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>
36 #include <bin/lttng-consumerd/health-consumerd.h>
37 #include <common/common.h>
38 #include <common/sessiond-comm/sessiond-comm.h>
39 #include <common/relayd/relayd.h>
40 #include <common/compat/fcntl.h>
41 #include <common/compat/endian.h>
42 #include <common/consumer/consumer-metadata-cache.h>
43 #include <common/consumer/consumer-stream.h>
44 #include <common/consumer/consumer-timer.h>
45 #include <common/utils.h>
46 #include <common/index/index.h>
48 #include "ust-consumer.h"
50 #define INT_MAX_STR_LEN 12 /* includes \0 */
52 extern struct lttng_consumer_global_data consumer_data
;
53 extern int consumer_poll_timeout
;
56 * Free channel object and all streams associated with it. This MUST be used
57 * only and only if the channel has _NEVER_ been added to the global channel
60 static void destroy_channel(struct lttng_consumer_channel
*channel
)
62 struct lttng_consumer_stream
*stream
, *stmp
;
66 DBG("UST consumer cleaning stream list");
68 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
73 cds_list_del(&stream
->send_node
);
74 ustctl_destroy_stream(stream
->ustream
);
79 * If a channel is available meaning that was created before the streams
83 lttng_ustconsumer_del_channel(channel
);
84 lttng_ustconsumer_free_channel(channel
);
90 * Add channel to internal consumer state.
92 * Returns 0 on success or else a negative value.
94 static int add_channel(struct lttng_consumer_channel
*channel
,
95 struct lttng_consumer_local_data
*ctx
)
102 if (ctx
->on_recv_channel
!= NULL
) {
103 ret
= ctx
->on_recv_channel(channel
);
105 ret
= consumer_add_channel(channel
, ctx
);
106 } else if (ret
< 0) {
107 /* Most likely an ENOMEM. */
108 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
112 ret
= consumer_add_channel(channel
, ctx
);
115 DBG("UST consumer channel added (key: %" PRIu64
")", channel
->key
);
122 * Allocate and return a consumer channel object.
124 static struct lttng_consumer_channel
*allocate_channel(uint64_t session_id
,
125 const char *pathname
, const char *name
, uid_t uid
, gid_t gid
,
126 uint64_t relayd_id
, uint64_t key
, enum lttng_event_output output
,
127 uint64_t tracefile_size
, uint64_t tracefile_count
,
128 uint64_t session_id_per_pid
, unsigned int monitor
,
129 unsigned int live_timer_interval
,
130 const char *root_shm_path
, const char *shm_path
)
135 return consumer_allocate_channel(key
, session_id
, pathname
, name
, uid
,
136 gid
, relayd_id
, output
, tracefile_size
,
137 tracefile_count
, session_id_per_pid
, monitor
,
138 live_timer_interval
, root_shm_path
, shm_path
);
142 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
143 * error value if applicable is set in it else it is kept untouched.
145 * Return NULL on error else the newly allocated stream object.
147 static struct lttng_consumer_stream
*allocate_stream(int cpu
, int key
,
148 struct lttng_consumer_channel
*channel
,
149 struct lttng_consumer_local_data
*ctx
, int *_alloc_ret
)
152 struct lttng_consumer_stream
*stream
= NULL
;
157 stream
= consumer_allocate_stream(channel
->key
,
159 LTTNG_CONSUMER_ACTIVE_STREAM
,
169 if (stream
== NULL
) {
173 * We could not find the channel. Can happen if cpu hotplug
174 * happens while tearing down.
176 DBG3("Could not find channel");
181 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
187 stream
->chan
= channel
;
191 *_alloc_ret
= alloc_ret
;
197 * Send the given stream pointer to the corresponding thread.
199 * Returns 0 on success else a negative value.
201 static int send_stream_to_thread(struct lttng_consumer_stream
*stream
,
202 struct lttng_consumer_local_data
*ctx
)
205 struct lttng_pipe
*stream_pipe
;
207 /* Get the right pipe where the stream will be sent. */
208 if (stream
->metadata_flag
) {
209 ret
= consumer_add_metadata_stream(stream
);
211 ERR("Consumer add metadata stream %" PRIu64
" failed.",
215 stream_pipe
= ctx
->consumer_metadata_pipe
;
217 ret
= consumer_add_data_stream(stream
);
219 ERR("Consumer add stream %" PRIu64
" failed.",
223 stream_pipe
= ctx
->consumer_data_pipe
;
227 * From this point on, the stream's ownership has been moved away from
228 * the channel and becomes globally visible.
230 stream
->globally_visible
= 1;
232 ret
= lttng_pipe_write(stream_pipe
, &stream
, sizeof(stream
));
234 ERR("Consumer write %s stream to pipe %d",
235 stream
->metadata_flag
? "metadata" : "data",
236 lttng_pipe_get_writefd(stream_pipe
));
237 if (stream
->metadata_flag
) {
238 consumer_del_stream_for_metadata(stream
);
240 consumer_del_stream_for_data(stream
);
248 int get_stream_shm_path(char *stream_shm_path
, const char *shm_path
, int cpu
)
250 char cpu_nr
[INT_MAX_STR_LEN
]; /* int max len */
253 strncpy(stream_shm_path
, shm_path
, PATH_MAX
);
254 stream_shm_path
[PATH_MAX
- 1] = '\0';
255 ret
= snprintf(cpu_nr
, INT_MAX_STR_LEN
, "%i", cpu
);
260 strncat(stream_shm_path
, cpu_nr
,
261 PATH_MAX
- strlen(stream_shm_path
) - 1);
268 * Create streams for the given channel using liblttng-ust-ctl.
270 * Return 0 on success else a negative value.
272 static int create_ust_streams(struct lttng_consumer_channel
*channel
,
273 struct lttng_consumer_local_data
*ctx
)
276 struct ustctl_consumer_stream
*ustream
;
277 struct lttng_consumer_stream
*stream
;
283 * While a stream is available from ustctl. When NULL is returned, we've
284 * reached the end of the possible stream for the channel.
286 while ((ustream
= ustctl_create_stream(channel
->uchan
, cpu
))) {
288 int ust_metadata_pipe
[2];
290 health_code_update();
292 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& channel
->monitor
) {
293 ret
= utils_create_pipe_cloexec_nonblock(ust_metadata_pipe
);
295 ERR("Create ust metadata poll pipe");
298 wait_fd
= ust_metadata_pipe
[0];
300 wait_fd
= ustctl_stream_get_wait_fd(ustream
);
303 /* Allocate consumer stream object. */
304 stream
= allocate_stream(cpu
, wait_fd
, channel
, ctx
, &ret
);
308 stream
->ustream
= ustream
;
310 * Store it so we can save multiple function calls afterwards since
311 * this value is used heavily in the stream threads. This is UST
312 * specific so this is why it's done after allocation.
314 stream
->wait_fd
= wait_fd
;
317 * Increment channel refcount since the channel reference has now been
318 * assigned in the allocation process above.
320 if (stream
->chan
->monitor
) {
321 uatomic_inc(&stream
->chan
->refcount
);
325 * Order is important this is why a list is used. On error, the caller
326 * should clean this list.
328 cds_list_add_tail(&stream
->send_node
, &channel
->streams
.head
);
330 ret
= ustctl_get_max_subbuf_size(stream
->ustream
,
331 &stream
->max_sb_size
);
333 ERR("ustctl_get_max_subbuf_size failed for stream %s",
338 /* Do actions once stream has been received. */
339 if (ctx
->on_recv_stream
) {
340 ret
= ctx
->on_recv_stream(stream
);
346 DBG("UST consumer add stream %s (key: %" PRIu64
") with relayd id %" PRIu64
,
347 stream
->name
, stream
->key
, stream
->relayd_stream_id
);
349 /* Set next CPU stream. */
350 channel
->streams
.count
= ++cpu
;
352 /* Keep stream reference when creating metadata. */
353 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
354 channel
->metadata_stream
= stream
;
355 if (channel
->monitor
) {
356 /* Set metadata poll pipe if we created one */
357 memcpy(stream
->ust_metadata_poll_pipe
,
359 sizeof(ust_metadata_pipe
));
372 * create_posix_shm is never called concurrently within a process.
375 int create_posix_shm(void)
377 char tmp_name
[NAME_MAX
];
380 ret
= snprintf(tmp_name
, NAME_MAX
, "/ust-shm-consumer-%d", getpid());
386 * Allocate shm, and immediately unlink its shm oject, keeping
387 * only the file descriptor as a reference to the object.
388 * We specifically do _not_ use the / at the beginning of the
389 * pathname so that some OS implementations can keep it local to
390 * the process (POSIX leaves this implementation-defined).
392 shmfd
= shm_open(tmp_name
, O_CREAT
| O_EXCL
| O_RDWR
, 0700);
397 ret
= shm_unlink(tmp_name
);
398 if (ret
< 0 && errno
!= ENOENT
) {
399 PERROR("shm_unlink");
400 goto error_shm_release
;
413 static int open_ust_stream_fd(struct lttng_consumer_channel
*channel
,
414 struct ustctl_consumer_channel_attr
*attr
,
417 char shm_path
[PATH_MAX
];
420 if (!channel
->shm_path
[0]) {
421 return create_posix_shm();
423 ret
= get_stream_shm_path(shm_path
, channel
->shm_path
, cpu
);
427 return run_as_open(shm_path
,
428 O_RDWR
| O_CREAT
| O_EXCL
, S_IRUSR
| S_IWUSR
,
429 channel
->uid
, channel
->gid
);
436 * Create an UST channel with the given attributes and send it to the session
437 * daemon using the ust ctl API.
439 * Return 0 on success or else a negative value.
441 static int create_ust_channel(struct lttng_consumer_channel
*channel
,
442 struct ustctl_consumer_channel_attr
*attr
,
443 struct ustctl_consumer_channel
**ust_chanp
)
445 int ret
, nr_stream_fds
, i
, j
;
447 struct ustctl_consumer_channel
*ust_channel
;
453 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
454 "subbuf_size: %" PRIu64
", num_subbuf: %" PRIu64
", "
455 "switch_timer_interval: %u, read_timer_interval: %u, "
456 "output: %d, type: %d", attr
->overwrite
, attr
->subbuf_size
,
457 attr
->num_subbuf
, attr
->switch_timer_interval
,
458 attr
->read_timer_interval
, attr
->output
, attr
->type
);
460 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
)
463 nr_stream_fds
= ustctl_get_nr_stream_per_channel();
464 stream_fds
= zmalloc(nr_stream_fds
* sizeof(*stream_fds
));
469 for (i
= 0; i
< nr_stream_fds
; i
++) {
470 stream_fds
[i
] = open_ust_stream_fd(channel
, attr
, i
);
471 if (stream_fds
[i
] < 0) {
476 ust_channel
= ustctl_create_channel(attr
, stream_fds
, nr_stream_fds
);
481 channel
->nr_stream_fds
= nr_stream_fds
;
482 channel
->stream_fds
= stream_fds
;
483 *ust_chanp
= ust_channel
;
489 for (j
= i
- 1; j
>= 0; j
--) {
492 closeret
= close(stream_fds
[j
]);
496 if (channel
->shm_path
[0]) {
497 char shm_path
[PATH_MAX
];
499 closeret
= get_stream_shm_path(shm_path
,
500 channel
->shm_path
, j
);
502 ERR("Cannot get stream shm path");
504 closeret
= run_as_unlink(shm_path
,
505 channel
->uid
, channel
->gid
);
507 PERROR("unlink %s", shm_path
);
511 /* Try to rmdir all directories under shm_path root. */
512 if (channel
->root_shm_path
[0]) {
513 (void) run_as_rmdir_recursive(channel
->root_shm_path
,
514 channel
->uid
, channel
->gid
);
522 * Send a single given stream to the session daemon using the sock.
524 * Return 0 on success else a negative value.
526 static int send_sessiond_stream(int sock
, struct lttng_consumer_stream
*stream
)
533 DBG("UST consumer sending stream %" PRIu64
" to sessiond", stream
->key
);
535 /* Send stream to session daemon. */
536 ret
= ustctl_send_stream_to_sessiond(sock
, stream
->ustream
);
546 * Send channel to sessiond.
548 * Return 0 on success or else a negative value.
550 static int send_sessiond_channel(int sock
,
551 struct lttng_consumer_channel
*channel
,
552 struct lttng_consumer_local_data
*ctx
, int *relayd_error
)
554 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
555 struct lttng_consumer_stream
*stream
;
556 uint64_t net_seq_idx
= -1ULL;
562 DBG("UST consumer sending channel %s to sessiond", channel
->name
);
564 if (channel
->relayd_id
!= (uint64_t) -1ULL) {
565 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
567 health_code_update();
569 /* Try to send the stream to the relayd if one is available. */
570 ret
= consumer_send_relayd_stream(stream
, stream
->chan
->pathname
);
573 * Flag that the relayd was the problem here probably due to a
574 * communicaton error on the socket.
579 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
581 if (net_seq_idx
== -1ULL) {
582 net_seq_idx
= stream
->net_seq_idx
;
587 /* Inform sessiond that we are about to send channel and streams. */
588 ret
= consumer_send_status_msg(sock
, ret_code
);
589 if (ret
< 0 || ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
591 * Either the session daemon is not responding or the relayd died so we
597 /* Send channel to sessiond. */
598 ret
= ustctl_send_channel_to_sessiond(sock
, channel
->uchan
);
603 ret
= ustctl_channel_close_wakeup_fd(channel
->uchan
);
608 /* The channel was sent successfully to the sessiond at this point. */
609 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
611 health_code_update();
613 /* Send stream to session daemon. */
614 ret
= send_sessiond_stream(sock
, stream
);
620 /* Tell sessiond there is no more stream. */
621 ret
= ustctl_send_stream_to_sessiond(sock
, NULL
);
626 DBG("UST consumer NULL stream sent to sessiond");
631 if (ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
638 * Creates a channel and streams and add the channel it to the channel internal
639 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
642 * Return 0 on success or else, a negative value is returned and the channel
643 * MUST be destroyed by consumer_del_channel().
645 static int ask_channel(struct lttng_consumer_local_data
*ctx
, int sock
,
646 struct lttng_consumer_channel
*channel
,
647 struct ustctl_consumer_channel_attr
*attr
)
656 * This value is still used by the kernel consumer since for the kernel,
657 * the stream ownership is not IN the consumer so we need to have the
658 * number of left stream that needs to be initialized so we can know when
659 * to delete the channel (see consumer.c).
661 * As for the user space tracer now, the consumer creates and sends the
662 * stream to the session daemon which only sends them to the application
663 * once every stream of a channel is received making this value useless
664 * because we they will be added to the poll thread before the application
665 * receives them. This ensures that a stream can not hang up during
666 * initilization of a channel.
668 channel
->nb_init_stream_left
= 0;
670 /* The reply msg status is handled in the following call. */
671 ret
= create_ust_channel(channel
, attr
, &channel
->uchan
);
676 channel
->wait_fd
= ustctl_channel_get_wait_fd(channel
->uchan
);
679 * For the snapshots (no monitor), we create the metadata streams
680 * on demand, not during the channel creation.
682 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& !channel
->monitor
) {
687 /* Open all streams for this channel. */
688 ret
= create_ust_streams(channel
, ctx
);
698 * Send all stream of a channel to the right thread handling it.
700 * On error, return a negative value else 0 on success.
702 static int send_streams_to_thread(struct lttng_consumer_channel
*channel
,
703 struct lttng_consumer_local_data
*ctx
)
706 struct lttng_consumer_stream
*stream
, *stmp
;
711 /* Send streams to the corresponding thread. */
712 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
715 health_code_update();
717 /* Sending the stream to the thread. */
718 ret
= send_stream_to_thread(stream
, ctx
);
721 * If we are unable to send the stream to the thread, there is
722 * a big problem so just stop everything.
724 /* Remove node from the channel stream list. */
725 cds_list_del(&stream
->send_node
);
729 /* Remove node from the channel stream list. */
730 cds_list_del(&stream
->send_node
);
739 * Flush channel's streams using the given key to retrieve the channel.
741 * Return 0 on success else an LTTng error code.
743 static int flush_channel(uint64_t chan_key
)
746 struct lttng_consumer_channel
*channel
;
747 struct lttng_consumer_stream
*stream
;
749 struct lttng_ht_iter iter
;
751 DBG("UST consumer flush channel key %" PRIu64
, chan_key
);
754 channel
= consumer_find_channel(chan_key
);
756 ERR("UST consumer flush channel %" PRIu64
" not found", chan_key
);
757 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
761 ht
= consumer_data
.stream_per_chan_id_ht
;
763 /* For each stream of the channel id, flush it. */
764 cds_lfht_for_each_entry_duplicate(ht
->ht
,
765 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
766 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
768 health_code_update();
770 pthread_mutex_lock(&stream
->lock
);
771 if (!stream
->quiescent
) {
772 ustctl_flush_buffer(stream
->ustream
, 0);
773 stream
->quiescent
= true;
775 pthread_mutex_unlock(&stream
->lock
);
783 * Clear quiescent state from channel's streams using the given key to
784 * retrieve the channel.
786 * Return 0 on success else an LTTng error code.
788 static int clear_quiescent_channel(uint64_t chan_key
)
791 struct lttng_consumer_channel
*channel
;
792 struct lttng_consumer_stream
*stream
;
794 struct lttng_ht_iter iter
;
796 DBG("UST consumer clear quiescent channel key %" PRIu64
, chan_key
);
799 channel
= consumer_find_channel(chan_key
);
801 ERR("UST consumer clear quiescent channel %" PRIu64
" not found", chan_key
);
802 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
806 ht
= consumer_data
.stream_per_chan_id_ht
;
808 /* For each stream of the channel id, clear quiescent state. */
809 cds_lfht_for_each_entry_duplicate(ht
->ht
,
810 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
811 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
813 health_code_update();
815 pthread_mutex_lock(&stream
->lock
);
816 stream
->quiescent
= false;
817 pthread_mutex_unlock(&stream
->lock
);
825 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
826 * RCU read side lock MUST be acquired before calling this function.
828 * Return 0 on success else an LTTng error code.
830 static int close_metadata(uint64_t chan_key
)
833 struct lttng_consumer_channel
*channel
;
835 DBG("UST consumer close metadata key %" PRIu64
, chan_key
);
837 channel
= consumer_find_channel(chan_key
);
840 * This is possible if the metadata thread has issue a delete because
841 * the endpoint point of the stream hung up. There is no way the
842 * session daemon can know about it thus use a DBG instead of an actual
845 DBG("UST consumer close metadata %" PRIu64
" not found", chan_key
);
846 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
850 pthread_mutex_lock(&consumer_data
.lock
);
851 pthread_mutex_lock(&channel
->lock
);
853 if (cds_lfht_is_node_deleted(&channel
->node
.node
)) {
857 lttng_ustconsumer_close_metadata(channel
);
860 pthread_mutex_unlock(&channel
->lock
);
861 pthread_mutex_unlock(&consumer_data
.lock
);
867 * RCU read side lock MUST be acquired before calling this function.
869 * Return 0 on success else an LTTng error code.
871 static int setup_metadata(struct lttng_consumer_local_data
*ctx
, uint64_t key
)
874 struct lttng_consumer_channel
*metadata
;
876 DBG("UST consumer setup metadata key %" PRIu64
, key
);
878 metadata
= consumer_find_channel(key
);
880 ERR("UST consumer push metadata %" PRIu64
" not found", key
);
881 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
886 * In no monitor mode, the metadata channel has no stream(s) so skip the
887 * ownership transfer to the metadata thread.
889 if (!metadata
->monitor
) {
890 DBG("Metadata channel in no monitor");
896 * Send metadata stream to relayd if one available. Availability is
897 * known if the stream is still in the list of the channel.
899 if (cds_list_empty(&metadata
->streams
.head
)) {
900 ERR("Metadata channel key %" PRIu64
", no stream available.", key
);
901 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
902 goto error_no_stream
;
905 /* Send metadata stream to relayd if needed. */
906 if (metadata
->metadata_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
907 ret
= consumer_send_relayd_stream(metadata
->metadata_stream
,
910 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
913 ret
= consumer_send_relayd_streams_sent(
914 metadata
->metadata_stream
->net_seq_idx
);
916 ret
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
921 ret
= send_streams_to_thread(metadata
, ctx
);
924 * If we are unable to send the stream to the thread, there is
925 * a big problem so just stop everything.
927 ret
= LTTCOMM_CONSUMERD_FATAL
;
930 /* List MUST be empty after or else it could be reused. */
931 assert(cds_list_empty(&metadata
->streams
.head
));
938 * Delete metadata channel on error. At this point, the metadata stream can
939 * NOT be monitored by the metadata thread thus having the guarantee that
940 * the stream is still in the local stream list of the channel. This call
941 * will make sure to clean that list.
943 consumer_stream_destroy(metadata
->metadata_stream
, NULL
);
944 cds_list_del(&metadata
->metadata_stream
->send_node
);
945 metadata
->metadata_stream
= NULL
;
952 * Snapshot the whole metadata.
954 * Returns 0 on success, < 0 on error
956 static int snapshot_metadata(uint64_t key
, char *path
, uint64_t relayd_id
,
957 struct lttng_consumer_local_data
*ctx
)
960 struct lttng_consumer_channel
*metadata_channel
;
961 struct lttng_consumer_stream
*metadata_stream
;
966 DBG("UST consumer snapshot metadata with key %" PRIu64
" at path %s",
971 metadata_channel
= consumer_find_channel(key
);
972 if (!metadata_channel
) {
973 ERR("UST snapshot metadata channel not found for key %" PRIu64
,
978 assert(!metadata_channel
->monitor
);
980 health_code_update();
983 * Ask the sessiond if we have new metadata waiting and update the
984 * consumer metadata cache.
986 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 1);
991 health_code_update();
994 * The metadata stream is NOT created in no monitor mode when the channel
995 * is created on a sessiond ask channel command.
997 ret
= create_ust_streams(metadata_channel
, ctx
);
1002 metadata_stream
= metadata_channel
->metadata_stream
;
1003 assert(metadata_stream
);
1005 if (relayd_id
!= (uint64_t) -1ULL) {
1006 metadata_stream
->net_seq_idx
= relayd_id
;
1007 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
1012 ret
= utils_create_stream_file(path
, metadata_stream
->name
,
1013 metadata_stream
->chan
->tracefile_size
,
1014 metadata_stream
->tracefile_count_current
,
1015 metadata_stream
->uid
, metadata_stream
->gid
, NULL
);
1019 metadata_stream
->out_fd
= ret
;
1020 metadata_stream
->tracefile_size_current
= 0;
1024 health_code_update();
1026 ret
= lttng_consumer_read_subbuffer(metadata_stream
, ctx
);
1034 * Clean up the stream completly because the next snapshot will use a new
1037 consumer_stream_destroy(metadata_stream
, NULL
);
1038 cds_list_del(&metadata_stream
->send_node
);
1039 metadata_channel
->metadata_stream
= NULL
;
1047 * Take a snapshot of all the stream of a channel.
1049 * Returns 0 on success, < 0 on error
1051 static int snapshot_channel(uint64_t key
, char *path
, uint64_t relayd_id
,
1052 uint64_t nb_packets_per_stream
, struct lttng_consumer_local_data
*ctx
)
1055 unsigned use_relayd
= 0;
1056 unsigned long consumed_pos
, produced_pos
;
1057 struct lttng_consumer_channel
*channel
;
1058 struct lttng_consumer_stream
*stream
;
1065 if (relayd_id
!= (uint64_t) -1ULL) {
1069 channel
= consumer_find_channel(key
);
1071 ERR("UST snapshot channel not found for key %" PRIu64
, key
);
1075 assert(!channel
->monitor
);
1076 DBG("UST consumer snapshot channel %" PRIu64
, key
);
1078 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
1079 health_code_update();
1081 /* Lock stream because we are about to change its state. */
1082 pthread_mutex_lock(&stream
->lock
);
1083 stream
->net_seq_idx
= relayd_id
;
1086 ret
= consumer_send_relayd_stream(stream
, path
);
1091 ret
= utils_create_stream_file(path
, stream
->name
,
1092 stream
->chan
->tracefile_size
,
1093 stream
->tracefile_count_current
,
1094 stream
->uid
, stream
->gid
, NULL
);
1098 stream
->out_fd
= ret
;
1099 stream
->tracefile_size_current
= 0;
1101 DBG("UST consumer snapshot stream %s/%s (%" PRIu64
")", path
,
1102 stream
->name
, stream
->key
);
1106 * If tracing is active, we want to perform a "full" buffer flush.
1107 * Else, if quiescent, it has already been done by the prior stop.
1109 if (!stream
->quiescent
) {
1110 ustctl_flush_buffer(stream
->ustream
, 0);
1113 ret
= lttng_ustconsumer_take_snapshot(stream
);
1115 ERR("Taking UST snapshot");
1119 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
1121 ERR("Produced UST snapshot position");
1125 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1127 ERR("Consumerd UST snapshot position");
1132 * The original value is sent back if max stream size is larger than
1133 * the possible size of the snapshot. Also, we assume that the session
1134 * daemon should never send a maximum stream size that is lower than
1137 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
1138 produced_pos
, nb_packets_per_stream
,
1139 stream
->max_sb_size
);
1141 while (consumed_pos
< produced_pos
) {
1143 unsigned long len
, padded_len
;
1145 health_code_update();
1147 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1149 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1151 if (ret
!= -EAGAIN
) {
1152 PERROR("ustctl_get_subbuf snapshot");
1153 goto error_close_stream
;
1155 DBG("UST consumer get subbuf failed. Skipping it.");
1156 consumed_pos
+= stream
->max_sb_size
;
1157 stream
->chan
->lost_packets
++;
1161 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1163 ERR("Snapshot ustctl_get_subbuf_size");
1164 goto error_put_subbuf
;
1167 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1169 ERR("Snapshot ustctl_get_padded_subbuf_size");
1170 goto error_put_subbuf
;
1173 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, len
,
1174 padded_len
- len
, NULL
);
1176 if (read_len
!= len
) {
1178 goto error_put_subbuf
;
1181 if (read_len
!= padded_len
) {
1183 goto error_put_subbuf
;
1187 ret
= ustctl_put_subbuf(stream
->ustream
);
1189 ERR("Snapshot ustctl_put_subbuf");
1190 goto error_close_stream
;
1192 consumed_pos
+= stream
->max_sb_size
;
1195 /* Simply close the stream so we can use it on the next snapshot. */
1196 consumer_stream_close(stream
);
1197 pthread_mutex_unlock(&stream
->lock
);
1204 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1205 ERR("Snapshot ustctl_put_subbuf");
1208 consumer_stream_close(stream
);
1210 pthread_mutex_unlock(&stream
->lock
);
1217 * When a channel has finished the rotation of all its streams, inform the
1221 int rotate_notify_sessiond(struct lttng_consumer_local_data
*ctx
,
1227 ret
= write(ctx
->channel_rotate_pipe
, &key
, sizeof(key
));
1228 } while (ret
== -1 && errno
== EINTR
);
1230 PERROR("write to the channel rotate pipe");
1232 DBG("Sent channel rotation notification for channel key %"
1240 * Performs the stream rotation for the rotate session feature if needed.
1241 * It must be called with the stream and channel locks held.
1243 * FIXME: find a way to lock the chan without deadlock. same for kernel.
1245 * Return 0 on success, a negative number of error.
1248 int stream_rotation(struct lttng_consumer_local_data
*ctx
,
1249 struct lttng_consumer_stream
*stream
)
1252 unsigned long consumed_pos
;
1254 if (!stream
->rotate_position
&& !stream
->rotate_ready
) {
1260 * If we don't have the rotate_ready flag, check the consumed position
1261 * to determine if we need to rotate.
1263 if (!stream
->rotate_ready
) {
1264 ret
= lttng_ustconsumer_sample_snapshot_positions(stream
);
1266 ERR("Taking UST snapshot positions");
1270 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1272 ERR("Produced UST snapshot position");
1276 fprintf(stderr
, "packet %lu, pos %lu\n", stream
->key
, consumed_pos
);
1277 /* Rotate position not reached yet. */
1278 if (consumed_pos
< stream
->rotate_position
) {
1282 fprintf(stderr
, "Rotate position %lu (expected %lu) reached for stream %lu\n",
1283 consumed_pos
, stream
->rotate_position
, stream
->key
);
1285 fprintf(stderr
, "Rotate position reached for stream %lu\n",
1289 ret
= close(stream
->out_fd
);
1291 PERROR("Closing tracefile");
1295 fprintf(stderr
, "Rotating stream %lu to %s/%s\n", stream
->key
,
1296 stream
->chan
->pathname
, stream
->name
);
1297 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
1298 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
1299 stream
->uid
, stream
->gid
, NULL
);
1303 stream
->out_fd
= ret
;
1304 stream
->tracefile_size_current
= 0;
1306 if (!stream
->metadata_flag
) {
1307 struct lttng_index_file
*index_file
;
1309 lttng_index_file_put(stream
->index_file
);
1311 index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1312 stream
->name
, stream
->uid
, stream
->gid
,
1313 stream
->chan
->tracefile_size
,
1314 stream
->tracefile_count_current
,
1315 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1319 stream
->index_file
= index_file
;
1320 stream
->out_fd_offset
= 0;
1323 * Reset the position pushed from the metadata cache so it
1324 * will write from the beginning on the next push.
1326 stream
->ust_metadata_pushed
= 0;
1328 * Wakeup the metadata thread so it dumps the metadata cache
1331 consumer_metadata_wakeup_pipe(stream
->chan
);
1334 stream
->rotate_position
= 0;
1335 stream
->rotate_ready
= 0;
1337 if (--stream
->chan
->nr_stream_rotate_pending
== 0) {
1338 rotate_notify_sessiond(ctx
, stream
->chan
->key
);
1339 fprintf(stderr
, "SENT %lu\n", stream
->chan
->key
);
1352 * Sample the rotate position for all the streams of a channel.
1354 * Returns 0 on success, < 0 on error
1357 int lttng_ustconsumer_rotate_channel(uint64_t key
, char *path
,
1358 uint64_t relayd_id
, uint32_t metadata
,
1359 struct lttng_consumer_local_data
*ctx
)
1362 struct lttng_consumer_channel
*channel
;
1363 struct lttng_consumer_stream
*stream
;
1364 struct lttng_ht_iter iter
;
1365 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
1367 /* FIXME: metadata param is useless */
1369 DBG("UST consumer sample rotate position for channel %" PRIu64
, key
);
1373 channel
= consumer_find_channel(key
);
1375 ERR("No channel found for key %" PRIu64
, key
);
1379 pthread_mutex_lock(&channel
->lock
);
1380 snprintf(channel
->pathname
, PATH_MAX
, "%s", path
);
1382 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1383 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
1384 ht
->match_fct
, &channel
->key
, &iter
.iter
,
1385 stream
, node_channel_id
.node
) {
1386 uint64_t consumed_pos
;
1387 health_code_update();
1390 * Lock stream because we are about to change its state.
1392 pthread_mutex_lock(&stream
->lock
);
1393 ret
= lttng_ustconsumer_sample_snapshot_positions(stream
);
1395 ERR("Taking UST snapshot positions");
1399 ret
= lttng_ustconsumer_get_produced_snapshot(stream
,
1400 &stream
->rotate_position
);
1402 ERR("Produced UST snapshot position");
1405 fprintf(stderr
, "Stream %lu should rotate after %lu to %s\n",
1406 stream
->key
, stream
->rotate_position
,
1408 lttng_ustconsumer_get_consumed_snapshot(stream
,
1410 fprintf(stderr
, "consumed %lu\n", consumed_pos
);
1411 if (consumed_pos
== stream
->rotate_position
) {
1412 stream
->rotate_ready
= 1;
1413 fprintf(stderr
, "Stream %lu ready to rotate to %s\n",
1414 stream
->key
, channel
->pathname
);
1416 channel
->nr_stream_rotate_pending
++;
1418 ustctl_flush_buffer(stream
->ustream
, 1);
1420 pthread_mutex_unlock(&stream
->lock
);
1424 goto end_unlock_channel
;
1427 pthread_mutex_unlock(&stream
->lock
);
1429 pthread_mutex_unlock(&channel
->lock
);
1436 * Rotate all the ready streams.
1438 * This is especially important for low throughput streams that have already
1439 * been consumed, we cannot wait for their next packet to perform the
1442 * Returns 0 on success, < 0 on error
1445 int lttng_ustconsumer_rotate_ready_streams(uint64_t key
,
1446 struct lttng_consumer_local_data
*ctx
)
1449 struct lttng_consumer_channel
*channel
;
1450 struct lttng_consumer_stream
*stream
;
1451 struct lttng_ht_iter iter
;
1452 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
1456 channel
= consumer_find_channel(key
);
1458 ERR("No channel found for key %" PRIu64
, key
);
1462 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1463 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
1464 ht
->match_fct
, &channel
->key
, &iter
.iter
,
1465 stream
, node_channel_id
.node
) {
1466 health_code_update();
1469 * Lock stream because we are about to change its state.
1471 pthread_mutex_lock(&stream
->lock
);
1472 if (stream
->rotate_ready
== 0) {
1473 pthread_mutex_unlock(&stream
->lock
);
1476 ret
= stream_rotation(ctx
, stream
);
1478 pthread_mutex_unlock(&stream
->lock
);
1479 ERR("Stream rotation error");
1483 pthread_mutex_unlock(&stream
->lock
);
1494 * Receive the metadata updates from the sessiond. Supports receiving
1495 * overlapping metadata, but is needs to always belong to a contiguous
1496 * range starting from 0.
1497 * Be careful about the locks held when calling this function: it needs
1498 * the metadata cache flush to concurrently progress in order to
1501 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1502 uint64_t len
, uint64_t version
,
1503 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
1505 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1508 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1510 metadata_str
= zmalloc(len
* sizeof(char));
1511 if (!metadata_str
) {
1512 PERROR("zmalloc metadata string");
1513 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1517 health_code_update();
1519 /* Receive metadata string. */
1520 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1522 /* Session daemon is dead so return gracefully. */
1527 health_code_update();
1529 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1530 ret
= consumer_metadata_cache_write(channel
, offset
, len
, version
,
1533 /* Unable to handle metadata. Notify session daemon. */
1534 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1536 * Skip metadata flush on write error since the offset and len might
1537 * not have been updated which could create an infinite loop below when
1538 * waiting for the metadata cache to be flushed.
1540 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1543 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1548 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1549 DBG("Waiting for metadata to be flushed");
1551 health_code_update();
1553 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1563 * Receive command from session daemon and process it.
1565 * Return 1 on success else a negative value or 0.
1567 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1568 int sock
, struct pollfd
*consumer_sockpoll
)
1571 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1572 struct lttcomm_consumer_msg msg
;
1573 struct lttng_consumer_channel
*channel
= NULL
;
1575 health_code_update();
1577 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1578 if (ret
!= sizeof(msg
)) {
1579 DBG("Consumer received unexpected message size %zd (expects %zu)",
1582 * The ret value might 0 meaning an orderly shutdown but this is ok
1583 * since the caller handles this.
1586 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1592 health_code_update();
1595 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1597 health_code_update();
1599 /* relayd needs RCU read-side lock */
1602 switch (msg
.cmd_type
) {
1603 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1605 /* Session daemon status message are handled in the following call. */
1606 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1607 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
1608 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
1609 msg
.u
.relayd_sock
.relayd_session_id
);
1612 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1614 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1615 struct consumer_relayd_sock_pair
*relayd
;
1617 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1619 /* Get relayd reference if exists. */
1620 relayd
= consumer_find_relayd(index
);
1621 if (relayd
== NULL
) {
1622 DBG("Unable to find relayd %" PRIu64
, index
);
1623 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1627 * Each relayd socket pair has a refcount of stream attached to it
1628 * which tells if the relayd is still active or not depending on the
1631 * This will set the destroy flag of the relayd object and destroy it
1632 * if the refcount reaches zero when called.
1634 * The destroy can happen either here or when a stream fd hangs up.
1637 consumer_flag_relayd_for_destroy(relayd
);
1640 goto end_msg_sessiond
;
1642 case LTTNG_CONSUMER_UPDATE_STREAM
:
1647 case LTTNG_CONSUMER_DATA_PENDING
:
1649 int ret
, is_data_pending
;
1650 uint64_t id
= msg
.u
.data_pending
.session_id
;
1652 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1654 is_data_pending
= consumer_data_pending(id
);
1656 /* Send back returned value to session daemon */
1657 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1658 sizeof(is_data_pending
));
1660 DBG("Error when sending the data pending ret code: %d", ret
);
1665 * No need to send back a status message since the data pending
1666 * returned value is the response.
1670 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1673 struct ustctl_consumer_channel_attr attr
;
1675 /* Create a plain object and reserve a channel key. */
1676 channel
= allocate_channel(msg
.u
.ask_channel
.session_id
,
1677 msg
.u
.ask_channel
.pathname
, msg
.u
.ask_channel
.name
,
1678 msg
.u
.ask_channel
.uid
, msg
.u
.ask_channel
.gid
,
1679 msg
.u
.ask_channel
.relayd_id
, msg
.u
.ask_channel
.key
,
1680 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1681 msg
.u
.ask_channel
.tracefile_size
,
1682 msg
.u
.ask_channel
.tracefile_count
,
1683 msg
.u
.ask_channel
.session_id_per_pid
,
1684 msg
.u
.ask_channel
.monitor
,
1685 msg
.u
.ask_channel
.live_timer_interval
,
1686 msg
.u
.ask_channel
.root_shm_path
,
1687 msg
.u
.ask_channel
.shm_path
);
1689 goto end_channel_error
;
1693 * Assign UST application UID to the channel. This value is ignored for
1694 * per PID buffers. This is specific to UST thus setting this after the
1697 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1699 /* Build channel attributes from received message. */
1700 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1701 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1702 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1703 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1704 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1705 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1706 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1707 attr
.blocking_timeout
= msg
.u
.ask_channel
.blocking_timeout
;
1709 /* Match channel buffer type to the UST abi. */
1710 switch (msg
.u
.ask_channel
.output
) {
1711 case LTTNG_EVENT_MMAP
:
1713 attr
.output
= LTTNG_UST_MMAP
;
1717 /* Translate and save channel type. */
1718 switch (msg
.u
.ask_channel
.type
) {
1719 case LTTNG_UST_CHAN_PER_CPU
:
1720 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1721 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1723 * Set refcount to 1 for owner. Below, we will
1724 * pass ownership to the
1725 * consumer_thread_channel_poll() thread.
1727 channel
->refcount
= 1;
1729 case LTTNG_UST_CHAN_METADATA
:
1730 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1731 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1738 health_code_update();
1740 ret
= ask_channel(ctx
, sock
, channel
, &attr
);
1742 goto end_channel_error
;
1745 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1746 ret
= consumer_metadata_cache_allocate(channel
);
1748 ERR("Allocating metadata cache");
1749 goto end_channel_error
;
1751 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1752 attr
.switch_timer_interval
= 0;
1754 int monitor_start_ret
;
1756 consumer_timer_live_start(channel
,
1757 msg
.u
.ask_channel
.live_timer_interval
);
1758 monitor_start_ret
= consumer_timer_monitor_start(
1760 msg
.u
.ask_channel
.monitor_timer_interval
);
1761 if (monitor_start_ret
< 0) {
1762 ERR("Starting channel monitoring timer failed");
1763 goto end_channel_error
;
1767 health_code_update();
1770 * Add the channel to the internal state AFTER all streams were created
1771 * and successfully sent to session daemon. This way, all streams must
1772 * be ready before this channel is visible to the threads.
1773 * If add_channel succeeds, ownership of the channel is
1774 * passed to consumer_thread_channel_poll().
1776 ret
= add_channel(channel
, ctx
);
1778 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1779 if (channel
->switch_timer_enabled
== 1) {
1780 consumer_timer_switch_stop(channel
);
1782 consumer_metadata_cache_destroy(channel
);
1784 if (channel
->live_timer_enabled
== 1) {
1785 consumer_timer_live_stop(channel
);
1787 if (channel
->monitor_timer_enabled
== 1) {
1788 consumer_timer_monitor_stop(channel
);
1790 goto end_channel_error
;
1793 health_code_update();
1796 * Channel and streams are now created. Inform the session daemon that
1797 * everything went well and should wait to receive the channel and
1798 * streams with ustctl API.
1800 ret
= consumer_send_status_channel(sock
, channel
);
1803 * There is probably a problem on the socket.
1810 case LTTNG_CONSUMER_GET_CHANNEL
:
1812 int ret
, relayd_err
= 0;
1813 uint64_t key
= msg
.u
.get_channel
.key
;
1814 struct lttng_consumer_channel
*channel
;
1816 channel
= consumer_find_channel(key
);
1818 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1819 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1820 goto end_msg_sessiond
;
1823 health_code_update();
1825 /* Send everything to sessiond. */
1826 ret
= send_sessiond_channel(sock
, channel
, ctx
, &relayd_err
);
1830 * We were unable to send to the relayd the stream so avoid
1831 * sending back a fatal error to the thread since this is OK
1832 * and the consumer can continue its work. The above call
1833 * has sent the error status message to the sessiond.
1838 * The communicaton was broken hence there is a bad state between
1839 * the consumer and sessiond so stop everything.
1844 health_code_update();
1847 * In no monitor mode, the streams ownership is kept inside the channel
1848 * so don't send them to the data thread.
1850 if (!channel
->monitor
) {
1851 goto end_msg_sessiond
;
1854 ret
= send_streams_to_thread(channel
, ctx
);
1857 * If we are unable to send the stream to the thread, there is
1858 * a big problem so just stop everything.
1862 /* List MUST be empty after or else it could be reused. */
1863 assert(cds_list_empty(&channel
->streams
.head
));
1864 goto end_msg_sessiond
;
1866 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1868 uint64_t key
= msg
.u
.destroy_channel
.key
;
1871 * Only called if streams have not been sent to stream
1872 * manager thread. However, channel has been sent to
1873 * channel manager thread.
1875 notify_thread_del_channel(ctx
, key
);
1876 goto end_msg_sessiond
;
1878 case LTTNG_CONSUMER_CLOSE_METADATA
:
1882 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1887 goto end_msg_sessiond
;
1889 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1893 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1898 goto end_msg_sessiond
;
1900 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL
:
1904 ret
= clear_quiescent_channel(
1905 msg
.u
.clear_quiescent_channel
.key
);
1910 goto end_msg_sessiond
;
1912 case LTTNG_CONSUMER_PUSH_METADATA
:
1915 uint64_t len
= msg
.u
.push_metadata
.len
;
1916 uint64_t key
= msg
.u
.push_metadata
.key
;
1917 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1918 uint64_t version
= msg
.u
.push_metadata
.version
;
1919 struct lttng_consumer_channel
*channel
;
1921 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1924 channel
= consumer_find_channel(key
);
1927 * This is possible if the metadata creation on the consumer side
1928 * is in flight vis-a-vis a concurrent push metadata from the
1929 * session daemon. Simply return that the channel failed and the
1930 * session daemon will handle that message correctly considering
1931 * that this race is acceptable thus the DBG() statement here.
1933 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1934 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1935 goto end_msg_sessiond
;
1938 health_code_update();
1942 * There is nothing to receive. We have simply
1943 * checked whether the channel can be found.
1945 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1946 goto end_msg_sessiond
;
1949 /* Tell session daemon we are ready to receive the metadata. */
1950 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1952 /* Somehow, the session daemon is not responding anymore. */
1956 health_code_update();
1958 /* Wait for more data. */
1959 health_poll_entry();
1960 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1966 health_code_update();
1968 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1969 len
, version
, channel
, 0, 1);
1971 /* error receiving from sessiond */
1975 goto end_msg_sessiond
;
1978 case LTTNG_CONSUMER_SETUP_METADATA
:
1982 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1986 goto end_msg_sessiond
;
1988 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1990 if (msg
.u
.snapshot_channel
.metadata
) {
1991 ret
= snapshot_metadata(msg
.u
.snapshot_channel
.key
,
1992 msg
.u
.snapshot_channel
.pathname
,
1993 msg
.u
.snapshot_channel
.relayd_id
,
1996 ERR("Snapshot metadata failed");
1997 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
2000 ret
= snapshot_channel(msg
.u
.snapshot_channel
.key
,
2001 msg
.u
.snapshot_channel
.pathname
,
2002 msg
.u
.snapshot_channel
.relayd_id
,
2003 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
2006 ERR("Snapshot channel failed");
2007 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
2011 health_code_update();
2012 ret
= consumer_send_status_msg(sock
, ret_code
);
2014 /* Somehow, the session daemon is not responding anymore. */
2017 health_code_update();
2020 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
2023 uint64_t discarded_events
;
2024 struct lttng_ht_iter iter
;
2025 struct lttng_ht
*ht
;
2026 struct lttng_consumer_stream
*stream
;
2027 uint64_t id
= msg
.u
.discarded_events
.session_id
;
2028 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
2030 DBG("UST consumer discarded events command for session id %"
2033 pthread_mutex_lock(&consumer_data
.lock
);
2035 ht
= consumer_data
.stream_list_ht
;
2038 * We only need a reference to the channel, but they are not
2039 * directly indexed, so we just use the first matching stream
2040 * to extract the information we need, we default to 0 if not
2041 * found (no events are dropped if the channel is not yet in
2044 discarded_events
= 0;
2045 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2046 ht
->hash_fct(&id
, lttng_ht_seed
),
2048 &iter
.iter
, stream
, node_session_id
.node
) {
2049 if (stream
->chan
->key
== key
) {
2050 discarded_events
= stream
->chan
->discarded_events
;
2054 pthread_mutex_unlock(&consumer_data
.lock
);
2057 DBG("UST consumer discarded events command for session id %"
2058 PRIu64
", channel key %" PRIu64
, id
, key
);
2060 health_code_update();
2062 /* Send back returned value to session daemon */
2063 ret
= lttcomm_send_unix_sock(sock
, &discarded_events
, sizeof(discarded_events
));
2065 PERROR("send discarded events");
2071 case LTTNG_CONSUMER_LOST_PACKETS
:
2074 uint64_t lost_packets
;
2075 struct lttng_ht_iter iter
;
2076 struct lttng_ht
*ht
;
2077 struct lttng_consumer_stream
*stream
;
2078 uint64_t id
= msg
.u
.lost_packets
.session_id
;
2079 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
2081 DBG("UST consumer lost packets command for session id %"
2084 pthread_mutex_lock(&consumer_data
.lock
);
2086 ht
= consumer_data
.stream_list_ht
;
2089 * We only need a reference to the channel, but they are not
2090 * directly indexed, so we just use the first matching stream
2091 * to extract the information we need, we default to 0 if not
2092 * found (no packets lost if the channel is not yet in use).
2095 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2096 ht
->hash_fct(&id
, lttng_ht_seed
),
2098 &iter
.iter
, stream
, node_session_id
.node
) {
2099 if (stream
->chan
->key
== key
) {
2100 lost_packets
= stream
->chan
->lost_packets
;
2104 pthread_mutex_unlock(&consumer_data
.lock
);
2107 DBG("UST consumer lost packets command for session id %"
2108 PRIu64
", channel key %" PRIu64
, id
, key
);
2110 health_code_update();
2112 /* Send back returned value to session daemon */
2113 ret
= lttcomm_send_unix_sock(sock
, &lost_packets
,
2114 sizeof(lost_packets
));
2116 PERROR("send lost packets");
2122 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
2124 int channel_monitor_pipe
;
2126 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
2127 /* Successfully received the command's type. */
2128 ret
= consumer_send_status_msg(sock
, ret_code
);
2133 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
2135 if (ret
!= sizeof(channel_monitor_pipe
)) {
2136 ERR("Failed to receive channel monitor pipe");
2140 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
2141 ret
= consumer_timer_thread_set_channel_monitor_pipe(
2142 channel_monitor_pipe
);
2146 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
2147 /* Set the pipe as non-blocking. */
2148 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
2150 PERROR("fcntl get flags of the channel monitoring pipe");
2155 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
2156 flags
| O_NONBLOCK
);
2158 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
2161 DBG("Channel monitor pipe set as non-blocking");
2163 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
2165 goto end_msg_sessiond
;
2167 case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE
:
2169 int channel_rotate_pipe
;
2172 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
2173 /* Successfully received the command's type. */
2174 ret
= consumer_send_status_msg(sock
, ret_code
);
2179 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_rotate_pipe
,
2181 if (ret
!= sizeof(channel_rotate_pipe
)) {
2182 ERR("Failed to receive channel rotate pipe");
2186 DBG("Received channel rotate pipe (%d)", channel_rotate_pipe
);
2187 ctx
->channel_rotate_pipe
= channel_rotate_pipe
;
2188 /* Set the pipe as non-blocking. */
2189 ret
= fcntl(channel_rotate_pipe
, F_GETFL
, 0);
2191 PERROR("fcntl get flags of the channel rotate pipe");
2196 ret
= fcntl(channel_rotate_pipe
, F_SETFL
,
2197 flags
| O_NONBLOCK
);
2199 PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
2202 DBG("Channel rotate pipe set as non-blocking");
2203 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
2204 ret
= consumer_send_status_msg(sock
, ret_code
);
2210 case LTTNG_CONSUMER_ROTATE_CHANNEL
:
2212 ret
= lttng_ustconsumer_rotate_channel(msg
.u
.rotate_channel
.key
,
2213 msg
.u
.rotate_channel
.pathname
,
2214 msg
.u
.rotate_channel
.relayd_id
,
2215 msg
.u
.rotate_channel
.metadata
,
2218 ERR("Rotate channel failed");
2219 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2222 health_code_update();
2224 ret
= consumer_send_status_msg(sock
, ret_code
);
2226 /* Somehow, the session daemon is not responding anymore. */
2231 * Rotate the streams that are ready right now.
2232 * FIXME: this is a second consecutive iteration over the
2233 * streams in a channel, there is probably a better way to
2234 * handle this, but it needs to be after the
2235 * consumer_send_status_msg() call.
2237 ret
= lttng_ustconsumer_rotate_ready_streams(
2238 msg
.u
.rotate_channel
.key
, ctx
);
2240 ERR("Rotate channel failed");
2241 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2252 health_code_update();
2255 * Return 1 to indicate success since the 0 value can be a socket
2256 * shutdown during the recv() or send() call.
2262 * The returned value here is not useful since either way we'll return 1 to
2263 * the caller because the session daemon socket management is done
2264 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2266 ret
= consumer_send_status_msg(sock
, ret_code
);
2272 health_code_update();
2278 * Free channel here since no one has a reference to it. We don't
2279 * free after that because a stream can store this pointer.
2281 destroy_channel(channel
);
2283 /* We have to send a status channel message indicating an error. */
2284 ret
= consumer_send_status_channel(sock
, NULL
);
2286 /* Stop everything if session daemon can not be notified. */
2291 health_code_update();
2296 /* This will issue a consumer stop. */
2301 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2302 * compiled out, we isolate it in this library.
2304 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream
*stream
,
2308 assert(stream
->ustream
);
2310 return ustctl_get_mmap_read_offset(stream
->ustream
, off
);
2314 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2315 * compiled out, we isolate it in this library.
2317 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream
*stream
)
2320 assert(stream
->ustream
);
2322 return ustctl_get_mmap_base(stream
->ustream
);
2326 * Take a snapshot for a specific stream.
2328 * Returns 0 on success, < 0 on error
2330 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2333 assert(stream
->ustream
);
2335 return ustctl_snapshot(stream
->ustream
);
2339 * Sample consumed and produced positions for a specific stream.
2341 * Returns 0 on success, < 0 on error.
2343 int lttng_ustconsumer_sample_snapshot_positions(
2344 struct lttng_consumer_stream
*stream
)
2347 assert(stream
->ustream
);
2349 return ustctl_snapshot_sample_positions(stream
->ustream
);
2353 * Get the produced position
2355 * Returns 0 on success, < 0 on error
2357 int lttng_ustconsumer_get_produced_snapshot(
2358 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2361 assert(stream
->ustream
);
2364 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
2368 * Get the consumed position
2370 * Returns 0 on success, < 0 on error
2372 int lttng_ustconsumer_get_consumed_snapshot(
2373 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2376 assert(stream
->ustream
);
2379 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
2382 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
2386 assert(stream
->ustream
);
2388 ustctl_flush_buffer(stream
->ustream
, producer
);
2391 int lttng_ustconsumer_get_current_timestamp(
2392 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
2395 assert(stream
->ustream
);
2398 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
2401 int lttng_ustconsumer_get_sequence_number(
2402 struct lttng_consumer_stream
*stream
, uint64_t *seq
)
2405 assert(stream
->ustream
);
2408 return ustctl_get_sequence_number(stream
->ustream
, seq
);
2412 * Called when the stream signals the consumer that it has hung up.
2414 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
2417 assert(stream
->ustream
);
2419 pthread_mutex_lock(&stream
->lock
);
2420 if (!stream
->quiescent
) {
2421 ustctl_flush_buffer(stream
->ustream
, 0);
2422 stream
->quiescent
= true;
2424 pthread_mutex_unlock(&stream
->lock
);
2425 stream
->hangup_flush_done
= 1;
2428 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
2433 assert(chan
->uchan
);
2435 if (chan
->switch_timer_enabled
== 1) {
2436 consumer_timer_switch_stop(chan
);
2438 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
2441 ret
= close(chan
->stream_fds
[i
]);
2445 if (chan
->shm_path
[0]) {
2446 char shm_path
[PATH_MAX
];
2448 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
2450 ERR("Cannot get stream shm path");
2452 ret
= run_as_unlink(shm_path
, chan
->uid
, chan
->gid
);
2454 PERROR("unlink %s", shm_path
);
2460 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
2463 assert(chan
->uchan
);
2465 consumer_metadata_cache_destroy(chan
);
2466 ustctl_destroy_channel(chan
->uchan
);
2467 /* Try to rmdir all directories under shm_path root. */
2468 if (chan
->root_shm_path
[0]) {
2469 (void) run_as_rmdir_recursive(chan
->root_shm_path
,
2470 chan
->uid
, chan
->gid
);
2472 free(chan
->stream_fds
);
2475 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
2478 assert(stream
->ustream
);
2480 if (stream
->chan
->switch_timer_enabled
== 1) {
2481 consumer_timer_switch_stop(stream
->chan
);
2483 ustctl_destroy_stream(stream
->ustream
);
2486 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
2489 assert(stream
->ustream
);
2491 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
2494 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
2497 assert(stream
->ustream
);
2499 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
2503 * Populate index values of a UST stream. Values are set in big endian order.
2505 * Return 0 on success or else a negative value.
2507 static int get_index_values(struct ctf_packet_index
*index
,
2508 struct ustctl_consumer_stream
*ustream
)
2512 ret
= ustctl_get_timestamp_begin(ustream
, &index
->timestamp_begin
);
2514 PERROR("ustctl_get_timestamp_begin");
2517 index
->timestamp_begin
= htobe64(index
->timestamp_begin
);
2519 ret
= ustctl_get_timestamp_end(ustream
, &index
->timestamp_end
);
2521 PERROR("ustctl_get_timestamp_end");
2524 index
->timestamp_end
= htobe64(index
->timestamp_end
);
2526 ret
= ustctl_get_events_discarded(ustream
, &index
->events_discarded
);
2528 PERROR("ustctl_get_events_discarded");
2531 index
->events_discarded
= htobe64(index
->events_discarded
);
2533 ret
= ustctl_get_content_size(ustream
, &index
->content_size
);
2535 PERROR("ustctl_get_content_size");
2538 index
->content_size
= htobe64(index
->content_size
);
2540 ret
= ustctl_get_packet_size(ustream
, &index
->packet_size
);
2542 PERROR("ustctl_get_packet_size");
2545 index
->packet_size
= htobe64(index
->packet_size
);
2547 ret
= ustctl_get_stream_id(ustream
, &index
->stream_id
);
2549 PERROR("ustctl_get_stream_id");
2552 index
->stream_id
= htobe64(index
->stream_id
);
2554 ret
= ustctl_get_instance_id(ustream
, &index
->stream_instance_id
);
2556 PERROR("ustctl_get_instance_id");
2559 index
->stream_instance_id
= htobe64(index
->stream_instance_id
);
2561 ret
= ustctl_get_sequence_number(ustream
, &index
->packet_seq_num
);
2563 PERROR("ustctl_get_sequence_number");
2566 index
->packet_seq_num
= htobe64(index
->packet_seq_num
);
2573 void metadata_stream_reset_cache(struct lttng_consumer_stream
*stream
,
2574 struct consumer_metadata_cache
*cache
)
2576 DBG("Metadata stream update to version %" PRIu64
,
2578 stream
->ust_metadata_pushed
= 0;
2579 stream
->metadata_version
= cache
->version
;
2580 stream
->reset_metadata_flag
= 1;
2584 * Check if the version of the metadata stream and metadata cache match.
2585 * If the cache got updated, reset the metadata stream.
2586 * The stream lock and metadata cache lock MUST be held.
2587 * Return 0 on success, a negative value on error.
2590 int metadata_stream_check_version(struct lttng_consumer_stream
*stream
)
2593 struct consumer_metadata_cache
*cache
= stream
->chan
->metadata_cache
;
2595 if (cache
->version
== stream
->metadata_version
) {
2598 metadata_stream_reset_cache(stream
, cache
);
2605 * Write up to one packet from the metadata cache to the channel.
2607 * Returns the number of bytes pushed in the cache, or a negative value
2611 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
2616 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2617 ret
= metadata_stream_check_version(stream
);
2621 if (stream
->chan
->metadata_cache
->max_offset
2622 == stream
->ust_metadata_pushed
) {
2627 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
2628 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
2629 stream
->chan
->metadata_cache
->max_offset
2630 - stream
->ust_metadata_pushed
);
2631 assert(write_len
!= 0);
2632 if (write_len
< 0) {
2633 ERR("Writing one metadata packet");
2637 stream
->ust_metadata_pushed
+= write_len
;
2639 assert(stream
->chan
->metadata_cache
->max_offset
>=
2640 stream
->ust_metadata_pushed
);
2644 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2650 * Sync metadata meaning request them to the session daemon and snapshot to the
2651 * metadata thread can consumer them.
2653 * Metadata stream lock is held here, but we need to release it when
2654 * interacting with sessiond, else we cause a deadlock with live
2655 * awaiting on metadata to be pushed out.
2657 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2658 * is empty or a negative value on error.
2660 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data
*ctx
,
2661 struct lttng_consumer_stream
*metadata
)
2669 pthread_mutex_unlock(&metadata
->lock
);
2671 * Request metadata from the sessiond, but don't wait for the flush
2672 * because we locked the metadata thread.
2674 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata
->chan
, 0, 0);
2675 pthread_mutex_lock(&metadata
->lock
);
2680 ret
= commit_one_metadata_packet(metadata
);
2683 } else if (ret
> 0) {
2687 ustctl_flush_buffer(metadata
->ustream
, 1);
2688 ret
= ustctl_snapshot(metadata
->ustream
);
2690 if (errno
!= EAGAIN
) {
2691 ERR("Sync metadata, taking UST snapshot");
2694 DBG("No new metadata when syncing them.");
2695 /* No new metadata, exit. */
2701 * After this flush, we still need to extract metadata.
2712 * Return 0 on success else a negative value.
2714 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2715 struct lttng_consumer_local_data
*ctx
)
2718 struct ustctl_consumer_stream
*ustream
;
2723 ustream
= stream
->ustream
;
2726 * First, we are going to check if there is a new subbuffer available
2727 * before reading the stream wait_fd.
2729 /* Get the next subbuffer */
2730 ret
= ustctl_get_next_subbuf(ustream
);
2732 /* No more data found, flag the stream. */
2733 stream
->has_data
= 0;
2738 ret
= ustctl_put_subbuf(ustream
);
2741 /* This stream still has data. Flag it and wake up the data thread. */
2742 stream
->has_data
= 1;
2744 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2747 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2748 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2753 /* The wake up pipe has been notified. */
2754 ctx
->has_wakeup
= 1;
2763 int update_stream_stats(struct lttng_consumer_stream
*stream
)
2766 uint64_t seq
, discarded
;
2768 ret
= ustctl_get_sequence_number(stream
->ustream
, &seq
);
2770 PERROR("ustctl_get_sequence_number");
2774 * Start the sequence when we extract the first packet in case we don't
2775 * start at 0 (for example if a consumer is not connected to the
2776 * session immediately after the beginning).
2778 if (stream
->last_sequence_number
== -1ULL) {
2779 stream
->last_sequence_number
= seq
;
2780 } else if (seq
> stream
->last_sequence_number
) {
2781 stream
->chan
->lost_packets
+= seq
-
2782 stream
->last_sequence_number
- 1;
2784 /* seq <= last_sequence_number */
2785 ERR("Sequence number inconsistent : prev = %" PRIu64
2786 ", current = %" PRIu64
,
2787 stream
->last_sequence_number
, seq
);
2791 stream
->last_sequence_number
= seq
;
2793 ret
= ustctl_get_events_discarded(stream
->ustream
, &discarded
);
2795 PERROR("kernctl_get_events_discarded");
2798 if (discarded
< stream
->last_discarded_events
) {
2800 * Overflow has occurred. We assume only one wrap-around
2803 stream
->chan
->discarded_events
+=
2804 (1ULL << (CAA_BITS_PER_LONG
- 1)) -
2805 stream
->last_discarded_events
+ discarded
;
2807 stream
->chan
->discarded_events
+= discarded
-
2808 stream
->last_discarded_events
;
2810 stream
->last_discarded_events
= discarded
;
2818 * Read subbuffer from the given stream.
2820 * Stream lock MUST be acquired.
2822 * Return 0 on success else a negative value.
2824 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
2825 struct lttng_consumer_local_data
*ctx
)
2827 unsigned long len
, subbuf_size
, padding
;
2828 int err
, write_index
= 1, rotation_ret
;
2830 struct ustctl_consumer_stream
*ustream
;
2831 struct ctf_packet_index index
;
2834 assert(stream
->ustream
);
2837 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream
->wait_fd
,
2840 /* Ease our life for what's next. */
2841 ustream
= stream
->ustream
;
2844 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2845 * error if we cannot read this one byte (read returns 0), or if the error
2846 * is EAGAIN or EWOULDBLOCK.
2848 * This is only done when the stream is monitored by a thread, before the
2849 * flush is done after a hangup and if the stream is not flagged with data
2850 * since there might be nothing to consume in the wait fd but still have
2851 * data available flagged by the consumer wake up pipe.
2853 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2857 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2858 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2865 /* Get the next subbuffer */
2866 err
= ustctl_get_next_subbuf(ustream
);
2869 * Populate metadata info if the existing info has
2870 * already been read.
2872 if (stream
->metadata_flag
) {
2873 ret
= commit_one_metadata_packet(stream
);
2877 ustctl_flush_buffer(stream
->ustream
, 1);
2881 ret
= err
; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2883 * This is a debug message even for single-threaded consumer,
2884 * because poll() have more relaxed criterions than get subbuf,
2885 * so get_subbuf may fail for short race windows where poll()
2886 * would issue wakeups.
2888 DBG("Reserving sub buffer failed (everything is normal, "
2889 "it is due to concurrency) [ret: %d]", err
);
2892 assert(stream
->chan
->output
== CONSUMER_CHANNEL_MMAP
);
2894 if (!stream
->metadata_flag
) {
2895 index
.offset
= htobe64(stream
->out_fd_offset
);
2896 ret
= get_index_values(&index
, ustream
);
2898 err
= ustctl_put_subbuf(ustream
);
2903 /* Update the stream's sequence and discarded events count. */
2904 ret
= update_stream_stats(stream
);
2906 PERROR("kernctl_get_events_discarded");
2907 err
= ustctl_put_subbuf(ustream
);
2915 /* Get the full padded subbuffer size */
2916 err
= ustctl_get_padded_subbuf_size(ustream
, &len
);
2919 /* Get subbuffer data size (without padding) */
2920 err
= ustctl_get_subbuf_size(ustream
, &subbuf_size
);
2923 /* Make sure we don't get a subbuffer size bigger than the padded */
2924 assert(len
>= subbuf_size
);
2926 padding
= len
- subbuf_size
;
2927 /* write the subbuffer to the tracefile */
2928 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
, padding
, &index
);
2930 * The mmap operation should write subbuf_size amount of data when network
2931 * streaming or the full padding (len) size when we are _not_ streaming.
2933 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
2934 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
2936 * Display the error but continue processing to try to release the
2937 * subbuffer. This is a DBG statement since any unexpected kill or
2938 * signal, the application gets unregistered, relayd gets closed or
2939 * anything that affects the buffer lifetime will trigger this error.
2940 * So, for the sake of the user, don't print this error since it can
2941 * happen and it is OK with the code flow.
2943 DBG("Error writing to tracefile "
2944 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2945 ret
, len
, subbuf_size
);
2948 err
= ustctl_put_next_subbuf(ustream
);
2952 * This will consumer the byte on the wait_fd if and only if there is not
2953 * next subbuffer to be acquired.
2955 if (!stream
->metadata_flag
) {
2956 ret
= notify_if_more_data(stream
, ctx
);
2962 /* Write index if needed. */
2967 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
2969 * In live, block until all the metadata is sent.
2971 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2972 assert(!stream
->missed_metadata_flush
);
2973 stream
->waiting_on_metadata
= true;
2974 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2976 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
2978 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2979 stream
->waiting_on_metadata
= false;
2980 if (stream
->missed_metadata_flush
) {
2981 stream
->missed_metadata_flush
= false;
2982 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2983 (void) consumer_flush_ust_index(stream
);
2985 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2993 assert(!stream
->metadata_flag
);
2994 err
= consumer_stream_write_index(stream
, &index
);
3000 /* FIXME: do we need this lock, it causes deadlocks when called
3001 * at the same time with lttng_ustconsumer_rotate_channel ? */
3002 // pthread_mutex_lock(&stream->chan->lock);
3003 rotation_ret
= stream_rotation(ctx
, stream
);
3004 if (rotation_ret
< 0) {
3005 // pthread_mutex_unlock(&stream->chan->lock);
3007 ERR("Stream rotation error");
3010 // pthread_mutex_unlock(&stream->chan->lock);
3015 * Called when a stream is created.
3017 * Return 0 on success or else a negative value.
3019 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3025 /* Don't create anything if this is set for streaming. */
3026 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
) {
3027 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
3028 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
3029 stream
->uid
, stream
->gid
, NULL
);
3033 stream
->out_fd
= ret
;
3034 stream
->tracefile_size_current
= 0;
3036 if (!stream
->metadata_flag
) {
3037 struct lttng_index_file
*index_file
;
3039 index_file
= lttng_index_file_create(stream
->chan
->pathname
,
3040 stream
->name
, stream
->uid
, stream
->gid
,
3041 stream
->chan
->tracefile_size
,
3042 stream
->tracefile_count_current
,
3043 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
3047 assert(!stream
->index_file
);
3048 stream
->index_file
= index_file
;
3058 * Check if data is still being extracted from the buffers for a specific
3059 * stream. Consumer data lock MUST be acquired before calling this function
3060 * and the stream lock.
3062 * Return 1 if the traced data are still getting read else 0 meaning that the
3063 * data is available for trace viewer reading.
3065 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
3070 assert(stream
->ustream
);
3072 DBG("UST consumer checking data pending");
3074 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
3079 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
3080 uint64_t contiguous
, pushed
;
3082 /* Ease our life a bit. */
3083 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
3084 pushed
= stream
->ust_metadata_pushed
;
3087 * We can simply check whether all contiguously available data
3088 * has been pushed to the ring buffer, since the push operation
3089 * is performed within get_next_subbuf(), and because both
3090 * get_next_subbuf() and put_next_subbuf() are issued atomically
3091 * thanks to the stream lock within
3092 * lttng_ustconsumer_read_subbuffer(). This basically means that
3093 * whetnever ust_metadata_pushed is incremented, the associated
3094 * metadata has been consumed from the metadata stream.
3096 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
3097 contiguous
, pushed
);
3098 assert(((int64_t) (contiguous
- pushed
)) >= 0);
3099 if ((contiguous
!= pushed
) ||
3100 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
3101 ret
= 1; /* Data is pending */
3105 ret
= ustctl_get_next_subbuf(stream
->ustream
);
3108 * There is still data so let's put back this
3111 ret
= ustctl_put_subbuf(stream
->ustream
);
3113 ret
= 1; /* Data is pending */
3118 /* Data is NOT pending so ready to be read. */
3126 * Stop a given metadata channel timer if enabled and close the wait fd which
3127 * is the poll pipe of the metadata stream.
3129 * This MUST be called with the metadata channel acquired.
3131 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
3136 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
3138 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
3140 if (metadata
->switch_timer_enabled
== 1) {
3141 consumer_timer_switch_stop(metadata
);
3144 if (!metadata
->metadata_stream
) {
3149 * Closing write side so the thread monitoring the stream wakes up if any
3150 * and clean the metadata stream.
3152 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
3153 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
3155 PERROR("closing metadata pipe write side");
3157 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
3165 * Close every metadata stream wait fd of the metadata hash table. This
3166 * function MUST be used very carefully so not to run into a race between the
3167 * metadata thread handling streams and this function closing their wait fd.
3169 * For UST, this is used when the session daemon hangs up. Its the metadata
3170 * producer so calling this is safe because we are assured that no state change
3171 * can occur in the metadata thread for the streams in the hash table.
3173 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
3175 struct lttng_ht_iter iter
;
3176 struct lttng_consumer_stream
*stream
;
3178 assert(metadata_ht
);
3179 assert(metadata_ht
->ht
);
3181 DBG("UST consumer closing all metadata streams");
3184 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
3187 health_code_update();
3189 pthread_mutex_lock(&stream
->chan
->lock
);
3190 lttng_ustconsumer_close_metadata(stream
->chan
);
3191 pthread_mutex_unlock(&stream
->chan
->lock
);
3197 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
3201 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
3203 ERR("Unable to close wakeup fd");
3208 * Please refer to consumer-timer.c before adding any lock within this
3209 * function or any of its callees. Timers have a very strict locking
3210 * semantic with respect to teardown. Failure to respect this semantic
3211 * introduces deadlocks.
3213 * DON'T hold the metadata lock when calling this function, else this
3214 * can cause deadlock involving consumer awaiting for metadata to be
3215 * pushed out due to concurrent interaction with the session daemon.
3217 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
3218 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
3220 struct lttcomm_metadata_request_msg request
;
3221 struct lttcomm_consumer_msg msg
;
3222 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3223 uint64_t len
, key
, offset
, version
;
3227 assert(channel
->metadata_cache
);
3229 memset(&request
, 0, sizeof(request
));
3231 /* send the metadata request to sessiond */
3232 switch (consumer_data
.type
) {
3233 case LTTNG_CONSUMER64_UST
:
3234 request
.bits_per_long
= 64;
3236 case LTTNG_CONSUMER32_UST
:
3237 request
.bits_per_long
= 32;
3240 request
.bits_per_long
= 0;
3244 request
.session_id
= channel
->session_id
;
3245 request
.session_id_per_pid
= channel
->session_id_per_pid
;
3247 * Request the application UID here so the metadata of that application can
3248 * be sent back. The channel UID corresponds to the user UID of the session
3249 * used for the rights on the stream file(s).
3251 request
.uid
= channel
->ust_app_uid
;
3252 request
.key
= channel
->key
;
3254 DBG("Sending metadata request to sessiond, session id %" PRIu64
3255 ", per-pid %" PRIu64
", app UID %u and channek key %" PRIu64
,
3256 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
3259 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
3261 health_code_update();
3263 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
3266 ERR("Asking metadata to sessiond");
3270 health_code_update();
3272 /* Receive the metadata from sessiond */
3273 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
3275 if (ret
!= sizeof(msg
)) {
3276 DBG("Consumer received unexpected message size %d (expects %zu)",
3278 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
3280 * The ret value might 0 meaning an orderly shutdown but this is ok
3281 * since the caller handles this.
3286 health_code_update();
3288 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
3289 /* No registry found */
3290 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3294 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
3295 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
3300 len
= msg
.u
.push_metadata
.len
;
3301 key
= msg
.u
.push_metadata
.key
;
3302 offset
= msg
.u
.push_metadata
.target_offset
;
3303 version
= msg
.u
.push_metadata
.version
;
3305 assert(key
== channel
->key
);
3307 DBG("No new metadata to receive for key %" PRIu64
, key
);
3310 health_code_update();
3312 /* Tell session daemon we are ready to receive the metadata. */
3313 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3314 LTTCOMM_CONSUMERD_SUCCESS
);
3315 if (ret
< 0 || len
== 0) {
3317 * Somehow, the session daemon is not responding anymore or there is
3318 * nothing to receive.
3323 health_code_update();
3325 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
3326 key
, offset
, len
, version
, channel
, timer
, wait
);
3329 * Only send the status msg if the sessiond is alive meaning a positive
3332 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
3337 health_code_update();
3339 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
3344 * Return the ustctl call for the get stream id.
3346 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
3347 uint64_t *stream_id
)
3352 return ustctl_get_stream_id(stream
->ustream
, stream_id
);