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 and relayd if applicable.
545 * Return 0 on success or else a negative value.
547 static int send_channel_to_sessiond_and_relayd(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 DBG("Sending stream %" PRIu64
" of channel \"%s\" to relayd",
568 stream
->key
, channel
->name
);
569 ret
= consumer_send_relayd_stream(stream
, stream
->chan
->pathname
);
572 * Flag that the relayd was the problem here probably due to a
573 * communicaton error on the socket.
578 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
580 if (net_seq_idx
== -1ULL) {
581 net_seq_idx
= stream
->net_seq_idx
;
586 /* Inform sessiond that we are about to send channel and streams. */
587 ret
= consumer_send_status_msg(sock
, ret_code
);
588 if (ret
< 0 || ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
590 * Either the session daemon is not responding or the relayd died so we
596 /* Send channel to sessiond. */
597 ret
= ustctl_send_channel_to_sessiond(sock
, channel
->uchan
);
602 ret
= ustctl_channel_close_wakeup_fd(channel
->uchan
);
607 /* The channel was sent successfully to the sessiond at this point. */
608 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
610 health_code_update();
612 /* Send stream to session daemon. */
613 ret
= send_sessiond_stream(sock
, stream
);
619 /* Tell sessiond there is no more stream. */
620 ret
= ustctl_send_stream_to_sessiond(sock
, NULL
);
625 DBG("UST consumer NULL stream sent to sessiond");
630 if (ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
637 * Creates a channel and streams and add the channel it to the channel internal
638 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
641 * Return 0 on success or else, a negative value is returned and the channel
642 * MUST be destroyed by consumer_del_channel().
644 static int ask_channel(struct lttng_consumer_local_data
*ctx
, int sock
,
645 struct lttng_consumer_channel
*channel
,
646 struct ustctl_consumer_channel_attr
*attr
)
655 * This value is still used by the kernel consumer since for the kernel,
656 * the stream ownership is not IN the consumer so we need to have the
657 * number of left stream that needs to be initialized so we can know when
658 * to delete the channel (see consumer.c).
660 * As for the user space tracer now, the consumer creates and sends the
661 * stream to the session daemon which only sends them to the application
662 * once every stream of a channel is received making this value useless
663 * because we they will be added to the poll thread before the application
664 * receives them. This ensures that a stream can not hang up during
665 * initilization of a channel.
667 channel
->nb_init_stream_left
= 0;
669 /* The reply msg status is handled in the following call. */
670 ret
= create_ust_channel(channel
, attr
, &channel
->uchan
);
675 channel
->wait_fd
= ustctl_channel_get_wait_fd(channel
->uchan
);
678 * For the snapshots (no monitor), we create the metadata streams
679 * on demand, not during the channel creation.
681 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& !channel
->monitor
) {
686 /* Open all streams for this channel. */
687 ret
= create_ust_streams(channel
, ctx
);
697 * Send all stream of a channel to the right thread handling it.
699 * On error, return a negative value else 0 on success.
701 static int send_streams_to_thread(struct lttng_consumer_channel
*channel
,
702 struct lttng_consumer_local_data
*ctx
)
705 struct lttng_consumer_stream
*stream
, *stmp
;
710 /* Send streams to the corresponding thread. */
711 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
714 health_code_update();
716 /* Sending the stream to the thread. */
717 ret
= send_stream_to_thread(stream
, ctx
);
720 * If we are unable to send the stream to the thread, there is
721 * a big problem so just stop everything.
732 * Flush channel's streams using the given key to retrieve the channel.
734 * Return 0 on success else an LTTng error code.
736 static int flush_channel(uint64_t chan_key
)
739 struct lttng_consumer_channel
*channel
;
740 struct lttng_consumer_stream
*stream
;
742 struct lttng_ht_iter iter
;
744 DBG("UST consumer flush channel key %" PRIu64
, chan_key
);
747 channel
= consumer_find_channel(chan_key
);
749 ERR("UST consumer flush channel %" PRIu64
" not found", chan_key
);
750 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
754 ht
= consumer_data
.stream_per_chan_id_ht
;
756 /* For each stream of the channel id, flush it. */
757 cds_lfht_for_each_entry_duplicate(ht
->ht
,
758 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
759 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
761 health_code_update();
763 pthread_mutex_lock(&stream
->lock
);
764 if (!stream
->quiescent
) {
765 ustctl_flush_buffer(stream
->ustream
, 0);
766 stream
->quiescent
= true;
768 pthread_mutex_unlock(&stream
->lock
);
776 * Clear quiescent state from channel's streams using the given key to
777 * retrieve the channel.
779 * Return 0 on success else an LTTng error code.
781 static int clear_quiescent_channel(uint64_t chan_key
)
784 struct lttng_consumer_channel
*channel
;
785 struct lttng_consumer_stream
*stream
;
787 struct lttng_ht_iter iter
;
789 DBG("UST consumer clear quiescent channel key %" PRIu64
, chan_key
);
792 channel
= consumer_find_channel(chan_key
);
794 ERR("UST consumer clear quiescent channel %" PRIu64
" not found", chan_key
);
795 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
799 ht
= consumer_data
.stream_per_chan_id_ht
;
801 /* For each stream of the channel id, clear quiescent state. */
802 cds_lfht_for_each_entry_duplicate(ht
->ht
,
803 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
804 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
806 health_code_update();
808 pthread_mutex_lock(&stream
->lock
);
809 stream
->quiescent
= false;
810 pthread_mutex_unlock(&stream
->lock
);
818 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
819 * RCU read side lock MUST be acquired before calling this function.
821 * Return 0 on success else an LTTng error code.
823 static int close_metadata(uint64_t chan_key
)
826 struct lttng_consumer_channel
*channel
;
827 unsigned int channel_monitor
;
829 DBG("UST consumer close metadata key %" PRIu64
, chan_key
);
831 channel
= consumer_find_channel(chan_key
);
834 * This is possible if the metadata thread has issue a delete because
835 * the endpoint point of the stream hung up. There is no way the
836 * session daemon can know about it thus use a DBG instead of an actual
839 DBG("UST consumer close metadata %" PRIu64
" not found", chan_key
);
840 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
844 pthread_mutex_lock(&consumer_data
.lock
);
845 pthread_mutex_lock(&channel
->lock
);
846 channel_monitor
= channel
->monitor
;
847 if (cds_lfht_is_node_deleted(&channel
->node
.node
)) {
851 lttng_ustconsumer_close_metadata(channel
);
852 pthread_mutex_unlock(&channel
->lock
);
853 pthread_mutex_unlock(&consumer_data
.lock
);
856 * The ownership of a metadata channel depends on the type of
857 * session to which it belongs. In effect, the monitor flag is checked
858 * to determine if this metadata channel is in "snapshot" mode or not.
860 * In the non-snapshot case, the metadata channel is created along with
861 * a single stream which will remain present until the metadata channel
862 * is destroyed (on the destruction of its session). In this case, the
863 * metadata stream in "monitored" by the metadata poll thread and holds
864 * the ownership of its channel.
866 * Closing the metadata will cause the metadata stream's "metadata poll
867 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
868 * thread which will teardown the metadata stream which, in return,
869 * deletes the metadata channel.
871 * In the snapshot case, the metadata stream is created and destroyed
872 * on every snapshot record. Since the channel doesn't have an owner
873 * other than the session daemon, it is safe to destroy it immediately
874 * on reception of the CLOSE_METADATA command.
876 if (!channel_monitor
) {
878 * The channel and consumer_data locks must be
879 * released before this call since consumer_del_channel
880 * re-acquires the channel and consumer_data locks to teardown
881 * the channel and queue its reclamation by the "call_rcu"
884 consumer_del_channel(channel
);
889 pthread_mutex_unlock(&channel
->lock
);
890 pthread_mutex_unlock(&consumer_data
.lock
);
896 * RCU read side lock MUST be acquired before calling this function.
898 * Return 0 on success else an LTTng error code.
900 static int setup_metadata(struct lttng_consumer_local_data
*ctx
, uint64_t key
)
903 struct lttng_consumer_channel
*metadata
;
905 DBG("UST consumer setup metadata key %" PRIu64
, key
);
907 metadata
= consumer_find_channel(key
);
909 ERR("UST consumer push metadata %" PRIu64
" not found", key
);
910 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
915 * In no monitor mode, the metadata channel has no stream(s) so skip the
916 * ownership transfer to the metadata thread.
918 if (!metadata
->monitor
) {
919 DBG("Metadata channel in no monitor");
925 * Send metadata stream to relayd if one available. Availability is
926 * known if the stream is still in the list of the channel.
928 if (cds_list_empty(&metadata
->streams
.head
)) {
929 ERR("Metadata channel key %" PRIu64
", no stream available.", key
);
930 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
931 goto error_no_stream
;
934 /* Send metadata stream to relayd if needed. */
935 if (metadata
->metadata_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
936 ret
= consumer_send_relayd_stream(metadata
->metadata_stream
,
939 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
942 ret
= consumer_send_relayd_streams_sent(
943 metadata
->metadata_stream
->net_seq_idx
);
945 ret
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
951 * Ownership of metadata stream is passed along. Freeing is handled by
954 ret
= send_streams_to_thread(metadata
, ctx
);
957 * If we are unable to send the stream to the thread, there is
958 * a big problem so just stop everything.
960 ret
= LTTCOMM_CONSUMERD_FATAL
;
961 goto send_streams_error
;
963 /* List MUST be empty after or else it could be reused. */
964 assert(cds_list_empty(&metadata
->streams
.head
));
971 * Delete metadata channel on error. At this point, the metadata stream can
972 * NOT be monitored by the metadata thread thus having the guarantee that
973 * the stream is still in the local stream list of the channel. This call
974 * will make sure to clean that list.
976 consumer_stream_destroy(metadata
->metadata_stream
, NULL
);
977 cds_list_del(&metadata
->metadata_stream
->send_node
);
978 metadata
->metadata_stream
= NULL
;
986 * Snapshot the whole metadata.
988 * Returns 0 on success, < 0 on error
990 static int snapshot_metadata(uint64_t key
, char *path
, uint64_t relayd_id
,
991 struct lttng_consumer_local_data
*ctx
)
994 struct lttng_consumer_channel
*metadata_channel
;
995 struct lttng_consumer_stream
*metadata_stream
;
1000 DBG("UST consumer snapshot metadata with key %" PRIu64
" at path %s",
1005 metadata_channel
= consumer_find_channel(key
);
1006 if (!metadata_channel
) {
1007 ERR("UST snapshot metadata channel not found for key %" PRIu64
,
1012 assert(!metadata_channel
->monitor
);
1014 health_code_update();
1017 * Ask the sessiond if we have new metadata waiting and update the
1018 * consumer metadata cache.
1020 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 1);
1025 health_code_update();
1028 * The metadata stream is NOT created in no monitor mode when the channel
1029 * is created on a sessiond ask channel command.
1031 ret
= create_ust_streams(metadata_channel
, ctx
);
1036 metadata_stream
= metadata_channel
->metadata_stream
;
1037 assert(metadata_stream
);
1039 if (relayd_id
!= (uint64_t) -1ULL) {
1040 metadata_stream
->net_seq_idx
= relayd_id
;
1041 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
1046 ret
= utils_create_stream_file(path
, metadata_stream
->name
,
1047 metadata_stream
->chan
->tracefile_size
,
1048 metadata_stream
->tracefile_count_current
,
1049 metadata_stream
->uid
, metadata_stream
->gid
, NULL
);
1053 metadata_stream
->out_fd
= ret
;
1054 metadata_stream
->tracefile_size_current
= 0;
1058 health_code_update();
1060 ret
= lttng_consumer_read_subbuffer(metadata_stream
, ctx
);
1068 * Clean up the stream completly because the next snapshot will use a new
1071 consumer_stream_destroy(metadata_stream
, NULL
);
1072 cds_list_del(&metadata_stream
->send_node
);
1073 metadata_channel
->metadata_stream
= NULL
;
1081 * Take a snapshot of all the stream of a channel.
1083 * Returns 0 on success, < 0 on error
1085 static int snapshot_channel(uint64_t key
, char *path
, uint64_t relayd_id
,
1086 uint64_t nb_packets_per_stream
, struct lttng_consumer_local_data
*ctx
)
1089 unsigned use_relayd
= 0;
1090 unsigned long consumed_pos
, produced_pos
;
1091 struct lttng_consumer_channel
*channel
;
1092 struct lttng_consumer_stream
*stream
;
1099 if (relayd_id
!= (uint64_t) -1ULL) {
1103 channel
= consumer_find_channel(key
);
1105 ERR("UST snapshot channel not found for key %" PRIu64
, key
);
1109 assert(!channel
->monitor
);
1110 DBG("UST consumer snapshot channel %" PRIu64
, key
);
1112 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
1113 health_code_update();
1115 /* Lock stream because we are about to change its state. */
1116 pthread_mutex_lock(&stream
->lock
);
1117 stream
->net_seq_idx
= relayd_id
;
1120 ret
= consumer_send_relayd_stream(stream
, path
);
1125 ret
= utils_create_stream_file(path
, stream
->name
,
1126 stream
->chan
->tracefile_size
,
1127 stream
->tracefile_count_current
,
1128 stream
->uid
, stream
->gid
, NULL
);
1132 stream
->out_fd
= ret
;
1133 stream
->tracefile_size_current
= 0;
1135 DBG("UST consumer snapshot stream %s/%s (%" PRIu64
")", path
,
1136 stream
->name
, stream
->key
);
1140 * If tracing is active, we want to perform a "full" buffer flush.
1141 * Else, if quiescent, it has already been done by the prior stop.
1143 if (!stream
->quiescent
) {
1144 ustctl_flush_buffer(stream
->ustream
, 0);
1147 ret
= lttng_ustconsumer_take_snapshot(stream
);
1149 ERR("Taking UST snapshot");
1153 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
1155 ERR("Produced UST snapshot position");
1159 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1161 ERR("Consumerd UST snapshot position");
1166 * The original value is sent back if max stream size is larger than
1167 * the possible size of the snapshot. Also, we assume that the session
1168 * daemon should never send a maximum stream size that is lower than
1171 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
1172 produced_pos
, nb_packets_per_stream
,
1173 stream
->max_sb_size
);
1175 while (consumed_pos
< produced_pos
) {
1177 unsigned long len
, padded_len
;
1179 health_code_update();
1181 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1183 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1185 if (ret
!= -EAGAIN
) {
1186 PERROR("ustctl_get_subbuf snapshot");
1187 goto error_close_stream
;
1189 DBG("UST consumer get subbuf failed. Skipping it.");
1190 consumed_pos
+= stream
->max_sb_size
;
1191 stream
->chan
->lost_packets
++;
1195 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1197 ERR("Snapshot ustctl_get_subbuf_size");
1198 goto error_put_subbuf
;
1201 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1203 ERR("Snapshot ustctl_get_padded_subbuf_size");
1204 goto error_put_subbuf
;
1207 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, len
,
1208 padded_len
- len
, NULL
);
1210 if (read_len
!= len
) {
1212 goto error_put_subbuf
;
1215 if (read_len
!= padded_len
) {
1217 goto error_put_subbuf
;
1221 ret
= ustctl_put_subbuf(stream
->ustream
);
1223 ERR("Snapshot ustctl_put_subbuf");
1224 goto error_close_stream
;
1226 consumed_pos
+= stream
->max_sb_size
;
1229 /* Simply close the stream so we can use it on the next snapshot. */
1230 consumer_stream_close(stream
);
1231 pthread_mutex_unlock(&stream
->lock
);
1238 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1239 ERR("Snapshot ustctl_put_subbuf");
1242 consumer_stream_close(stream
);
1244 pthread_mutex_unlock(&stream
->lock
);
1251 * Receive the metadata updates from the sessiond. Supports receiving
1252 * overlapping metadata, but is needs to always belong to a contiguous
1253 * range starting from 0.
1254 * Be careful about the locks held when calling this function: it needs
1255 * the metadata cache flush to concurrently progress in order to
1258 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1259 uint64_t len
, uint64_t version
,
1260 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
1262 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1265 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1267 metadata_str
= zmalloc(len
* sizeof(char));
1268 if (!metadata_str
) {
1269 PERROR("zmalloc metadata string");
1270 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1274 health_code_update();
1276 /* Receive metadata string. */
1277 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1279 /* Session daemon is dead so return gracefully. */
1284 health_code_update();
1286 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1287 ret
= consumer_metadata_cache_write(channel
, offset
, len
, version
,
1290 /* Unable to handle metadata. Notify session daemon. */
1291 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1293 * Skip metadata flush on write error since the offset and len might
1294 * not have been updated which could create an infinite loop below when
1295 * waiting for the metadata cache to be flushed.
1297 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1300 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1305 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1306 DBG("Waiting for metadata to be flushed");
1308 health_code_update();
1310 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1320 * Receive command from session daemon and process it.
1322 * Return 1 on success else a negative value or 0.
1324 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1325 int sock
, struct pollfd
*consumer_sockpoll
)
1328 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1329 struct lttcomm_consumer_msg msg
;
1330 struct lttng_consumer_channel
*channel
= NULL
;
1332 health_code_update();
1334 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1335 if (ret
!= sizeof(msg
)) {
1336 DBG("Consumer received unexpected message size %zd (expects %zu)",
1339 * The ret value might 0 meaning an orderly shutdown but this is ok
1340 * since the caller handles this.
1343 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1349 health_code_update();
1352 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1354 health_code_update();
1356 /* relayd needs RCU read-side lock */
1359 switch (msg
.cmd_type
) {
1360 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1362 /* Session daemon status message are handled in the following call. */
1363 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1364 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
1365 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
1366 msg
.u
.relayd_sock
.relayd_session_id
);
1369 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1371 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1372 struct consumer_relayd_sock_pair
*relayd
;
1374 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1376 /* Get relayd reference if exists. */
1377 relayd
= consumer_find_relayd(index
);
1378 if (relayd
== NULL
) {
1379 DBG("Unable to find relayd %" PRIu64
, index
);
1380 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1384 * Each relayd socket pair has a refcount of stream attached to it
1385 * which tells if the relayd is still active or not depending on the
1388 * This will set the destroy flag of the relayd object and destroy it
1389 * if the refcount reaches zero when called.
1391 * The destroy can happen either here or when a stream fd hangs up.
1394 consumer_flag_relayd_for_destroy(relayd
);
1397 goto end_msg_sessiond
;
1399 case LTTNG_CONSUMER_UPDATE_STREAM
:
1404 case LTTNG_CONSUMER_DATA_PENDING
:
1406 int ret
, is_data_pending
;
1407 uint64_t id
= msg
.u
.data_pending
.session_id
;
1409 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1411 is_data_pending
= consumer_data_pending(id
);
1413 /* Send back returned value to session daemon */
1414 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1415 sizeof(is_data_pending
));
1417 DBG("Error when sending the data pending ret code: %d", ret
);
1422 * No need to send back a status message since the data pending
1423 * returned value is the response.
1427 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1430 struct ustctl_consumer_channel_attr attr
;
1432 /* Create a plain object and reserve a channel key. */
1433 channel
= allocate_channel(msg
.u
.ask_channel
.session_id
,
1434 msg
.u
.ask_channel
.pathname
, msg
.u
.ask_channel
.name
,
1435 msg
.u
.ask_channel
.uid
, msg
.u
.ask_channel
.gid
,
1436 msg
.u
.ask_channel
.relayd_id
, msg
.u
.ask_channel
.key
,
1437 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1438 msg
.u
.ask_channel
.tracefile_size
,
1439 msg
.u
.ask_channel
.tracefile_count
,
1440 msg
.u
.ask_channel
.session_id_per_pid
,
1441 msg
.u
.ask_channel
.monitor
,
1442 msg
.u
.ask_channel
.live_timer_interval
,
1443 msg
.u
.ask_channel
.root_shm_path
,
1444 msg
.u
.ask_channel
.shm_path
);
1446 goto end_channel_error
;
1450 * Assign UST application UID to the channel. This value is ignored for
1451 * per PID buffers. This is specific to UST thus setting this after the
1454 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1456 /* Build channel attributes from received message. */
1457 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1458 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1459 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1460 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1461 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1462 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1463 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1464 attr
.blocking_timeout
= msg
.u
.ask_channel
.blocking_timeout
;
1466 /* Match channel buffer type to the UST abi. */
1467 switch (msg
.u
.ask_channel
.output
) {
1468 case LTTNG_EVENT_MMAP
:
1470 attr
.output
= LTTNG_UST_MMAP
;
1474 /* Translate and save channel type. */
1475 switch (msg
.u
.ask_channel
.type
) {
1476 case LTTNG_UST_CHAN_PER_CPU
:
1477 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1478 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1480 * Set refcount to 1 for owner. Below, we will
1481 * pass ownership to the
1482 * consumer_thread_channel_poll() thread.
1484 channel
->refcount
= 1;
1486 case LTTNG_UST_CHAN_METADATA
:
1487 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1488 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1495 health_code_update();
1497 ret
= ask_channel(ctx
, sock
, channel
, &attr
);
1499 goto end_channel_error
;
1502 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1503 ret
= consumer_metadata_cache_allocate(channel
);
1505 ERR("Allocating metadata cache");
1506 goto end_channel_error
;
1508 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1509 attr
.switch_timer_interval
= 0;
1511 int monitor_start_ret
;
1513 consumer_timer_live_start(channel
,
1514 msg
.u
.ask_channel
.live_timer_interval
);
1515 monitor_start_ret
= consumer_timer_monitor_start(
1517 msg
.u
.ask_channel
.monitor_timer_interval
);
1518 if (monitor_start_ret
< 0) {
1519 ERR("Starting channel monitoring timer failed");
1520 goto end_channel_error
;
1524 health_code_update();
1527 * Add the channel to the internal state AFTER all streams were created
1528 * and successfully sent to session daemon. This way, all streams must
1529 * be ready before this channel is visible to the threads.
1530 * If add_channel succeeds, ownership of the channel is
1531 * passed to consumer_thread_channel_poll().
1533 ret
= add_channel(channel
, ctx
);
1535 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1536 if (channel
->switch_timer_enabled
== 1) {
1537 consumer_timer_switch_stop(channel
);
1539 consumer_metadata_cache_destroy(channel
);
1541 if (channel
->live_timer_enabled
== 1) {
1542 consumer_timer_live_stop(channel
);
1544 if (channel
->monitor_timer_enabled
== 1) {
1545 consumer_timer_monitor_stop(channel
);
1547 goto end_channel_error
;
1550 health_code_update();
1553 * Channel and streams are now created. Inform the session daemon that
1554 * everything went well and should wait to receive the channel and
1555 * streams with ustctl API.
1557 ret
= consumer_send_status_channel(sock
, channel
);
1560 * There is probably a problem on the socket.
1567 case LTTNG_CONSUMER_GET_CHANNEL
:
1569 int ret
, relayd_err
= 0;
1570 uint64_t key
= msg
.u
.get_channel
.key
;
1571 struct lttng_consumer_channel
*channel
;
1573 channel
= consumer_find_channel(key
);
1575 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1576 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1577 goto end_msg_sessiond
;
1580 health_code_update();
1582 /* Send the channel to sessiond (and relayd, if applicable). */
1583 ret
= send_channel_to_sessiond_and_relayd(sock
, channel
, ctx
,
1588 * We were unable to send to the relayd the stream so avoid
1589 * sending back a fatal error to the thread since this is OK
1590 * and the consumer can continue its work. The above call
1591 * has sent the error status message to the sessiond.
1596 * The communicaton was broken hence there is a bad state between
1597 * the consumer and sessiond so stop everything.
1602 health_code_update();
1605 * In no monitor mode, the streams ownership is kept inside the channel
1606 * so don't send them to the data thread.
1608 if (!channel
->monitor
) {
1609 goto end_msg_sessiond
;
1612 ret
= send_streams_to_thread(channel
, ctx
);
1615 * If we are unable to send the stream to the thread, there is
1616 * a big problem so just stop everything.
1620 /* List MUST be empty after or else it could be reused. */
1621 assert(cds_list_empty(&channel
->streams
.head
));
1622 goto end_msg_sessiond
;
1624 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1626 uint64_t key
= msg
.u
.destroy_channel
.key
;
1629 * Only called if streams have not been sent to stream
1630 * manager thread. However, channel has been sent to
1631 * channel manager thread.
1633 notify_thread_del_channel(ctx
, key
);
1634 goto end_msg_sessiond
;
1636 case LTTNG_CONSUMER_CLOSE_METADATA
:
1640 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1645 goto end_msg_sessiond
;
1647 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1651 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1656 goto end_msg_sessiond
;
1658 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL
:
1662 ret
= clear_quiescent_channel(
1663 msg
.u
.clear_quiescent_channel
.key
);
1668 goto end_msg_sessiond
;
1670 case LTTNG_CONSUMER_PUSH_METADATA
:
1673 uint64_t len
= msg
.u
.push_metadata
.len
;
1674 uint64_t key
= msg
.u
.push_metadata
.key
;
1675 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1676 uint64_t version
= msg
.u
.push_metadata
.version
;
1677 struct lttng_consumer_channel
*channel
;
1679 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1682 channel
= consumer_find_channel(key
);
1685 * This is possible if the metadata creation on the consumer side
1686 * is in flight vis-a-vis a concurrent push metadata from the
1687 * session daemon. Simply return that the channel failed and the
1688 * session daemon will handle that message correctly considering
1689 * that this race is acceptable thus the DBG() statement here.
1691 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1692 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1693 goto end_msg_sessiond
;
1696 health_code_update();
1700 * There is nothing to receive. We have simply
1701 * checked whether the channel can be found.
1703 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1704 goto end_msg_sessiond
;
1707 /* Tell session daemon we are ready to receive the metadata. */
1708 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1710 /* Somehow, the session daemon is not responding anymore. */
1714 health_code_update();
1716 /* Wait for more data. */
1717 health_poll_entry();
1718 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1724 health_code_update();
1726 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1727 len
, version
, channel
, 0, 1);
1729 /* error receiving from sessiond */
1733 goto end_msg_sessiond
;
1736 case LTTNG_CONSUMER_SETUP_METADATA
:
1740 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1744 goto end_msg_sessiond
;
1746 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1748 if (msg
.u
.snapshot_channel
.metadata
) {
1749 ret
= snapshot_metadata(msg
.u
.snapshot_channel
.key
,
1750 msg
.u
.snapshot_channel
.pathname
,
1751 msg
.u
.snapshot_channel
.relayd_id
,
1754 ERR("Snapshot metadata failed");
1755 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1758 ret
= snapshot_channel(msg
.u
.snapshot_channel
.key
,
1759 msg
.u
.snapshot_channel
.pathname
,
1760 msg
.u
.snapshot_channel
.relayd_id
,
1761 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
1764 ERR("Snapshot channel failed");
1765 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1769 health_code_update();
1770 ret
= consumer_send_status_msg(sock
, ret_code
);
1772 /* Somehow, the session daemon is not responding anymore. */
1775 health_code_update();
1778 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
1781 uint64_t discarded_events
;
1782 struct lttng_ht_iter iter
;
1783 struct lttng_ht
*ht
;
1784 struct lttng_consumer_stream
*stream
;
1785 uint64_t id
= msg
.u
.discarded_events
.session_id
;
1786 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
1788 DBG("UST consumer discarded events command for session id %"
1791 pthread_mutex_lock(&consumer_data
.lock
);
1793 ht
= consumer_data
.stream_list_ht
;
1796 * We only need a reference to the channel, but they are not
1797 * directly indexed, so we just use the first matching stream
1798 * to extract the information we need, we default to 0 if not
1799 * found (no events are dropped if the channel is not yet in
1802 discarded_events
= 0;
1803 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1804 ht
->hash_fct(&id
, lttng_ht_seed
),
1806 &iter
.iter
, stream
, node_session_id
.node
) {
1807 if (stream
->chan
->key
== key
) {
1808 discarded_events
= stream
->chan
->discarded_events
;
1812 pthread_mutex_unlock(&consumer_data
.lock
);
1815 DBG("UST consumer discarded events command for session id %"
1816 PRIu64
", channel key %" PRIu64
, id
, key
);
1818 health_code_update();
1820 /* Send back returned value to session daemon */
1821 ret
= lttcomm_send_unix_sock(sock
, &discarded_events
, sizeof(discarded_events
));
1823 PERROR("send discarded events");
1829 case LTTNG_CONSUMER_LOST_PACKETS
:
1832 uint64_t lost_packets
;
1833 struct lttng_ht_iter iter
;
1834 struct lttng_ht
*ht
;
1835 struct lttng_consumer_stream
*stream
;
1836 uint64_t id
= msg
.u
.lost_packets
.session_id
;
1837 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
1839 DBG("UST consumer lost packets command for session id %"
1842 pthread_mutex_lock(&consumer_data
.lock
);
1844 ht
= consumer_data
.stream_list_ht
;
1847 * We only need a reference to the channel, but they are not
1848 * directly indexed, so we just use the first matching stream
1849 * to extract the information we need, we default to 0 if not
1850 * found (no packets lost if the channel is not yet in use).
1853 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1854 ht
->hash_fct(&id
, lttng_ht_seed
),
1856 &iter
.iter
, stream
, node_session_id
.node
) {
1857 if (stream
->chan
->key
== key
) {
1858 lost_packets
= stream
->chan
->lost_packets
;
1862 pthread_mutex_unlock(&consumer_data
.lock
);
1865 DBG("UST consumer lost packets command for session id %"
1866 PRIu64
", channel key %" PRIu64
, id
, key
);
1868 health_code_update();
1870 /* Send back returned value to session daemon */
1871 ret
= lttcomm_send_unix_sock(sock
, &lost_packets
,
1872 sizeof(lost_packets
));
1874 PERROR("send lost packets");
1880 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
1882 int channel_monitor_pipe
;
1884 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1885 /* Successfully received the command's type. */
1886 ret
= consumer_send_status_msg(sock
, ret_code
);
1891 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
1893 if (ret
!= sizeof(channel_monitor_pipe
)) {
1894 ERR("Failed to receive channel monitor pipe");
1898 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
1899 ret
= consumer_timer_thread_set_channel_monitor_pipe(
1900 channel_monitor_pipe
);
1904 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1905 /* Set the pipe as non-blocking. */
1906 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
1908 PERROR("fcntl get flags of the channel monitoring pipe");
1913 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
1914 flags
| O_NONBLOCK
);
1916 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1919 DBG("Channel monitor pipe set as non-blocking");
1921 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
1923 goto end_msg_sessiond
;
1925 case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE
:
1927 int channel_rotate_pipe
;
1930 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1931 /* Successfully received the command's type. */
1932 ret
= consumer_send_status_msg(sock
, ret_code
);
1937 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_rotate_pipe
, 1);
1938 if (ret
!= sizeof(channel_rotate_pipe
)) {
1939 ERR("Failed to receive channel rotate pipe");
1943 DBG("Received channel rotate pipe (%d)", channel_rotate_pipe
);
1944 ctx
->channel_rotate_pipe
= channel_rotate_pipe
;
1945 /* Set the pipe as non-blocking. */
1946 ret
= fcntl(channel_rotate_pipe
, F_GETFL
, 0);
1948 PERROR("fcntl get flags of the channel rotate pipe");
1953 ret
= fcntl(channel_rotate_pipe
, F_SETFL
, flags
| O_NONBLOCK
);
1955 PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
1958 DBG("Channel rotate pipe set as non-blocking");
1959 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1960 ret
= consumer_send_status_msg(sock
, ret_code
);
1966 case LTTNG_CONSUMER_ROTATE_CHANNEL
:
1969 * Sample the rotate position of all the streams in this channel.
1971 ret
= lttng_consumer_rotate_channel(msg
.u
.rotate_channel
.key
,
1972 msg
.u
.rotate_channel
.pathname
,
1973 msg
.u
.rotate_channel
.relayd_id
,
1974 msg
.u
.rotate_channel
.metadata
,
1975 msg
.u
.rotate_channel
.new_chunk_id
,
1978 ERR("Rotate channel failed");
1979 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1982 health_code_update();
1984 ret
= consumer_send_status_msg(sock
, ret_code
);
1986 /* Somehow, the session daemon is not responding anymore. */
1991 * Rotate the streams that are ready right now.
1992 * FIXME: this is a second consecutive iteration over the
1993 * streams in a channel, there is probably a better way to
1994 * handle this, but it needs to be after the
1995 * consumer_send_status_msg() call.
1997 ret
= lttng_consumer_rotate_ready_streams(
1998 msg
.u
.rotate_channel
.key
, ctx
);
2000 ERR("Rotate channel failed");
2001 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2005 case LTTNG_CONSUMER_ROTATE_RENAME
:
2007 DBG("Consumer rename session %" PRIu64
" after rotation",
2008 msg
.u
.rotate_rename
.session_id
);
2009 ret
= lttng_consumer_rotate_rename(msg
.u
.rotate_rename
.old_path
,
2010 msg
.u
.rotate_rename
.new_path
,
2011 msg
.u
.rotate_rename
.uid
,
2012 msg
.u
.rotate_rename
.gid
,
2013 msg
.u
.rotate_rename
.relayd_id
);
2015 ERR("Rotate rename failed");
2016 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
2019 health_code_update();
2021 ret
= consumer_send_status_msg(sock
, ret_code
);
2023 /* Somehow, the session daemon is not responding anymore. */
2028 case LTTNG_CONSUMER_ROTATE_PENDING_RELAY
:
2032 DBG("Consumer rotate pending on relay for session %" PRIu64
,
2033 msg
.u
.rotate_pending_relay
.session_id
);
2034 pending
= lttng_consumer_rotate_pending_relay(
2035 msg
.u
.rotate_pending_relay
.session_id
,
2036 msg
.u
.rotate_pending_relay
.relayd_id
,
2037 msg
.u
.rotate_pending_relay
.chunk_id
);
2039 ERR("Rotate pending relay failed");
2040 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
2043 health_code_update();
2045 ret
= consumer_send_status_msg(sock
, ret_code
);
2047 /* Somehow, the session daemon is not responding anymore. */
2051 /* Send back returned value to session daemon */
2052 ret
= lttcomm_send_unix_sock(sock
, &pending
, sizeof(pending
));
2054 PERROR("send data pending ret code");
2059 case LTTNG_CONSUMER_MKDIR
:
2061 DBG("Consumer mkdir %s in session %" PRIu64
,
2063 msg
.u
.mkdir
.session_id
);
2064 ret
= lttng_consumer_mkdir(msg
.u
.mkdir
.path
,
2067 msg
.u
.mkdir
.relayd_id
);
2069 ERR("consumer mkdir failed");
2070 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
2073 health_code_update();
2075 ret
= consumer_send_status_msg(sock
, ret_code
);
2077 /* Somehow, the session daemon is not responding anymore. */
2089 health_code_update();
2092 * Return 1 to indicate success since the 0 value can be a socket
2093 * shutdown during the recv() or send() call.
2099 * The returned value here is not useful since either way we'll return 1 to
2100 * the caller because the session daemon socket management is done
2101 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2103 ret
= consumer_send_status_msg(sock
, ret_code
);
2109 health_code_update();
2115 * Free channel here since no one has a reference to it. We don't
2116 * free after that because a stream can store this pointer.
2118 destroy_channel(channel
);
2120 /* We have to send a status channel message indicating an error. */
2121 ret
= consumer_send_status_channel(sock
, NULL
);
2123 /* Stop everything if session daemon can not be notified. */
2128 health_code_update();
2133 /* This will issue a consumer stop. */
2138 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2139 * compiled out, we isolate it in this library.
2141 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream
*stream
,
2145 assert(stream
->ustream
);
2147 return ustctl_get_mmap_read_offset(stream
->ustream
, off
);
2151 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2152 * compiled out, we isolate it in this library.
2154 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream
*stream
)
2157 assert(stream
->ustream
);
2159 return ustctl_get_mmap_base(stream
->ustream
);
2162 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream
*stream
,
2163 int producer_active
)
2166 assert(stream
->ustream
);
2168 ustctl_flush_buffer(stream
->ustream
, producer_active
);
2172 * Take a snapshot for a specific stream.
2174 * Returns 0 on success, < 0 on error
2176 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2179 assert(stream
->ustream
);
2181 return ustctl_snapshot(stream
->ustream
);
2185 * Sample consumed and produced positions for a specific stream.
2187 * Returns 0 on success, < 0 on error.
2189 int lttng_ustconsumer_sample_snapshot_positions(
2190 struct lttng_consumer_stream
*stream
)
2193 assert(stream
->ustream
);
2195 return ustctl_snapshot_sample_positions(stream
->ustream
);
2199 * Get the produced position
2201 * Returns 0 on success, < 0 on error
2203 int lttng_ustconsumer_get_produced_snapshot(
2204 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2207 assert(stream
->ustream
);
2210 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
2214 * Get the consumed position
2216 * Returns 0 on success, < 0 on error
2218 int lttng_ustconsumer_get_consumed_snapshot(
2219 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2222 assert(stream
->ustream
);
2225 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
2228 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
2232 assert(stream
->ustream
);
2234 ustctl_flush_buffer(stream
->ustream
, producer
);
2237 int lttng_ustconsumer_get_current_timestamp(
2238 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
2241 assert(stream
->ustream
);
2244 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
2247 int lttng_ustconsumer_get_sequence_number(
2248 struct lttng_consumer_stream
*stream
, uint64_t *seq
)
2251 assert(stream
->ustream
);
2254 return ustctl_get_sequence_number(stream
->ustream
, seq
);
2258 * Called when the stream signals the consumer that it has hung up.
2260 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
2263 assert(stream
->ustream
);
2265 pthread_mutex_lock(&stream
->lock
);
2266 if (!stream
->quiescent
) {
2267 ustctl_flush_buffer(stream
->ustream
, 0);
2268 stream
->quiescent
= true;
2270 pthread_mutex_unlock(&stream
->lock
);
2271 stream
->hangup_flush_done
= 1;
2274 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
2279 assert(chan
->uchan
);
2281 if (chan
->switch_timer_enabled
== 1) {
2282 consumer_timer_switch_stop(chan
);
2284 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
2287 ret
= close(chan
->stream_fds
[i
]);
2291 if (chan
->shm_path
[0]) {
2292 char shm_path
[PATH_MAX
];
2294 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
2296 ERR("Cannot get stream shm path");
2298 ret
= run_as_unlink(shm_path
, chan
->uid
, chan
->gid
);
2300 PERROR("unlink %s", shm_path
);
2306 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
2309 assert(chan
->uchan
);
2311 consumer_metadata_cache_destroy(chan
);
2312 ustctl_destroy_channel(chan
->uchan
);
2313 /* Try to rmdir all directories under shm_path root. */
2314 if (chan
->root_shm_path
[0]) {
2315 (void) run_as_rmdir_recursive(chan
->root_shm_path
,
2316 chan
->uid
, chan
->gid
);
2318 free(chan
->stream_fds
);
2321 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
2324 assert(stream
->ustream
);
2326 if (stream
->chan
->switch_timer_enabled
== 1) {
2327 consumer_timer_switch_stop(stream
->chan
);
2329 ustctl_destroy_stream(stream
->ustream
);
2332 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
2335 assert(stream
->ustream
);
2337 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
2340 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
2343 assert(stream
->ustream
);
2345 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
2349 * Populate index values of a UST stream. Values are set in big endian order.
2351 * Return 0 on success or else a negative value.
2353 static int get_index_values(struct ctf_packet_index
*index
,
2354 struct ustctl_consumer_stream
*ustream
)
2358 ret
= ustctl_get_timestamp_begin(ustream
, &index
->timestamp_begin
);
2360 PERROR("ustctl_get_timestamp_begin");
2363 index
->timestamp_begin
= htobe64(index
->timestamp_begin
);
2365 ret
= ustctl_get_timestamp_end(ustream
, &index
->timestamp_end
);
2367 PERROR("ustctl_get_timestamp_end");
2370 index
->timestamp_end
= htobe64(index
->timestamp_end
);
2372 ret
= ustctl_get_events_discarded(ustream
, &index
->events_discarded
);
2374 PERROR("ustctl_get_events_discarded");
2377 index
->events_discarded
= htobe64(index
->events_discarded
);
2379 ret
= ustctl_get_content_size(ustream
, &index
->content_size
);
2381 PERROR("ustctl_get_content_size");
2384 index
->content_size
= htobe64(index
->content_size
);
2386 ret
= ustctl_get_packet_size(ustream
, &index
->packet_size
);
2388 PERROR("ustctl_get_packet_size");
2391 index
->packet_size
= htobe64(index
->packet_size
);
2393 ret
= ustctl_get_stream_id(ustream
, &index
->stream_id
);
2395 PERROR("ustctl_get_stream_id");
2398 index
->stream_id
= htobe64(index
->stream_id
);
2400 ret
= ustctl_get_instance_id(ustream
, &index
->stream_instance_id
);
2402 PERROR("ustctl_get_instance_id");
2405 index
->stream_instance_id
= htobe64(index
->stream_instance_id
);
2407 ret
= ustctl_get_sequence_number(ustream
, &index
->packet_seq_num
);
2409 PERROR("ustctl_get_sequence_number");
2412 index
->packet_seq_num
= htobe64(index
->packet_seq_num
);
2419 void metadata_stream_reset_cache(struct lttng_consumer_stream
*stream
,
2420 struct consumer_metadata_cache
*cache
)
2422 DBG("Metadata stream update to version %" PRIu64
,
2424 stream
->ust_metadata_pushed
= 0;
2425 stream
->metadata_version
= cache
->version
;
2426 stream
->reset_metadata_flag
= 1;
2430 * Check if the version of the metadata stream and metadata cache match.
2431 * If the cache got updated, reset the metadata stream.
2432 * The stream lock and metadata cache lock MUST be held.
2433 * Return 0 on success, a negative value on error.
2436 int metadata_stream_check_version(struct lttng_consumer_stream
*stream
)
2439 struct consumer_metadata_cache
*cache
= stream
->chan
->metadata_cache
;
2441 if (cache
->version
== stream
->metadata_version
) {
2444 metadata_stream_reset_cache(stream
, cache
);
2451 * Write up to one packet from the metadata cache to the channel.
2453 * Returns the number of bytes pushed in the cache, or a negative value
2457 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
2462 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2463 ret
= metadata_stream_check_version(stream
);
2467 if (stream
->chan
->metadata_cache
->max_offset
2468 == stream
->ust_metadata_pushed
) {
2473 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
2474 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
2475 stream
->chan
->metadata_cache
->max_offset
2476 - stream
->ust_metadata_pushed
);
2477 assert(write_len
!= 0);
2478 if (write_len
< 0) {
2479 ERR("Writing one metadata packet");
2483 stream
->ust_metadata_pushed
+= write_len
;
2485 assert(stream
->chan
->metadata_cache
->max_offset
>=
2486 stream
->ust_metadata_pushed
);
2490 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2496 * Sync metadata meaning request them to the session daemon and snapshot to the
2497 * metadata thread can consumer them.
2499 * Metadata stream lock is held here, but we need to release it when
2500 * interacting with sessiond, else we cause a deadlock with live
2501 * awaiting on metadata to be pushed out.
2503 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2504 * is empty or a negative value on error.
2506 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data
*ctx
,
2507 struct lttng_consumer_stream
*metadata
)
2515 pthread_mutex_unlock(&metadata
->lock
);
2517 * Request metadata from the sessiond, but don't wait for the flush
2518 * because we locked the metadata thread.
2520 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata
->chan
, 0, 0);
2521 pthread_mutex_lock(&metadata
->lock
);
2526 ret
= commit_one_metadata_packet(metadata
);
2529 } else if (ret
> 0) {
2533 ustctl_flush_buffer(metadata
->ustream
, 1);
2534 ret
= ustctl_snapshot(metadata
->ustream
);
2536 if (errno
!= EAGAIN
) {
2537 ERR("Sync metadata, taking UST snapshot");
2540 DBG("No new metadata when syncing them.");
2541 /* No new metadata, exit. */
2547 * After this flush, we still need to extract metadata.
2558 * Return 0 on success else a negative value.
2560 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2561 struct lttng_consumer_local_data
*ctx
)
2564 struct ustctl_consumer_stream
*ustream
;
2569 ustream
= stream
->ustream
;
2572 * First, we are going to check if there is a new subbuffer available
2573 * before reading the stream wait_fd.
2575 /* Get the next subbuffer */
2576 ret
= ustctl_get_next_subbuf(ustream
);
2578 /* No more data found, flag the stream. */
2579 stream
->has_data
= 0;
2584 ret
= ustctl_put_subbuf(ustream
);
2587 /* This stream still has data. Flag it and wake up the data thread. */
2588 stream
->has_data
= 1;
2590 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2593 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2594 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2599 /* The wake up pipe has been notified. */
2600 ctx
->has_wakeup
= 1;
2609 int update_stream_stats(struct lttng_consumer_stream
*stream
)
2612 uint64_t seq
, discarded
;
2614 ret
= ustctl_get_sequence_number(stream
->ustream
, &seq
);
2616 PERROR("ustctl_get_sequence_number");
2620 * Start the sequence when we extract the first packet in case we don't
2621 * start at 0 (for example if a consumer is not connected to the
2622 * session immediately after the beginning).
2624 if (stream
->last_sequence_number
== -1ULL) {
2625 stream
->last_sequence_number
= seq
;
2626 } else if (seq
> stream
->last_sequence_number
) {
2627 stream
->chan
->lost_packets
+= seq
-
2628 stream
->last_sequence_number
- 1;
2630 /* seq <= last_sequence_number */
2631 ERR("Sequence number inconsistent : prev = %" PRIu64
2632 ", current = %" PRIu64
,
2633 stream
->last_sequence_number
, seq
);
2637 stream
->last_sequence_number
= seq
;
2639 ret
= ustctl_get_events_discarded(stream
->ustream
, &discarded
);
2641 PERROR("kernctl_get_events_discarded");
2644 if (discarded
< stream
->last_discarded_events
) {
2646 * Overflow has occurred. We assume only one wrap-around
2649 stream
->chan
->discarded_events
+=
2650 (1ULL << (CAA_BITS_PER_LONG
- 1)) -
2651 stream
->last_discarded_events
+ discarded
;
2653 stream
->chan
->discarded_events
+= discarded
-
2654 stream
->last_discarded_events
;
2656 stream
->last_discarded_events
= discarded
;
2664 * Read subbuffer from the given stream.
2666 * Stream lock MUST be acquired.
2668 * Return 0 on success else a negative value.
2670 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
2671 struct lttng_consumer_local_data
*ctx
, bool *rotated
)
2673 unsigned long len
, subbuf_size
, padding
;
2674 int err
, write_index
= 1, rotation_ret
;
2676 struct ustctl_consumer_stream
*ustream
;
2677 struct ctf_packet_index index
;
2680 assert(stream
->ustream
);
2683 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream
->wait_fd
,
2686 /* Ease our life for what's next. */
2687 ustream
= stream
->ustream
;
2690 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2691 * error if we cannot read this one byte (read returns 0), or if the error
2692 * is EAGAIN or EWOULDBLOCK.
2694 * This is only done when the stream is monitored by a thread, before the
2695 * flush is done after a hangup and if the stream is not flagged with data
2696 * since there might be nothing to consume in the wait fd but still have
2697 * data available flagged by the consumer wake up pipe.
2699 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2703 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2704 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2711 * If the stream was flagged to be ready for rotation before we extract the
2712 * next packet, rotate it now.
2714 if (stream
->rotate_ready
) {
2715 DBG("Rotate stream before extracting data");
2716 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
, rotated
);
2717 if (rotation_ret
< 0) {
2718 ERR("Stream rotation error");
2725 /* Get the next subbuffer */
2726 err
= ustctl_get_next_subbuf(ustream
);
2729 * Populate metadata info if the existing info has
2730 * already been read.
2732 if (stream
->metadata_flag
) {
2733 ret
= commit_one_metadata_packet(stream
);
2737 ustctl_flush_buffer(stream
->ustream
, 1);
2741 ret
= err
; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2743 * This is a debug message even for single-threaded consumer,
2744 * because poll() have more relaxed criterions than get subbuf,
2745 * so get_subbuf may fail for short race windows where poll()
2746 * would issue wakeups.
2748 DBG("Reserving sub buffer failed (everything is normal, "
2749 "it is due to concurrency) [ret: %d]", err
);
2752 assert(stream
->chan
->output
== CONSUMER_CHANNEL_MMAP
);
2754 if (!stream
->metadata_flag
) {
2755 index
.offset
= htobe64(stream
->out_fd_offset
);
2756 ret
= get_index_values(&index
, ustream
);
2758 err
= ustctl_put_subbuf(ustream
);
2763 /* Update the stream's sequence and discarded events count. */
2764 ret
= update_stream_stats(stream
);
2766 PERROR("kernctl_get_events_discarded");
2767 err
= ustctl_put_subbuf(ustream
);
2775 /* Get the full padded subbuffer size */
2776 err
= ustctl_get_padded_subbuf_size(ustream
, &len
);
2779 /* Get subbuffer data size (without padding) */
2780 err
= ustctl_get_subbuf_size(ustream
, &subbuf_size
);
2783 /* Make sure we don't get a subbuffer size bigger than the padded */
2784 assert(len
>= subbuf_size
);
2786 padding
= len
- subbuf_size
;
2788 /* write the subbuffer to the tracefile */
2789 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
, padding
, &index
);
2791 * The mmap operation should write subbuf_size amount of data when network
2792 * streaming or the full padding (len) size when we are _not_ streaming.
2794 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
2795 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
2797 * Display the error but continue processing to try to release the
2798 * subbuffer. This is a DBG statement since any unexpected kill or
2799 * signal, the application gets unregistered, relayd gets closed or
2800 * anything that affects the buffer lifetime will trigger this error.
2801 * So, for the sake of the user, don't print this error since it can
2802 * happen and it is OK with the code flow.
2804 DBG("Error writing to tracefile "
2805 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2806 ret
, len
, subbuf_size
);
2809 err
= ustctl_put_next_subbuf(ustream
);
2813 * This will consumer the byte on the wait_fd if and only if there is not
2814 * next subbuffer to be acquired.
2816 if (!stream
->metadata_flag
) {
2817 ret
= notify_if_more_data(stream
, ctx
);
2823 /* Write index if needed. */
2828 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
2830 * In live, block until all the metadata is sent.
2832 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2833 assert(!stream
->missed_metadata_flush
);
2834 stream
->waiting_on_metadata
= true;
2835 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2837 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
2839 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2840 stream
->waiting_on_metadata
= false;
2841 if (stream
->missed_metadata_flush
) {
2842 stream
->missed_metadata_flush
= false;
2843 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2844 (void) consumer_flush_ust_index(stream
);
2846 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2854 assert(!stream
->metadata_flag
);
2855 err
= consumer_stream_write_index(stream
, &index
);
2862 * After extracting the packet, we check if the stream is now ready to be
2863 * rotated and perform the action immediately.
2865 rotation_ret
= lttng_consumer_stream_is_rotate_ready(stream
);
2866 if (rotation_ret
== 1) {
2867 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
, rotated
);
2868 if (rotation_ret
< 0) {
2869 ERR("Stream rotation error");
2873 } else if (rotation_ret
< 0) {
2874 ERR("Checking if stream is ready to rotate");
2883 * Called when a stream is created.
2885 * Return 0 on success or else a negative value.
2887 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
2893 /* Don't create anything if this is set for streaming. */
2894 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
) {
2895 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
2896 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
2897 stream
->uid
, stream
->gid
, NULL
);
2901 stream
->out_fd
= ret
;
2902 stream
->tracefile_size_current
= 0;
2904 if (!stream
->metadata_flag
) {
2905 struct lttng_index_file
*index_file
;
2907 index_file
= lttng_index_file_create(stream
->chan
->pathname
,
2908 stream
->name
, stream
->uid
, stream
->gid
,
2909 stream
->chan
->tracefile_size
,
2910 stream
->tracefile_count_current
,
2911 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
2915 assert(!stream
->index_file
);
2916 stream
->index_file
= index_file
;
2926 * Check if data is still being extracted from the buffers for a specific
2927 * stream. Consumer data lock MUST be acquired before calling this function
2928 * and the stream lock.
2930 * Return 1 if the traced data are still getting read else 0 meaning that the
2931 * data is available for trace viewer reading.
2933 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
2938 assert(stream
->ustream
);
2940 DBG("UST consumer checking data pending");
2942 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
2947 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
2948 uint64_t contiguous
, pushed
;
2950 /* Ease our life a bit. */
2951 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
2952 pushed
= stream
->ust_metadata_pushed
;
2955 * We can simply check whether all contiguously available data
2956 * has been pushed to the ring buffer, since the push operation
2957 * is performed within get_next_subbuf(), and because both
2958 * get_next_subbuf() and put_next_subbuf() are issued atomically
2959 * thanks to the stream lock within
2960 * lttng_ustconsumer_read_subbuffer(). This basically means that
2961 * whetnever ust_metadata_pushed is incremented, the associated
2962 * metadata has been consumed from the metadata stream.
2964 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
2965 contiguous
, pushed
);
2966 assert(((int64_t) (contiguous
- pushed
)) >= 0);
2967 if ((contiguous
!= pushed
) ||
2968 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
2969 ret
= 1; /* Data is pending */
2973 ret
= ustctl_get_next_subbuf(stream
->ustream
);
2976 * There is still data so let's put back this
2979 ret
= ustctl_put_subbuf(stream
->ustream
);
2981 ret
= 1; /* Data is pending */
2986 /* Data is NOT pending so ready to be read. */
2994 * Stop a given metadata channel timer if enabled and close the wait fd which
2995 * is the poll pipe of the metadata stream.
2997 * This MUST be called with the metadata channel acquired.
2999 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
3004 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
3006 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
3008 if (metadata
->switch_timer_enabled
== 1) {
3009 consumer_timer_switch_stop(metadata
);
3012 if (!metadata
->metadata_stream
) {
3017 * Closing write side so the thread monitoring the stream wakes up if any
3018 * and clean the metadata stream.
3020 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
3021 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
3023 PERROR("closing metadata pipe write side");
3025 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
3033 * Close every metadata stream wait fd of the metadata hash table. This
3034 * function MUST be used very carefully so not to run into a race between the
3035 * metadata thread handling streams and this function closing their wait fd.
3037 * For UST, this is used when the session daemon hangs up. Its the metadata
3038 * producer so calling this is safe because we are assured that no state change
3039 * can occur in the metadata thread for the streams in the hash table.
3041 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
3043 struct lttng_ht_iter iter
;
3044 struct lttng_consumer_stream
*stream
;
3046 assert(metadata_ht
);
3047 assert(metadata_ht
->ht
);
3049 DBG("UST consumer closing all metadata streams");
3052 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
3055 health_code_update();
3057 pthread_mutex_lock(&stream
->chan
->lock
);
3058 lttng_ustconsumer_close_metadata(stream
->chan
);
3059 pthread_mutex_unlock(&stream
->chan
->lock
);
3065 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
3069 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
3071 ERR("Unable to close wakeup fd");
3076 * Please refer to consumer-timer.c before adding any lock within this
3077 * function or any of its callees. Timers have a very strict locking
3078 * semantic with respect to teardown. Failure to respect this semantic
3079 * introduces deadlocks.
3081 * DON'T hold the metadata lock when calling this function, else this
3082 * can cause deadlock involving consumer awaiting for metadata to be
3083 * pushed out due to concurrent interaction with the session daemon.
3085 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
3086 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
3088 struct lttcomm_metadata_request_msg request
;
3089 struct lttcomm_consumer_msg msg
;
3090 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3091 uint64_t len
, key
, offset
, version
;
3095 assert(channel
->metadata_cache
);
3097 memset(&request
, 0, sizeof(request
));
3099 /* send the metadata request to sessiond */
3100 switch (consumer_data
.type
) {
3101 case LTTNG_CONSUMER64_UST
:
3102 request
.bits_per_long
= 64;
3104 case LTTNG_CONSUMER32_UST
:
3105 request
.bits_per_long
= 32;
3108 request
.bits_per_long
= 0;
3112 request
.session_id
= channel
->session_id
;
3113 request
.session_id_per_pid
= channel
->session_id_per_pid
;
3115 * Request the application UID here so the metadata of that application can
3116 * be sent back. The channel UID corresponds to the user UID of the session
3117 * used for the rights on the stream file(s).
3119 request
.uid
= channel
->ust_app_uid
;
3120 request
.key
= channel
->key
;
3122 DBG("Sending metadata request to sessiond, session id %" PRIu64
3123 ", per-pid %" PRIu64
", app UID %u and channek key %" PRIu64
,
3124 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
3127 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
3129 health_code_update();
3131 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
3134 ERR("Asking metadata to sessiond");
3138 health_code_update();
3140 /* Receive the metadata from sessiond */
3141 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
3143 if (ret
!= sizeof(msg
)) {
3144 DBG("Consumer received unexpected message size %d (expects %zu)",
3146 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
3148 * The ret value might 0 meaning an orderly shutdown but this is ok
3149 * since the caller handles this.
3154 health_code_update();
3156 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
3157 /* No registry found */
3158 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3162 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
3163 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
3168 len
= msg
.u
.push_metadata
.len
;
3169 key
= msg
.u
.push_metadata
.key
;
3170 offset
= msg
.u
.push_metadata
.target_offset
;
3171 version
= msg
.u
.push_metadata
.version
;
3173 assert(key
== channel
->key
);
3175 DBG("No new metadata to receive for key %" PRIu64
, key
);
3178 health_code_update();
3180 /* Tell session daemon we are ready to receive the metadata. */
3181 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3182 LTTCOMM_CONSUMERD_SUCCESS
);
3183 if (ret
< 0 || len
== 0) {
3185 * Somehow, the session daemon is not responding anymore or there is
3186 * nothing to receive.
3191 health_code_update();
3193 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
3194 key
, offset
, len
, version
, channel
, timer
, wait
);
3197 * Only send the status msg if the sessiond is alive meaning a positive
3200 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
3205 health_code_update();
3207 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
3212 * Return the ustctl call for the get stream id.
3214 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
3215 uint64_t *stream_id
)
3220 return ustctl_get_stream_id(stream
->ustream
, stream_id
);