_metadata_pipe
[2];
267 health_code_update();
269 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& channel
->monitor
) {
270 ret
= utils_create_pipe_cloexec_nonblock(ust_metadata_pipe
);
272 ERR("Create ust metadata poll pipe");
275 wait_fd
= ust_metadata_pipe
[0];
277 wait_fd
= ustctl_stream_get_wait_fd(ustream
);
280 /* Allocate consumer stream object. */
281 stream
= allocate_stream(cpu
, wait_fd
, channel
, ctx
, &ret
);
285 stream
->ustream
= ustream
;
287 * Store it so we can save multiple function calls afterwards since
288 * this value is used heavily in the stream threads. This is UST
289 * specific so this is why it's done after allocation.
291 stream
->wait_fd
= wait_fd
;
294 * Increment channel refcount since the channel reference has now been
295 * assigned in the allocation process above.
297 if (stream
->chan
->monitor
) {
298 uatomic_inc(&stream
->chan
->refcount
);
301 pthread_mutex_lock(&stream
->lock
);
302 current_stream_lock
= &stream
->lock
;
304 * Order is important this is why a list is used. On error, the caller
305 * should clean this list.
307 cds_list_add_tail(&stream
->send_node
, &channel
->streams
.head
);
309 ret
= ustctl_get_max_subbuf_size(stream
->ustream
,
310 &stream
->max_sb_size
);
312 ERR("ustctl_get_max_subbuf_size failed for stream %s",
317 /* Do actions once stream has been received. */
318 if (ctx
->on_recv_stream
) {
319 ret
= ctx
->on_recv_stream(stream
);
325 DBG("UST consumer add stream %s (key: %" PRIu64
") with relayd id %" PRIu64
,
326 stream
->name
, stream
->key
, stream
->relayd_stream_id
);
328 /* Set next CPU stream. */
329 channel
->streams
.count
= ++cpu
;
331 /* Keep stream reference when creating metadata. */
332 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
333 channel
->metadata_stream
= stream
;
334 if (channel
->monitor
) {
335 /* Set metadata poll pipe if we created one */
336 memcpy(stream
->ust_metadata_poll_pipe
,
338 sizeof(ust_metadata_pipe
));
341 pthread_mutex_unlock(&stream
->lock
);
342 current_stream_lock
= NULL
;
349 if (current_stream_lock
) {
350 pthread_mutex_unlock(current_stream_lock
);
356 * create_posix_shm is never called concurrently within a process.
359 int create_posix_shm(void)
361 char tmp_name
[NAME_MAX
];
364 ret
= snprintf(tmp_name
, NAME_MAX
, "/ust-shm-consumer-%d", getpid());
370 * Allocate shm, and immediately unlink its shm oject, keeping
371 * only the file descriptor as a reference to the object.
372 * We specifically do _not_ use the / at the beginning of the
373 * pathname so that some OS implementations can keep it local to
374 * the process (POSIX leaves this implementation-defined).
376 shmfd
= shm_open(tmp_name
, O_CREAT
| O_EXCL
| O_RDWR
, 0700);
381 ret
= shm_unlink(tmp_name
);
382 if (ret
< 0 && errno
!= ENOENT
) {
383 PERROR("shm_unlink");
384 goto error_shm_release
;
397 static int open_ust_stream_fd(struct lttng_consumer_channel
*channel
, int cpu
,
398 const struct lttng_credentials
*session_credentials
)
400 char shm_path
[PATH_MAX
];
403 if (!channel
->shm_path
[0]) {
404 return create_posix_shm();
406 ret
= get_stream_shm_path(shm_path
, channel
->shm_path
, cpu
);
410 return run_as_open(shm_path
,
411 O_RDWR
| O_CREAT
| O_EXCL
, S_IRUSR
| S_IWUSR
,
412 session_credentials
->uid
, session_credentials
->gid
);
419 * Create an UST channel with the given attributes and send it to the session
420 * daemon using the ust ctl API.
422 * Return 0 on success or else a negative value.
424 static int create_ust_channel(struct lttng_consumer_channel
*channel
,
425 struct ustctl_consumer_channel_attr
*attr
,
426 struct ustctl_consumer_channel
**ust_chanp
)
428 int ret
, nr_stream_fds
, i
, j
;
430 struct ustctl_consumer_channel
*ust_channel
;
435 assert(channel
->buffer_credentials
.is_set
);
437 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
438 "subbuf_size: %" PRIu64
", num_subbuf: %" PRIu64
", "
439 "switch_timer_interval: %u, read_timer_interval: %u, "
440 "output: %d, type: %d", attr
->overwrite
, attr
->subbuf_size
,
441 attr
->num_subbuf
, attr
->switch_timer_interval
,
442 attr
->read_timer_interval
, attr
->output
, attr
->type
);
444 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
)
447 nr_stream_fds
= ustctl_get_nr_stream_per_channel();
448 stream_fds
= zmalloc(nr_stream_fds
* sizeof(*stream_fds
));
453 for (i
= 0; i
< nr_stream_fds
; i
++) {
454 stream_fds
[i
] = open_ust_stream_fd(channel
, i
,
455 &channel
->buffer_credentials
.value
);
456 if (stream_fds
[i
] < 0) {
461 ust_channel
= ustctl_create_channel(attr
, stream_fds
, nr_stream_fds
);
466 channel
->nr_stream_fds
= nr_stream_fds
;
467 channel
->stream_fds
= stream_fds
;
468 *ust_chanp
= ust_channel
;
474 for (j
= i
- 1; j
>= 0; j
--) {
477 closeret
= close(stream_fds
[j
]);
481 if (channel
->shm_path
[0]) {
482 char shm_path
[PATH_MAX
];
484 closeret
= get_stream_shm_path(shm_path
,
485 channel
->shm_path
, j
);
487 ERR("Cannot get stream shm path");
489 closeret
= run_as_unlink(shm_path
,
490 channel
->buffer_credentials
.value
.uid
,
491 channel
->buffer_credentials
.value
.gid
);
493 PERROR("unlink %s", shm_path
);
497 /* Try to rmdir all directories under shm_path root. */
498 if (channel
->root_shm_path
[0]) {
499 (void) run_as_rmdir_recursive(channel
->root_shm_path
,
500 channel
->buffer_credentials
.value
.uid
,
501 channel
->buffer_credentials
.value
.gid
,
502 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG
);
510 * Send a single given stream to the session daemon using the sock.
512 * Return 0 on success else a negative value.
514 static int send_sessiond_stream(int sock
, struct lttng_consumer_stream
*stream
)
521 DBG("UST consumer sending stream %" PRIu64
" to sessiond", stream
->key
);
523 /* Send stream to session daemon. */
524 ret
= ustctl_send_stream_to_sessiond(sock
, stream
->ustream
);
534 * Send channel to sessiond and relayd if applicable.
536 * Return 0 on success or else a negative value.
538 static int send_channel_to_sessiond_and_relayd(int sock
,
539 struct lttng_consumer_channel
*channel
,
540 struct lttng_consumer_local_data
*ctx
, int *relayd_error
)
542 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
543 struct lttng_consumer_stream
*stream
;
544 uint64_t net_seq_idx
= -1ULL;
550 DBG("UST consumer sending channel %s to sessiond", channel
->name
);
552 if (channel
->relayd_id
!= (uint64_t) -1ULL) {
553 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
555 health_code_update();
557 /* Try to send the stream to the relayd if one is available. */
558 DBG("Sending stream %" PRIu64
" of channel \"%s\" to relayd",
559 stream
->key
, channel
->name
);
560 ret
= consumer_send_relayd_stream(stream
, stream
->chan
->pathname
);
563 * Flag that the relayd was the problem here probably due to a
564 * communicaton error on the socket.
569 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
571 if (net_seq_idx
== -1ULL) {
572 net_seq_idx
= stream
->net_seq_idx
;
577 /* Inform sessiond that we are about to send channel and streams. */
578 ret
= consumer_send_status_msg(sock
, ret_code
);
579 if (ret
< 0 || ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
581 * Either the session daemon is not responding or the relayd died so we
587 /* Send channel to sessiond. */
588 ret
= ustctl_send_channel_to_sessiond(sock
, channel
->uchan
);
593 ret
= ustctl_channel_close_wakeup_fd(channel
->uchan
);
598 /* The channel was sent successfully to the sessiond at this point. */
599 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
601 health_code_update();
603 /* Send stream to session daemon. */
604 ret
= send_sessiond_stream(sock
, stream
);
610 /* Tell sessiond there is no more stream. */
611 ret
= ustctl_send_stream_to_sessiond(sock
, NULL
);
616 DBG("UST consumer NULL stream sent to sessiond");
621 if (ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
628 * Creates a channel and streams and add the channel it to the channel internal
629 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
632 * Return 0 on success or else, a negative value is returned and the channel
633 * MUST be destroyed by consumer_del_channel().
635 static int ask_channel(struct lttng_consumer_local_data
*ctx
,
636 struct lttng_consumer_channel
*channel
,
637 struct ustctl_consumer_channel_attr
*attr
)
646 * This value is still used by the kernel consumer since for the kernel,
647 * the stream ownership is not IN the consumer so we need to have the
648 * number of left stream that needs to be initialized so we can know when
649 * to delete the channel (see consumer.c).
651 * As for the user space tracer now, the consumer creates and sends the
652 * stream to the session daemon which only sends them to the application
653 * once every stream of a channel is received making this value useless
654 * because we they will be added to the poll thread before the application
655 * receives them. This ensures that a stream can not hang up during
656 * initilization of a channel.
658 channel
->nb_init_stream_left
= 0;
660 /* The reply msg status is handled in the following call. */
661 ret
= create_ust_channel(channel
, attr
, &channel
->uchan
);
666 channel
->wait_fd
= ustctl_channel_get_wait_fd(channel
->uchan
);
669 * For the snapshots (no monitor), we create the metadata streams
670 * on demand, not during the channel creation.
672 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& !channel
->monitor
) {
677 /* Open all streams for this channel. */
678 pthread_mutex_lock(&channel
->lock
);
679 ret
= create_ust_streams(channel
, ctx
);
680 pthread_mutex_unlock(&channel
->lock
);
690 * Send all stream of a channel to the right thread handling it.
692 * On error, return a negative value else 0 on success.
694 static int send_streams_to_thread(struct lttng_consumer_channel
*channel
,
695 struct lttng_consumer_local_data
*ctx
)
698 struct lttng_consumer_stream
*stream
, *stmp
;
703 /* Send streams to the corresponding thread. */
704 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
707 health_code_update();
709 /* Sending the stream to the thread. */
710 ret
= send_stream_to_thread(stream
, ctx
);
713 * If we are unable to send the stream to the thread, there is
714 * a big problem so just stop everything.
725 * Flush channel's streams using the given key to retrieve the channel.
727 * Return 0 on success else an LTTng error code.
729 static int flush_channel(uint64_t chan_key
)
732 struct lttng_consumer_channel
*channel
;
733 struct lttng_consumer_stream
*stream
;
735 struct lttng_ht_iter iter
;
737 DBG("UST consumer flush channel key %" PRIu64
, chan_key
);
740 channel
= consumer_find_channel(chan_key
);
742 ERR("UST consumer flush channel %" PRIu64
" not found", chan_key
);
743 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
747 ht
= consumer_data
.stream_per_chan_id_ht
;
749 /* For each stream of the channel id, flush it. */
750 cds_lfht_for_each_entry_duplicate(ht
->ht
,
751 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
752 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
754 health_code_update();
756 pthread_mutex_lock(&stream
->lock
);
759 * Protect against concurrent teardown of a stream.
761 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
765 if (!stream
->quiescent
) {
766 ustctl_flush_buffer(stream
->ustream
, 0);
767 stream
->quiescent
= true;
770 pthread_mutex_unlock(&stream
->lock
);
778 * Clear quiescent state from channel's streams using the given key to
779 * retrieve the channel.
781 * Return 0 on success else an LTTng error code.
783 static int clear_quiescent_channel(uint64_t chan_key
)
786 struct lttng_consumer_channel
*channel
;
787 struct lttng_consumer_stream
*stream
;
789 struct lttng_ht_iter iter
;
791 DBG("UST consumer clear quiescent channel key %" PRIu64
, chan_key
);
794 channel
= consumer_find_channel(chan_key
);
796 ERR("UST consumer clear quiescent channel %" PRIu64
" not found", chan_key
);
797 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
801 ht
= consumer_data
.stream_per_chan_id_ht
;
803 /* For each stream of the channel id, clear quiescent state. */
804 cds_lfht_for_each_entry_duplicate(ht
->ht
,
805 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
806 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
808 health_code_update();
810 pthread_mutex_lock(&stream
->lock
);
811 stream
->quiescent
= false;
812 pthread_mutex_unlock(&stream
->lock
);
820 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
822 * Return 0 on success else an LTTng error code.
824 static int close_metadata(uint64_t chan_key
)
827 struct lttng_consumer_channel
*channel
;
828 unsigned int channel_monitor
;
830 DBG("UST consumer close metadata key %" PRIu64
, chan_key
);
832 channel
= consumer_find_channel(chan_key
);
835 * This is possible if the metadata thread has issue a delete because
836 * the endpoint point of the stream hung up. There is no way the
837 * session daemon can know about it thus use a DBG instead of an actual
840 DBG("UST consumer close metadata %" PRIu64
" not found", chan_key
);
841 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
845 pthread_mutex_lock(&consumer_data
.lock
);
846 pthread_mutex_lock(&channel
->lock
);
847 channel_monitor
= channel
->monitor
;
848 if (cds_lfht_is_node_deleted(&channel
->node
.node
)) {
852 lttng_ustconsumer_close_metadata(channel
);
853 pthread_mutex_unlock(&channel
->lock
);
854 pthread_mutex_unlock(&consumer_data
.lock
);
857 * The ownership of a metadata channel depends on the type of
858 * session to which it belongs. In effect, the monitor flag is checked
859 * to determine if this metadata channel is in "snapshot" mode or not.
861 * In the non-snapshot case, the metadata channel is created along with
862 * a single stream which will remain present until the metadata channel
863 * is destroyed (on the destruction of its session). In this case, the
864 * metadata stream in "monitored" by the metadata poll thread and holds
865 * the ownership of its channel.
867 * Closing the metadata will cause the metadata stream's "metadata poll
868 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
869 * thread which will teardown the metadata stream which, in return,
870 * deletes the metadata channel.
872 * In the snapshot case, the metadata stream is created and destroyed
873 * on every snapshot record. Since the channel doesn't have an owner
874 * other than the session daemon, it is safe to destroy it immediately
875 * on reception of the CLOSE_METADATA command.
877 if (!channel_monitor
) {
879 * The channel and consumer_data locks must be
880 * released before this call since consumer_del_channel
881 * re-acquires the channel and consumer_data locks to teardown
882 * the channel and queue its reclamation by the "call_rcu"
885 consumer_del_channel(channel
);
890 pthread_mutex_unlock(&channel
->lock
);
891 pthread_mutex_unlock(&consumer_data
.lock
);
897 * RCU read side lock MUST be acquired before calling this function.
899 * Return 0 on success else an LTTng error code.
901 static int setup_metadata(struct lttng_consumer_local_data
*ctx
, uint64_t key
)
904 struct lttng_consumer_channel
*metadata
;
906 DBG("UST consumer setup metadata key %" PRIu64
, key
);
908 metadata
= consumer_find_channel(key
);
910 ERR("UST consumer push metadata %" PRIu64
" not found", key
);
911 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
916 * In no monitor mode, the metadata channel has no stream(s) so skip the
917 * ownership transfer to the metadata thread.
919 if (!metadata
->monitor
) {
920 DBG("Metadata channel in no monitor");
926 * Send metadata stream to relayd if one available. Availability is
927 * known if the stream is still in the list of the channel.
929 if (cds_list_empty(&metadata
->streams
.head
)) {
930 ERR("Metadata channel key %" PRIu64
", no stream available.", key
);
931 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
932 goto error_no_stream
;
935 /* Send metadata stream to relayd if needed. */
936 if (metadata
->metadata_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
937 ret
= consumer_send_relayd_stream(metadata
->metadata_stream
,
940 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
943 ret
= consumer_send_relayd_streams_sent(
944 metadata
->metadata_stream
->net_seq_idx
);
946 ret
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
952 * Ownership of metadata stream is passed along. Freeing is handled by
955 ret
= send_streams_to_thread(metadata
, ctx
);
958 * If we are unable to send the stream to the thread, there is
959 * a big problem so just stop everything.
961 ret
= LTTCOMM_CONSUMERD_FATAL
;
962 goto send_streams_error
;
964 /* List MUST be empty after or else it could be reused. */
965 assert(cds_list_empty(&metadata
->streams
.head
));
972 * Delete metadata channel on error. At this point, the metadata stream can
973 * NOT be monitored by the metadata thread thus having the guarantee that
974 * the stream is still in the local stream list of the channel. This call
975 * will make sure to clean that list.
977 consumer_stream_destroy(metadata
->metadata_stream
, NULL
);
978 cds_list_del(&metadata
->metadata_stream
->send_node
);
979 metadata
->metadata_stream
= NULL
;
987 * Snapshot the whole metadata.
988 * RCU read-side lock must be held by the caller.
990 * Returns 0 on success, < 0 on error
992 static int snapshot_metadata(struct lttng_consumer_channel
*metadata_channel
,
993 uint64_t key
, char *path
, uint64_t relayd_id
,
994 struct lttng_consumer_local_data
*ctx
)
997 struct lttng_consumer_stream
*metadata_stream
;
1002 DBG("UST consumer snapshot metadata with key %" PRIu64
" at path %s",
1007 assert(!metadata_channel
->monitor
);
1009 health_code_update();
1012 * Ask the sessiond if we have new metadata waiting and update the
1013 * consumer metadata cache.
1015 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 1);
1020 health_code_update();
1023 * The metadata stream is NOT created in no monitor mode when the channel
1024 * is created on a sessiond ask channel command.
1026 ret
= create_ust_streams(metadata_channel
, ctx
);
1031 metadata_stream
= metadata_channel
->metadata_stream
;
1032 assert(metadata_stream
);
1034 pthread_mutex_lock(&metadata_stream
->lock
);
1035 if (relayd_id
!= (uint64_t) -1ULL) {
1036 metadata_stream
->net_seq_idx
= relayd_id
;
1037 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
1039 ret
= consumer_stream_create_output_files(metadata_stream
,
1042 pthread_mutex_unlock(&metadata_stream
->lock
);
1048 health_code_update();
1050 ret
= lttng_consumer_read_subbuffer(metadata_stream
, ctx
, true);
1058 * Clean up the stream completly because the next snapshot will use a new
1061 consumer_stream_destroy(metadata_stream
, NULL
);
1062 cds_list_del(&metadata_stream
->send_node
);
1063 metadata_channel
->metadata_stream
= NULL
;
1071 int get_current_subbuf_addr(struct lttng_consumer_stream
*stream
,
1075 unsigned long mmap_offset
;
1076 const char *mmap_base
;
1078 mmap_base
= ustctl_get_mmap_base(stream
->ustream
);
1080 ERR("Failed to get mmap base for stream `%s`",
1086 ret
= ustctl_get_mmap_read_offset(stream
->ustream
, &mmap_offset
);
1088 ERR("Failed to get mmap offset for stream `%s`", stream
->name
);
1093 *addr
= mmap_base
+ mmap_offset
;
1100 * Take a snapshot of all the stream of a channel.
1101 * RCU read-side lock and the channel lock must be held by the caller.
1103 * Returns 0 on success, < 0 on error
1105 static int snapshot_channel(struct lttng_consumer_channel
*channel
,
1106 uint64_t key
, char *path
, uint64_t relayd_id
,
1107 uint64_t nb_packets_per_stream
,
1108 struct lttng_consumer_local_data
*ctx
)
1111 unsigned use_relayd
= 0;
1112 unsigned long consumed_pos
, produced_pos
;
1113 struct lttng_consumer_stream
*stream
;
1120 if (relayd_id
!= (uint64_t) -1ULL) {
1124 assert(!channel
->monitor
);
1125 DBG("UST consumer snapshot channel %" PRIu64
, key
);
1127 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
1128 health_code_update();
1130 /* Lock stream because we are about to change its state. */
1131 pthread_mutex_lock(&stream
->lock
);
1132 assert(channel
->trace_chunk
);
1133 if (!lttng_trace_chunk_get(channel
->trace_chunk
)) {
1135 * Can't happen barring an internal error as the channel
1136 * holds a reference to the trace chunk.
1138 ERR("Failed to acquire reference to channel's trace chunk");
1142 assert(!stream
->trace_chunk
);
1143 stream
->trace_chunk
= channel
->trace_chunk
;
1145 stream
->net_seq_idx
= relayd_id
;
1148 ret
= consumer_send_relayd_stream(stream
, path
);
1153 ret
= consumer_stream_create_output_files(stream
,
1158 DBG("UST consumer snapshot stream (%" PRIu64
")",
1163 * If tracing is active, we want to perform a "full" buffer flush.
1164 * Else, if quiescent, it has already been done by the prior stop.
1166 if (!stream
->quiescent
) {
1167 ustctl_flush_buffer(stream
->ustream
, 0);
1170 ret
= lttng_ustconsumer_take_snapshot(stream
);
1172 ERR("Taking UST snapshot");
1176 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
1178 ERR("Produced UST snapshot position");
1182 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1184 ERR("Consumerd UST snapshot position");
1189 * The original value is sent back if max stream size is larger than
1190 * the possible size of the snapshot. Also, we assume that the session
1191 * daemon should never send a maximum stream size that is lower than
1194 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
1195 produced_pos
, nb_packets_per_stream
,
1196 stream
->max_sb_size
);
1198 while ((long) (consumed_pos
- produced_pos
) < 0) {
1200 unsigned long len
, padded_len
;
1201 const char *subbuf_addr
;
1202 struct lttng_buffer_view subbuf_view
;
1204 health_code_update();
1206 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1208 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1210 if (ret
!= -EAGAIN
) {
1211 PERROR("ustctl_get_subbuf snapshot");
1212 goto error_close_stream
;
1214 DBG("UST consumer get subbuf failed. Skipping it.");
1215 consumed_pos
+= stream
->max_sb_size
;
1216 stream
->chan
->lost_packets
++;
1220 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1222 ERR("Snapshot ustctl_get_subbuf_size");
1223 goto error_put_subbuf
;
1226 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1228 ERR("Snapshot ustctl_get_padded_subbuf_size");
1229 goto error_put_subbuf
;
1232 ret
= get_current_subbuf_addr(stream
, &subbuf_addr
);
1234 goto error_put_subbuf
;
1237 subbuf_view
= lttng_buffer_view_init(
1238 subbuf_addr
, 0, padded_len
);
1239 read_len
= lttng_consumer_on_read_subbuffer_mmap(
1240 stream
, &subbuf_view
, padded_len
- len
);
1242 if (read_len
!= len
) {
1244 goto error_put_subbuf
;
1247 if (read_len
!= padded_len
) {
1249 goto error_put_subbuf
;
1253 ret
= ustctl_put_subbuf(stream
->ustream
);
1255 ERR("Snapshot ustctl_put_subbuf");
1256 goto error_close_stream
;
1258 consumed_pos
+= stream
->max_sb_size
;
1261 /* Simply close the stream so we can use it on the next snapshot. */
1262 consumer_stream_close(stream
);
1263 pthread_mutex_unlock(&stream
->lock
);
1270 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1271 ERR("Snapshot ustctl_put_subbuf");
1274 consumer_stream_close(stream
);
1276 pthread_mutex_unlock(&stream
->lock
);
1282 * Receive the metadata updates from the sessiond. Supports receiving
1283 * overlapping metadata, but is needs to always belong to a contiguous
1284 * range starting from 0.
1285 * Be careful about the locks held when calling this function: it needs
1286 * the metadata cache flush to concurrently progress in order to
1289 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1290 uint64_t len
, uint64_t version
,
1291 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
1293 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1296 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1298 metadata_str
= zmalloc(len
* sizeof(char));
1299 if (!metadata_str
) {
1300 PERROR("zmalloc metadata string");
1301 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1305 health_code_update();
1307 /* Receive metadata string. */
1308 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1310 /* Session daemon is dead so return gracefully. */
1315 health_code_update();
1317 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1318 ret
= consumer_metadata_cache_write(channel
, offset
, len
, version
,
1321 /* Unable to handle metadata. Notify session daemon. */
1322 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1324 * Skip metadata flush on write error since the offset and len might
1325 * not have been updated which could create an infinite loop below when
1326 * waiting for the metadata cache to be flushed.
1328 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1331 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1336 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1337 DBG("Waiting for metadata to be flushed");
1339 health_code_update();
1341 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1351 * Receive command from session daemon and process it.
1353 * Return 1 on success else a negative value or 0.
1355 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1356 int sock
, struct pollfd
*consumer_sockpoll
)
1359 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1360 struct lttcomm_consumer_msg msg
;
1361 struct lttng_consumer_channel
*channel
= NULL
;
1363 health_code_update();
1365 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1366 if (ret
!= sizeof(msg
)) {
1367 DBG("Consumer received unexpected message size %zd (expects %zu)",
1370 * The ret value might 0 meaning an orderly shutdown but this is ok
1371 * since the caller handles this.
1374 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1380 health_code_update();
1383 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1385 health_code_update();
1387 /* relayd needs RCU read-side lock */
1390 switch (msg
.cmd_type
) {
1391 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1393 /* Session daemon status message are handled in the following call. */
1394 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1395 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
1396 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
1397 msg
.u
.relayd_sock
.relayd_session_id
);
1400 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1402 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1403 struct consumer_relayd_sock_pair
*relayd
;
1405 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1407 /* Get relayd reference if exists. */
1408 relayd
= consumer_find_relayd(index
);
1409 if (relayd
== NULL
) {
1410 DBG("Unable to find relayd %" PRIu64
, index
);
1411 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1415 * Each relayd socket pair has a refcount of stream attached to it
1416 * which tells if the relayd is still active or not depending on the
1419 * This will set the destroy flag of the relayd object and destroy it
1420 * if the refcount reaches zero when called.
1422 * The destroy can happen either here or when a stream fd hangs up.
1425 consumer_flag_relayd_for_destroy(relayd
);
1428 goto end_msg_sessiond
;
1430 case LTTNG_CONSUMER_UPDATE_STREAM
:
1435 case LTTNG_CONSUMER_DATA_PENDING
:
1437 int ret
, is_data_pending
;
1438 uint64_t id
= msg
.u
.data_pending
.session_id
;
1440 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1442 is_data_pending
= consumer_data_pending(id
);
1444 /* Send back returned value to session daemon */
1445 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1446 sizeof(is_data_pending
));
1448 DBG("Error when sending the data pending ret code: %d", ret
);
1453 * No need to send back a status message since the data pending
1454 * returned value is the response.
1458 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1461 struct ustctl_consumer_channel_attr attr
;
1462 const uint64_t chunk_id
= msg
.u
.ask_channel
.chunk_id
.value
;
1463 const struct lttng_credentials buffer_credentials
= {
1464 .uid
= msg
.u
.ask_channel
.buffer_credentials
.uid
,
1465 .gid
= msg
.u
.ask_channel
.buffer_credentials
.gid
,
1468 /* Create a plain object and reserve a channel key. */
1469 channel
= consumer_allocate_channel(
1470 msg
.u
.ask_channel
.key
,
1471 msg
.u
.ask_channel
.session_id
,
1472 msg
.u
.ask_channel
.chunk_id
.is_set
?
1474 msg
.u
.ask_channel
.pathname
,
1475 msg
.u
.ask_channel
.name
,
1476 msg
.u
.ask_channel
.relayd_id
,
1477 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1478 msg
.u
.ask_channel
.tracefile_size
,
1479 msg
.u
.ask_channel
.tracefile_count
,
1480 msg
.u
.ask_channel
.session_id_per_pid
,
1481 msg
.u
.ask_channel
.monitor
,
1482 msg
.u
.ask_channel
.live_timer_interval
,
1483 msg
.u
.ask_channel
.is_live
,
1484 msg
.u
.ask_channel
.root_shm_path
,
1485 msg
.u
.ask_channel
.shm_path
);
1487 goto end_channel_error
;
1490 LTTNG_OPTIONAL_SET(&channel
->buffer_credentials
,
1491 buffer_credentials
);
1494 * Assign UST application UID to the channel. This value is ignored for
1495 * per PID buffers. This is specific to UST thus setting this after the
1498 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1500 /* Build channel attributes from received message. */
1501 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1502 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1503 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1504 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1505 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1506 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1507 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1508 attr
.blocking_timeout
= msg
.u
.ask_channel
.blocking_timeout
;
1510 /* Match channel buffer type to the UST abi. */
1511 switch (msg
.u
.ask_channel
.output
) {
1512 case LTTNG_EVENT_MMAP
:
1514 attr
.output
= LTTNG_UST_MMAP
;
1518 /* Translate and save channel type. */
1519 switch (msg
.u
.ask_channel
.type
) {
1520 case LTTNG_UST_CHAN_PER_CPU
:
1521 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1522 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1524 * Set refcount to 1 for owner. Below, we will
1525 * pass ownership to the
1526 * consumer_thread_channel_poll() thread.
1528 channel
->refcount
= 1;
1530 case LTTNG_UST_CHAN_METADATA
:
1531 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1532 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1539 health_code_update();
1541 ret
= ask_channel(ctx
, channel
, &attr
);
1543 goto end_channel_error
;
1546 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1547 ret
= consumer_metadata_cache_allocate(channel
);
1549 ERR("Allocating metadata cache");
1550 goto end_channel_error
;
1552 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1553 attr
.switch_timer_interval
= 0;
1555 int monitor_start_ret
;
1557 consumer_timer_live_start(channel
,
1558 msg
.u
.ask_channel
.live_timer_interval
);
1559 monitor_start_ret
= consumer_timer_monitor_start(
1561 msg
.u
.ask_channel
.monitor_timer_interval
);
1562 if (monitor_start_ret
< 0) {
1563 ERR("Starting channel monitoring timer failed");
1564 goto end_channel_error
;
1568 health_code_update();
1571 * Add the channel to the internal state AFTER all streams were created
1572 * and successfully sent to session daemon. This way, all streams must
1573 * be ready before this channel is visible to the threads.
1574 * If add_channel succeeds, ownership of the channel is
1575 * passed to consumer_thread_channel_poll().
1577 ret
= add_channel(channel
, ctx
);
1579 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1580 if (channel
->switch_timer_enabled
== 1) {
1581 consumer_timer_switch_stop(channel
);
1583 consumer_metadata_cache_destroy(channel
);
1585 if (channel
->live_timer_enabled
== 1) {
1586 consumer_timer_live_stop(channel
);
1588 if (channel
->monitor_timer_enabled
== 1) {
1589 consumer_timer_monitor_stop(channel
);
1591 goto end_channel_error
;
1594 health_code_update();
1597 * Channel and streams are now created. Inform the session daemon that
1598 * everything went well and should wait to receive the channel and
1599 * streams with ustctl API.
1601 ret
= consumer_send_status_channel(sock
, channel
);
1604 * There is probably a problem on the socket.
1611 case LTTNG_CONSUMER_GET_CHANNEL
:
1613 int ret
, relayd_err
= 0;
1614 uint64_t key
= msg
.u
.get_channel
.key
;
1615 struct lttng_consumer_channel
*channel
;
1617 channel
= consumer_find_channel(key
);
1619 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1620 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1621 goto end_get_channel
;
1624 health_code_update();
1626 /* Send the channel to sessiond (and relayd, if applicable). */
1627 ret
= send_channel_to_sessiond_and_relayd(sock
, channel
, ctx
,
1632 * We were unable to send to the relayd the stream so avoid
1633 * sending back a fatal error to the thread since this is OK
1634 * and the consumer can continue its work. The above call
1635 * has sent the error status message to the sessiond.
1637 goto end_get_channel_nosignal
;
1640 * The communicaton was broken hence there is a bad state between
1641 * the consumer and sessiond so stop everything.
1643 goto error_get_channel_fatal
;
1646 health_code_update();
1649 * In no monitor mode, the streams ownership is kept inside the channel
1650 * so don't send them to the data thread.
1652 if (!channel
->monitor
) {
1653 goto end_get_channel
;
1656 ret
= send_streams_to_thread(channel
, ctx
);
1659 * If we are unable to send the stream to the thread, there is
1660 * a big problem so just stop everything.
1662 goto error_get_channel_fatal
;
1664 /* List MUST be empty after or else it could be reused. */
1665 assert(cds_list_empty(&channel
->streams
.head
));
1667 goto end_msg_sessiond
;
1668 error_get_channel_fatal
:
1670 end_get_channel_nosignal
:
1673 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1675 uint64_t key
= msg
.u
.destroy_channel
.key
;
1678 * Only called if streams have not been sent to stream
1679 * manager thread. However, channel has been sent to
1680 * channel manager thread.
1682 notify_thread_del_channel(ctx
, key
);
1683 goto end_msg_sessiond
;
1685 case LTTNG_CONSUMER_CLOSE_METADATA
:
1689 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1694 goto end_msg_sessiond
;
1696 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1700 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1705 goto end_msg_sessiond
;
1707 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL
:
1711 ret
= clear_quiescent_channel(
1712 msg
.u
.clear_quiescent_channel
.key
);
1717 goto end_msg_sessiond
;
1719 case LTTNG_CONSUMER_PUSH_METADATA
:
1722 uint64_t len
= msg
.u
.push_metadata
.len
;
1723 uint64_t key
= msg
.u
.push_metadata
.key
;
1724 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1725 uint64_t version
= msg
.u
.push_metadata
.version
;
1726 struct lttng_consumer_channel
*channel
;
1728 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1731 channel
= consumer_find_channel(key
);
1734 * This is possible if the metadata creation on the consumer side
1735 * is in flight vis-a-vis a concurrent push metadata from the
1736 * session daemon. Simply return that the channel failed and the
1737 * session daemon will handle that message correctly considering
1738 * that this race is acceptable thus the DBG() statement here.
1740 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1741 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1742 goto end_push_metadata_msg_sessiond
;
1745 health_code_update();
1749 * There is nothing to receive. We have simply
1750 * checked whether the channel can be found.
1752 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1753 goto end_push_metadata_msg_sessiond
;
1756 /* Tell session daemon we are ready to receive the metadata. */
1757 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1759 /* Somehow, the session daemon is not responding anymore. */
1760 goto error_push_metadata_fatal
;
1763 health_code_update();
1765 /* Wait for more data. */
1766 health_poll_entry();
1767 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1770 goto error_push_metadata_fatal
;
1773 health_code_update();
1775 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1776 len
, version
, channel
, 0, 1);
1778 /* error receiving from sessiond */
1779 goto error_push_metadata_fatal
;
1782 goto end_push_metadata_msg_sessiond
;
1784 end_push_metadata_msg_sessiond
:
1785 goto end_msg_sessiond
;
1786 error_push_metadata_fatal
:
1789 case LTTNG_CONSUMER_SETUP_METADATA
:
1793 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1797 goto end_msg_sessiond
;
1799 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1801 struct lttng_consumer_channel
*channel
;
1802 uint64_t key
= msg
.u
.snapshot_channel
.key
;
1804 channel
= consumer_find_channel(key
);
1806 DBG("UST snapshot channel not found for key %" PRIu64
, key
);
1807 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1809 if (msg
.u
.snapshot_channel
.metadata
) {
1810 ret
= snapshot_metadata(channel
, key
,
1811 msg
.u
.snapshot_channel
.pathname
,
1812 msg
.u
.snapshot_channel
.relayd_id
,
1815 ERR("Snapshot metadata failed");
1816 ret_code
= LTTCOMM_CONSUMERD_SNAPSHOT_FAILED
;
1819 ret
= snapshot_channel(channel
, key
,
1820 msg
.u
.snapshot_channel
.pathname
,
1821 msg
.u
.snapshot_channel
.relayd_id
,
1822 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
1825 ERR("Snapshot channel failed");
1826 ret_code
= LTTCOMM_CONSUMERD_SNAPSHOT_FAILED
;
1830 health_code_update();
1831 ret
= consumer_send_status_msg(sock
, ret_code
);
1833 /* Somehow, the session daemon is not responding anymore. */
1836 health_code_update();
1839 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
1842 uint64_t discarded_events
;
1843 struct lttng_ht_iter iter
;
1844 struct lttng_ht
*ht
;
1845 struct lttng_consumer_stream
*stream
;
1846 uint64_t id
= msg
.u
.discarded_events
.session_id
;
1847 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
1849 DBG("UST consumer discarded events command for session id %"
1852 pthread_mutex_lock(&consumer_data
.lock
);
1854 ht
= consumer_data
.stream_list_ht
;
1857 * We only need a reference to the channel, but they are not
1858 * directly indexed, so we just use the first matching stream
1859 * to extract the information we need, we default to 0 if not
1860 * found (no events are dropped if the channel is not yet in
1863 discarded_events
= 0;
1864 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1865 ht
->hash_fct(&id
, lttng_ht_seed
),
1867 &iter
.iter
, stream
, node_session_id
.node
) {
1868 if (stream
->chan
->key
== key
) {
1869 discarded_events
= stream
->chan
->discarded_events
;
1873 pthread_mutex_unlock(&consumer_data
.lock
);
1876 DBG("UST consumer discarded events command for session id %"
1877 PRIu64
", channel key %" PRIu64
, id
, key
);
1879 health_code_update();
1881 /* Send back returned value to session daemon */
1882 ret
= lttcomm_send_unix_sock(sock
, &discarded_events
, sizeof(discarded_events
));
1884 PERROR("send discarded events");
1890 case LTTNG_CONSUMER_LOST_PACKETS
:
1893 uint64_t lost_packets
;
1894 struct lttng_ht_iter iter
;
1895 struct lttng_ht
*ht
;
1896 struct lttng_consumer_stream
*stream
;
1897 uint64_t id
= msg
.u
.lost_packets
.session_id
;
1898 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
1900 DBG("UST consumer lost packets command for session id %"
1903 pthread_mutex_lock(&consumer_data
.lock
);
1905 ht
= consumer_data
.stream_list_ht
;
1908 * We only need a reference to the channel, but they are not
1909 * directly indexed, so we just use the first matching stream
1910 * to extract the information we need, we default to 0 if not
1911 * found (no packets lost if the channel is not yet in use).
1914 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1915 ht
->hash_fct(&id
, lttng_ht_seed
),
1917 &iter
.iter
, stream
, node_session_id
.node
) {
1918 if (stream
->chan
->key
== key
) {
1919 lost_packets
= stream
->chan
->lost_packets
;
1923 pthread_mutex_unlock(&consumer_data
.lock
);
1926 DBG("UST consumer lost packets command for session id %"
1927 PRIu64
", channel key %" PRIu64
, id
, key
);
1929 health_code_update();
1931 /* Send back returned value to session daemon */
1932 ret
= lttcomm_send_unix_sock(sock
, &lost_packets
,
1933 sizeof(lost_packets
));
1935 PERROR("send lost packets");
1941 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
1943 int channel_monitor_pipe
;
1945 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1946 /* Successfully received the command's type. */
1947 ret
= consumer_send_status_msg(sock
, ret_code
);
1952 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
1954 if (ret
!= sizeof(channel_monitor_pipe
)) {
1955 ERR("Failed to receive channel monitor pipe");
1959 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
1960 ret
= consumer_timer_thread_set_channel_monitor_pipe(
1961 channel_monitor_pipe
);
1965 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1966 /* Set the pipe as non-blocking. */
1967 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
1969 PERROR("fcntl get flags of the channel monitoring pipe");
1974 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
1975 flags
| O_NONBLOCK
);
1977 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1980 DBG("Channel monitor pipe set as non-blocking");
1982 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
1984 goto end_msg_sessiond
;
1986 case LTTNG_CONSUMER_ROTATE_CHANNEL
:
1988 struct lttng_consumer_channel
*channel
;
1989 uint64_t key
= msg
.u
.rotate_channel
.key
;
1991 channel
= consumer_find_channel(key
);
1993 DBG("Channel %" PRIu64
" not found", key
);
1994 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1997 * Sample the rotate position of all the streams in
2000 ret
= lttng_consumer_rotate_channel(channel
, key
,
2001 msg
.u
.rotate_channel
.relayd_id
,
2002 msg
.u
.rotate_channel
.metadata
,
2005 ERR("Rotate channel failed");
2006 ret_code
= LTTCOMM_CONSUMERD_ROTATION_FAIL
;
2009 health_code_update();
2011 ret
= consumer_send_status_msg(sock
, ret_code
);
2013 /* Somehow, the session daemon is not responding anymore. */
2014 goto end_rotate_channel_nosignal
;
2018 * Rotate the streams that are ready right now.
2019 * FIXME: this is a second consecutive iteration over the
2020 * streams in a channel, there is probably a better way to
2021 * handle this, but it needs to be after the
2022 * consumer_send_status_msg() call.
2025 ret
= lttng_consumer_rotate_ready_streams(
2028 ERR("Rotate channel failed");
2032 end_rotate_channel_nosignal
:
2035 case LTTNG_CONSUMER_CLEAR_CHANNEL
:
2037 struct lttng_consumer_channel
*channel
;
2038 uint64_t key
= msg
.u
.clear_channel
.key
;
2040 channel
= consumer_find_channel(key
);
2042 DBG("Channel %" PRIu64
" not found", key
);
2043 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2045 ret
= lttng_consumer_clear_channel(channel
);
2047 ERR("Clear channel failed key %" PRIu64
, key
);
2051 health_code_update();
2053 ret
= consumer_send_status_msg(sock
, ret_code
);
2055 /* Somehow, the session daemon is not responding anymore. */
2060 case LTTNG_CONSUMER_INIT
:
2062 ret_code
= lttng_consumer_init_command(ctx
,
2063 msg
.u
.init
.sessiond_uuid
);
2064 health_code_update();
2065 ret
= consumer_send_status_msg(sock
, ret_code
);
2067 /* Somehow, the session daemon is not responding anymore. */
2072 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK
:
2074 const struct lttng_credentials credentials
= {
2075 .uid
= msg
.u
.create_trace_chunk
.credentials
.value
.uid
,
2076 .gid
= msg
.u
.create_trace_chunk
.credentials
.value
.gid
,
2078 const bool is_local_trace
=
2079 !msg
.u
.create_trace_chunk
.relayd_id
.is_set
;
2080 const uint64_t relayd_id
=
2081 msg
.u
.create_trace_chunk
.relayd_id
.value
;
2082 const char *chunk_override_name
=
2083 *msg
.u
.create_trace_chunk
.override_name
?
2084 msg
.u
.create_trace_chunk
.override_name
:
2086 struct lttng_directory_handle
*chunk_directory_handle
= NULL
;
2089 * The session daemon will only provide a chunk directory file
2090 * descriptor for local traces.
2092 if (is_local_trace
) {
2095 /* Acnowledge the reception of the command. */
2096 ret
= consumer_send_status_msg(sock
,
2097 LTTCOMM_CONSUMERD_SUCCESS
);
2099 /* Somehow, the session daemon is not responding anymore. */
2104 * Receive trace chunk domain dirfd.
2106 ret
= lttcomm_recv_fds_unix_sock(sock
, &chunk_dirfd
, 1);
2107 if (ret
!= sizeof(chunk_dirfd
)) {
2108 ERR("Failed to receive trace chunk domain directory file descriptor");
2112 DBG("Received trace chunk domain directory fd (%d)",
2114 chunk_directory_handle
= lttng_directory_handle_create_from_dirfd(
2116 if (!chunk_directory_handle
) {
2117 ERR("Failed to initialize chunk domain directory handle from directory file descriptor");
2118 if (close(chunk_dirfd
)) {
2119 PERROR("Failed to close chunk directory file descriptor");
2125 ret_code
= lttng_consumer_create_trace_chunk(
2126 !is_local_trace
? &relayd_id
: NULL
,
2127 msg
.u
.create_trace_chunk
.session_id
,
2128 msg
.u
.create_trace_chunk
.chunk_id
,
2129 (time_t) msg
.u
.create_trace_chunk
2130 .creation_timestamp
,
2131 chunk_override_name
,
2132 msg
.u
.create_trace_chunk
.credentials
.is_set
?
2135 chunk_directory_handle
);
2136 lttng_directory_handle_put(chunk_directory_handle
);
2137 goto end_msg_sessiond
;
2139 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK
:
2141 enum lttng_trace_chunk_command_type close_command
=
2142 msg
.u
.close_trace_chunk
.close_command
.value
;
2143 const uint64_t relayd_id
=
2144 msg
.u
.close_trace_chunk
.relayd_id
.value
;
2145 struct lttcomm_consumer_close_trace_chunk_reply reply
;
2146 char closed_trace_chunk_path
[LTTNG_PATH_MAX
];
2149 ret_code
= lttng_consumer_close_trace_chunk(
2150 msg
.u
.close_trace_chunk
.relayd_id
.is_set
?
2153 msg
.u
.close_trace_chunk
.session_id
,
2154 msg
.u
.close_trace_chunk
.chunk_id
,
2155 (time_t) msg
.u
.close_trace_chunk
.close_timestamp
,
2156 msg
.u
.close_trace_chunk
.close_command
.is_set
?
2158 NULL
, closed_trace_chunk_path
);
2159 reply
.ret_code
= ret_code
;
2160 reply
.path_length
= strlen(closed_trace_chunk_path
) + 1;
2161 ret
= lttcomm_send_unix_sock(sock
, &reply
, sizeof(reply
));
2162 if (ret
!= sizeof(reply
)) {
2165 ret
= lttcomm_send_unix_sock(sock
, closed_trace_chunk_path
,
2167 if (ret
!= reply
.path_length
) {
2172 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS
:
2174 const uint64_t relayd_id
=
2175 msg
.u
.trace_chunk_exists
.relayd_id
.value
;
2177 ret_code
= lttng_consumer_trace_chunk_exists(
2178 msg
.u
.trace_chunk_exists
.relayd_id
.is_set
?
2180 msg
.u
.trace_chunk_exists
.session_id
,
2181 msg
.u
.trace_chunk_exists
.chunk_id
);
2182 goto end_msg_sessiond
;
2184 case LTTNG_CONSUMER_OPEN_CHANNEL_PACKETS
:
2186 const uint64_t key
= msg
.u
.open_channel_packets
.key
;
2187 struct lttng_consumer_channel
*channel
=
2188 consumer_find_channel(key
);
2191 pthread_mutex_lock(&channel
->lock
);
2192 ret_code
= lttng_consumer_open_channel_packets(channel
);
2193 pthread_mutex_unlock(&channel
->lock
);
2196 * The channel could have disappeared in per-pid
2199 DBG("Channel %" PRIu64
" not found", key
);
2200 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2203 health_code_update();
2204 goto end_msg_sessiond
;
2212 * Return 1 to indicate success since the 0 value can be a socket
2213 * shutdown during the recv() or send() call.
2220 * The returned value here is not useful since either way we'll return 1 to
2221 * the caller because the session daemon socket management is done
2222 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2224 ret
= consumer_send_status_msg(sock
, ret_code
);
2234 * Free channel here since no one has a reference to it. We don't
2235 * free after that because a stream can store this pointer.
2237 destroy_channel(channel
);
2239 /* We have to send a status channel message indicating an error. */
2240 ret
= consumer_send_status_channel(sock
, NULL
);
2242 /* Stop everything if session daemon can not be notified. */
2249 /* This will issue a consumer stop. */
2255 health_code_update();
2259 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream
*stream
,
2260 int producer_active
)
2263 assert(stream
->ustream
);
2265 ustctl_flush_buffer(stream
->ustream
, producer_active
);
2269 * Take a snapshot for a specific stream.
2271 * Returns 0 on success, < 0 on error
2273 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2276 assert(stream
->ustream
);
2278 return ustctl_snapshot(stream
->ustream
);
2282 * Sample consumed and produced positions for a specific stream.
2284 * Returns 0 on success, < 0 on error.
2286 int lttng_ustconsumer_sample_snapshot_positions(
2287 struct lttng_consumer_stream
*stream
)
2290 assert(stream
->ustream
);
2292 return ustctl_snapshot_sample_positions(stream
->ustream
);
2296 * Get the produced position
2298 * Returns 0 on success, < 0 on error
2300 int lttng_ustconsumer_get_produced_snapshot(
2301 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2304 assert(stream
->ustream
);
2307 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
2311 * Get the consumed position
2313 * Returns 0 on success, < 0 on error
2315 int lttng_ustconsumer_get_consumed_snapshot(
2316 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2319 assert(stream
->ustream
);
2322 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
2325 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
2329 assert(stream
->ustream
);
2331 ustctl_flush_buffer(stream
->ustream
, producer
);
2334 void lttng_ustconsumer_clear_buffer(struct lttng_consumer_stream
*stream
)
2337 assert(stream
->ustream
);
2339 ustctl_clear_buffer(stream
->ustream
);
2342 int lttng_ustconsumer_get_current_timestamp(
2343 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
2346 assert(stream
->ustream
);
2349 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
2352 int lttng_ustconsumer_get_sequence_number(
2353 struct lttng_consumer_stream
*stream
, uint64_t *seq
)
2356 assert(stream
->ustream
);
2359 return ustctl_get_sequence_number(stream
->ustream
, seq
);
2363 * Called when the stream signals the consumer that it has hung up.
2365 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
2368 assert(stream
->ustream
);
2370 pthread_mutex_lock(&stream
->lock
);
2371 if (!stream
->quiescent
) {
2372 ustctl_flush_buffer(stream
->ustream
, 0);
2373 stream
->quiescent
= true;
2375 pthread_mutex_unlock(&stream
->lock
);
2376 stream
->hangup_flush_done
= 1;
2379 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
2384 assert(chan
->uchan
);
2385 assert(chan
->buffer_credentials
.is_set
);
2387 if (chan
->switch_timer_enabled
== 1) {
2388 consumer_timer_switch_stop(chan
);
2390 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
2393 ret
= close(chan
->stream_fds
[i
]);
2397 if (chan
->shm_path
[0]) {
2398 char shm_path
[PATH_MAX
];
2400 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
2402 ERR("Cannot get stream shm path");
2404 ret
= run_as_unlink(shm_path
,
2405 chan
->buffer_credentials
.value
.uid
,
2406 chan
->buffer_credentials
.value
.gid
);
2408 PERROR("unlink %s", shm_path
);
2414 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
2417 assert(chan
->uchan
);
2418 assert(chan
->buffer_credentials
.is_set
);
2420 consumer_metadata_cache_destroy(chan
);
2421 ustctl_destroy_channel(chan
->uchan
);
2422 /* Try to rmdir all directories under shm_path root. */
2423 if (chan
->root_shm_path
[0]) {
2424 (void) run_as_rmdir_recursive(chan
->root_shm_path
,
2425 chan
->buffer_credentials
.value
.uid
,
2426 chan
->buffer_credentials
.value
.gid
,
2427 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG
);
2429 free(chan
->stream_fds
);
2432 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
2435 assert(stream
->ustream
);
2437 if (stream
->chan
->switch_timer_enabled
== 1) {
2438 consumer_timer_switch_stop(stream
->chan
);
2440 ustctl_destroy_stream(stream
->ustream
);
2443 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
2446 assert(stream
->ustream
);
2448 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
2451 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
2454 assert(stream
->ustream
);
2456 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
2460 void metadata_stream_reset_cache_consumed_position(
2461 struct lttng_consumer_stream
*stream
)
2463 DBG("Reset metadata cache of session %" PRIu64
,
2464 stream
->chan
->session_id
);
2465 stream
->ust_metadata_pushed
= 0;
2469 * Write up to one packet from the metadata cache to the channel.
2471 * Returns the number of bytes pushed from the cache into the ring buffer, or a
2472 * negative value on error.
2475 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
2480 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2481 if (stream
->chan
->metadata_cache
->max_offset
==
2482 stream
->ust_metadata_pushed
) {
2484 * In the context of a user space metadata channel, a
2485 * change in version can be detected in two ways:
2486 * 1) During the pre-consume of the `read_subbuffer` loop,
2487 * 2) When populating the metadata ring buffer (i.e. here).
2489 * This function is invoked when there is no metadata
2490 * available in the ring-buffer. If all data was consumed
2491 * up to the size of the metadata cache, there is no metadata
2492 * to insert in the ring-buffer.
2494 * However, the metadata version could still have changed (a
2495 * regeneration without any new data will yield the same cache
2498 * The cache's version is checked for a version change and the
2499 * consumed position is reset if one occurred.
2501 * This check is only necessary for the user space domain as
2502 * it has to manage the cache explicitly. If this reset was not
2503 * performed, no metadata would be consumed (and no reset would
2504 * occur as part of the pre-consume) until the metadata size
2505 * exceeded the cache size.
2507 if (stream
->metadata_version
!=
2508 stream
->chan
->metadata_cache
->version
) {
2509 metadata_stream_reset_cache_consumed_position(stream
);
2510 consumer_stream_metadata_set_version(stream
,
2511 stream
->chan
->metadata_cache
->version
);
2518 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
2519 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
2520 stream
->chan
->metadata_cache
->max_offset
2521 - stream
->ust_metadata_pushed
);
2522 assert(write_len
!= 0);
2523 if (write_len
< 0) {
2524 ERR("Writing one metadata packet");
2528 stream
->ust_metadata_pushed
+= write_len
;
2530 assert(stream
->chan
->metadata_cache
->max_offset
>=
2531 stream
->ust_metadata_pushed
);
2535 * Switch packet (but don't open the next one) on every commit of
2536 * a metadata packet. Since the subbuffer is fully filled (with padding,
2537 * if needed), the stream is "quiescent" after this commit.
2539 ustctl_flush_buffer(stream
->ustream
, 1);
2540 stream
->quiescent
= true;
2542 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2548 * Sync metadata meaning request them to the session daemon and snapshot to the
2549 * metadata thread can consumer them.
2551 * Metadata stream lock is held here, but we need to release it when
2552 * interacting with sessiond, else we cause a deadlock with live
2553 * awaiting on metadata to be pushed out.
2555 * The RCU read side lock must be held by the caller.
2557 enum sync_metadata_status
lttng_ustconsumer_sync_metadata(
2558 struct lttng_consumer_local_data
*ctx
,
2559 struct lttng_consumer_stream
*metadata_stream
)
2562 enum sync_metadata_status status
;
2563 struct lttng_consumer_channel
*metadata_channel
;
2566 assert(metadata_stream
);
2568 metadata_channel
= metadata_stream
->chan
;
2569 pthread_mutex_unlock(&metadata_stream
->lock
);
2571 * Request metadata from the sessiond, but don't wait for the flush
2572 * because we locked the metadata thread.
2574 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 0);
2575 pthread_mutex_lock(&metadata_stream
->lock
);
2577 status
= SYNC_METADATA_STATUS_ERROR
;
2582 * The metadata stream and channel can be deleted while the
2583 * metadata stream lock was released. The streamed is checked
2584 * for deletion before we use it further.
2586 * Note that it is safe to access a logically-deleted stream since its
2587 * existence is still guaranteed by the RCU read side lock. However,
2588 * it should no longer be used. The close/deletion of the metadata
2589 * channel and stream already guarantees that all metadata has been
2590 * consumed. Therefore, there is nothing left to do in this function.
2592 if (consumer_stream_is_deleted(metadata_stream
)) {
2593 DBG("Metadata stream %" PRIu64
" was deleted during the metadata synchronization",
2594 metadata_stream
->key
);
2595 status
= SYNC_METADATA_STATUS_NO_DATA
;
2599 ret
= commit_one_metadata_packet(metadata_stream
);
2601 status
= SYNC_METADATA_STATUS_ERROR
;
2603 } else if (ret
> 0) {
2604 status
= SYNC_METADATA_STATUS_NEW_DATA
;
2605 } else /* ret == 0 */ {
2606 status
= SYNC_METADATA_STATUS_NO_DATA
;
2610 ret
= ustctl_snapshot(metadata_stream
->ustream
);
2612 ERR("Failed to take a snapshot of the metadata ring-buffer positions, ret = %d", ret
);
2613 status
= SYNC_METADATA_STATUS_ERROR
;
2622 * Return 0 on success else a negative value.
2624 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2625 struct lttng_consumer_local_data
*ctx
)
2628 struct ustctl_consumer_stream
*ustream
;
2633 ustream
= stream
->ustream
;
2636 * First, we are going to check if there is a new subbuffer available
2637 * before reading the stream wait_fd.
2639 /* Get the next subbuffer */
2640 ret
= ustctl_get_next_subbuf(ustream
);
2642 /* No more data found, flag the stream. */
2643 stream
->has_data
= 0;
2648 ret
= ustctl_put_subbuf(ustream
);
2651 /* This stream still has data. Flag it and wake up the data thread. */
2652 stream
->has_data
= 1;
2654 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2657 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2658 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2663 /* The wake up pipe has been notified. */
2664 ctx
->has_wakeup
= 1;
2672 static int consumer_stream_ust_on_wake_up(struct lttng_consumer_stream
*stream
)
2677 * We can consume the 1 byte written into the wait_fd by
2678 * UST. Don't trigger error if we cannot read this one byte
2679 * (read returns 0), or if the error is EAGAIN or EWOULDBLOCK.
2681 * This is only done when the stream is monitored by a thread,
2682 * before the flush is done after a hangup and if the stream
2683 * is not flagged with data since there might be nothing to
2684 * consume in the wait fd but still have data available
2685 * flagged by the consumer wake up pipe.
2687 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2691 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2692 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2700 static int extract_common_subbuffer_info(struct lttng_consumer_stream
*stream
,
2701 struct stream_subbuffer
*subbuf
)
2705 ret
= ustctl_get_subbuf_size(
2706 stream
->ustream
, &subbuf
->info
.data
.subbuf_size
);
2711 ret
= ustctl_get_padded_subbuf_size(
2712 stream
->ustream
, &subbuf
->info
.data
.padded_subbuf_size
);
2721 static int extract_metadata_subbuffer_info(struct lttng_consumer_stream
*stream
,
2722 struct stream_subbuffer
*subbuf
)
2726 ret
= extract_common_subbuffer_info(stream
, subbuf
);
2731 subbuf
->info
.metadata
.version
= stream
->metadata_version
;
2737 static int extract_data_subbuffer_info(struct lttng_consumer_stream
*stream
,
2738 struct stream_subbuffer
*subbuf
)
2742 ret
= extract_common_subbuffer_info(stream
, subbuf
);
2747 ret
= ustctl_get_packet_size(
2748 stream
->ustream
, &subbuf
->info
.data
.packet_size
);
2750 PERROR("Failed to get sub-buffer packet size");
2754 ret
= ustctl_get_content_size(
2755 stream
->ustream
, &subbuf
->info
.data
.content_size
);
2757 PERROR("Failed to get sub-buffer content size");
2761 ret
= ustctl_get_timestamp_begin(
2762 stream
->ustream
, &subbuf
->info
.data
.timestamp_begin
);
2764 PERROR("Failed to get sub-buffer begin timestamp");
2768 ret
= ustctl_get_timestamp_end(
2769 stream
->ustream
, &subbuf
->info
.data
.timestamp_end
);
2771 PERROR("Failed to get sub-buffer end timestamp");
2775 ret
= ustctl_get_events_discarded(
2776 stream
->ustream
, &subbuf
->info
.data
.events_discarded
);
2778 PERROR("Failed to get sub-buffer events discarded count");
2782 ret
= ustctl_get_sequence_number(stream
->ustream
,
2783 &subbuf
->info
.data
.sequence_number
.value
);
2785 /* May not be supported by older LTTng-modules. */
2786 if (ret
!= -ENOTTY
) {
2787 PERROR("Failed to get sub-buffer sequence number");
2791 subbuf
->info
.data
.sequence_number
.is_set
= true;
2794 ret
= ustctl_get_stream_id(
2795 stream
->ustream
, &subbuf
->info
.data
.stream_id
);
2797 PERROR("Failed to get stream id");
2801 ret
= ustctl_get_instance_id(stream
->ustream
,
2802 &subbuf
->info
.data
.stream_instance_id
.value
);
2804 /* May not be supported by older LTTng-modules. */
2805 if (ret
!= -ENOTTY
) {
2806 PERROR("Failed to get stream instance id");
2810 subbuf
->info
.data
.stream_instance_id
.is_set
= true;
2816 static int get_next_subbuffer_common(struct lttng_consumer_stream
*stream
,
2817 struct stream_subbuffer
*subbuffer
)
2822 ret
= stream
->read_subbuffer_ops
.extract_subbuffer_info(
2828 ret
= get_current_subbuf_addr(stream
, &addr
);
2833 subbuffer
->buffer
.buffer
= lttng_buffer_view_init(
2834 addr
, 0, subbuffer
->info
.data
.padded_subbuf_size
);
2835 assert(subbuffer
->buffer
.buffer
.data
!= NULL
);
2840 static int get_next_subbuffer(struct lttng_consumer_stream
*stream
,
2841 struct stream_subbuffer
*subbuffer
)
2845 ret
= ustctl_get_next_subbuf(stream
->ustream
);
2850 ret
= get_next_subbuffer_common(stream
, subbuffer
);
2858 static int get_next_subbuffer_metadata(struct lttng_consumer_stream
*stream
,
2859 struct stream_subbuffer
*subbuffer
)
2866 unsigned long consumed_pos
, produced_pos
;
2869 ret
= ustctl_get_next_subbuf(stream
->ustream
);
2871 got_subbuffer
= true;
2873 got_subbuffer
= false;
2874 if (ret
!= -EAGAIN
) {
2881 * Determine if the cache is empty and ensure that a sub-buffer
2882 * is made available if the cache is not empty.
2884 if (!got_subbuffer
) {
2885 ret
= commit_one_metadata_packet(stream
);
2886 if (ret
< 0 && ret
!= -ENOBUFS
) {
2888 } else if (ret
== 0) {
2889 /* Not an error, the cache is empty. */
2894 cache_empty
= false;
2897 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2898 cache_empty
= stream
->chan
->metadata_cache
->max_offset
==
2899 stream
->ust_metadata_pushed
;
2900 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2902 } while (!got_subbuffer
);
2904 /* Populate sub-buffer infos and view. */
2905 ret
= get_next_subbuffer_common(stream
, subbuffer
);
2910 ret
= lttng_ustconsumer_sample_snapshot_positions(stream
);
2913 * -EAGAIN is not expected since we got a sub-buffer and haven't
2914 * pushed the consumption position yet (on put_next).
2916 PERROR("Failed to take a snapshot of metadata buffer positions");
2920 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
2922 PERROR("Failed to get metadata consumed position");
2926 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
2928 PERROR("Failed to get metadata produced position");
2932 /* Last sub-buffer of the ring buffer ? */
2933 buffer_empty
= (consumed_pos
+ stream
->max_sb_size
) == produced_pos
;
2936 * The sessiond registry lock ensures that coherent units of metadata
2937 * are pushed to the consumer daemon at once. Hence, if a sub-buffer is
2938 * acquired, the cache is empty, and it is the only available sub-buffer
2939 * available, it is safe to assume that it is "coherent".
2941 coherent
= got_subbuffer
&& cache_empty
&& buffer_empty
;
2943 LTTNG_OPTIONAL_SET(&subbuffer
->info
.metadata
.coherent
, coherent
);
2948 static int put_next_subbuffer(struct lttng_consumer_stream
*stream
,
2949 struct stream_subbuffer
*subbuffer
)
2951 const int ret
= ustctl_put_next_subbuf(stream
->ustream
);
2957 static int signal_metadata(struct lttng_consumer_stream
*stream
,
2958 struct lttng_consumer_local_data
*ctx
)
2960 return pthread_cond_broadcast(&stream
->metadata_rdv
) ? -errno
: 0;
2963 static int lttng_ustconsumer_set_stream_ops(
2964 struct lttng_consumer_stream
*stream
)
2968 stream
->read_subbuffer_ops
.on_wake_up
= consumer_stream_ust_on_wake_up
;
2969 if (stream
->metadata_flag
) {
2970 stream
->read_subbuffer_ops
.get_next_subbuffer
=
2971 get_next_subbuffer_metadata
;
2972 stream
->read_subbuffer_ops
.extract_subbuffer_info
=
2973 extract_metadata_subbuffer_info
;
2974 stream
->read_subbuffer_ops
.reset_metadata
=
2975 metadata_stream_reset_cache_consumed_position
;
2976 if (stream
->chan
->is_live
) {
2977 stream
->read_subbuffer_ops
.on_sleep
= signal_metadata
;
2978 ret
= consumer_stream_enable_metadata_bucketization(
2985 stream
->read_subbuffer_ops
.get_next_subbuffer
=
2987 stream
->read_subbuffer_ops
.extract_subbuffer_info
=
2988 extract_data_subbuffer_info
;
2989 stream
->read_subbuffer_ops
.on_sleep
= notify_if_more_data
;
2990 if (stream
->chan
->is_live
) {
2991 stream
->read_subbuffer_ops
.send_live_beacon
=
2992 consumer_flush_ust_index
;
2996 stream
->read_subbuffer_ops
.put_next_subbuffer
= put_next_subbuffer
;
3002 * Called when a stream is created.
3004 * Return 0 on success or else a negative value.
3006 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3013 * Don't create anything if this is set for streaming or if there is
3014 * no current trace chunk on the parent channel.
3016 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
&&
3017 stream
->chan
->trace_chunk
) {
3018 ret
= consumer_stream_create_output_files(stream
, true);
3024 lttng_ustconsumer_set_stream_ops(stream
);
3032 * Check if data is still being extracted from the buffers for a specific
3033 * stream. Consumer data lock MUST be acquired before calling this function
3034 * and the stream lock.
3036 * Return 1 if the traced data are still getting read else 0 meaning that the
3037 * data is available for trace viewer reading.
3039 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
3044 assert(stream
->ustream
);
3046 DBG("UST consumer checking data pending");
3048 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
3053 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
3054 uint64_t contiguous
, pushed
;
3056 /* Ease our life a bit. */
3057 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
3058 pushed
= stream
->ust_metadata_pushed
;
3061 * We can simply check whether all contiguously available data
3062 * has been pushed to the ring buffer, since the push operation
3063 * is performed within get_next_subbuf(), and because both
3064 * get_next_subbuf() and put_next_subbuf() are issued atomically
3065 * thanks to the stream lock within
3066 * lttng_ustconsumer_read_subbuffer(). This basically means that
3067 * whetnever ust_metadata_pushed is incremented, the associated
3068 * metadata has been consumed from the metadata stream.
3070 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
3071 contiguous
, pushed
);
3072 assert(((int64_t) (contiguous
- pushed
)) >= 0);
3073 if ((contiguous
!= pushed
) ||
3074 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
3075 ret
= 1; /* Data is pending */
3079 ret
= ustctl_get_next_subbuf(stream
->ustream
);
3082 * There is still data so let's put back this
3085 ret
= ustctl_put_subbuf(stream
->ustream
);
3087 ret
= 1; /* Data is pending */
3092 /* Data is NOT pending so ready to be read. */
3100 * Stop a given metadata channel timer if enabled and close the wait fd which
3101 * is the poll pipe of the metadata stream.
3103 * This MUST be called with the metadata channel lock acquired.
3105 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
3110 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
3112 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
3114 if (metadata
->switch_timer_enabled
== 1) {
3115 consumer_timer_switch_stop(metadata
);
3118 if (!metadata
->metadata_stream
) {
3123 * Closing write side so the thread monitoring the stream wakes up if any
3124 * and clean the metadata stream.
3126 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
3127 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
3129 PERROR("closing metadata pipe write side");
3131 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
3139 * Close every metadata stream wait fd of the metadata hash table. This
3140 * function MUST be used very carefully so not to run into a race between the
3141 * metadata thread handling streams and this function closing their wait fd.
3143 * For UST, this is used when the session daemon hangs up. Its the metadata
3144 * producer so calling this is safe because we are assured that no state change
3145 * can occur in the metadata thread for the streams in the hash table.
3147 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
3149 struct lttng_ht_iter iter
;
3150 struct lttng_consumer_stream
*stream
;
3152 assert(metadata_ht
);
3153 assert(metadata_ht
->ht
);
3155 DBG("UST consumer closing all metadata streams");
3158 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
3161 health_code_update();
3163 pthread_mutex_lock(&stream
->chan
->lock
);
3164 lttng_ustconsumer_close_metadata(stream
->chan
);
3165 pthread_mutex_unlock(&stream
->chan
->lock
);
3171 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
3175 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
3177 ERR("Unable to close wakeup fd");
3182 * Please refer to consumer-timer.c before adding any lock within this
3183 * function or any of its callees. Timers have a very strict locking
3184 * semantic with respect to teardown. Failure to respect this semantic
3185 * introduces deadlocks.
3187 * DON'T hold the metadata lock when calling this function, else this
3188 * can cause deadlock involving consumer awaiting for metadata to be
3189 * pushed out due to concurrent interaction with the session daemon.
3191 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
3192 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
3194 struct lttcomm_metadata_request_msg request
;
3195 struct lttcomm_consumer_msg msg
;
3196 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3197 uint64_t len
, key
, offset
, version
;
3201 assert(channel
->metadata_cache
);
3203 memset(&request
, 0, sizeof(request
));
3205 /* send the metadata request to sessiond */
3206 switch (consumer_data
.type
) {
3207 case LTTNG_CONSUMER64_UST
:
3208 request
.bits_per_long
= 64;
3210 case LTTNG_CONSUMER32_UST
:
3211 request
.bits_per_long
= 32;
3214 request
.bits_per_long
= 0;
3218 request
.session_id
= channel
->session_id
;
3219 request
.session_id_per_pid
= channel
->session_id_per_pid
;
3221 * Request the application UID here so the metadata of that application can
3222 * be sent back. The channel UID corresponds to the user UID of the session
3223 * used for the rights on the stream file(s).
3225 request
.uid
= channel
->ust_app_uid
;
3226 request
.key
= channel
->key
;
3228 DBG("Sending metadata request to sessiond, session id %" PRIu64
3229 ", per-pid %" PRIu64
", app UID %u and channel key %" PRIu64
,
3230 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
3233 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
3235 health_code_update();
3237 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
3240 ERR("Asking metadata to sessiond");
3244 health_code_update();
3246 /* Receive the metadata from sessiond */
3247 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
3249 if (ret
!= sizeof(msg
)) {
3250 DBG("Consumer received unexpected message size %d (expects %zu)",
3252 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
3254 * The ret value might 0 meaning an orderly shutdown but this is ok
3255 * since the caller handles this.
3260 health_code_update();
3262 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
3263 /* No registry found */
3264 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3268 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
3269 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
3274 len
= msg
.u
.push_metadata
.len
;
3275 key
= msg
.u
.push_metadata
.key
;
3276 offset
= msg
.u
.push_metadata
.target_offset
;
3277 version
= msg
.u
.push_metadata
.version
;
3279 assert(key
== channel
->key
);
3281 DBG("No new metadata to receive for key %" PRIu64
, key
);
3284 health_code_update();
3286 /* Tell session daemon we are ready to receive the metadata. */
3287 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3288 LTTCOMM_CONSUMERD_SUCCESS
);
3289 if (ret
< 0 || len
== 0) {
3291 * Somehow, the session daemon is not responding anymore or there is
3292 * nothing to receive.
3297 health_code_update();
3299 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
3300 key
, offset
, len
, version
, channel
, timer
, wait
);
3303 * Only send the status msg if the sessiond is alive meaning a positive
3306 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
3311 health_code_update();
3313 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
3318 * Return the ustctl call for the get stream id.
3320 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
3321 uint64_t *stream_id
)
3326 return ustctl_get_stream_id(stream
->ustream
, stream_id
);