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
);
186 consumer_stream_copy_ro_channel_values(stream
, channel
);
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
,
574 * Flag that the relayd was the problem here probably due to a
575 * communicaton error on the socket.
580 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
582 if (net_seq_idx
== -1ULL) {
583 net_seq_idx
= stream
->net_seq_idx
;
588 /* Inform sessiond that we are about to send channel and streams. */
589 ret
= consumer_send_status_msg(sock
, ret_code
);
590 if (ret
< 0 || ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
592 * Either the session daemon is not responding or the relayd died so we
598 /* Send channel to sessiond. */
599 ret
= ustctl_send_channel_to_sessiond(sock
, channel
->uchan
);
604 ret
= ustctl_channel_close_wakeup_fd(channel
->uchan
);
609 /* The channel was sent successfully to the sessiond at this point. */
610 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
612 health_code_update();
614 /* Send stream to session daemon. */
615 ret
= send_sessiond_stream(sock
, stream
);
621 /* Tell sessiond there is no more stream. */
622 ret
= ustctl_send_stream_to_sessiond(sock
, NULL
);
627 DBG("UST consumer NULL stream sent to sessiond");
632 if (ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
639 * Creates a channel and streams and add the channel it to the channel internal
640 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
643 * Return 0 on success or else, a negative value is returned and the channel
644 * MUST be destroyed by consumer_del_channel().
646 static int ask_channel(struct lttng_consumer_local_data
*ctx
, int sock
,
647 struct lttng_consumer_channel
*channel
,
648 struct ustctl_consumer_channel_attr
*attr
)
657 * This value is still used by the kernel consumer since for the kernel,
658 * the stream ownership is not IN the consumer so we need to have the
659 * number of left stream that needs to be initialized so we can know when
660 * to delete the channel (see consumer.c).
662 * As for the user space tracer now, the consumer creates and sends the
663 * stream to the session daemon which only sends them to the application
664 * once every stream of a channel is received making this value useless
665 * because we they will be added to the poll thread before the application
666 * receives them. This ensures that a stream can not hang up during
667 * initilization of a channel.
669 channel
->nb_init_stream_left
= 0;
671 /* The reply msg status is handled in the following call. */
672 ret
= create_ust_channel(channel
, attr
, &channel
->uchan
);
677 channel
->wait_fd
= ustctl_channel_get_wait_fd(channel
->uchan
);
680 * For the snapshots (no monitor), we create the metadata streams
681 * on demand, not during the channel creation.
683 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& !channel
->monitor
) {
688 /* Open all streams for this channel. */
689 ret
= create_ust_streams(channel
, ctx
);
699 * Send all stream of a channel to the right thread handling it.
701 * On error, return a negative value else 0 on success.
703 static int send_streams_to_thread(struct lttng_consumer_channel
*channel
,
704 struct lttng_consumer_local_data
*ctx
)
707 struct lttng_consumer_stream
*stream
, *stmp
;
712 /* Send streams to the corresponding thread. */
713 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
716 health_code_update();
718 /* Sending the stream to the thread. */
719 ret
= send_stream_to_thread(stream
, ctx
);
722 * If we are unable to send the stream to the thread, there is
723 * a big problem so just stop everything.
725 /* Remove node from the channel stream list. */
726 cds_list_del(&stream
->send_node
);
730 /* Remove node from the channel stream list. */
731 cds_list_del(&stream
->send_node
);
740 * Flush channel's streams using the given key to retrieve the channel.
742 * Return 0 on success else an LTTng error code.
744 static int flush_channel(uint64_t chan_key
)
747 struct lttng_consumer_channel
*channel
;
748 struct lttng_consumer_stream
*stream
;
750 struct lttng_ht_iter iter
;
752 DBG("UST consumer flush channel key %" PRIu64
, chan_key
);
755 channel
= consumer_find_channel(chan_key
);
757 ERR("UST consumer flush channel %" PRIu64
" not found", chan_key
);
758 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
762 ht
= consumer_data
.stream_per_chan_id_ht
;
764 /* For each stream of the channel id, flush it. */
765 cds_lfht_for_each_entry_duplicate(ht
->ht
,
766 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
767 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
769 health_code_update();
771 pthread_mutex_lock(&stream
->lock
);
772 if (!stream
->quiescent
) {
773 ustctl_flush_buffer(stream
->ustream
, 0);
774 stream
->quiescent
= true;
776 pthread_mutex_unlock(&stream
->lock
);
784 * Clear quiescent state from channel's streams using the given key to
785 * retrieve the channel.
787 * Return 0 on success else an LTTng error code.
789 static int clear_quiescent_channel(uint64_t chan_key
)
792 struct lttng_consumer_channel
*channel
;
793 struct lttng_consumer_stream
*stream
;
795 struct lttng_ht_iter iter
;
797 DBG("UST consumer clear quiescent channel key %" PRIu64
, chan_key
);
800 channel
= consumer_find_channel(chan_key
);
802 ERR("UST consumer clear quiescent channel %" PRIu64
" not found", chan_key
);
803 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
807 ht
= consumer_data
.stream_per_chan_id_ht
;
809 /* For each stream of the channel id, clear quiescent state. */
810 cds_lfht_for_each_entry_duplicate(ht
->ht
,
811 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
812 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
814 health_code_update();
816 pthread_mutex_lock(&stream
->lock
);
817 stream
->quiescent
= false;
818 pthread_mutex_unlock(&stream
->lock
);
826 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
827 * RCU read side lock MUST be acquired before calling this function.
829 * Return 0 on success else an LTTng error code.
831 static int close_metadata(uint64_t chan_key
)
834 struct lttng_consumer_channel
*channel
;
836 DBG("UST consumer close metadata key %" PRIu64
, chan_key
);
838 channel
= consumer_find_channel(chan_key
);
841 * This is possible if the metadata thread has issue a delete because
842 * the endpoint point of the stream hung up. There is no way the
843 * session daemon can know about it thus use a DBG instead of an actual
846 DBG("UST consumer close metadata %" PRIu64
" not found", chan_key
);
847 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
851 pthread_mutex_lock(&consumer_data
.lock
);
852 pthread_mutex_lock(&channel
->lock
);
854 if (cds_lfht_is_node_deleted(&channel
->node
.node
)) {
858 lttng_ustconsumer_close_metadata(channel
);
861 pthread_mutex_unlock(&channel
->lock
);
862 pthread_mutex_unlock(&consumer_data
.lock
);
868 * RCU read side lock MUST be acquired before calling this function.
870 * Return 0 on success else an LTTng error code.
872 static int setup_metadata(struct lttng_consumer_local_data
*ctx
, uint64_t key
)
875 struct lttng_consumer_channel
*metadata
;
877 DBG("UST consumer setup metadata key %" PRIu64
, key
);
879 metadata
= consumer_find_channel(key
);
881 ERR("UST consumer push metadata %" PRIu64
" not found", key
);
882 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
887 * In no monitor mode, the metadata channel has no stream(s) so skip the
888 * ownership transfer to the metadata thread.
890 if (!metadata
->monitor
) {
891 DBG("Metadata channel in no monitor");
897 * Send metadata stream to relayd if one available. Availability is
898 * known if the stream is still in the list of the channel.
900 if (cds_list_empty(&metadata
->streams
.head
)) {
901 ERR("Metadata channel key %" PRIu64
", no stream available.", key
);
902 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
903 goto error_no_stream
;
906 /* Send metadata stream to relayd if needed. */
907 if (metadata
->metadata_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
908 ret
= consumer_send_relayd_stream(metadata
->metadata_stream
,
909 metadata
->pathname
, LTTNG_DOMAIN_UST
);
911 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
914 ret
= consumer_send_relayd_streams_sent(
915 metadata
->metadata_stream
->net_seq_idx
);
917 ret
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
922 ret
= send_streams_to_thread(metadata
, ctx
);
925 * If we are unable to send the stream to the thread, there is
926 * a big problem so just stop everything.
928 ret
= LTTCOMM_CONSUMERD_FATAL
;
931 /* List MUST be empty after or else it could be reused. */
932 assert(cds_list_empty(&metadata
->streams
.head
));
939 * Delete metadata channel on error. At this point, the metadata stream can
940 * NOT be monitored by the metadata thread thus having the guarantee that
941 * the stream is still in the local stream list of the channel. This call
942 * will make sure to clean that list.
944 consumer_stream_destroy(metadata
->metadata_stream
, NULL
);
945 cds_list_del(&metadata
->metadata_stream
->send_node
);
946 metadata
->metadata_stream
= NULL
;
953 * Snapshot the whole metadata.
955 * Returns 0 on success, < 0 on error
957 static int snapshot_metadata(uint64_t key
, char *path
, uint64_t relayd_id
,
958 struct lttng_consumer_local_data
*ctx
)
961 struct lttng_consumer_channel
*metadata_channel
;
962 struct lttng_consumer_stream
*metadata_stream
;
967 DBG("UST consumer snapshot metadata with key %" PRIu64
" at path %s",
972 metadata_channel
= consumer_find_channel(key
);
973 if (!metadata_channel
) {
974 ERR("UST snapshot metadata channel not found for key %" PRIu64
,
979 assert(!metadata_channel
->monitor
);
981 health_code_update();
984 * Ask the sessiond if we have new metadata waiting and update the
985 * consumer metadata cache.
987 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 1);
992 health_code_update();
995 * The metadata stream is NOT created in no monitor mode when the channel
996 * is created on a sessiond ask channel command.
998 ret
= create_ust_streams(metadata_channel
, ctx
);
1003 metadata_stream
= metadata_channel
->metadata_stream
;
1004 assert(metadata_stream
);
1006 if (relayd_id
!= (uint64_t) -1ULL) {
1007 metadata_stream
->net_seq_idx
= relayd_id
;
1008 ret
= consumer_send_relayd_stream(metadata_stream
, path
,
1014 ret
= utils_create_stream_file(path
, metadata_stream
->name
,
1015 metadata_stream
->chan
->tracefile_size
,
1016 metadata_stream
->tracefile_count_current
,
1017 metadata_stream
->uid
, metadata_stream
->gid
, NULL
);
1021 metadata_stream
->out_fd
= ret
;
1022 metadata_stream
->tracefile_size_current
= 0;
1026 health_code_update();
1028 ret
= lttng_consumer_read_subbuffer(metadata_stream
, ctx
);
1036 * Clean up the stream completly because the next snapshot will use a new
1039 consumer_stream_destroy(metadata_stream
, NULL
);
1040 cds_list_del(&metadata_stream
->send_node
);
1041 metadata_channel
->metadata_stream
= NULL
;
1049 * Take a snapshot of all the stream of a channel.
1051 * Returns 0 on success, < 0 on error
1053 static int snapshot_channel(uint64_t key
, char *path
, uint64_t relayd_id
,
1054 uint64_t nb_packets_per_stream
, struct lttng_consumer_local_data
*ctx
)
1057 unsigned use_relayd
= 0;
1058 unsigned long consumed_pos
, produced_pos
;
1059 struct lttng_consumer_channel
*channel
;
1060 struct lttng_consumer_stream
*stream
;
1067 if (relayd_id
!= (uint64_t) -1ULL) {
1071 channel
= consumer_find_channel(key
);
1073 ERR("UST snapshot channel not found for key %" PRIu64
, key
);
1077 assert(!channel
->monitor
);
1078 DBG("UST consumer snapshot channel %" PRIu64
, key
);
1080 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
1081 health_code_update();
1083 /* Lock stream because we are about to change its state. */
1084 pthread_mutex_lock(&stream
->lock
);
1085 stream
->net_seq_idx
= relayd_id
;
1088 ret
= consumer_send_relayd_stream(stream
, path
,
1094 ret
= utils_create_stream_file(path
, stream
->name
,
1095 stream
->chan
->tracefile_size
,
1096 stream
->tracefile_count_current
,
1097 stream
->uid
, stream
->gid
, NULL
);
1101 stream
->out_fd
= ret
;
1102 stream
->tracefile_size_current
= 0;
1104 DBG("UST consumer snapshot stream %s/%s (%" PRIu64
")", path
,
1105 stream
->name
, stream
->key
);
1109 * If tracing is active, we want to perform a "full" buffer flush.
1110 * Else, if quiescent, it has already been done by the prior stop.
1112 if (!stream
->quiescent
) {
1113 ustctl_flush_buffer(stream
->ustream
, 0);
1116 ret
= lttng_ustconsumer_take_snapshot(stream
);
1118 ERR("Taking UST snapshot");
1122 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
1124 ERR("Produced UST snapshot position");
1128 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1130 ERR("Consumerd UST snapshot position");
1135 * The original value is sent back if max stream size is larger than
1136 * the possible size of the snapshot. Also, we assume that the session
1137 * daemon should never send a maximum stream size that is lower than
1140 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
1141 produced_pos
, nb_packets_per_stream
,
1142 stream
->max_sb_size
);
1144 while (consumed_pos
< produced_pos
) {
1146 unsigned long len
, padded_len
;
1148 health_code_update();
1150 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1152 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1154 if (ret
!= -EAGAIN
) {
1155 PERROR("ustctl_get_subbuf snapshot");
1156 goto error_close_stream
;
1158 DBG("UST consumer get subbuf failed. Skipping it.");
1159 consumed_pos
+= stream
->max_sb_size
;
1160 stream
->chan
->lost_packets
++;
1164 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1166 ERR("Snapshot ustctl_get_subbuf_size");
1167 goto error_put_subbuf
;
1170 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1172 ERR("Snapshot ustctl_get_padded_subbuf_size");
1173 goto error_put_subbuf
;
1176 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, len
,
1177 padded_len
- len
, NULL
);
1179 if (read_len
!= len
) {
1181 goto error_put_subbuf
;
1184 if (read_len
!= padded_len
) {
1186 goto error_put_subbuf
;
1190 ret
= ustctl_put_subbuf(stream
->ustream
);
1192 ERR("Snapshot ustctl_put_subbuf");
1193 goto error_close_stream
;
1195 consumed_pos
+= stream
->max_sb_size
;
1198 /* Simply close the stream so we can use it on the next snapshot. */
1199 consumer_stream_close(stream
);
1200 pthread_mutex_unlock(&stream
->lock
);
1207 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1208 ERR("Snapshot ustctl_put_subbuf");
1211 consumer_stream_close(stream
);
1213 pthread_mutex_unlock(&stream
->lock
);
1220 * Receive the metadata updates from the sessiond. Supports receiving
1221 * overlapping metadata, but is needs to always belong to a contiguous
1222 * range starting from 0.
1223 * Be careful about the locks held when calling this function: it needs
1224 * the metadata cache flush to concurrently progress in order to
1227 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1228 uint64_t len
, uint64_t version
,
1229 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
1231 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1234 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1236 metadata_str
= zmalloc(len
* sizeof(char));
1237 if (!metadata_str
) {
1238 PERROR("zmalloc metadata string");
1239 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1243 health_code_update();
1245 /* Receive metadata string. */
1246 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1248 /* Session daemon is dead so return gracefully. */
1253 health_code_update();
1255 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1256 ret
= consumer_metadata_cache_write(channel
, offset
, len
, version
,
1259 /* Unable to handle metadata. Notify session daemon. */
1260 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1262 * Skip metadata flush on write error since the offset and len might
1263 * not have been updated which could create an infinite loop below when
1264 * waiting for the metadata cache to be flushed.
1266 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1269 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1274 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1275 DBG("Waiting for metadata to be flushed");
1277 health_code_update();
1279 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1289 * Receive command from session daemon and process it.
1291 * Return 1 on success else a negative value or 0.
1293 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1294 int sock
, struct pollfd
*consumer_sockpoll
)
1297 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1298 struct lttcomm_consumer_msg msg
;
1299 struct lttng_consumer_channel
*channel
= NULL
;
1301 health_code_update();
1303 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1304 if (ret
!= sizeof(msg
)) {
1305 DBG("Consumer received unexpected message size %zd (expects %zu)",
1308 * The ret value might 0 meaning an orderly shutdown but this is ok
1309 * since the caller handles this.
1312 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1318 health_code_update();
1321 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1323 health_code_update();
1325 /* relayd needs RCU read-side lock */
1328 switch (msg
.cmd_type
) {
1329 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1331 /* Session daemon status message are handled in the following call. */
1332 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1333 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
1334 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
1335 msg
.u
.relayd_sock
.relayd_session_id
);
1338 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1340 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1341 struct consumer_relayd_sock_pair
*relayd
;
1343 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1345 /* Get relayd reference if exists. */
1346 relayd
= consumer_find_relayd(index
);
1347 if (relayd
== NULL
) {
1348 DBG("Unable to find relayd %" PRIu64
, index
);
1349 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1353 * Each relayd socket pair has a refcount of stream attached to it
1354 * which tells if the relayd is still active or not depending on the
1357 * This will set the destroy flag of the relayd object and destroy it
1358 * if the refcount reaches zero when called.
1360 * The destroy can happen either here or when a stream fd hangs up.
1363 consumer_flag_relayd_for_destroy(relayd
);
1366 goto end_msg_sessiond
;
1368 case LTTNG_CONSUMER_UPDATE_STREAM
:
1373 case LTTNG_CONSUMER_DATA_PENDING
:
1375 int ret
, is_data_pending
;
1376 uint64_t id
= msg
.u
.data_pending
.session_id
;
1378 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1380 is_data_pending
= consumer_data_pending(id
);
1382 /* Send back returned value to session daemon */
1383 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1384 sizeof(is_data_pending
));
1386 DBG("Error when sending the data pending ret code: %d", ret
);
1391 * No need to send back a status message since the data pending
1392 * returned value is the response.
1396 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1399 struct ustctl_consumer_channel_attr attr
;
1401 /* Create a plain object and reserve a channel key. */
1402 channel
= allocate_channel(msg
.u
.ask_channel
.session_id
,
1403 msg
.u
.ask_channel
.pathname
, msg
.u
.ask_channel
.name
,
1404 msg
.u
.ask_channel
.uid
, msg
.u
.ask_channel
.gid
,
1405 msg
.u
.ask_channel
.relayd_id
, msg
.u
.ask_channel
.key
,
1406 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1407 msg
.u
.ask_channel
.tracefile_size
,
1408 msg
.u
.ask_channel
.tracefile_count
,
1409 msg
.u
.ask_channel
.session_id_per_pid
,
1410 msg
.u
.ask_channel
.monitor
,
1411 msg
.u
.ask_channel
.live_timer_interval
,
1412 msg
.u
.ask_channel
.root_shm_path
,
1413 msg
.u
.ask_channel
.shm_path
);
1415 goto end_channel_error
;
1419 * Assign UST application UID to the channel. This value is ignored for
1420 * per PID buffers. This is specific to UST thus setting this after the
1423 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1425 /* Build channel attributes from received message. */
1426 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1427 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1428 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1429 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1430 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1431 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1432 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1433 attr
.blocking_timeout
= msg
.u
.ask_channel
.blocking_timeout
;
1435 /* Match channel buffer type to the UST abi. */
1436 switch (msg
.u
.ask_channel
.output
) {
1437 case LTTNG_EVENT_MMAP
:
1439 attr
.output
= LTTNG_UST_MMAP
;
1443 /* Translate and save channel type. */
1444 switch (msg
.u
.ask_channel
.type
) {
1445 case LTTNG_UST_CHAN_PER_CPU
:
1446 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1447 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1449 * Set refcount to 1 for owner. Below, we will
1450 * pass ownership to the
1451 * consumer_thread_channel_poll() thread.
1453 channel
->refcount
= 1;
1455 case LTTNG_UST_CHAN_METADATA
:
1456 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1457 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1464 health_code_update();
1466 ret
= ask_channel(ctx
, sock
, channel
, &attr
);
1468 goto end_channel_error
;
1471 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1472 ret
= consumer_metadata_cache_allocate(channel
);
1474 ERR("Allocating metadata cache");
1475 goto end_channel_error
;
1477 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1478 attr
.switch_timer_interval
= 0;
1480 int monitor_start_ret
;
1482 consumer_timer_live_start(channel
,
1483 msg
.u
.ask_channel
.live_timer_interval
);
1484 monitor_start_ret
= consumer_timer_monitor_start(
1486 msg
.u
.ask_channel
.monitor_timer_interval
);
1487 if (monitor_start_ret
< 0) {
1488 ERR("Starting channel monitoring timer failed");
1489 goto end_channel_error
;
1493 health_code_update();
1496 * Add the channel to the internal state AFTER all streams were created
1497 * and successfully sent to session daemon. This way, all streams must
1498 * be ready before this channel is visible to the threads.
1499 * If add_channel succeeds, ownership of the channel is
1500 * passed to consumer_thread_channel_poll().
1502 ret
= add_channel(channel
, ctx
);
1504 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1505 if (channel
->switch_timer_enabled
== 1) {
1506 consumer_timer_switch_stop(channel
);
1508 consumer_metadata_cache_destroy(channel
);
1510 if (channel
->live_timer_enabled
== 1) {
1511 consumer_timer_live_stop(channel
);
1513 if (channel
->monitor_timer_enabled
== 1) {
1514 consumer_timer_monitor_stop(channel
);
1516 goto end_channel_error
;
1519 health_code_update();
1522 * Channel and streams are now created. Inform the session daemon that
1523 * everything went well and should wait to receive the channel and
1524 * streams with ustctl API.
1526 ret
= consumer_send_status_channel(sock
, channel
);
1529 * There is probably a problem on the socket.
1536 case LTTNG_CONSUMER_GET_CHANNEL
:
1538 int ret
, relayd_err
= 0;
1539 uint64_t key
= msg
.u
.get_channel
.key
;
1540 struct lttng_consumer_channel
*channel
;
1542 channel
= consumer_find_channel(key
);
1544 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1545 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1546 goto end_msg_sessiond
;
1549 health_code_update();
1551 /* Send everything to sessiond. */
1552 ret
= send_sessiond_channel(sock
, channel
, ctx
, &relayd_err
);
1556 * We were unable to send to the relayd the stream so avoid
1557 * sending back a fatal error to the thread since this is OK
1558 * and the consumer can continue its work. The above call
1559 * has sent the error status message to the sessiond.
1564 * The communicaton was broken hence there is a bad state between
1565 * the consumer and sessiond so stop everything.
1570 health_code_update();
1573 * In no monitor mode, the streams ownership is kept inside the channel
1574 * so don't send them to the data thread.
1576 if (!channel
->monitor
) {
1577 goto end_msg_sessiond
;
1580 ret
= send_streams_to_thread(channel
, ctx
);
1583 * If we are unable to send the stream to the thread, there is
1584 * a big problem so just stop everything.
1588 /* List MUST be empty after or else it could be reused. */
1589 assert(cds_list_empty(&channel
->streams
.head
));
1590 goto end_msg_sessiond
;
1592 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1594 uint64_t key
= msg
.u
.destroy_channel
.key
;
1597 * Only called if streams have not been sent to stream
1598 * manager thread. However, channel has been sent to
1599 * channel manager thread.
1601 notify_thread_del_channel(ctx
, key
);
1602 goto end_msg_sessiond
;
1604 case LTTNG_CONSUMER_CLOSE_METADATA
:
1608 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1613 goto end_msg_sessiond
;
1615 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1619 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1624 goto end_msg_sessiond
;
1626 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL
:
1630 ret
= clear_quiescent_channel(
1631 msg
.u
.clear_quiescent_channel
.key
);
1636 goto end_msg_sessiond
;
1638 case LTTNG_CONSUMER_PUSH_METADATA
:
1641 uint64_t len
= msg
.u
.push_metadata
.len
;
1642 uint64_t key
= msg
.u
.push_metadata
.key
;
1643 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1644 uint64_t version
= msg
.u
.push_metadata
.version
;
1645 struct lttng_consumer_channel
*channel
;
1647 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1650 channel
= consumer_find_channel(key
);
1653 * This is possible if the metadata creation on the consumer side
1654 * is in flight vis-a-vis a concurrent push metadata from the
1655 * session daemon. Simply return that the channel failed and the
1656 * session daemon will handle that message correctly considering
1657 * that this race is acceptable thus the DBG() statement here.
1659 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1660 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1661 goto end_msg_sessiond
;
1664 health_code_update();
1668 * There is nothing to receive. We have simply
1669 * checked whether the channel can be found.
1671 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1672 goto end_msg_sessiond
;
1675 /* Tell session daemon we are ready to receive the metadata. */
1676 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1678 /* Somehow, the session daemon is not responding anymore. */
1682 health_code_update();
1684 /* Wait for more data. */
1685 health_poll_entry();
1686 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1692 health_code_update();
1694 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1695 len
, version
, channel
, 0, 1);
1697 /* error receiving from sessiond */
1701 goto end_msg_sessiond
;
1704 case LTTNG_CONSUMER_SETUP_METADATA
:
1708 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1712 goto end_msg_sessiond
;
1714 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1716 if (msg
.u
.snapshot_channel
.metadata
) {
1717 ret
= snapshot_metadata(msg
.u
.snapshot_channel
.key
,
1718 msg
.u
.snapshot_channel
.pathname
,
1719 msg
.u
.snapshot_channel
.relayd_id
,
1722 ERR("Snapshot metadata failed");
1723 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1726 ret
= snapshot_channel(msg
.u
.snapshot_channel
.key
,
1727 msg
.u
.snapshot_channel
.pathname
,
1728 msg
.u
.snapshot_channel
.relayd_id
,
1729 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
1732 ERR("Snapshot channel failed");
1733 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1737 health_code_update();
1738 ret
= consumer_send_status_msg(sock
, ret_code
);
1740 /* Somehow, the session daemon is not responding anymore. */
1743 health_code_update();
1746 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
1749 uint64_t discarded_events
;
1750 struct lttng_ht_iter iter
;
1751 struct lttng_ht
*ht
;
1752 struct lttng_consumer_stream
*stream
;
1753 uint64_t id
= msg
.u
.discarded_events
.session_id
;
1754 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
1756 DBG("UST consumer discarded events command for session id %"
1759 pthread_mutex_lock(&consumer_data
.lock
);
1761 ht
= consumer_data
.stream_list_ht
;
1764 * We only need a reference to the channel, but they are not
1765 * directly indexed, so we just use the first matching stream
1766 * to extract the information we need, we default to 0 if not
1767 * found (no events are dropped if the channel is not yet in
1770 discarded_events
= 0;
1771 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1772 ht
->hash_fct(&id
, lttng_ht_seed
),
1774 &iter
.iter
, stream
, node_session_id
.node
) {
1775 if (stream
->chan
->key
== key
) {
1776 discarded_events
= stream
->chan
->discarded_events
;
1780 pthread_mutex_unlock(&consumer_data
.lock
);
1783 DBG("UST consumer discarded events command for session id %"
1784 PRIu64
", channel key %" PRIu64
, id
, key
);
1786 health_code_update();
1788 /* Send back returned value to session daemon */
1789 ret
= lttcomm_send_unix_sock(sock
, &discarded_events
, sizeof(discarded_events
));
1791 PERROR("send discarded events");
1797 case LTTNG_CONSUMER_LOST_PACKETS
:
1800 uint64_t lost_packets
;
1801 struct lttng_ht_iter iter
;
1802 struct lttng_ht
*ht
;
1803 struct lttng_consumer_stream
*stream
;
1804 uint64_t id
= msg
.u
.lost_packets
.session_id
;
1805 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
1807 DBG("UST consumer lost packets command for session id %"
1810 pthread_mutex_lock(&consumer_data
.lock
);
1812 ht
= consumer_data
.stream_list_ht
;
1815 * We only need a reference to the channel, but they are not
1816 * directly indexed, so we just use the first matching stream
1817 * to extract the information we need, we default to 0 if not
1818 * found (no packets lost if the channel is not yet in use).
1821 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1822 ht
->hash_fct(&id
, lttng_ht_seed
),
1824 &iter
.iter
, stream
, node_session_id
.node
) {
1825 if (stream
->chan
->key
== key
) {
1826 lost_packets
= stream
->chan
->lost_packets
;
1830 pthread_mutex_unlock(&consumer_data
.lock
);
1833 DBG("UST consumer lost packets command for session id %"
1834 PRIu64
", channel key %" PRIu64
, id
, key
);
1836 health_code_update();
1838 /* Send back returned value to session daemon */
1839 ret
= lttcomm_send_unix_sock(sock
, &lost_packets
,
1840 sizeof(lost_packets
));
1842 PERROR("send lost packets");
1848 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
1850 int channel_monitor_pipe
;
1852 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1853 /* Successfully received the command's type. */
1854 ret
= consumer_send_status_msg(sock
, ret_code
);
1859 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
1861 if (ret
!= sizeof(channel_monitor_pipe
)) {
1862 ERR("Failed to receive channel monitor pipe");
1866 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
1867 ret
= consumer_timer_thread_set_channel_monitor_pipe(
1868 channel_monitor_pipe
);
1872 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1873 /* Set the pipe as non-blocking. */
1874 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
1876 PERROR("fcntl get flags of the channel monitoring pipe");
1881 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
1882 flags
| O_NONBLOCK
);
1884 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1887 DBG("Channel monitor pipe set as non-blocking");
1889 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
1891 goto end_msg_sessiond
;
1893 case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE
:
1895 int channel_rotate_pipe
;
1898 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1899 /* Successfully received the command's type. */
1900 ret
= consumer_send_status_msg(sock
, ret_code
);
1905 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_rotate_pipe
,
1907 if (ret
!= sizeof(channel_rotate_pipe
)) {
1908 ERR("Failed to receive channel rotate pipe");
1912 DBG("Received channel rotate pipe (%d)", channel_rotate_pipe
);
1913 ctx
->channel_rotate_pipe
= channel_rotate_pipe
;
1914 /* Set the pipe as non-blocking. */
1915 ret
= fcntl(channel_rotate_pipe
, F_GETFL
, 0);
1917 PERROR("fcntl get flags of the channel rotate pipe");
1922 ret
= fcntl(channel_rotate_pipe
, F_SETFL
,
1923 flags
| O_NONBLOCK
);
1925 PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
1928 DBG("Channel rotate pipe set as non-blocking");
1929 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1930 ret
= consumer_send_status_msg(sock
, ret_code
);
1936 case LTTNG_CONSUMER_ROTATE_CHANNEL
:
1938 ret
= lttng_consumer_rotate_channel(msg
.u
.rotate_channel
.key
,
1939 msg
.u
.rotate_channel
.pathname
,
1940 msg
.u
.rotate_channel
.relayd_id
,
1941 msg
.u
.rotate_channel
.metadata
,
1942 msg
.u
.rotate_channel
.new_chunk_id
,
1945 ERR("Rotate channel failed");
1946 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1949 health_code_update();
1951 ret
= consumer_send_status_msg(sock
, ret_code
);
1953 /* Somehow, the session daemon is not responding anymore. */
1958 * Rotate the streams that are ready right now.
1959 * FIXME: this is a second consecutive iteration over the
1960 * streams in a channel, there is probably a better way to
1961 * handle this, but it needs to be after the
1962 * consumer_send_status_msg() call.
1964 ret
= lttng_consumer_rotate_ready_streams(
1965 msg
.u
.rotate_channel
.key
, ctx
);
1967 ERR("Rotate channel failed");
1968 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1972 case LTTNG_CONSUMER_ROTATE_RENAME
:
1974 DBG("Consumer rename session %" PRIu64
" after rotation",
1975 msg
.u
.rotate_rename
.session_id
);
1976 ret
= lttng_consumer_rotate_rename(msg
.u
.rotate_rename
.current_path
,
1977 msg
.u
.rotate_rename
.new_path
,
1978 msg
.u
.rotate_rename
.uid
,
1979 msg
.u
.rotate_rename
.gid
,
1980 msg
.u
.rotate_rename
.relayd_id
);
1982 ERR("Rotate rename failed");
1983 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1986 health_code_update();
1988 ret
= consumer_send_status_msg(sock
, ret_code
);
1990 /* Somehow, the session daemon is not responding anymore. */
1995 case LTTNG_CONSUMER_ROTATE_PENDING_RELAY
:
1997 DBG("Consumer rotate pending on relay for session %" PRIu64
,
1998 msg
.u
.rotate_pending_relay
.session_id
);
1999 ret
= lttng_consumer_rotate_pending_relay(
2000 msg
.u
.rotate_pending_relay
.session_id
,
2001 msg
.u
.rotate_pending_relay
.relayd_id
,
2002 msg
.u
.rotate_pending_relay
.chunk_id
);
2004 ERR("Rotate pending relay failed");
2005 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2008 health_code_update();
2010 ret
= consumer_send_status_msg(sock
, ret_code
);
2012 /* Somehow, the session daemon is not responding anymore. */
2024 health_code_update();
2027 * Return 1 to indicate success since the 0 value can be a socket
2028 * shutdown during the recv() or send() call.
2034 * The returned value here is not useful since either way we'll return 1 to
2035 * the caller because the session daemon socket management is done
2036 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2038 ret
= consumer_send_status_msg(sock
, ret_code
);
2044 health_code_update();
2050 * Free channel here since no one has a reference to it. We don't
2051 * free after that because a stream can store this pointer.
2053 destroy_channel(channel
);
2055 /* We have to send a status channel message indicating an error. */
2056 ret
= consumer_send_status_channel(sock
, NULL
);
2058 /* Stop everything if session daemon can not be notified. */
2063 health_code_update();
2068 /* This will issue a consumer stop. */
2073 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2074 * compiled out, we isolate it in this library.
2076 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream
*stream
,
2080 assert(stream
->ustream
);
2082 return ustctl_get_mmap_read_offset(stream
->ustream
, off
);
2086 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2087 * compiled out, we isolate it in this library.
2089 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream
*stream
)
2092 assert(stream
->ustream
);
2094 return ustctl_get_mmap_base(stream
->ustream
);
2097 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream
*stream
,
2098 int producer_active
)
2101 assert(stream
->ustream
);
2103 ustctl_flush_buffer(stream
->ustream
, producer_active
);
2107 * Take a snapshot for a specific stream.
2109 * Returns 0 on success, < 0 on error
2111 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2114 assert(stream
->ustream
);
2116 return ustctl_snapshot(stream
->ustream
);
2120 * Sample consumed and produced positions for a specific stream.
2122 * Returns 0 on success, < 0 on error.
2124 int lttng_ustconsumer_sample_snapshot_positions(
2125 struct lttng_consumer_stream
*stream
)
2128 assert(stream
->ustream
);
2130 return ustctl_snapshot_sample_positions(stream
->ustream
);
2134 * Get the produced position
2136 * Returns 0 on success, < 0 on error
2138 int lttng_ustconsumer_get_produced_snapshot(
2139 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2142 assert(stream
->ustream
);
2145 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
2149 * Get the consumed position
2151 * Returns 0 on success, < 0 on error
2153 int lttng_ustconsumer_get_consumed_snapshot(
2154 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2157 assert(stream
->ustream
);
2160 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
2163 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
2167 assert(stream
->ustream
);
2169 ustctl_flush_buffer(stream
->ustream
, producer
);
2172 int lttng_ustconsumer_get_current_timestamp(
2173 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
2176 assert(stream
->ustream
);
2179 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
2182 int lttng_ustconsumer_get_sequence_number(
2183 struct lttng_consumer_stream
*stream
, uint64_t *seq
)
2186 assert(stream
->ustream
);
2189 return ustctl_get_sequence_number(stream
->ustream
, seq
);
2193 * Called when the stream signals the consumer that it has hung up.
2195 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
2198 assert(stream
->ustream
);
2200 pthread_mutex_lock(&stream
->lock
);
2201 if (!stream
->quiescent
) {
2202 ustctl_flush_buffer(stream
->ustream
, 0);
2203 stream
->quiescent
= true;
2205 pthread_mutex_unlock(&stream
->lock
);
2206 stream
->hangup_flush_done
= 1;
2209 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
2214 assert(chan
->uchan
);
2216 if (chan
->switch_timer_enabled
== 1) {
2217 consumer_timer_switch_stop(chan
);
2219 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
2222 ret
= close(chan
->stream_fds
[i
]);
2226 if (chan
->shm_path
[0]) {
2227 char shm_path
[PATH_MAX
];
2229 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
2231 ERR("Cannot get stream shm path");
2233 ret
= run_as_unlink(shm_path
, chan
->uid
, chan
->gid
);
2235 PERROR("unlink %s", shm_path
);
2241 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
2244 assert(chan
->uchan
);
2246 consumer_metadata_cache_destroy(chan
);
2247 ustctl_destroy_channel(chan
->uchan
);
2248 /* Try to rmdir all directories under shm_path root. */
2249 if (chan
->root_shm_path
[0]) {
2250 (void) run_as_rmdir_recursive(chan
->root_shm_path
,
2251 chan
->uid
, chan
->gid
);
2253 free(chan
->stream_fds
);
2256 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
2259 assert(stream
->ustream
);
2261 if (stream
->chan
->switch_timer_enabled
== 1) {
2262 consumer_timer_switch_stop(stream
->chan
);
2264 ustctl_destroy_stream(stream
->ustream
);
2267 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
2270 assert(stream
->ustream
);
2272 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
2275 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
2278 assert(stream
->ustream
);
2280 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
2284 * Populate index values of a UST stream. Values are set in big endian order.
2286 * Return 0 on success or else a negative value.
2288 static int get_index_values(struct ctf_packet_index
*index
,
2289 struct ustctl_consumer_stream
*ustream
)
2293 ret
= ustctl_get_timestamp_begin(ustream
, &index
->timestamp_begin
);
2295 PERROR("ustctl_get_timestamp_begin");
2298 index
->timestamp_begin
= htobe64(index
->timestamp_begin
);
2300 ret
= ustctl_get_timestamp_end(ustream
, &index
->timestamp_end
);
2302 PERROR("ustctl_get_timestamp_end");
2305 index
->timestamp_end
= htobe64(index
->timestamp_end
);
2307 ret
= ustctl_get_events_discarded(ustream
, &index
->events_discarded
);
2309 PERROR("ustctl_get_events_discarded");
2312 index
->events_discarded
= htobe64(index
->events_discarded
);
2314 ret
= ustctl_get_content_size(ustream
, &index
->content_size
);
2316 PERROR("ustctl_get_content_size");
2319 index
->content_size
= htobe64(index
->content_size
);
2321 ret
= ustctl_get_packet_size(ustream
, &index
->packet_size
);
2323 PERROR("ustctl_get_packet_size");
2326 index
->packet_size
= htobe64(index
->packet_size
);
2328 ret
= ustctl_get_stream_id(ustream
, &index
->stream_id
);
2330 PERROR("ustctl_get_stream_id");
2333 index
->stream_id
= htobe64(index
->stream_id
);
2335 ret
= ustctl_get_instance_id(ustream
, &index
->stream_instance_id
);
2337 PERROR("ustctl_get_instance_id");
2340 index
->stream_instance_id
= htobe64(index
->stream_instance_id
);
2342 ret
= ustctl_get_sequence_number(ustream
, &index
->packet_seq_num
);
2344 PERROR("ustctl_get_sequence_number");
2347 index
->packet_seq_num
= htobe64(index
->packet_seq_num
);
2354 void metadata_stream_reset_cache(struct lttng_consumer_stream
*stream
,
2355 struct consumer_metadata_cache
*cache
)
2357 DBG("Metadata stream update to version %" PRIu64
,
2359 stream
->ust_metadata_pushed
= 0;
2360 stream
->metadata_version
= cache
->version
;
2361 stream
->reset_metadata_flag
= 1;
2365 * Check if the version of the metadata stream and metadata cache match.
2366 * If the cache got updated, reset the metadata stream.
2367 * The stream lock and metadata cache lock MUST be held.
2368 * Return 0 on success, a negative value on error.
2371 int metadata_stream_check_version(struct lttng_consumer_stream
*stream
)
2374 struct consumer_metadata_cache
*cache
= stream
->chan
->metadata_cache
;
2376 if (cache
->version
== stream
->metadata_version
) {
2379 metadata_stream_reset_cache(stream
, cache
);
2386 * Write up to one packet from the metadata cache to the channel.
2388 * Returns the number of bytes pushed in the cache, or a negative value
2392 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
2397 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2398 ret
= metadata_stream_check_version(stream
);
2402 if (stream
->chan
->metadata_cache
->max_offset
2403 == stream
->ust_metadata_pushed
) {
2408 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
2409 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
2410 stream
->chan
->metadata_cache
->max_offset
2411 - stream
->ust_metadata_pushed
);
2412 assert(write_len
!= 0);
2413 if (write_len
< 0) {
2414 ERR("Writing one metadata packet");
2418 stream
->ust_metadata_pushed
+= write_len
;
2420 assert(stream
->chan
->metadata_cache
->max_offset
>=
2421 stream
->ust_metadata_pushed
);
2425 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2431 * Sync metadata meaning request them to the session daemon and snapshot to the
2432 * metadata thread can consumer them.
2434 * Metadata stream lock is held here, but we need to release it when
2435 * interacting with sessiond, else we cause a deadlock with live
2436 * awaiting on metadata to be pushed out.
2438 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2439 * is empty or a negative value on error.
2441 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data
*ctx
,
2442 struct lttng_consumer_stream
*metadata
)
2450 pthread_mutex_unlock(&metadata
->lock
);
2452 * Request metadata from the sessiond, but don't wait for the flush
2453 * because we locked the metadata thread.
2455 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata
->chan
, 0, 0);
2456 pthread_mutex_lock(&metadata
->lock
);
2461 ret
= commit_one_metadata_packet(metadata
);
2464 } else if (ret
> 0) {
2468 ustctl_flush_buffer(metadata
->ustream
, 1);
2469 ret
= ustctl_snapshot(metadata
->ustream
);
2471 if (errno
!= EAGAIN
) {
2472 ERR("Sync metadata, taking UST snapshot");
2475 DBG("No new metadata when syncing them.");
2476 /* No new metadata, exit. */
2482 * After this flush, we still need to extract metadata.
2493 * Return 0 on success else a negative value.
2495 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2496 struct lttng_consumer_local_data
*ctx
)
2499 struct ustctl_consumer_stream
*ustream
;
2504 ustream
= stream
->ustream
;
2507 * First, we are going to check if there is a new subbuffer available
2508 * before reading the stream wait_fd.
2510 /* Get the next subbuffer */
2511 ret
= ustctl_get_next_subbuf(ustream
);
2513 /* No more data found, flag the stream. */
2514 stream
->has_data
= 0;
2519 ret
= ustctl_put_subbuf(ustream
);
2522 /* This stream still has data. Flag it and wake up the data thread. */
2523 stream
->has_data
= 1;
2525 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2528 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2529 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2534 /* The wake up pipe has been notified. */
2535 ctx
->has_wakeup
= 1;
2544 int update_stream_stats(struct lttng_consumer_stream
*stream
)
2547 uint64_t seq
, discarded
;
2549 ret
= ustctl_get_sequence_number(stream
->ustream
, &seq
);
2551 PERROR("ustctl_get_sequence_number");
2555 * Start the sequence when we extract the first packet in case we don't
2556 * start at 0 (for example if a consumer is not connected to the
2557 * session immediately after the beginning).
2559 if (stream
->last_sequence_number
== -1ULL) {
2560 stream
->last_sequence_number
= seq
;
2561 } else if (seq
> stream
->last_sequence_number
) {
2562 stream
->chan
->lost_packets
+= seq
-
2563 stream
->last_sequence_number
- 1;
2565 /* seq <= last_sequence_number */
2566 ERR("Sequence number inconsistent : prev = %" PRIu64
2567 ", current = %" PRIu64
,
2568 stream
->last_sequence_number
, seq
);
2572 stream
->last_sequence_number
= seq
;
2574 ret
= ustctl_get_events_discarded(stream
->ustream
, &discarded
);
2576 PERROR("kernctl_get_events_discarded");
2579 if (discarded
< stream
->last_discarded_events
) {
2581 * Overflow has occurred. We assume only one wrap-around
2584 stream
->chan
->discarded_events
+=
2585 (1ULL << (CAA_BITS_PER_LONG
- 1)) -
2586 stream
->last_discarded_events
+ discarded
;
2588 stream
->chan
->discarded_events
+= discarded
-
2589 stream
->last_discarded_events
;
2591 stream
->last_discarded_events
= discarded
;
2599 * Read subbuffer from the given stream.
2601 * Stream lock MUST be acquired.
2603 * Return 0 on success else a negative value.
2605 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
2606 struct lttng_consumer_local_data
*ctx
)
2608 unsigned long len
, subbuf_size
, padding
;
2609 int err
, write_index
= 1, rotation_ret
, rotate_ready
;
2611 struct ustctl_consumer_stream
*ustream
;
2612 struct ctf_packet_index index
;
2615 assert(stream
->ustream
);
2618 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream
->wait_fd
,
2621 /* Ease our life for what's next. */
2622 ustream
= stream
->ustream
;
2625 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2626 * error if we cannot read this one byte (read returns 0), or if the error
2627 * is EAGAIN or EWOULDBLOCK.
2629 * This is only done when the stream is monitored by a thread, before the
2630 * flush is done after a hangup and if the stream is not flagged with data
2631 * since there might be nothing to consume in the wait fd but still have
2632 * data available flagged by the consumer wake up pipe.
2634 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2638 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2639 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2646 /* Get the next subbuffer */
2647 err
= ustctl_get_next_subbuf(ustream
);
2650 * Populate metadata info if the existing info has
2651 * already been read.
2653 if (stream
->metadata_flag
) {
2654 ret
= commit_one_metadata_packet(stream
);
2658 ustctl_flush_buffer(stream
->ustream
, 1);
2662 ret
= err
; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2664 * This is a debug message even for single-threaded consumer,
2665 * because poll() have more relaxed criterions than get subbuf,
2666 * so get_subbuf may fail for short race windows where poll()
2667 * would issue wakeups.
2669 DBG("Reserving sub buffer failed (everything is normal, "
2670 "it is due to concurrency) [ret: %d]", err
);
2673 assert(stream
->chan
->output
== CONSUMER_CHANNEL_MMAP
);
2675 if (!stream
->metadata_flag
) {
2676 index
.offset
= htobe64(stream
->out_fd_offset
);
2677 ret
= get_index_values(&index
, ustream
);
2679 err
= ustctl_put_subbuf(ustream
);
2684 /* Update the stream's sequence and discarded events count. */
2685 ret
= update_stream_stats(stream
);
2687 PERROR("kernctl_get_events_discarded");
2688 err
= ustctl_put_subbuf(ustream
);
2696 /* Get the full padded subbuffer size */
2697 err
= ustctl_get_padded_subbuf_size(ustream
, &len
);
2700 /* Get subbuffer data size (without padding) */
2701 err
= ustctl_get_subbuf_size(ustream
, &subbuf_size
);
2704 /* Make sure we don't get a subbuffer size bigger than the padded */
2705 assert(len
>= subbuf_size
);
2707 padding
= len
- subbuf_size
;
2709 rotate_ready
= lttng_consumer_stream_is_rotate_ready(stream
, len
);
2710 if (rotate_ready
< 0) {
2711 ERR("Failed to check if stream is ready for rotation");
2713 err
= ustctl_put_subbuf(ustream
);
2717 stream
->rotate_ready
= rotate_ready
;
2719 /* write the subbuffer to the tracefile */
2720 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
, padding
, &index
);
2722 * The mmap operation should write subbuf_size amount of data when network
2723 * streaming or the full padding (len) size when we are _not_ streaming.
2725 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
2726 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
2728 * Display the error but continue processing to try to release the
2729 * subbuffer. This is a DBG statement since any unexpected kill or
2730 * signal, the application gets unregistered, relayd gets closed or
2731 * anything that affects the buffer lifetime will trigger this error.
2732 * So, for the sake of the user, don't print this error since it can
2733 * happen and it is OK with the code flow.
2735 DBG("Error writing to tracefile "
2736 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2737 ret
, len
, subbuf_size
);
2740 err
= ustctl_put_next_subbuf(ustream
);
2744 * This will consumer the byte on the wait_fd if and only if there is not
2745 * next subbuffer to be acquired.
2747 if (!stream
->metadata_flag
) {
2748 ret
= notify_if_more_data(stream
, ctx
);
2754 /* Write index if needed. */
2759 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
2761 * In live, block until all the metadata is sent.
2763 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2764 assert(!stream
->missed_metadata_flush
);
2765 stream
->waiting_on_metadata
= true;
2766 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2768 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
2770 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2771 stream
->waiting_on_metadata
= false;
2772 if (stream
->missed_metadata_flush
) {
2773 stream
->missed_metadata_flush
= false;
2774 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2775 (void) consumer_flush_ust_index(stream
);
2777 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2785 assert(!stream
->metadata_flag
);
2786 err
= consumer_stream_write_index(stream
, &index
);
2792 if (stream
->rotate_ready
) {
2793 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
);
2794 if (rotation_ret
< 0) {
2796 ERR("Stream rotation error");
2805 * Called when a stream is created.
2807 * Return 0 on success or else a negative value.
2809 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
2815 /* Don't create anything if this is set for streaming. */
2816 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
) {
2817 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
2818 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
2819 stream
->uid
, stream
->gid
, NULL
);
2823 stream
->out_fd
= ret
;
2824 stream
->tracefile_size_current
= 0;
2826 if (!stream
->metadata_flag
) {
2827 struct lttng_index_file
*index_file
;
2829 index_file
= lttng_index_file_create(stream
->chan
->pathname
,
2830 stream
->name
, stream
->uid
, stream
->gid
,
2831 stream
->chan
->tracefile_size
,
2832 stream
->tracefile_count_current
,
2833 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
2837 assert(!stream
->index_file
);
2838 stream
->index_file
= index_file
;
2848 * Check if data is still being extracted from the buffers for a specific
2849 * stream. Consumer data lock MUST be acquired before calling this function
2850 * and the stream lock.
2852 * Return 1 if the traced data are still getting read else 0 meaning that the
2853 * data is available for trace viewer reading.
2855 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
2860 assert(stream
->ustream
);
2862 DBG("UST consumer checking data pending");
2864 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
2869 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
2870 uint64_t contiguous
, pushed
;
2872 /* Ease our life a bit. */
2873 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
2874 pushed
= stream
->ust_metadata_pushed
;
2877 * We can simply check whether all contiguously available data
2878 * has been pushed to the ring buffer, since the push operation
2879 * is performed within get_next_subbuf(), and because both
2880 * get_next_subbuf() and put_next_subbuf() are issued atomically
2881 * thanks to the stream lock within
2882 * lttng_ustconsumer_read_subbuffer(). This basically means that
2883 * whetnever ust_metadata_pushed is incremented, the associated
2884 * metadata has been consumed from the metadata stream.
2886 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
2887 contiguous
, pushed
);
2888 assert(((int64_t) (contiguous
- pushed
)) >= 0);
2889 if ((contiguous
!= pushed
) ||
2890 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
2891 ret
= 1; /* Data is pending */
2895 ret
= ustctl_get_next_subbuf(stream
->ustream
);
2898 * There is still data so let's put back this
2901 ret
= ustctl_put_subbuf(stream
->ustream
);
2903 ret
= 1; /* Data is pending */
2908 /* Data is NOT pending so ready to be read. */
2916 * Stop a given metadata channel timer if enabled and close the wait fd which
2917 * is the poll pipe of the metadata stream.
2919 * This MUST be called with the metadata channel acquired.
2921 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
2926 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
2928 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
2930 if (metadata
->switch_timer_enabled
== 1) {
2931 consumer_timer_switch_stop(metadata
);
2934 if (!metadata
->metadata_stream
) {
2939 * Closing write side so the thread monitoring the stream wakes up if any
2940 * and clean the metadata stream.
2942 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
2943 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
2945 PERROR("closing metadata pipe write side");
2947 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
2955 * Close every metadata stream wait fd of the metadata hash table. This
2956 * function MUST be used very carefully so not to run into a race between the
2957 * metadata thread handling streams and this function closing their wait fd.
2959 * For UST, this is used when the session daemon hangs up. Its the metadata
2960 * producer so calling this is safe because we are assured that no state change
2961 * can occur in the metadata thread for the streams in the hash table.
2963 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
2965 struct lttng_ht_iter iter
;
2966 struct lttng_consumer_stream
*stream
;
2968 assert(metadata_ht
);
2969 assert(metadata_ht
->ht
);
2971 DBG("UST consumer closing all metadata streams");
2974 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
2977 health_code_update();
2979 pthread_mutex_lock(&stream
->chan
->lock
);
2980 lttng_ustconsumer_close_metadata(stream
->chan
);
2981 pthread_mutex_unlock(&stream
->chan
->lock
);
2987 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
2991 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
2993 ERR("Unable to close wakeup fd");
2998 * Please refer to consumer-timer.c before adding any lock within this
2999 * function or any of its callees. Timers have a very strict locking
3000 * semantic with respect to teardown. Failure to respect this semantic
3001 * introduces deadlocks.
3003 * DON'T hold the metadata lock when calling this function, else this
3004 * can cause deadlock involving consumer awaiting for metadata to be
3005 * pushed out due to concurrent interaction with the session daemon.
3007 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
3008 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
3010 struct lttcomm_metadata_request_msg request
;
3011 struct lttcomm_consumer_msg msg
;
3012 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3013 uint64_t len
, key
, offset
, version
;
3017 assert(channel
->metadata_cache
);
3019 memset(&request
, 0, sizeof(request
));
3021 /* send the metadata request to sessiond */
3022 switch (consumer_data
.type
) {
3023 case LTTNG_CONSUMER64_UST
:
3024 request
.bits_per_long
= 64;
3026 case LTTNG_CONSUMER32_UST
:
3027 request
.bits_per_long
= 32;
3030 request
.bits_per_long
= 0;
3034 request
.session_id
= channel
->session_id
;
3035 request
.session_id_per_pid
= channel
->session_id_per_pid
;
3037 * Request the application UID here so the metadata of that application can
3038 * be sent back. The channel UID corresponds to the user UID of the session
3039 * used for the rights on the stream file(s).
3041 request
.uid
= channel
->ust_app_uid
;
3042 request
.key
= channel
->key
;
3044 DBG("Sending metadata request to sessiond, session id %" PRIu64
3045 ", per-pid %" PRIu64
", app UID %u and channek key %" PRIu64
,
3046 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
3049 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
3051 health_code_update();
3053 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
3056 ERR("Asking metadata to sessiond");
3060 health_code_update();
3062 /* Receive the metadata from sessiond */
3063 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
3065 if (ret
!= sizeof(msg
)) {
3066 DBG("Consumer received unexpected message size %d (expects %zu)",
3068 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
3070 * The ret value might 0 meaning an orderly shutdown but this is ok
3071 * since the caller handles this.
3076 health_code_update();
3078 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
3079 /* No registry found */
3080 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3084 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
3085 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
3090 len
= msg
.u
.push_metadata
.len
;
3091 key
= msg
.u
.push_metadata
.key
;
3092 offset
= msg
.u
.push_metadata
.target_offset
;
3093 version
= msg
.u
.push_metadata
.version
;
3095 assert(key
== channel
->key
);
3097 DBG("No new metadata to receive for key %" PRIu64
, key
);
3100 health_code_update();
3102 /* Tell session daemon we are ready to receive the metadata. */
3103 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3104 LTTCOMM_CONSUMERD_SUCCESS
);
3105 if (ret
< 0 || len
== 0) {
3107 * Somehow, the session daemon is not responding anymore or there is
3108 * nothing to receive.
3113 health_code_update();
3115 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
3116 key
, offset
, len
, version
, channel
, timer
, wait
);
3119 * Only send the status msg if the sessiond is alive meaning a positive
3122 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
3127 health_code_update();
3129 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
3134 * Return the ustctl call for the get stream id.
3136 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
3137 uint64_t *stream_id
)
3142 return ustctl_get_stream_id(stream
->ustream
, stream_id
);