2 * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@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, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
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.
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
33 #include <sys/types.h>
36 #include <urcu/futex.h>
37 #include <urcu/uatomic.h>
42 #include <lttng/lttng.h>
43 #include <common/common.h>
44 #include <common/compat/poll.h>
45 #include <common/compat/socket.h>
46 #include <common/defaults.h>
47 #include <common/futex.h>
48 #include <common/sessiond-comm/sessiond-comm.h>
49 #include <common/sessiond-comm/inet.h>
50 #include <common/sessiond-comm/relayd.h>
51 #include <common/uri.h>
52 #include <common/utils.h>
56 #include "lttng-relayd.h"
57 #include "lttng-viewer.h"
59 #include "health-relayd.h"
61 static struct lttng_uri
*live_uri
;
64 * Quit pipe for all threads. This permits a single cancellation point
65 * for all threads when receiving an event on the pipe.
67 static int live_thread_quit_pipe
[2] = { -1, -1 };
70 * This pipe is used to inform the worker thread that a command is queued and
71 * ready to be processed.
73 static int live_relay_cmd_pipe
[2] = { -1, -1 };
75 /* Shared between threads */
76 static int live_dispatch_thread_exit
;
78 static pthread_t live_listener_thread
;
79 static pthread_t live_dispatcher_thread
;
80 static pthread_t live_worker_thread
;
83 * Relay command queue.
85 * The live_thread_listener and live_thread_dispatcher communicate with this
88 static struct relay_cmd_queue viewer_cmd_queue
;
90 static uint64_t last_relay_viewer_session_id
;
104 * Write to writable pipe used to notify a thread.
107 int notify_thread_pipe(int wpipe
)
111 ret
= lttng_write(wpipe
, "!", 1);
113 PERROR("write poll pipe");
120 * Stop all threads by closing the thread quit pipe.
123 void stop_threads(void)
127 /* Stopping all threads */
128 DBG("Terminating all live threads");
129 ret
= notify_thread_pipe(live_thread_quit_pipe
[1]);
131 ERR("write error on thread quit pipe");
134 /* Dispatch thread */
135 CMM_STORE_SHARED(live_dispatch_thread_exit
, 1);
136 futex_nto1_wake(&viewer_cmd_queue
.futex
);
140 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
143 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
147 if (events
== NULL
|| size
== 0) {
152 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
158 ret
= lttng_poll_add(events
, live_thread_quit_pipe
[0], LPOLLIN
);
170 * Check if the thread quit pipe was triggered.
172 * Return 1 if it was triggered else 0;
175 int check_thread_quit_pipe(int fd
, uint32_t events
)
177 if (fd
== live_thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
185 * Create and init socket from uri.
188 struct lttcomm_sock
*init_socket(struct lttng_uri
*uri
)
191 struct lttcomm_sock
*sock
= NULL
;
193 sock
= lttcomm_alloc_sock_from_uri(uri
);
195 ERR("Allocating socket");
199 ret
= lttcomm_create_sock(sock
);
203 DBG("Listening on sock %d for live", sock
->fd
);
205 ret
= sock
->ops
->bind(sock
);
210 ret
= sock
->ops
->listen(sock
, -1);
220 lttcomm_destroy_sock(sock
);
226 * This thread manages the listening for new connections on the network
229 void *thread_listener(void *data
)
231 int i
, ret
, pollfd
, err
= -1;
233 uint32_t revents
, nb_fd
;
234 struct lttng_poll_event events
;
235 struct lttcomm_sock
*live_control_sock
;
237 DBG("[thread] Relay live listener started");
239 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_LISTENER
);
241 health_code_update();
243 live_control_sock
= init_socket(live_uri
);
244 if (!live_control_sock
) {
245 goto error_sock_control
;
249 * Pass 3 as size here for the thread quit pipe, control and data socket.
251 ret
= create_thread_poll_set(&events
, 2);
253 goto error_create_poll
;
256 /* Add the control socket */
257 ret
= lttng_poll_add(&events
, live_control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
263 health_code_update();
265 DBG("Listener accepting live viewers connections");
269 ret
= lttng_poll_wait(&events
, -1);
273 * Restart interrupted system call.
275 if (errno
== EINTR
) {
282 DBG("Relay new viewer connection received");
283 for (i
= 0; i
< nb_fd
; i
++) {
284 health_code_update();
286 /* Fetch once the poll data */
287 revents
= LTTNG_POLL_GETEV(&events
, i
);
288 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
290 /* Thread quit pipe has been closed. Killing thread. */
291 ret
= check_thread_quit_pipe(pollfd
, revents
);
297 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
298 ERR("socket poll error");
300 } else if (revents
& LPOLLIN
) {
302 * Get allocated in this thread, enqueued to a global queue,
303 * dequeued and freed in the worker thread.
305 struct relay_command
*relay_cmd
;
306 struct lttcomm_sock
*newsock
;
308 relay_cmd
= zmalloc(sizeof(*relay_cmd
));
310 PERROR("relay command zmalloc");
314 assert(pollfd
== live_control_sock
->fd
);
315 newsock
= live_control_sock
->ops
->accept(live_control_sock
);
317 PERROR("accepting control sock");
321 DBG("Relay viewer connection accepted socket %d", newsock
->fd
);
322 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
325 PERROR("setsockopt inet");
326 lttcomm_destroy_sock(newsock
);
330 relay_cmd
->sock
= newsock
;
333 * Lock free enqueue the request.
335 cds_wfq_enqueue(&viewer_cmd_queue
.queue
, &relay_cmd
->node
);
338 * Wake the dispatch queue futex. Implicit memory
339 * barrier with the exchange in cds_wfq_enqueue.
341 futex_nto1_wake(&viewer_cmd_queue
.futex
);
349 lttng_poll_clean(&events
);
351 if (live_control_sock
->fd
>= 0) {
352 ret
= live_control_sock
->ops
->close(live_control_sock
);
357 lttcomm_destroy_sock(live_control_sock
);
361 DBG("Live viewer listener thread exited with error");
363 health_unregister(health_relayd
);
364 DBG("Live viewer listener thread cleanup complete");
370 * This thread manages the dispatching of the requests to worker threads
373 void *thread_dispatcher(void *data
)
377 struct cds_wfq_node
*node
;
378 struct relay_command
*relay_cmd
= NULL
;
380 DBG("[thread] Live viewer relay dispatcher started");
382 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER
);
384 health_code_update();
386 while (!CMM_LOAD_SHARED(live_dispatch_thread_exit
)) {
387 health_code_update();
389 /* Atomically prepare the queue futex */
390 futex_nto1_prepare(&viewer_cmd_queue
.futex
);
393 health_code_update();
395 /* Dequeue commands */
396 node
= cds_wfq_dequeue_blocking(&viewer_cmd_queue
.queue
);
398 DBG("Woken up but nothing in the live-viewer "
399 "relay command queue");
400 /* Continue thread execution */
404 relay_cmd
= caa_container_of(node
, struct relay_command
, node
);
405 DBG("Dispatching viewer request waiting on sock %d",
406 relay_cmd
->sock
->fd
);
409 * Inform worker thread of the new request. This call is blocking
410 * so we can be assured that the data will be read at some point in
411 * time or wait to the end of the world :)
413 ret
= lttng_write(live_relay_cmd_pipe
[1], relay_cmd
,
416 if (ret
< sizeof(struct relay_command
)) {
417 PERROR("write cmd pipe");
420 } while (node
!= NULL
);
422 /* Futex wait on queue. Blocking call on futex() */
424 futex_nto1_wait(&viewer_cmd_queue
.futex
);
428 /* Normal exit, no error */
434 ERR("Health error occurred in %s", __func__
);
436 health_unregister(health_relayd
);
437 DBG("Live viewer dispatch thread dying");
443 * Establish connection with the viewer and check the versions.
445 * Return 0 on success or else negative value.
448 int viewer_connect(struct relay_command
*cmd
)
451 struct lttng_viewer_connect reply
, msg
;
455 cmd
->version_check_done
= 1;
457 health_code_update();
459 /* Get version from the other side. */
460 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
461 if (ret
< 0 || ret
!= sizeof(msg
)) {
463 /* Orderly shutdown. Not necessary to print an error. */
464 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
466 ERR("Relay failed to receive the version values.");
472 health_code_update();
474 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
475 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
477 /* Major versions must be the same */
478 if (reply
.major
!= be32toh(msg
.major
)) {
479 DBG("Incompatible major versions (%u vs %u)", reply
.major
,
485 cmd
->major
= reply
.major
;
486 /* We adapt to the lowest compatible version */
487 if (reply
.minor
<= be32toh(msg
.minor
)) {
488 cmd
->minor
= reply
.minor
;
490 cmd
->minor
= be32toh(msg
.minor
);
493 if (be32toh(msg
.type
) == VIEWER_CLIENT_COMMAND
) {
494 cmd
->type
= RELAY_VIEWER_COMMAND
;
495 } else if (be32toh(msg
.type
) == VIEWER_CLIENT_NOTIFICATION
) {
496 cmd
->type
= RELAY_VIEWER_NOTIFICATION
;
498 ERR("Unknown connection type : %u", be32toh(msg
.type
));
503 reply
.major
= htobe32(reply
.major
);
504 reply
.minor
= htobe32(reply
.minor
);
505 if (cmd
->type
== RELAY_VIEWER_COMMAND
) {
506 reply
.viewer_session_id
= htobe64(++last_relay_viewer_session_id
);
509 health_code_update();
511 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
512 sizeof(struct lttng_viewer_connect
), 0);
514 ERR("Relay sending version");
517 health_code_update();
519 DBG("Version check done using protocol %u.%u", cmd
->major
, cmd
->minor
);
527 * Send the viewer the list of current sessions.
529 * Return 0 on success or else a negative value.
532 int viewer_list_sessions(struct relay_command
*cmd
,
533 struct lttng_ht
*sessions_ht
)
536 struct lttng_viewer_list_sessions session_list
;
538 long approx_before
, approx_after
;
539 struct lttng_ht_node_ulong
*node
;
540 struct lttng_ht_iter iter
;
541 struct lttng_viewer_session send_session
;
542 struct relay_session
*session
;
544 DBG("List sessions received");
546 if (cmd
->version_check_done
== 0) {
547 ERR("Trying to list sessions before version check");
553 cds_lfht_count_nodes(sessions_ht
->ht
, &approx_before
, &count
, &approx_after
);
554 session_list
.sessions_count
= htobe32(count
);
556 health_code_update();
558 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &session_list
,
559 sizeof(session_list
), 0);
561 ERR("Relay sending sessions list");
565 health_code_update();
567 cds_lfht_for_each_entry(sessions_ht
->ht
, &iter
.iter
, node
, node
) {
568 health_code_update();
570 node
= lttng_ht_iter_get_node_ulong(&iter
);
574 session
= caa_container_of(node
, struct relay_session
, session_n
);
576 strncpy(send_session
.session_name
, session
->session_name
,
577 sizeof(send_session
.session_name
));
578 strncpy(send_session
.hostname
, session
->hostname
,
579 sizeof(send_session
.hostname
));
580 send_session
.id
= htobe64(session
->id
);
581 send_session
.live_timer
= htobe32(session
->live_timer
);
582 send_session
.clients
= htobe32(session
->viewer_attached
);
583 send_session
.streams
= htobe32(session
->stream_count
);
585 health_code_update();
587 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &send_session
,
588 sizeof(send_session
), 0);
590 ERR("Relay sending session info");
594 health_code_update();
609 * Open index file using a given viewer stream.
611 * Return 0 on success or else a negative value.
613 static int open_index(struct relay_viewer_stream
*stream
)
616 char fullpath
[PATH_MAX
];
617 struct lttng_packet_index_file_hdr hdr
;
619 if (stream
->tracefile_size
> 0) {
620 /* For now we don't support on-disk ring buffer. */
625 ret
= snprintf(fullpath
, sizeof(fullpath
), "%s/" DEFAULT_INDEX_DIR
"/%s"
626 DEFAULT_INDEX_FILE_SUFFIX
, stream
->path_name
,
627 stream
->channel_name
);
629 PERROR("snprintf index path");
633 DBG("Opening index file %s in read only", fullpath
);
634 ret
= open(fullpath
, O_RDONLY
);
636 if (errno
== ENOENT
) {
640 PERROR("opening index in read-only");
644 stream
->index_read_fd
= ret
;
645 DBG("Opening index file %s in read only, (fd: %d)", fullpath
, ret
);
647 ret
= lttng_read(stream
->index_read_fd
, &hdr
, sizeof(hdr
));
648 if (ret
< sizeof(hdr
)) {
649 PERROR("Reading index header");
652 if (strncmp(hdr
.magic
, INDEX_MAGIC
, sizeof(hdr
.magic
)) != 0) {
653 ERR("Invalid header magic");
657 if (be32toh(hdr
.index_major
) != INDEX_MAJOR
||
658 be32toh(hdr
.index_minor
) != INDEX_MINOR
) {
659 ERR("Invalid header version");
671 * Allocate and init a new viewer_stream.
673 * Copies the values from the stream passed in parameter and insert the new
674 * stream in the viewer_streams_ht.
676 * MUST be called with rcu_read_lock held.
678 * Returns 0 on success or a negative value on error.
681 int init_viewer_stream(struct relay_stream
*stream
, int seek_last
)
684 struct relay_viewer_stream
*viewer_stream
;
688 viewer_stream
= zmalloc(sizeof(*viewer_stream
));
689 if (!viewer_stream
) {
690 PERROR("relay viewer stream zmalloc");
695 viewer_stream
->read_fd
= -1;
696 viewer_stream
->index_read_fd
= -1;
697 viewer_stream
->session_id
= stream
->session
->id
;
698 viewer_stream
->stream_handle
= stream
->stream_handle
;
699 viewer_stream
->path_name
= strndup(stream
->path_name
,
700 LTTNG_VIEWER_PATH_MAX
);
701 viewer_stream
->channel_name
= strndup(stream
->channel_name
,
702 LTTNG_VIEWER_NAME_MAX
);
703 viewer_stream
->total_index_received
= stream
->total_index_received
;
704 viewer_stream
->tracefile_size
= stream
->tracefile_size
;
705 viewer_stream
->tracefile_count
= stream
->tracefile_count
;
706 viewer_stream
->metadata_flag
= stream
->metadata_flag
;
708 if (seek_last
&& viewer_stream
->total_index_received
> 0) {
709 ret
= open_index(viewer_stream
);
713 ret
= lseek(viewer_stream
->index_read_fd
,
714 viewer_stream
->total_index_received
*
715 sizeof(struct lttng_packet_index
),
720 viewer_stream
->last_sent_index
=
721 viewer_stream
->total_index_received
;
725 * This is to avoid a race between the initialization of this object and
726 * the close of the given stream. If the stream is unable to find this
727 * viewer stream when closing, this copy will at least take the latest
730 viewer_stream
->total_index_received
= stream
->total_index_received
;
733 * The deletion of this ctf_trace object is only done in a call RCU of the
734 * relay stream making it valid as long as we have the read side lock.
736 viewer_stream
->ctf_trace
= stream
->ctf_trace
;
737 uatomic_inc(&viewer_stream
->ctf_trace
->refcount
);
739 lttng_ht_node_init_u64(&viewer_stream
->stream_n
, stream
->stream_handle
);
740 lttng_ht_add_unique_u64(viewer_streams_ht
, &viewer_stream
->stream_n
);
749 * Send the viewer the list of current sessions.
752 int viewer_attach_session(struct relay_command
*cmd
,
753 struct lttng_ht
*sessions_ht
)
755 int ret
, send_streams
= 0, nb_streams
= 0;
756 struct lttng_viewer_attach_session_request request
;
757 struct lttng_viewer_attach_session_response response
;
758 struct lttng_viewer_stream send_stream
;
759 struct relay_stream
*stream
;
760 struct relay_viewer_stream
*viewer_stream
;
761 struct lttng_ht_node_ulong
*node
;
762 struct lttng_ht_node_u64
*node64
;
763 struct lttng_ht_iter iter
;
764 struct relay_session
*session
;
770 DBG("Attach session received");
772 if (cmd
->version_check_done
== 0) {
773 ERR("Trying to attach session before version check");
778 health_code_update();
780 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &request
, sizeof(request
), 0);
781 if (ret
< 0 || ret
!= sizeof(request
)) {
783 /* Orderly shutdown. Not necessary to print an error. */
784 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
786 ERR("Relay failed to receive the attach parameters.");
792 health_code_update();
795 lttng_ht_lookup(sessions_ht
,
796 (void *)((unsigned long) be64toh(request
.session_id
)), &iter
);
797 node
= lttng_ht_iter_get_node_ulong(&iter
);
799 DBG("Relay session %" PRIu64
" not found",
800 be64toh(request
.session_id
));
801 response
.status
= htobe32(VIEWER_ATTACH_UNK
);
805 session
= caa_container_of(node
, struct relay_session
, session_n
);
806 if (cmd
->session_id
== session
->id
) {
807 /* Same viewer already attached, just send the stream list. */
809 response
.status
= htobe32(VIEWER_ATTACH_OK
);
810 } else if (session
->viewer_attached
!= 0) {
811 DBG("Already a viewer attached");
812 response
.status
= htobe32(VIEWER_ATTACH_ALREADY
);
814 } else if (session
->live_timer
== 0) {
815 DBG("Not live session");
816 response
.status
= htobe32(VIEWER_ATTACH_NOT_LIVE
);
819 session
->viewer_attached
++;
821 response
.status
= htobe32(VIEWER_ATTACH_OK
);
822 cmd
->session_id
= session
->id
;
823 cmd
->session
= session
;
826 switch (be32toh(request
.seek
)) {
827 case VIEWER_SEEK_BEGINNING
:
828 /* Default behaviour. */
830 case VIEWER_SEEK_LAST
:
834 ERR("Wrong seek parameter");
835 response
.status
= htobe32(VIEWER_ATTACH_SEEK_ERR
);
841 /* We should only be there if we have a session to attach to. */
845 * Fill the viewer_streams_ht to count the number of streams
846 * ready to be sent and avoid concurrency issues on the
847 * relay_streams_ht and don't rely on a total session stream count.
849 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, node
, node
) {
850 struct relay_viewer_stream
*vstream
;
852 health_code_update();
854 node
= lttng_ht_iter_get_node_ulong(&iter
);
858 stream
= caa_container_of(node
, struct relay_stream
, stream_n
);
859 if (stream
->session
!= cmd
->session
) {
864 * Don't send streams with no ctf_trace, they are not ready to be
867 if (!stream
->ctf_trace
) {
871 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
);
873 ret
= init_viewer_stream(stream
, seek_last
);
880 response
.streams_count
= htobe32(nb_streams
);
884 health_code_update();
885 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &response
, sizeof(response
), 0);
887 ERR("Relay sending viewer attach response");
890 health_code_update();
893 * Unknown or busy session, just return gracefully, the viewer knows what
901 /* We should only be there if we have a session to attach to. */
903 cds_lfht_for_each_entry(viewer_streams_ht
->ht
, &iter
.iter
, node
, node
) {
904 health_code_update();
906 node64
= lttng_ht_iter_get_node_u64(&iter
);
910 viewer_stream
= caa_container_of(node64
, struct relay_viewer_stream
,
912 if (viewer_stream
->session_id
!= cmd
->session
->id
) {
916 send_stream
.id
= htobe64(viewer_stream
->stream_handle
);
917 send_stream
.ctf_trace_id
= htobe64(viewer_stream
->ctf_trace
->id
);
918 send_stream
.metadata_flag
= htobe32(viewer_stream
->metadata_flag
);
919 strncpy(send_stream
.path_name
, viewer_stream
->path_name
,
920 sizeof(send_stream
.path_name
));
921 strncpy(send_stream
.channel_name
, viewer_stream
->channel_name
,
922 sizeof(send_stream
.channel_name
));
924 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &send_stream
,
925 sizeof(send_stream
), 0);
927 ERR("Relay sending stream %" PRIu64
, viewer_stream
->stream_handle
);
930 DBG("Sent stream %" PRIu64
" to viewer", viewer_stream
->stream_handle
);
942 * Get viewer stream from stream id.
944 * RCU read side lock MUST be acquired.
946 struct relay_viewer_stream
*live_find_viewer_stream_by_id(uint64_t stream_id
)
948 struct lttng_ht_node_u64
*node
;
949 struct lttng_ht_iter iter
;
950 struct relay_viewer_stream
*stream
= NULL
;
952 lttng_ht_lookup(viewer_streams_ht
, &stream_id
, &iter
);
953 node
= lttng_ht_iter_get_node_u64(&iter
);
955 DBG("Relay viewer stream %" PRIu64
" not found", stream_id
);
958 stream
= caa_container_of(node
, struct relay_viewer_stream
, stream_n
);
965 * Send the next index for a stream.
967 * Return 0 on success or else a negative value.
970 int viewer_get_next_index(struct relay_command
*cmd
,
971 struct lttng_ht
*sessions_ht
)
974 struct lttng_viewer_get_next_index request_index
;
975 struct lttng_viewer_index viewer_index
;
976 struct lttng_packet_index packet_index
;
977 struct relay_viewer_stream
*vstream
;
978 struct relay_stream
*rstream
;
983 DBG("Viewer get next index");
985 if (cmd
->version_check_done
== 0) {
986 ERR("Trying to request index before version check");
991 health_code_update();
992 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &request_index
,
993 sizeof(request_index
), 0);
994 if (ret
< 0 || ret
!= sizeof(request_index
)) {
996 ERR("Relay didn't receive the whole packet");
999 health_code_update();
1002 vstream
= live_find_viewer_stream_by_id(be64toh(request_index
.stream_id
));
1008 memset(&viewer_index
, 0, sizeof(viewer_index
));
1011 * The viewer should not ask for index on metadata stream.
1013 if (vstream
->metadata_flag
) {
1014 viewer_index
.status
= htobe32(VIEWER_INDEX_HUP
);
1018 /* First time, we open the index file */
1019 if (vstream
->index_read_fd
< 0) {
1020 ret
= open_index(vstream
);
1021 if (ret
== -ENOENT
) {
1023 * The index is created only when the first data packet arrives, it
1024 * might not be ready at the beginning of the session
1026 viewer_index
.status
= htobe32(VIEWER_INDEX_RETRY
);
1028 } else if (ret
< 0) {
1029 viewer_index
.status
= htobe32(VIEWER_INDEX_ERR
);
1034 rstream
= relay_stream_find_by_id(vstream
->stream_handle
);
1036 if (rstream
->beacon_ts_end
!= -1ULL &&
1037 vstream
->last_sent_index
== rstream
->total_index_received
) {
1038 viewer_index
.status
= htobe32(VIEWER_INDEX_INACTIVE
);
1039 viewer_index
.timestamp_end
= htobe64(rstream
->beacon_ts_end
);
1043 if (rstream
->total_index_received
<= vstream
->last_sent_index
) {
1044 /* No new index to send, retry later. */
1045 viewer_index
.status
= htobe32(VIEWER_INDEX_RETRY
);
1048 } else if (!rstream
&&
1049 vstream
->total_index_received
== vstream
->last_sent_index
) {
1050 /* Last index sent and stream closed */
1051 viewer_index
.status
= htobe32(VIEWER_INDEX_HUP
);
1055 if (!vstream
->ctf_trace
->metadata_received
||
1056 vstream
->ctf_trace
->metadata_received
>
1057 vstream
->ctf_trace
->metadata_sent
) {
1058 viewer_index
.flags
|= LTTNG_VIEWER_FLAG_NEW_METADATA
;
1061 ret
= lttng_read(vstream
->index_read_fd
, &packet_index
,
1062 sizeof(packet_index
));
1063 if (ret
< sizeof(packet_index
)) {
1064 PERROR("Relay reading index file");
1065 viewer_index
.status
= htobe32(VIEWER_INDEX_ERR
);
1067 viewer_index
.status
= htobe32(VIEWER_INDEX_OK
);
1068 vstream
->last_sent_index
++;
1072 * Indexes are stored in big endian, no need to switch before sending.
1074 viewer_index
.offset
= packet_index
.offset
;
1075 viewer_index
.packet_size
= packet_index
.packet_size
;
1076 viewer_index
.content_size
= packet_index
.content_size
;
1077 viewer_index
.timestamp_begin
= packet_index
.timestamp_begin
;
1078 viewer_index
.timestamp_end
= packet_index
.timestamp_end
;
1079 viewer_index
.events_discarded
= packet_index
.events_discarded
;
1080 viewer_index
.stream_id
= packet_index
.stream_id
;
1083 viewer_index
.flags
= htobe32(viewer_index
.flags
);
1084 health_code_update();
1085 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &viewer_index
,
1086 sizeof(viewer_index
), 0);
1088 ERR("Relay index to viewer");
1091 health_code_update();
1093 DBG("Index %" PRIu64
"for stream %" PRIu64
"sent",
1094 vstream
->last_sent_index
, vstream
->stream_handle
);
1105 * Send the next index for a stream
1107 * Return 0 on success or else a negative value.
1110 int viewer_get_packet(struct relay_command
*cmd
)
1112 int ret
, send_data
= 0;
1116 struct lttng_viewer_get_packet get_packet_info
;
1117 struct lttng_viewer_trace_packet reply
;
1118 struct relay_viewer_stream
*stream
;
1122 DBG2("Relay get data packet");
1124 if (cmd
->version_check_done
== 0) {
1125 ERR("Trying to get packet before version check");
1130 health_code_update();
1131 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &get_packet_info
,
1132 sizeof(get_packet_info
), 0);
1133 if (ret
< 0 || ret
!= sizeof(get_packet_info
)) {
1135 ERR("Relay didn't receive the whole packet");
1138 health_code_update();
1140 /* From this point on, the error label can be reached. */
1141 memset(&reply
, 0, sizeof(reply
));
1144 stream
= live_find_viewer_stream_by_id(be64toh(get_packet_info
.stream_id
));
1148 assert(stream
->ctf_trace
);
1151 * First time we read this stream, we need open the tracefile, we should
1152 * only arrive here if an index has already been sent to the viewer, so the
1153 * tracefile must exist, if it does not it is a fatal error.
1155 if (stream
->read_fd
< 0) {
1156 char fullpath
[PATH_MAX
];
1158 ret
= snprintf(fullpath
, PATH_MAX
, "%s/%s", stream
->path_name
,
1159 stream
->channel_name
);
1163 ret
= open(fullpath
, O_RDONLY
);
1165 PERROR("Relay opening trace file");
1168 stream
->read_fd
= ret
;
1171 if (!stream
->ctf_trace
->metadata_received
||
1172 stream
->ctf_trace
->metadata_received
>
1173 stream
->ctf_trace
->metadata_sent
) {
1174 reply
.status
= htobe32(VIEWER_GET_PACKET_ERR
);
1175 reply
.flags
|= LTTNG_VIEWER_FLAG_NEW_METADATA
;
1179 len
= be32toh(get_packet_info
.len
);
1180 data
= zmalloc(len
);
1182 PERROR("relay data zmalloc");
1186 ret
= lseek(stream
->read_fd
, be64toh(get_packet_info
.offset
), SEEK_SET
);
1191 read_len
= lttng_read(stream
->read_fd
, data
, len
);
1192 if (read_len
< len
) {
1193 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64
,
1194 stream
->read_fd
, be64toh(get_packet_info
.offset
));
1197 reply
.status
= htobe32(VIEWER_GET_PACKET_OK
);
1198 reply
.len
= htobe32(len
);
1203 reply
.status
= htobe32(VIEWER_GET_PACKET_ERR
);
1206 reply
.flags
= htobe32(reply
.flags
);
1208 health_code_update();
1209 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1211 ERR("Relay data header to viewer");
1214 health_code_update();
1217 health_code_update();
1218 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, data
, len
, 0);
1220 ERR("Relay send data to viewer");
1223 health_code_update();
1226 DBG("Sent %u bytes for stream %" PRIu64
, len
,
1227 be64toh(get_packet_info
.stream_id
));
1238 * Send the session's metadata
1240 * Return 0 on success else a negative value.
1243 int viewer_get_metadata(struct relay_command
*cmd
)
1249 struct lttng_viewer_get_metadata request
;
1250 struct lttng_viewer_metadata_packet reply
;
1251 struct relay_viewer_stream
*stream
;
1255 DBG("Relay get metadata");
1257 if (cmd
->version_check_done
== 0) {
1258 ERR("Trying to get metadata before version check");
1263 health_code_update();
1264 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &request
,
1265 sizeof(request
), 0);
1266 if (ret
< 0 || ret
!= sizeof(request
)) {
1268 ERR("Relay didn't receive the whole packet");
1271 health_code_update();
1274 stream
= live_find_viewer_stream_by_id(be64toh(request
.stream_id
));
1275 if (!stream
|| !stream
->metadata_flag
) {
1276 ERR("Invalid metadata stream");
1279 assert(stream
->ctf_trace
);
1280 assert(stream
->ctf_trace
->metadata_sent
<=
1281 stream
->ctf_trace
->metadata_received
);
1283 len
= stream
->ctf_trace
->metadata_received
-
1284 stream
->ctf_trace
->metadata_sent
;
1286 reply
.status
= htobe32(VIEWER_NO_NEW_METADATA
);
1290 /* first time, we open the metadata file */
1291 if (stream
->read_fd
< 0) {
1292 char fullpath
[PATH_MAX
];
1294 ret
= snprintf(fullpath
, PATH_MAX
, "%s/%s", stream
->path_name
,
1295 stream
->channel_name
);
1299 ret
= open(fullpath
, O_RDONLY
);
1301 PERROR("Relay opening metadata file");
1304 stream
->read_fd
= ret
;
1307 reply
.len
= htobe64(len
);
1308 data
= zmalloc(len
);
1310 PERROR("viewer metadata zmalloc");
1314 read_len
= lttng_read(stream
->read_fd
, data
, len
);
1315 if (read_len
< len
) {
1316 PERROR("Relay reading metadata file");
1319 stream
->ctf_trace
->metadata_sent
+= read_len
;
1320 reply
.status
= htobe32(VIEWER_METADATA_OK
);
1324 reply
.status
= htobe32(VIEWER_METADATA_ERR
);
1327 health_code_update();
1328 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1330 ERR("Relay data header to viewer");
1333 health_code_update();
1336 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, data
, len
, 0);
1338 ERR("Relay send data to viewer");
1343 DBG("Sent %" PRIu64
" bytes of metadata for stream %" PRIu64
, len
,
1344 be64toh(request
.stream_id
));
1346 DBG("Metadata sent");
1356 * live_relay_unknown_command: send -1 if received unknown command
1359 void live_relay_unknown_command(struct relay_command
*cmd
)
1361 struct lttcomm_relayd_generic_reply reply
;
1364 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1365 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1366 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1368 ERR("Relay sending unknown command");
1373 * Process the commands received on the control socket
1376 int process_control(struct lttng_viewer_cmd
*recv_hdr
,
1377 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1381 switch (be32toh(recv_hdr
->cmd
)) {
1382 case VIEWER_CONNECT
:
1383 ret
= viewer_connect(cmd
);
1385 case VIEWER_LIST_SESSIONS
:
1386 ret
= viewer_list_sessions(cmd
, sessions_ht
);
1388 case VIEWER_ATTACH_SESSION
:
1389 ret
= viewer_attach_session(cmd
, sessions_ht
);
1391 case VIEWER_GET_NEXT_INDEX
:
1392 ret
= viewer_get_next_index(cmd
, sessions_ht
);
1394 case VIEWER_GET_PACKET
:
1395 ret
= viewer_get_packet(cmd
);
1397 case VIEWER_GET_METADATA
:
1398 ret
= viewer_get_metadata(cmd
);
1401 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr
->cmd
));
1402 live_relay_unknown_command(cmd
);
1412 void cleanup_poll_connection(struct lttng_poll_event
*events
, int pollfd
)
1418 lttng_poll_del(events
, pollfd
);
1420 ret
= close(pollfd
);
1422 ERR("Closing pollfd %d", pollfd
);
1427 * Create and add connection to the given hash table.
1429 * Return poll add value or else -1 on error.
1432 int add_connection(int fd
, struct lttng_poll_event
*events
,
1433 struct lttng_ht
*relay_connections_ht
)
1436 struct relay_command
*relay_connection
;
1439 assert(relay_connections_ht
);
1441 relay_connection
= zmalloc(sizeof(struct relay_command
));
1442 if (relay_connection
== NULL
) {
1443 PERROR("Relay command zmalloc");
1447 ret
= lttng_read(fd
, relay_connection
, sizeof(*relay_connection
));
1448 if (ret
< sizeof(*relay_connection
)) {
1449 PERROR("read relay cmd pipe");
1453 lttng_ht_node_init_ulong(&relay_connection
->sock_n
,
1454 (unsigned long) relay_connection
->sock
->fd
);
1456 lttng_ht_add_unique_ulong(relay_connections_ht
,
1457 &relay_connection
->sock_n
);
1460 return lttng_poll_add(events
, relay_connection
->sock
->fd
,
1461 LPOLLIN
| LPOLLRDHUP
);
1464 free(relay_connection
);
1470 void deferred_free_connection(struct rcu_head
*head
)
1472 struct relay_command
*relay_connection
=
1473 caa_container_of(head
, struct relay_command
, rcu_node
);
1475 if (relay_connection
->session
&&
1476 relay_connection
->session
->viewer_attached
> 0) {
1477 relay_connection
->session
->viewer_attached
--;
1479 lttcomm_destroy_sock(relay_connection
->sock
);
1480 free(relay_connection
);
1484 void deferred_free_viewer_stream(struct rcu_head
*head
)
1486 struct relay_viewer_stream
*stream
=
1487 caa_container_of(head
, struct relay_viewer_stream
, rcu_node
);
1489 if (stream
->ctf_trace
) {
1490 uatomic_dec(&stream
->ctf_trace
->refcount
);
1491 assert(uatomic_read(&stream
->ctf_trace
->refcount
) >= 0);
1492 if (uatomic_read(&stream
->ctf_trace
->refcount
) == 0) {
1493 DBG("Freeing ctf_trace %" PRIu64
, stream
->ctf_trace
->id
);
1494 free(stream
->ctf_trace
);
1498 free(stream
->path_name
);
1499 free(stream
->channel_name
);
1504 void viewer_del_streams(uint64_t session_id
)
1507 struct relay_viewer_stream
*stream
;
1508 struct lttng_ht_node_u64
*node
;
1509 struct lttng_ht_iter iter
;
1512 cds_lfht_for_each_entry(viewer_streams_ht
->ht
, &iter
.iter
, node
, node
) {
1513 health_code_update();
1515 node
= lttng_ht_iter_get_node_u64(&iter
);
1520 stream
= caa_container_of(node
, struct relay_viewer_stream
, stream_n
);
1521 if (stream
->session_id
!= session_id
) {
1525 if (stream
->read_fd
> 0) {
1526 ret
= close(stream
->read_fd
);
1528 PERROR("close read_fd");
1531 if (stream
->index_read_fd
> 0) {
1532 ret
= close(stream
->index_read_fd
);
1534 PERROR("close index_read_fd");
1537 if (stream
->metadata_flag
&& stream
->ctf_trace
) {
1538 stream
->ctf_trace
->metadata_sent
= 0;
1540 ret
= lttng_ht_del(viewer_streams_ht
, &iter
);
1542 call_rcu(&stream
->rcu_node
, deferred_free_viewer_stream
);
1548 * Delete and free a connection.
1550 * RCU read side lock MUST be acquired.
1553 void del_connection(struct lttng_ht
*relay_connections_ht
,
1554 struct lttng_ht_iter
*iter
, struct relay_command
*relay_connection
)
1558 assert(relay_connections_ht
);
1560 assert(relay_connection
);
1562 ret
= lttng_ht_del(relay_connections_ht
, iter
);
1565 viewer_del_streams(relay_connection
->session_id
);
1567 call_rcu(&relay_connection
->rcu_node
, deferred_free_connection
);
1571 * This thread does the actual work
1574 void *thread_worker(void *data
)
1578 struct relay_command
*relay_connection
;
1579 struct lttng_poll_event events
;
1580 struct lttng_ht
*relay_connections_ht
;
1581 struct lttng_ht_node_ulong
*node
;
1582 struct lttng_ht_iter iter
;
1583 struct lttng_viewer_cmd recv_hdr
;
1584 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
1585 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
1587 DBG("[thread] Live viewer relay worker started");
1589 rcu_register_thread();
1591 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_WORKER
);
1593 /* table of connections indexed on socket */
1594 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
1595 if (!relay_connections_ht
) {
1596 goto relay_connections_ht_error
;
1599 ret
= create_thread_poll_set(&events
, 2);
1601 goto error_poll_create
;
1604 ret
= lttng_poll_add(&events
, live_relay_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1613 health_code_update();
1615 /* Infinite blocking call, waiting for transmission */
1616 DBG3("Relayd live viewer worker thread polling...");
1617 health_poll_entry();
1618 ret
= lttng_poll_wait(&events
, -1);
1622 * Restart interrupted system call.
1624 if (errno
== EINTR
) {
1633 * Process control. The control connection is prioritised so we don't
1634 * starve it with high throughput tracing data on the data
1637 for (i
= 0; i
< nb_fd
; i
++) {
1638 /* Fetch once the poll data */
1639 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1640 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1642 health_code_update();
1644 /* Thread quit pipe has been closed. Killing thread. */
1645 ret
= check_thread_quit_pipe(pollfd
, revents
);
1651 /* Inspect the relay cmd pipe for new connection */
1652 if (pollfd
== live_relay_cmd_pipe
[0]) {
1653 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1654 ERR("Relay live pipe error");
1656 } else if (revents
& LPOLLIN
) {
1657 DBG("Relay live viewer command received");
1658 ret
= add_connection(live_relay_cmd_pipe
[0],
1659 &events
, relay_connections_ht
);
1664 } else if (revents
) {
1666 lttng_ht_lookup(relay_connections_ht
,
1667 (void *)((unsigned long) pollfd
), &iter
);
1668 node
= lttng_ht_iter_get_node_ulong(&iter
);
1670 DBG2("Relay viewer sock %d not found", pollfd
);
1674 relay_connection
= caa_container_of(node
, struct relay_command
,
1677 if (revents
& (LPOLLERR
)) {
1678 cleanup_poll_connection(&events
, pollfd
);
1679 del_connection(relay_connections_ht
, &iter
,
1681 } else if (revents
& (LPOLLHUP
| LPOLLRDHUP
)) {
1682 DBG("Viewer socket %d hung up", pollfd
);
1683 cleanup_poll_connection(&events
, pollfd
);
1684 del_connection(relay_connections_ht
, &iter
,
1686 } else if (revents
& LPOLLIN
) {
1687 ret
= relay_connection
->sock
->ops
->recvmsg(
1688 relay_connection
->sock
, &recv_hdr
,
1689 sizeof(struct lttng_viewer_cmd
),
1691 /* connection closed */
1693 cleanup_poll_connection(&events
, pollfd
);
1694 del_connection(relay_connections_ht
, &iter
,
1696 DBG("Viewer control connection closed with %d",
1699 if (relay_connection
->session
) {
1700 DBG2("Relay viewer worker receiving data for "
1701 "session: %" PRIu64
,
1702 relay_connection
->session
->id
);
1704 ret
= process_control(&recv_hdr
, relay_connection
,
1707 /* Clear the session on error. */
1708 cleanup_poll_connection(&events
, pollfd
);
1709 del_connection(relay_connections_ht
, &iter
,
1711 DBG("Viewer connection closed with %d", pollfd
);
1722 lttng_poll_clean(&events
);
1724 /* empty the hash table and free the memory */
1726 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, node
, node
) {
1727 health_code_update();
1729 node
= lttng_ht_iter_get_node_ulong(&iter
);
1734 relay_connection
= caa_container_of(node
, struct relay_command
,
1736 del_connection(relay_connections_ht
, &iter
, relay_connection
);
1740 lttng_ht_destroy(relay_connections_ht
);
1741 relay_connections_ht_error
:
1742 /* Close relay cmd pipes */
1743 utils_close_pipe(live_relay_cmd_pipe
);
1745 DBG("Viewer worker thread exited with error");
1747 DBG("Viewer worker thread cleanup complete");
1750 ERR("Health error occurred in %s", __func__
);
1752 health_unregister(health_relayd
);
1754 rcu_unregister_thread();
1759 * Create the relay command pipe to wake thread_manage_apps.
1760 * Closed in cleanup().
1762 static int create_relay_cmd_pipe(void)
1766 ret
= utils_create_pipe_cloexec(live_relay_cmd_pipe
);
1771 void live_stop_threads(void)
1778 ret
= pthread_join(live_listener_thread
, &status
);
1780 PERROR("pthread_join live listener");
1781 goto error
; /* join error, exit without cleanup */
1784 ret
= pthread_join(live_worker_thread
, &status
);
1786 PERROR("pthread_join live worker");
1787 goto error
; /* join error, exit without cleanup */
1790 ret
= pthread_join(live_dispatcher_thread
, &status
);
1792 PERROR("pthread_join live dispatcher");
1793 goto error
; /* join error, exit without cleanup */
1805 int live_start_threads(struct lttng_uri
*uri
,
1806 struct relay_local_data
*relay_ctx
, int quit_pipe
[2])
1815 live_thread_quit_pipe
[0] = quit_pipe
[0];
1816 live_thread_quit_pipe
[1] = quit_pipe
[1];
1818 /* Check if daemon is UID = 0 */
1819 is_root
= !getuid();
1822 if (live_uri
->port
< 1024) {
1823 ERR("Need to be root to use ports < 1024");
1829 /* Setup the thread apps communication pipe. */
1830 if ((ret
= create_relay_cmd_pipe()) < 0) {
1834 /* Init relay command queue. */
1835 cds_wfq_init(&viewer_cmd_queue
.queue
);
1837 /* Set up max poll set size */
1838 lttng_poll_set_max_size();
1840 /* Setup the dispatcher thread */
1841 ret
= pthread_create(&live_dispatcher_thread
, NULL
,
1842 thread_dispatcher
, (void *) NULL
);
1844 PERROR("pthread_create viewer dispatcher");
1845 goto exit_dispatcher
;
1848 /* Setup the worker thread */
1849 ret
= pthread_create(&live_worker_thread
, NULL
,
1850 thread_worker
, relay_ctx
);
1852 PERROR("pthread_create viewer worker");
1856 /* Setup the listener thread */
1857 ret
= pthread_create(&live_listener_thread
, NULL
,
1858 thread_listener
, (void *) NULL
);
1860 PERROR("pthread_create viewer listener");
1868 ret
= pthread_join(live_listener_thread
, &status
);
1870 PERROR("pthread_join live listener");
1871 goto error
; /* join error, exit without cleanup */
1875 ret
= pthread_join(live_worker_thread
, &status
);
1877 PERROR("pthread_join live worker");
1878 goto error
; /* join error, exit without cleanup */
1882 ret
= pthread_join(live_dispatcher_thread
, &status
);
1884 PERROR("pthread_join live dispatcher");
1885 goto error
; /* join error, exit without cleanup */