2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 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-metadata-cache.h>
43 #include <common/consumer-stream.h>
44 #include <common/consumer-timer.h>
45 #include <common/utils.h>
46 #include <common/index/index.h>
48 #include "ust-consumer.h"
50 #define UINT_MAX_STR_LEN 11 /* includes \0 */
52 extern struct lttng_consumer_global_data consumer_data
;
53 extern int consumer_poll_timeout
;
54 extern volatile int consumer_quit
;
57 * Free channel object and all streams associated with it. This MUST be used
58 * only and only if the channel has _NEVER_ been added to the global channel
61 static void destroy_channel(struct lttng_consumer_channel
*channel
)
63 struct lttng_consumer_stream
*stream
, *stmp
;
67 DBG("UST consumer cleaning stream list");
69 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
74 cds_list_del(&stream
->send_node
);
75 ustctl_destroy_stream(stream
->ustream
);
80 * If a channel is available meaning that was created before the streams
84 lttng_ustconsumer_del_channel(channel
);
85 lttng_ustconsumer_free_channel(channel
);
91 * Add channel to internal consumer state.
93 * Returns 0 on success or else a negative value.
95 static int add_channel(struct lttng_consumer_channel
*channel
,
96 struct lttng_consumer_local_data
*ctx
)
103 if (ctx
->on_recv_channel
!= NULL
) {
104 ret
= ctx
->on_recv_channel(channel
);
106 ret
= consumer_add_channel(channel
, ctx
);
107 } else if (ret
< 0) {
108 /* Most likely an ENOMEM. */
109 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
113 ret
= consumer_add_channel(channel
, ctx
);
116 DBG("UST consumer channel added (key: %" PRIu64
")", channel
->key
);
123 * Allocate and return a consumer channel object.
125 static struct lttng_consumer_channel
*allocate_channel(uint64_t session_id
,
126 const char *pathname
, const char *name
, uid_t uid
, gid_t gid
,
127 uint64_t relayd_id
, uint64_t key
, enum lttng_event_output output
,
128 uint64_t tracefile_size
, uint64_t tracefile_count
,
129 uint64_t session_id_per_pid
, unsigned int monitor
,
130 unsigned int live_timer_interval
,
131 const char *root_shm_path
, const char *shm_path
)
136 return consumer_allocate_channel(key
, session_id
, pathname
, name
, uid
,
137 gid
, relayd_id
, output
, tracefile_size
,
138 tracefile_count
, session_id_per_pid
, monitor
,
139 live_timer_interval
, root_shm_path
, shm_path
);
143 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
144 * error value if applicable is set in it else it is kept untouched.
146 * Return NULL on error else the newly allocated stream object.
148 static struct lttng_consumer_stream
*allocate_stream(int cpu
, int key
,
149 struct lttng_consumer_channel
*channel
,
150 struct lttng_consumer_local_data
*ctx
, int *_alloc_ret
)
153 struct lttng_consumer_stream
*stream
= NULL
;
158 stream
= consumer_allocate_stream(channel
->key
,
160 LTTNG_CONSUMER_ACTIVE_STREAM
,
170 if (stream
== NULL
) {
174 * We could not find the channel. Can happen if cpu hotplug
175 * happens while tearing down.
177 DBG3("Could not find channel");
182 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
188 stream
->chan
= channel
;
192 *_alloc_ret
= alloc_ret
;
198 * Send the given stream pointer to the corresponding thread.
200 * Returns 0 on success else a negative value.
202 static int send_stream_to_thread(struct lttng_consumer_stream
*stream
,
203 struct lttng_consumer_local_data
*ctx
)
206 struct lttng_pipe
*stream_pipe
;
208 /* Get the right pipe where the stream will be sent. */
209 if (stream
->metadata_flag
) {
210 ret
= consumer_add_metadata_stream(stream
);
212 ERR("Consumer add metadata stream %" PRIu64
" failed.",
216 stream_pipe
= ctx
->consumer_metadata_pipe
;
218 ret
= consumer_add_data_stream(stream
);
220 ERR("Consumer add stream %" PRIu64
" failed.",
224 stream_pipe
= ctx
->consumer_data_pipe
;
228 * From this point on, the stream's ownership has been moved away from
229 * the channel and becomes globally visible.
231 stream
->globally_visible
= 1;
233 ret
= lttng_pipe_write(stream_pipe
, &stream
, sizeof(stream
));
235 ERR("Consumer write %s stream to pipe %d",
236 stream
->metadata_flag
? "metadata" : "data",
237 lttng_pipe_get_writefd(stream_pipe
));
238 if (stream
->metadata_flag
) {
239 consumer_del_stream_for_metadata(stream
);
241 consumer_del_stream_for_data(stream
);
249 int get_stream_shm_path(char *stream_shm_path
, const char *shm_path
, int cpu
)
251 char cpu_nr
[UINT_MAX_STR_LEN
]; /* unsigned int max len */
254 strncpy(stream_shm_path
, shm_path
, PATH_MAX
);
255 stream_shm_path
[PATH_MAX
- 1] = '\0';
256 ret
= snprintf(cpu_nr
, UINT_MAX_STR_LEN
, "%u", cpu
);
261 strncat(stream_shm_path
, cpu_nr
,
262 PATH_MAX
- strlen(stream_shm_path
) - 1);
269 * Create streams for the given channel using liblttng-ust-ctl.
271 * Return 0 on success else a negative value.
273 static int create_ust_streams(struct lttng_consumer_channel
*channel
,
274 struct lttng_consumer_local_data
*ctx
)
277 struct ustctl_consumer_stream
*ustream
;
278 struct lttng_consumer_stream
*stream
;
284 * While a stream is available from ustctl. When NULL is returned, we've
285 * reached the end of the possible stream for the channel.
287 while ((ustream
= ustctl_create_stream(channel
->uchan
, cpu
))) {
289 int ust_metadata_pipe
[2];
291 health_code_update();
293 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
&& channel
->monitor
) {
294 ret
= utils_create_pipe_cloexec_nonblock(ust_metadata_pipe
);
296 ERR("Create ust metadata poll pipe");
299 wait_fd
= ust_metadata_pipe
[0];
301 wait_fd
= ustctl_stream_get_wait_fd(ustream
);
304 /* Allocate consumer stream object. */
305 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
,
415 struct ustctl_consumer_channel_attr
*attr
,
418 char shm_path
[PATH_MAX
];
421 if (!channel
->shm_path
[0]) {
422 return create_posix_shm();
424 ret
= get_stream_shm_path(shm_path
, channel
->shm_path
, cpu
);
428 return run_as_open(shm_path
,
429 O_RDWR
| O_CREAT
| O_EXCL
, S_IRUSR
| S_IWUSR
,
430 channel
->uid
, channel
->gid
);
437 * Create an UST channel with the given attributes and send it to the session
438 * daemon using the ust ctl API.
440 * Return 0 on success or else a negative value.
442 static int create_ust_channel(struct lttng_consumer_channel
*channel
,
443 struct ustctl_consumer_channel_attr
*attr
,
444 struct ustctl_consumer_channel
**ust_chanp
)
446 int ret
, nr_stream_fds
, i
, j
;
448 struct ustctl_consumer_channel
*ust_channel
;
454 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
455 "subbuf_size: %" PRIu64
", num_subbuf: %" PRIu64
", "
456 "switch_timer_interval: %u, read_timer_interval: %u, "
457 "output: %d, type: %d", attr
->overwrite
, attr
->subbuf_size
,
458 attr
->num_subbuf
, attr
->switch_timer_interval
,
459 attr
->read_timer_interval
, attr
->output
, attr
->type
);
461 if (channel
->type
== CONSUMER_CHANNEL_TYPE_METADATA
)
464 nr_stream_fds
= ustctl_get_nr_stream_per_channel();
465 stream_fds
= zmalloc(nr_stream_fds
* sizeof(*stream_fds
));
470 for (i
= 0; i
< nr_stream_fds
; i
++) {
471 stream_fds
[i
] = open_ust_stream_fd(channel
, attr
, i
);
472 if (stream_fds
[i
] < 0) {
477 ust_channel
= ustctl_create_channel(attr
, stream_fds
, nr_stream_fds
);
482 channel
->nr_stream_fds
= nr_stream_fds
;
483 channel
->stream_fds
= stream_fds
;
484 *ust_chanp
= ust_channel
;
490 for (j
= i
- 1; j
>= 0; j
--) {
493 closeret
= close(stream_fds
[j
]);
497 if (channel
->shm_path
[0]) {
498 char shm_path
[PATH_MAX
];
500 closeret
= get_stream_shm_path(shm_path
,
501 channel
->shm_path
, j
);
503 ERR("Cannot get stream shm path");
505 closeret
= run_as_unlink(shm_path
,
506 channel
->uid
, channel
->gid
);
508 PERROR("unlink %s", shm_path
);
512 /* Try to rmdir all directories under shm_path root. */
513 if (channel
->root_shm_path
[0]) {
514 (void) run_as_recursive_rmdir(channel
->root_shm_path
,
515 channel
->uid
, channel
->gid
);
523 * Send a single given stream to the session daemon using the sock.
525 * Return 0 on success else a negative value.
527 static int send_sessiond_stream(int sock
, struct lttng_consumer_stream
*stream
)
534 DBG("UST consumer sending stream %" PRIu64
" to sessiond", stream
->key
);
536 /* Send stream to session daemon. */
537 ret
= ustctl_send_stream_to_sessiond(sock
, stream
->ustream
);
547 * Send channel to sessiond.
549 * Return 0 on success or else a negative value.
551 static int send_sessiond_channel(int sock
,
552 struct lttng_consumer_channel
*channel
,
553 struct lttng_consumer_local_data
*ctx
, int *relayd_error
)
555 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
556 struct lttng_consumer_stream
*stream
;
557 uint64_t net_seq_idx
= -1ULL;
563 DBG("UST consumer sending channel %s to sessiond", channel
->name
);
565 if (channel
->relayd_id
!= (uint64_t) -1ULL) {
566 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
568 health_code_update();
570 /* Try to send the stream to the relayd if one is available. */
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
, 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 ustctl_flush_buffer(stream
->ustream
, 1);
779 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
780 * RCU read side lock MUST be acquired before calling this function.
782 * Return 0 on success else an LTTng error code.
784 static int close_metadata(uint64_t chan_key
)
787 struct lttng_consumer_channel
*channel
;
789 DBG("UST consumer close metadata key %" PRIu64
, chan_key
);
791 channel
= consumer_find_channel(chan_key
);
794 * This is possible if the metadata thread has issue a delete because
795 * the endpoint point of the stream hung up. There is no way the
796 * session daemon can know about it thus use a DBG instead of an actual
799 DBG("UST consumer close metadata %" PRIu64
" not found", chan_key
);
800 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
804 pthread_mutex_lock(&consumer_data
.lock
);
805 pthread_mutex_lock(&channel
->lock
);
807 if (cds_lfht_is_node_deleted(&channel
->node
.node
)) {
811 lttng_ustconsumer_close_metadata(channel
);
814 pthread_mutex_unlock(&channel
->lock
);
815 pthread_mutex_unlock(&consumer_data
.lock
);
821 * RCU read side lock MUST be acquired before calling this function.
823 * Return 0 on success else an LTTng error code.
825 static int setup_metadata(struct lttng_consumer_local_data
*ctx
, uint64_t key
)
828 struct lttng_consumer_channel
*metadata
;
830 DBG("UST consumer setup metadata key %" PRIu64
, key
);
832 metadata
= consumer_find_channel(key
);
834 ERR("UST consumer push metadata %" PRIu64
" not found", key
);
835 ret
= LTTNG_ERR_UST_CHAN_NOT_FOUND
;
840 * In no monitor mode, the metadata channel has no stream(s) so skip the
841 * ownership transfer to the metadata thread.
843 if (!metadata
->monitor
) {
844 DBG("Metadata channel in no monitor");
850 * Send metadata stream to relayd if one available. Availability is
851 * known if the stream is still in the list of the channel.
853 if (cds_list_empty(&metadata
->streams
.head
)) {
854 ERR("Metadata channel key %" PRIu64
", no stream available.", key
);
855 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
856 goto error_no_stream
;
859 /* Send metadata stream to relayd if needed. */
860 if (metadata
->metadata_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
861 ret
= consumer_send_relayd_stream(metadata
->metadata_stream
,
864 ret
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
867 ret
= consumer_send_relayd_streams_sent(
868 metadata
->metadata_stream
->net_seq_idx
);
870 ret
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
875 ret
= send_streams_to_thread(metadata
, ctx
);
878 * If we are unable to send the stream to the thread, there is
879 * a big problem so just stop everything.
881 ret
= LTTCOMM_CONSUMERD_FATAL
;
884 /* List MUST be empty after or else it could be reused. */
885 assert(cds_list_empty(&metadata
->streams
.head
));
892 * Delete metadata channel on error. At this point, the metadata stream can
893 * NOT be monitored by the metadata thread thus having the guarantee that
894 * the stream is still in the local stream list of the channel. This call
895 * will make sure to clean that list.
897 consumer_stream_destroy(metadata
->metadata_stream
, NULL
);
898 cds_list_del(&metadata
->metadata_stream
->send_node
);
899 metadata
->metadata_stream
= NULL
;
906 * Snapshot the whole metadata.
908 * Returns 0 on success, < 0 on error
910 static int snapshot_metadata(uint64_t key
, char *path
, uint64_t relayd_id
,
911 struct lttng_consumer_local_data
*ctx
)
914 struct lttng_consumer_channel
*metadata_channel
;
915 struct lttng_consumer_stream
*metadata_stream
;
920 DBG("UST consumer snapshot metadata with key %" PRIu64
" at path %s",
925 metadata_channel
= consumer_find_channel(key
);
926 if (!metadata_channel
) {
927 ERR("UST snapshot metadata channel not found for key %" PRIu64
,
932 assert(!metadata_channel
->monitor
);
934 health_code_update();
937 * Ask the sessiond if we have new metadata waiting and update the
938 * consumer metadata cache.
940 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata_channel
, 0, 1);
945 health_code_update();
948 * The metadata stream is NOT created in no monitor mode when the channel
949 * is created on a sessiond ask channel command.
951 ret
= create_ust_streams(metadata_channel
, ctx
);
956 metadata_stream
= metadata_channel
->metadata_stream
;
957 assert(metadata_stream
);
959 if (relayd_id
!= (uint64_t) -1ULL) {
960 metadata_stream
->net_seq_idx
= relayd_id
;
961 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
966 ret
= utils_create_stream_file(path
, metadata_stream
->name
,
967 metadata_stream
->chan
->tracefile_size
,
968 metadata_stream
->tracefile_count_current
,
969 metadata_stream
->uid
, metadata_stream
->gid
, NULL
);
973 metadata_stream
->out_fd
= ret
;
974 metadata_stream
->tracefile_size_current
= 0;
978 health_code_update();
980 ret
= lttng_consumer_read_subbuffer(metadata_stream
, ctx
);
988 * Clean up the stream completly because the next snapshot will use a new
991 consumer_stream_destroy(metadata_stream
, NULL
);
992 cds_list_del(&metadata_stream
->send_node
);
993 metadata_channel
->metadata_stream
= NULL
;
1001 * Take a snapshot of all the stream of a channel.
1003 * Returns 0 on success, < 0 on error
1005 static int snapshot_channel(uint64_t key
, char *path
, uint64_t relayd_id
,
1006 uint64_t nb_packets_per_stream
, struct lttng_consumer_local_data
*ctx
)
1009 unsigned use_relayd
= 0;
1010 unsigned long consumed_pos
, produced_pos
;
1011 struct lttng_consumer_channel
*channel
;
1012 struct lttng_consumer_stream
*stream
;
1019 if (relayd_id
!= (uint64_t) -1ULL) {
1023 channel
= consumer_find_channel(key
);
1025 ERR("UST snapshot channel not found for key %" PRIu64
, key
);
1029 assert(!channel
->monitor
);
1030 DBG("UST consumer snapshot channel %" PRIu64
, key
);
1032 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
1034 health_code_update();
1036 /* Lock stream because we are about to change its state. */
1037 pthread_mutex_lock(&stream
->lock
);
1038 stream
->net_seq_idx
= relayd_id
;
1041 ret
= consumer_send_relayd_stream(stream
, path
);
1046 ret
= utils_create_stream_file(path
, stream
->name
,
1047 stream
->chan
->tracefile_size
,
1048 stream
->tracefile_count_current
,
1049 stream
->uid
, stream
->gid
, NULL
);
1053 stream
->out_fd
= ret
;
1054 stream
->tracefile_size_current
= 0;
1056 DBG("UST consumer snapshot stream %s/%s (%" PRIu64
")", path
,
1057 stream
->name
, stream
->key
);
1059 if (relayd_id
!= -1ULL) {
1060 ret
= consumer_send_relayd_streams_sent(relayd_id
);
1066 ustctl_flush_buffer(stream
->ustream
, 1);
1068 ret
= lttng_ustconsumer_take_snapshot(stream
);
1070 ERR("Taking UST snapshot");
1074 ret
= lttng_ustconsumer_get_produced_snapshot(stream
, &produced_pos
);
1076 ERR("Produced UST snapshot position");
1080 ret
= lttng_ustconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
1082 ERR("Consumerd UST snapshot position");
1087 * The original value is sent back if max stream size is larger than
1088 * the possible size of the snapshot. Also, we assume that the session
1089 * daemon should never send a maximum stream size that is lower than
1092 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
1093 produced_pos
, nb_packets_per_stream
,
1094 stream
->max_sb_size
);
1096 while (consumed_pos
< produced_pos
) {
1098 unsigned long len
, padded_len
;
1100 health_code_update();
1102 DBG("UST consumer taking snapshot at pos %lu", consumed_pos
);
1104 ret
= ustctl_get_subbuf(stream
->ustream
, &consumed_pos
);
1106 if (ret
!= -EAGAIN
) {
1107 PERROR("ustctl_get_subbuf snapshot");
1108 goto error_close_stream
;
1110 DBG("UST consumer get subbuf failed. Skipping it.");
1111 consumed_pos
+= stream
->max_sb_size
;
1115 ret
= ustctl_get_subbuf_size(stream
->ustream
, &len
);
1117 ERR("Snapshot ustctl_get_subbuf_size");
1118 goto error_put_subbuf
;
1121 ret
= ustctl_get_padded_subbuf_size(stream
->ustream
, &padded_len
);
1123 ERR("Snapshot ustctl_get_padded_subbuf_size");
1124 goto error_put_subbuf
;
1127 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, len
,
1128 padded_len
- len
, NULL
);
1130 if (read_len
!= len
) {
1132 goto error_put_subbuf
;
1135 if (read_len
!= padded_len
) {
1137 goto error_put_subbuf
;
1141 ret
= ustctl_put_subbuf(stream
->ustream
);
1143 ERR("Snapshot ustctl_put_subbuf");
1144 goto error_close_stream
;
1146 consumed_pos
+= stream
->max_sb_size
;
1149 /* Simply close the stream so we can use it on the next snapshot. */
1150 consumer_stream_close(stream
);
1151 pthread_mutex_unlock(&stream
->lock
);
1158 if (ustctl_put_subbuf(stream
->ustream
) < 0) {
1159 ERR("Snapshot ustctl_put_subbuf");
1162 consumer_stream_close(stream
);
1164 pthread_mutex_unlock(&stream
->lock
);
1171 * Receive the metadata updates from the sessiond. Supports receiving
1172 * overlapping metadata, but is needs to always belong to a contiguous
1173 * range starting from 0.
1174 * Be careful about the locks held when calling this function: it needs
1175 * the metadata cache flush to concurrently progress in order to
1178 int lttng_ustconsumer_recv_metadata(int sock
, uint64_t key
, uint64_t offset
,
1179 uint64_t len
, struct lttng_consumer_channel
*channel
,
1180 int timer
, int wait
)
1182 int ret
, ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1185 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
, len
);
1187 metadata_str
= zmalloc(len
* sizeof(char));
1188 if (!metadata_str
) {
1189 PERROR("zmalloc metadata string");
1190 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
1194 health_code_update();
1196 /* Receive metadata string. */
1197 ret
= lttcomm_recv_unix_sock(sock
, metadata_str
, len
);
1199 /* Session daemon is dead so return gracefully. */
1204 health_code_update();
1206 pthread_mutex_lock(&channel
->metadata_cache
->lock
);
1207 ret
= consumer_metadata_cache_write(channel
, offset
, len
, metadata_str
);
1209 /* Unable to handle metadata. Notify session daemon. */
1210 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1212 * Skip metadata flush on write error since the offset and len might
1213 * not have been updated which could create an infinite loop below when
1214 * waiting for the metadata cache to be flushed.
1216 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1219 pthread_mutex_unlock(&channel
->metadata_cache
->lock
);
1224 while (consumer_metadata_cache_flushed(channel
, offset
+ len
, timer
)) {
1225 DBG("Waiting for metadata to be flushed");
1227 health_code_update();
1229 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME
);
1239 * Receive command from session daemon and process it.
1241 * Return 1 on success else a negative value or 0.
1243 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1244 int sock
, struct pollfd
*consumer_sockpoll
)
1247 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1248 struct lttcomm_consumer_msg msg
;
1249 struct lttng_consumer_channel
*channel
= NULL
;
1251 health_code_update();
1253 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
1254 if (ret
!= sizeof(msg
)) {
1255 DBG("Consumer received unexpected message size %zd (expects %zu)",
1258 * The ret value might 0 meaning an orderly shutdown but this is ok
1259 * since the caller handles this.
1262 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
1268 health_code_update();
1271 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
1273 health_code_update();
1275 /* relayd needs RCU read-side lock */
1278 switch (msg
.cmd_type
) {
1279 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
1281 /* Session daemon status message are handled in the following call. */
1282 ret
= consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
1283 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
1284 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
1285 msg
.u
.relayd_sock
.relayd_session_id
);
1288 case LTTNG_CONSUMER_DESTROY_RELAYD
:
1290 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
1291 struct consumer_relayd_sock_pair
*relayd
;
1293 DBG("UST consumer destroying relayd %" PRIu64
, index
);
1295 /* Get relayd reference if exists. */
1296 relayd
= consumer_find_relayd(index
);
1297 if (relayd
== NULL
) {
1298 DBG("Unable to find relayd %" PRIu64
, index
);
1299 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
1303 * Each relayd socket pair has a refcount of stream attached to it
1304 * which tells if the relayd is still active or not depending on the
1307 * This will set the destroy flag of the relayd object and destroy it
1308 * if the refcount reaches zero when called.
1310 * The destroy can happen either here or when a stream fd hangs up.
1313 consumer_flag_relayd_for_destroy(relayd
);
1316 goto end_msg_sessiond
;
1318 case LTTNG_CONSUMER_UPDATE_STREAM
:
1323 case LTTNG_CONSUMER_DATA_PENDING
:
1325 int ret
, is_data_pending
;
1326 uint64_t id
= msg
.u
.data_pending
.session_id
;
1328 DBG("UST consumer data pending command for id %" PRIu64
, id
);
1330 is_data_pending
= consumer_data_pending(id
);
1332 /* Send back returned value to session daemon */
1333 ret
= lttcomm_send_unix_sock(sock
, &is_data_pending
,
1334 sizeof(is_data_pending
));
1336 DBG("Error when sending the data pending ret code: %d", ret
);
1341 * No need to send back a status message since the data pending
1342 * returned value is the response.
1346 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION
:
1349 struct ustctl_consumer_channel_attr attr
;
1351 /* Create a plain object and reserve a channel key. */
1352 channel
= allocate_channel(msg
.u
.ask_channel
.session_id
,
1353 msg
.u
.ask_channel
.pathname
, msg
.u
.ask_channel
.name
,
1354 msg
.u
.ask_channel
.uid
, msg
.u
.ask_channel
.gid
,
1355 msg
.u
.ask_channel
.relayd_id
, msg
.u
.ask_channel
.key
,
1356 (enum lttng_event_output
) msg
.u
.ask_channel
.output
,
1357 msg
.u
.ask_channel
.tracefile_size
,
1358 msg
.u
.ask_channel
.tracefile_count
,
1359 msg
.u
.ask_channel
.session_id_per_pid
,
1360 msg
.u
.ask_channel
.monitor
,
1361 msg
.u
.ask_channel
.live_timer_interval
,
1362 msg
.u
.ask_channel
.root_shm_path
,
1363 msg
.u
.ask_channel
.shm_path
);
1365 goto end_channel_error
;
1369 * Assign UST application UID to the channel. This value is ignored for
1370 * per PID buffers. This is specific to UST thus setting this after the
1373 channel
->ust_app_uid
= msg
.u
.ask_channel
.ust_app_uid
;
1375 /* Build channel attributes from received message. */
1376 attr
.subbuf_size
= msg
.u
.ask_channel
.subbuf_size
;
1377 attr
.num_subbuf
= msg
.u
.ask_channel
.num_subbuf
;
1378 attr
.overwrite
= msg
.u
.ask_channel
.overwrite
;
1379 attr
.switch_timer_interval
= msg
.u
.ask_channel
.switch_timer_interval
;
1380 attr
.read_timer_interval
= msg
.u
.ask_channel
.read_timer_interval
;
1381 attr
.chan_id
= msg
.u
.ask_channel
.chan_id
;
1382 memcpy(attr
.uuid
, msg
.u
.ask_channel
.uuid
, sizeof(attr
.uuid
));
1384 /* Match channel buffer type to the UST abi. */
1385 switch (msg
.u
.ask_channel
.output
) {
1386 case LTTNG_EVENT_MMAP
:
1388 attr
.output
= LTTNG_UST_MMAP
;
1392 /* Translate and save channel type. */
1393 switch (msg
.u
.ask_channel
.type
) {
1394 case LTTNG_UST_CHAN_PER_CPU
:
1395 channel
->type
= CONSUMER_CHANNEL_TYPE_DATA
;
1396 attr
.type
= LTTNG_UST_CHAN_PER_CPU
;
1398 * Set refcount to 1 for owner. Below, we will
1399 * pass ownership to the
1400 * consumer_thread_channel_poll() thread.
1402 channel
->refcount
= 1;
1404 case LTTNG_UST_CHAN_METADATA
:
1405 channel
->type
= CONSUMER_CHANNEL_TYPE_METADATA
;
1406 attr
.type
= LTTNG_UST_CHAN_METADATA
;
1413 health_code_update();
1415 ret
= ask_channel(ctx
, sock
, channel
, &attr
);
1417 goto end_channel_error
;
1420 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1421 ret
= consumer_metadata_cache_allocate(channel
);
1423 ERR("Allocating metadata cache");
1424 goto end_channel_error
;
1426 consumer_timer_switch_start(channel
, attr
.switch_timer_interval
);
1427 attr
.switch_timer_interval
= 0;
1429 consumer_timer_live_start(channel
,
1430 msg
.u
.ask_channel
.live_timer_interval
);
1433 health_code_update();
1436 * Add the channel to the internal state AFTER all streams were created
1437 * and successfully sent to session daemon. This way, all streams must
1438 * be ready before this channel is visible to the threads.
1439 * If add_channel succeeds, ownership of the channel is
1440 * passed to consumer_thread_channel_poll().
1442 ret
= add_channel(channel
, ctx
);
1444 if (msg
.u
.ask_channel
.type
== LTTNG_UST_CHAN_METADATA
) {
1445 if (channel
->switch_timer_enabled
== 1) {
1446 consumer_timer_switch_stop(channel
);
1448 consumer_metadata_cache_destroy(channel
);
1450 if (channel
->live_timer_enabled
== 1) {
1451 consumer_timer_live_stop(channel
);
1453 goto end_channel_error
;
1456 health_code_update();
1459 * Channel and streams are now created. Inform the session daemon that
1460 * everything went well and should wait to receive the channel and
1461 * streams with ustctl API.
1463 ret
= consumer_send_status_channel(sock
, channel
);
1466 * There is probably a problem on the socket.
1473 case LTTNG_CONSUMER_GET_CHANNEL
:
1475 int ret
, relayd_err
= 0;
1476 uint64_t key
= msg
.u
.get_channel
.key
;
1477 struct lttng_consumer_channel
*channel
;
1479 channel
= consumer_find_channel(key
);
1481 ERR("UST consumer get channel key %" PRIu64
" not found", key
);
1482 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
1483 goto end_msg_sessiond
;
1486 health_code_update();
1488 /* Send everything to sessiond. */
1489 ret
= send_sessiond_channel(sock
, channel
, ctx
, &relayd_err
);
1493 * We were unable to send to the relayd the stream so avoid
1494 * sending back a fatal error to the thread since this is OK
1495 * and the consumer can continue its work. The above call
1496 * has sent the error status message to the sessiond.
1501 * The communicaton was broken hence there is a bad state between
1502 * the consumer and sessiond so stop everything.
1507 health_code_update();
1510 * In no monitor mode, the streams ownership is kept inside the channel
1511 * so don't send them to the data thread.
1513 if (!channel
->monitor
) {
1514 goto end_msg_sessiond
;
1517 ret
= send_streams_to_thread(channel
, ctx
);
1520 * If we are unable to send the stream to the thread, there is
1521 * a big problem so just stop everything.
1525 /* List MUST be empty after or else it could be reused. */
1526 assert(cds_list_empty(&channel
->streams
.head
));
1527 goto end_msg_sessiond
;
1529 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
1531 uint64_t key
= msg
.u
.destroy_channel
.key
;
1534 * Only called if streams have not been sent to stream
1535 * manager thread. However, channel has been sent to
1536 * channel manager thread.
1538 notify_thread_del_channel(ctx
, key
);
1539 goto end_msg_sessiond
;
1541 case LTTNG_CONSUMER_CLOSE_METADATA
:
1545 ret
= close_metadata(msg
.u
.close_metadata
.key
);
1550 goto end_msg_sessiond
;
1552 case LTTNG_CONSUMER_FLUSH_CHANNEL
:
1556 ret
= flush_channel(msg
.u
.flush_channel
.key
);
1561 goto end_msg_sessiond
;
1563 case LTTNG_CONSUMER_PUSH_METADATA
:
1566 uint64_t len
= msg
.u
.push_metadata
.len
;
1567 uint64_t key
= msg
.u
.push_metadata
.key
;
1568 uint64_t offset
= msg
.u
.push_metadata
.target_offset
;
1569 struct lttng_consumer_channel
*channel
;
1571 DBG("UST consumer push metadata key %" PRIu64
" of len %" PRIu64
, key
,
1574 channel
= consumer_find_channel(key
);
1577 * This is possible if the metadata creation on the consumer side
1578 * is in flight vis-a-vis a concurrent push metadata from the
1579 * session daemon. Simply return that the channel failed and the
1580 * session daemon will handle that message correctly considering
1581 * that this race is acceptable thus the DBG() statement here.
1583 DBG("UST consumer push metadata %" PRIu64
" not found", key
);
1584 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1585 goto end_msg_sessiond
;
1588 health_code_update();
1592 * There is nothing to receive. We have simply
1593 * checked whether the channel can be found.
1595 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1596 goto end_msg_sessiond
;
1599 /* Tell session daemon we are ready to receive the metadata. */
1600 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
1602 /* Somehow, the session daemon is not responding anymore. */
1606 health_code_update();
1608 /* Wait for more data. */
1609 health_poll_entry();
1610 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
1616 health_code_update();
1618 ret
= lttng_ustconsumer_recv_metadata(sock
, key
, offset
,
1619 len
, channel
, 0, 1);
1621 /* error receiving from sessiond */
1625 goto end_msg_sessiond
;
1628 case LTTNG_CONSUMER_SETUP_METADATA
:
1632 ret
= setup_metadata(ctx
, msg
.u
.setup_metadata
.key
);
1636 goto end_msg_sessiond
;
1638 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
1640 if (msg
.u
.snapshot_channel
.metadata
) {
1641 ret
= snapshot_metadata(msg
.u
.snapshot_channel
.key
,
1642 msg
.u
.snapshot_channel
.pathname
,
1643 msg
.u
.snapshot_channel
.relayd_id
,
1646 ERR("Snapshot metadata failed");
1647 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
1650 ret
= snapshot_channel(msg
.u
.snapshot_channel
.key
,
1651 msg
.u
.snapshot_channel
.pathname
,
1652 msg
.u
.snapshot_channel
.relayd_id
,
1653 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
1656 ERR("Snapshot channel failed");
1657 ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
1661 health_code_update();
1662 ret
= consumer_send_status_msg(sock
, ret_code
);
1664 /* Somehow, the session daemon is not responding anymore. */
1667 health_code_update();
1677 health_code_update();
1680 * Return 1 to indicate success since the 0 value can be a socket
1681 * shutdown during the recv() or send() call.
1687 * The returned value here is not useful since either way we'll return 1 to
1688 * the caller because the session daemon socket management is done
1689 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1691 ret
= consumer_send_status_msg(sock
, ret_code
);
1697 health_code_update();
1703 * Free channel here since no one has a reference to it. We don't
1704 * free after that because a stream can store this pointer.
1706 destroy_channel(channel
);
1708 /* We have to send a status channel message indicating an error. */
1709 ret
= consumer_send_status_channel(sock
, NULL
);
1711 /* Stop everything if session daemon can not be notified. */
1716 health_code_update();
1721 /* This will issue a consumer stop. */
1726 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1727 * compiled out, we isolate it in this library.
1729 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream
*stream
,
1733 assert(stream
->ustream
);
1735 return ustctl_get_mmap_read_offset(stream
->ustream
, off
);
1739 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1740 * compiled out, we isolate it in this library.
1742 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream
*stream
)
1745 assert(stream
->ustream
);
1747 return ustctl_get_mmap_base(stream
->ustream
);
1751 * Take a snapshot for a specific fd
1753 * Returns 0 on success, < 0 on error
1755 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
1758 assert(stream
->ustream
);
1760 return ustctl_snapshot(stream
->ustream
);
1764 * Get the produced position
1766 * Returns 0 on success, < 0 on error
1768 int lttng_ustconsumer_get_produced_snapshot(
1769 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
1772 assert(stream
->ustream
);
1775 return ustctl_snapshot_get_produced(stream
->ustream
, pos
);
1779 * Get the consumed position
1781 * Returns 0 on success, < 0 on error
1783 int lttng_ustconsumer_get_consumed_snapshot(
1784 struct lttng_consumer_stream
*stream
, unsigned long *pos
)
1787 assert(stream
->ustream
);
1790 return ustctl_snapshot_get_consumed(stream
->ustream
, pos
);
1793 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream
*stream
,
1797 assert(stream
->ustream
);
1799 ustctl_flush_buffer(stream
->ustream
, producer
);
1802 int lttng_ustconsumer_get_current_timestamp(
1803 struct lttng_consumer_stream
*stream
, uint64_t *ts
)
1806 assert(stream
->ustream
);
1809 return ustctl_get_current_timestamp(stream
->ustream
, ts
);
1813 * Called when the stream signal the consumer that it has hang up.
1815 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream
*stream
)
1818 assert(stream
->ustream
);
1820 ustctl_flush_buffer(stream
->ustream
, 0);
1821 stream
->hangup_flush_done
= 1;
1824 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel
*chan
)
1829 assert(chan
->uchan
);
1831 if (chan
->switch_timer_enabled
== 1) {
1832 consumer_timer_switch_stop(chan
);
1834 for (i
= 0; i
< chan
->nr_stream_fds
; i
++) {
1837 ret
= close(chan
->stream_fds
[i
]);
1841 if (chan
->shm_path
[0]) {
1842 char shm_path
[PATH_MAX
];
1844 ret
= get_stream_shm_path(shm_path
, chan
->shm_path
, i
);
1846 ERR("Cannot get stream shm path");
1848 ret
= run_as_unlink(shm_path
, chan
->uid
, chan
->gid
);
1850 PERROR("unlink %s", shm_path
);
1854 /* Try to rmdir all directories under shm_path root. */
1855 if (chan
->root_shm_path
[0]) {
1856 (void) run_as_recursive_rmdir(chan
->root_shm_path
,
1857 chan
->uid
, chan
->gid
);
1861 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel
*chan
)
1864 assert(chan
->uchan
);
1866 consumer_metadata_cache_destroy(chan
);
1867 ustctl_destroy_channel(chan
->uchan
);
1868 free(chan
->stream_fds
);
1871 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream
*stream
)
1874 assert(stream
->ustream
);
1876 if (stream
->chan
->switch_timer_enabled
== 1) {
1877 consumer_timer_switch_stop(stream
->chan
);
1879 ustctl_destroy_stream(stream
->ustream
);
1882 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream
*stream
)
1885 assert(stream
->ustream
);
1887 return ustctl_stream_get_wakeup_fd(stream
->ustream
);
1890 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream
*stream
)
1893 assert(stream
->ustream
);
1895 return ustctl_stream_close_wakeup_fd(stream
->ustream
);
1899 * Populate index values of a UST stream. Values are set in big endian order.
1901 * Return 0 on success or else a negative value.
1903 static int get_index_values(struct ctf_packet_index
*index
,
1904 struct ustctl_consumer_stream
*ustream
)
1908 ret
= ustctl_get_timestamp_begin(ustream
, &index
->timestamp_begin
);
1910 PERROR("ustctl_get_timestamp_begin");
1913 index
->timestamp_begin
= htobe64(index
->timestamp_begin
);
1915 ret
= ustctl_get_timestamp_end(ustream
, &index
->timestamp_end
);
1917 PERROR("ustctl_get_timestamp_end");
1920 index
->timestamp_end
= htobe64(index
->timestamp_end
);
1922 ret
= ustctl_get_events_discarded(ustream
, &index
->events_discarded
);
1924 PERROR("ustctl_get_events_discarded");
1927 index
->events_discarded
= htobe64(index
->events_discarded
);
1929 ret
= ustctl_get_content_size(ustream
, &index
->content_size
);
1931 PERROR("ustctl_get_content_size");
1934 index
->content_size
= htobe64(index
->content_size
);
1936 ret
= ustctl_get_packet_size(ustream
, &index
->packet_size
);
1938 PERROR("ustctl_get_packet_size");
1941 index
->packet_size
= htobe64(index
->packet_size
);
1943 ret
= ustctl_get_stream_id(ustream
, &index
->stream_id
);
1945 PERROR("ustctl_get_stream_id");
1948 index
->stream_id
= htobe64(index
->stream_id
);
1955 * Write up to one packet from the metadata cache to the channel.
1957 * Returns the number of bytes pushed in the cache, or a negative value
1961 int commit_one_metadata_packet(struct lttng_consumer_stream
*stream
)
1966 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
1967 if (stream
->chan
->metadata_cache
->max_offset
1968 == stream
->ust_metadata_pushed
) {
1973 write_len
= ustctl_write_one_packet_to_channel(stream
->chan
->uchan
,
1974 &stream
->chan
->metadata_cache
->data
[stream
->ust_metadata_pushed
],
1975 stream
->chan
->metadata_cache
->max_offset
1976 - stream
->ust_metadata_pushed
);
1977 assert(write_len
!= 0);
1978 if (write_len
< 0) {
1979 ERR("Writing one metadata packet");
1983 stream
->ust_metadata_pushed
+= write_len
;
1985 assert(stream
->chan
->metadata_cache
->max_offset
>=
1986 stream
->ust_metadata_pushed
);
1990 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
1996 * Sync metadata meaning request them to the session daemon and snapshot to the
1997 * metadata thread can consumer them.
1999 * Metadata stream lock is held here, but we need to release it when
2000 * interacting with sessiond, else we cause a deadlock with live
2001 * awaiting on metadata to be pushed out.
2003 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2004 * is empty or a negative value on error.
2006 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data
*ctx
,
2007 struct lttng_consumer_stream
*metadata
)
2015 pthread_mutex_unlock(&metadata
->lock
);
2017 * Request metadata from the sessiond, but don't wait for the flush
2018 * because we locked the metadata thread.
2020 ret
= lttng_ustconsumer_request_metadata(ctx
, metadata
->chan
, 0, 0);
2024 pthread_mutex_lock(&metadata
->lock
);
2026 ret
= commit_one_metadata_packet(metadata
);
2029 } else if (ret
> 0) {
2033 ustctl_flush_buffer(metadata
->ustream
, 1);
2034 ret
= ustctl_snapshot(metadata
->ustream
);
2036 if (errno
!= EAGAIN
) {
2037 ERR("Sync metadata, taking UST snapshot");
2040 DBG("No new metadata when syncing them.");
2041 /* No new metadata, exit. */
2047 * After this flush, we still need to extract metadata.
2058 * Return 0 on success else a negative value.
2060 static int notify_if_more_data(struct lttng_consumer_stream
*stream
,
2061 struct lttng_consumer_local_data
*ctx
)
2064 struct ustctl_consumer_stream
*ustream
;
2069 ustream
= stream
->ustream
;
2072 * First, we are going to check if there is a new subbuffer available
2073 * before reading the stream wait_fd.
2075 /* Get the next subbuffer */
2076 ret
= ustctl_get_next_subbuf(ustream
);
2078 /* No more data found, flag the stream. */
2079 stream
->has_data
= 0;
2084 ret
= ustctl_put_subbuf(ustream
);
2087 /* This stream still has data. Flag it and wake up the data thread. */
2088 stream
->has_data
= 1;
2090 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !ctx
->has_wakeup
) {
2093 writelen
= lttng_pipe_write(ctx
->consumer_wakeup_pipe
, "!", 1);
2094 if (writelen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2099 /* The wake up pipe has been notified. */
2100 ctx
->has_wakeup
= 1;
2109 * Read subbuffer from the given stream.
2111 * Stream lock MUST be acquired.
2113 * Return 0 on success else a negative value.
2115 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
2116 struct lttng_consumer_local_data
*ctx
)
2118 unsigned long len
, subbuf_size
, padding
;
2119 int err
, write_index
= 1;
2121 struct ustctl_consumer_stream
*ustream
;
2122 struct ctf_packet_index index
;
2125 assert(stream
->ustream
);
2128 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream
->wait_fd
,
2131 /* Ease our life for what's next. */
2132 ustream
= stream
->ustream
;
2135 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2136 * error if we cannot read this one byte (read returns 0), or if the error
2137 * is EAGAIN or EWOULDBLOCK.
2139 * This is only done when the stream is monitored by a thread, before the
2140 * flush is done after a hangup and if the stream is not flagged with data
2141 * since there might be nothing to consume in the wait fd but still have
2142 * data available flagged by the consumer wake up pipe.
2144 if (stream
->monitor
&& !stream
->hangup_flush_done
&& !stream
->has_data
) {
2148 readlen
= lttng_read(stream
->wait_fd
, &dummy
, 1);
2149 if (readlen
< 0 && errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2156 /* Get the next subbuffer */
2157 err
= ustctl_get_next_subbuf(ustream
);
2160 * Populate metadata info if the existing info has
2161 * already been read.
2163 if (stream
->metadata_flag
) {
2164 ret
= commit_one_metadata_packet(stream
);
2168 ustctl_flush_buffer(stream
->ustream
, 1);
2172 ret
= err
; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2174 * This is a debug message even for single-threaded consumer,
2175 * because poll() have more relaxed criterions than get subbuf,
2176 * so get_subbuf may fail for short race windows where poll()
2177 * would issue wakeups.
2179 DBG("Reserving sub buffer failed (everything is normal, "
2180 "it is due to concurrency) [ret: %d]", err
);
2183 assert(stream
->chan
->output
== CONSUMER_CHANNEL_MMAP
);
2185 if (!stream
->metadata_flag
) {
2186 index
.offset
= htobe64(stream
->out_fd_offset
);
2187 ret
= get_index_values(&index
, ustream
);
2195 /* Get the full padded subbuffer size */
2196 err
= ustctl_get_padded_subbuf_size(ustream
, &len
);
2199 /* Get subbuffer data size (without padding) */
2200 err
= ustctl_get_subbuf_size(ustream
, &subbuf_size
);
2203 /* Make sure we don't get a subbuffer size bigger than the padded */
2204 assert(len
>= subbuf_size
);
2206 padding
= len
- subbuf_size
;
2207 /* write the subbuffer to the tracefile */
2208 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
, padding
, &index
);
2210 * The mmap operation should write subbuf_size amount of data when network
2211 * streaming or the full padding (len) size when we are _not_ streaming.
2213 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
2214 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
2216 * Display the error but continue processing to try to release the
2217 * subbuffer. This is a DBG statement since any unexpected kill or
2218 * signal, the application gets unregistered, relayd gets closed or
2219 * anything that affects the buffer lifetime will trigger this error.
2220 * So, for the sake of the user, don't print this error since it can
2221 * happen and it is OK with the code flow.
2223 DBG("Error writing to tracefile "
2224 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2225 ret
, len
, subbuf_size
);
2228 err
= ustctl_put_next_subbuf(ustream
);
2232 * This will consumer the byte on the wait_fd if and only if there is not
2233 * next subbuffer to be acquired.
2235 if (!stream
->metadata_flag
) {
2236 ret
= notify_if_more_data(stream
, ctx
);
2242 /* Write index if needed. */
2247 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
2249 * In live, block until all the metadata is sent.
2251 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2252 assert(!stream
->missed_metadata_flush
);
2253 stream
->waiting_on_metadata
= true;
2254 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2256 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
2258 pthread_mutex_lock(&stream
->metadata_timer_lock
);
2259 stream
->waiting_on_metadata
= false;
2260 if (stream
->missed_metadata_flush
) {
2261 stream
->missed_metadata_flush
= false;
2262 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2263 (void) consumer_flush_ust_index(stream
);
2265 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
2273 assert(!stream
->metadata_flag
);
2274 err
= consumer_stream_write_index(stream
, &index
);
2284 * Called when a stream is created.
2286 * Return 0 on success or else a negative value.
2288 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
2294 /* Don't create anything if this is set for streaming. */
2295 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
) {
2296 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
2297 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
2298 stream
->uid
, stream
->gid
, NULL
);
2302 stream
->out_fd
= ret
;
2303 stream
->tracefile_size_current
= 0;
2305 if (!stream
->metadata_flag
) {
2306 ret
= index_create_file(stream
->chan
->pathname
,
2307 stream
->name
, stream
->uid
, stream
->gid
,
2308 stream
->chan
->tracefile_size
,
2309 stream
->tracefile_count_current
);
2313 stream
->index_fd
= ret
;
2323 * Check if data is still being extracted from the buffers for a specific
2324 * stream. Consumer data lock MUST be acquired before calling this function
2325 * and the stream lock.
2327 * Return 1 if the traced data are still getting read else 0 meaning that the
2328 * data is available for trace viewer reading.
2330 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream
*stream
)
2335 assert(stream
->ustream
);
2337 DBG("UST consumer checking data pending");
2339 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
2344 if (stream
->chan
->type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
2345 uint64_t contiguous
, pushed
;
2347 /* Ease our life a bit. */
2348 contiguous
= stream
->chan
->metadata_cache
->max_offset
;
2349 pushed
= stream
->ust_metadata_pushed
;
2352 * We can simply check whether all contiguously available data
2353 * has been pushed to the ring buffer, since the push operation
2354 * is performed within get_next_subbuf(), and because both
2355 * get_next_subbuf() and put_next_subbuf() are issued atomically
2356 * thanks to the stream lock within
2357 * lttng_ustconsumer_read_subbuffer(). This basically means that
2358 * whetnever ust_metadata_pushed is incremented, the associated
2359 * metadata has been consumed from the metadata stream.
2361 DBG("UST consumer metadata pending check: contiguous %" PRIu64
" vs pushed %" PRIu64
,
2362 contiguous
, pushed
);
2363 assert(((int64_t) (contiguous
- pushed
)) >= 0);
2364 if ((contiguous
!= pushed
) ||
2365 (((int64_t) contiguous
- pushed
) > 0 || contiguous
== 0)) {
2366 ret
= 1; /* Data is pending */
2370 ret
= ustctl_get_next_subbuf(stream
->ustream
);
2373 * There is still data so let's put back this
2376 ret
= ustctl_put_subbuf(stream
->ustream
);
2378 ret
= 1; /* Data is pending */
2383 /* Data is NOT pending so ready to be read. */
2391 * Stop a given metadata channel timer if enabled and close the wait fd which
2392 * is the poll pipe of the metadata stream.
2394 * This MUST be called with the metadata channel acquired.
2396 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel
*metadata
)
2401 assert(metadata
->type
== CONSUMER_CHANNEL_TYPE_METADATA
);
2403 DBG("Closing metadata channel key %" PRIu64
, metadata
->key
);
2405 if (metadata
->switch_timer_enabled
== 1) {
2406 consumer_timer_switch_stop(metadata
);
2409 if (!metadata
->metadata_stream
) {
2414 * Closing write side so the thread monitoring the stream wakes up if any
2415 * and clean the metadata stream.
2417 if (metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] >= 0) {
2418 ret
= close(metadata
->metadata_stream
->ust_metadata_poll_pipe
[1]);
2420 PERROR("closing metadata pipe write side");
2422 metadata
->metadata_stream
->ust_metadata_poll_pipe
[1] = -1;
2430 * Close every metadata stream wait fd of the metadata hash table. This
2431 * function MUST be used very carefully so not to run into a race between the
2432 * metadata thread handling streams and this function closing their wait fd.
2434 * For UST, this is used when the session daemon hangs up. Its the metadata
2435 * producer so calling this is safe because we are assured that no state change
2436 * can occur in the metadata thread for the streams in the hash table.
2438 void lttng_ustconsumer_close_all_metadata(struct lttng_ht
*metadata_ht
)
2440 struct lttng_ht_iter iter
;
2441 struct lttng_consumer_stream
*stream
;
2443 assert(metadata_ht
);
2444 assert(metadata_ht
->ht
);
2446 DBG("UST consumer closing all metadata streams");
2449 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
,
2452 health_code_update();
2454 pthread_mutex_lock(&stream
->chan
->lock
);
2455 lttng_ustconsumer_close_metadata(stream
->chan
);
2456 pthread_mutex_unlock(&stream
->chan
->lock
);
2462 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream
*stream
)
2466 ret
= ustctl_stream_close_wakeup_fd(stream
->ustream
);
2468 ERR("Unable to close wakeup fd");
2473 * Please refer to consumer-timer.c before adding any lock within this
2474 * function or any of its callees. Timers have a very strict locking
2475 * semantic with respect to teardown. Failure to respect this semantic
2476 * introduces deadlocks.
2478 * DON'T hold the metadata lock when calling this function, else this
2479 * can cause deadlock involving consumer awaiting for metadata to be
2480 * pushed out due to concurrent interaction with the session daemon.
2482 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data
*ctx
,
2483 struct lttng_consumer_channel
*channel
, int timer
, int wait
)
2485 struct lttcomm_metadata_request_msg request
;
2486 struct lttcomm_consumer_msg msg
;
2487 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
2488 uint64_t len
, key
, offset
;
2492 assert(channel
->metadata_cache
);
2494 memset(&request
, 0, sizeof(request
));
2496 /* send the metadata request to sessiond */
2497 switch (consumer_data
.type
) {
2498 case LTTNG_CONSUMER64_UST
:
2499 request
.bits_per_long
= 64;
2501 case LTTNG_CONSUMER32_UST
:
2502 request
.bits_per_long
= 32;
2505 request
.bits_per_long
= 0;
2509 request
.session_id
= channel
->session_id
;
2510 request
.session_id_per_pid
= channel
->session_id_per_pid
;
2512 * Request the application UID here so the metadata of that application can
2513 * be sent back. The channel UID corresponds to the user UID of the session
2514 * used for the rights on the stream file(s).
2516 request
.uid
= channel
->ust_app_uid
;
2517 request
.key
= channel
->key
;
2519 DBG("Sending metadata request to sessiond, session id %" PRIu64
2520 ", per-pid %" PRIu64
", app UID %u and channek key %" PRIu64
,
2521 request
.session_id
, request
.session_id_per_pid
, request
.uid
,
2524 pthread_mutex_lock(&ctx
->metadata_socket_lock
);
2526 health_code_update();
2528 ret
= lttcomm_send_unix_sock(ctx
->consumer_metadata_socket
, &request
,
2531 ERR("Asking metadata to sessiond");
2535 health_code_update();
2537 /* Receive the metadata from sessiond */
2538 ret
= lttcomm_recv_unix_sock(ctx
->consumer_metadata_socket
, &msg
,
2540 if (ret
!= sizeof(msg
)) {
2541 DBG("Consumer received unexpected message size %d (expects %zu)",
2543 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
2545 * The ret value might 0 meaning an orderly shutdown but this is ok
2546 * since the caller handles this.
2551 health_code_update();
2553 if (msg
.cmd_type
== LTTNG_ERR_UND
) {
2554 /* No registry found */
2555 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
,
2559 } else if (msg
.cmd_type
!= LTTNG_CONSUMER_PUSH_METADATA
) {
2560 ERR("Unexpected cmd_type received %d", msg
.cmd_type
);
2565 len
= msg
.u
.push_metadata
.len
;
2566 key
= msg
.u
.push_metadata
.key
;
2567 offset
= msg
.u
.push_metadata
.target_offset
;
2569 assert(key
== channel
->key
);
2571 DBG("No new metadata to receive for key %" PRIu64
, key
);
2574 health_code_update();
2576 /* Tell session daemon we are ready to receive the metadata. */
2577 ret
= consumer_send_status_msg(ctx
->consumer_metadata_socket
,
2578 LTTCOMM_CONSUMERD_SUCCESS
);
2579 if (ret
< 0 || len
== 0) {
2581 * Somehow, the session daemon is not responding anymore or there is
2582 * nothing to receive.
2587 health_code_update();
2589 ret
= lttng_ustconsumer_recv_metadata(ctx
->consumer_metadata_socket
,
2590 key
, offset
, len
, channel
, timer
, wait
);
2593 * Only send the status msg if the sessiond is alive meaning a positive
2596 (void) consumer_send_status_msg(ctx
->consumer_metadata_socket
, ret
);
2601 health_code_update();
2603 pthread_mutex_unlock(&ctx
->metadata_socket_lock
);
2608 * Return the ustctl call for the get stream id.
2610 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream
*stream
,
2611 uint64_t *stream_id
)
2616 return ustctl_get_stream_id(stream
->ustream
, stream_id
);