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/index/index.h>
49 #include <common/sessiond-comm/sessiond-comm.h>
50 #include <common/sessiond-comm/inet.h>
51 #include <common/sessiond-comm/relayd.h>
52 #include <common/uri.h>
53 #include <common/utils.h>
57 #include "lttng-relayd.h"
59 #include "health-relayd.h"
60 #include "testpoint.h"
61 #include "viewer-stream.h"
64 #include "ctf-trace.h"
65 #include "connection.h"
67 static struct lttng_uri
*live_uri
;
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_conn_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_conn_queue viewer_conn_queue
;
90 static uint64_t last_relay_viewer_session_id
;
104 * Receive a request buffer using a given socket, destination allocated buffer
107 * Return the size of the received message or else a negative value on error
108 * with errno being set by recvmsg() syscall.
111 ssize_t
recv_request(struct lttcomm_sock
*sock
, void *buf
, size_t size
)
118 ret
= sock
->ops
->recvmsg(sock
, buf
, size
, 0);
119 if (ret
< 0 || ret
!= size
) {
121 /* Orderly shutdown. Not necessary to print an error. */
122 DBG("Socket %d did an orderly shutdown", sock
->fd
);
124 ERR("Relay failed to receive request.");
133 * Send a response buffer using a given socket, source allocated buffer of
136 * Return the size of the sent message or else a negative value on error with
137 * errno being set by sendmsg() syscall.
140 ssize_t
send_response(struct lttcomm_sock
*sock
, void *buf
, size_t size
)
147 ret
= sock
->ops
->sendmsg(sock
, buf
, size
, 0);
149 ERR("Relayd failed to send response.");
156 * Atomically check if new streams got added in the session since the last
157 * check and reset the flag to 0.
159 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
163 int check_new_streams(uint64_t session_id
, struct lttng_ht
*sessions_ht
)
166 unsigned long current_val
;
167 struct relay_session
*session
;
171 session
= session_find_by_id(sessions_ht
, session_id
);
173 DBG("Relay session %" PRIu64
" not found", session_id
);
178 current_val
= uatomic_cmpxchg(&session
->new_streams
, 1, 0);
186 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
187 * this function should ignore the sent flag or not.
189 * Return 0 on success or else a negative value.
192 ssize_t
send_viewer_streams(struct lttcomm_sock
*sock
,
193 struct relay_session
*session
, unsigned int ignore_sent_flag
)
196 struct lttng_viewer_stream send_stream
;
197 struct lttng_ht_iter iter
;
198 struct relay_viewer_stream
*vstream
;
204 cds_lfht_for_each_entry(viewer_streams_ht
->ht
, &iter
.iter
, vstream
,
206 struct ctf_trace
*ctf_trace
;
208 health_code_update();
210 /* Ignore if not the same session. */
211 if (vstream
->session_id
!= session
->id
||
212 (!ignore_sent_flag
&& vstream
->sent_flag
)) {
216 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
,
220 send_stream
.id
= htobe64(vstream
->stream_handle
);
221 send_stream
.ctf_trace_id
= htobe64(ctf_trace
->id
);
222 send_stream
.metadata_flag
= htobe32(vstream
->metadata_flag
);
223 strncpy(send_stream
.path_name
, vstream
->path_name
,
224 sizeof(send_stream
.path_name
));
225 strncpy(send_stream
.channel_name
, vstream
->channel_name
,
226 sizeof(send_stream
.channel_name
));
228 DBG("Sending stream %" PRIu64
" to viewer", vstream
->stream_handle
);
229 ret
= send_response(sock
, &send_stream
, sizeof(send_stream
));
233 vstream
->sent_flag
= 1;
244 * Create every viewer stream possible for the given session with the seek
245 * type. Three counters *can* be return which are in order the total amount of
246 * viewer stream of the session, the number of unsent stream and the number of
247 * stream created. Those counters can be NULL and thus will be ignored.
249 * Return 0 on success or else a negative value.
252 int make_viewer_streams(struct relay_session
*session
,
253 enum lttng_viewer_seek seek_t
, uint32_t *nb_total
, uint32_t *nb_unsent
,
254 uint32_t *nb_created
)
257 struct lttng_ht_iter iter
;
258 struct ctf_trace
*ctf_trace
;
263 * This is to make sure we create viewer streams for a full received
264 * channel. For instance, if we have 8 streams for a channel that are
265 * concurrently being flagged ready, we can end up creating just a subset
266 * of the 8 streams (the ones that are flagged). This lock avoids this
269 pthread_mutex_lock(&session
->viewer_ready_lock
);
272 * Create viewer streams for relay streams that are ready to be used for a
273 * the given session id only.
276 cds_lfht_for_each_entry(session
->ctf_traces_ht
->ht
, &iter
.iter
, ctf_trace
,
278 struct relay_stream
*stream
;
280 health_code_update();
282 if (ctf_trace
->invalid_flag
) {
286 cds_list_for_each_entry(stream
, &ctf_trace
->stream_list
, trace_list
) {
287 struct relay_viewer_stream
*vstream
;
289 if (!stream
->viewer_ready
) {
293 vstream
= viewer_stream_find_by_id(stream
->stream_handle
);
295 vstream
= viewer_stream_create(stream
, seek_t
, ctf_trace
);
300 /* Acquire reference to ctf_trace. */
301 ctf_trace_get_ref(ctf_trace
);
304 /* Update number of created stream counter. */
307 } else if (!vstream
->sent_flag
&& nb_unsent
) {
308 /* Update number of unsent stream counter. */
311 /* Update number of total stream counter. */
322 pthread_mutex_unlock(&session
->viewer_ready_lock
);
327 * Write to writable pipe used to notify a thread.
330 int notify_thread_pipe(int wpipe
)
334 ret
= lttng_write(wpipe
, "!", 1);
336 PERROR("write poll pipe");
343 * Stop all threads by closing the thread quit pipe.
346 void stop_threads(void)
350 /* Stopping all threads */
351 DBG("Terminating all live threads");
352 ret
= notify_thread_pipe(live_conn_pipe
[1]);
354 ERR("write error on thread quit pipe");
357 /* Dispatch thread */
358 CMM_STORE_SHARED(live_dispatch_thread_exit
, 1);
359 futex_nto1_wake(&viewer_conn_queue
.futex
);
363 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
366 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
370 if (events
== NULL
|| size
== 0) {
375 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
381 ret
= lttng_poll_add(events
, live_conn_pipe
[0], LPOLLIN
| LPOLLERR
);
393 * Check if the thread quit pipe was triggered.
395 * Return 1 if it was triggered else 0;
398 int check_live_conn_pipe(int fd
, uint32_t events
)
400 if (fd
== live_conn_pipe
[0] && (events
& LPOLLIN
)) {
408 * Create and init socket from uri.
411 struct lttcomm_sock
*init_socket(struct lttng_uri
*uri
)
414 struct lttcomm_sock
*sock
= NULL
;
416 sock
= lttcomm_alloc_sock_from_uri(uri
);
418 ERR("Allocating socket");
422 ret
= lttcomm_create_sock(sock
);
426 DBG("Listening on sock %d for live", sock
->fd
);
428 ret
= sock
->ops
->bind(sock
);
433 ret
= sock
->ops
->listen(sock
, -1);
443 lttcomm_destroy_sock(sock
);
449 * This thread manages the listening for new connections on the network
452 void *thread_listener(void *data
)
454 int i
, ret
, pollfd
, err
= -1;
455 uint32_t revents
, nb_fd
;
456 struct lttng_poll_event events
;
457 struct lttcomm_sock
*live_control_sock
;
459 DBG("[thread] Relay live listener started");
461 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_LISTENER
);
463 health_code_update();
465 live_control_sock
= init_socket(live_uri
);
466 if (!live_control_sock
) {
467 goto error_sock_control
;
470 /* Pass 2 as size here for the thread quit pipe and control sockets. */
471 ret
= create_thread_poll_set(&events
, 2);
473 goto error_create_poll
;
476 /* Add the control socket */
477 ret
= lttng_poll_add(&events
, live_control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
482 lttng_relay_notify_ready();
484 if (testpoint(relayd_thread_live_listener
)) {
485 goto error_testpoint
;
489 health_code_update();
491 DBG("Listener accepting live viewers connections");
495 ret
= lttng_poll_wait(&events
, -1);
499 * Restart interrupted system call.
501 if (errno
== EINTR
) {
508 DBG("Relay new viewer connection received");
509 for (i
= 0; i
< nb_fd
; i
++) {
510 health_code_update();
512 /* Fetch once the poll data */
513 revents
= LTTNG_POLL_GETEV(&events
, i
);
514 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
516 /* Thread quit pipe has been closed. Killing thread. */
517 ret
= check_live_conn_pipe(pollfd
, revents
);
523 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
524 ERR("socket poll error");
526 } else if (revents
& LPOLLIN
) {
528 * Get allocated in this thread, enqueued to a global queue,
529 * dequeued and freed in the worker thread.
532 struct relay_connection
*new_conn
;
533 struct lttcomm_sock
*newsock
;
535 new_conn
= connection_create();
540 newsock
= live_control_sock
->ops
->accept(live_control_sock
);
542 PERROR("accepting control sock");
543 connection_free(new_conn
);
546 DBG("Relay viewer connection accepted socket %d", newsock
->fd
);
548 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
551 PERROR("setsockopt inet");
552 lttcomm_destroy_sock(newsock
);
553 connection_free(new_conn
);
556 new_conn
->sock
= newsock
;
558 /* Enqueue request for the dispatcher thread. */
559 cds_wfq_enqueue(&viewer_conn_queue
.queue
, &new_conn
->qnode
);
562 * Wake the dispatch queue futex. Implicit memory barrier with
563 * the exchange in cds_wfq_enqueue.
565 futex_nto1_wake(&viewer_conn_queue
.futex
);
574 lttng_poll_clean(&events
);
576 if (live_control_sock
->fd
>= 0) {
577 ret
= live_control_sock
->ops
->close(live_control_sock
);
582 lttcomm_destroy_sock(live_control_sock
);
586 DBG("Live viewer listener thread exited with error");
588 health_unregister(health_relayd
);
589 DBG("Live viewer listener thread cleanup complete");
595 * This thread manages the dispatching of the requests to worker threads
598 void *thread_dispatcher(void *data
)
602 struct cds_wfq_node
*node
;
603 struct relay_connection
*conn
= NULL
;
605 DBG("[thread] Live viewer relay dispatcher started");
607 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER
);
609 if (testpoint(relayd_thread_live_dispatcher
)) {
610 goto error_testpoint
;
613 health_code_update();
615 while (!CMM_LOAD_SHARED(live_dispatch_thread_exit
)) {
616 health_code_update();
618 /* Atomically prepare the queue futex */
619 futex_nto1_prepare(&viewer_conn_queue
.futex
);
622 health_code_update();
624 /* Dequeue commands */
625 node
= cds_wfq_dequeue_blocking(&viewer_conn_queue
.queue
);
627 DBG("Woken up but nothing in the live-viewer "
628 "relay command queue");
629 /* Continue thread execution */
632 conn
= caa_container_of(node
, struct relay_connection
, qnode
);
633 DBG("Dispatching viewer request waiting on sock %d",
637 * Inform worker thread of the new request. This call is blocking
638 * so we can be assured that the data will be read at some point in
639 * time or wait to the end of the world :)
641 ret
= lttng_write(live_conn_pipe
[1], &conn
, sizeof(conn
));
643 PERROR("write conn pipe");
644 connection_destroy(conn
);
647 } while (node
!= NULL
);
649 /* Futex wait on queue. Blocking call on futex() */
651 futex_nto1_wait(&viewer_conn_queue
.futex
);
655 /* Normal exit, no error */
662 ERR("Health error occurred in %s", __func__
);
664 health_unregister(health_relayd
);
665 DBG("Live viewer dispatch thread dying");
671 * Establish connection with the viewer and check the versions.
673 * Return 0 on success or else negative value.
676 int viewer_connect(struct relay_connection
*conn
)
679 struct lttng_viewer_connect reply
, msg
;
683 conn
->version_check_done
= 1;
685 health_code_update();
687 DBG("Viewer is establishing a connection to the relayd.");
689 ret
= recv_request(conn
->sock
, &msg
, sizeof(msg
));
694 health_code_update();
696 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
697 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
699 /* Major versions must be the same */
700 if (reply
.major
!= be32toh(msg
.major
)) {
701 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
702 reply
.major
, be32toh(msg
.major
));
707 conn
->major
= reply
.major
;
708 /* We adapt to the lowest compatible version */
709 if (reply
.minor
<= be32toh(msg
.minor
)) {
710 conn
->minor
= reply
.minor
;
712 conn
->minor
= be32toh(msg
.minor
);
715 if (be32toh(msg
.type
) == LTTNG_VIEWER_CLIENT_COMMAND
) {
716 conn
->type
= RELAY_VIEWER_COMMAND
;
717 } else if (be32toh(msg
.type
) == LTTNG_VIEWER_CLIENT_NOTIFICATION
) {
718 conn
->type
= RELAY_VIEWER_NOTIFICATION
;
720 ERR("Unknown connection type : %u", be32toh(msg
.type
));
725 reply
.major
= htobe32(reply
.major
);
726 reply
.minor
= htobe32(reply
.minor
);
727 if (conn
->type
== RELAY_VIEWER_COMMAND
) {
728 reply
.viewer_session_id
= htobe64(++last_relay_viewer_session_id
);
731 health_code_update();
733 ret
= send_response(conn
->sock
, &reply
, sizeof(reply
));
738 health_code_update();
740 DBG("Version check done using protocol %u.%u", conn
->major
, conn
->minor
);
748 * Send the viewer the list of current sessions.
750 * Return 0 on success or else a negative value.
753 int viewer_list_sessions(struct relay_connection
*conn
)
756 struct lttng_viewer_list_sessions session_list
;
758 long approx_before
, approx_after
;
759 struct lttng_ht_iter iter
;
760 struct lttng_viewer_session send_session
;
761 struct relay_session
*session
;
763 DBG("List sessions received");
766 cds_lfht_count_nodes(conn
->sessions_ht
->ht
, &approx_before
, &count
,
768 session_list
.sessions_count
= htobe32(count
);
770 health_code_update();
772 ret
= send_response(conn
->sock
, &session_list
, sizeof(session_list
));
777 health_code_update();
779 cds_lfht_for_each_entry(conn
->sessions_ht
->ht
, &iter
.iter
, session
,
781 health_code_update();
783 strncpy(send_session
.session_name
, session
->session_name
,
784 sizeof(send_session
.session_name
));
785 strncpy(send_session
.hostname
, session
->hostname
,
786 sizeof(send_session
.hostname
));
787 send_session
.id
= htobe64(session
->id
);
788 send_session
.live_timer
= htobe32(session
->live_timer
);
789 send_session
.clients
= htobe32(session
->viewer_refcount
);
790 send_session
.streams
= htobe32(session
->stream_count
);
792 health_code_update();
794 ret
= send_response(conn
->sock
, &send_session
, sizeof(send_session
));
799 health_code_update();
813 * Send the viewer the list of current sessions.
816 int viewer_get_new_streams(struct relay_connection
*conn
)
818 int ret
, send_streams
= 0;
819 uint32_t nb_created
= 0, nb_unsent
= 0, nb_streams
= 0;
820 struct lttng_viewer_new_streams_request request
;
821 struct lttng_viewer_new_streams_response response
;
822 struct relay_session
*session
;
826 DBG("Get new streams received");
828 health_code_update();
830 /* Receive the request from the connected client. */
831 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
836 health_code_update();
839 session
= session_find_by_id(conn
->sessions_ht
,
840 be64toh(request
.session_id
));
842 DBG("Relay session %" PRIu64
" not found",
843 be64toh(request
.session_id
));
844 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR
);
848 if (conn
->session_id
== session
->id
) {
849 /* We confirmed the viewer is asking for the same session. */
851 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_OK
);
854 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR
);
862 ret
= make_viewer_streams(session
, LTTNG_VIEWER_SEEK_LAST
, NULL
, &nb_unsent
,
867 /* Only send back the newly created streams with the unsent ones. */
868 nb_streams
= nb_created
+ nb_unsent
;
869 response
.streams_count
= htobe32(nb_streams
);
872 health_code_update();
873 ret
= send_response(conn
->sock
, &response
, sizeof(response
));
877 health_code_update();
880 * Unknown or empty session, just return gracefully, the viewer knows what
883 if (!send_streams
|| !nb_streams
) {
889 * Send stream and *DON'T* ignore the sent flag so every viewer streams
890 * that were not sent from that point will be sent to the viewer.
892 ret
= send_viewer_streams(conn
->sock
, session
, 0);
904 * Send the viewer the list of current sessions.
907 int viewer_attach_session(struct relay_connection
*conn
)
909 int send_streams
= 0;
911 uint32_t nb_streams
= 0;
912 enum lttng_viewer_seek seek_type
;
913 struct lttng_viewer_attach_session_request request
;
914 struct lttng_viewer_attach_session_response response
;
915 struct relay_session
*session
;
919 health_code_update();
921 /* Receive the request from the connected client. */
922 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
927 health_code_update();
929 if (!conn
->viewer_session
) {
930 DBG("Client trying to attach before creating a live viewer session");
931 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION
);
936 session
= session_find_by_id(conn
->sessions_ht
,
937 be64toh(request
.session_id
));
939 DBG("Relay session %" PRIu64
" not found",
940 be64toh(request
.session_id
));
941 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_UNK
);
944 session_viewer_attach(session
);
945 DBG("Attach session ID %" PRIu64
" received", be64toh(request
.session_id
));
947 if (uatomic_read(&session
->viewer_refcount
) > 1) {
948 DBG("Already a viewer attached");
949 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_ALREADY
);
950 session_viewer_detach(session
);
952 } else if (session
->live_timer
== 0) {
953 DBG("Not live session");
954 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE
);
958 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_OK
);
959 conn
->session_id
= session
->id
;
960 conn
->session
= session
;
963 switch (be32toh(request
.seek
)) {
964 case LTTNG_VIEWER_SEEK_BEGINNING
:
965 case LTTNG_VIEWER_SEEK_LAST
:
966 seek_type
= be32toh(request
.seek
);
969 ERR("Wrong seek parameter");
970 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR
);
979 ret
= make_viewer_streams(session
, seek_type
, &nb_streams
, NULL
, NULL
);
983 response
.streams_count
= htobe32(nb_streams
);
986 health_code_update();
987 ret
= send_response(conn
->sock
, &response
, sizeof(response
));
991 health_code_update();
994 * Unknown or empty session, just return gracefully, the viewer knows what
997 if (!send_streams
|| !nb_streams
) {
1002 /* Send stream and ignore the sent flag. */
1003 ret
= send_viewer_streams(conn
->sock
, session
, 1);
1015 * Send the next index for a stream.
1017 * Return 0 on success or else a negative value.
1020 int viewer_get_next_index(struct relay_connection
*conn
)
1023 struct lttng_viewer_get_next_index request_index
;
1024 struct lttng_viewer_index viewer_index
;
1025 struct ctf_packet_index packet_index
;
1026 struct relay_viewer_stream
*vstream
;
1027 struct relay_stream
*rstream
;
1028 struct ctf_trace
*ctf_trace
;
1029 struct relay_session
*session
;
1033 DBG("Viewer get next index");
1035 health_code_update();
1037 ret
= recv_request(conn
->sock
, &request_index
, sizeof(request_index
));
1041 health_code_update();
1044 session
= session_find_by_id(conn
->sessions_ht
, conn
->session_id
);
1050 vstream
= viewer_stream_find_by_id(be64toh(request_index
.stream_id
));
1056 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
, vstream
->path_name
);
1059 memset(&viewer_index
, 0, sizeof(viewer_index
));
1062 * The viewer should not ask for index on metadata stream.
1064 if (vstream
->metadata_flag
) {
1065 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1069 /* First time, we open the index file */
1070 if (vstream
->index_read_fd
< 0) {
1071 ret
= index_open(vstream
->path_name
, vstream
->channel_name
,
1072 vstream
->tracefile_count
, vstream
->tracefile_count_current
);
1073 if (ret
== -ENOENT
) {
1075 * The index is created only when the first data packet arrives, it
1076 * might not be ready at the beginning of the session
1078 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1080 } else if (ret
< 0) {
1081 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1084 vstream
->index_read_fd
= ret
;
1087 rstream
= stream_find_by_id(relay_streams_ht
, vstream
->stream_handle
);
1090 if (!rstream
->close_flag
) {
1091 if (vstream
->abort_flag
) {
1092 /* Rotate on abort (overwrite). */
1093 DBG("Viewer rotate because of overwrite");
1094 ret
= viewer_stream_rotate(vstream
, rstream
);
1097 } else if (ret
== 1) {
1098 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1099 viewer_stream_delete(vstream
);
1100 viewer_stream_destroy(ctf_trace
, vstream
);
1103 /* ret == 0 means successful so we continue. */
1106 pthread_mutex_lock(&rstream
->viewer_stream_rotation_lock
);
1107 if (rstream
->tracefile_count_current
== vstream
->tracefile_count_current
) {
1108 if (rstream
->beacon_ts_end
!= -1ULL &&
1109 vstream
->last_sent_index
== rstream
->total_index_received
) {
1110 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_INACTIVE
);
1111 viewer_index
.timestamp_end
= htobe64(rstream
->beacon_ts_end
);
1112 pthread_mutex_unlock(&rstream
->viewer_stream_rotation_lock
);
1114 } else if (rstream
->total_index_received
<= vstream
->last_sent_index
1115 && !vstream
->close_write_flag
) {
1117 * Reader and writer are working in the same tracefile, so we care
1118 * about the number of index received and sent. Otherwise, we read
1121 pthread_mutex_unlock(&rstream
->viewer_stream_rotation_lock
);
1122 /* No new index to send, retry later. */
1123 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1127 pthread_mutex_unlock(&rstream
->viewer_stream_rotation_lock
);
1128 } else if (rstream
->close_flag
&& vstream
->close_write_flag
&&
1129 vstream
->total_index_received
== vstream
->last_sent_index
) {
1130 /* Last index sent and current tracefile closed in write */
1131 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1132 viewer_stream_delete(vstream
);
1133 viewer_stream_destroy(ctf_trace
, vstream
);
1136 vstream
->close_write_flag
= 1;
1139 if (!ctf_trace
->metadata_received
||
1140 ctf_trace
->metadata_received
> ctf_trace
->metadata_sent
) {
1141 viewer_index
.flags
|= LTTNG_VIEWER_FLAG_NEW_METADATA
;
1144 ret
= check_new_streams(vstream
->session_id
, conn
->sessions_ht
);
1147 } else if (ret
== 1) {
1148 viewer_index
.flags
|= LTTNG_VIEWER_FLAG_NEW_STREAM
;
1151 pthread_mutex_lock(&vstream
->overwrite_lock
);
1152 if (vstream
->abort_flag
) {
1154 * The file is being overwritten by the writer, we cannot * use it.
1156 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1157 pthread_mutex_unlock(&vstream
->overwrite_lock
);
1158 ret
= viewer_stream_rotate(vstream
, rstream
);
1161 } else if (ret
== 1) {
1162 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1163 viewer_stream_delete(vstream
);
1164 viewer_stream_destroy(ctf_trace
, vstream
);
1170 ret
= lttng_read(vstream
->index_read_fd
, &packet_index
,
1171 sizeof(packet_index
));
1172 pthread_mutex_unlock(&vstream
->overwrite_lock
);
1173 if (ret
< sizeof(packet_index
)) {
1175 * The tracefile is closed in write, so we read up to EOF.
1177 if (vstream
->close_write_flag
== 1) {
1178 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1179 /* Rotate on normal EOF */
1180 ret
= viewer_stream_rotate(vstream
, rstream
);
1183 } else if (ret
== 1) {
1184 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1185 viewer_stream_delete(vstream
);
1186 viewer_stream_destroy(ctf_trace
, vstream
);
1190 PERROR("Relay reading index file %d", vstream
->index_read_fd
);
1191 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1195 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_OK
);
1196 vstream
->last_sent_index
++;
1200 * Indexes are stored in big endian, no need to switch before sending.
1202 viewer_index
.offset
= packet_index
.offset
;
1203 viewer_index
.packet_size
= packet_index
.packet_size
;
1204 viewer_index
.content_size
= packet_index
.content_size
;
1205 viewer_index
.timestamp_begin
= packet_index
.timestamp_begin
;
1206 viewer_index
.timestamp_end
= packet_index
.timestamp_end
;
1207 viewer_index
.events_discarded
= packet_index
.events_discarded
;
1208 viewer_index
.stream_id
= packet_index
.stream_id
;
1211 viewer_index
.flags
= htobe32(viewer_index
.flags
);
1212 health_code_update();
1214 ret
= send_response(conn
->sock
, &viewer_index
, sizeof(viewer_index
));
1218 health_code_update();
1220 DBG("Index %" PRIu64
" for stream %" PRIu64
" sent",
1221 vstream
->last_sent_index
, vstream
->stream_handle
);
1231 * Send the next index for a stream
1233 * Return 0 on success or else a negative value.
1236 int viewer_get_packet(struct relay_connection
*conn
)
1238 int ret
, send_data
= 0;
1242 struct lttng_viewer_get_packet get_packet_info
;
1243 struct lttng_viewer_trace_packet reply
;
1244 struct relay_viewer_stream
*stream
;
1245 struct ctf_trace
*ctf_trace
;
1249 DBG2("Relay get data packet");
1251 health_code_update();
1253 ret
= recv_request(conn
->sock
, &get_packet_info
, sizeof(get_packet_info
));
1257 health_code_update();
1259 /* From this point on, the error label can be reached. */
1260 memset(&reply
, 0, sizeof(reply
));
1263 stream
= viewer_stream_find_by_id(be64toh(get_packet_info
.stream_id
));
1268 ctf_trace
= ctf_trace_find_by_path(conn
->session
->ctf_traces_ht
,
1273 * First time we read this stream, we need open the tracefile, we should
1274 * only arrive here if an index has already been sent to the viewer, so the
1275 * tracefile must exist, if it does not it is a fatal error.
1277 if (stream
->read_fd
< 0) {
1278 char fullpath
[PATH_MAX
];
1280 if (stream
->tracefile_count
> 0) {
1281 ret
= snprintf(fullpath
, PATH_MAX
, "%s/%s_%" PRIu64
, stream
->path_name
,
1282 stream
->channel_name
,
1283 stream
->tracefile_count_current
);
1285 ret
= snprintf(fullpath
, PATH_MAX
, "%s/%s", stream
->path_name
,
1286 stream
->channel_name
);
1291 ret
= open(fullpath
, O_RDONLY
);
1293 PERROR("Relay opening trace file");
1296 stream
->read_fd
= ret
;
1299 if (!ctf_trace
->metadata_received
||
1300 ctf_trace
->metadata_received
> ctf_trace
->metadata_sent
) {
1301 reply
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_ERR
);
1302 reply
.flags
|= LTTNG_VIEWER_FLAG_NEW_METADATA
;
1306 ret
= check_new_streams(stream
->session_id
, conn
->sessions_ht
);
1309 } else if (ret
== 1) {
1310 reply
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_ERR
);
1311 reply
.flags
|= LTTNG_VIEWER_FLAG_NEW_STREAM
;
1315 len
= be32toh(get_packet_info
.len
);
1316 data
= zmalloc(len
);
1318 PERROR("relay data zmalloc");
1322 ret
= lseek(stream
->read_fd
, be64toh(get_packet_info
.offset
), SEEK_SET
);
1325 * If the read fd was closed by the streaming side, the
1326 * abort_flag will be set to 1, otherwise it is an error.
1328 if (stream
->abort_flag
== 0) {
1332 reply
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_EOF
);
1335 read_len
= lttng_read(stream
->read_fd
, data
, len
);
1336 if (read_len
< len
) {
1338 * If the read fd was closed by the streaming side, the
1339 * abort_flag will be set to 1, otherwise it is an error.
1341 if (stream
->abort_flag
== 0) {
1342 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64
,
1344 be64toh(get_packet_info
.offset
));
1347 reply
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_EOF
);
1351 reply
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_OK
);
1352 reply
.len
= htobe32(len
);
1357 reply
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_ERR
);
1360 reply
.flags
= htobe32(reply
.flags
);
1362 health_code_update();
1364 ret
= send_response(conn
->sock
, &reply
, sizeof(reply
));
1368 health_code_update();
1371 health_code_update();
1372 ret
= send_response(conn
->sock
, data
, len
);
1376 health_code_update();
1379 DBG("Sent %u bytes for stream %" PRIu64
, len
,
1380 be64toh(get_packet_info
.stream_id
));
1391 * Send the session's metadata
1393 * Return 0 on success else a negative value.
1396 int viewer_get_metadata(struct relay_connection
*conn
)
1402 struct lttng_viewer_get_metadata request
;
1403 struct lttng_viewer_metadata_packet reply
;
1404 struct relay_viewer_stream
*stream
;
1405 struct ctf_trace
*ctf_trace
;
1409 DBG("Relay get metadata");
1411 health_code_update();
1413 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
1417 health_code_update();
1420 stream
= viewer_stream_find_by_id(be64toh(request
.stream_id
));
1421 if (!stream
|| !stream
->metadata_flag
) {
1422 ERR("Invalid metadata stream");
1426 ctf_trace
= ctf_trace_find_by_path(conn
->session
->ctf_traces_ht
,
1429 assert(ctf_trace
->metadata_sent
<= ctf_trace
->metadata_received
);
1431 len
= ctf_trace
->metadata_received
- ctf_trace
->metadata_sent
;
1433 reply
.status
= htobe32(LTTNG_VIEWER_NO_NEW_METADATA
);
1437 /* first time, we open the metadata file */
1438 if (stream
->read_fd
< 0) {
1439 char fullpath
[PATH_MAX
];
1441 ret
= snprintf(fullpath
, PATH_MAX
, "%s/%s", stream
->path_name
,
1442 stream
->channel_name
);
1446 ret
= open(fullpath
, O_RDONLY
);
1448 PERROR("Relay opening metadata file");
1451 stream
->read_fd
= ret
;
1454 reply
.len
= htobe64(len
);
1455 data
= zmalloc(len
);
1457 PERROR("viewer metadata zmalloc");
1461 read_len
= lttng_read(stream
->read_fd
, data
, len
);
1462 if (read_len
< len
) {
1463 PERROR("Relay reading metadata file");
1466 ctf_trace
->metadata_sent
+= read_len
;
1467 reply
.status
= htobe32(LTTNG_VIEWER_METADATA_OK
);
1471 reply
.status
= htobe32(LTTNG_VIEWER_METADATA_ERR
);
1474 health_code_update();
1475 ret
= send_response(conn
->sock
, &reply
, sizeof(reply
));
1479 health_code_update();
1482 ret
= send_response(conn
->sock
, data
, len
);
1488 DBG("Sent %" PRIu64
" bytes of metadata for stream %" PRIu64
, len
,
1489 be64toh(request
.stream_id
));
1491 DBG("Metadata sent");
1501 * Create a viewer session.
1503 * Return 0 on success or else a negative value.
1506 int viewer_create_session(struct relay_connection
*conn
)
1509 struct lttng_viewer_create_session_response resp
;
1511 DBG("Viewer create session received");
1513 resp
.status
= htobe32(LTTNG_VIEWER_CREATE_SESSION_OK
);
1514 conn
->viewer_session
= zmalloc(sizeof(conn
->viewer_session
));
1515 if (!conn
->viewer_session
) {
1516 ERR("Allocation viewer session");
1517 resp
.status
= htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR
);
1520 CDS_INIT_LIST_HEAD(&conn
->viewer_session
->sessions_head
);
1523 health_code_update();
1524 ret
= send_response(conn
->sock
, &resp
, sizeof(resp
));
1528 health_code_update();
1537 * live_relay_unknown_command: send -1 if received unknown command
1540 void live_relay_unknown_command(struct relay_connection
*conn
)
1542 struct lttcomm_relayd_generic_reply reply
;
1544 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1545 (void) send_response(conn
->sock
, &reply
, sizeof(reply
));
1549 * Process the commands received on the control socket
1552 int process_control(struct lttng_viewer_cmd
*recv_hdr
,
1553 struct relay_connection
*conn
)
1561 msg_value
= be32toh(recv_hdr
->cmd
);
1564 * Make sure we've done the version check before any command other then a
1565 * new client connection.
1567 if (msg_value
!= LTTNG_VIEWER_CONNECT
&& !conn
->version_check_done
) {
1568 ERR("Viewer conn value %" PRIu32
" before version check", msg_value
);
1573 switch (msg_value
) {
1574 case LTTNG_VIEWER_CONNECT
:
1575 ret
= viewer_connect(conn
);
1577 case LTTNG_VIEWER_LIST_SESSIONS
:
1578 ret
= viewer_list_sessions(conn
);
1580 case LTTNG_VIEWER_ATTACH_SESSION
:
1581 ret
= viewer_attach_session(conn
);
1583 case LTTNG_VIEWER_GET_NEXT_INDEX
:
1584 ret
= viewer_get_next_index(conn
);
1586 case LTTNG_VIEWER_GET_PACKET
:
1587 ret
= viewer_get_packet(conn
);
1589 case LTTNG_VIEWER_GET_METADATA
:
1590 ret
= viewer_get_metadata(conn
);
1592 case LTTNG_VIEWER_GET_NEW_STREAMS
:
1593 ret
= viewer_get_new_streams(conn
);
1595 case LTTNG_VIEWER_CREATE_SESSION
:
1596 ret
= viewer_create_session(conn
);
1599 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr
->cmd
));
1600 live_relay_unknown_command(conn
);
1610 void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
1616 (void) lttng_poll_del(events
, pollfd
);
1618 ret
= close(pollfd
);
1620 ERR("Closing pollfd %d", pollfd
);
1625 * Delete all streams for a specific session ID.
1627 static void destroy_viewer_streams_by_session(struct relay_session
*session
)
1629 struct relay_viewer_stream
*stream
;
1630 struct lttng_ht_iter iter
;
1635 cds_lfht_for_each_entry(viewer_streams_ht
->ht
, &iter
.iter
, stream
,
1637 struct ctf_trace
*ctf_trace
;
1639 health_code_update();
1640 if (stream
->session_id
!= session
->id
) {
1644 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
,
1648 viewer_stream_delete(stream
);
1650 if (stream
->metadata_flag
) {
1651 ctf_trace
->metadata_sent
= 0;
1652 ctf_trace
->viewer_metadata_stream
= NULL
;
1655 viewer_stream_destroy(ctf_trace
, stream
);
1660 static void try_destroy_streams(struct relay_session
*session
)
1662 struct ctf_trace
*ctf_trace
;
1663 struct lttng_ht_iter iter
;
1667 cds_lfht_for_each_entry(session
->ctf_traces_ht
->ht
, &iter
.iter
, ctf_trace
,
1669 /* Attempt to destroy the ctf trace of that session. */
1670 ctf_trace_try_destroy(session
, ctf_trace
);
1675 * Delete and destroy a connection.
1677 * RCU read side lock MUST be acquired.
1679 static void destroy_connection(struct lttng_ht
*relay_connections_ht
,
1680 struct relay_connection
*conn
)
1682 struct relay_session
*session
;
1684 assert(relay_connections_ht
);
1687 DBG("Cleaning connection of session ID %" PRIu64
, conn
->session_id
);
1689 connection_delete(relay_connections_ht
, conn
);
1692 session
= session_find_by_id(conn
->sessions_ht
, conn
->session_id
);
1695 * Very important that this is done before destroying the session so we
1696 * can put back every viewer stream reference from the ctf_trace.
1698 destroy_viewer_streams_by_session(session
);
1699 try_destroy_streams(session
);
1700 session_viewer_try_destroy(conn
->sessions_ht
, session
);
1704 connection_destroy(conn
);
1708 * This thread does the actual work
1711 void *thread_worker(void *data
)
1715 struct relay_connection
*conn
;
1716 struct lttng_poll_event events
;
1717 struct lttng_ht
*relay_connections_ht
;
1718 struct lttng_ht_iter iter
;
1719 struct lttng_viewer_cmd recv_hdr
;
1720 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
1721 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
1723 DBG("[thread] Live viewer relay worker started");
1725 rcu_register_thread();
1727 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_WORKER
);
1729 if (testpoint(relayd_thread_live_worker
)) {
1730 goto error_testpoint
;
1733 /* table of connections indexed on socket */
1734 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
1735 if (!relay_connections_ht
) {
1736 goto relay_connections_ht_error
;
1739 ret
= create_thread_poll_set(&events
, 2);
1741 goto error_poll_create
;
1744 ret
= lttng_poll_add(&events
, live_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1753 health_code_update();
1755 /* Infinite blocking call, waiting for transmission */
1756 DBG3("Relayd live viewer worker thread polling...");
1757 health_poll_entry();
1758 ret
= lttng_poll_wait(&events
, -1);
1762 * Restart interrupted system call.
1764 if (errno
== EINTR
) {
1773 * Process control. The control connection is prioritised so we don't
1774 * starve it with high throughput tracing data on the data
1777 for (i
= 0; i
< nb_fd
; i
++) {
1778 /* Fetch once the poll data */
1779 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1780 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1782 health_code_update();
1784 /* Thread quit pipe has been closed. Killing thread. */
1785 ret
= check_live_conn_pipe(pollfd
, revents
);
1791 /* Inspect the relay conn pipe for new connection */
1792 if (pollfd
== live_conn_pipe
[0]) {
1793 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1794 ERR("Relay live pipe error");
1796 } else if (revents
& LPOLLIN
) {
1797 ret
= lttng_read(live_conn_pipe
[0], &conn
, sizeof(conn
));
1801 conn
->sessions_ht
= sessions_ht
;
1802 connection_init(conn
);
1803 lttng_poll_add(&events
, conn
->sock
->fd
,
1804 LPOLLIN
| LPOLLRDHUP
);
1806 lttng_ht_add_unique_ulong(relay_connections_ht
,
1809 DBG("Connection socket %d added", conn
->sock
->fd
);
1813 conn
= connection_find_by_sock(relay_connections_ht
, pollfd
);
1814 /* If not found, there is a synchronization issue. */
1817 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1818 cleanup_connection_pollfd(&events
, pollfd
);
1819 destroy_connection(relay_connections_ht
, conn
);
1820 } else if (revents
& LPOLLIN
) {
1821 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &recv_hdr
,
1822 sizeof(recv_hdr
), 0);
1824 /* Connection closed */
1825 cleanup_connection_pollfd(&events
, pollfd
);
1826 destroy_connection(relay_connections_ht
, conn
);
1827 DBG("Viewer control conn closed with %d", pollfd
);
1829 ret
= process_control(&recv_hdr
, conn
);
1831 /* Clear the session on error. */
1832 cleanup_connection_pollfd(&events
, pollfd
);
1833 destroy_connection(relay_connections_ht
, conn
);
1834 DBG("Viewer connection closed with %d", pollfd
);
1845 lttng_poll_clean(&events
);
1847 /* Cleanup reamaining connection object. */
1849 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, conn
,
1851 health_code_update();
1852 destroy_connection(relay_connections_ht
, conn
);
1856 lttng_ht_destroy(relay_connections_ht
);
1857 relay_connections_ht_error
:
1858 /* Close relay conn pipes */
1859 utils_close_pipe(live_conn_pipe
);
1861 DBG("Viewer worker thread exited with error");
1863 DBG("Viewer worker thread cleanup complete");
1867 ERR("Health error occurred in %s", __func__
);
1869 health_unregister(health_relayd
);
1871 rcu_unregister_thread();
1876 * Create the relay command pipe to wake thread_manage_apps.
1877 * Closed in cleanup().
1879 static int create_conn_pipe(void)
1883 ret
= utils_create_pipe_cloexec(live_conn_pipe
);
1888 void live_stop_threads(void)
1895 ret
= pthread_join(live_listener_thread
, &status
);
1897 PERROR("pthread_join live listener");
1898 goto error
; /* join error, exit without cleanup */
1901 ret
= pthread_join(live_worker_thread
, &status
);
1903 PERROR("pthread_join live worker");
1904 goto error
; /* join error, exit without cleanup */
1907 ret
= pthread_join(live_dispatcher_thread
, &status
);
1909 PERROR("pthread_join live dispatcher");
1910 goto error
; /* join error, exit without cleanup */
1922 int live_start_threads(struct lttng_uri
*uri
,
1923 struct relay_local_data
*relay_ctx
)
1932 /* Check if daemon is UID = 0 */
1933 is_root
= !getuid();
1936 if (live_uri
->port
< 1024) {
1937 ERR("Need to be root to use ports < 1024");
1943 /* Setup the thread apps communication pipe. */
1944 if ((ret
= create_conn_pipe()) < 0) {
1948 /* Init relay command queue. */
1949 cds_wfq_init(&viewer_conn_queue
.queue
);
1951 /* Set up max poll set size */
1952 lttng_poll_set_max_size();
1954 /* Setup the dispatcher thread */
1955 ret
= pthread_create(&live_dispatcher_thread
, NULL
,
1956 thread_dispatcher
, (void *) NULL
);
1958 PERROR("pthread_create viewer dispatcher");
1959 goto exit_dispatcher
;
1962 /* Setup the worker thread */
1963 ret
= pthread_create(&live_worker_thread
, NULL
,
1964 thread_worker
, relay_ctx
);
1966 PERROR("pthread_create viewer worker");
1970 /* Setup the listener thread */
1971 ret
= pthread_create(&live_listener_thread
, NULL
,
1972 thread_listener
, (void *) NULL
);
1974 PERROR("pthread_create viewer listener");
1982 ret
= pthread_join(live_listener_thread
, &status
);
1984 PERROR("pthread_join live listener");
1985 goto error
; /* join error, exit without cleanup */
1989 ret
= pthread_join(live_worker_thread
, &status
);
1991 PERROR("pthread_join live worker");
1992 goto error
; /* join error, exit without cleanup */
1996 ret
= pthread_join(live_dispatcher_thread
, &status
);
1998 PERROR("pthread_join live dispatcher");
1999 goto error
; /* join error, exit without cleanup */