2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
22 #include <lttng/ust-ctl.h>
28 #include <sys/socket.h>
30 #include <sys/types.h>
33 #include <urcu/list.h>
37 #include <bin/lttng-consumerd/health-consumerd.h>
38 #include <common/common.h>
39 #include <common/sessiond-comm/sessiond-comm.h>
40 #include <common/relayd/relayd.h>
41 #include <common/compat/fcntl.h>
42 #include <common/compat/endian.h>
43 #include <common/consumer/consumer-metadata-cache.h>
44 #include <common/consumer/consumer-stream.h>
45 #include <common/consumer/consumer-timer.h>
46 #include <common/utils.h>
47 #include <common/index/index.h>
49 #include "ust-consumer.h"
51 #define INT_MAX_STR_LEN 12 /* includes \0 */
53 extern struct lttng_consumer_global_data consumer_data
;
54 extern int consumer_poll_timeout
;
57 * Free channel object and all streams associated with it. This MUST be used
58 * only and only if the channel has _NEVER_ been added to the global channel
61 static void destroy_channel(struct lttng_consumer_channel
*channel
)
63 struct lttng_consumer_stream
*stream
, *stmp
;
67 DBG("UST consumer cleaning stream list");
69 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
74 cds_list_del(&stream
->send_node
);
75 ustctl_destroy_stream(stream
->ustream
);
80 * If a channel is available meaning that was created before the streams
84 lttng_ustconsumer_del_channel(channel
);
85 lttng_ustconsumer_free_channel(channel
);
91 * Add channel to internal consumer state.
93 * Returns 0 on success or else a negative value.
95 static int add_channel(struct lttng_consumer_channel
*channel
,
96 struct lttng_consumer_local_data
*ctx
)
103 if (ctx
->on_recv_channel
!= NULL
) {
104 ret
= ctx
->on_recv_channel(channel
);
106 ret
= consumer_add_channel(channel
, ctx
);
107 } else if (ret
< 0) {
108 /* Most likely an ENOMEM. */
109 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
113 ret
= consumer_add_channel(channel
, ctx
);
116 DBG("UST consumer channel added (key: %" PRIu64
")", channel
->key
);
123 * Allocate and return a consumer channel object.
125 static struct lttng_consumer_channel
*allocate_channel(uint64_t session_id
,
126 const char *pathname
, const char *name
, uid_t uid
, gid_t gid
,
127 uint64_t relayd_id
, uint64_t key
, enum lttng_event_output output
,
128 uint64_t tracefile_size
, uint64_t tracefile_count
,
129 uint64_t session_id_per_pid
, unsigned int monitor
,
130 unsigned int live_timer_interval
,
131 const char *root_shm_path
, const char *shm_path
)
136 return consumer_allocate_channel(key
, session_id
, pathname
, name
, uid
,
137 gid
, relayd_id
, output
, tracefile_size
,
138 tracefile_count
, session_id_per_pid
, monitor
,
139 live_timer_interval
, root_shm_path
, shm_path
);
143 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
144 * error value if applicable is set in it else it is kept untouched.
146 * Return NULL on error else the newly allocated stream object.
148 static struct lttng_consumer_stream
*allocate_stream(int cpu
, int key
,
149 struct lttng_consumer_channel
*channel
,
150 struct lttng_consumer_local_data
*ctx
, int *_alloc_ret
,
151 uint64_t trace_archive_id
)
154 struct lttng_consumer_stream
*stream
= NULL
;
159 stream
= consumer_allocate_stream(channel
->key
,
161 LTTNG_CONSUMER_ACTIVE_STREAM
,
172 if (stream
== NULL
) {
176 * We could not find the channel. Can happen if cpu hotplug
177 * happens while tearing down.
179 DBG3("Could not find channel");
184 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
190 consumer_stream_update_channel_attributes(stream
, channel
);
191 stream
->chan
= channel
;
195 *_alloc_ret
= alloc_ret
;
201 * Send the given stream pointer to the corresponding thread.
203 * Returns 0 on success else a negative value.
205 static int send_stream_to_thread(struct lttng_consumer_stream
*stream
,
206 struct lttng_consumer_local_data
*ctx
)
209 struct lttng_pipe
*stream_pipe
;
211 /* Get the right pipe where the stream will be sent. */
212 if (stream
->metadata_flag
) {
213 consumer_add_metadata_stream(stream
);
214 stream_pipe
= ctx
->consumer_metadata_pipe
;
216 consumer_add_data_stream(stream
);
217 stream_pipe
= ctx
->consumer_data_pipe
;
221 * From this point on, the stream's ownership has been moved away from
222 * the channel and it becomes globally visible. Hence, remove it from
223 * the local stream list to prevent the stream from being both local and
226 stream
->globally_visible
= 1;
227 cds_list_del(&stream
->send_node
);
229 ret
= lttng_pipe_write(stream_pipe
, &stream
, sizeof(stream
));
231 ERR("Consumer write %s stream to pipe %d",
232 stream
->metadata_flag
? "metadata" : "data",
233 lttng_pipe_get_writefd(stream_pipe
));
234 if (stream
->metadata_flag
) {
235 consumer_del_stream_for_metadata(stream
);
237 consumer_del_stream_for_data(stream
);
247 int get_stream_shm_path(char *stream_shm_path
, const char *shm_path
, int cpu
)
249 char cpu_nr
[INT_MAX_STR_LEN
]; /* int max len */
252 strncpy(stream_shm_path
, shm_path
, PATH_MAX
);
253 stream_shm_path
[PATH_MAX
- 1] = '\0';
254 ret
= snprintf(cpu_nr
, INT_MAX_STR_LEN
, "%i", cpu
);
259 strncat(stream_shm_path
, cpu_nr
,
260 PATH_MAX
- strlen(stream_shm_path
) - 1);
267 * Create streams for the given channel using liblttng-ust-ctl.
269 * Return 0 on success else a negative value.
271 static int create_ust_streams(struct lttng_consumer_channel
*channel
,
272 struct lttng_consumer_local_data
*ctx
,
273 uint64_t trace_archive_id
)
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
,
309 stream
->ustream
= ustream
;
311 * Store it so we can save multiple function calls afterwards since
312 * this value is used heavily in the stream threads. This is UST
313 * specific so this is why it's done after allocation.
315 stream
->wait_fd
= wait_fd
;
318 * Increment channel refcount since the channel reference has now been
319 * assigned in the allocation process above.
321 if (stream
->chan
->monitor
) {
322 uatomic_inc(&stream
->chan
->refcount
);
326 * Order is important this is why a list is used. On error, the caller
327 * should clean this list.
329 cds_list_add_tail(&stream
->send_node
, &channel
->streams
.head
);
331 ret
= ustctl_get_max_subbuf_size(stream
->ustream
,
332 &stream
->max_sb_size
);
334 ERR("ustctl_get_max_subbuf_size failed for stream %s",
339 /* Do actions once stream has been received. */
340 if (ctx
->on_recv_stream
) {
341 ret
= ctx
->on_recv_stream(stream
);
347 DBG("UST consumer add stream %s (key: %" PRIu64
") with relayd id %" PRIu64
,
348 stream
->name
, stream
->key
, stream
->relayd_stream_id
);
350 /* Set next CPU stream. */
351 channel
->streams
.count
= ++cpu
;
353 /* Keep stream reference when creating metadata. */
354 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
355 channel
->metadata_stream
= stream
;
356 if (channel
->monitor
) {
357 /* Set metadata poll pipe if we created one */
358 memcpy(stream
->ust_metadata_poll_pipe
,
360 sizeof(ust_metadata_pipe
));
373 * create_posix_shm is never called concurrently within a process.
376 int create_posix_shm(void)
378 char tmp_name
[NAME_MAX
];
381 ret
= snprintf(tmp_name
, NAME_MAX
, "/ust-shm-consumer-%d", getpid());
387 * Allocate shm, and immediately unlink its shm oject, keeping
388 * only the file descriptor as a reference to the object.
389 * We specifically do _not_ use the / at the beginning of the
390 * pathname so that some OS implementations can keep it local to
391 * the process (POSIX leaves this implementation-defined).
393 shmfd
= shm_open(tmp_name
, O_CREAT
| O_EXCL
| O_RDWR
, 0700);
398 ret
= shm_unlink(tmp_name
);
399 if (ret
< 0 && errno
!= ENOENT
) {
400 PERROR("shm_unlink");
401 goto error_shm_release
;
414 static int open_ust_stream_fd(struct lttng_consumer_channel
*channel
, int cpu
)
416 char shm_path
[PATH_MAX
];
419 if (!channel
->shm_path
[0]) {
420 return create_posix_shm();
422 ret
= get_stream_shm_path(shm_path
, channel
->shm_path
, cpu
);
426 return run_as_open(shm_path
,
427 O_RDWR
| O_CREAT
| O_EXCL
, S_IRUSR
| S_IWUSR
,
428 channel
->uid
, channel
->gid
);
435 * Create an UST channel with the given attributes and send it to the session
436 * daemon using the ust ctl API.
438 * Return 0 on success or else a negative value.
440 static int create_ust_channel(struct lttng_consumer_channel
*channel
,
441 struct ustctl_consumer_channel_attr
*attr
,
442 struct ustctl_consumer_channel
**ust_chanp
)
444 int ret
, nr_stream_fds
, i
, j
;
446 struct ustctl_consumer_channel
*ust_channel
;
452 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
453 "subbuf_size: %" PRIu64
", num_subbuf: %" PRIu64
", "
454 "switch_timer_interval: %u, read_timer_interval: %u, "
455 "output: %d, type: %d", attr
->overwrite
, attr
->subbuf_size
,
456 attr
->num_subbuf
, attr
->switch_timer_interval
,
457 attr
->read_timer_interval
, attr
->output
, attr
->type
);
459 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
)
462 nr_stream_fds
= ustctl_get_nr_stream_per_channel();
463 stream_fds
= zmalloc(nr_stream_fds
* sizeof(*stream_fds
));
468 for (i
= 0; i
< nr_stream_fds
; i
++) {
469 stream_fds
[i
] = open_ust_stream_fd(channel
, i
);
470 if (stream_fds
[i
] < 0) {
475 ust_channel
= ustctl_create_channel(attr
, stream_fds
, nr_stream_fds
);
480 channel
->nr_stream_fds
= nr_stream_fds
;
481 channel
->stream_fds
= stream_fds
;
482 *ust_chanp
= ust_channel
;
488 for (j
= i
- 1; j
>= 0; j
--) {
491 closeret
= close(stream_fds
[j
]);
495 if (channel
->shm_path
[0]) {
496 char shm_path
[PATH_MAX
];
498 closeret
= get_stream_shm_path(shm_path
,
499 channel
->shm_path
, j
);
501 ERR("Cannot get stream shm path");
503 closeret
= run_as_unlink(shm_path
,
504 channel
->uid
, channel
->gid
);
506 PERROR("unlink %s", shm_path
);
510 /* Try to rmdir all directories under shm_path root. */
511 if (channel
->root_shm_path
[0]) {
512 (void) run_as_rmdir_recursive(channel
->root_shm_path
,
513 channel
->uid
, channel
->gid
);
521 * Send a single given stream to the session daemon using the sock.
523 * Return 0 on success else a negative value.
525 static int send_sessiond_stream(int sock
, struct lttng_consumer_stream
*stream
)
532 DBG("UST consumer sending stream %" PRIu64
" to sessiond", stream
->key
);
534 /* Send stream to session daemon. */
535 ret
= ustctl_send_stream_to_sessiond(sock
, stream
->ustream
);
545 * Send channel to sessiond and relayd if applicable.
547 * Return 0 on success or else a negative value.
549 static int send_channel_to_sessiond_and_relayd(int sock
,
550 struct lttng_consumer_channel
*channel
,
551 struct lttng_consumer_local_data
*ctx
, int *relayd_error
)
553 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
554 struct lttng_consumer_stream
*stream
;
555 uint64_t net_seq_idx
= -1ULL;
561 DBG("UST consumer sending channel %s to sessiond", channel
->name
);
563 if (channel
->relayd_id
!= (uint64_t) -1ULL) {
564 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
566 health_code_update();
568 /* Try to send the stream to the relayd if one is available. */
569 DBG("Sending stream %" PRIu64
" of channel \"%s\" to relayd",
570 stream
->key
, channel
->name
);
571 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
,
647 struct lttng_consumer_channel
*channel
,
648 struct ustctl_consumer_channel_attr
*attr
,
649 uint64_t trace_archive_id
)
658 * This value is still used by the kernel consumer since for the kernel,
659 * the stream ownership is not IN the consumer so we need to have the
660 * number of left stream that needs to be initialized so we can know when
661 * to delete the channel (see consumer.c).
663 * As for the user space tracer now, the consumer creates and sends the
664 * stream to the session daemon which only sends them to the application
665 * once every stream of a channel is received making this value useless
666 * because we they will be added to the poll thread before the application
667 * receives them. This ensures that a stream can not hang up during
668 * initilization of a channel.
670 channel
->nb_init_stream_left
= 0;
672 /* The reply msg status is handled in the following call. */
673 ret
= create_ust_channel(channel
, attr
, &channel
->uchan
);
678 channel
->wait_fd
= ustctl_channel_get_wait_fd(channel
->uchan
);
681 * For the snapshots (no monitor), we create the metadata streams
682 * on demand, not during the channel creation.
684 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& !channel
->monitor
) {
689 /* Open all streams for this channel. */
690 ret
= create_ust_streams(channel
, ctx
, trace_archive_id
);
700 * Send all stream of a channel to the right thread handling it.
702 * On error, return a negative value else 0 on success.
704 static int send_streams_to_thread(struct lttng_consumer_channel
*channel
,
705 struct lttng_consumer_local_data
*ctx
)
708 struct lttng_consumer_stream
*stream
, *stmp
;
713 /* Send streams to the corresponding thread. */
714 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
717 health_code_update();
719 /* Sending the stream to the thread. */
720 ret
= send_stream_to_thread(stream
, ctx
);
723 * If we are unable to send the stream to the thread, there is
724 * a big problem so just stop everything.
735 * Flush channel's streams using the given key to retrieve the channel.
737 * Return 0 on success else an LTTng error code.
739 static int flush_channel(uint64_t chan_key
)
742 struct lttng_consumer_channel
*channel
;
743 struct lttng_consumer_stream
*stream
;
745 struct lttng_ht_iter iter
;
747 DBG("UST consumer flush channel key %" PRIu64
, chan_key
);
750 channel
= consumer_find_channel(chan_key
);
752 ERR("UST consumer flush channel %" PRIu64
" not found", chan_key
);
753 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
757 ht
= consumer_data
.stream_per_chan_id_ht
;
759 /* For each stream of the channel id, flush it. */
760 cds_lfht_for_each_entry_duplicate(ht
->ht
,
761 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
762 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
764 health_code_update();
766 pthread_mutex_lock(&stream
->lock
);
767 if (!stream
->quiescent
) {
768 ustctl_flush_buffer(stream
->ustream
, 0);
769 stream
->quiescent
= true;
771 pthread_mutex_unlock(&stream
->lock
);
779 * Clear quiescent state from channel's streams using the given key to
780 * retrieve the channel.
782 * Return 0 on success else an LTTng error code.
784 static int clear_quiescent_channel(uint64_t chan_key
)
787 struct lttng_consumer_channel
*channel
;
788 struct lttng_consumer_stream
*stream
;
790 struct lttng_ht_iter iter
;
792 DBG("UST consumer clear quiescent channel key %" PRIu64
, chan_key
);
795 channel
= consumer_find_channel(chan_key
);
797 ERR("UST consumer clear quiescent channel %" PRIu64
" not found", chan_key
);
798 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
802 ht
= consumer_data
.stream_per_chan_id_ht
;
804 /* For each stream of the channel id, clear quiescent state. */
805 cds_lfht_for_each_entry_duplicate(ht
->ht
,
806 ht
->hash_fct(&channel
->key
, lttng_ht_seed
), ht
->match_fct
,
807 &channel
->key
, &iter
.iter
, stream
, node_channel_id
.node
) {
809 health_code_update();
811 pthread_mutex_lock(&stream
->lock
);
812 stream
->quiescent
= false;
813 pthread_mutex_unlock(&stream
->lock
);
821 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
822 * RCU read side lock MUST be acquired before calling this function.
824 * Return 0 on success else an LTTng error code.
826 static int close_metadata(uint64_t chan_key
)
829 struct lttng_consumer_channel
*channel
;
830 unsigned int channel_monitor
;
832 DBG("UST consumer close metadata key %" PRIu64
, chan_key
);
834 channel
= consumer_find_channel(chan_key
);
837 * This is possible if the metadata thread has issue a delete because
838 * the endpoint point of the stream hung up. There is no way the
839 * session daemon can know about it thus use a DBG instead of an actual
842 DBG("UST consumer close metadata %" PRIu64
" not found", chan_key
);
843 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
847 pthread_mutex_lock(&consumer_data
.lock
);
848 pthread_mutex_lock(&channel
->lock
);
849 channel_monitor
= channel
->monitor
;
850 if (cds_lfht_is_node_deleted(&channel
->node
.node
)) {
854 lttng_ustconsumer_close_metadata(channel
);
855 pthread_mutex_unlock(&channel
->lock
);
856 pthread_mutex_unlock(&consumer_data
.lock
);
859 * The ownership of a metadata channel depends on the type of
860 * session to which it belongs. In effect, the monitor flag is checked
861 * to determine if this metadata channel is in "snapshot" mode or not.
863 * In the non-snapshot case, the metadata channel is created along with
864 * a single stream which will remain present until the metadata channel
865 * is destroyed (on the destruction of its session). In this case, the
866 * metadata stream in "monitored" by the metadata poll thread and holds
867 * the ownership of its channel.
869 * Closing the metadata will cause the metadata stream's "metadata poll
870 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
871 * thread which will teardown the metadata stream which, in return,
872 * deletes the metadata channel.
874 * In the snapshot case, the metadata stream is created and destroyed
875 * on every snapshot record. Since the channel doesn't have an owner
876 * other than the session daemon, it is safe to destroy it immediately
877 * on reception of the CLOSE_METADATA command.
879 if (!channel_monitor
) {
881 * The channel and consumer_data locks must be
882 * released before this call since consumer_del_channel
883 * re-acquires the channel and consumer_data locks to teardown
884 * the channel and queue its reclamation by the "call_rcu"
887 consumer_del_channel(channel
);
892 pthread_mutex_unlock(&channel
->lock
);
893 pthread_mutex_unlock(&consumer_data
.lock
);
899 * RCU read side lock MUST be acquired before calling this function.
901 * Return 0 on success else an LTTng error code.
903 static int setup_metadata(struct lttng_consumer_local_data
*ctx
, uint64_t key
)
906 struct lttng_consumer_channel
*metadata
;
908 DBG("UST consumer setup metadata key %" PRIu64
, key
);
910 metadata
= consumer_find_channel(key
);
912 ERR("UST consumer push metadata %" PRIu64
" not found", key
);
913 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
918 * In no monitor mode, the metadata channel has no stream(s) so skip the
919 * ownership transfer to the metadata thread.
921 if (!metadata
->monitor
) {
922 DBG("Metadata channel in no monitor");
928 * Send metadata stream to relayd if one available. Availability is
929 * known if the stream is still in the list of the channel.
931 if (cds_list_empty(&metadata
->streams
.head
)) {
932 ERR("Metadata channel key %" PRIu64
", no stream available.", key
);
933 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
934 goto error_no_stream
;
937 /* Send metadata stream to relayd if needed. */
938 if (metadata
->metadata_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
939 ret
= consumer_send_relayd_stream(metadata
->metadata_stream
,
942 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
945 ret
= consumer_send_relayd_streams_sent(
946 metadata
->metadata_stream
->net_seq_idx
);
948 ret
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
954 * Ownership of metadata stream is passed along. Freeing is handled by
957 ret
= send_streams_to_thread(metadata
, ctx
);
960 * If we are unable to send the stream to the thread, there is
961 * a big problem so just stop everything.
963 ret
= LTTCOMM_CONSUMERD_FATAL
;
964 goto send_streams_error
;
966 /* List MUST be empty after or else it could be reused. */
967 assert(cds_list_empty(&metadata
->streams
.head
));
974 * Delete metadata channel on error. At this point, the metadata stream can
975 * NOT be monitored by the metadata thread thus having the guarantee that
976 * the stream is still in the local stream list of the channel. This call
977 * will make sure to clean that list.
979 consumer_stream_destroy(metadata
->metadata_stream
, NULL
);
980 cds_list_del(&metadata
->metadata_stream
->send_node
);
981 metadata
->metadata_stream
= NULL
;
989 * Snapshot the whole metadata.
991 * Returns 0 on success, < 0 on error
993 static int snapshot_metadata(uint64_t key
, char *path
, uint64_t relayd_id
,
994 struct lttng_consumer_local_data
*ctx
,
995 uint64_t trace_archive_id
)
998 struct lttng_consumer_channel
*metadata_channel
;
999 struct lttng_consumer_stream
*metadata_stream
;
1004 DBG("UST consumer snapshot metadata with key %" PRIu64
" at path %s",
1009 metadata_channel
= consumer_find_channel(key
);
1010 if (!metadata_channel
) {
1011 ERR("UST snapshot metadata channel not found for key %" PRIu64
,
1016 assert(!metadata_channel
->monitor
);
1018 health_code_update();
1021 * Ask the sessiond if we have new metadata waiting and update the
1022 * consumer metadata cache.
1024 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 1);
1029 health_code_update();
1032 * The metadata stream is NOT created in no monitor mode when the channel
1033 * is created on a sessiond ask channel command.
1035 ret
= create_ust_streams(metadata_channel
, ctx
, trace_archive_id
);
1040 metadata_stream
= metadata_channel
->metadata_stream
;
1041 assert(metadata_stream
);
1043 if (relayd_id
!= (uint64_t) -1ULL) {
1044 metadata_stream
->net_seq_idx
= relayd_id
;
1045 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
1050 ret
= utils_create_stream_file(path
, metadata_stream
->name
,
1051 metadata_stream
->chan
->tracefile_size
,
1052 metadata_stream
->tracefile_count_current
,
1053 metadata_stream
->uid
, metadata_stream
->gid
, NULL
);
1057 metadata_stream
->out_fd
= ret
;
1058 metadata_stream
->tracefile_size_current
= 0;
1062 health_code_update();
1064 ret
= lttng_consumer_read_subbuffer(metadata_stream
, ctx
);
1072 * Clean up the stream completly because the next snapshot will use a new
1075 consumer_stream_destroy(metadata_stream
, NULL
);
1076 cds_list_del(&metadata_stream
->send_node
);
1077 metadata_channel
->metadata_stream
= NULL
;
1085 * Take a snapshot of all the stream of a channel.
1087 * Returns 0 on success, < 0 on error
1089 static int snapshot_channel(uint64_t key
, char *path
, uint64_t relayd_id
,
1090 uint64_t nb_packets_per_stream
,
1091 struct lttng_consumer_local_data
*ctx
)
1094 unsigned use_relayd
= 0;
1095 unsigned long consumed_pos
, produced_pos
;
1096 struct lttng_consumer_channel
*channel
;
1097 struct lttng_consumer_stream
*stream
;
1104 if (relayd_id
!= (uint64_t) -1ULL) {
1108 channel
= consumer_find_channel(key
);
1110 ERR("UST snapshot channel not found for key %" PRIu64
, key
);
1114 assert(!channel
->monitor
);
1115 DBG("UST consumer snapshot channel %" PRIu64
, key
);
1117 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
1118 health_code_update();
1120 /* Lock stream because we are about to change its state. */
1121 pthread_mutex_lock(&stream
->lock
);
1122 stream
->net_seq_idx
= relayd_id
;
1125 ret
= consumer_send_relayd_stream(stream
, path
);
1130 ret
= utils_create_stream_file(path
, stream
->name
,
1131 stream
->chan
->tracefile_size
,
1132 stream
->tracefile_count_current
,
1133 stream
->uid
, stream
->gid
, NULL
);
1137 stream
->out_fd
= ret
;
1138 stream
->tracefile_size_current
= 0;
1140 DBG("UST consumer snapshot stream %s/%s (%" PRIu64
")", path
,
1141 stream
->name
, stream
->key
);
1145 * If tracing is active, we want to perform a "full" buffer flush.
1146 * Else, if quiescent, it has already been done by the prior stop.
1148 if (!stream
->quiescent
) {
1149 ustctl_flush_buffer(stream
->ustream
, 0);
1152 ret
= lttng_ustconsumer_take_snapshot(stream
);
1154 ERR("Taking UST snapshot");
1158 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
1160 ERR("Produced UST snapshot position");
1164 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1166 ERR("Consumerd UST snapshot position");
1171 * The original value is sent back if max stream size is larger than
1172 * the possible size of the snapshot. Also, we assume that the session
1173 * daemon should never send a maximum stream size that is lower than
1176 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
1177 produced_pos
, nb_packets_per_stream
,
1178 stream
->max_sb_size
);
1180 while (consumed_pos
< produced_pos
) {
1182 unsigned long len
, padded_len
;
1184 health_code_update();
1186 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1188 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1190 if (ret
!= -EAGAIN
) {
1191 PERROR("ustctl_get_subbuf snapshot");
1192 goto error_close_stream
;
1194 DBG("UST consumer get subbuf failed. Skipping it.");
1195 consumed_pos
+= stream
->max_sb_size
;
1196 stream
->chan
->lost_packets
++;
1200 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1202 ERR("Snapshot ustctl_get_subbuf_size");
1203 goto error_put_subbuf
;
1206 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1208 ERR("Snapshot ustctl_get_padded_subbuf_size");
1209 goto error_put_subbuf
;
1212 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, len
,
1213 padded_len
- len
, NULL
);
1215 if (read_len
!= len
) {
1217 goto error_put_subbuf
;
1220 if (read_len
!= padded_len
) {
1222 goto error_put_subbuf
;
1226 ret
= ustctl_put_subbuf(stream
->ustream
);
1228 ERR("Snapshot ustctl_put_subbuf");
1229 goto error_close_stream
;
1231 consumed_pos
+= stream
->max_sb_size
;
1234 /* Simply close the stream so we can use it on the next snapshot. */
1235 consumer_stream_close(stream
);
1236 pthread_mutex_unlock(&stream
->lock
);
1243 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1244 ERR("Snapshot ustctl_put_subbuf");
1247 consumer_stream_close(stream
);
1249 pthread_mutex_unlock(&stream
->lock
);
1256 * Receive the metadata updates from the sessiond. Supports receiving
1257 * overlapping metadata, but is needs to always belong to a contiguous
1258 * range starting from 0.
1259 * Be careful about the locks held when calling this function: it needs
1260 * the metadata cache flush to concurrently progress in order to
1263 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1264 uint64_t len
, uint64_t version
,
1265 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
1267 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1270 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1272 metadata_str
= zmalloc(len
* sizeof(char));
1273 if (!metadata_str
) {
1274 PERROR("zmalloc metadata string");
1275 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1279 health_code_update();
1281 /* Receive metadata string. */
1282 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1284 /* Session daemon is dead so return gracefully. */
1289 health_code_update();
1291 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1292 ret
= consumer_metadata_cache_write(channel
, offset
, len
, version
,
1295 /* Unable to handle metadata. Notify session daemon. */
1296 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1298 * Skip metadata flush on write error since the offset and len might
1299 * not have been updated which could create an infinite loop below when
1300 * waiting for the metadata cache to be flushed.
1302 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1305 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1310 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1311 DBG("Waiting for metadata to be flushed");
1313 health_code_update();
1315 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1325 * Receive command from session daemon and process it.
1327 * Return 1 on success else a negative value or 0.
1329 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1330 int sock
, struct pollfd
*consumer_sockpoll
)
1333 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1334 struct lttcomm_consumer_msg msg
;
1335 struct lttng_consumer_channel
*channel
= NULL
;
1337 health_code_update();
1339 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1340 if (ret
!= sizeof(msg
)) {
1341 DBG("Consumer received unexpected message size %zd (expects %zu)",
1344 * The ret value might 0 meaning an orderly shutdown but this is ok
1345 * since the caller handles this.
1348 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1354 health_code_update();
1357 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1359 health_code_update();
1361 /* relayd needs RCU read-side lock */
1364 switch (msg
.cmd_type
) {
1365 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1367 /* Session daemon status message are handled in the following call. */
1368 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1369 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
1370 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
1371 msg
.u
.relayd_sock
.relayd_session_id
);
1374 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1376 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1377 struct consumer_relayd_sock_pair
*relayd
;
1379 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1381 /* Get relayd reference if exists. */
1382 relayd
= consumer_find_relayd(index
);
1383 if (relayd
== NULL
) {
1384 DBG("Unable to find relayd %" PRIu64
, index
);
1385 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1389 * Each relayd socket pair has a refcount of stream attached to it
1390 * which tells if the relayd is still active or not depending on the
1393 * This will set the destroy flag of the relayd object and destroy it
1394 * if the refcount reaches zero when called.
1396 * The destroy can happen either here or when a stream fd hangs up.
1399 consumer_flag_relayd_for_destroy(relayd
);
1402 goto end_msg_sessiond
;
1404 case LTTNG_CONSUMER_UPDATE_STREAM
:
1409 case LTTNG_CONSUMER_DATA_PENDING
:
1411 int ret
, is_data_pending
;
1412 uint64_t id
= msg
.u
.data_pending
.session_id
;
1414 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1416 is_data_pending
= consumer_data_pending(id
);
1418 /* Send back returned value to session daemon */
1419 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1420 sizeof(is_data_pending
));
1422 DBG("Error when sending the data pending ret code: %d", ret
);
1427 * No need to send back a status message since the data pending
1428 * returned value is the response.
1432 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1435 struct ustctl_consumer_channel_attr attr
;
1437 /* Create a plain object and reserve a channel key. */
1438 channel
= allocate_channel(msg
.u
.ask_channel
.session_id
,
1439 msg
.u
.ask_channel
.pathname
, msg
.u
.ask_channel
.name
,
1440 msg
.u
.ask_channel
.uid
, msg
.u
.ask_channel
.gid
,
1441 msg
.u
.ask_channel
.relayd_id
, msg
.u
.ask_channel
.key
,
1442 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1443 msg
.u
.ask_channel
.tracefile_size
,
1444 msg
.u
.ask_channel
.tracefile_count
,
1445 msg
.u
.ask_channel
.session_id_per_pid
,
1446 msg
.u
.ask_channel
.monitor
,
1447 msg
.u
.ask_channel
.live_timer_interval
,
1448 msg
.u
.ask_channel
.root_shm_path
,
1449 msg
.u
.ask_channel
.shm_path
);
1451 goto end_channel_error
;
1455 * Assign UST application UID to the channel. This value is ignored for
1456 * per PID buffers. This is specific to UST thus setting this after the
1459 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1461 /* Build channel attributes from received message. */
1462 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1463 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1464 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1465 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1466 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1467 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1468 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1469 attr
.blocking_timeout
= msg
.u
.ask_channel
.blocking_timeout
;
1471 /* Match channel buffer type to the UST abi. */
1472 switch (msg
.u
.ask_channel
.output
) {
1473 case LTTNG_EVENT_MMAP
:
1475 attr
.output
= LTTNG_UST_MMAP
;
1479 /* Translate and save channel type. */
1480 switch (msg
.u
.ask_channel
.type
) {
1481 case LTTNG_UST_CHAN_PER_CPU
:
1482 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1483 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1485 * Set refcount to 1 for owner. Below, we will
1486 * pass ownership to the
1487 * consumer_thread_channel_poll() thread.
1489 channel
->refcount
= 1;
1491 case LTTNG_UST_CHAN_METADATA
:
1492 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1493 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1500 health_code_update();
1502 ret
= ask_channel(ctx
, channel
, &attr
,
1503 msg
.u
.ask_channel
.trace_archive_id
);
1505 goto end_channel_error
;
1508 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1509 ret
= consumer_metadata_cache_allocate(channel
);
1511 ERR("Allocating metadata cache");
1512 goto end_channel_error
;
1514 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1515 attr
.switch_timer_interval
= 0;
1517 int monitor_start_ret
;
1519 consumer_timer_live_start(channel
,
1520 msg
.u
.ask_channel
.live_timer_interval
);
1521 monitor_start_ret
= consumer_timer_monitor_start(
1523 msg
.u
.ask_channel
.monitor_timer_interval
);
1524 if (monitor_start_ret
< 0) {
1525 ERR("Starting channel monitoring timer failed");
1526 goto end_channel_error
;
1530 health_code_update();
1533 * Add the channel to the internal state AFTER all streams were created
1534 * and successfully sent to session daemon. This way, all streams must
1535 * be ready before this channel is visible to the threads.
1536 * If add_channel succeeds, ownership of the channel is
1537 * passed to consumer_thread_channel_poll().
1539 ret
= add_channel(channel
, ctx
);
1541 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1542 if (channel
->switch_timer_enabled
== 1) {
1543 consumer_timer_switch_stop(channel
);
1545 consumer_metadata_cache_destroy(channel
);
1547 if (channel
->live_timer_enabled
== 1) {
1548 consumer_timer_live_stop(channel
);
1550 if (channel
->monitor_timer_enabled
== 1) {
1551 consumer_timer_monitor_stop(channel
);
1553 goto end_channel_error
;
1556 health_code_update();
1559 * Channel and streams are now created. Inform the session daemon that
1560 * everything went well and should wait to receive the channel and
1561 * streams with ustctl API.
1563 ret
= consumer_send_status_channel(sock
, channel
);
1566 * There is probably a problem on the socket.
1573 case LTTNG_CONSUMER_GET_CHANNEL
:
1575 int ret
, relayd_err
= 0;
1576 uint64_t key
= msg
.u
.get_channel
.key
;
1577 struct lttng_consumer_channel
*channel
;
1579 channel
= consumer_find_channel(key
);
1581 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1582 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1583 goto end_msg_sessiond
;
1586 health_code_update();
1588 /* Send the channel to sessiond (and relayd, if applicable). */
1589 ret
= send_channel_to_sessiond_and_relayd(sock
, channel
, ctx
,
1594 * We were unable to send to the relayd the stream so avoid
1595 * sending back a fatal error to the thread since this is OK
1596 * and the consumer can continue its work. The above call
1597 * has sent the error status message to the sessiond.
1602 * The communicaton was broken hence there is a bad state between
1603 * the consumer and sessiond so stop everything.
1608 health_code_update();
1611 * In no monitor mode, the streams ownership is kept inside the channel
1612 * so don't send them to the data thread.
1614 if (!channel
->monitor
) {
1615 goto end_msg_sessiond
;
1618 ret
= send_streams_to_thread(channel
, ctx
);
1621 * If we are unable to send the stream to the thread, there is
1622 * a big problem so just stop everything.
1626 /* List MUST be empty after or else it could be reused. */
1627 assert(cds_list_empty(&channel
->streams
.head
));
1628 goto end_msg_sessiond
;
1630 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1632 uint64_t key
= msg
.u
.destroy_channel
.key
;
1635 * Only called if streams have not been sent to stream
1636 * manager thread. However, channel has been sent to
1637 * channel manager thread.
1639 notify_thread_del_channel(ctx
, key
);
1640 goto end_msg_sessiond
;
1642 case LTTNG_CONSUMER_CLOSE_METADATA
:
1646 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1651 goto end_msg_sessiond
;
1653 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1657 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1662 goto end_msg_sessiond
;
1664 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL
:
1668 ret
= clear_quiescent_channel(
1669 msg
.u
.clear_quiescent_channel
.key
);
1674 goto end_msg_sessiond
;
1676 case LTTNG_CONSUMER_PUSH_METADATA
:
1679 uint64_t len
= msg
.u
.push_metadata
.len
;
1680 uint64_t key
= msg
.u
.push_metadata
.key
;
1681 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1682 uint64_t version
= msg
.u
.push_metadata
.version
;
1683 struct lttng_consumer_channel
*channel
;
1685 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1688 channel
= consumer_find_channel(key
);
1691 * This is possible if the metadata creation on the consumer side
1692 * is in flight vis-a-vis a concurrent push metadata from the
1693 * session daemon. Simply return that the channel failed and the
1694 * session daemon will handle that message correctly considering
1695 * that this race is acceptable thus the DBG() statement here.
1697 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1698 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1699 goto end_msg_sessiond
;
1702 health_code_update();
1706 * There is nothing to receive. We have simply
1707 * checked whether the channel can be found.
1709 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1710 goto end_msg_sessiond
;
1713 /* Tell session daemon we are ready to receive the metadata. */
1714 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1716 /* Somehow, the session daemon is not responding anymore. */
1720 health_code_update();
1722 /* Wait for more data. */
1723 health_poll_entry();
1724 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1730 health_code_update();
1732 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1733 len
, version
, channel
, 0, 1);
1735 /* error receiving from sessiond */
1739 goto end_msg_sessiond
;
1742 case LTTNG_CONSUMER_SETUP_METADATA
:
1746 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1750 goto end_msg_sessiond
;
1752 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1754 if (msg
.u
.snapshot_channel
.metadata
) {
1755 ret
= snapshot_metadata(msg
.u
.snapshot_channel
.key
,
1756 msg
.u
.snapshot_channel
.pathname
,
1757 msg
.u
.snapshot_channel
.relayd_id
,
1759 msg
.u
.snapshot_channel
.trace_archive_id
);
1761 ERR("Snapshot metadata failed");
1762 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1765 ret
= snapshot_channel(msg
.u
.snapshot_channel
.key
,
1766 msg
.u
.snapshot_channel
.pathname
,
1767 msg
.u
.snapshot_channel
.relayd_id
,
1768 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
1771 ERR("Snapshot channel failed");
1772 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1776 health_code_update();
1777 ret
= consumer_send_status_msg(sock
, ret_code
);
1779 /* Somehow, the session daemon is not responding anymore. */
1782 health_code_update();
1785 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
1788 uint64_t discarded_events
;
1789 struct lttng_ht_iter iter
;
1790 struct lttng_ht
*ht
;
1791 struct lttng_consumer_stream
*stream
;
1792 uint64_t id
= msg
.u
.discarded_events
.session_id
;
1793 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
1795 DBG("UST consumer discarded events command for session id %"
1798 pthread_mutex_lock(&consumer_data
.lock
);
1800 ht
= consumer_data
.stream_list_ht
;
1803 * We only need a reference to the channel, but they are not
1804 * directly indexed, so we just use the first matching stream
1805 * to extract the information we need, we default to 0 if not
1806 * found (no events are dropped if the channel is not yet in
1809 discarded_events
= 0;
1810 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1811 ht
->hash_fct(&id
, lttng_ht_seed
),
1813 &iter
.iter
, stream
, node_session_id
.node
) {
1814 if (stream
->chan
->key
== key
) {
1815 discarded_events
= stream
->chan
->discarded_events
;
1819 pthread_mutex_unlock(&consumer_data
.lock
);
1822 DBG("UST consumer discarded events command for session id %"
1823 PRIu64
", channel key %" PRIu64
, id
, key
);
1825 health_code_update();
1827 /* Send back returned value to session daemon */
1828 ret
= lttcomm_send_unix_sock(sock
, &discarded_events
, sizeof(discarded_events
));
1830 PERROR("send discarded events");
1836 case LTTNG_CONSUMER_LOST_PACKETS
:
1839 uint64_t lost_packets
;
1840 struct lttng_ht_iter iter
;
1841 struct lttng_ht
*ht
;
1842 struct lttng_consumer_stream
*stream
;
1843 uint64_t id
= msg
.u
.lost_packets
.session_id
;
1844 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
1846 DBG("UST consumer lost packets command for session id %"
1849 pthread_mutex_lock(&consumer_data
.lock
);
1851 ht
= consumer_data
.stream_list_ht
;
1854 * We only need a reference to the channel, but they are not
1855 * directly indexed, so we just use the first matching stream
1856 * to extract the information we need, we default to 0 if not
1857 * found (no packets lost if the channel is not yet in use).
1860 cds_lfht_for_each_entry_duplicate(ht
->ht
,
1861 ht
->hash_fct(&id
, lttng_ht_seed
),
1863 &iter
.iter
, stream
, node_session_id
.node
) {
1864 if (stream
->chan
->key
== key
) {
1865 lost_packets
= stream
->chan
->lost_packets
;
1869 pthread_mutex_unlock(&consumer_data
.lock
);
1872 DBG("UST consumer lost packets command for session id %"
1873 PRIu64
", channel key %" PRIu64
, id
, key
);
1875 health_code_update();
1877 /* Send back returned value to session daemon */
1878 ret
= lttcomm_send_unix_sock(sock
, &lost_packets
,
1879 sizeof(lost_packets
));
1881 PERROR("send lost packets");
1887 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
1889 int channel_monitor_pipe
;
1891 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1892 /* Successfully received the command's type. */
1893 ret
= consumer_send_status_msg(sock
, ret_code
);
1898 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
1900 if (ret
!= sizeof(channel_monitor_pipe
)) {
1901 ERR("Failed to receive channel monitor pipe");
1905 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
1906 ret
= consumer_timer_thread_set_channel_monitor_pipe(
1907 channel_monitor_pipe
);
1911 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1912 /* Set the pipe as non-blocking. */
1913 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
1915 PERROR("fcntl get flags of the channel monitoring pipe");
1920 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
1921 flags
| O_NONBLOCK
);
1923 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1926 DBG("Channel monitor pipe set as non-blocking");
1928 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
1930 goto end_msg_sessiond
;
1932 case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE
:
1934 int channel_rotate_pipe
;
1937 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1938 /* Successfully received the command's type. */
1939 ret
= consumer_send_status_msg(sock
, ret_code
);
1944 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_rotate_pipe
, 1);
1945 if (ret
!= sizeof(channel_rotate_pipe
)) {
1946 ERR("Failed to receive channel rotate pipe");
1950 DBG("Received channel rotate pipe (%d)", channel_rotate_pipe
);
1951 ctx
->channel_rotate_pipe
= channel_rotate_pipe
;
1952 /* Set the pipe as non-blocking. */
1953 ret
= fcntl(channel_rotate_pipe
, F_GETFL
, 0);
1955 PERROR("fcntl get flags of the channel rotate pipe");
1960 ret
= fcntl(channel_rotate_pipe
, F_SETFL
, flags
| O_NONBLOCK
);
1962 PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
1965 DBG("Channel rotate pipe set as non-blocking");
1966 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1967 ret
= consumer_send_status_msg(sock
, ret_code
);
1973 case LTTNG_CONSUMER_ROTATE_CHANNEL
:
1976 * Sample the rotate position of all the streams in this channel.
1978 ret
= lttng_consumer_rotate_channel(msg
.u
.rotate_channel
.key
,
1979 msg
.u
.rotate_channel
.pathname
,
1980 msg
.u
.rotate_channel
.relayd_id
,
1981 msg
.u
.rotate_channel
.metadata
,
1982 msg
.u
.rotate_channel
.new_chunk_id
,
1985 ERR("Rotate channel failed");
1986 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1989 health_code_update();
1991 ret
= consumer_send_status_msg(sock
, ret_code
);
1993 /* Somehow, the session daemon is not responding anymore. */
1998 * Rotate the streams that are ready right now.
1999 * FIXME: this is a second consecutive iteration over the
2000 * streams in a channel, there is probably a better way to
2001 * handle this, but it needs to be after the
2002 * consumer_send_status_msg() call.
2004 ret
= lttng_consumer_rotate_ready_streams(
2005 msg
.u
.rotate_channel
.key
, ctx
);
2007 ERR("Rotate channel failed");
2008 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
2012 case LTTNG_CONSUMER_ROTATE_RENAME
:
2014 DBG("Consumer rename session %" PRIu64
" after rotation",
2015 msg
.u
.rotate_rename
.session_id
);
2016 ret
= lttng_consumer_rotate_rename(msg
.u
.rotate_rename
.old_path
,
2017 msg
.u
.rotate_rename
.new_path
,
2018 msg
.u
.rotate_rename
.uid
,
2019 msg
.u
.rotate_rename
.gid
,
2020 msg
.u
.rotate_rename
.relayd_id
);
2022 ERR("Rotate rename failed");
2023 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
2026 health_code_update();
2028 ret
= consumer_send_status_msg(sock
, ret_code
);
2030 /* Somehow, the session daemon is not responding anymore. */
2035 case LTTNG_CONSUMER_ROTATE_PENDING_RELAY
:
2039 DBG("Consumer rotate pending on relay for session %" PRIu64
,
2040 msg
.u
.rotate_pending_relay
.session_id
);
2041 pending
= lttng_consumer_rotate_pending_relay(
2042 msg
.u
.rotate_pending_relay
.session_id
,
2043 msg
.u
.rotate_pending_relay
.relayd_id
,
2044 msg
.u
.rotate_pending_relay
.chunk_id
);
2046 ERR("Rotate pending relay failed");
2047 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
2050 health_code_update();
2052 ret
= consumer_send_status_msg(sock
, ret_code
);
2054 /* Somehow, the session daemon is not responding anymore. */
2058 /* Send back returned value to session daemon */
2059 ret
= lttcomm_send_unix_sock(sock
, &pending
, sizeof(pending
));
2061 PERROR("send data pending ret code");
2066 case LTTNG_CONSUMER_MKDIR
:
2068 DBG("Consumer mkdir %s in session %" PRIu64
,
2070 msg
.u
.mkdir
.session_id
);
2071 ret
= lttng_consumer_mkdir(msg
.u
.mkdir
.path
,
2074 msg
.u
.mkdir
.relayd_id
);
2076 ERR("consumer mkdir failed");
2077 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
2080 health_code_update();
2082 ret
= consumer_send_status_msg(sock
, ret_code
);
2084 /* Somehow, the session daemon is not responding anymore. */
2096 health_code_update();
2099 * Return 1 to indicate success since the 0 value can be a socket
2100 * shutdown during the recv() or send() call.
2106 * The returned value here is not useful since either way we'll return 1 to
2107 * the caller because the session daemon socket management is done
2108 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2110 ret
= consumer_send_status_msg(sock
, ret_code
);
2116 health_code_update();
2122 * Free channel here since no one has a reference to it. We don't
2123 * free after that because a stream can store this pointer.
2125 destroy_channel(channel
);
2127 /* We have to send a status channel message indicating an error. */
2128 ret
= consumer_send_status_channel(sock
, NULL
);
2130 /* Stop everything if session daemon can not be notified. */
2135 health_code_update();
2140 /* This will issue a consumer stop. */
2145 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2146 * compiled out, we isolate it in this library.
2148 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream
*stream
,
2152 assert(stream
->ustream
);
2154 return ustctl_get_mmap_read_offset(stream
->ustream
, off
);
2158 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2159 * compiled out, we isolate it in this library.
2161 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream
*stream
)
2164 assert(stream
->ustream
);
2166 return ustctl_get_mmap_base(stream
->ustream
);
2169 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream
*stream
,
2170 int producer_active
)
2173 assert(stream
->ustream
);
2175 ustctl_flush_buffer(stream
->ustream
, producer_active
);
2179 * Take a snapshot for a specific stream.
2181 * Returns 0 on success, < 0 on error
2183 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2186 assert(stream
->ustream
);
2188 return ustctl_snapshot(stream
->ustream
);
2192 * Sample consumed and produced positions for a specific stream.
2194 * Returns 0 on success, < 0 on error.
2196 int lttng_ustconsumer_sample_snapshot_positions(
2197 struct lttng_consumer_stream
*stream
)
2200 assert(stream
->ustream
);
2202 return ustctl_snapshot_sample_positions(stream
->ustream
);
2206 * Get the produced position
2208 * Returns 0 on success, < 0 on error
2210 int lttng_ustconsumer_get_produced_snapshot(
2211 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2214 assert(stream
->ustream
);
2217 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
2221 * Get the consumed position
2223 * Returns 0 on success, < 0 on error
2225 int lttng_ustconsumer_get_consumed_snapshot(
2226 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
2229 assert(stream
->ustream
);
2232 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
2235 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
2239 assert(stream
->ustream
);
2241 ustctl_flush_buffer(stream
->ustream
, producer
);
2244 int lttng_ustconsumer_get_current_timestamp(
2245 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
2248 assert(stream
->ustream
);
2251 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
2254 int lttng_ustconsumer_get_sequence_number(
2255 struct lttng_consumer_stream
*stream
, uint64_t *seq
)
2258 assert(stream
->ustream
);
2261 return ustctl_get_sequence_number(stream
->ustream
, seq
);
2265 * Called when the stream signals the consumer that it has hung up.
2267 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
2270 assert(stream
->ustream
);
2272 pthread_mutex_lock(&stream
->lock
);
2273 if (!stream
->quiescent
) {
2274 ustctl_flush_buffer(stream
->ustream
, 0);
2275 stream
->quiescent
= true;
2277 pthread_mutex_unlock(&stream
->lock
);
2278 stream
->hangup_flush_done
= 1;
2281 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
2286 assert(chan
->uchan
);
2288 if (chan
->switch_timer_enabled
== 1) {
2289 consumer_timer_switch_stop(chan
);
2291 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
2294 ret
= close(chan
->stream_fds
[i
]);
2298 if (chan
->shm_path
[0]) {
2299 char shm_path
[PATH_MAX
];
2301 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
2303 ERR("Cannot get stream shm path");
2305 ret
= run_as_unlink(shm_path
, chan
->uid
, chan
->gid
);
2307 PERROR("unlink %s", shm_path
);
2313 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
2316 assert(chan
->uchan
);
2318 consumer_metadata_cache_destroy(chan
);
2319 ustctl_destroy_channel(chan
->uchan
);
2320 /* Try to rmdir all directories under shm_path root. */
2321 if (chan
->root_shm_path
[0]) {
2322 (void) run_as_rmdir_recursive(chan
->root_shm_path
,
2323 chan
->uid
, chan
->gid
);
2325 free(chan
->stream_fds
);
2328 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
2331 assert(stream
->ustream
);
2333 if (stream
->chan
->switch_timer_enabled
== 1) {
2334 consumer_timer_switch_stop(stream
->chan
);
2336 ustctl_destroy_stream(stream
->ustream
);
2339 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
2342 assert(stream
->ustream
);
2344 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
2347 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
2350 assert(stream
->ustream
);
2352 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
2356 * Populate index values of a UST stream. Values are set in big endian order.
2358 * Return 0 on success or else a negative value.
2360 static int get_index_values(struct ctf_packet_index
*index
,
2361 struct ustctl_consumer_stream
*ustream
)
2365 ret
= ustctl_get_timestamp_begin(ustream
, &index
->timestamp_begin
);
2367 PERROR("ustctl_get_timestamp_begin");
2370 index
->timestamp_begin
= htobe64(index
->timestamp_begin
);
2372 ret
= ustctl_get_timestamp_end(ustream
, &index
->timestamp_end
);
2374 PERROR("ustctl_get_timestamp_end");
2377 index
->timestamp_end
= htobe64(index
->timestamp_end
);
2379 ret
= ustctl_get_events_discarded(ustream
, &index
->events_discarded
);
2381 PERROR("ustctl_get_events_discarded");
2384 index
->events_discarded
= htobe64(index
->events_discarded
);
2386 ret
= ustctl_get_content_size(ustream
, &index
->content_size
);
2388 PERROR("ustctl_get_content_size");
2391 index
->content_size
= htobe64(index
->content_size
);
2393 ret
= ustctl_get_packet_size(ustream
, &index
->packet_size
);
2395 PERROR("ustctl_get_packet_size");
2398 index
->packet_size
= htobe64(index
->packet_size
);
2400 ret
= ustctl_get_stream_id(ustream
, &index
->stream_id
);
2402 PERROR("ustctl_get_stream_id");
2405 index
->stream_id
= htobe64(index
->stream_id
);
2407 ret
= ustctl_get_instance_id(ustream
, &index
->stream_instance_id
);
2409 PERROR("ustctl_get_instance_id");
2412 index
->stream_instance_id
= htobe64(index
->stream_instance_id
);
2414 ret
= ustctl_get_sequence_number(ustream
, &index
->packet_seq_num
);
2416 PERROR("ustctl_get_sequence_number");
2419 index
->packet_seq_num
= htobe64(index
->packet_seq_num
);
2426 void metadata_stream_reset_cache(struct lttng_consumer_stream
*stream
,
2427 struct consumer_metadata_cache
*cache
)
2429 DBG("Metadata stream update to version %" PRIu64
,
2431 stream
->ust_metadata_pushed
= 0;
2432 stream
->metadata_version
= cache
->version
;
2433 stream
->reset_metadata_flag
= 1;
2437 * Check if the version of the metadata stream and metadata cache match.
2438 * If the cache got updated, reset the metadata stream.
2439 * The stream lock and metadata cache lock MUST be held.
2440 * Return 0 on success, a negative value on error.
2443 int metadata_stream_check_version(struct lttng_consumer_stream
*stream
)
2446 struct consumer_metadata_cache
*cache
= stream
->chan
->metadata_cache
;
2448 if (cache
->version
== stream
->metadata_version
) {
2451 metadata_stream_reset_cache(stream
, cache
);
2458 * Write up to one packet from the metadata cache to the channel.
2460 * Returns the number of bytes pushed in the cache, or a negative value
2464 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
2469 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2470 ret
= metadata_stream_check_version(stream
);
2474 if (stream
->chan
->metadata_cache
->max_offset
2475 == stream
->ust_metadata_pushed
) {
2480 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
2481 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
2482 stream
->chan
->metadata_cache
->max_offset
2483 - stream
->ust_metadata_pushed
);
2484 assert(write_len
!= 0);
2485 if (write_len
< 0) {
2486 ERR("Writing one metadata packet");
2490 stream
->ust_metadata_pushed
+= write_len
;
2492 assert(stream
->chan
->metadata_cache
->max_offset
>=
2493 stream
->ust_metadata_pushed
);
2497 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2503 * Sync metadata meaning request them to the session daemon and snapshot to the
2504 * metadata thread can consumer them.
2506 * Metadata stream lock is held here, but we need to release it when
2507 * interacting with sessiond, else we cause a deadlock with live
2508 * awaiting on metadata to be pushed out.
2510 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2511 * is empty or a negative value on error.
2513 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data
*ctx
,
2514 struct lttng_consumer_stream
*metadata
)
2522 pthread_mutex_unlock(&metadata
->lock
);
2524 * Request metadata from the sessiond, but don't wait for the flush
2525 * because we locked the metadata thread.
2527 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata
->chan
, 0, 0);
2528 pthread_mutex_lock(&metadata
->lock
);
2533 ret
= commit_one_metadata_packet(metadata
);
2536 } else if (ret
> 0) {
2540 ustctl_flush_buffer(metadata
->ustream
, 1);
2541 ret
= ustctl_snapshot(metadata
->ustream
);
2543 if (errno
!= EAGAIN
) {
2544 ERR("Sync metadata, taking UST snapshot");
2547 DBG("No new metadata when syncing them.");
2548 /* No new metadata, exit. */
2554 * After this flush, we still need to extract metadata.
2565 * Return 0 on success else a negative value.
2567 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2568 struct lttng_consumer_local_data
*ctx
)
2571 struct ustctl_consumer_stream
*ustream
;
2576 ustream
= stream
->ustream
;
2579 * First, we are going to check if there is a new subbuffer available
2580 * before reading the stream wait_fd.
2582 /* Get the next subbuffer */
2583 ret
= ustctl_get_next_subbuf(ustream
);
2585 /* No more data found, flag the stream. */
2586 stream
->has_data
= 0;
2591 ret
= ustctl_put_subbuf(ustream
);
2594 /* This stream still has data. Flag it and wake up the data thread. */
2595 stream
->has_data
= 1;
2597 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2600 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2601 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2606 /* The wake up pipe has been notified. */
2607 ctx
->has_wakeup
= 1;
2616 int update_stream_stats(struct lttng_consumer_stream
*stream
)
2619 uint64_t seq
, discarded
;
2621 ret
= ustctl_get_sequence_number(stream
->ustream
, &seq
);
2623 PERROR("ustctl_get_sequence_number");
2627 * Start the sequence when we extract the first packet in case we don't
2628 * start at 0 (for example if a consumer is not connected to the
2629 * session immediately after the beginning).
2631 if (stream
->last_sequence_number
== -1ULL) {
2632 stream
->last_sequence_number
= seq
;
2633 } else if (seq
> stream
->last_sequence_number
) {
2634 stream
->chan
->lost_packets
+= seq
-
2635 stream
->last_sequence_number
- 1;
2637 /* seq <= last_sequence_number */
2638 ERR("Sequence number inconsistent : prev = %" PRIu64
2639 ", current = %" PRIu64
,
2640 stream
->last_sequence_number
, seq
);
2644 stream
->last_sequence_number
= seq
;
2646 ret
= ustctl_get_events_discarded(stream
->ustream
, &discarded
);
2648 PERROR("kernctl_get_events_discarded");
2651 if (discarded
< stream
->last_discarded_events
) {
2653 * Overflow has occurred. We assume only one wrap-around
2656 stream
->chan
->discarded_events
+=
2657 (1ULL << (CAA_BITS_PER_LONG
- 1)) -
2658 stream
->last_discarded_events
+ discarded
;
2660 stream
->chan
->discarded_events
+= discarded
-
2661 stream
->last_discarded_events
;
2663 stream
->last_discarded_events
= discarded
;
2671 * Read subbuffer from the given stream.
2673 * Stream lock MUST be acquired.
2675 * Return 0 on success else a negative value.
2677 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
2678 struct lttng_consumer_local_data
*ctx
, bool *rotated
)
2680 unsigned long len
, subbuf_size
, padding
;
2681 int err
, write_index
= 1, rotation_ret
;
2683 struct ustctl_consumer_stream
*ustream
;
2684 struct ctf_packet_index index
;
2687 assert(stream
->ustream
);
2690 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream
->wait_fd
,
2693 /* Ease our life for what's next. */
2694 ustream
= stream
->ustream
;
2697 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2698 * error if we cannot read this one byte (read returns 0), or if the error
2699 * is EAGAIN or EWOULDBLOCK.
2701 * This is only done when the stream is monitored by a thread, before the
2702 * flush is done after a hangup and if the stream is not flagged with data
2703 * since there might be nothing to consume in the wait fd but still have
2704 * data available flagged by the consumer wake up pipe.
2706 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2710 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2711 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2718 * If the stream was flagged to be ready for rotation before we extract the
2719 * next packet, rotate it now.
2721 if (stream
->rotate_ready
) {
2722 DBG("Rotate stream before extracting data");
2723 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
, rotated
);
2724 if (rotation_ret
< 0) {
2725 ERR("Stream rotation error");
2732 /* Get the next subbuffer */
2733 err
= ustctl_get_next_subbuf(ustream
);
2736 * Populate metadata info if the existing info has
2737 * already been read.
2739 if (stream
->metadata_flag
) {
2740 ret
= commit_one_metadata_packet(stream
);
2744 ustctl_flush_buffer(stream
->ustream
, 1);
2748 ret
= err
; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2750 * This is a debug message even for single-threaded consumer,
2751 * because poll() have more relaxed criterions than get subbuf,
2752 * so get_subbuf may fail for short race windows where poll()
2753 * would issue wakeups.
2755 DBG("Reserving sub buffer failed (everything is normal, "
2756 "it is due to concurrency) [ret: %d]", err
);
2759 assert(stream
->chan
->output
== CONSUMER_CHANNEL_MMAP
);
2761 if (!stream
->metadata_flag
) {
2762 index
.offset
= htobe64(stream
->out_fd_offset
);
2763 ret
= get_index_values(&index
, ustream
);
2765 err
= ustctl_put_subbuf(ustream
);
2770 /* Update the stream's sequence and discarded events count. */
2771 ret
= update_stream_stats(stream
);
2773 PERROR("kernctl_get_events_discarded");
2774 err
= ustctl_put_subbuf(ustream
);
2782 /* Get the full padded subbuffer size */
2783 err
= ustctl_get_padded_subbuf_size(ustream
, &len
);
2786 /* Get subbuffer data size (without padding) */
2787 err
= ustctl_get_subbuf_size(ustream
, &subbuf_size
);
2790 /* Make sure we don't get a subbuffer size bigger than the padded */
2791 assert(len
>= subbuf_size
);
2793 padding
= len
- subbuf_size
;
2795 /* write the subbuffer to the tracefile */
2796 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
, padding
, &index
);
2798 * The mmap operation should write subbuf_size amount of data when network
2799 * streaming or the full padding (len) size when we are _not_ streaming.
2801 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
2802 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
2804 * Display the error but continue processing to try to release the
2805 * subbuffer. This is a DBG statement since any unexpected kill or
2806 * signal, the application gets unregistered, relayd gets closed or
2807 * anything that affects the buffer lifetime will trigger this error.
2808 * So, for the sake of the user, don't print this error since it can
2809 * happen and it is OK with the code flow.
2811 DBG("Error writing to tracefile "
2812 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2813 ret
, len
, subbuf_size
);
2816 err
= ustctl_put_next_subbuf(ustream
);
2820 * This will consumer the byte on the wait_fd if and only if there is not
2821 * next subbuffer to be acquired.
2823 if (!stream
->metadata_flag
) {
2824 ret
= notify_if_more_data(stream
, ctx
);
2830 /* Write index if needed. */
2835 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
2837 * In live, block until all the metadata is sent.
2839 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2840 assert(!stream
->missed_metadata_flush
);
2841 stream
->waiting_on_metadata
= true;
2842 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2844 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
2846 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2847 stream
->waiting_on_metadata
= false;
2848 if (stream
->missed_metadata_flush
) {
2849 stream
->missed_metadata_flush
= false;
2850 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2851 (void) consumer_flush_ust_index(stream
);
2853 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2861 assert(!stream
->metadata_flag
);
2862 err
= consumer_stream_write_index(stream
, &index
);
2869 * After extracting the packet, we check if the stream is now ready to be
2870 * rotated and perform the action immediately.
2872 rotation_ret
= lttng_consumer_stream_is_rotate_ready(stream
);
2873 if (rotation_ret
== 1) {
2874 rotation_ret
= lttng_consumer_rotate_stream(ctx
, stream
, rotated
);
2875 if (rotation_ret
< 0) {
2876 ERR("Stream rotation error");
2880 } else if (rotation_ret
< 0) {
2881 ERR("Checking if stream is ready to rotate");
2890 * Called when a stream is created.
2892 * Return 0 on success or else a negative value.
2894 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
2900 /* Don't create anything if this is set for streaming. */
2901 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
) {
2902 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
2903 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
2904 stream
->uid
, stream
->gid
, NULL
);
2908 stream
->out_fd
= ret
;
2909 stream
->tracefile_size_current
= 0;
2911 if (!stream
->metadata_flag
) {
2912 struct lttng_index_file
*index_file
;
2914 index_file
= lttng_index_file_create(stream
->chan
->pathname
,
2915 stream
->name
, stream
->uid
, stream
->gid
,
2916 stream
->chan
->tracefile_size
,
2917 stream
->tracefile_count_current
,
2918 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
2922 assert(!stream
->index_file
);
2923 stream
->index_file
= index_file
;
2933 * Check if data is still being extracted from the buffers for a specific
2934 * stream. Consumer data lock MUST be acquired before calling this function
2935 * and the stream lock.
2937 * Return 1 if the traced data are still getting read else 0 meaning that the
2938 * data is available for trace viewer reading.
2940 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
2945 assert(stream
->ustream
);
2947 DBG("UST consumer checking data pending");
2949 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
2954 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
2955 uint64_t contiguous
, pushed
;
2957 /* Ease our life a bit. */
2958 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
2959 pushed
= stream
->ust_metadata_pushed
;
2962 * We can simply check whether all contiguously available data
2963 * has been pushed to the ring buffer, since the push operation
2964 * is performed within get_next_subbuf(), and because both
2965 * get_next_subbuf() and put_next_subbuf() are issued atomically
2966 * thanks to the stream lock within
2967 * lttng_ustconsumer_read_subbuffer(). This basically means that
2968 * whetnever ust_metadata_pushed is incremented, the associated
2969 * metadata has been consumed from the metadata stream.
2971 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
2972 contiguous
, pushed
);
2973 assert(((int64_t) (contiguous
- pushed
)) >= 0);
2974 if ((contiguous
!= pushed
) ||
2975 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
2976 ret
= 1; /* Data is pending */
2980 ret
= ustctl_get_next_subbuf(stream
->ustream
);
2983 * There is still data so let's put back this
2986 ret
= ustctl_put_subbuf(stream
->ustream
);
2988 ret
= 1; /* Data is pending */
2993 /* Data is NOT pending so ready to be read. */
3001 * Stop a given metadata channel timer if enabled and close the wait fd which
3002 * is the poll pipe of the metadata stream.
3004 * This MUST be called with the metadata channel acquired.
3006 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
3011 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
3013 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
3015 if (metadata
->switch_timer_enabled
== 1) {
3016 consumer_timer_switch_stop(metadata
);
3019 if (!metadata
->metadata_stream
) {
3024 * Closing write side so the thread monitoring the stream wakes up if any
3025 * and clean the metadata stream.
3027 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
3028 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
3030 PERROR("closing metadata pipe write side");
3032 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
3040 * Close every metadata stream wait fd of the metadata hash table. This
3041 * function MUST be used very carefully so not to run into a race between the
3042 * metadata thread handling streams and this function closing their wait fd.
3044 * For UST, this is used when the session daemon hangs up. Its the metadata
3045 * producer so calling this is safe because we are assured that no state change
3046 * can occur in the metadata thread for the streams in the hash table.
3048 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
3050 struct lttng_ht_iter iter
;
3051 struct lttng_consumer_stream
*stream
;
3053 assert(metadata_ht
);
3054 assert(metadata_ht
->ht
);
3056 DBG("UST consumer closing all metadata streams");
3059 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
3062 health_code_update();
3064 pthread_mutex_lock(&stream
->chan
->lock
);
3065 lttng_ustconsumer_close_metadata(stream
->chan
);
3066 pthread_mutex_unlock(&stream
->chan
->lock
);
3072 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
3076 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
3078 ERR("Unable to close wakeup fd");
3083 * Please refer to consumer-timer.c before adding any lock within this
3084 * function or any of its callees. Timers have a very strict locking
3085 * semantic with respect to teardown. Failure to respect this semantic
3086 * introduces deadlocks.
3088 * DON'T hold the metadata lock when calling this function, else this
3089 * can cause deadlock involving consumer awaiting for metadata to be
3090 * pushed out due to concurrent interaction with the session daemon.
3092 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
3093 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
3095 struct lttcomm_metadata_request_msg request
;
3096 struct lttcomm_consumer_msg msg
;
3097 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3098 uint64_t len
, key
, offset
, version
;
3102 assert(channel
->metadata_cache
);
3104 memset(&request
, 0, sizeof(request
));
3106 /* send the metadata request to sessiond */
3107 switch (consumer_data
.type
) {
3108 case LTTNG_CONSUMER64_UST
:
3109 request
.bits_per_long
= 64;
3111 case LTTNG_CONSUMER32_UST
:
3112 request
.bits_per_long
= 32;
3115 request
.bits_per_long
= 0;
3119 request
.session_id
= channel
->session_id
;
3120 request
.session_id_per_pid
= channel
->session_id_per_pid
;
3122 * Request the application UID here so the metadata of that application can
3123 * be sent back. The channel UID corresponds to the user UID of the session
3124 * used for the rights on the stream file(s).
3126 request
.uid
= channel
->ust_app_uid
;
3127 request
.key
= channel
->key
;
3129 DBG("Sending metadata request to sessiond, session id %" PRIu64
3130 ", per-pid %" PRIu64
", app UID %u and channel key %" PRIu64
,
3131 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
3134 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
3136 health_code_update();
3138 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
3141 ERR("Asking metadata to sessiond");
3145 health_code_update();
3147 /* Receive the metadata from sessiond */
3148 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
3150 if (ret
!= sizeof(msg
)) {
3151 DBG("Consumer received unexpected message size %d (expects %zu)",
3153 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
3155 * The ret value might 0 meaning an orderly shutdown but this is ok
3156 * since the caller handles this.
3161 health_code_update();
3163 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
3164 /* No registry found */
3165 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3169 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
3170 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
3175 len
= msg
.u
.push_metadata
.len
;
3176 key
= msg
.u
.push_metadata
.key
;
3177 offset
= msg
.u
.push_metadata
.target_offset
;
3178 version
= msg
.u
.push_metadata
.version
;
3180 assert(key
== channel
->key
);
3182 DBG("No new metadata to receive for key %" PRIu64
, key
);
3185 health_code_update();
3187 /* Tell session daemon we are ready to receive the metadata. */
3188 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
3189 LTTCOMM_CONSUMERD_SUCCESS
);
3190 if (ret
< 0 || len
== 0) {
3192 * Somehow, the session daemon is not responding anymore or there is
3193 * nothing to receive.
3198 health_code_update();
3200 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
3201 key
, offset
, len
, version
, channel
, timer
, wait
);
3204 * Only send the status msg if the sessiond is alive meaning a positive
3207 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
3212 health_code_update();
3214 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
3219 * Return the ustctl call for the get stream id.
3221 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
3222 uint64_t *stream_id
)
3227 return ustctl_get_stream_id(stream
->ustream
, stream_id
);