2 * Copyright (C) 2011 EfficiOS Inc.
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_init(&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_init(&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 metadata
->metadata_stream
= NULL
;
986 * Snapshot the whole metadata.
987 * RCU read-side lock must be held by the caller.
989 * Returns 0 on success, < 0 on error
991 static int snapshot_metadata(struct lttng_consumer_channel
*metadata_channel
,
992 uint64_t key
, char *path
, uint64_t relayd_id
,
993 struct lttng_consumer_local_data
*ctx
)
996 struct lttng_consumer_stream
*metadata_stream
;
1001 DBG("UST consumer snapshot metadata with key %" PRIu64
" at path %s",
1006 assert(!metadata_channel
->monitor
);
1008 health_code_update();
1011 * Ask the sessiond if we have new metadata waiting and update the
1012 * consumer metadata cache.
1014 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 1);
1019 health_code_update();
1022 * The metadata stream is NOT created in no monitor mode when the channel
1023 * is created on a sessiond ask channel command.
1025 ret
= create_ust_streams(metadata_channel
, ctx
);
1030 metadata_stream
= metadata_channel
->metadata_stream
;
1031 assert(metadata_stream
);
1033 metadata_stream
->read_subbuffer_ops
.lock(metadata_stream
);
1034 if (relayd_id
!= (uint64_t) -1ULL) {
1035 metadata_stream
->net_seq_idx
= relayd_id
;
1036 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
1038 ret
= consumer_stream_create_output_files(metadata_stream
,
1046 health_code_update();
1047 ret
= lttng_consumer_read_subbuffer(metadata_stream
, ctx
, true);
1054 metadata_stream
->read_subbuffer_ops
.unlock(metadata_stream
);
1056 * Clean up the stream completely because the next snapshot will use a
1057 * new metadata stream.
1059 consumer_stream_destroy(metadata_stream
, NULL
);
1060 metadata_channel
->metadata_stream
= NULL
;
1068 int get_current_subbuf_addr(struct lttng_consumer_stream
*stream
,
1072 unsigned long mmap_offset
;
1073 const char *mmap_base
;
1075 mmap_base
= ustctl_get_mmap_base(stream
->ustream
);
1077 ERR("Failed to get mmap base for stream `%s`",
1083 ret
= ustctl_get_mmap_read_offset(stream
->ustream
, &mmap_offset
);
1085 ERR("Failed to get mmap offset for stream `%s`", stream
->name
);
1090 *addr
= mmap_base
+ mmap_offset
;
1097 * Take a snapshot of all the stream of a channel.
1098 * RCU read-side lock and the channel lock must be held by the caller.
1100 * Returns 0 on success, < 0 on error
1102 static int snapshot_channel(struct lttng_consumer_channel
*channel
,
1103 uint64_t key
, char *path
, uint64_t relayd_id
,
1104 uint64_t nb_packets_per_stream
,
1105 struct lttng_consumer_local_data
*ctx
)
1108 unsigned use_relayd
= 0;
1109 unsigned long consumed_pos
, produced_pos
;
1110 struct lttng_consumer_stream
*stream
;
1117 if (relayd_id
!= (uint64_t) -1ULL) {
1121 assert(!channel
->monitor
);
1122 DBG("UST consumer snapshot channel %" PRIu64
, key
);
1124 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
1125 health_code_update();
1127 /* Lock stream because we are about to change its state. */
1128 pthread_mutex_lock(&stream
->lock
);
1129 assert(channel
->trace_chunk
);
1130 if (!lttng_trace_chunk_get(channel
->trace_chunk
)) {
1132 * Can't happen barring an internal error as the channel
1133 * holds a reference to the trace chunk.
1135 ERR("Failed to acquire reference to channel's trace chunk");
1139 assert(!stream
->trace_chunk
);
1140 stream
->trace_chunk
= channel
->trace_chunk
;
1142 stream
->net_seq_idx
= relayd_id
;
1145 ret
= consumer_send_relayd_stream(stream
, path
);
1147 goto error_close_stream
;
1150 ret
= consumer_stream_create_output_files(stream
,
1153 goto error_close_stream
;
1155 DBG("UST consumer snapshot stream (%" PRIu64
")",
1160 * If tracing is active, we want to perform a "full" buffer flush.
1161 * Else, if quiescent, it has already been done by the prior stop.
1163 if (!stream
->quiescent
) {
1164 ustctl_flush_buffer(stream
->ustream
, 0);
1167 ret
= lttng_ustconsumer_take_snapshot(stream
);
1169 ERR("Taking UST snapshot");
1170 goto error_close_stream
;
1173 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
1175 ERR("Produced UST snapshot position");
1176 goto error_close_stream
;
1179 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1181 ERR("Consumerd UST snapshot position");
1182 goto error_close_stream
;
1186 * The original value is sent back if max stream size is larger than
1187 * the possible size of the snapshot. Also, we assume that the session
1188 * daemon should never send a maximum stream size that is lower than
1191 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
1192 produced_pos
, nb_packets_per_stream
,
1193 stream
->max_sb_size
);
1195 while ((long) (consumed_pos
- produced_pos
) < 0) {
1197 unsigned long len
, padded_len
;
1198 const char *subbuf_addr
;
1199 struct lttng_buffer_view subbuf_view
;
1201 health_code_update();
1203 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1205 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1207 if (ret
!= -EAGAIN
) {
1208 PERROR("ustctl_get_subbuf snapshot");
1209 goto error_close_stream
;
1211 DBG("UST consumer get subbuf failed. Skipping it.");
1212 consumed_pos
+= stream
->max_sb_size
;
1213 stream
->chan
->lost_packets
++;
1217 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1219 ERR("Snapshot ustctl_get_subbuf_size");
1220 goto error_put_subbuf
;
1223 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1225 ERR("Snapshot ustctl_get_padded_subbuf_size");
1226 goto error_put_subbuf
;
1229 ret
= get_current_subbuf_addr(stream
, &subbuf_addr
);
1231 goto error_put_subbuf
;
1234 subbuf_view
= lttng_buffer_view_init(
1235 subbuf_addr
, 0, padded_len
);
1236 read_len
= lttng_consumer_on_read_subbuffer_mmap(
1237 stream
, &subbuf_view
, padded_len
- len
);
1239 if (read_len
!= len
) {
1241 goto error_put_subbuf
;
1244 if (read_len
!= padded_len
) {
1246 goto error_put_subbuf
;
1250 ret
= ustctl_put_subbuf(stream
->ustream
);
1252 ERR("Snapshot ustctl_put_subbuf");
1253 goto error_close_stream
;
1255 consumed_pos
+= stream
->max_sb_size
;
1258 /* Simply close the stream so we can use it on the next snapshot. */
1259 consumer_stream_close(stream
);
1260 pthread_mutex_unlock(&stream
->lock
);
1267 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1268 ERR("Snapshot ustctl_put_subbuf");
1271 consumer_stream_close(stream
);
1273 pthread_mutex_unlock(&stream
->lock
);
1279 void metadata_stream_reset_cache_consumed_position(
1280 struct lttng_consumer_stream
*stream
)
1282 ASSERT_LOCKED(stream
->lock
);
1284 DBG("Reset metadata cache of session %" PRIu64
,
1285 stream
->chan
->session_id
);
1286 stream
->ust_metadata_pushed
= 0;
1290 * Receive the metadata updates from the sessiond. Supports receiving
1291 * overlapping metadata, but is needs to always belong to a contiguous
1292 * range starting from 0.
1293 * Be careful about the locks held when calling this function: it needs
1294 * the metadata cache flush to concurrently progress in order to
1297 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1298 uint64_t len
, uint64_t version
,
1299 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
1301 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1303 enum consumer_metadata_cache_write_status cache_write_status
;
1305 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1307 metadata_str
= zmalloc(len
* sizeof(char));
1308 if (!metadata_str
) {
1309 PERROR("zmalloc metadata string");
1310 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1314 health_code_update();
1316 /* Receive metadata string. */
1317 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1319 /* Session daemon is dead so return gracefully. */
1324 health_code_update();
1326 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1327 cache_write_status
= consumer_metadata_cache_write(
1328 channel
, offset
, len
, version
, metadata_str
);
1329 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1330 switch (cache_write_status
) {
1331 case CONSUMER_METADATA_CACHE_WRITE_STATUS_NO_CHANGE
:
1333 * The write entirely overlapped with existing contents of the
1334 * same metadata version (same content); there is nothing to do.
1337 case CONSUMER_METADATA_CACHE_WRITE_STATUS_INVALIDATED
:
1339 * The metadata cache was invalidated (previously pushed
1340 * content has been overwritten). Reset the stream's consumed
1341 * metadata position to ensure the metadata poll thread consumes
1346 * channel::metadata_stream can be null when the metadata
1347 * channel is under a snapshot session type. No need to update
1348 * the stream position in that scenario.
1350 if (channel
->metadata_stream
!= NULL
) {
1351 pthread_mutex_lock(&channel
->metadata_stream
->lock
);
1352 metadata_stream_reset_cache_consumed_position(
1353 channel
->metadata_stream
);
1354 pthread_mutex_unlock(&channel
->metadata_stream
->lock
);
1357 case CONSUMER_METADATA_CACHE_WRITE_STATUS_APPENDED_CONTENT
:
1359 * In both cases, the metadata poll thread has new data to
1362 ret
= consumer_metadata_wakeup_pipe(channel
);
1364 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1368 case CONSUMER_METADATA_CACHE_WRITE_STATUS_ERROR
:
1369 /* Unable to handle metadata. Notify session daemon. */
1370 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1372 * Skip metadata flush on write error since the offset and len might
1373 * not have been updated which could create an infinite loop below when
1374 * waiting for the metadata cache to be flushed.
1384 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1385 DBG("Waiting for metadata to be flushed");
1387 health_code_update();
1389 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1399 * Receive command from session daemon and process it.
1401 * Return 1 on success else a negative value or 0.
1403 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1404 int sock
, struct pollfd
*consumer_sockpoll
)
1407 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1408 struct lttcomm_consumer_msg msg
;
1409 struct lttng_consumer_channel
*channel
= NULL
;
1411 health_code_update();
1413 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1414 if (ret
!= sizeof(msg
)) {
1415 DBG("Consumer received unexpected message size %zd (expects %zu)",
1418 * The ret value might 0 meaning an orderly shutdown but this is ok
1419 * since the caller handles this.
1422 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1428 health_code_update();
1431 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1433 health_code_update();
1435 /* relayd needs RCU read-side lock */
1438 switch (msg
.cmd_type
) {
1439 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1441 uint32_t major
= msg
.u
.relayd_sock
.major
;
1442 uint32_t minor
= msg
.u
.relayd_sock
.minor
;
1443 enum lttcomm_sock_proto protocol
=
1444 (enum lttcomm_sock_proto
) msg
.u
.relayd_sock
1445 .relayd_socket_protocol
;
1447 /* Session daemon status message are handled in the following call. */
1448 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1449 msg
.u
.relayd_sock
.type
, ctx
, sock
,
1450 consumer_sockpoll
, msg
.u
.relayd_sock
.session_id
,
1451 msg
.u
.relayd_sock
.relayd_session_id
, major
,
1455 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1457 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1458 struct consumer_relayd_sock_pair
*relayd
;
1460 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1462 /* Get relayd reference if exists. */
1463 relayd
= consumer_find_relayd(index
);
1464 if (relayd
== NULL
) {
1465 DBG("Unable to find relayd %" PRIu64
, index
);
1466 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1470 * Each relayd socket pair has a refcount of stream attached to it
1471 * which tells if the relayd is still active or not depending on the
1474 * This will set the destroy flag of the relayd object and destroy it
1475 * if the refcount reaches zero when called.
1477 * The destroy can happen either here or when a stream fd hangs up.
1480 consumer_flag_relayd_for_destroy(relayd
);
1483 goto end_msg_sessiond
;
1485 case LTTNG_CONSUMER_UPDATE_STREAM
:
1490 case LTTNG_CONSUMER_DATA_PENDING
:
1492 int ret
, is_data_pending
;
1493 uint64_t id
= msg
.u
.data_pending
.session_id
;
1495 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1497 is_data_pending
= consumer_data_pending(id
);
1499 /* Send back returned value to session daemon */
1500 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1501 sizeof(is_data_pending
));
1503 DBG("Error when sending the data pending ret code: %d", ret
);
1508 * No need to send back a status message since the data pending
1509 * returned value is the response.
1513 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1516 struct ustctl_consumer_channel_attr attr
;
1517 const uint64_t chunk_id
= msg
.u
.ask_channel
.chunk_id
.value
;
1518 const struct lttng_credentials buffer_credentials
= {
1519 .uid
= msg
.u
.ask_channel
.buffer_credentials
.uid
,
1520 .gid
= msg
.u
.ask_channel
.buffer_credentials
.gid
,
1523 /* Create a plain object and reserve a channel key. */
1524 channel
= consumer_allocate_channel(
1525 msg
.u
.ask_channel
.key
,
1526 msg
.u
.ask_channel
.session_id
,
1527 msg
.u
.ask_channel
.chunk_id
.is_set
?
1529 msg
.u
.ask_channel
.pathname
,
1530 msg
.u
.ask_channel
.name
,
1531 msg
.u
.ask_channel
.relayd_id
,
1532 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1533 msg
.u
.ask_channel
.tracefile_size
,
1534 msg
.u
.ask_channel
.tracefile_count
,
1535 msg
.u
.ask_channel
.session_id_per_pid
,
1536 msg
.u
.ask_channel
.monitor
,
1537 msg
.u
.ask_channel
.live_timer_interval
,
1538 msg
.u
.ask_channel
.is_live
,
1539 msg
.u
.ask_channel
.root_shm_path
,
1540 msg
.u
.ask_channel
.shm_path
);
1542 goto end_channel_error
;
1545 LTTNG_OPTIONAL_SET(&channel
->buffer_credentials
,
1546 buffer_credentials
);
1549 * Assign UST application UID to the channel. This value is ignored for
1550 * per PID buffers. This is specific to UST thus setting this after the
1553 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1555 /* Build channel attributes from received message. */
1556 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1557 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1558 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1559 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1560 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1561 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1562 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1563 attr
.blocking_timeout
= msg
.u
.ask_channel
.blocking_timeout
;
1565 /* Match channel buffer type to the UST abi. */
1566 switch (msg
.u
.ask_channel
.output
) {
1567 case LTTNG_EVENT_MMAP
:
1569 attr
.output
= LTTNG_UST_MMAP
;
1573 /* Translate and save channel type. */
1574 switch (msg
.u
.ask_channel
.type
) {
1575 case LTTNG_UST_CHAN_PER_CPU
:
1576 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1577 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1579 * Set refcount to 1 for owner. Below, we will
1580 * pass ownership to the
1581 * consumer_thread_channel_poll() thread.
1583 channel
->refcount
= 1;
1585 case LTTNG_UST_CHAN_METADATA
:
1586 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1587 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1594 health_code_update();
1596 ret
= ask_channel(ctx
, channel
, &attr
);
1598 goto end_channel_error
;
1601 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1602 ret
= consumer_metadata_cache_allocate(channel
);
1604 ERR("Allocating metadata cache");
1605 goto end_channel_error
;
1607 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1608 attr
.switch_timer_interval
= 0;
1610 int monitor_start_ret
;
1612 consumer_timer_live_start(channel
,
1613 msg
.u
.ask_channel
.live_timer_interval
);
1614 monitor_start_ret
= consumer_timer_monitor_start(
1616 msg
.u
.ask_channel
.monitor_timer_interval
);
1617 if (monitor_start_ret
< 0) {
1618 ERR("Starting channel monitoring timer failed");
1619 goto end_channel_error
;
1623 health_code_update();
1626 * Add the channel to the internal state AFTER all streams were created
1627 * and successfully sent to session daemon. This way, all streams must
1628 * be ready before this channel is visible to the threads.
1629 * If add_channel succeeds, ownership of the channel is
1630 * passed to consumer_thread_channel_poll().
1632 ret
= add_channel(channel
, ctx
);
1634 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1635 if (channel
->switch_timer_enabled
== 1) {
1636 consumer_timer_switch_stop(channel
);
1638 consumer_metadata_cache_destroy(channel
);
1640 if (channel
->live_timer_enabled
== 1) {
1641 consumer_timer_live_stop(channel
);
1643 if (channel
->monitor_timer_enabled
== 1) {
1644 consumer_timer_monitor_stop(channel
);
1646 goto end_channel_error
;
1649 health_code_update();
1652 * Channel and streams are now created. Inform the session daemon that
1653 * everything went well and should wait to receive the channel and
1654 * streams with ustctl API.
1656 ret
= consumer_send_status_channel(sock
, channel
);
1659 * There is probably a problem on the socket.
1666 case LTTNG_CONSUMER_GET_CHANNEL
:
1668 int ret
, relayd_err
= 0;
1669 uint64_t key
= msg
.u
.get_channel
.key
;
1670 struct lttng_consumer_channel
*channel
;
1672 channel
= consumer_find_channel(key
);
1674 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1675 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1676 goto end_get_channel
;
1679 health_code_update();
1681 /* Send the channel to sessiond (and relayd, if applicable). */
1682 ret
= send_channel_to_sessiond_and_relayd(sock
, channel
, ctx
,
1687 * We were unable to send to the relayd the stream so avoid
1688 * sending back a fatal error to the thread since this is OK
1689 * and the consumer can continue its work. The above call
1690 * has sent the error status message to the sessiond.
1692 goto end_get_channel_nosignal
;
1695 * The communicaton was broken hence there is a bad state between
1696 * the consumer and sessiond so stop everything.
1698 goto error_get_channel_fatal
;
1701 health_code_update();
1704 * In no monitor mode, the streams ownership is kept inside the channel
1705 * so don't send them to the data thread.
1707 if (!channel
->monitor
) {
1708 goto end_get_channel
;
1711 ret
= send_streams_to_thread(channel
, ctx
);
1714 * If we are unable to send the stream to the thread, there is
1715 * a big problem so just stop everything.
1717 goto error_get_channel_fatal
;
1719 /* List MUST be empty after or else it could be reused. */
1720 assert(cds_list_empty(&channel
->streams
.head
));
1722 goto end_msg_sessiond
;
1723 error_get_channel_fatal
:
1725 end_get_channel_nosignal
:
1728 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1730 uint64_t key
= msg
.u
.destroy_channel
.key
;
1733 * Only called if streams have not been sent to stream
1734 * manager thread. However, channel has been sent to
1735 * channel manager thread.
1737 notify_thread_del_channel(ctx
, key
);
1738 goto end_msg_sessiond
;
1740 case LTTNG_CONSUMER_CLOSE_METADATA
:
1744 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1749 goto end_msg_sessiond
;
1751 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1755 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1760 goto end_msg_sessiond
;
1762 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL
:
1766 ret
= clear_quiescent_channel(
1767 msg
.u
.clear_quiescent_channel
.key
);
1772 goto end_msg_sessiond
;
1774 case LTTNG_CONSUMER_PUSH_METADATA
:
1777 uint64_t len
= msg
.u
.push_metadata
.len
;
1778 uint64_t key
= msg
.u
.push_metadata
.key
;
1779 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1780 uint64_t version
= msg
.u
.push_metadata
.version
;
1781 struct lttng_consumer_channel
*channel
;
1783 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1786 channel
= consumer_find_channel(key
);
1789 * This is possible if the metadata creation on the consumer side
1790 * is in flight vis-a-vis a concurrent push metadata from the
1791 * session daemon. Simply return that the channel failed and the
1792 * session daemon will handle that message correctly considering
1793 * that this race is acceptable thus the DBG() statement here.
1795 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1796 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1797 goto end_push_metadata_msg_sessiond
;
1800 health_code_update();
1804 * There is nothing to receive. We have simply
1805 * checked whether the channel can be found.
1807 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1808 goto end_push_metadata_msg_sessiond
;
1811 /* Tell session daemon we are ready to receive the metadata. */
1812 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1814 /* Somehow, the session daemon is not responding anymore. */
1815 goto error_push_metadata_fatal
;
1818 health_code_update();
1820 /* Wait for more data. */
1821 health_poll_entry();
1822 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1825 goto error_push_metadata_fatal
;
1828 health_code_update();
1830 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1831 len
, version
, channel
, 0, 1);
1833 /* error receiving from sessiond */
1834 goto error_push_metadata_fatal
;
1837 goto end_push_metadata_msg_sessiond
;
1839 end_push_metadata_msg_sessiond
:
1840 goto end_msg_sessiond
;
1841 error_push_metadata_fatal
:
1844 case LTTNG_CONSUMER_SETUP_METADATA
:
1848 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1852 goto end_msg_sessiond
;
1854 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1856 struct lttng_consumer_channel
*channel
;
1857 uint64_t key
= msg
.u
.snapshot_channel
.key
;
1859 channel
= consumer_find_channel(key
);
1861 DBG("UST snapshot channel not found for key %" PRIu64
, key
);
1862 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1864 if (msg
.u
.snapshot_channel
.metadata
) {
1865 ret
= snapshot_metadata(channel
, key
,
1866 msg
.u
.snapshot_channel
.pathname
,
1867 msg
.u
.snapshot_channel
.relayd_id
,
1870 ERR("Snapshot metadata failed");
1871 ret_code
= LTTCOMM_CONSUMERD_SNAPSHOT_FAILED
;
1874 ret
= snapshot_channel(channel
, key
,
1875 msg
.u
.snapshot_channel
.pathname
,
1876 msg
.u
.snapshot_channel
.relayd_id
,
1877 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
1880 ERR("Snapshot channel failed");
1881 ret_code
= LTTCOMM_CONSUMERD_SNAPSHOT_FAILED
;
1885 health_code_update();
1886 ret
= consumer_send_status_msg(sock
, ret_code
);
1888 /* Somehow, the session daemon is not responding anymore. */
1891 health_code_update();
1894 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
1897 uint64_t discarded_events
;
1898 struct lttng_ht_iter iter
;
1899 struct lttng_ht
*ht
;
1900 struct lttng_consumer_stream
*stream
;
1901 uint64_t id
= msg
.u
.discarded_events
.session_id
;
1902 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
1904 DBG("UST consumer discarded events command for session id %"
1907 pthread_mutex_lock(&consumer_data
.lock
);
1909 ht
= consumer_data
.stream_list_ht
;
1912 * We only need a reference to the channel, but they are not
1913 * directly indexed, so we just use the first matching stream
1914 * to extract the information we need, we default to 0 if not
1915 * found (no events are dropped if the channel is not yet in
1918 discarded_events
= 0;
1919 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1920 ht
->hash_fct(&id
, lttng_ht_seed
),
1922 &iter
.iter
, stream
, node_session_id
.node
) {
1923 if (stream
->chan
->key
== key
) {
1924 discarded_events
= stream
->chan
->discarded_events
;
1928 pthread_mutex_unlock(&consumer_data
.lock
);
1931 DBG("UST consumer discarded events command for session id %"
1932 PRIu64
", channel key %" PRIu64
, id
, key
);
1934 health_code_update();
1936 /* Send back returned value to session daemon */
1937 ret
= lttcomm_send_unix_sock(sock
, &discarded_events
, sizeof(discarded_events
));
1939 PERROR("send discarded events");
1945 case LTTNG_CONSUMER_LOST_PACKETS
:
1948 uint64_t lost_packets
;
1949 struct lttng_ht_iter iter
;
1950 struct lttng_ht
*ht
;
1951 struct lttng_consumer_stream
*stream
;
1952 uint64_t id
= msg
.u
.lost_packets
.session_id
;
1953 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
1955 DBG("UST consumer lost packets command for session id %"
1958 pthread_mutex_lock(&consumer_data
.lock
);
1960 ht
= consumer_data
.stream_list_ht
;
1963 * We only need a reference to the channel, but they are not
1964 * directly indexed, so we just use the first matching stream
1965 * to extract the information we need, we default to 0 if not
1966 * found (no packets lost if the channel is not yet in use).
1969 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1970 ht
->hash_fct(&id
, lttng_ht_seed
),
1972 &iter
.iter
, stream
, node_session_id
.node
) {
1973 if (stream
->chan
->key
== key
) {
1974 lost_packets
= stream
->chan
->lost_packets
;
1978 pthread_mutex_unlock(&consumer_data
.lock
);
1981 DBG("UST consumer lost packets command for session id %"
1982 PRIu64
", channel key %" PRIu64
, id
, key
);
1984 health_code_update();
1986 /* Send back returned value to session daemon */
1987 ret
= lttcomm_send_unix_sock(sock
, &lost_packets
,
1988 sizeof(lost_packets
));
1990 PERROR("send lost packets");
1996 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
1998 int channel_monitor_pipe
;
2000 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
2001 /* Successfully received the command's type. */
2002 ret
= consumer_send_status_msg(sock
, ret_code
);
2007 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
2009 if (ret
!= sizeof(channel_monitor_pipe
)) {
2010 ERR("Failed to receive channel monitor pipe");
2014 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
2015 ret
= consumer_timer_thread_set_channel_monitor_pipe(
2016 channel_monitor_pipe
);
2020 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
2021 /* Set the pipe as non-blocking. */
2022 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
2024 PERROR("fcntl get flags of the channel monitoring pipe");
2029 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
2030 flags
| O_NONBLOCK
);
2032 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
2035 DBG("Channel monitor pipe set as non-blocking");
2037 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
2039 goto end_msg_sessiond
;
2041 case LTTNG_CONSUMER_ROTATE_CHANNEL
:
2043 struct lttng_consumer_channel
*channel
;
2044 uint64_t key
= msg
.u
.rotate_channel
.key
;
2046 channel
= consumer_find_channel(key
);
2048 DBG("Channel %" PRIu64
" not found", key
);
2049 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2052 * Sample the rotate position of all the streams in
2055 ret
= lttng_consumer_rotate_channel(channel
, key
,
2056 msg
.u
.rotate_channel
.relayd_id
,
2057 msg
.u
.rotate_channel
.metadata
,
2060 ERR("Rotate channel failed");
2061 ret_code
= LTTCOMM_CONSUMERD_ROTATION_FAIL
;
2064 health_code_update();
2066 ret
= consumer_send_status_msg(sock
, ret_code
);
2068 /* Somehow, the session daemon is not responding anymore. */
2069 goto end_rotate_channel_nosignal
;
2073 * Rotate the streams that are ready right now.
2074 * FIXME: this is a second consecutive iteration over the
2075 * streams in a channel, there is probably a better way to
2076 * handle this, but it needs to be after the
2077 * consumer_send_status_msg() call.
2080 ret
= lttng_consumer_rotate_ready_streams(
2083 ERR("Rotate channel failed");
2087 end_rotate_channel_nosignal
:
2090 case LTTNG_CONSUMER_CLEAR_CHANNEL
:
2092 struct lttng_consumer_channel
*channel
;
2093 uint64_t key
= msg
.u
.clear_channel
.key
;
2095 channel
= consumer_find_channel(key
);
2097 DBG("Channel %" PRIu64
" not found", key
);
2098 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2100 ret
= lttng_consumer_clear_channel(channel
);
2102 ERR("Clear channel failed key %" PRIu64
, key
);
2106 health_code_update();
2108 ret
= consumer_send_status_msg(sock
, ret_code
);
2110 /* Somehow, the session daemon is not responding anymore. */
2115 case LTTNG_CONSUMER_INIT
:
2117 ret_code
= lttng_consumer_init_command(ctx
,
2118 msg
.u
.init
.sessiond_uuid
);
2119 health_code_update();
2120 ret
= consumer_send_status_msg(sock
, ret_code
);
2122 /* Somehow, the session daemon is not responding anymore. */
2127 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK
:
2129 const struct lttng_credentials credentials
= {
2130 .uid
= msg
.u
.create_trace_chunk
.credentials
.value
.uid
,
2131 .gid
= msg
.u
.create_trace_chunk
.credentials
.value
.gid
,
2133 const bool is_local_trace
=
2134 !msg
.u
.create_trace_chunk
.relayd_id
.is_set
;
2135 const uint64_t relayd_id
=
2136 msg
.u
.create_trace_chunk
.relayd_id
.value
;
2137 const char *chunk_override_name
=
2138 *msg
.u
.create_trace_chunk
.override_name
?
2139 msg
.u
.create_trace_chunk
.override_name
:
2141 struct lttng_directory_handle
*chunk_directory_handle
= NULL
;
2144 * The session daemon will only provide a chunk directory file
2145 * descriptor for local traces.
2147 if (is_local_trace
) {
2150 /* Acnowledge the reception of the command. */
2151 ret
= consumer_send_status_msg(sock
,
2152 LTTCOMM_CONSUMERD_SUCCESS
);
2154 /* Somehow, the session daemon is not responding anymore. */
2159 * Receive trace chunk domain dirfd.
2161 ret
= lttcomm_recv_fds_unix_sock(sock
, &chunk_dirfd
, 1);
2162 if (ret
!= sizeof(chunk_dirfd
)) {
2163 ERR("Failed to receive trace chunk domain directory file descriptor");
2167 DBG("Received trace chunk domain directory fd (%d)",
2169 chunk_directory_handle
= lttng_directory_handle_create_from_dirfd(
2171 if (!chunk_directory_handle
) {
2172 ERR("Failed to initialize chunk domain directory handle from directory file descriptor");
2173 if (close(chunk_dirfd
)) {
2174 PERROR("Failed to close chunk directory file descriptor");
2180 ret_code
= lttng_consumer_create_trace_chunk(
2181 !is_local_trace
? &relayd_id
: NULL
,
2182 msg
.u
.create_trace_chunk
.session_id
,
2183 msg
.u
.create_trace_chunk
.chunk_id
,
2184 (time_t) msg
.u
.create_trace_chunk
2185 .creation_timestamp
,
2186 chunk_override_name
,
2187 msg
.u
.create_trace_chunk
.credentials
.is_set
?
2190 chunk_directory_handle
);
2191 lttng_directory_handle_put(chunk_directory_handle
);
2192 goto end_msg_sessiond
;
2194 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK
:
2196 enum lttng_trace_chunk_command_type close_command
=
2197 msg
.u
.close_trace_chunk
.close_command
.value
;
2198 const uint64_t relayd_id
=
2199 msg
.u
.close_trace_chunk
.relayd_id
.value
;
2200 struct lttcomm_consumer_close_trace_chunk_reply reply
;
2201 char closed_trace_chunk_path
[LTTNG_PATH_MAX
] = {};
2204 ret_code
= lttng_consumer_close_trace_chunk(
2205 msg
.u
.close_trace_chunk
.relayd_id
.is_set
?
2208 msg
.u
.close_trace_chunk
.session_id
,
2209 msg
.u
.close_trace_chunk
.chunk_id
,
2210 (time_t) msg
.u
.close_trace_chunk
.close_timestamp
,
2211 msg
.u
.close_trace_chunk
.close_command
.is_set
?
2213 NULL
, closed_trace_chunk_path
);
2214 reply
.ret_code
= ret_code
;
2215 reply
.path_length
= strlen(closed_trace_chunk_path
) + 1;
2216 ret
= lttcomm_send_unix_sock(sock
, &reply
, sizeof(reply
));
2217 if (ret
!= sizeof(reply
)) {
2220 ret
= lttcomm_send_unix_sock(sock
, closed_trace_chunk_path
,
2222 if (ret
!= reply
.path_length
) {
2227 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS
:
2229 const uint64_t relayd_id
=
2230 msg
.u
.trace_chunk_exists
.relayd_id
.value
;
2232 ret_code
= lttng_consumer_trace_chunk_exists(
2233 msg
.u
.trace_chunk_exists
.relayd_id
.is_set
?
2235 msg
.u
.trace_chunk_exists
.session_id
,
2236 msg
.u
.trace_chunk_exists
.chunk_id
);
2237 goto end_msg_sessiond
;
2239 case LTTNG_CONSUMER_OPEN_CHANNEL_PACKETS
:
2241 const uint64_t key
= msg
.u
.open_channel_packets
.key
;
2242 struct lttng_consumer_channel
*channel
=
2243 consumer_find_channel(key
);
2246 pthread_mutex_lock(&channel
->lock
);
2247 ret_code
= lttng_consumer_open_channel_packets(channel
);
2248 pthread_mutex_unlock(&channel
->lock
);
2251 * The channel could have disappeared in per-pid
2254 DBG("Channel %" PRIu64
" not found", key
);
2255 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2258 health_code_update();
2259 goto end_msg_sessiond
;
2267 * Return 1 to indicate success since the 0 value can be a socket
2268 * shutdown during the recv() or send() call.
2275 * The returned value here is not useful since either way we'll return 1 to
2276 * the caller because the session daemon socket management is done
2277 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2279 ret
= consumer_send_status_msg(sock
, ret_code
);
2289 * Free channel here since no one has a reference to it. We don't
2290 * free after that because a stream can store this pointer.
2292 destroy_channel(channel
);
2294 /* We have to send a status channel message indicating an error. */
2295 ret
= consumer_send_status_channel(sock
, NULL
);
2297 /* Stop everything if session daemon can not be notified. */
2304 /* This will issue a consumer stop. */
2310 health_code_update();
2314 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream
*stream
,
2315 int producer_active
)
2318 assert(stream
->ustream
);
2320 ustctl_flush_buffer(stream
->ustream
, producer_active
);
2324 * Take a snapshot for a specific stream.
2326 * Returns 0 on success, < 0 on error
2328 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2331 assert(stream
->ustream
);
2333 return ustctl_snapshot(stream
->ustream
);
2337 * Sample consumed and produced positions for a specific stream.
2339 * Returns 0 on success, < 0 on error.
2341 int lttng_ustconsumer_sample_snapshot_positions(
2342 struct lttng_consumer_stream
*stream
)
2345 assert(stream
->ustream
);
2347 return ustctl_snapshot_sample_positions(stream
->ustream
);
2351 * Get the produced position
2353 * Returns 0 on success, < 0 on error
2355 int lttng_ustconsumer_get_produced_snapshot(
2356 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2359 assert(stream
->ustream
);
2362 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
2366 * Get the consumed position
2368 * Returns 0 on success, < 0 on error
2370 int lttng_ustconsumer_get_consumed_snapshot(
2371 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2374 assert(stream
->ustream
);
2377 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
2380 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
2384 assert(stream
->ustream
);
2386 ustctl_flush_buffer(stream
->ustream
, producer
);
2389 void lttng_ustconsumer_clear_buffer(struct lttng_consumer_stream
*stream
)
2392 assert(stream
->ustream
);
2394 ustctl_clear_buffer(stream
->ustream
);
2397 int lttng_ustconsumer_get_current_timestamp(
2398 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
2401 assert(stream
->ustream
);
2404 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
2407 int lttng_ustconsumer_get_sequence_number(
2408 struct lttng_consumer_stream
*stream
, uint64_t *seq
)
2411 assert(stream
->ustream
);
2414 return ustctl_get_sequence_number(stream
->ustream
, seq
);
2418 * Called when the stream signals the consumer that it has hung up.
2420 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
2423 assert(stream
->ustream
);
2425 pthread_mutex_lock(&stream
->lock
);
2426 if (!stream
->quiescent
) {
2427 ustctl_flush_buffer(stream
->ustream
, 0);
2428 stream
->quiescent
= true;
2431 stream
->hangup_flush_done
= 1;
2432 pthread_mutex_unlock(&stream
->lock
);
2435 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
2440 assert(chan
->uchan
);
2441 assert(chan
->buffer_credentials
.is_set
);
2443 if (chan
->switch_timer_enabled
== 1) {
2444 consumer_timer_switch_stop(chan
);
2446 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
2449 ret
= close(chan
->stream_fds
[i
]);
2453 if (chan
->shm_path
[0]) {
2454 char shm_path
[PATH_MAX
];
2456 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
2458 ERR("Cannot get stream shm path");
2460 ret
= run_as_unlink(shm_path
,
2461 chan
->buffer_credentials
.value
.uid
,
2462 chan
->buffer_credentials
.value
.gid
);
2464 PERROR("unlink %s", shm_path
);
2470 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
2473 assert(chan
->uchan
);
2474 assert(chan
->buffer_credentials
.is_set
);
2476 consumer_metadata_cache_destroy(chan
);
2477 ustctl_destroy_channel(chan
->uchan
);
2478 /* Try to rmdir all directories under shm_path root. */
2479 if (chan
->root_shm_path
[0]) {
2480 (void) run_as_rmdir_recursive(chan
->root_shm_path
,
2481 chan
->buffer_credentials
.value
.uid
,
2482 chan
->buffer_credentials
.value
.gid
,
2483 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG
);
2485 free(chan
->stream_fds
);
2488 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
2491 assert(stream
->ustream
);
2493 if (stream
->chan
->switch_timer_enabled
== 1) {
2494 consumer_timer_switch_stop(stream
->chan
);
2496 ustctl_destroy_stream(stream
->ustream
);
2499 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
2502 assert(stream
->ustream
);
2504 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
2507 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
2510 assert(stream
->ustream
);
2512 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
2516 * Write up to one packet from the metadata cache to the channel.
2518 * Returns the number of bytes pushed from the cache into the ring buffer, or a
2519 * negative value on error.
2522 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
2527 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2528 if (stream
->chan
->metadata_cache
->max_offset
==
2529 stream
->ust_metadata_pushed
) {
2531 * In the context of a user space metadata channel, a
2532 * change in version can be detected in two ways:
2533 * 1) During the pre-consume of the `read_subbuffer` loop,
2534 * 2) When populating the metadata ring buffer (i.e. here).
2536 * This function is invoked when there is no metadata
2537 * available in the ring-buffer. If all data was consumed
2538 * up to the size of the metadata cache, there is no metadata
2539 * to insert in the ring-buffer.
2541 * However, the metadata version could still have changed (a
2542 * regeneration without any new data will yield the same cache
2545 * The cache's version is checked for a version change and the
2546 * consumed position is reset if one occurred.
2548 * This check is only necessary for the user space domain as
2549 * it has to manage the cache explicitly. If this reset was not
2550 * performed, no metadata would be consumed (and no reset would
2551 * occur as part of the pre-consume) until the metadata size
2552 * exceeded the cache size.
2554 if (stream
->metadata_version
!=
2555 stream
->chan
->metadata_cache
->version
) {
2556 metadata_stream_reset_cache_consumed_position(stream
);
2557 consumer_stream_metadata_set_version(stream
,
2558 stream
->chan
->metadata_cache
->version
);
2565 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
2566 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
2567 stream
->chan
->metadata_cache
->max_offset
2568 - stream
->ust_metadata_pushed
);
2569 assert(write_len
!= 0);
2570 if (write_len
< 0) {
2571 ERR("Writing one metadata packet");
2575 stream
->ust_metadata_pushed
+= write_len
;
2577 assert(stream
->chan
->metadata_cache
->max_offset
>=
2578 stream
->ust_metadata_pushed
);
2582 * Switch packet (but don't open the next one) on every commit of
2583 * a metadata packet. Since the subbuffer is fully filled (with padding,
2584 * if needed), the stream is "quiescent" after this commit.
2586 ustctl_flush_buffer(stream
->ustream
, 1);
2587 stream
->quiescent
= true;
2589 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2595 * Sync metadata meaning request them to the session daemon and snapshot to the
2596 * metadata thread can consumer them.
2598 * Metadata stream lock is held here, but we need to release it when
2599 * interacting with sessiond, else we cause a deadlock with live
2600 * awaiting on metadata to be pushed out.
2602 * The RCU read side lock must be held by the caller.
2604 enum sync_metadata_status
lttng_ustconsumer_sync_metadata(
2605 struct lttng_consumer_local_data
*ctx
,
2606 struct lttng_consumer_stream
*metadata_stream
)
2609 enum sync_metadata_status status
;
2610 struct lttng_consumer_channel
*metadata_channel
;
2613 assert(metadata_stream
);
2615 metadata_channel
= metadata_stream
->chan
;
2616 pthread_mutex_unlock(&metadata_stream
->lock
);
2618 * Request metadata from the sessiond, but don't wait for the flush
2619 * because we locked the metadata thread.
2621 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 0);
2622 pthread_mutex_lock(&metadata_stream
->lock
);
2624 status
= SYNC_METADATA_STATUS_ERROR
;
2629 * The metadata stream and channel can be deleted while the
2630 * metadata stream lock was released. The streamed is checked
2631 * for deletion before we use it further.
2633 * Note that it is safe to access a logically-deleted stream since its
2634 * existence is still guaranteed by the RCU read side lock. However,
2635 * it should no longer be used. The close/deletion of the metadata
2636 * channel and stream already guarantees that all metadata has been
2637 * consumed. Therefore, there is nothing left to do in this function.
2639 if (consumer_stream_is_deleted(metadata_stream
)) {
2640 DBG("Metadata stream %" PRIu64
" was deleted during the metadata synchronization",
2641 metadata_stream
->key
);
2642 status
= SYNC_METADATA_STATUS_NO_DATA
;
2646 ret
= commit_one_metadata_packet(metadata_stream
);
2648 status
= SYNC_METADATA_STATUS_ERROR
;
2650 } else if (ret
> 0) {
2651 status
= SYNC_METADATA_STATUS_NEW_DATA
;
2652 } else /* ret == 0 */ {
2653 status
= SYNC_METADATA_STATUS_NO_DATA
;
2657 ret
= ustctl_snapshot(metadata_stream
->ustream
);
2659 ERR("Failed to take a snapshot of the metadata ring-buffer positions, ret = %d", ret
);
2660 status
= SYNC_METADATA_STATUS_ERROR
;
2669 * Return 0 on success else a negative value.
2671 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2672 struct lttng_consumer_local_data
*ctx
)
2675 struct ustctl_consumer_stream
*ustream
;
2680 ustream
= stream
->ustream
;
2683 * First, we are going to check if there is a new subbuffer available
2684 * before reading the stream wait_fd.
2686 /* Get the next subbuffer */
2687 ret
= ustctl_get_next_subbuf(ustream
);
2689 /* No more data found, flag the stream. */
2690 stream
->has_data
= 0;
2695 ret
= ustctl_put_subbuf(ustream
);
2698 /* This stream still has data. Flag it and wake up the data thread. */
2699 stream
->has_data
= 1;
2701 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2704 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2705 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2710 /* The wake up pipe has been notified. */
2711 ctx
->has_wakeup
= 1;
2719 static int consumer_stream_ust_on_wake_up(struct lttng_consumer_stream
*stream
)
2724 * We can consume the 1 byte written into the wait_fd by
2725 * UST. Don't trigger error if we cannot read this one byte
2726 * (read returns 0), or if the error is EAGAIN or EWOULDBLOCK.
2728 * This is only done when the stream is monitored by a thread,
2729 * before the flush is done after a hangup and if the stream
2730 * is not flagged with data since there might be nothing to
2731 * consume in the wait fd but still have data available
2732 * flagged by the consumer wake up pipe.
2734 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2738 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2739 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2747 static int extract_common_subbuffer_info(struct lttng_consumer_stream
*stream
,
2748 struct stream_subbuffer
*subbuf
)
2752 ret
= ustctl_get_subbuf_size(
2753 stream
->ustream
, &subbuf
->info
.data
.subbuf_size
);
2758 ret
= ustctl_get_padded_subbuf_size(
2759 stream
->ustream
, &subbuf
->info
.data
.padded_subbuf_size
);
2768 static int extract_metadata_subbuffer_info(struct lttng_consumer_stream
*stream
,
2769 struct stream_subbuffer
*subbuf
)
2773 ret
= extract_common_subbuffer_info(stream
, subbuf
);
2778 subbuf
->info
.metadata
.version
= stream
->metadata_version
;
2784 static int extract_data_subbuffer_info(struct lttng_consumer_stream
*stream
,
2785 struct stream_subbuffer
*subbuf
)
2789 ret
= extract_common_subbuffer_info(stream
, subbuf
);
2794 ret
= ustctl_get_packet_size(
2795 stream
->ustream
, &subbuf
->info
.data
.packet_size
);
2797 PERROR("Failed to get sub-buffer packet size");
2801 ret
= ustctl_get_content_size(
2802 stream
->ustream
, &subbuf
->info
.data
.content_size
);
2804 PERROR("Failed to get sub-buffer content size");
2808 ret
= ustctl_get_timestamp_begin(
2809 stream
->ustream
, &subbuf
->info
.data
.timestamp_begin
);
2811 PERROR("Failed to get sub-buffer begin timestamp");
2815 ret
= ustctl_get_timestamp_end(
2816 stream
->ustream
, &subbuf
->info
.data
.timestamp_end
);
2818 PERROR("Failed to get sub-buffer end timestamp");
2822 ret
= ustctl_get_events_discarded(
2823 stream
->ustream
, &subbuf
->info
.data
.events_discarded
);
2825 PERROR("Failed to get sub-buffer events discarded count");
2829 ret
= ustctl_get_sequence_number(stream
->ustream
,
2830 &subbuf
->info
.data
.sequence_number
.value
);
2832 /* May not be supported by older LTTng-modules. */
2833 if (ret
!= -ENOTTY
) {
2834 PERROR("Failed to get sub-buffer sequence number");
2838 subbuf
->info
.data
.sequence_number
.is_set
= true;
2841 ret
= ustctl_get_stream_id(
2842 stream
->ustream
, &subbuf
->info
.data
.stream_id
);
2844 PERROR("Failed to get stream id");
2848 ret
= ustctl_get_instance_id(stream
->ustream
,
2849 &subbuf
->info
.data
.stream_instance_id
.value
);
2851 /* May not be supported by older LTTng-modules. */
2852 if (ret
!= -ENOTTY
) {
2853 PERROR("Failed to get stream instance id");
2857 subbuf
->info
.data
.stream_instance_id
.is_set
= true;
2863 static int get_next_subbuffer_common(struct lttng_consumer_stream
*stream
,
2864 struct stream_subbuffer
*subbuffer
)
2869 ret
= stream
->read_subbuffer_ops
.extract_subbuffer_info(
2875 ret
= get_current_subbuf_addr(stream
, &addr
);
2880 subbuffer
->buffer
.buffer
= lttng_buffer_view_init(
2881 addr
, 0, subbuffer
->info
.data
.padded_subbuf_size
);
2882 assert(subbuffer
->buffer
.buffer
.data
!= NULL
);
2887 static int get_next_subbuffer(struct lttng_consumer_stream
*stream
,
2888 struct stream_subbuffer
*subbuffer
)
2892 ret
= ustctl_get_next_subbuf(stream
->ustream
);
2897 ret
= get_next_subbuffer_common(stream
, subbuffer
);
2905 static int get_next_subbuffer_metadata(struct lttng_consumer_stream
*stream
,
2906 struct stream_subbuffer
*subbuffer
)
2913 unsigned long consumed_pos
, produced_pos
;
2916 ret
= ustctl_get_next_subbuf(stream
->ustream
);
2918 got_subbuffer
= true;
2920 got_subbuffer
= false;
2921 if (ret
!= -EAGAIN
) {
2928 * Determine if the cache is empty and ensure that a sub-buffer
2929 * is made available if the cache is not empty.
2931 if (!got_subbuffer
) {
2932 ret
= commit_one_metadata_packet(stream
);
2933 if (ret
< 0 && ret
!= -ENOBUFS
) {
2935 } else if (ret
== 0) {
2936 /* Not an error, the cache is empty. */
2941 cache_empty
= false;
2944 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2945 cache_empty
= stream
->chan
->metadata_cache
->max_offset
==
2946 stream
->ust_metadata_pushed
;
2947 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2949 } while (!got_subbuffer
);
2951 /* Populate sub-buffer infos and view. */
2952 ret
= get_next_subbuffer_common(stream
, subbuffer
);
2957 ret
= lttng_ustconsumer_sample_snapshot_positions(stream
);
2960 * -EAGAIN is not expected since we got a sub-buffer and haven't
2961 * pushed the consumption position yet (on put_next).
2963 PERROR("Failed to take a snapshot of metadata buffer positions");
2967 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
2969 PERROR("Failed to get metadata consumed position");
2973 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
2975 PERROR("Failed to get metadata produced position");
2979 /* Last sub-buffer of the ring buffer ? */
2980 buffer_empty
= (consumed_pos
+ stream
->max_sb_size
) == produced_pos
;
2983 * The sessiond registry lock ensures that coherent units of metadata
2984 * are pushed to the consumer daemon at once. Hence, if a sub-buffer is
2985 * acquired, the cache is empty, and it is the only available sub-buffer
2986 * available, it is safe to assume that it is "coherent".
2988 coherent
= got_subbuffer
&& cache_empty
&& buffer_empty
;
2990 LTTNG_OPTIONAL_SET(&subbuffer
->info
.metadata
.coherent
, coherent
);
2995 static int put_next_subbuffer(struct lttng_consumer_stream
*stream
,
2996 struct stream_subbuffer
*subbuffer
)
2998 const int ret
= ustctl_put_next_subbuf(stream
->ustream
);
3004 static int signal_metadata(struct lttng_consumer_stream
*stream
,
3005 struct lttng_consumer_local_data
*ctx
)
3007 return pthread_cond_broadcast(&stream
->metadata_rdv
) ? -errno
: 0;
3010 static int lttng_ustconsumer_set_stream_ops(
3011 struct lttng_consumer_stream
*stream
)
3015 stream
->read_subbuffer_ops
.on_wake_up
= consumer_stream_ust_on_wake_up
;
3016 if (stream
->metadata_flag
) {
3017 stream
->read_subbuffer_ops
.get_next_subbuffer
=
3018 get_next_subbuffer_metadata
;
3019 stream
->read_subbuffer_ops
.extract_subbuffer_info
=
3020 extract_metadata_subbuffer_info
;
3021 stream
->read_subbuffer_ops
.reset_metadata
=
3022 metadata_stream_reset_cache_consumed_position
;
3023 if (stream
->chan
->is_live
) {
3024 stream
->read_subbuffer_ops
.on_sleep
= signal_metadata
;
3025 ret
= consumer_stream_enable_metadata_bucketization(
3032 stream
->read_subbuffer_ops
.get_next_subbuffer
=
3034 stream
->read_subbuffer_ops
.extract_subbuffer_info
=
3035 extract_data_subbuffer_info
;
3036 stream
->read_subbuffer_ops
.on_sleep
= notify_if_more_data
;
3037 if (stream
->chan
->is_live
) {
3038 stream
->read_subbuffer_ops
.send_live_beacon
=
3039 consumer_flush_ust_index
;
3043 stream
->read_subbuffer_ops
.put_next_subbuffer
= put_next_subbuffer
;
3049 * Called when a stream is created.
3051 * Return 0 on success or else a negative value.
3053 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3060 * Don't create anything if this is set for streaming or if there is
3061 * no current trace chunk on the parent channel.
3063 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
&&
3064 stream
->chan
->trace_chunk
) {
3065 ret
= consumer_stream_create_output_files(stream
, true);
3071 lttng_ustconsumer_set_stream_ops(stream
);
3079 * Check if data is still being extracted from the buffers for a specific
3080 * stream. Consumer data lock MUST be acquired before calling this function
3081 * and the stream lock.
3083 * Return 1 if the traced data are still getting read else 0 meaning that the
3084 * data is available for trace viewer reading.
3086 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
3091 assert(stream
->ustream
);
3092 ASSERT_LOCKED(stream
->lock
);
3094 DBG("UST consumer checking data pending");
3096 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
3101 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
3102 uint64_t contiguous
, pushed
;
3104 /* Ease our life a bit. */
3105 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
3106 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
3107 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
3108 pushed
= stream
->ust_metadata_pushed
;
3111 * We can simply check whether all contiguously available data
3112 * has been pushed to the ring buffer, since the push operation
3113 * is performed within get_next_subbuf(), and because both
3114 * get_next_subbuf() and put_next_subbuf() are issued atomically
3115 * thanks to the stream lock within
3116 * lttng_ustconsumer_read_subbuffer(). This basically means that
3117 * whetnever ust_metadata_pushed is incremented, the associated
3118 * metadata has been consumed from the metadata stream.
3120 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
3121 contiguous
, pushed
);
3122 assert(((int64_t) (contiguous
- pushed
)) >= 0);
3123 if ((contiguous
!= pushed
) ||
3124 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
3125 ret
= 1; /* Data is pending */
3129 ret
= ustctl_get_next_subbuf(stream
->ustream
);
3132 * There is still data so let's put back this
3135 ret
= ustctl_put_subbuf(stream
->ustream
);
3137 ret
= 1; /* Data is pending */
3142 /* Data is NOT pending so ready to be read. */
3150 * Stop a given metadata channel timer if enabled and close the wait fd which
3151 * is the poll pipe of the metadata stream.
3153 * This MUST be called with the metadata channel lock acquired.
3155 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
3160 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
3162 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
3164 if (metadata
->switch_timer_enabled
== 1) {
3165 consumer_timer_switch_stop(metadata
);
3168 if (!metadata
->metadata_stream
) {
3173 * Closing write side so the thread monitoring the stream wakes up if any
3174 * and clean the metadata stream.
3176 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
3177 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
3179 PERROR("closing metadata pipe write side");
3181 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
3189 * Close every metadata stream wait fd of the metadata hash table. This
3190 * function MUST be used very carefully so not to run into a race between the
3191 * metadata thread handling streams and this function closing their wait fd.
3193 * For UST, this is used when the session daemon hangs up. Its the metadata
3194 * producer so calling this is safe because we are assured that no state change
3195 * can occur in the metadata thread for the streams in the hash table.
3197 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
3199 struct lttng_ht_iter iter
;
3200 struct lttng_consumer_stream
*stream
;
3202 assert(metadata_ht
);
3203 assert(metadata_ht
->ht
);
3205 DBG("UST consumer closing all metadata streams");
3208 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
3211 health_code_update();
3213 pthread_mutex_lock(&stream
->chan
->lock
);
3214 lttng_ustconsumer_close_metadata(stream
->chan
);
3215 pthread_mutex_unlock(&stream
->chan
->lock
);
3221 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
3225 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
3227 ERR("Unable to close wakeup fd");
3232 * Please refer to consumer-timer.c before adding any lock within this
3233 * function or any of its callees. Timers have a very strict locking
3234 * semantic with respect to teardown. Failure to respect this semantic
3235 * introduces deadlocks.
3237 * DON'T hold the metadata lock when calling this function, else this
3238 * can cause deadlock involving consumer awaiting for metadata to be
3239 * pushed out due to concurrent interaction with the session daemon.
3241 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
3242 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
3244 struct lttcomm_metadata_request_msg request
;
3245 struct lttcomm_consumer_msg msg
;
3246 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3247 uint64_t len
, key
, offset
, version
;
3251 assert(channel
->metadata_cache
);
3253 memset(&request
, 0, sizeof(request
));
3255 /* send the metadata request to sessiond */
3256 switch (consumer_data
.type
) {
3257 case LTTNG_CONSUMER64_UST
:
3258 request
.bits_per_long
= 64;
3260 case LTTNG_CONSUMER32_UST
:
3261 request
.bits_per_long
= 32;
3264 request
.bits_per_long
= 0;
3268 request
.session_id
= channel
->session_id
;
3269 request
.session_id_per_pid
= channel
->session_id_per_pid
;
3271 * Request the application UID here so the metadata of that application can
3272 * be sent back. The channel UID corresponds to the user UID of the session
3273 * used for the rights on the stream file(s).
3275 request
.uid
= channel
->ust_app_uid
;
3276 request
.key
= channel
->key
;
3278 DBG("Sending metadata request to sessiond, session id %" PRIu64
3279 ", per-pid %" PRIu64
", app UID %u and channel key %" PRIu64
,
3280 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
3283 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
3285 health_code_update();
3287 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
3290 ERR("Asking metadata to sessiond");
3294 health_code_update();
3296 /* Receive the metadata from sessiond */
3297 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
3299 if (ret
!= sizeof(msg
)) {
3300 DBG("Consumer received unexpected message size %d (expects %zu)",
3302 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
3304 * The ret value might 0 meaning an orderly shutdown but this is ok
3305 * since the caller handles this.
3310 health_code_update();
3312 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
3313 /* No registry found */
3314 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3318 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
3319 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
3324 len
= msg
.u
.push_metadata
.len
;
3325 key
= msg
.u
.push_metadata
.key
;
3326 offset
= msg
.u
.push_metadata
.target_offset
;
3327 version
= msg
.u
.push_metadata
.version
;
3329 assert(key
== channel
->key
);
3331 DBG("No new metadata to receive for key %" PRIu64
, key
);
3334 health_code_update();
3336 /* Tell session daemon we are ready to receive the metadata. */
3337 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3338 LTTCOMM_CONSUMERD_SUCCESS
);
3339 if (ret
< 0 || len
== 0) {
3341 * Somehow, the session daemon is not responding anymore or there is
3342 * nothing to receive.
3347 health_code_update();
3349 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
3350 key
, offset
, len
, version
, channel
, timer
, wait
);
3353 * Only send the status msg if the sessiond is alive meaning a positive
3356 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
3361 health_code_update();
3363 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
3368 * Return the ustctl call for the get stream id.
3370 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
3371 uint64_t *stream_id
)
3376 return ustctl_get_stream_id(stream
->ustream
, stream_id
);