2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
23 #include <lttng/ust-ctl.h>
29 #include <sys/socket.h>
31 #include <sys/types.h>
34 #include <urcu/list.h>
38 #include <bin/lttng-consumerd/health-consumerd.h>
39 #include <common/common.h>
40 #include <common/sessiond-comm/sessiond-comm.h>
41 #include <common/relayd/relayd.h>
42 #include <common/compat/fcntl.h>
43 #include <common/compat/endian.h>
44 #include <common/consumer/consumer-metadata-cache.h>
45 #include <common/consumer/consumer-stream.h>
46 #include <common/consumer/consumer-timer.h>
47 #include <common/utils.h>
48 #include <common/index/index.h>
50 #include "ust-consumer.h"
52 #define INT_MAX_STR_LEN 12 /* includes \0 */
54 extern struct lttng_consumer_global_data consumer_data
;
55 extern int consumer_poll_timeout
;
58 * Free channel object and all streams associated with it. This MUST be used
59 * only and only if the channel has _NEVER_ been added to the global channel
62 static void destroy_channel(struct lttng_consumer_channel
*channel
)
64 struct lttng_consumer_stream
*stream
, *stmp
;
68 DBG("UST consumer cleaning stream list");
70 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
75 cds_list_del(&stream
->send_node
);
76 ustctl_destroy_stream(stream
->ustream
);
77 lttng_trace_chunk_put(stream
->trace_chunk
);
82 * If a channel is available meaning that was created before the streams
86 lttng_ustconsumer_del_channel(channel
);
87 lttng_ustconsumer_free_channel(channel
);
93 * Add channel to internal consumer state.
95 * Returns 0 on success or else a negative value.
97 static int add_channel(struct lttng_consumer_channel
*channel
,
98 struct lttng_consumer_local_data
*ctx
)
105 if (ctx
->on_recv_channel
!= NULL
) {
106 ret
= ctx
->on_recv_channel(channel
);
108 ret
= consumer_add_channel(channel
, ctx
);
109 } else if (ret
< 0) {
110 /* Most likely an ENOMEM. */
111 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
115 ret
= consumer_add_channel(channel
, ctx
);
118 DBG("UST consumer channel added (key: %" PRIu64
")", channel
->key
);
125 * Allocate and return a consumer channel object.
127 static struct lttng_consumer_channel
*allocate_channel(uint64_t session_id
,
128 const uint64_t *chunk_id
, const char *pathname
, const char *name
,
129 uint64_t relayd_id
, uint64_t key
, enum lttng_event_output output
,
130 uint64_t tracefile_size
, uint64_t tracefile_count
,
131 uint64_t session_id_per_pid
, unsigned int monitor
,
132 unsigned int live_timer_interval
,
133 const char *root_shm_path
, const char *shm_path
)
138 return consumer_allocate_channel(key
, session_id
, chunk_id
, pathname
,
139 name
, relayd_id
, output
, tracefile_size
,
140 tracefile_count
, session_id_per_pid
, monitor
,
141 live_timer_interval
, root_shm_path
, shm_path
);
145 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
146 * error value if applicable is set in it else it is kept untouched.
148 * Return NULL on error else the newly allocated stream object.
150 static struct lttng_consumer_stream
*allocate_stream(int cpu
, int key
,
151 struct lttng_consumer_channel
*channel
,
152 struct lttng_consumer_local_data
*ctx
, int *_alloc_ret
)
155 struct lttng_consumer_stream
*stream
= NULL
;
160 stream
= consumer_allocate_stream(
167 channel
->trace_chunk
,
172 if (stream
== NULL
) {
176 * We could not find the channel. Can happen if cpu hotplug
177 * happens while tearing down.
179 DBG3("Could not find channel");
184 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
190 consumer_stream_update_channel_attributes(stream
, channel
);
194 *_alloc_ret
= alloc_ret
;
200 * Send the given stream pointer to the corresponding thread.
202 * Returns 0 on success else a negative value.
204 static int send_stream_to_thread(struct lttng_consumer_stream
*stream
,
205 struct lttng_consumer_local_data
*ctx
)
208 struct lttng_pipe
*stream_pipe
;
210 /* Get the right pipe where the stream will be sent. */
211 if (stream
->metadata_flag
) {
212 consumer_add_metadata_stream(stream
);
213 stream_pipe
= ctx
->consumer_metadata_pipe
;
215 consumer_add_data_stream(stream
);
216 stream_pipe
= ctx
->consumer_data_pipe
;
220 * From this point on, the stream's ownership has been moved away from
221 * the channel and it becomes globally visible. Hence, remove it from
222 * the local stream list to prevent the stream from being both local and
225 stream
->globally_visible
= 1;
226 cds_list_del(&stream
->send_node
);
228 ret
= lttng_pipe_write(stream_pipe
, &stream
, sizeof(stream
));
230 ERR("Consumer write %s stream to pipe %d",
231 stream
->metadata_flag
? "metadata" : "data",
232 lttng_pipe_get_writefd(stream_pipe
));
233 if (stream
->metadata_flag
) {
234 consumer_del_stream_for_metadata(stream
);
236 consumer_del_stream_for_data(stream
);
246 int get_stream_shm_path(char *stream_shm_path
, const char *shm_path
, int cpu
)
248 char cpu_nr
[INT_MAX_STR_LEN
]; /* int max len */
251 strncpy(stream_shm_path
, shm_path
, PATH_MAX
);
252 stream_shm_path
[PATH_MAX
- 1] = '\0';
253 ret
= snprintf(cpu_nr
, INT_MAX_STR_LEN
, "%i", cpu
);
258 strncat(stream_shm_path
, cpu_nr
,
259 PATH_MAX
- strlen(stream_shm_path
) - 1);
266 * Create streams for the given channel using liblttng-ust-ctl.
267 * The channel lock must be acquired by the caller.
269 * Return 0 on success else a negative value.
271 static int create_ust_streams(struct lttng_consumer_channel
*channel
,
272 struct lttng_consumer_local_data
*ctx
)
275 struct ustctl_consumer_stream
*ustream
;
276 struct lttng_consumer_stream
*stream
;
277 pthread_mutex_t
*current_stream_lock
= NULL
;
283 * While a stream is available from ustctl. When NULL is returned, we've
284 * reached the end of the possible stream for the channel.
286 while ((ustream
= ustctl_create_stream(channel
->uchan
, cpu
))) {
288 int ust_metadata_pipe
[2];
290 health_code_update();
292 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& channel
->monitor
) {
293 ret
= utils_create_pipe_cloexec_nonblock(ust_metadata_pipe
);
295 ERR("Create ust metadata poll pipe");
298 wait_fd
= ust_metadata_pipe
[0];
300 wait_fd
= ustctl_stream_get_wait_fd(ustream
);
303 /* Allocate consumer stream object. */
304 stream
= allocate_stream(cpu
, wait_fd
, channel
, ctx
, &ret
);
308 stream
->ustream
= ustream
;
310 * Store it so we can save multiple function calls afterwards since
311 * this value is used heavily in the stream threads. This is UST
312 * specific so this is why it's done after allocation.
314 stream
->wait_fd
= wait_fd
;
317 * Increment channel refcount since the channel reference has now been
318 * assigned in the allocation process above.
320 if (stream
->chan
->monitor
) {
321 uatomic_inc(&stream
->chan
->refcount
);
324 pthread_mutex_lock(&stream
->lock
);
325 current_stream_lock
= &stream
->lock
;
327 * Order is important this is why a list is used. On error, the caller
328 * should clean this list.
330 cds_list_add_tail(&stream
->send_node
, &channel
->streams
.head
);
332 ret
= ustctl_get_max_subbuf_size(stream
->ustream
,
333 &stream
->max_sb_size
);
335 ERR("ustctl_get_max_subbuf_size failed for stream %s",
340 /* Do actions once stream has been received. */
341 if (ctx
->on_recv_stream
) {
342 ret
= ctx
->on_recv_stream(stream
);
348 DBG("UST consumer add stream %s (key: %" PRIu64
") with relayd id %" PRIu64
,
349 stream
->name
, stream
->key
, stream
->relayd_stream_id
);
351 /* Set next CPU stream. */
352 channel
->streams
.count
= ++cpu
;
354 /* Keep stream reference when creating metadata. */
355 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
356 channel
->metadata_stream
= stream
;
357 if (channel
->monitor
) {
358 /* Set metadata poll pipe if we created one */
359 memcpy(stream
->ust_metadata_poll_pipe
,
361 sizeof(ust_metadata_pipe
));
364 pthread_mutex_unlock(&stream
->lock
);
365 current_stream_lock
= NULL
;
372 if (current_stream_lock
) {
373 pthread_mutex_unlock(current_stream_lock
);
379 * create_posix_shm is never called concurrently within a process.
382 int create_posix_shm(void)
384 char tmp_name
[NAME_MAX
];
387 ret
= snprintf(tmp_name
, NAME_MAX
, "/ust-shm-consumer-%d", getpid());
393 * Allocate shm, and immediately unlink its shm oject, keeping
394 * only the file descriptor as a reference to the object.
395 * We specifically do _not_ use the / at the beginning of the
396 * pathname so that some OS implementations can keep it local to
397 * the process (POSIX leaves this implementation-defined).
399 shmfd
= shm_open(tmp_name
, O_CREAT
| O_EXCL
| O_RDWR
, 0700);
404 ret
= shm_unlink(tmp_name
);
405 if (ret
< 0 && errno
!= ENOENT
) {
406 PERROR("shm_unlink");
407 goto error_shm_release
;
420 static int open_ust_stream_fd(struct lttng_consumer_channel
*channel
, int cpu
,
421 const struct lttng_credentials
*session_credentials
)
423 char shm_path
[PATH_MAX
];
426 if (!channel
->shm_path
[0]) {
427 return create_posix_shm();
429 ret
= get_stream_shm_path(shm_path
, channel
->shm_path
, cpu
);
433 return run_as_open(shm_path
,
434 O_RDWR
| O_CREAT
| O_EXCL
, S_IRUSR
| S_IWUSR
,
435 session_credentials
->uid
, session_credentials
->gid
);
442 * Create an UST channel with the given attributes and send it to the session
443 * daemon using the ust ctl API.
445 * Return 0 on success or else a negative value.
447 static int create_ust_channel(struct lttng_consumer_channel
*channel
,
448 struct ustctl_consumer_channel_attr
*attr
,
449 struct ustctl_consumer_channel
**ust_chanp
)
451 int ret
, nr_stream_fds
, i
, j
;
453 struct ustctl_consumer_channel
*ust_channel
;
458 assert(channel
->buffer_credentials
.is_set
);
460 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
461 "subbuf_size: %" PRIu64
", num_subbuf: %" PRIu64
", "
462 "switch_timer_interval: %u, read_timer_interval: %u, "
463 "output: %d, type: %d", attr
->overwrite
, attr
->subbuf_size
,
464 attr
->num_subbuf
, attr
->switch_timer_interval
,
465 attr
->read_timer_interval
, attr
->output
, attr
->type
);
467 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
)
470 nr_stream_fds
= ustctl_get_nr_stream_per_channel();
471 stream_fds
= zmalloc(nr_stream_fds
* sizeof(*stream_fds
));
476 for (i
= 0; i
< nr_stream_fds
; i
++) {
477 stream_fds
[i
] = open_ust_stream_fd(channel
, i
,
478 &channel
->buffer_credentials
.value
);
479 if (stream_fds
[i
] < 0) {
484 ust_channel
= ustctl_create_channel(attr
, stream_fds
, nr_stream_fds
);
489 channel
->nr_stream_fds
= nr_stream_fds
;
490 channel
->stream_fds
= stream_fds
;
491 *ust_chanp
= ust_channel
;
497 for (j
= i
- 1; j
>= 0; j
--) {
500 closeret
= close(stream_fds
[j
]);
504 if (channel
->shm_path
[0]) {
505 char shm_path
[PATH_MAX
];
507 closeret
= get_stream_shm_path(shm_path
,
508 channel
->shm_path
, j
);
510 ERR("Cannot get stream shm path");
512 closeret
= run_as_unlink(shm_path
,
513 channel
->buffer_credentials
.value
.uid
,
514 channel
->buffer_credentials
.value
.gid
);
516 PERROR("unlink %s", shm_path
);
520 /* Try to rmdir all directories under shm_path root. */
521 if (channel
->root_shm_path
[0]) {
522 (void) run_as_rmdir_recursive(channel
->root_shm_path
,
523 channel
->buffer_credentials
.value
.uid
,
524 channel
->buffer_credentials
.value
.gid
,
525 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG
);
533 * Send a single given stream to the session daemon using the sock.
535 * Return 0 on success else a negative value.
537 static int send_sessiond_stream(int sock
, struct lttng_consumer_stream
*stream
)
544 DBG("UST consumer sending stream %" PRIu64
" to sessiond", stream
->key
);
546 /* Send stream to session daemon. */
547 ret
= ustctl_send_stream_to_sessiond(sock
, stream
->ustream
);
557 * Send channel to sessiond and relayd if applicable.
559 * Return 0 on success or else a negative value.
561 static int send_channel_to_sessiond_and_relayd(int sock
,
562 struct lttng_consumer_channel
*channel
,
563 struct lttng_consumer_local_data
*ctx
, int *relayd_error
)
565 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
566 struct lttng_consumer_stream
*stream
;
567 uint64_t net_seq_idx
= -1ULL;
573 DBG("UST consumer sending channel %s to sessiond", channel
->name
);
575 if (channel
->relayd_id
!= (uint64_t) -1ULL) {
576 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
578 health_code_update();
580 /* Try to send the stream to the relayd if one is available. */
581 DBG("Sending stream %" PRIu64
" of channel \"%s\" to relayd",
582 stream
->key
, channel
->name
);
583 ret
= consumer_send_relayd_stream(stream
, stream
->chan
->pathname
);
586 * Flag that the relayd was the problem here probably due to a
587 * communicaton error on the socket.
592 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
594 if (net_seq_idx
== -1ULL) {
595 net_seq_idx
= stream
->net_seq_idx
;
600 /* Inform sessiond that we are about to send channel and streams. */
601 ret
= consumer_send_status_msg(sock
, ret_code
);
602 if (ret
< 0 || ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
604 * Either the session daemon is not responding or the relayd died so we
610 /* Send channel to sessiond. */
611 ret
= ustctl_send_channel_to_sessiond(sock
, channel
->uchan
);
616 ret
= ustctl_channel_close_wakeup_fd(channel
->uchan
);
621 /* The channel was sent successfully to the sessiond at this point. */
622 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
624 health_code_update();
626 /* Send stream to session daemon. */
627 ret
= send_sessiond_stream(sock
, stream
);
633 /* Tell sessiond there is no more stream. */
634 ret
= ustctl_send_stream_to_sessiond(sock
, NULL
);
639 DBG("UST consumer NULL stream sent to sessiond");
644 if (ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
651 * Creates a channel and streams and add the channel it to the channel internal
652 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
655 * Return 0 on success or else, a negative value is returned and the channel
656 * MUST be destroyed by consumer_del_channel().
658 static int ask_channel(struct lttng_consumer_local_data
*ctx
,
659 struct lttng_consumer_channel
*channel
,
660 struct ustctl_consumer_channel_attr
*attr
)
669 * This value is still used by the kernel consumer since for the kernel,
670 * the stream ownership is not IN the consumer so we need to have the
671 * number of left stream that needs to be initialized so we can know when
672 * to delete the channel (see consumer.c).
674 * As for the user space tracer now, the consumer creates and sends the
675 * stream to the session daemon which only sends them to the application
676 * once every stream of a channel is received making this value useless
677 * because we they will be added to the poll thread before the application
678 * receives them. This ensures that a stream can not hang up during
679 * initilization of a channel.
681 channel
->nb_init_stream_left
= 0;
683 /* The reply msg status is handled in the following call. */
684 ret
= create_ust_channel(channel
, attr
, &channel
->uchan
);
689 channel
->wait_fd
= ustctl_channel_get_wait_fd(channel
->uchan
);
692 * For the snapshots (no monitor), we create the metadata streams
693 * on demand, not during the channel creation.
695 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& !channel
->monitor
) {
700 /* Open all streams for this channel. */
701 pthread_mutex_lock(&channel
->lock
);
702 ret
= create_ust_streams(channel
, ctx
);
703 pthread_mutex_unlock(&channel
->lock
);
713 * Send all stream of a channel to the right thread handling it.
715 * On error, return a negative value else 0 on success.
717 static int send_streams_to_thread(struct lttng_consumer_channel
*channel
,
718 struct lttng_consumer_local_data
*ctx
)
721 struct lttng_consumer_stream
*stream
, *stmp
;
726 /* Send streams to the corresponding thread. */
727 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
730 health_code_update();
732 /* Sending the stream to the thread. */
733 ret
= send_stream_to_thread(stream
, ctx
);
736 * If we are unable to send the stream to the thread, there is
737 * a big problem so just stop everything.
748 * Flush channel's streams using the given key to retrieve the channel.
750 * Return 0 on success else an LTTng error code.
752 static int flush_channel(uint64_t chan_key
)
755 struct lttng_consumer_channel
*channel
;
756 struct lttng_consumer_stream
*stream
;
758 struct lttng_ht_iter iter
;
760 DBG("UST consumer flush channel key %" PRIu64
, chan_key
);
763 channel
= consumer_find_channel(chan_key
);
765 ERR("UST consumer flush channel %" PRIu64
" not found", chan_key
);
766 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
770 ht
= consumer_data
.stream_per_chan_id_ht
;
772 /* For each stream of the channel id, flush it. */
773 cds_lfht_for_each_entry_duplicate(ht
->ht
,
774 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
775 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
777 health_code_update();
779 pthread_mutex_lock(&stream
->lock
);
782 * Protect against concurrent teardown of a stream.
784 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
788 if (!stream
->quiescent
) {
789 ustctl_flush_buffer(stream
->ustream
, 0);
790 stream
->quiescent
= true;
793 pthread_mutex_unlock(&stream
->lock
);
801 * Clear quiescent state from channel's streams using the given key to
802 * retrieve the channel.
804 * Return 0 on success else an LTTng error code.
806 static int clear_quiescent_channel(uint64_t chan_key
)
809 struct lttng_consumer_channel
*channel
;
810 struct lttng_consumer_stream
*stream
;
812 struct lttng_ht_iter iter
;
814 DBG("UST consumer clear quiescent channel key %" PRIu64
, chan_key
);
817 channel
= consumer_find_channel(chan_key
);
819 ERR("UST consumer clear quiescent channel %" PRIu64
" not found", chan_key
);
820 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
824 ht
= consumer_data
.stream_per_chan_id_ht
;
826 /* For each stream of the channel id, clear quiescent state. */
827 cds_lfht_for_each_entry_duplicate(ht
->ht
,
828 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
829 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
831 health_code_update();
833 pthread_mutex_lock(&stream
->lock
);
834 stream
->quiescent
= false;
835 pthread_mutex_unlock(&stream
->lock
);
843 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
845 * Return 0 on success else an LTTng error code.
847 static int close_metadata(uint64_t chan_key
)
850 struct lttng_consumer_channel
*channel
;
851 unsigned int channel_monitor
;
853 DBG("UST consumer close metadata key %" PRIu64
, chan_key
);
855 channel
= consumer_find_channel(chan_key
);
858 * This is possible if the metadata thread has issue a delete because
859 * the endpoint point of the stream hung up. There is no way the
860 * session daemon can know about it thus use a DBG instead of an actual
863 DBG("UST consumer close metadata %" PRIu64
" not found", chan_key
);
864 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
868 pthread_mutex_lock(&consumer_data
.lock
);
869 pthread_mutex_lock(&channel
->lock
);
870 channel_monitor
= channel
->monitor
;
871 if (cds_lfht_is_node_deleted(&channel
->node
.node
)) {
875 lttng_ustconsumer_close_metadata(channel
);
876 pthread_mutex_unlock(&channel
->lock
);
877 pthread_mutex_unlock(&consumer_data
.lock
);
880 * The ownership of a metadata channel depends on the type of
881 * session to which it belongs. In effect, the monitor flag is checked
882 * to determine if this metadata channel is in "snapshot" mode or not.
884 * In the non-snapshot case, the metadata channel is created along with
885 * a single stream which will remain present until the metadata channel
886 * is destroyed (on the destruction of its session). In this case, the
887 * metadata stream in "monitored" by the metadata poll thread and holds
888 * the ownership of its channel.
890 * Closing the metadata will cause the metadata stream's "metadata poll
891 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
892 * thread which will teardown the metadata stream which, in return,
893 * deletes the metadata channel.
895 * In the snapshot case, the metadata stream is created and destroyed
896 * on every snapshot record. Since the channel doesn't have an owner
897 * other than the session daemon, it is safe to destroy it immediately
898 * on reception of the CLOSE_METADATA command.
900 if (!channel_monitor
) {
902 * The channel and consumer_data locks must be
903 * released before this call since consumer_del_channel
904 * re-acquires the channel and consumer_data locks to teardown
905 * the channel and queue its reclamation by the "call_rcu"
908 consumer_del_channel(channel
);
913 pthread_mutex_unlock(&channel
->lock
);
914 pthread_mutex_unlock(&consumer_data
.lock
);
920 * RCU read side lock MUST be acquired before calling this function.
922 * Return 0 on success else an LTTng error code.
924 static int setup_metadata(struct lttng_consumer_local_data
*ctx
, uint64_t key
)
927 struct lttng_consumer_channel
*metadata
;
929 DBG("UST consumer setup metadata key %" PRIu64
, key
);
931 metadata
= consumer_find_channel(key
);
933 ERR("UST consumer push metadata %" PRIu64
" not found", key
);
934 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
939 * In no monitor mode, the metadata channel has no stream(s) so skip the
940 * ownership transfer to the metadata thread.
942 if (!metadata
->monitor
) {
943 DBG("Metadata channel in no monitor");
949 * Send metadata stream to relayd if one available. Availability is
950 * known if the stream is still in the list of the channel.
952 if (cds_list_empty(&metadata
->streams
.head
)) {
953 ERR("Metadata channel key %" PRIu64
", no stream available.", key
);
954 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
955 goto error_no_stream
;
958 /* Send metadata stream to relayd if needed. */
959 if (metadata
->metadata_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
960 ret
= consumer_send_relayd_stream(metadata
->metadata_stream
,
963 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
966 ret
= consumer_send_relayd_streams_sent(
967 metadata
->metadata_stream
->net_seq_idx
);
969 ret
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
975 * Ownership of metadata stream is passed along. Freeing is handled by
978 ret
= send_streams_to_thread(metadata
, ctx
);
981 * If we are unable to send the stream to the thread, there is
982 * a big problem so just stop everything.
984 ret
= LTTCOMM_CONSUMERD_FATAL
;
985 goto send_streams_error
;
987 /* List MUST be empty after or else it could be reused. */
988 assert(cds_list_empty(&metadata
->streams
.head
));
995 * Delete metadata channel on error. At this point, the metadata stream can
996 * NOT be monitored by the metadata thread thus having the guarantee that
997 * the stream is still in the local stream list of the channel. This call
998 * will make sure to clean that list.
1000 consumer_stream_destroy(metadata
->metadata_stream
, NULL
);
1001 cds_list_del(&metadata
->metadata_stream
->send_node
);
1002 metadata
->metadata_stream
= NULL
;
1010 * Snapshot the whole metadata.
1011 * RCU read-side lock must be held by the caller.
1013 * Returns 0 on success, < 0 on error
1015 static int snapshot_metadata(struct lttng_consumer_channel
*metadata_channel
,
1016 uint64_t key
, char *path
, uint64_t relayd_id
,
1017 struct lttng_consumer_local_data
*ctx
)
1020 struct lttng_consumer_stream
*metadata_stream
;
1025 DBG("UST consumer snapshot metadata with key %" PRIu64
" at path %s",
1030 assert(!metadata_channel
->monitor
);
1032 health_code_update();
1035 * Ask the sessiond if we have new metadata waiting and update the
1036 * consumer metadata cache.
1038 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 1);
1043 health_code_update();
1046 * The metadata stream is NOT created in no monitor mode when the channel
1047 * is created on a sessiond ask channel command.
1049 ret
= create_ust_streams(metadata_channel
, ctx
);
1054 metadata_stream
= metadata_channel
->metadata_stream
;
1055 assert(metadata_stream
);
1057 pthread_mutex_lock(&metadata_stream
->lock
);
1058 if (relayd_id
!= (uint64_t) -1ULL) {
1059 metadata_stream
->net_seq_idx
= relayd_id
;
1060 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
1062 ret
= consumer_stream_create_output_files(metadata_stream
,
1065 pthread_mutex_unlock(&metadata_stream
->lock
);
1071 health_code_update();
1073 ret
= lttng_consumer_read_subbuffer(metadata_stream
, ctx
);
1081 * Clean up the stream completly because the next snapshot will use a new
1084 consumer_stream_destroy(metadata_stream
, NULL
);
1085 cds_list_del(&metadata_stream
->send_node
);
1086 metadata_channel
->metadata_stream
= NULL
;
1094 int get_current_subbuf_addr(struct lttng_consumer_stream
*stream
,
1098 unsigned long mmap_offset
;
1099 const char *mmap_base
;
1101 mmap_base
= ustctl_get_mmap_base(stream
->ustream
);
1103 ERR("Failed to get mmap base for stream `%s`",
1109 ret
= ustctl_get_mmap_read_offset(stream
->ustream
, &mmap_offset
);
1111 ERR("Failed to get mmap offset for stream `%s`", stream
->name
);
1116 *addr
= mmap_base
+ mmap_offset
;
1123 * Take a snapshot of all the stream of a channel.
1124 * RCU read-side lock and the channel lock must be held by the caller.
1126 * Returns 0 on success, < 0 on error
1128 static int snapshot_channel(struct lttng_consumer_channel
*channel
,
1129 uint64_t key
, char *path
, uint64_t relayd_id
,
1130 uint64_t nb_packets_per_stream
,
1131 struct lttng_consumer_local_data
*ctx
)
1134 unsigned use_relayd
= 0;
1135 unsigned long consumed_pos
, produced_pos
;
1136 struct lttng_consumer_stream
*stream
;
1143 if (relayd_id
!= (uint64_t) -1ULL) {
1147 assert(!channel
->monitor
);
1148 DBG("UST consumer snapshot channel %" PRIu64
, key
);
1150 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
1151 health_code_update();
1153 /* Lock stream because we are about to change its state. */
1154 pthread_mutex_lock(&stream
->lock
);
1155 assert(channel
->trace_chunk
);
1156 if (!lttng_trace_chunk_get(channel
->trace_chunk
)) {
1158 * Can't happen barring an internal error as the channel
1159 * holds a reference to the trace chunk.
1161 ERR("Failed to acquire reference to channel's trace chunk");
1165 assert(!stream
->trace_chunk
);
1166 stream
->trace_chunk
= channel
->trace_chunk
;
1168 stream
->net_seq_idx
= relayd_id
;
1171 ret
= consumer_send_relayd_stream(stream
, path
);
1176 ret
= consumer_stream_create_output_files(stream
,
1181 DBG("UST consumer snapshot stream (%" PRIu64
")",
1186 * If tracing is active, we want to perform a "full" buffer flush.
1187 * Else, if quiescent, it has already been done by the prior stop.
1189 if (!stream
->quiescent
) {
1190 ustctl_flush_buffer(stream
->ustream
, 0);
1193 ret
= lttng_ustconsumer_take_snapshot(stream
);
1195 ERR("Taking UST snapshot");
1199 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
1201 ERR("Produced UST snapshot position");
1205 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1207 ERR("Consumerd UST snapshot position");
1212 * The original value is sent back if max stream size is larger than
1213 * the possible size of the snapshot. Also, we assume that the session
1214 * daemon should never send a maximum stream size that is lower than
1217 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
1218 produced_pos
, nb_packets_per_stream
,
1219 stream
->max_sb_size
);
1221 while ((long) (consumed_pos
- produced_pos
) < 0) {
1223 unsigned long len
, padded_len
;
1224 const char *subbuf_addr
;
1225 struct lttng_buffer_view subbuf_view
;
1227 health_code_update();
1229 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1231 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1233 if (ret
!= -EAGAIN
) {
1234 PERROR("ustctl_get_subbuf snapshot");
1235 goto error_close_stream
;
1237 DBG("UST consumer get subbuf failed. Skipping it.");
1238 consumed_pos
+= stream
->max_sb_size
;
1239 stream
->chan
->lost_packets
++;
1243 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1245 ERR("Snapshot ustctl_get_subbuf_size");
1246 goto error_put_subbuf
;
1249 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1251 ERR("Snapshot ustctl_get_padded_subbuf_size");
1252 goto error_put_subbuf
;
1255 ret
= get_current_subbuf_addr(stream
, &subbuf_addr
);
1257 goto error_put_subbuf
;
1260 subbuf_view
= lttng_buffer_view_init(
1261 subbuf_addr
, 0, padded_len
);
1262 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
,
1263 stream
, &subbuf_view
, padded_len
- len
,
1266 if (read_len
!= len
) {
1268 goto error_put_subbuf
;
1271 if (read_len
!= padded_len
) {
1273 goto error_put_subbuf
;
1277 ret
= ustctl_put_subbuf(stream
->ustream
);
1279 ERR("Snapshot ustctl_put_subbuf");
1280 goto error_close_stream
;
1282 consumed_pos
+= stream
->max_sb_size
;
1285 /* Simply close the stream so we can use it on the next snapshot. */
1286 consumer_stream_close(stream
);
1287 pthread_mutex_unlock(&stream
->lock
);
1294 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1295 ERR("Snapshot ustctl_put_subbuf");
1298 consumer_stream_close(stream
);
1300 pthread_mutex_unlock(&stream
->lock
);
1306 * Receive the metadata updates from the sessiond. Supports receiving
1307 * overlapping metadata, but is needs to always belong to a contiguous
1308 * range starting from 0.
1309 * Be careful about the locks held when calling this function: it needs
1310 * the metadata cache flush to concurrently progress in order to
1313 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1314 uint64_t len
, uint64_t version
,
1315 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
1317 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1320 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1322 metadata_str
= zmalloc(len
* sizeof(char));
1323 if (!metadata_str
) {
1324 PERROR("zmalloc metadata string");
1325 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1329 health_code_update();
1331 /* Receive metadata string. */
1332 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1334 /* Session daemon is dead so return gracefully. */
1339 health_code_update();
1341 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1342 ret
= consumer_metadata_cache_write(channel
, offset
, len
, version
,
1345 /* Unable to handle metadata. Notify session daemon. */
1346 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1348 * Skip metadata flush on write error since the offset and len might
1349 * not have been updated which could create an infinite loop below when
1350 * waiting for the metadata cache to be flushed.
1352 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1355 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1360 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1361 DBG("Waiting for metadata to be flushed");
1363 health_code_update();
1365 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1375 * Receive command from session daemon and process it.
1377 * Return 1 on success else a negative value or 0.
1379 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1380 int sock
, struct pollfd
*consumer_sockpoll
)
1383 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1384 struct lttcomm_consumer_msg msg
;
1385 struct lttng_consumer_channel
*channel
= NULL
;
1387 health_code_update();
1389 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1390 if (ret
!= sizeof(msg
)) {
1391 DBG("Consumer received unexpected message size %zd (expects %zu)",
1394 * The ret value might 0 meaning an orderly shutdown but this is ok
1395 * since the caller handles this.
1398 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1404 health_code_update();
1407 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1409 health_code_update();
1411 /* relayd needs RCU read-side lock */
1414 switch (msg
.cmd_type
) {
1415 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1417 /* Session daemon status message are handled in the following call. */
1418 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1419 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
1420 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
1421 msg
.u
.relayd_sock
.relayd_session_id
);
1424 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1426 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1427 struct consumer_relayd_sock_pair
*relayd
;
1429 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1431 /* Get relayd reference if exists. */
1432 relayd
= consumer_find_relayd(index
);
1433 if (relayd
== NULL
) {
1434 DBG("Unable to find relayd %" PRIu64
, index
);
1435 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1439 * Each relayd socket pair has a refcount of stream attached to it
1440 * which tells if the relayd is still active or not depending on the
1443 * This will set the destroy flag of the relayd object and destroy it
1444 * if the refcount reaches zero when called.
1446 * The destroy can happen either here or when a stream fd hangs up.
1449 consumer_flag_relayd_for_destroy(relayd
);
1452 goto end_msg_sessiond
;
1454 case LTTNG_CONSUMER_UPDATE_STREAM
:
1459 case LTTNG_CONSUMER_DATA_PENDING
:
1461 int ret
, is_data_pending
;
1462 uint64_t id
= msg
.u
.data_pending
.session_id
;
1464 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1466 is_data_pending
= consumer_data_pending(id
);
1468 /* Send back returned value to session daemon */
1469 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1470 sizeof(is_data_pending
));
1472 DBG("Error when sending the data pending ret code: %d", ret
);
1477 * No need to send back a status message since the data pending
1478 * returned value is the response.
1482 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1485 struct ustctl_consumer_channel_attr attr
;
1486 const uint64_t chunk_id
= msg
.u
.ask_channel
.chunk_id
.value
;
1487 const struct lttng_credentials buffer_credentials
= {
1488 .uid
= msg
.u
.ask_channel
.buffer_credentials
.uid
,
1489 .gid
= msg
.u
.ask_channel
.buffer_credentials
.gid
,
1492 /* Create a plain object and reserve a channel key. */
1493 channel
= allocate_channel(msg
.u
.ask_channel
.session_id
,
1494 msg
.u
.ask_channel
.chunk_id
.is_set
?
1496 msg
.u
.ask_channel
.pathname
,
1497 msg
.u
.ask_channel
.name
,
1498 msg
.u
.ask_channel
.relayd_id
,
1499 msg
.u
.ask_channel
.key
,
1500 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1501 msg
.u
.ask_channel
.tracefile_size
,
1502 msg
.u
.ask_channel
.tracefile_count
,
1503 msg
.u
.ask_channel
.session_id_per_pid
,
1504 msg
.u
.ask_channel
.monitor
,
1505 msg
.u
.ask_channel
.live_timer_interval
,
1506 msg
.u
.ask_channel
.root_shm_path
,
1507 msg
.u
.ask_channel
.shm_path
);
1509 goto end_channel_error
;
1512 LTTNG_OPTIONAL_SET(&channel
->buffer_credentials
,
1513 buffer_credentials
);
1516 * Assign UST application UID to the channel. This value is ignored for
1517 * per PID buffers. This is specific to UST thus setting this after the
1520 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1522 /* Build channel attributes from received message. */
1523 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1524 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1525 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1526 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1527 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1528 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1529 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1530 attr
.blocking_timeout
= msg
.u
.ask_channel
.blocking_timeout
;
1532 /* Match channel buffer type to the UST abi. */
1533 switch (msg
.u
.ask_channel
.output
) {
1534 case LTTNG_EVENT_MMAP
:
1536 attr
.output
= LTTNG_UST_MMAP
;
1540 /* Translate and save channel type. */
1541 switch (msg
.u
.ask_channel
.type
) {
1542 case LTTNG_UST_CHAN_PER_CPU
:
1543 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1544 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1546 * Set refcount to 1 for owner. Below, we will
1547 * pass ownership to the
1548 * consumer_thread_channel_poll() thread.
1550 channel
->refcount
= 1;
1552 case LTTNG_UST_CHAN_METADATA
:
1553 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1554 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1561 health_code_update();
1563 ret
= ask_channel(ctx
, channel
, &attr
);
1565 goto end_channel_error
;
1568 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1569 ret
= consumer_metadata_cache_allocate(channel
);
1571 ERR("Allocating metadata cache");
1572 goto end_channel_error
;
1574 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1575 attr
.switch_timer_interval
= 0;
1577 int monitor_start_ret
;
1579 consumer_timer_live_start(channel
,
1580 msg
.u
.ask_channel
.live_timer_interval
);
1581 monitor_start_ret
= consumer_timer_monitor_start(
1583 msg
.u
.ask_channel
.monitor_timer_interval
);
1584 if (monitor_start_ret
< 0) {
1585 ERR("Starting channel monitoring timer failed");
1586 goto end_channel_error
;
1590 health_code_update();
1593 * Add the channel to the internal state AFTER all streams were created
1594 * and successfully sent to session daemon. This way, all streams must
1595 * be ready before this channel is visible to the threads.
1596 * If add_channel succeeds, ownership of the channel is
1597 * passed to consumer_thread_channel_poll().
1599 ret
= add_channel(channel
, ctx
);
1601 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1602 if (channel
->switch_timer_enabled
== 1) {
1603 consumer_timer_switch_stop(channel
);
1605 consumer_metadata_cache_destroy(channel
);
1607 if (channel
->live_timer_enabled
== 1) {
1608 consumer_timer_live_stop(channel
);
1610 if (channel
->monitor_timer_enabled
== 1) {
1611 consumer_timer_monitor_stop(channel
);
1613 goto end_channel_error
;
1616 health_code_update();
1619 * Channel and streams are now created. Inform the session daemon that
1620 * everything went well and should wait to receive the channel and
1621 * streams with ustctl API.
1623 ret
= consumer_send_status_channel(sock
, channel
);
1626 * There is probably a problem on the socket.
1633 case LTTNG_CONSUMER_GET_CHANNEL
:
1635 int ret
, relayd_err
= 0;
1636 uint64_t key
= msg
.u
.get_channel
.key
;
1637 struct lttng_consumer_channel
*channel
;
1639 channel
= consumer_find_channel(key
);
1641 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1642 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1643 goto end_get_channel
;
1646 health_code_update();
1648 /* Send the channel to sessiond (and relayd, if applicable). */
1649 ret
= send_channel_to_sessiond_and_relayd(sock
, channel
, ctx
,
1654 * We were unable to send to the relayd the stream so avoid
1655 * sending back a fatal error to the thread since this is OK
1656 * and the consumer can continue its work. The above call
1657 * has sent the error status message to the sessiond.
1659 goto end_get_channel_nosignal
;
1662 * The communicaton was broken hence there is a bad state between
1663 * the consumer and sessiond so stop everything.
1665 goto error_get_channel_fatal
;
1668 health_code_update();
1671 * In no monitor mode, the streams ownership is kept inside the channel
1672 * so don't send them to the data thread.
1674 if (!channel
->monitor
) {
1675 goto end_get_channel
;
1678 ret
= send_streams_to_thread(channel
, ctx
);
1681 * If we are unable to send the stream to the thread, there is
1682 * a big problem so just stop everything.
1684 goto error_get_channel_fatal
;
1686 /* List MUST be empty after or else it could be reused. */
1687 assert(cds_list_empty(&channel
->streams
.head
));
1689 goto end_msg_sessiond
;
1690 error_get_channel_fatal
:
1692 end_get_channel_nosignal
:
1695 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1697 uint64_t key
= msg
.u
.destroy_channel
.key
;
1700 * Only called if streams have not been sent to stream
1701 * manager thread. However, channel has been sent to
1702 * channel manager thread.
1704 notify_thread_del_channel(ctx
, key
);
1705 goto end_msg_sessiond
;
1707 case LTTNG_CONSUMER_CLOSE_METADATA
:
1711 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1716 goto end_msg_sessiond
;
1718 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1722 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1727 goto end_msg_sessiond
;
1729 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL
:
1733 ret
= clear_quiescent_channel(
1734 msg
.u
.clear_quiescent_channel
.key
);
1739 goto end_msg_sessiond
;
1741 case LTTNG_CONSUMER_PUSH_METADATA
:
1744 uint64_t len
= msg
.u
.push_metadata
.len
;
1745 uint64_t key
= msg
.u
.push_metadata
.key
;
1746 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1747 uint64_t version
= msg
.u
.push_metadata
.version
;
1748 struct lttng_consumer_channel
*channel
;
1750 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1753 channel
= consumer_find_channel(key
);
1756 * This is possible if the metadata creation on the consumer side
1757 * is in flight vis-a-vis a concurrent push metadata from the
1758 * session daemon. Simply return that the channel failed and the
1759 * session daemon will handle that message correctly considering
1760 * that this race is acceptable thus the DBG() statement here.
1762 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1763 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1764 goto end_push_metadata_msg_sessiond
;
1767 health_code_update();
1771 * There is nothing to receive. We have simply
1772 * checked whether the channel can be found.
1774 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1775 goto end_push_metadata_msg_sessiond
;
1778 /* Tell session daemon we are ready to receive the metadata. */
1779 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1781 /* Somehow, the session daemon is not responding anymore. */
1782 goto error_push_metadata_fatal
;
1785 health_code_update();
1787 /* Wait for more data. */
1788 health_poll_entry();
1789 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1792 goto error_push_metadata_fatal
;
1795 health_code_update();
1797 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1798 len
, version
, channel
, 0, 1);
1800 /* error receiving from sessiond */
1801 goto error_push_metadata_fatal
;
1804 goto end_push_metadata_msg_sessiond
;
1806 end_push_metadata_msg_sessiond
:
1807 goto end_msg_sessiond
;
1808 error_push_metadata_fatal
:
1811 case LTTNG_CONSUMER_SETUP_METADATA
:
1815 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1819 goto end_msg_sessiond
;
1821 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1823 struct lttng_consumer_channel
*channel
;
1824 uint64_t key
= msg
.u
.snapshot_channel
.key
;
1826 channel
= consumer_find_channel(key
);
1828 DBG("UST snapshot channel not found for key %" PRIu64
, key
);
1829 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1831 if (msg
.u
.snapshot_channel
.metadata
) {
1832 ret
= snapshot_metadata(channel
, key
,
1833 msg
.u
.snapshot_channel
.pathname
,
1834 msg
.u
.snapshot_channel
.relayd_id
,
1837 ERR("Snapshot metadata failed");
1838 ret_code
= LTTCOMM_CONSUMERD_SNAPSHOT_FAILED
;
1841 ret
= snapshot_channel(channel
, key
,
1842 msg
.u
.snapshot_channel
.pathname
,
1843 msg
.u
.snapshot_channel
.relayd_id
,
1844 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
1847 ERR("Snapshot channel failed");
1848 ret_code
= LTTCOMM_CONSUMERD_SNAPSHOT_FAILED
;
1852 health_code_update();
1853 ret
= consumer_send_status_msg(sock
, ret_code
);
1855 /* Somehow, the session daemon is not responding anymore. */
1858 health_code_update();
1861 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
1864 uint64_t discarded_events
;
1865 struct lttng_ht_iter iter
;
1866 struct lttng_ht
*ht
;
1867 struct lttng_consumer_stream
*stream
;
1868 uint64_t id
= msg
.u
.discarded_events
.session_id
;
1869 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
1871 DBG("UST consumer discarded events command for session id %"
1874 pthread_mutex_lock(&consumer_data
.lock
);
1876 ht
= consumer_data
.stream_list_ht
;
1879 * We only need a reference to the channel, but they are not
1880 * directly indexed, so we just use the first matching stream
1881 * to extract the information we need, we default to 0 if not
1882 * found (no events are dropped if the channel is not yet in
1885 discarded_events
= 0;
1886 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1887 ht
->hash_fct(&id
, lttng_ht_seed
),
1889 &iter
.iter
, stream
, node_session_id
.node
) {
1890 if (stream
->chan
->key
== key
) {
1891 discarded_events
= stream
->chan
->discarded_events
;
1895 pthread_mutex_unlock(&consumer_data
.lock
);
1898 DBG("UST consumer discarded events command for session id %"
1899 PRIu64
", channel key %" PRIu64
, id
, key
);
1901 health_code_update();
1903 /* Send back returned value to session daemon */
1904 ret
= lttcomm_send_unix_sock(sock
, &discarded_events
, sizeof(discarded_events
));
1906 PERROR("send discarded events");
1912 case LTTNG_CONSUMER_LOST_PACKETS
:
1915 uint64_t lost_packets
;
1916 struct lttng_ht_iter iter
;
1917 struct lttng_ht
*ht
;
1918 struct lttng_consumer_stream
*stream
;
1919 uint64_t id
= msg
.u
.lost_packets
.session_id
;
1920 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
1922 DBG("UST consumer lost packets command for session id %"
1925 pthread_mutex_lock(&consumer_data
.lock
);
1927 ht
= consumer_data
.stream_list_ht
;
1930 * We only need a reference to the channel, but they are not
1931 * directly indexed, so we just use the first matching stream
1932 * to extract the information we need, we default to 0 if not
1933 * found (no packets lost if the channel is not yet in use).
1936 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1937 ht
->hash_fct(&id
, lttng_ht_seed
),
1939 &iter
.iter
, stream
, node_session_id
.node
) {
1940 if (stream
->chan
->key
== key
) {
1941 lost_packets
= stream
->chan
->lost_packets
;
1945 pthread_mutex_unlock(&consumer_data
.lock
);
1948 DBG("UST consumer lost packets command for session id %"
1949 PRIu64
", channel key %" PRIu64
, id
, key
);
1951 health_code_update();
1953 /* Send back returned value to session daemon */
1954 ret
= lttcomm_send_unix_sock(sock
, &lost_packets
,
1955 sizeof(lost_packets
));
1957 PERROR("send lost packets");
1963 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
1965 int channel_monitor_pipe
;
1967 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1968 /* Successfully received the command's type. */
1969 ret
= consumer_send_status_msg(sock
, ret_code
);
1974 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
1976 if (ret
!= sizeof(channel_monitor_pipe
)) {
1977 ERR("Failed to receive channel monitor pipe");
1981 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
1982 ret
= consumer_timer_thread_set_channel_monitor_pipe(
1983 channel_monitor_pipe
);
1987 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1988 /* Set the pipe as non-blocking. */
1989 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
1991 PERROR("fcntl get flags of the channel monitoring pipe");
1996 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
1997 flags
| O_NONBLOCK
);
1999 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
2002 DBG("Channel monitor pipe set as non-blocking");
2004 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
2006 goto end_msg_sessiond
;
2008 case LTTNG_CONSUMER_ROTATE_CHANNEL
:
2010 struct lttng_consumer_channel
*channel
;
2011 uint64_t key
= msg
.u
.rotate_channel
.key
;
2013 channel
= consumer_find_channel(key
);
2015 DBG("Channel %" PRIu64
" not found", key
);
2016 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2019 * Sample the rotate position of all the streams in
2022 ret
= lttng_consumer_rotate_channel(channel
, key
,
2023 msg
.u
.rotate_channel
.relayd_id
,
2024 msg
.u
.rotate_channel
.metadata
,
2027 ERR("Rotate channel failed");
2028 ret_code
= LTTCOMM_CONSUMERD_ROTATION_FAIL
;
2031 health_code_update();
2033 ret
= consumer_send_status_msg(sock
, ret_code
);
2035 /* Somehow, the session daemon is not responding anymore. */
2036 goto end_rotate_channel_nosignal
;
2040 * Rotate the streams that are ready right now.
2041 * FIXME: this is a second consecutive iteration over the
2042 * streams in a channel, there is probably a better way to
2043 * handle this, but it needs to be after the
2044 * consumer_send_status_msg() call.
2047 ret
= lttng_consumer_rotate_ready_streams(
2050 ERR("Rotate channel failed");
2054 end_rotate_channel_nosignal
:
2057 case LTTNG_CONSUMER_INIT
:
2059 ret_code
= lttng_consumer_init_command(ctx
,
2060 msg
.u
.init
.sessiond_uuid
);
2061 health_code_update();
2062 ret
= consumer_send_status_msg(sock
, ret_code
);
2064 /* Somehow, the session daemon is not responding anymore. */
2069 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK
:
2071 const struct lttng_credentials credentials
= {
2072 .uid
= msg
.u
.create_trace_chunk
.credentials
.value
.uid
,
2073 .gid
= msg
.u
.create_trace_chunk
.credentials
.value
.gid
,
2075 const bool is_local_trace
=
2076 !msg
.u
.create_trace_chunk
.relayd_id
.is_set
;
2077 const uint64_t relayd_id
=
2078 msg
.u
.create_trace_chunk
.relayd_id
.value
;
2079 const char *chunk_override_name
=
2080 *msg
.u
.create_trace_chunk
.override_name
?
2081 msg
.u
.create_trace_chunk
.override_name
:
2083 LTTNG_OPTIONAL(struct lttng_directory_handle
) chunk_directory_handle
=
2084 LTTNG_OPTIONAL_INIT
;
2087 * The session daemon will only provide a chunk directory file
2088 * descriptor for local traces.
2090 if (is_local_trace
) {
2093 /* Acnowledge the reception of the command. */
2094 ret
= consumer_send_status_msg(sock
,
2095 LTTCOMM_CONSUMERD_SUCCESS
);
2097 /* Somehow, the session daemon is not responding anymore. */
2101 ret
= lttcomm_recv_fds_unix_sock(sock
, &chunk_dirfd
, 1);
2102 if (ret
!= sizeof(chunk_dirfd
)) {
2103 ERR("Failed to receive trace chunk directory file descriptor");
2107 DBG("Received trace chunk directory fd (%d)",
2109 ret
= lttng_directory_handle_init_from_dirfd(
2110 &chunk_directory_handle
.value
,
2113 ERR("Failed to initialize chunk directory handle from directory file descriptor");
2114 if (close(chunk_dirfd
)) {
2115 PERROR("Failed to close chunk directory file descriptor");
2119 chunk_directory_handle
.is_set
= true;
2122 ret_code
= lttng_consumer_create_trace_chunk(
2123 !is_local_trace
? &relayd_id
: NULL
,
2124 msg
.u
.create_trace_chunk
.session_id
,
2125 msg
.u
.create_trace_chunk
.chunk_id
,
2126 (time_t) msg
.u
.create_trace_chunk
2127 .creation_timestamp
,
2128 chunk_override_name
,
2129 msg
.u
.create_trace_chunk
.credentials
.is_set
?
2132 chunk_directory_handle
.is_set
?
2133 &chunk_directory_handle
.value
:
2136 if (chunk_directory_handle
.is_set
) {
2137 lttng_directory_handle_fini(
2138 &chunk_directory_handle
.value
);
2140 goto end_msg_sessiond
;
2142 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK
:
2144 enum lttng_trace_chunk_command_type close_command
=
2145 msg
.u
.close_trace_chunk
.close_command
.value
;
2146 const uint64_t relayd_id
=
2147 msg
.u
.close_trace_chunk
.relayd_id
.value
;
2148 struct lttcomm_consumer_close_trace_chunk_reply reply
;
2149 char closed_trace_chunk_path
[LTTNG_PATH_MAX
];
2152 ret_code
= lttng_consumer_close_trace_chunk(
2153 msg
.u
.close_trace_chunk
.relayd_id
.is_set
?
2156 msg
.u
.close_trace_chunk
.session_id
,
2157 msg
.u
.close_trace_chunk
.chunk_id
,
2158 (time_t) msg
.u
.close_trace_chunk
.close_timestamp
,
2159 msg
.u
.close_trace_chunk
.close_command
.is_set
?
2161 NULL
, closed_trace_chunk_path
);
2162 reply
.ret_code
= ret_code
;
2163 reply
.path_length
= strlen(closed_trace_chunk_path
) + 1;
2164 ret
= lttcomm_send_unix_sock(sock
, &reply
, sizeof(reply
));
2165 if (ret
!= sizeof(reply
)) {
2168 ret
= lttcomm_send_unix_sock(sock
, closed_trace_chunk_path
,
2170 if (ret
!= reply
.path_length
) {
2175 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS
:
2177 const uint64_t relayd_id
=
2178 msg
.u
.trace_chunk_exists
.relayd_id
.value
;
2180 ret_code
= lttng_consumer_trace_chunk_exists(
2181 msg
.u
.trace_chunk_exists
.relayd_id
.is_set
?
2183 msg
.u
.trace_chunk_exists
.session_id
,
2184 msg
.u
.trace_chunk_exists
.chunk_id
);
2185 goto end_msg_sessiond
;
2193 * Return 1 to indicate success since the 0 value can be a socket
2194 * shutdown during the recv() or send() call.
2201 * The returned value here is not useful since either way we'll return 1 to
2202 * the caller because the session daemon socket management is done
2203 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2205 ret
= consumer_send_status_msg(sock
, ret_code
);
2215 * Free channel here since no one has a reference to it. We don't
2216 * free after that because a stream can store this pointer.
2218 destroy_channel(channel
);
2220 /* We have to send a status channel message indicating an error. */
2221 ret
= consumer_send_status_channel(sock
, NULL
);
2223 /* Stop everything if session daemon can not be notified. */
2230 /* This will issue a consumer stop. */
2236 health_code_update();
2240 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream
*stream
,
2241 int producer_active
)
2244 assert(stream
->ustream
);
2246 ustctl_flush_buffer(stream
->ustream
, producer_active
);
2250 * Take a snapshot for a specific stream.
2252 * Returns 0 on success, < 0 on error
2254 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2257 assert(stream
->ustream
);
2259 return ustctl_snapshot(stream
->ustream
);
2263 * Sample consumed and produced positions for a specific stream.
2265 * Returns 0 on success, < 0 on error.
2267 int lttng_ustconsumer_sample_snapshot_positions(
2268 struct lttng_consumer_stream
*stream
)
2271 assert(stream
->ustream
);
2273 return ustctl_snapshot_sample_positions(stream
->ustream
);
2277 * Get the produced position
2279 * Returns 0 on success, < 0 on error
2281 int lttng_ustconsumer_get_produced_snapshot(
2282 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2285 assert(stream
->ustream
);
2288 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
2292 * Get the consumed position
2294 * Returns 0 on success, < 0 on error
2296 int lttng_ustconsumer_get_consumed_snapshot(
2297 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2300 assert(stream
->ustream
);
2303 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
2306 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
2310 assert(stream
->ustream
);
2312 ustctl_flush_buffer(stream
->ustream
, producer
);
2315 int lttng_ustconsumer_get_current_timestamp(
2316 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
2319 assert(stream
->ustream
);
2322 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
2325 int lttng_ustconsumer_get_sequence_number(
2326 struct lttng_consumer_stream
*stream
, uint64_t *seq
)
2329 assert(stream
->ustream
);
2332 return ustctl_get_sequence_number(stream
->ustream
, seq
);
2336 * Called when the stream signals the consumer that it has hung up.
2338 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
2341 assert(stream
->ustream
);
2343 pthread_mutex_lock(&stream
->lock
);
2344 if (!stream
->quiescent
) {
2345 ustctl_flush_buffer(stream
->ustream
, 0);
2346 stream
->quiescent
= true;
2348 pthread_mutex_unlock(&stream
->lock
);
2349 stream
->hangup_flush_done
= 1;
2352 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
2357 assert(chan
->uchan
);
2358 assert(chan
->buffer_credentials
.is_set
);
2360 if (chan
->switch_timer_enabled
== 1) {
2361 consumer_timer_switch_stop(chan
);
2363 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
2366 ret
= close(chan
->stream_fds
[i
]);
2370 if (chan
->shm_path
[0]) {
2371 char shm_path
[PATH_MAX
];
2373 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
2375 ERR("Cannot get stream shm path");
2377 ret
= run_as_unlink(shm_path
,
2378 chan
->buffer_credentials
.value
.uid
,
2379 chan
->buffer_credentials
.value
.gid
);
2381 PERROR("unlink %s", shm_path
);
2387 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
2390 assert(chan
->uchan
);
2391 assert(chan
->buffer_credentials
.is_set
);
2393 consumer_metadata_cache_destroy(chan
);
2394 ustctl_destroy_channel(chan
->uchan
);
2395 /* Try to rmdir all directories under shm_path root. */
2396 if (chan
->root_shm_path
[0]) {
2397 (void) run_as_rmdir_recursive(chan
->root_shm_path
,
2398 chan
->buffer_credentials
.value
.uid
,
2399 chan
->buffer_credentials
.value
.gid
,
2400 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG
);
2402 free(chan
->stream_fds
);
2405 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
2408 assert(stream
->ustream
);
2410 if (stream
->chan
->switch_timer_enabled
== 1) {
2411 consumer_timer_switch_stop(stream
->chan
);
2413 ustctl_destroy_stream(stream
->ustream
);
2416 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
2419 assert(stream
->ustream
);
2421 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
2424 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
2427 assert(stream
->ustream
);
2429 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
2433 * Populate index values of a UST stream. Values are set in big endian order.
2435 * Return 0 on success or else a negative value.
2437 static int get_index_values(struct ctf_packet_index
*index
,
2438 struct ustctl_consumer_stream
*ustream
)
2441 uint64_t packet_size
, content_size
, timestamp_begin
, timestamp_end
,
2442 events_discarded
, stream_id
, stream_instance_id
,
2445 ret
= ustctl_get_timestamp_begin(ustream
, ×tamp_begin
);
2447 PERROR("ustctl_get_timestamp_begin");
2451 ret
= ustctl_get_timestamp_end(ustream
, ×tamp_end
);
2453 PERROR("ustctl_get_timestamp_end");
2457 ret
= ustctl_get_events_discarded(ustream
, &events_discarded
);
2459 PERROR("ustctl_get_events_discarded");
2463 ret
= ustctl_get_content_size(ustream
, &content_size
);
2465 PERROR("ustctl_get_content_size");
2469 ret
= ustctl_get_packet_size(ustream
, &packet_size
);
2471 PERROR("ustctl_get_packet_size");
2475 ret
= ustctl_get_stream_id(ustream
, &stream_id
);
2477 PERROR("ustctl_get_stream_id");
2481 ret
= ustctl_get_instance_id(ustream
, &stream_instance_id
);
2483 PERROR("ustctl_get_instance_id");
2487 ret
= ustctl_get_sequence_number(ustream
, &packet_seq_num
);
2489 PERROR("ustctl_get_sequence_number");
2493 *index
= (typeof(*index
)) {
2494 .offset
= index
->offset
,
2495 .packet_size
= htobe64(packet_size
),
2496 .content_size
= htobe64(content_size
),
2497 .timestamp_begin
= htobe64(timestamp_begin
),
2498 .timestamp_end
= htobe64(timestamp_end
),
2499 .events_discarded
= htobe64(events_discarded
),
2500 .stream_id
= htobe64(stream_id
),
2501 .stream_instance_id
= htobe64(stream_instance_id
),
2502 .packet_seq_num
= htobe64(packet_seq_num
),
2510 void metadata_stream_reset_cache(struct lttng_consumer_stream
*stream
,
2511 struct consumer_metadata_cache
*cache
)
2513 DBG("Metadata stream update to version %" PRIu64
,
2515 stream
->ust_metadata_pushed
= 0;
2516 stream
->metadata_version
= cache
->version
;
2517 stream
->reset_metadata_flag
= 1;
2521 * Check if the version of the metadata stream and metadata cache match.
2522 * If the cache got updated, reset the metadata stream.
2523 * The stream lock and metadata cache lock MUST be held.
2524 * Return 0 on success, a negative value on error.
2527 int metadata_stream_check_version(struct lttng_consumer_stream
*stream
)
2530 struct consumer_metadata_cache
*cache
= stream
->chan
->metadata_cache
;
2532 if (cache
->version
== stream
->metadata_version
) {
2535 metadata_stream_reset_cache(stream
, cache
);
2542 * Write up to one packet from the metadata cache to the channel.
2544 * Returns the number of bytes pushed in the cache, or a negative value
2548 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
2553 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2554 ret
= metadata_stream_check_version(stream
);
2558 if (stream
->chan
->metadata_cache
->max_offset
2559 == stream
->ust_metadata_pushed
) {
2564 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
2565 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
2566 stream
->chan
->metadata_cache
->max_offset
2567 - stream
->ust_metadata_pushed
);
2568 assert(write_len
!= 0);
2569 if (write_len
< 0) {
2570 ERR("Writing one metadata packet");
2574 stream
->ust_metadata_pushed
+= write_len
;
2576 assert(stream
->chan
->metadata_cache
->max_offset
>=
2577 stream
->ust_metadata_pushed
);
2581 * Switch packet (but don't open the next one) on every commit of
2582 * a metadata packet. Since the subbuffer is fully filled (with padding,
2583 * if needed), the stream is "quiescent" after this commit.
2585 ustctl_flush_buffer(stream
->ustream
, 1);
2586 stream
->quiescent
= true;
2588 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2594 * Sync metadata meaning request them to the session daemon and snapshot to the
2595 * metadata thread can consumer them.
2597 * Metadata stream lock is held here, but we need to release it when
2598 * interacting with sessiond, else we cause a deadlock with live
2599 * awaiting on metadata to be pushed out.
2601 * The RCU read side lock must be held by the caller.
2603 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2604 * is empty or a negative value on error.
2606 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data
*ctx
,
2607 struct lttng_consumer_stream
*metadata_stream
)
2611 struct lttng_consumer_channel
*metadata_channel
;
2614 assert(metadata_stream
);
2616 metadata_channel
= metadata_stream
->chan
;
2617 pthread_mutex_unlock(&metadata_stream
->lock
);
2619 * Request metadata from the sessiond, but don't wait for the flush
2620 * because we locked the metadata thread.
2622 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 0);
2623 pthread_mutex_lock(&metadata_stream
->lock
);
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
);
2646 ret
= commit_one_metadata_packet(metadata_stream
);
2649 } else if (ret
> 0) {
2653 ret
= ustctl_snapshot(metadata_stream
->ustream
);
2655 if (errno
!= EAGAIN
) {
2656 ERR("Sync metadata, taking UST snapshot");
2659 DBG("No new metadata when syncing them.");
2660 /* No new metadata, exit. */
2666 * After this flush, we still need to extract metadata.
2677 * Return 0 on success else a negative value.
2679 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2680 struct lttng_consumer_local_data
*ctx
)
2683 struct ustctl_consumer_stream
*ustream
;
2688 ustream
= stream
->ustream
;
2691 * First, we are going to check if there is a new subbuffer available
2692 * before reading the stream wait_fd.
2694 /* Get the next subbuffer */
2695 ret
= ustctl_get_next_subbuf(ustream
);
2697 /* No more data found, flag the stream. */
2698 stream
->has_data
= 0;
2703 ret
= ustctl_put_subbuf(ustream
);
2706 /* This stream still has data. Flag it and wake up the data thread. */
2707 stream
->has_data
= 1;
2709 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2712 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2713 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2718 /* The wake up pipe has been notified. */
2719 ctx
->has_wakeup
= 1;
2728 int update_stream_stats(struct lttng_consumer_stream
*stream
)
2731 uint64_t seq
, discarded
;
2733 ret
= ustctl_get_sequence_number(stream
->ustream
, &seq
);
2735 PERROR("ustctl_get_sequence_number");
2739 * Start the sequence when we extract the first packet in case we don't
2740 * start at 0 (for example if a consumer is not connected to the
2741 * session immediately after the beginning).
2743 if (stream
->last_sequence_number
== -1ULL) {
2744 stream
->last_sequence_number
= seq
;
2745 } else if (seq
> stream
->last_sequence_number
) {
2746 stream
->chan
->lost_packets
+= seq
-
2747 stream
->last_sequence_number
- 1;
2749 /* seq <= last_sequence_number */
2750 ERR("Sequence number inconsistent : prev = %" PRIu64
2751 ", current = %" PRIu64
,
2752 stream
->last_sequence_number
, seq
);
2756 stream
->last_sequence_number
= seq
;
2758 ret
= ustctl_get_events_discarded(stream
->ustream
, &discarded
);
2760 PERROR("kernctl_get_events_discarded");
2763 if (discarded
< stream
->last_discarded_events
) {
2765 * Overflow has occurred. We assume only one wrap-around
2768 stream
->chan
->discarded_events
+=
2769 (1ULL << (CAA_BITS_PER_LONG
- 1)) -
2770 stream
->last_discarded_events
+ discarded
;
2772 stream
->chan
->discarded_events
+= discarded
-
2773 stream
->last_discarded_events
;
2775 stream
->last_discarded_events
= discarded
;
2783 * Read subbuffer from the given stream.
2785 * Stream and channel locks MUST be acquired by the caller.
2787 * Return 0 on success else a negative value.
2789 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
2790 struct lttng_consumer_local_data
*ctx
)
2792 unsigned long len
, subbuf_size
, padding
;
2793 int err
, write_index
= 1, rotation_ret
;
2795 struct ustctl_consumer_stream
*ustream
;
2796 struct ctf_packet_index index
;
2797 const char *subbuf_addr
;
2798 struct lttng_buffer_view subbuf_view
;
2801 assert(stream
->ustream
);
2804 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream
->wait_fd
,
2807 /* Ease our life for what's next. */
2808 ustream
= stream
->ustream
;
2811 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2812 * error if we cannot read this one byte (read returns 0), or if the error
2813 * is EAGAIN or EWOULDBLOCK.
2815 * This is only done when the stream is monitored by a thread, before the
2816 * flush is done after a hangup and if the stream is not flagged with data
2817 * since there might be nothing to consume in the wait fd but still have
2818 * data available flagged by the consumer wake up pipe.
2820 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2824 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2825 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2832 * If the stream was flagged to be ready for rotation before we extract the
2833 * next packet, rotate it now.
2835 if (stream
->rotate_ready
) {
2836 DBG("Rotate stream before extracting data");
2837 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
);
2838 if (rotation_ret
< 0) {
2839 ERR("Stream rotation error");
2846 /* Get the next subbuffer */
2847 err
= ustctl_get_next_subbuf(ustream
);
2850 * Populate metadata info if the existing info has
2851 * already been read.
2853 if (stream
->metadata_flag
) {
2854 ret
= commit_one_metadata_packet(stream
);
2861 ret
= err
; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2863 * This is a debug message even for single-threaded consumer,
2864 * because poll() have more relaxed criterions than get subbuf,
2865 * so get_subbuf may fail for short race windows where poll()
2866 * would issue wakeups.
2868 DBG("Reserving sub buffer failed (everything is normal, "
2869 "it is due to concurrency) [ret: %d]", err
);
2872 assert(stream
->chan
->output
== CONSUMER_CHANNEL_MMAP
);
2874 if (!stream
->metadata_flag
) {
2875 index
.offset
= htobe64(stream
->out_fd_offset
);
2876 ret
= get_index_values(&index
, ustream
);
2878 err
= ustctl_put_subbuf(ustream
);
2883 /* Update the stream's sequence and discarded events count. */
2884 ret
= update_stream_stats(stream
);
2886 PERROR("kernctl_get_events_discarded");
2887 err
= ustctl_put_subbuf(ustream
);
2895 /* Get the full padded subbuffer size */
2896 err
= ustctl_get_padded_subbuf_size(ustream
, &len
);
2899 /* Get subbuffer data size (without padding) */
2900 err
= ustctl_get_subbuf_size(ustream
, &subbuf_size
);
2903 /* Make sure we don't get a subbuffer size bigger than the padded */
2904 assert(len
>= subbuf_size
);
2906 padding
= len
- subbuf_size
;
2908 ret
= get_current_subbuf_addr(stream
, &subbuf_addr
);
2911 goto error_put_subbuf
;
2914 subbuf_view
= lttng_buffer_view_init(subbuf_addr
, 0, len
);
2916 /* write the subbuffer to the tracefile */
2917 ret
= lttng_consumer_on_read_subbuffer_mmap(
2918 ctx
, stream
, &subbuf_view
, padding
, &index
);
2920 * The mmap operation should write subbuf_size amount of data when
2921 * network streaming or the full padding (len) size when we are _not_
2924 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
2925 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
2927 * Display the error but continue processing to try to release the
2928 * subbuffer. This is a DBG statement since any unexpected kill or
2929 * signal, the application gets unregistered, relayd gets closed or
2930 * anything that affects the buffer lifetime will trigger this error.
2931 * So, for the sake of the user, don't print this error since it can
2932 * happen and it is OK with the code flow.
2934 DBG("Error writing to tracefile "
2935 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2936 ret
, len
, subbuf_size
);
2940 err
= ustctl_put_next_subbuf(ustream
);
2944 * This will consumer the byte on the wait_fd if and only if there is not
2945 * next subbuffer to be acquired.
2947 if (!stream
->metadata_flag
) {
2948 ret
= notify_if_more_data(stream
, ctx
);
2954 /* Write index if needed. */
2959 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
2961 * In live, block until all the metadata is sent.
2963 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2964 assert(!stream
->missed_metadata_flush
);
2965 stream
->waiting_on_metadata
= true;
2966 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2968 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
2970 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2971 stream
->waiting_on_metadata
= false;
2972 if (stream
->missed_metadata_flush
) {
2973 stream
->missed_metadata_flush
= false;
2974 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2975 (void) consumer_flush_ust_index(stream
);
2977 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2985 assert(!stream
->metadata_flag
);
2986 err
= consumer_stream_write_index(stream
, &index
);
2993 * After extracting the packet, we check if the stream is now ready to be
2994 * rotated and perform the action immediately.
2996 rotation_ret
= lttng_consumer_stream_is_rotate_ready(stream
);
2997 if (rotation_ret
== 1) {
2998 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
);
2999 if (rotation_ret
< 0) {
3000 ERR("Stream rotation error");
3004 } else if (rotation_ret
< 0) {
3005 ERR("Checking if stream is ready to rotate");
3014 * Called when a stream is created.
3016 * Return 0 on success or else a negative value.
3018 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3025 * Don't create anything if this is set for streaming or if there is
3026 * no current trace chunk on the parent channel.
3028 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
&&
3029 stream
->chan
->trace_chunk
) {
3030 ret
= consumer_stream_create_output_files(stream
, true);
3042 * Check if data is still being extracted from the buffers for a specific
3043 * stream. Consumer data lock MUST be acquired before calling this function
3044 * and the stream lock.
3046 * Return 1 if the traced data are still getting read else 0 meaning that the
3047 * data is available for trace viewer reading.
3049 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
3054 assert(stream
->ustream
);
3056 DBG("UST consumer checking data pending");
3058 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
3063 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
3064 uint64_t contiguous
, pushed
;
3066 /* Ease our life a bit. */
3067 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
3068 pushed
= stream
->ust_metadata_pushed
;
3071 * We can simply check whether all contiguously available data
3072 * has been pushed to the ring buffer, since the push operation
3073 * is performed within get_next_subbuf(), and because both
3074 * get_next_subbuf() and put_next_subbuf() are issued atomically
3075 * thanks to the stream lock within
3076 * lttng_ustconsumer_read_subbuffer(). This basically means that
3077 * whetnever ust_metadata_pushed is incremented, the associated
3078 * metadata has been consumed from the metadata stream.
3080 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
3081 contiguous
, pushed
);
3082 assert(((int64_t) (contiguous
- pushed
)) >= 0);
3083 if ((contiguous
!= pushed
) ||
3084 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
3085 ret
= 1; /* Data is pending */
3089 ret
= ustctl_get_next_subbuf(stream
->ustream
);
3092 * There is still data so let's put back this
3095 ret
= ustctl_put_subbuf(stream
->ustream
);
3097 ret
= 1; /* Data is pending */
3102 /* Data is NOT pending so ready to be read. */
3110 * Stop a given metadata channel timer if enabled and close the wait fd which
3111 * is the poll pipe of the metadata stream.
3113 * This MUST be called with the metadata channel lock acquired.
3115 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
3120 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
3122 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
3124 if (metadata
->switch_timer_enabled
== 1) {
3125 consumer_timer_switch_stop(metadata
);
3128 if (!metadata
->metadata_stream
) {
3133 * Closing write side so the thread monitoring the stream wakes up if any
3134 * and clean the metadata stream.
3136 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
3137 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
3139 PERROR("closing metadata pipe write side");
3141 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
3149 * Close every metadata stream wait fd of the metadata hash table. This
3150 * function MUST be used very carefully so not to run into a race between the
3151 * metadata thread handling streams and this function closing their wait fd.
3153 * For UST, this is used when the session daemon hangs up. Its the metadata
3154 * producer so calling this is safe because we are assured that no state change
3155 * can occur in the metadata thread for the streams in the hash table.
3157 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
3159 struct lttng_ht_iter iter
;
3160 struct lttng_consumer_stream
*stream
;
3162 assert(metadata_ht
);
3163 assert(metadata_ht
->ht
);
3165 DBG("UST consumer closing all metadata streams");
3168 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
3171 health_code_update();
3173 pthread_mutex_lock(&stream
->chan
->lock
);
3174 lttng_ustconsumer_close_metadata(stream
->chan
);
3175 pthread_mutex_unlock(&stream
->chan
->lock
);
3181 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
3185 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
3187 ERR("Unable to close wakeup fd");
3192 * Please refer to consumer-timer.c before adding any lock within this
3193 * function or any of its callees. Timers have a very strict locking
3194 * semantic with respect to teardown. Failure to respect this semantic
3195 * introduces deadlocks.
3197 * DON'T hold the metadata lock when calling this function, else this
3198 * can cause deadlock involving consumer awaiting for metadata to be
3199 * pushed out due to concurrent interaction with the session daemon.
3201 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
3202 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
3204 struct lttcomm_metadata_request_msg request
;
3205 struct lttcomm_consumer_msg msg
;
3206 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3207 uint64_t len
, key
, offset
, version
;
3211 assert(channel
->metadata_cache
);
3213 memset(&request
, 0, sizeof(request
));
3215 /* send the metadata request to sessiond */
3216 switch (consumer_data
.type
) {
3217 case LTTNG_CONSUMER64_UST
:
3218 request
.bits_per_long
= 64;
3220 case LTTNG_CONSUMER32_UST
:
3221 request
.bits_per_long
= 32;
3224 request
.bits_per_long
= 0;
3228 request
.session_id
= channel
->session_id
;
3229 request
.session_id_per_pid
= channel
->session_id_per_pid
;
3231 * Request the application UID here so the metadata of that application can
3232 * be sent back. The channel UID corresponds to the user UID of the session
3233 * used for the rights on the stream file(s).
3235 request
.uid
= channel
->ust_app_uid
;
3236 request
.key
= channel
->key
;
3238 DBG("Sending metadata request to sessiond, session id %" PRIu64
3239 ", per-pid %" PRIu64
", app UID %u and channel key %" PRIu64
,
3240 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
3243 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
3245 health_code_update();
3247 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
3250 ERR("Asking metadata to sessiond");
3254 health_code_update();
3256 /* Receive the metadata from sessiond */
3257 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
3259 if (ret
!= sizeof(msg
)) {
3260 DBG("Consumer received unexpected message size %d (expects %zu)",
3262 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
3264 * The ret value might 0 meaning an orderly shutdown but this is ok
3265 * since the caller handles this.
3270 health_code_update();
3272 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
3273 /* No registry found */
3274 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3278 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
3279 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
3284 len
= msg
.u
.push_metadata
.len
;
3285 key
= msg
.u
.push_metadata
.key
;
3286 offset
= msg
.u
.push_metadata
.target_offset
;
3287 version
= msg
.u
.push_metadata
.version
;
3289 assert(key
== channel
->key
);
3291 DBG("No new metadata to receive for key %" PRIu64
, key
);
3294 health_code_update();
3296 /* Tell session daemon we are ready to receive the metadata. */
3297 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3298 LTTCOMM_CONSUMERD_SUCCESS
);
3299 if (ret
< 0 || len
== 0) {
3301 * Somehow, the session daemon is not responding anymore or there is
3302 * nothing to receive.
3307 health_code_update();
3309 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
3310 key
, offset
, len
, version
, channel
, timer
, wait
);
3313 * Only send the status msg if the sessiond is alive meaning a positive
3316 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
3321 health_code_update();
3323 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
3328 * Return the ustctl call for the get stream id.
3330 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
3331 uint64_t *stream_id
)
3336 return ustctl_get_stream_id(stream
->ustream
, stream_id
);