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.
22 #include <lttng/ust-ctl.h>
28 #include <sys/socket.h>
30 #include <sys/types.h>
33 #include <urcu/list.h>
37 #include <bin/lttng-consumerd/health-consumerd.h>
38 #include <common/common.h>
39 #include <common/sessiond-comm/sessiond-comm.h>
40 #include <common/relayd/relayd.h>
41 #include <common/compat/fcntl.h>
42 #include <common/compat/endian.h>
43 #include <common/consumer/consumer-metadata-cache.h>
44 #include <common/consumer/consumer-stream.h>
45 #include <common/consumer/consumer-timer.h>
46 #include <common/utils.h>
47 #include <common/index/index.h>
49 #include "ust-consumer.h"
51 #define INT_MAX_STR_LEN 12 /* includes \0 */
53 extern struct lttng_consumer_global_data consumer_data
;
54 extern int consumer_poll_timeout
;
57 * Free channel object and all streams associated with it. This MUST be used
58 * only and only if the channel has _NEVER_ been added to the global channel
61 static void destroy_channel(struct lttng_consumer_channel
*channel
)
63 struct lttng_consumer_stream
*stream
, *stmp
;
67 DBG("UST consumer cleaning stream list");
69 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
74 cds_list_del(&stream
->send_node
);
75 ustctl_destroy_stream(stream
->ustream
);
80 * If a channel is available meaning that was created before the streams
84 lttng_ustconsumer_del_channel(channel
);
85 lttng_ustconsumer_free_channel(channel
);
91 * Add channel to internal consumer state.
93 * Returns 0 on success or else a negative value.
95 static int add_channel(struct lttng_consumer_channel
*channel
,
96 struct lttng_consumer_local_data
*ctx
)
103 if (ctx
->on_recv_channel
!= NULL
) {
104 ret
= ctx
->on_recv_channel(channel
);
106 ret
= consumer_add_channel(channel
, ctx
);
107 } else if (ret
< 0) {
108 /* Most likely an ENOMEM. */
109 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
113 ret
= consumer_add_channel(channel
, ctx
);
116 DBG("UST consumer channel added (key: %" PRIu64
")", channel
->key
);
123 * Allocate and return a consumer channel object.
125 static struct lttng_consumer_channel
*allocate_channel(uint64_t session_id
,
126 const char *pathname
, const char *name
, uid_t uid
, gid_t gid
,
127 uint64_t relayd_id
, uint64_t key
, enum lttng_event_output output
,
128 uint64_t tracefile_size
, uint64_t tracefile_count
,
129 uint64_t session_id_per_pid
, unsigned int monitor
,
130 unsigned int live_timer_interval
,
131 const char *root_shm_path
, const char *shm_path
)
136 return consumer_allocate_channel(key
, session_id
, pathname
, name
, uid
,
137 gid
, relayd_id
, output
, tracefile_size
,
138 tracefile_count
, session_id_per_pid
, monitor
,
139 live_timer_interval
, root_shm_path
, shm_path
);
143 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
144 * error value if applicable is set in it else it is kept untouched.
146 * Return NULL on error else the newly allocated stream object.
148 static struct lttng_consumer_stream
*allocate_stream(int cpu
, int key
,
149 struct lttng_consumer_channel
*channel
,
150 struct lttng_consumer_local_data
*ctx
, int *_alloc_ret
)
153 struct lttng_consumer_stream
*stream
= NULL
;
158 stream
= consumer_allocate_stream(channel
->key
,
160 LTTNG_CONSUMER_ACTIVE_STREAM
,
170 if (stream
== NULL
) {
174 * We could not find the channel. Can happen if cpu hotplug
175 * happens while tearing down.
177 DBG3("Could not find channel");
182 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
188 consumer_stream_update_channel_attributes(stream
, channel
);
189 stream
->chan
= channel
;
193 *_alloc_ret
= alloc_ret
;
199 * Send the given stream pointer to the corresponding thread.
201 * Returns 0 on success else a negative value.
203 static int send_stream_to_thread(struct lttng_consumer_stream
*stream
,
204 struct lttng_consumer_local_data
*ctx
)
207 struct lttng_pipe
*stream_pipe
;
209 /* Get the right pipe where the stream will be sent. */
210 if (stream
->metadata_flag
) {
211 consumer_add_metadata_stream(stream
);
212 stream_pipe
= ctx
->consumer_metadata_pipe
;
214 consumer_add_data_stream(stream
);
215 stream_pipe
= ctx
->consumer_data_pipe
;
219 * From this point on, the stream's ownership has been moved away from
220 * the channel and it becomes globally visible. Hence, remove it from
221 * the local stream list to prevent the stream from being both local and
224 stream
->globally_visible
= 1;
225 cds_list_del(&stream
->send_node
);
227 ret
= lttng_pipe_write(stream_pipe
, &stream
, sizeof(stream
));
229 ERR("Consumer write %s stream to pipe %d",
230 stream
->metadata_flag
? "metadata" : "data",
231 lttng_pipe_get_writefd(stream_pipe
));
232 if (stream
->metadata_flag
) {
233 consumer_del_stream_for_metadata(stream
);
235 consumer_del_stream_for_data(stream
);
245 int get_stream_shm_path(char *stream_shm_path
, const char *shm_path
, int cpu
)
247 char cpu_nr
[INT_MAX_STR_LEN
]; /* int max len */
250 strncpy(stream_shm_path
, shm_path
, PATH_MAX
);
251 stream_shm_path
[PATH_MAX
- 1] = '\0';
252 ret
= snprintf(cpu_nr
, INT_MAX_STR_LEN
, "%i", cpu
);
257 strncat(stream_shm_path
, cpu_nr
,
258 PATH_MAX
- strlen(stream_shm_path
) - 1);
265 * Create streams for the given channel using liblttng-ust-ctl.
267 * Return 0 on success else a negative value.
269 static int create_ust_streams(struct lttng_consumer_channel
*channel
,
270 struct lttng_consumer_local_data
*ctx
)
273 struct ustctl_consumer_stream
*ustream
;
274 struct lttng_consumer_stream
*stream
;
280 * While a stream is available from ustctl. When NULL is returned, we've
281 * reached the end of the possible stream for the channel.
283 while ((ustream
= ustctl_create_stream(channel
->uchan
, cpu
))) {
285 int ust_metadata_pipe
[2];
287 health_code_update();
289 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& channel
->monitor
) {
290 ret
= utils_create_pipe_cloexec_nonblock(ust_metadata_pipe
);
292 ERR("Create ust metadata poll pipe");
295 wait_fd
= ust_metadata_pipe
[0];
297 wait_fd
= ustctl_stream_get_wait_fd(ustream
);
300 /* Allocate consumer stream object. */
301 stream
= allocate_stream(cpu
, wait_fd
, channel
, ctx
, &ret
);
305 stream
->ustream
= ustream
;
307 * Store it so we can save multiple function calls afterwards since
308 * this value is used heavily in the stream threads. This is UST
309 * specific so this is why it's done after allocation.
311 stream
->wait_fd
= wait_fd
;
314 * Increment channel refcount since the channel reference has now been
315 * assigned in the allocation process above.
317 if (stream
->chan
->monitor
) {
318 uatomic_inc(&stream
->chan
->refcount
);
322 * Order is important this is why a list is used. On error, the caller
323 * should clean this list.
325 cds_list_add_tail(&stream
->send_node
, &channel
->streams
.head
);
327 ret
= ustctl_get_max_subbuf_size(stream
->ustream
,
328 &stream
->max_sb_size
);
330 ERR("ustctl_get_max_subbuf_size failed for stream %s",
335 /* Do actions once stream has been received. */
336 if (ctx
->on_recv_stream
) {
337 ret
= ctx
->on_recv_stream(stream
);
343 DBG("UST consumer add stream %s (key: %" PRIu64
") with relayd id %" PRIu64
,
344 stream
->name
, stream
->key
, stream
->relayd_stream_id
);
346 /* Set next CPU stream. */
347 channel
->streams
.count
= ++cpu
;
349 /* Keep stream reference when creating metadata. */
350 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
351 channel
->metadata_stream
= stream
;
352 if (channel
->monitor
) {
353 /* Set metadata poll pipe if we created one */
354 memcpy(stream
->ust_metadata_poll_pipe
,
356 sizeof(ust_metadata_pipe
));
369 * create_posix_shm is never called concurrently within a process.
372 int create_posix_shm(void)
374 char tmp_name
[NAME_MAX
];
377 ret
= snprintf(tmp_name
, NAME_MAX
, "/ust-shm-consumer-%d", getpid());
383 * Allocate shm, and immediately unlink its shm oject, keeping
384 * only the file descriptor as a reference to the object.
385 * We specifically do _not_ use the / at the beginning of the
386 * pathname so that some OS implementations can keep it local to
387 * the process (POSIX leaves this implementation-defined).
389 shmfd
= shm_open(tmp_name
, O_CREAT
| O_EXCL
| O_RDWR
, 0700);
394 ret
= shm_unlink(tmp_name
);
395 if (ret
< 0 && errno
!= ENOENT
) {
396 PERROR("shm_unlink");
397 goto error_shm_release
;
410 static int open_ust_stream_fd(struct lttng_consumer_channel
*channel
,
411 struct ustctl_consumer_channel_attr
*attr
,
414 char shm_path
[PATH_MAX
];
417 if (!channel
->shm_path
[0]) {
418 return create_posix_shm();
420 ret
= get_stream_shm_path(shm_path
, channel
->shm_path
, cpu
);
424 return run_as_open(shm_path
,
425 O_RDWR
| O_CREAT
| O_EXCL
, S_IRUSR
| S_IWUSR
,
426 channel
->uid
, channel
->gid
);
433 * Create an UST channel with the given attributes and send it to the session
434 * daemon using the ust ctl API.
436 * Return 0 on success or else a negative value.
438 static int create_ust_channel(struct lttng_consumer_channel
*channel
,
439 struct ustctl_consumer_channel_attr
*attr
,
440 struct ustctl_consumer_channel
**ust_chanp
)
442 int ret
, nr_stream_fds
, i
, j
;
444 struct ustctl_consumer_channel
*ust_channel
;
450 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
451 "subbuf_size: %" PRIu64
", num_subbuf: %" PRIu64
", "
452 "switch_timer_interval: %u, read_timer_interval: %u, "
453 "output: %d, type: %d", attr
->overwrite
, attr
->subbuf_size
,
454 attr
->num_subbuf
, attr
->switch_timer_interval
,
455 attr
->read_timer_interval
, attr
->output
, attr
->type
);
457 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
)
460 nr_stream_fds
= ustctl_get_nr_stream_per_channel();
461 stream_fds
= zmalloc(nr_stream_fds
* sizeof(*stream_fds
));
466 for (i
= 0; i
< nr_stream_fds
; i
++) {
467 stream_fds
[i
] = open_ust_stream_fd(channel
, attr
, i
);
468 if (stream_fds
[i
] < 0) {
473 ust_channel
= ustctl_create_channel(attr
, stream_fds
, nr_stream_fds
);
478 channel
->nr_stream_fds
= nr_stream_fds
;
479 channel
->stream_fds
= stream_fds
;
480 *ust_chanp
= ust_channel
;
486 for (j
= i
- 1; j
>= 0; j
--) {
489 closeret
= close(stream_fds
[j
]);
493 if (channel
->shm_path
[0]) {
494 char shm_path
[PATH_MAX
];
496 closeret
= get_stream_shm_path(shm_path
,
497 channel
->shm_path
, j
);
499 ERR("Cannot get stream shm path");
501 closeret
= run_as_unlink(shm_path
,
502 channel
->uid
, channel
->gid
);
504 PERROR("unlink %s", shm_path
);
508 /* Try to rmdir all directories under shm_path root. */
509 if (channel
->root_shm_path
[0]) {
510 (void) run_as_rmdir_recursive(channel
->root_shm_path
,
511 channel
->uid
, channel
->gid
);
519 * Send a single given stream to the session daemon using the sock.
521 * Return 0 on success else a negative value.
523 static int send_sessiond_stream(int sock
, struct lttng_consumer_stream
*stream
)
530 DBG("UST consumer sending stream %" PRIu64
" to sessiond", stream
->key
);
532 /* Send stream to session daemon. */
533 ret
= ustctl_send_stream_to_sessiond(sock
, stream
->ustream
);
543 * Send channel to sessiond.
545 * Return 0 on success or else a negative value.
547 static int send_sessiond_channel(int sock
,
548 struct lttng_consumer_channel
*channel
,
549 struct lttng_consumer_local_data
*ctx
, int *relayd_error
)
551 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
552 struct lttng_consumer_stream
*stream
;
553 uint64_t net_seq_idx
= -1ULL;
559 DBG("UST consumer sending channel %s to sessiond", channel
->name
);
561 if (channel
->relayd_id
!= (uint64_t) -1ULL) {
562 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
564 health_code_update();
566 /* Try to send the stream to the relayd if one is available. */
567 ret
= consumer_send_relayd_stream(stream
, stream
->chan
->pathname
);
570 * Flag that the relayd was the problem here probably due to a
571 * communicaton error on the socket.
576 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
578 if (net_seq_idx
== -1ULL) {
579 net_seq_idx
= stream
->net_seq_idx
;
584 /* Inform sessiond that we are about to send channel and streams. */
585 ret
= consumer_send_status_msg(sock
, ret_code
);
586 if (ret
< 0 || ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
588 * Either the session daemon is not responding or the relayd died so we
594 /* Send channel to sessiond. */
595 ret
= ustctl_send_channel_to_sessiond(sock
, channel
->uchan
);
600 ret
= ustctl_channel_close_wakeup_fd(channel
->uchan
);
605 /* The channel was sent successfully to the sessiond at this point. */
606 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
608 health_code_update();
610 /* Send stream to session daemon. */
611 ret
= send_sessiond_stream(sock
, stream
);
617 /* Tell sessiond there is no more stream. */
618 ret
= ustctl_send_stream_to_sessiond(sock
, NULL
);
623 DBG("UST consumer NULL stream sent to sessiond");
628 if (ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
635 * Creates a channel and streams and add the channel it to the channel internal
636 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
639 * Return 0 on success or else, a negative value is returned and the channel
640 * MUST be destroyed by consumer_del_channel().
642 static int ask_channel(struct lttng_consumer_local_data
*ctx
, int sock
,
643 struct lttng_consumer_channel
*channel
,
644 struct ustctl_consumer_channel_attr
*attr
)
653 * This value is still used by the kernel consumer since for the kernel,
654 * the stream ownership is not IN the consumer so we need to have the
655 * number of left stream that needs to be initialized so we can know when
656 * to delete the channel (see consumer.c).
658 * As for the user space tracer now, the consumer creates and sends the
659 * stream to the session daemon which only sends them to the application
660 * once every stream of a channel is received making this value useless
661 * because we they will be added to the poll thread before the application
662 * receives them. This ensures that a stream can not hang up during
663 * initilization of a channel.
665 channel
->nb_init_stream_left
= 0;
667 /* The reply msg status is handled in the following call. */
668 ret
= create_ust_channel(channel
, attr
, &channel
->uchan
);
673 channel
->wait_fd
= ustctl_channel_get_wait_fd(channel
->uchan
);
676 * For the snapshots (no monitor), we create the metadata streams
677 * on demand, not during the channel creation.
679 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& !channel
->monitor
) {
684 /* Open all streams for this channel. */
685 ret
= create_ust_streams(channel
, ctx
);
695 * Send all stream of a channel to the right thread handling it.
697 * On error, return a negative value else 0 on success.
699 static int send_streams_to_thread(struct lttng_consumer_channel
*channel
,
700 struct lttng_consumer_local_data
*ctx
)
703 struct lttng_consumer_stream
*stream
, *stmp
;
708 /* Send streams to the corresponding thread. */
709 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
712 health_code_update();
714 /* Sending the stream to the thread. */
715 ret
= send_stream_to_thread(stream
, ctx
);
718 * If we are unable to send the stream to the thread, there is
719 * a big problem so just stop everything.
730 * Flush channel's streams using the given key to retrieve the channel.
732 * Return 0 on success else an LTTng error code.
734 static int flush_channel(uint64_t chan_key
)
737 struct lttng_consumer_channel
*channel
;
738 struct lttng_consumer_stream
*stream
;
740 struct lttng_ht_iter iter
;
742 DBG("UST consumer flush channel key %" PRIu64
, chan_key
);
745 channel
= consumer_find_channel(chan_key
);
747 ERR("UST consumer flush channel %" PRIu64
" not found", chan_key
);
748 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
752 ht
= consumer_data
.stream_per_chan_id_ht
;
754 /* For each stream of the channel id, flush it. */
755 cds_lfht_for_each_entry_duplicate(ht
->ht
,
756 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
757 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
759 health_code_update();
761 pthread_mutex_lock(&stream
->lock
);
762 if (!stream
->quiescent
) {
763 ustctl_flush_buffer(stream
->ustream
, 0);
764 stream
->quiescent
= true;
766 pthread_mutex_unlock(&stream
->lock
);
774 * Clear quiescent state from channel's streams using the given key to
775 * retrieve the channel.
777 * Return 0 on success else an LTTng error code.
779 static int clear_quiescent_channel(uint64_t chan_key
)
782 struct lttng_consumer_channel
*channel
;
783 struct lttng_consumer_stream
*stream
;
785 struct lttng_ht_iter iter
;
787 DBG("UST consumer clear quiescent channel key %" PRIu64
, chan_key
);
790 channel
= consumer_find_channel(chan_key
);
792 ERR("UST consumer clear quiescent channel %" PRIu64
" not found", chan_key
);
793 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
797 ht
= consumer_data
.stream_per_chan_id_ht
;
799 /* For each stream of the channel id, clear quiescent state. */
800 cds_lfht_for_each_entry_duplicate(ht
->ht
,
801 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
802 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
804 health_code_update();
806 pthread_mutex_lock(&stream
->lock
);
807 stream
->quiescent
= false;
808 pthread_mutex_unlock(&stream
->lock
);
816 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
817 * RCU read side lock MUST be acquired before calling this function.
819 * Return 0 on success else an LTTng error code.
821 static int close_metadata(uint64_t chan_key
)
824 struct lttng_consumer_channel
*channel
;
825 unsigned int channel_monitor
;
827 DBG("UST consumer close metadata key %" PRIu64
, chan_key
);
829 channel
= consumer_find_channel(chan_key
);
832 * This is possible if the metadata thread has issue a delete because
833 * the endpoint point of the stream hung up. There is no way the
834 * session daemon can know about it thus use a DBG instead of an actual
837 DBG("UST consumer close metadata %" PRIu64
" not found", chan_key
);
838 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
842 pthread_mutex_lock(&consumer_data
.lock
);
843 pthread_mutex_lock(&channel
->lock
);
844 channel_monitor
= channel
->monitor
;
845 if (cds_lfht_is_node_deleted(&channel
->node
.node
)) {
849 lttng_ustconsumer_close_metadata(channel
);
850 pthread_mutex_unlock(&channel
->lock
);
851 pthread_mutex_unlock(&consumer_data
.lock
);
854 * The ownership of a metadata channel depends on the type of
855 * session to which it belongs. In effect, the monitor flag is checked
856 * to determine if this metadata channel is in "snapshot" mode or not.
858 * In the non-snapshot case, the metadata channel is created along with
859 * a single stream which will remain present until the metadata channel
860 * is destroyed (on the destruction of its session). In this case, the
861 * metadata stream in "monitored" by the metadata poll thread and holds
862 * the ownership of its channel.
864 * Closing the metadata will cause the metadata stream's "metadata poll
865 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
866 * thread which will teardown the metadata stream which, in return,
867 * deletes the metadata channel.
869 * In the snapshot case, the metadata stream is created and destroyed
870 * on every snapshot record. Since the channel doesn't have an owner
871 * other than the session daemon, it is safe to destroy it immediately
872 * on reception of the CLOSE_METADATA command.
874 if (!channel_monitor
) {
876 * The channel and consumer_data locks must be
877 * released before this call since consumer_del_channel
878 * re-acquires the channel and consumer_data locks to teardown
879 * the channel and queue its reclamation by the "call_rcu"
882 consumer_del_channel(channel
);
887 pthread_mutex_unlock(&channel
->lock
);
888 pthread_mutex_unlock(&consumer_data
.lock
);
894 * RCU read side lock MUST be acquired before calling this function.
896 * Return 0 on success else an LTTng error code.
898 static int setup_metadata(struct lttng_consumer_local_data
*ctx
, uint64_t key
)
901 struct lttng_consumer_channel
*metadata
;
903 DBG("UST consumer setup metadata key %" PRIu64
, key
);
905 metadata
= consumer_find_channel(key
);
907 ERR("UST consumer push metadata %" PRIu64
" not found", key
);
908 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
913 * In no monitor mode, the metadata channel has no stream(s) so skip the
914 * ownership transfer to the metadata thread.
916 if (!metadata
->monitor
) {
917 DBG("Metadata channel in no monitor");
923 * Send metadata stream to relayd if one available. Availability is
924 * known if the stream is still in the list of the channel.
926 if (cds_list_empty(&metadata
->streams
.head
)) {
927 ERR("Metadata channel key %" PRIu64
", no stream available.", key
);
928 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
929 goto error_no_stream
;
932 /* Send metadata stream to relayd if needed. */
933 if (metadata
->metadata_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
934 ret
= consumer_send_relayd_stream(metadata
->metadata_stream
,
937 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
940 ret
= consumer_send_relayd_streams_sent(
941 metadata
->metadata_stream
->net_seq_idx
);
943 ret
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
949 * Ownership of metadata stream is passed along. Freeing is handled by
952 ret
= send_streams_to_thread(metadata
, ctx
);
955 * If we are unable to send the stream to the thread, there is
956 * a big problem so just stop everything.
958 ret
= LTTCOMM_CONSUMERD_FATAL
;
959 goto send_streams_error
;
961 /* List MUST be empty after or else it could be reused. */
962 assert(cds_list_empty(&metadata
->streams
.head
));
969 * Delete metadata channel on error. At this point, the metadata stream can
970 * NOT be monitored by the metadata thread thus having the guarantee that
971 * the stream is still in the local stream list of the channel. This call
972 * will make sure to clean that list.
974 consumer_stream_destroy(metadata
->metadata_stream
, NULL
);
975 cds_list_del(&metadata
->metadata_stream
->send_node
);
976 metadata
->metadata_stream
= NULL
;
984 * Snapshot the whole metadata.
986 * Returns 0 on success, < 0 on error
988 static int snapshot_metadata(uint64_t key
, char *path
, uint64_t relayd_id
,
989 struct lttng_consumer_local_data
*ctx
)
992 struct lttng_consumer_channel
*metadata_channel
;
993 struct lttng_consumer_stream
*metadata_stream
;
998 DBG("UST consumer snapshot metadata with key %" PRIu64
" at path %s",
1003 metadata_channel
= consumer_find_channel(key
);
1004 if (!metadata_channel
) {
1005 ERR("UST snapshot metadata channel not found for key %" PRIu64
,
1010 assert(!metadata_channel
->monitor
);
1012 health_code_update();
1015 * Ask the sessiond if we have new metadata waiting and update the
1016 * consumer metadata cache.
1018 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 1);
1023 health_code_update();
1026 * The metadata stream is NOT created in no monitor mode when the channel
1027 * is created on a sessiond ask channel command.
1029 ret
= create_ust_streams(metadata_channel
, ctx
);
1034 metadata_stream
= metadata_channel
->metadata_stream
;
1035 assert(metadata_stream
);
1037 if (relayd_id
!= (uint64_t) -1ULL) {
1038 metadata_stream
->net_seq_idx
= relayd_id
;
1039 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
1044 ret
= utils_create_stream_file(path
, metadata_stream
->name
,
1045 metadata_stream
->chan
->tracefile_size
,
1046 metadata_stream
->tracefile_count_current
,
1047 metadata_stream
->uid
, metadata_stream
->gid
, NULL
);
1051 metadata_stream
->out_fd
= ret
;
1052 metadata_stream
->tracefile_size_current
= 0;
1056 health_code_update();
1058 ret
= lttng_consumer_read_subbuffer(metadata_stream
, ctx
);
1066 * Clean up the stream completly because the next snapshot will use a new
1069 consumer_stream_destroy(metadata_stream
, NULL
);
1070 cds_list_del(&metadata_stream
->send_node
);
1071 metadata_channel
->metadata_stream
= NULL
;
1079 * Take a snapshot of all the stream of a channel.
1081 * Returns 0 on success, < 0 on error
1083 static int snapshot_channel(uint64_t key
, char *path
, uint64_t relayd_id
,
1084 uint64_t nb_packets_per_stream
, struct lttng_consumer_local_data
*ctx
)
1087 unsigned use_relayd
= 0;
1088 unsigned long consumed_pos
, produced_pos
;
1089 struct lttng_consumer_channel
*channel
;
1090 struct lttng_consumer_stream
*stream
;
1097 if (relayd_id
!= (uint64_t) -1ULL) {
1101 channel
= consumer_find_channel(key
);
1103 ERR("UST snapshot channel not found for key %" PRIu64
, key
);
1107 assert(!channel
->monitor
);
1108 DBG("UST consumer snapshot channel %" PRIu64
, key
);
1110 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
1111 health_code_update();
1113 /* Lock stream because we are about to change its state. */
1114 pthread_mutex_lock(&stream
->lock
);
1115 stream
->net_seq_idx
= relayd_id
;
1118 ret
= consumer_send_relayd_stream(stream
, path
);
1123 ret
= utils_create_stream_file(path
, stream
->name
,
1124 stream
->chan
->tracefile_size
,
1125 stream
->tracefile_count_current
,
1126 stream
->uid
, stream
->gid
, NULL
);
1130 stream
->out_fd
= ret
;
1131 stream
->tracefile_size_current
= 0;
1133 DBG("UST consumer snapshot stream %s/%s (%" PRIu64
")", path
,
1134 stream
->name
, stream
->key
);
1138 * If tracing is active, we want to perform a "full" buffer flush.
1139 * Else, if quiescent, it has already been done by the prior stop.
1141 if (!stream
->quiescent
) {
1142 ustctl_flush_buffer(stream
->ustream
, 0);
1145 ret
= lttng_ustconsumer_take_snapshot(stream
);
1147 ERR("Taking UST snapshot");
1151 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
1153 ERR("Produced UST snapshot position");
1157 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1159 ERR("Consumerd UST snapshot position");
1164 * The original value is sent back if max stream size is larger than
1165 * the possible size of the snapshot. Also, we assume that the session
1166 * daemon should never send a maximum stream size that is lower than
1169 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
1170 produced_pos
, nb_packets_per_stream
,
1171 stream
->max_sb_size
);
1173 while (consumed_pos
< produced_pos
) {
1175 unsigned long len
, padded_len
;
1177 health_code_update();
1179 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1181 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1183 if (ret
!= -EAGAIN
) {
1184 PERROR("ustctl_get_subbuf snapshot");
1185 goto error_close_stream
;
1187 DBG("UST consumer get subbuf failed. Skipping it.");
1188 consumed_pos
+= stream
->max_sb_size
;
1189 stream
->chan
->lost_packets
++;
1193 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1195 ERR("Snapshot ustctl_get_subbuf_size");
1196 goto error_put_subbuf
;
1199 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1201 ERR("Snapshot ustctl_get_padded_subbuf_size");
1202 goto error_put_subbuf
;
1205 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, len
,
1206 padded_len
- len
, NULL
);
1208 if (read_len
!= len
) {
1210 goto error_put_subbuf
;
1213 if (read_len
!= padded_len
) {
1215 goto error_put_subbuf
;
1219 ret
= ustctl_put_subbuf(stream
->ustream
);
1221 ERR("Snapshot ustctl_put_subbuf");
1222 goto error_close_stream
;
1224 consumed_pos
+= stream
->max_sb_size
;
1227 /* Simply close the stream so we can use it on the next snapshot. */
1228 consumer_stream_close(stream
);
1229 pthread_mutex_unlock(&stream
->lock
);
1236 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1237 ERR("Snapshot ustctl_put_subbuf");
1240 consumer_stream_close(stream
);
1242 pthread_mutex_unlock(&stream
->lock
);
1249 * Receive the metadata updates from the sessiond. Supports receiving
1250 * overlapping metadata, but is needs to always belong to a contiguous
1251 * range starting from 0.
1252 * Be careful about the locks held when calling this function: it needs
1253 * the metadata cache flush to concurrently progress in order to
1256 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1257 uint64_t len
, uint64_t version
,
1258 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
1260 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1263 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1265 metadata_str
= zmalloc(len
* sizeof(char));
1266 if (!metadata_str
) {
1267 PERROR("zmalloc metadata string");
1268 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1272 health_code_update();
1274 /* Receive metadata string. */
1275 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1277 /* Session daemon is dead so return gracefully. */
1282 health_code_update();
1284 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1285 ret
= consumer_metadata_cache_write(channel
, offset
, len
, version
,
1288 /* Unable to handle metadata. Notify session daemon. */
1289 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1291 * Skip metadata flush on write error since the offset and len might
1292 * not have been updated which could create an infinite loop below when
1293 * waiting for the metadata cache to be flushed.
1295 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1298 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1303 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1304 DBG("Waiting for metadata to be flushed");
1306 health_code_update();
1308 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1318 * Receive command from session daemon and process it.
1320 * Return 1 on success else a negative value or 0.
1322 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1323 int sock
, struct pollfd
*consumer_sockpoll
)
1326 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1327 struct lttcomm_consumer_msg msg
;
1328 struct lttng_consumer_channel
*channel
= NULL
;
1330 health_code_update();
1332 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1333 if (ret
!= sizeof(msg
)) {
1334 DBG("Consumer received unexpected message size %zd (expects %zu)",
1337 * The ret value might 0 meaning an orderly shutdown but this is ok
1338 * since the caller handles this.
1341 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1347 health_code_update();
1350 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1352 health_code_update();
1354 /* relayd needs RCU read-side lock */
1357 switch (msg
.cmd_type
) {
1358 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1360 /* Session daemon status message are handled in the following call. */
1361 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1362 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
1363 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
1364 msg
.u
.relayd_sock
.relayd_session_id
);
1367 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1369 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1370 struct consumer_relayd_sock_pair
*relayd
;
1372 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1374 /* Get relayd reference if exists. */
1375 relayd
= consumer_find_relayd(index
);
1376 if (relayd
== NULL
) {
1377 DBG("Unable to find relayd %" PRIu64
, index
);
1378 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1382 * Each relayd socket pair has a refcount of stream attached to it
1383 * which tells if the relayd is still active or not depending on the
1386 * This will set the destroy flag of the relayd object and destroy it
1387 * if the refcount reaches zero when called.
1389 * The destroy can happen either here or when a stream fd hangs up.
1392 consumer_flag_relayd_for_destroy(relayd
);
1395 goto end_msg_sessiond
;
1397 case LTTNG_CONSUMER_UPDATE_STREAM
:
1402 case LTTNG_CONSUMER_DATA_PENDING
:
1404 int ret
, is_data_pending
;
1405 uint64_t id
= msg
.u
.data_pending
.session_id
;
1407 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1409 is_data_pending
= consumer_data_pending(id
);
1411 /* Send back returned value to session daemon */
1412 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1413 sizeof(is_data_pending
));
1415 DBG("Error when sending the data pending ret code: %d", ret
);
1420 * No need to send back a status message since the data pending
1421 * returned value is the response.
1425 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1428 struct ustctl_consumer_channel_attr attr
;
1430 /* Create a plain object and reserve a channel key. */
1431 channel
= allocate_channel(msg
.u
.ask_channel
.session_id
,
1432 msg
.u
.ask_channel
.pathname
, msg
.u
.ask_channel
.name
,
1433 msg
.u
.ask_channel
.uid
, msg
.u
.ask_channel
.gid
,
1434 msg
.u
.ask_channel
.relayd_id
, msg
.u
.ask_channel
.key
,
1435 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1436 msg
.u
.ask_channel
.tracefile_size
,
1437 msg
.u
.ask_channel
.tracefile_count
,
1438 msg
.u
.ask_channel
.session_id_per_pid
,
1439 msg
.u
.ask_channel
.monitor
,
1440 msg
.u
.ask_channel
.live_timer_interval
,
1441 msg
.u
.ask_channel
.root_shm_path
,
1442 msg
.u
.ask_channel
.shm_path
);
1444 goto end_channel_error
;
1448 * Assign UST application UID to the channel. This value is ignored for
1449 * per PID buffers. This is specific to UST thus setting this after the
1452 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1454 /* Build channel attributes from received message. */
1455 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1456 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1457 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1458 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1459 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1460 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1461 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1462 attr
.blocking_timeout
= msg
.u
.ask_channel
.blocking_timeout
;
1464 /* Match channel buffer type to the UST abi. */
1465 switch (msg
.u
.ask_channel
.output
) {
1466 case LTTNG_EVENT_MMAP
:
1468 attr
.output
= LTTNG_UST_MMAP
;
1472 /* Translate and save channel type. */
1473 switch (msg
.u
.ask_channel
.type
) {
1474 case LTTNG_UST_CHAN_PER_CPU
:
1475 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1476 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1478 * Set refcount to 1 for owner. Below, we will
1479 * pass ownership to the
1480 * consumer_thread_channel_poll() thread.
1482 channel
->refcount
= 1;
1484 case LTTNG_UST_CHAN_METADATA
:
1485 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1486 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1493 health_code_update();
1495 ret
= ask_channel(ctx
, sock
, channel
, &attr
);
1497 goto end_channel_error
;
1500 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1501 ret
= consumer_metadata_cache_allocate(channel
);
1503 ERR("Allocating metadata cache");
1504 goto end_channel_error
;
1506 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1507 attr
.switch_timer_interval
= 0;
1509 int monitor_start_ret
;
1511 consumer_timer_live_start(channel
,
1512 msg
.u
.ask_channel
.live_timer_interval
);
1513 monitor_start_ret
= consumer_timer_monitor_start(
1515 msg
.u
.ask_channel
.monitor_timer_interval
);
1516 if (monitor_start_ret
< 0) {
1517 ERR("Starting channel monitoring timer failed");
1518 goto end_channel_error
;
1522 health_code_update();
1525 * Add the channel to the internal state AFTER all streams were created
1526 * and successfully sent to session daemon. This way, all streams must
1527 * be ready before this channel is visible to the threads.
1528 * If add_channel succeeds, ownership of the channel is
1529 * passed to consumer_thread_channel_poll().
1531 ret
= add_channel(channel
, ctx
);
1533 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1534 if (channel
->switch_timer_enabled
== 1) {
1535 consumer_timer_switch_stop(channel
);
1537 consumer_metadata_cache_destroy(channel
);
1539 if (channel
->live_timer_enabled
== 1) {
1540 consumer_timer_live_stop(channel
);
1542 if (channel
->monitor_timer_enabled
== 1) {
1543 consumer_timer_monitor_stop(channel
);
1545 goto end_channel_error
;
1548 health_code_update();
1551 * Channel and streams are now created. Inform the session daemon that
1552 * everything went well and should wait to receive the channel and
1553 * streams with ustctl API.
1555 ret
= consumer_send_status_channel(sock
, channel
);
1558 * There is probably a problem on the socket.
1565 case LTTNG_CONSUMER_GET_CHANNEL
:
1567 int ret
, relayd_err
= 0;
1568 uint64_t key
= msg
.u
.get_channel
.key
;
1569 struct lttng_consumer_channel
*channel
;
1571 channel
= consumer_find_channel(key
);
1573 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1574 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1575 goto end_msg_sessiond
;
1578 health_code_update();
1580 /* Send everything to sessiond. */
1581 ret
= send_sessiond_channel(sock
, channel
, ctx
, &relayd_err
);
1585 * We were unable to send to the relayd the stream so avoid
1586 * sending back a fatal error to the thread since this is OK
1587 * and the consumer can continue its work. The above call
1588 * has sent the error status message to the sessiond.
1593 * The communicaton was broken hence there is a bad state between
1594 * the consumer and sessiond so stop everything.
1599 health_code_update();
1602 * In no monitor mode, the streams ownership is kept inside the channel
1603 * so don't send them to the data thread.
1605 if (!channel
->monitor
) {
1606 goto end_msg_sessiond
;
1609 ret
= send_streams_to_thread(channel
, ctx
);
1612 * If we are unable to send the stream to the thread, there is
1613 * a big problem so just stop everything.
1617 /* List MUST be empty after or else it could be reused. */
1618 assert(cds_list_empty(&channel
->streams
.head
));
1619 goto end_msg_sessiond
;
1621 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1623 uint64_t key
= msg
.u
.destroy_channel
.key
;
1626 * Only called if streams have not been sent to stream
1627 * manager thread. However, channel has been sent to
1628 * channel manager thread.
1630 notify_thread_del_channel(ctx
, key
);
1631 goto end_msg_sessiond
;
1633 case LTTNG_CONSUMER_CLOSE_METADATA
:
1637 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1642 goto end_msg_sessiond
;
1644 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1648 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1653 goto end_msg_sessiond
;
1655 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL
:
1659 ret
= clear_quiescent_channel(
1660 msg
.u
.clear_quiescent_channel
.key
);
1665 goto end_msg_sessiond
;
1667 case LTTNG_CONSUMER_PUSH_METADATA
:
1670 uint64_t len
= msg
.u
.push_metadata
.len
;
1671 uint64_t key
= msg
.u
.push_metadata
.key
;
1672 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1673 uint64_t version
= msg
.u
.push_metadata
.version
;
1674 struct lttng_consumer_channel
*channel
;
1676 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1679 channel
= consumer_find_channel(key
);
1682 * This is possible if the metadata creation on the consumer side
1683 * is in flight vis-a-vis a concurrent push metadata from the
1684 * session daemon. Simply return that the channel failed and the
1685 * session daemon will handle that message correctly considering
1686 * that this race is acceptable thus the DBG() statement here.
1688 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1689 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1690 goto end_msg_sessiond
;
1693 health_code_update();
1697 * There is nothing to receive. We have simply
1698 * checked whether the channel can be found.
1700 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1701 goto end_msg_sessiond
;
1704 /* Tell session daemon we are ready to receive the metadata. */
1705 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1707 /* Somehow, the session daemon is not responding anymore. */
1711 health_code_update();
1713 /* Wait for more data. */
1714 health_poll_entry();
1715 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1721 health_code_update();
1723 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1724 len
, version
, channel
, 0, 1);
1726 /* error receiving from sessiond */
1730 goto end_msg_sessiond
;
1733 case LTTNG_CONSUMER_SETUP_METADATA
:
1737 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1741 goto end_msg_sessiond
;
1743 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1745 if (msg
.u
.snapshot_channel
.metadata
) {
1746 ret
= snapshot_metadata(msg
.u
.snapshot_channel
.key
,
1747 msg
.u
.snapshot_channel
.pathname
,
1748 msg
.u
.snapshot_channel
.relayd_id
,
1751 ERR("Snapshot metadata failed");
1752 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1755 ret
= snapshot_channel(msg
.u
.snapshot_channel
.key
,
1756 msg
.u
.snapshot_channel
.pathname
,
1757 msg
.u
.snapshot_channel
.relayd_id
,
1758 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
1761 ERR("Snapshot channel failed");
1762 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1766 health_code_update();
1767 ret
= consumer_send_status_msg(sock
, ret_code
);
1769 /* Somehow, the session daemon is not responding anymore. */
1772 health_code_update();
1775 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
1778 uint64_t discarded_events
;
1779 struct lttng_ht_iter iter
;
1780 struct lttng_ht
*ht
;
1781 struct lttng_consumer_stream
*stream
;
1782 uint64_t id
= msg
.u
.discarded_events
.session_id
;
1783 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
1785 DBG("UST consumer discarded events command for session id %"
1788 pthread_mutex_lock(&consumer_data
.lock
);
1790 ht
= consumer_data
.stream_list_ht
;
1793 * We only need a reference to the channel, but they are not
1794 * directly indexed, so we just use the first matching stream
1795 * to extract the information we need, we default to 0 if not
1796 * found (no events are dropped if the channel is not yet in
1799 discarded_events
= 0;
1800 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1801 ht
->hash_fct(&id
, lttng_ht_seed
),
1803 &iter
.iter
, stream
, node_session_id
.node
) {
1804 if (stream
->chan
->key
== key
) {
1805 discarded_events
= stream
->chan
->discarded_events
;
1809 pthread_mutex_unlock(&consumer_data
.lock
);
1812 DBG("UST consumer discarded events command for session id %"
1813 PRIu64
", channel key %" PRIu64
, id
, key
);
1815 health_code_update();
1817 /* Send back returned value to session daemon */
1818 ret
= lttcomm_send_unix_sock(sock
, &discarded_events
, sizeof(discarded_events
));
1820 PERROR("send discarded events");
1826 case LTTNG_CONSUMER_LOST_PACKETS
:
1829 uint64_t lost_packets
;
1830 struct lttng_ht_iter iter
;
1831 struct lttng_ht
*ht
;
1832 struct lttng_consumer_stream
*stream
;
1833 uint64_t id
= msg
.u
.lost_packets
.session_id
;
1834 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
1836 DBG("UST consumer lost packets command for session id %"
1839 pthread_mutex_lock(&consumer_data
.lock
);
1841 ht
= consumer_data
.stream_list_ht
;
1844 * We only need a reference to the channel, but they are not
1845 * directly indexed, so we just use the first matching stream
1846 * to extract the information we need, we default to 0 if not
1847 * found (no packets lost if the channel is not yet in use).
1850 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1851 ht
->hash_fct(&id
, lttng_ht_seed
),
1853 &iter
.iter
, stream
, node_session_id
.node
) {
1854 if (stream
->chan
->key
== key
) {
1855 lost_packets
= stream
->chan
->lost_packets
;
1859 pthread_mutex_unlock(&consumer_data
.lock
);
1862 DBG("UST consumer lost packets command for session id %"
1863 PRIu64
", channel key %" PRIu64
, id
, key
);
1865 health_code_update();
1867 /* Send back returned value to session daemon */
1868 ret
= lttcomm_send_unix_sock(sock
, &lost_packets
,
1869 sizeof(lost_packets
));
1871 PERROR("send lost packets");
1877 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
1879 int channel_monitor_pipe
;
1881 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1882 /* Successfully received the command's type. */
1883 ret
= consumer_send_status_msg(sock
, ret_code
);
1888 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
1890 if (ret
!= sizeof(channel_monitor_pipe
)) {
1891 ERR("Failed to receive channel monitor pipe");
1895 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
1896 ret
= consumer_timer_thread_set_channel_monitor_pipe(
1897 channel_monitor_pipe
);
1901 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1902 /* Set the pipe as non-blocking. */
1903 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
1905 PERROR("fcntl get flags of the channel monitoring pipe");
1910 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
1911 flags
| O_NONBLOCK
);
1913 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1916 DBG("Channel monitor pipe set as non-blocking");
1918 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
1920 goto end_msg_sessiond
;
1922 case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE
:
1924 int channel_rotate_pipe
;
1927 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1928 /* Successfully received the command's type. */
1929 ret
= consumer_send_status_msg(sock
, ret_code
);
1934 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_rotate_pipe
, 1);
1935 if (ret
!= sizeof(channel_rotate_pipe
)) {
1936 ERR("Failed to receive channel rotate pipe");
1940 DBG("Received channel rotate pipe (%d)", channel_rotate_pipe
);
1941 ctx
->channel_rotate_pipe
= channel_rotate_pipe
;
1942 /* Set the pipe as non-blocking. */
1943 ret
= fcntl(channel_rotate_pipe
, F_GETFL
, 0);
1945 PERROR("fcntl get flags of the channel rotate pipe");
1950 ret
= fcntl(channel_rotate_pipe
, F_SETFL
, flags
| O_NONBLOCK
);
1952 PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
1955 DBG("Channel rotate pipe set as non-blocking");
1956 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1957 ret
= consumer_send_status_msg(sock
, ret_code
);
1963 case LTTNG_CONSUMER_ROTATE_CHANNEL
:
1966 * Sample the rotate position of all the streams in this channel.
1968 ret
= lttng_consumer_rotate_channel(msg
.u
.rotate_channel
.key
,
1969 msg
.u
.rotate_channel
.pathname
,
1970 msg
.u
.rotate_channel
.relayd_id
,
1971 msg
.u
.rotate_channel
.metadata
,
1972 msg
.u
.rotate_channel
.new_chunk_id
,
1975 ERR("Rotate channel failed");
1976 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1979 health_code_update();
1981 ret
= consumer_send_status_msg(sock
, ret_code
);
1983 /* Somehow, the session daemon is not responding anymore. */
1988 * Rotate the streams that are ready right now.
1989 * FIXME: this is a second consecutive iteration over the
1990 * streams in a channel, there is probably a better way to
1991 * handle this, but it needs to be after the
1992 * consumer_send_status_msg() call.
1994 ret
= lttng_consumer_rotate_ready_streams(
1995 msg
.u
.rotate_channel
.key
, ctx
);
1997 ERR("Rotate channel failed");
1998 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2002 case LTTNG_CONSUMER_ROTATE_RENAME
:
2004 DBG("Consumer rename session %" PRIu64
" after rotation",
2005 msg
.u
.rotate_rename
.session_id
);
2006 ret
= lttng_consumer_rotate_rename(msg
.u
.rotate_rename
.old_path
,
2007 msg
.u
.rotate_rename
.new_path
,
2008 msg
.u
.rotate_rename
.uid
,
2009 msg
.u
.rotate_rename
.gid
,
2010 msg
.u
.rotate_rename
.relayd_id
);
2012 ERR("Rotate rename failed");
2013 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
2016 health_code_update();
2018 ret
= consumer_send_status_msg(sock
, ret_code
);
2020 /* Somehow, the session daemon is not responding anymore. */
2025 case LTTNG_CONSUMER_ROTATE_PENDING_RELAY
:
2029 DBG("Consumer rotate pending on relay for session %" PRIu64
,
2030 msg
.u
.rotate_pending_relay
.session_id
);
2031 pending
= lttng_consumer_rotate_pending_relay(
2032 msg
.u
.rotate_pending_relay
.session_id
,
2033 msg
.u
.rotate_pending_relay
.relayd_id
,
2034 msg
.u
.rotate_pending_relay
.chunk_id
);
2036 ERR("Rotate pending relay failed");
2037 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
2040 health_code_update();
2042 ret
= consumer_send_status_msg(sock
, ret_code
);
2044 /* Somehow, the session daemon is not responding anymore. */
2048 /* Send back returned value to session daemon */
2049 ret
= lttcomm_send_unix_sock(sock
, &pending
, sizeof(pending
));
2051 PERROR("send data pending ret code");
2056 case LTTNG_CONSUMER_MKDIR
:
2058 DBG("Consumer mkdir %s in session %" PRIu64
,
2060 msg
.u
.mkdir
.session_id
);
2061 ret
= lttng_consumer_mkdir(msg
.u
.mkdir
.path
,
2064 msg
.u
.mkdir
.relayd_id
);
2066 ERR("consumer mkdir failed");
2067 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
2070 health_code_update();
2072 ret
= consumer_send_status_msg(sock
, ret_code
);
2074 /* Somehow, the session daemon is not responding anymore. */
2086 health_code_update();
2089 * Return 1 to indicate success since the 0 value can be a socket
2090 * shutdown during the recv() or send() call.
2096 * The returned value here is not useful since either way we'll return 1 to
2097 * the caller because the session daemon socket management is done
2098 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2100 ret
= consumer_send_status_msg(sock
, ret_code
);
2106 health_code_update();
2112 * Free channel here since no one has a reference to it. We don't
2113 * free after that because a stream can store this pointer.
2115 destroy_channel(channel
);
2117 /* We have to send a status channel message indicating an error. */
2118 ret
= consumer_send_status_channel(sock
, NULL
);
2120 /* Stop everything if session daemon can not be notified. */
2125 health_code_update();
2130 /* This will issue a consumer stop. */
2135 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2136 * compiled out, we isolate it in this library.
2138 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream
*stream
,
2142 assert(stream
->ustream
);
2144 return ustctl_get_mmap_read_offset(stream
->ustream
, off
);
2148 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2149 * compiled out, we isolate it in this library.
2151 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream
*stream
)
2154 assert(stream
->ustream
);
2156 return ustctl_get_mmap_base(stream
->ustream
);
2159 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream
*stream
,
2160 int producer_active
)
2163 assert(stream
->ustream
);
2165 ustctl_flush_buffer(stream
->ustream
, producer_active
);
2169 * Take a snapshot for a specific stream.
2171 * Returns 0 on success, < 0 on error
2173 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2176 assert(stream
->ustream
);
2178 return ustctl_snapshot(stream
->ustream
);
2182 * Sample consumed and produced positions for a specific stream.
2184 * Returns 0 on success, < 0 on error.
2186 int lttng_ustconsumer_sample_snapshot_positions(
2187 struct lttng_consumer_stream
*stream
)
2190 assert(stream
->ustream
);
2192 return ustctl_snapshot_sample_positions(stream
->ustream
);
2196 * Get the produced position
2198 * Returns 0 on success, < 0 on error
2200 int lttng_ustconsumer_get_produced_snapshot(
2201 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2204 assert(stream
->ustream
);
2207 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
2211 * Get the consumed position
2213 * Returns 0 on success, < 0 on error
2215 int lttng_ustconsumer_get_consumed_snapshot(
2216 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2219 assert(stream
->ustream
);
2222 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
2225 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
2229 assert(stream
->ustream
);
2231 ustctl_flush_buffer(stream
->ustream
, producer
);
2234 int lttng_ustconsumer_get_current_timestamp(
2235 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
2238 assert(stream
->ustream
);
2241 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
2244 int lttng_ustconsumer_get_sequence_number(
2245 struct lttng_consumer_stream
*stream
, uint64_t *seq
)
2248 assert(stream
->ustream
);
2251 return ustctl_get_sequence_number(stream
->ustream
, seq
);
2255 * Called when the stream signals the consumer that it has hung up.
2257 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
2260 assert(stream
->ustream
);
2262 pthread_mutex_lock(&stream
->lock
);
2263 if (!stream
->quiescent
) {
2264 ustctl_flush_buffer(stream
->ustream
, 0);
2265 stream
->quiescent
= true;
2267 pthread_mutex_unlock(&stream
->lock
);
2268 stream
->hangup_flush_done
= 1;
2271 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
2276 assert(chan
->uchan
);
2278 if (chan
->switch_timer_enabled
== 1) {
2279 consumer_timer_switch_stop(chan
);
2281 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
2284 ret
= close(chan
->stream_fds
[i
]);
2288 if (chan
->shm_path
[0]) {
2289 char shm_path
[PATH_MAX
];
2291 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
2293 ERR("Cannot get stream shm path");
2295 ret
= run_as_unlink(shm_path
, chan
->uid
, chan
->gid
);
2297 PERROR("unlink %s", shm_path
);
2303 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
2306 assert(chan
->uchan
);
2308 consumer_metadata_cache_destroy(chan
);
2309 ustctl_destroy_channel(chan
->uchan
);
2310 /* Try to rmdir all directories under shm_path root. */
2311 if (chan
->root_shm_path
[0]) {
2312 (void) run_as_rmdir_recursive(chan
->root_shm_path
,
2313 chan
->uid
, chan
->gid
);
2315 free(chan
->stream_fds
);
2318 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
2321 assert(stream
->ustream
);
2323 if (stream
->chan
->switch_timer_enabled
== 1) {
2324 consumer_timer_switch_stop(stream
->chan
);
2326 ustctl_destroy_stream(stream
->ustream
);
2329 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
2332 assert(stream
->ustream
);
2334 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
2337 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
2340 assert(stream
->ustream
);
2342 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
2346 * Populate index values of a UST stream. Values are set in big endian order.
2348 * Return 0 on success or else a negative value.
2350 static int get_index_values(struct ctf_packet_index
*index
,
2351 struct ustctl_consumer_stream
*ustream
)
2355 ret
= ustctl_get_timestamp_begin(ustream
, &index
->timestamp_begin
);
2357 PERROR("ustctl_get_timestamp_begin");
2360 index
->timestamp_begin
= htobe64(index
->timestamp_begin
);
2362 ret
= ustctl_get_timestamp_end(ustream
, &index
->timestamp_end
);
2364 PERROR("ustctl_get_timestamp_end");
2367 index
->timestamp_end
= htobe64(index
->timestamp_end
);
2369 ret
= ustctl_get_events_discarded(ustream
, &index
->events_discarded
);
2371 PERROR("ustctl_get_events_discarded");
2374 index
->events_discarded
= htobe64(index
->events_discarded
);
2376 ret
= ustctl_get_content_size(ustream
, &index
->content_size
);
2378 PERROR("ustctl_get_content_size");
2381 index
->content_size
= htobe64(index
->content_size
);
2383 ret
= ustctl_get_packet_size(ustream
, &index
->packet_size
);
2385 PERROR("ustctl_get_packet_size");
2388 index
->packet_size
= htobe64(index
->packet_size
);
2390 ret
= ustctl_get_stream_id(ustream
, &index
->stream_id
);
2392 PERROR("ustctl_get_stream_id");
2395 index
->stream_id
= htobe64(index
->stream_id
);
2397 ret
= ustctl_get_instance_id(ustream
, &index
->stream_instance_id
);
2399 PERROR("ustctl_get_instance_id");
2402 index
->stream_instance_id
= htobe64(index
->stream_instance_id
);
2404 ret
= ustctl_get_sequence_number(ustream
, &index
->packet_seq_num
);
2406 PERROR("ustctl_get_sequence_number");
2409 index
->packet_seq_num
= htobe64(index
->packet_seq_num
);
2416 void metadata_stream_reset_cache(struct lttng_consumer_stream
*stream
,
2417 struct consumer_metadata_cache
*cache
)
2419 DBG("Metadata stream update to version %" PRIu64
,
2421 stream
->ust_metadata_pushed
= 0;
2422 stream
->metadata_version
= cache
->version
;
2423 stream
->reset_metadata_flag
= 1;
2427 * Check if the version of the metadata stream and metadata cache match.
2428 * If the cache got updated, reset the metadata stream.
2429 * The stream lock and metadata cache lock MUST be held.
2430 * Return 0 on success, a negative value on error.
2433 int metadata_stream_check_version(struct lttng_consumer_stream
*stream
)
2436 struct consumer_metadata_cache
*cache
= stream
->chan
->metadata_cache
;
2438 if (cache
->version
== stream
->metadata_version
) {
2441 metadata_stream_reset_cache(stream
, cache
);
2448 * Write up to one packet from the metadata cache to the channel.
2450 * Returns the number of bytes pushed in the cache, or a negative value
2454 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
2459 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2460 ret
= metadata_stream_check_version(stream
);
2464 if (stream
->chan
->metadata_cache
->max_offset
2465 == stream
->ust_metadata_pushed
) {
2470 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
2471 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
2472 stream
->chan
->metadata_cache
->max_offset
2473 - stream
->ust_metadata_pushed
);
2474 assert(write_len
!= 0);
2475 if (write_len
< 0) {
2476 ERR("Writing one metadata packet");
2480 stream
->ust_metadata_pushed
+= write_len
;
2482 assert(stream
->chan
->metadata_cache
->max_offset
>=
2483 stream
->ust_metadata_pushed
);
2487 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2493 * Sync metadata meaning request them to the session daemon and snapshot to the
2494 * metadata thread can consumer them.
2496 * Metadata stream lock is held here, but we need to release it when
2497 * interacting with sessiond, else we cause a deadlock with live
2498 * awaiting on metadata to be pushed out.
2500 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2501 * is empty or a negative value on error.
2503 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data
*ctx
,
2504 struct lttng_consumer_stream
*metadata
)
2512 pthread_mutex_unlock(&metadata
->lock
);
2514 * Request metadata from the sessiond, but don't wait for the flush
2515 * because we locked the metadata thread.
2517 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata
->chan
, 0, 0);
2518 pthread_mutex_lock(&metadata
->lock
);
2523 ret
= commit_one_metadata_packet(metadata
);
2526 } else if (ret
> 0) {
2530 ustctl_flush_buffer(metadata
->ustream
, 1);
2531 ret
= ustctl_snapshot(metadata
->ustream
);
2533 if (errno
!= EAGAIN
) {
2534 ERR("Sync metadata, taking UST snapshot");
2537 DBG("No new metadata when syncing them.");
2538 /* No new metadata, exit. */
2544 * After this flush, we still need to extract metadata.
2555 * Return 0 on success else a negative value.
2557 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2558 struct lttng_consumer_local_data
*ctx
)
2561 struct ustctl_consumer_stream
*ustream
;
2566 ustream
= stream
->ustream
;
2569 * First, we are going to check if there is a new subbuffer available
2570 * before reading the stream wait_fd.
2572 /* Get the next subbuffer */
2573 ret
= ustctl_get_next_subbuf(ustream
);
2575 /* No more data found, flag the stream. */
2576 stream
->has_data
= 0;
2581 ret
= ustctl_put_subbuf(ustream
);
2584 /* This stream still has data. Flag it and wake up the data thread. */
2585 stream
->has_data
= 1;
2587 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2590 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2591 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2596 /* The wake up pipe has been notified. */
2597 ctx
->has_wakeup
= 1;
2606 int update_stream_stats(struct lttng_consumer_stream
*stream
)
2609 uint64_t seq
, discarded
;
2611 ret
= ustctl_get_sequence_number(stream
->ustream
, &seq
);
2613 PERROR("ustctl_get_sequence_number");
2617 * Start the sequence when we extract the first packet in case we don't
2618 * start at 0 (for example if a consumer is not connected to the
2619 * session immediately after the beginning).
2621 if (stream
->last_sequence_number
== -1ULL) {
2622 stream
->last_sequence_number
= seq
;
2623 } else if (seq
> stream
->last_sequence_number
) {
2624 stream
->chan
->lost_packets
+= seq
-
2625 stream
->last_sequence_number
- 1;
2627 /* seq <= last_sequence_number */
2628 ERR("Sequence number inconsistent : prev = %" PRIu64
2629 ", current = %" PRIu64
,
2630 stream
->last_sequence_number
, seq
);
2634 stream
->last_sequence_number
= seq
;
2636 ret
= ustctl_get_events_discarded(stream
->ustream
, &discarded
);
2638 PERROR("kernctl_get_events_discarded");
2641 if (discarded
< stream
->last_discarded_events
) {
2643 * Overflow has occurred. We assume only one wrap-around
2646 stream
->chan
->discarded_events
+=
2647 (1ULL << (CAA_BITS_PER_LONG
- 1)) -
2648 stream
->last_discarded_events
+ discarded
;
2650 stream
->chan
->discarded_events
+= discarded
-
2651 stream
->last_discarded_events
;
2653 stream
->last_discarded_events
= discarded
;
2661 * Read subbuffer from the given stream.
2663 * Stream lock MUST be acquired.
2665 * Return 0 on success else a negative value.
2667 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
2668 struct lttng_consumer_local_data
*ctx
, bool *rotated
)
2670 unsigned long len
, subbuf_size
, padding
;
2671 int err
, write_index
= 1, rotation_ret
;
2673 struct ustctl_consumer_stream
*ustream
;
2674 struct ctf_packet_index index
;
2677 assert(stream
->ustream
);
2680 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream
->wait_fd
,
2683 /* Ease our life for what's next. */
2684 ustream
= stream
->ustream
;
2687 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2688 * error if we cannot read this one byte (read returns 0), or if the error
2689 * is EAGAIN or EWOULDBLOCK.
2691 * This is only done when the stream is monitored by a thread, before the
2692 * flush is done after a hangup and if the stream is not flagged with data
2693 * since there might be nothing to consume in the wait fd but still have
2694 * data available flagged by the consumer wake up pipe.
2696 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2700 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2701 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2708 * If the stream was flagged to be ready for rotation before we extract the
2709 * next packet, rotate it now.
2711 if (stream
->rotate_ready
) {
2712 DBG("Rotate stream before extracting data");
2713 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
, rotated
);
2714 if (rotation_ret
< 0) {
2715 ERR("Stream rotation error");
2722 /* Get the next subbuffer */
2723 err
= ustctl_get_next_subbuf(ustream
);
2726 * Populate metadata info if the existing info has
2727 * already been read.
2729 if (stream
->metadata_flag
) {
2730 ret
= commit_one_metadata_packet(stream
);
2734 ustctl_flush_buffer(stream
->ustream
, 1);
2738 ret
= err
; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2740 * This is a debug message even for single-threaded consumer,
2741 * because poll() have more relaxed criterions than get subbuf,
2742 * so get_subbuf may fail for short race windows where poll()
2743 * would issue wakeups.
2745 DBG("Reserving sub buffer failed (everything is normal, "
2746 "it is due to concurrency) [ret: %d]", err
);
2749 assert(stream
->chan
->output
== CONSUMER_CHANNEL_MMAP
);
2751 if (!stream
->metadata_flag
) {
2752 index
.offset
= htobe64(stream
->out_fd_offset
);
2753 ret
= get_index_values(&index
, ustream
);
2755 err
= ustctl_put_subbuf(ustream
);
2760 /* Update the stream's sequence and discarded events count. */
2761 ret
= update_stream_stats(stream
);
2763 PERROR("kernctl_get_events_discarded");
2764 err
= ustctl_put_subbuf(ustream
);
2772 /* Get the full padded subbuffer size */
2773 err
= ustctl_get_padded_subbuf_size(ustream
, &len
);
2776 /* Get subbuffer data size (without padding) */
2777 err
= ustctl_get_subbuf_size(ustream
, &subbuf_size
);
2780 /* Make sure we don't get a subbuffer size bigger than the padded */
2781 assert(len
>= subbuf_size
);
2783 padding
= len
- subbuf_size
;
2785 /* write the subbuffer to the tracefile */
2786 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
, padding
, &index
);
2788 * The mmap operation should write subbuf_size amount of data when network
2789 * streaming or the full padding (len) size when we are _not_ streaming.
2791 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
2792 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
2794 * Display the error but continue processing to try to release the
2795 * subbuffer. This is a DBG statement since any unexpected kill or
2796 * signal, the application gets unregistered, relayd gets closed or
2797 * anything that affects the buffer lifetime will trigger this error.
2798 * So, for the sake of the user, don't print this error since it can
2799 * happen and it is OK with the code flow.
2801 DBG("Error writing to tracefile "
2802 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2803 ret
, len
, subbuf_size
);
2806 err
= ustctl_put_next_subbuf(ustream
);
2810 * This will consumer the byte on the wait_fd if and only if there is not
2811 * next subbuffer to be acquired.
2813 if (!stream
->metadata_flag
) {
2814 ret
= notify_if_more_data(stream
, ctx
);
2820 /* Write index if needed. */
2825 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
2827 * In live, block until all the metadata is sent.
2829 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2830 assert(!stream
->missed_metadata_flush
);
2831 stream
->waiting_on_metadata
= true;
2832 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2834 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
2836 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2837 stream
->waiting_on_metadata
= false;
2838 if (stream
->missed_metadata_flush
) {
2839 stream
->missed_metadata_flush
= false;
2840 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2841 (void) consumer_flush_ust_index(stream
);
2843 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2851 assert(!stream
->metadata_flag
);
2852 err
= consumer_stream_write_index(stream
, &index
);
2859 * After extracting the packet, we check if the stream is now ready to be
2860 * rotated and perform the action immediately.
2862 rotation_ret
= lttng_consumer_stream_is_rotate_ready(stream
);
2863 if (rotation_ret
== 1) {
2864 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
, rotated
);
2865 if (rotation_ret
< 0) {
2866 ERR("Stream rotation error");
2870 } else if (rotation_ret
< 0) {
2871 ERR("Checking if stream is ready to rotate");
2880 * Called when a stream is created.
2882 * Return 0 on success or else a negative value.
2884 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
2890 /* Don't create anything if this is set for streaming. */
2891 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
) {
2892 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
2893 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
2894 stream
->uid
, stream
->gid
, NULL
);
2898 stream
->out_fd
= ret
;
2899 stream
->tracefile_size_current
= 0;
2901 if (!stream
->metadata_flag
) {
2902 struct lttng_index_file
*index_file
;
2904 index_file
= lttng_index_file_create(stream
->chan
->pathname
,
2905 stream
->name
, stream
->uid
, stream
->gid
,
2906 stream
->chan
->tracefile_size
,
2907 stream
->tracefile_count_current
,
2908 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
2912 assert(!stream
->index_file
);
2913 stream
->index_file
= index_file
;
2923 * Check if data is still being extracted from the buffers for a specific
2924 * stream. Consumer data lock MUST be acquired before calling this function
2925 * and the stream lock.
2927 * Return 1 if the traced data are still getting read else 0 meaning that the
2928 * data is available for trace viewer reading.
2930 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
2935 assert(stream
->ustream
);
2937 DBG("UST consumer checking data pending");
2939 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
2944 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
2945 uint64_t contiguous
, pushed
;
2947 /* Ease our life a bit. */
2948 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
2949 pushed
= stream
->ust_metadata_pushed
;
2952 * We can simply check whether all contiguously available data
2953 * has been pushed to the ring buffer, since the push operation
2954 * is performed within get_next_subbuf(), and because both
2955 * get_next_subbuf() and put_next_subbuf() are issued atomically
2956 * thanks to the stream lock within
2957 * lttng_ustconsumer_read_subbuffer(). This basically means that
2958 * whetnever ust_metadata_pushed is incremented, the associated
2959 * metadata has been consumed from the metadata stream.
2961 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
2962 contiguous
, pushed
);
2963 assert(((int64_t) (contiguous
- pushed
)) >= 0);
2964 if ((contiguous
!= pushed
) ||
2965 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
2966 ret
= 1; /* Data is pending */
2970 ret
= ustctl_get_next_subbuf(stream
->ustream
);
2973 * There is still data so let's put back this
2976 ret
= ustctl_put_subbuf(stream
->ustream
);
2978 ret
= 1; /* Data is pending */
2983 /* Data is NOT pending so ready to be read. */
2991 * Stop a given metadata channel timer if enabled and close the wait fd which
2992 * is the poll pipe of the metadata stream.
2994 * This MUST be called with the metadata channel acquired.
2996 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
3001 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
3003 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
3005 if (metadata
->switch_timer_enabled
== 1) {
3006 consumer_timer_switch_stop(metadata
);
3009 if (!metadata
->metadata_stream
) {
3014 * Closing write side so the thread monitoring the stream wakes up if any
3015 * and clean the metadata stream.
3017 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
3018 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
3020 PERROR("closing metadata pipe write side");
3022 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
3030 * Close every metadata stream wait fd of the metadata hash table. This
3031 * function MUST be used very carefully so not to run into a race between the
3032 * metadata thread handling streams and this function closing their wait fd.
3034 * For UST, this is used when the session daemon hangs up. Its the metadata
3035 * producer so calling this is safe because we are assured that no state change
3036 * can occur in the metadata thread for the streams in the hash table.
3038 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
3040 struct lttng_ht_iter iter
;
3041 struct lttng_consumer_stream
*stream
;
3043 assert(metadata_ht
);
3044 assert(metadata_ht
->ht
);
3046 DBG("UST consumer closing all metadata streams");
3049 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
3052 health_code_update();
3054 pthread_mutex_lock(&stream
->chan
->lock
);
3055 lttng_ustconsumer_close_metadata(stream
->chan
);
3056 pthread_mutex_unlock(&stream
->chan
->lock
);
3062 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
3066 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
3068 ERR("Unable to close wakeup fd");
3073 * Please refer to consumer-timer.c before adding any lock within this
3074 * function or any of its callees. Timers have a very strict locking
3075 * semantic with respect to teardown. Failure to respect this semantic
3076 * introduces deadlocks.
3078 * DON'T hold the metadata lock when calling this function, else this
3079 * can cause deadlock involving consumer awaiting for metadata to be
3080 * pushed out due to concurrent interaction with the session daemon.
3082 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
3083 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
3085 struct lttcomm_metadata_request_msg request
;
3086 struct lttcomm_consumer_msg msg
;
3087 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3088 uint64_t len
, key
, offset
, version
;
3092 assert(channel
->metadata_cache
);
3094 memset(&request
, 0, sizeof(request
));
3096 /* send the metadata request to sessiond */
3097 switch (consumer_data
.type
) {
3098 case LTTNG_CONSUMER64_UST
:
3099 request
.bits_per_long
= 64;
3101 case LTTNG_CONSUMER32_UST
:
3102 request
.bits_per_long
= 32;
3105 request
.bits_per_long
= 0;
3109 request
.session_id
= channel
->session_id
;
3110 request
.session_id_per_pid
= channel
->session_id_per_pid
;
3112 * Request the application UID here so the metadata of that application can
3113 * be sent back. The channel UID corresponds to the user UID of the session
3114 * used for the rights on the stream file(s).
3116 request
.uid
= channel
->ust_app_uid
;
3117 request
.key
= channel
->key
;
3119 DBG("Sending metadata request to sessiond, session id %" PRIu64
3120 ", per-pid %" PRIu64
", app UID %u and channek key %" PRIu64
,
3121 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
3124 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
3126 health_code_update();
3128 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
3131 ERR("Asking metadata to sessiond");
3135 health_code_update();
3137 /* Receive the metadata from sessiond */
3138 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
3140 if (ret
!= sizeof(msg
)) {
3141 DBG("Consumer received unexpected message size %d (expects %zu)",
3143 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
3145 * The ret value might 0 meaning an orderly shutdown but this is ok
3146 * since the caller handles this.
3151 health_code_update();
3153 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
3154 /* No registry found */
3155 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3159 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
3160 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
3165 len
= msg
.u
.push_metadata
.len
;
3166 key
= msg
.u
.push_metadata
.key
;
3167 offset
= msg
.u
.push_metadata
.target_offset
;
3168 version
= msg
.u
.push_metadata
.version
;
3170 assert(key
== channel
->key
);
3172 DBG("No new metadata to receive for key %" PRIu64
, key
);
3175 health_code_update();
3177 /* Tell session daemon we are ready to receive the metadata. */
3178 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3179 LTTCOMM_CONSUMERD_SUCCESS
);
3180 if (ret
< 0 || len
== 0) {
3182 * Somehow, the session daemon is not responding anymore or there is
3183 * nothing to receive.
3188 health_code_update();
3190 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
3191 key
, offset
, len
, version
, channel
, timer
, wait
);
3194 * Only send the status msg if the sessiond is alive meaning a positive
3197 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
3202 health_code_update();
3204 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
3209 * Return the ustctl call for the get stream id.
3211 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
3212 uint64_t *stream_id
)
3217 return ustctl_get_stream_id(stream
->ustream
, stream_id
);