2 * Copyright (C) 2013 Julien Desfossez <jdesfossez@efficios.com>
3 * Copyright (C) 2013 David Goulet <dgoulet@efficios.com>
4 * Copyright (C) 2015 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 * SPDX-License-Identifier: GPL-2.0-only
22 #include <sys/mount.h>
23 #include <sys/resource.h>
24 #include <sys/socket.h>
26 #include <sys/types.h>
29 #include <urcu/futex.h>
30 #include <urcu/rculist.h>
31 #include <urcu/uatomic.h>
33 #include <common/common.h>
34 #include <common/compat/endian.h>
35 #include <common/compat/poll.h>
36 #include <common/compat/socket.h>
37 #include <common/defaults.h>
38 #include <common/fd-tracker/utils.h>
39 #include <common/fs-handle.h>
40 #include <common/futex.h>
41 #include <common/index/index.h>
42 #include <common/sessiond-comm/inet.h>
43 #include <common/sessiond-comm/relayd.h>
44 #include <common/sessiond-comm/sessiond-comm.h>
45 #include <common/uri.h>
46 #include <common/utils.h>
47 #include <lttng/lttng.h>
50 #include "connection.h"
51 #include "ctf-trace.h"
52 #include "health-relayd.h"
54 #include "lttng-relayd.h"
57 #include "testpoint.h"
59 #include "viewer-session.h"
60 #include "viewer-stream.h"
62 #define SESSION_BUF_DEFAULT_COUNT 16
64 static struct lttng_uri
*live_uri
;
67 * This pipe is used to inform the worker thread that a command is queued and
68 * ready to be processed.
70 static int live_conn_pipe
[2] = { -1, -1 };
72 /* Shared between threads */
73 static int live_dispatch_thread_exit
;
75 static pthread_t live_listener_thread
;
76 static pthread_t live_dispatcher_thread
;
77 static pthread_t live_worker_thread
;
80 * Relay command queue.
82 * The live_thread_listener and live_thread_dispatcher communicate with this
85 static struct relay_conn_queue viewer_conn_queue
;
87 static uint64_t last_relay_viewer_session_id
;
88 static pthread_mutex_t last_relay_viewer_session_id_lock
=
89 PTHREAD_MUTEX_INITIALIZER
;
95 void cleanup_relayd_live(void)
103 * Receive a request buffer using a given socket, destination allocated buffer
106 * Return the size of the received message or else a negative value on error
107 * with errno being set by recvmsg() syscall.
110 ssize_t
recv_request(struct lttcomm_sock
*sock
, void *buf
, size_t size
)
114 ret
= sock
->ops
->recvmsg(sock
, buf
, size
, 0);
115 if (ret
< 0 || ret
!= size
) {
117 /* Orderly shutdown. Not necessary to print an error. */
118 DBG("Socket %d did an orderly shutdown", sock
->fd
);
120 ERR("Relay failed to receive request.");
129 * Send a response buffer using a given socket, source allocated buffer of
132 * Return the size of the sent message or else a negative value on error with
133 * errno being set by sendmsg() syscall.
136 ssize_t
send_response(struct lttcomm_sock
*sock
, void *buf
, size_t size
)
140 ret
= sock
->ops
->sendmsg(sock
, buf
, size
, 0);
142 ERR("Relayd failed to send response.");
149 * Atomically check if new streams got added in one of the sessions attached
150 * and reset the flag to 0.
152 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
156 int check_new_streams(struct relay_connection
*conn
)
158 struct relay_session
*session
;
159 unsigned long current_val
;
162 if (!conn
->viewer_session
) {
166 cds_list_for_each_entry_rcu(session
,
167 &conn
->viewer_session
->session_list
,
168 viewer_session_node
) {
169 if (!session_get(session
)) {
172 current_val
= uatomic_cmpxchg(&session
->new_streams
, 1, 0);
174 session_put(session
);
185 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
186 * this function should ignore the sent flag or not.
188 * Return 0 on success or else a negative value.
191 ssize_t
send_viewer_streams(struct lttcomm_sock
*sock
,
192 uint64_t session_id
, unsigned int ignore_sent_flag
)
195 struct lttng_viewer_stream send_stream
;
196 struct lttng_ht_iter iter
;
197 struct relay_viewer_stream
*vstream
;
201 cds_lfht_for_each_entry(viewer_streams_ht
->ht
, &iter
.iter
, vstream
,
203 struct ctf_trace
*ctf_trace
;
205 health_code_update();
207 if (!viewer_stream_get(vstream
)) {
211 pthread_mutex_lock(&vstream
->stream
->lock
);
212 /* Ignore if not the same session. */
213 if (vstream
->stream
->trace
->session
->id
!= session_id
||
214 (!ignore_sent_flag
&& vstream
->sent_flag
)) {
215 pthread_mutex_unlock(&vstream
->stream
->lock
);
216 viewer_stream_put(vstream
);
220 ctf_trace
= vstream
->stream
->trace
;
221 send_stream
.id
= htobe64(vstream
->stream
->stream_handle
);
222 send_stream
.ctf_trace_id
= htobe64(ctf_trace
->id
);
223 send_stream
.metadata_flag
= htobe32(
224 vstream
->stream
->is_metadata
);
225 if (lttng_strncpy(send_stream
.path_name
, vstream
->path_name
,
226 sizeof(send_stream
.path_name
))) {
227 pthread_mutex_unlock(&vstream
->stream
->lock
);
228 viewer_stream_put(vstream
);
229 ret
= -1; /* Error. */
232 if (lttng_strncpy(send_stream
.channel_name
,
233 vstream
->channel_name
,
234 sizeof(send_stream
.channel_name
))) {
235 pthread_mutex_unlock(&vstream
->stream
->lock
);
236 viewer_stream_put(vstream
);
237 ret
= -1; /* Error. */
241 DBG("Sending stream %" PRIu64
" to viewer",
242 vstream
->stream
->stream_handle
);
243 vstream
->sent_flag
= 1;
244 pthread_mutex_unlock(&vstream
->stream
->lock
);
246 ret
= send_response(sock
, &send_stream
, sizeof(send_stream
));
247 viewer_stream_put(vstream
);
261 * Create every viewer stream possible for the given session with the seek
262 * type. Three counters *can* be return which are in order the total amount of
263 * viewer stream of the session, the number of unsent stream and the number of
264 * stream created. Those counters can be NULL and thus will be ignored.
266 * session must be locked to ensure that we see either none or all initial
267 * streams for a session, but no intermediate state..
269 * Return 0 on success or else a negative value.
271 static int make_viewer_streams(struct relay_session
*session
,
272 struct lttng_trace_chunk
*viewer_trace_chunk
,
273 enum lttng_viewer_seek seek_t
,
276 uint32_t *nb_created
,
280 struct lttng_ht_iter iter
;
281 struct ctf_trace
*ctf_trace
;
284 ASSERT_LOCKED(session
->lock
);
286 if (!viewer_trace_chunk
) {
287 ERR("Internal error: viewer session associated with session \"%s\" has a NULL trace chunk",
288 session
->session_name
);
293 if (session
->connection_closed
) {
298 * Create viewer streams for relay streams that are ready to be
299 * used for a the given session id only.
302 cds_lfht_for_each_entry(session
->ctf_traces_ht
->ht
, &iter
.iter
, ctf_trace
,
304 bool trace_has_metadata_stream
= false;
305 struct relay_stream
*stream
;
307 health_code_update();
309 if (!ctf_trace_get(ctf_trace
)) {
314 * Iterate over all the streams of the trace to see if we have a
317 cds_list_for_each_entry_rcu(
318 stream
, &ctf_trace
->stream_list
, stream_node
)
320 if (stream
->is_metadata
) {
321 trace_has_metadata_stream
= true;
327 * If there is no metadata stream in this trace at the moment
328 * and we never sent one to the viewer, skip the trace. We
329 * accept that the viewer will not see this trace at all.
331 if (!trace_has_metadata_stream
&&
332 !ctf_trace
->metadata_stream_sent_to_viewer
) {
333 ctf_trace_put(ctf_trace
);
337 cds_list_for_each_entry_rcu(stream
, &ctf_trace
->stream_list
, stream_node
) {
338 struct relay_viewer_stream
*vstream
;
340 if (!stream_get(stream
)) {
344 * stream published is protected by the session lock.
346 if (!stream
->published
) {
349 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
352 * Save that we sent the metadata stream to the
353 * viewer. So that we know what trace the viewer
356 if (stream
->is_metadata
) {
357 ctf_trace
->metadata_stream_sent_to_viewer
=
360 vstream
= viewer_stream_create(stream
,
361 viewer_trace_chunk
, seek_t
);
364 ctf_trace_put(ctf_trace
);
370 /* Update number of created stream counter. */
374 * Ensure a self-reference is preserved even
375 * after we have put our local reference.
377 if (!viewer_stream_get(vstream
)) {
378 ERR("Unable to get self-reference on viewer stream, logic error.");
382 if (!vstream
->sent_flag
&& nb_unsent
) {
383 /* Update number of unsent stream counter. */
387 /* Update number of total stream counter. */
389 if (stream
->is_metadata
) {
390 if (!stream
->closed
||
391 stream
->metadata_received
> vstream
->metadata_sent
) {
395 if (!stream
->closed
||
396 !(((int64_t) (stream
->prev_data_seq
- stream
->last_net_seq_num
)) >= 0)) {
402 /* Put local reference. */
403 viewer_stream_put(vstream
);
407 ctf_trace_put(ctf_trace
);
418 int relayd_live_stop(void)
420 /* Stop dispatch thread */
421 CMM_STORE_SHARED(live_dispatch_thread_exit
, 1);
422 futex_nto1_wake(&viewer_conn_queue
.futex
);
427 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
430 int create_named_thread_poll_set(struct lttng_poll_event
*events
,
431 int size
, const char *name
)
435 if (events
== NULL
|| size
== 0) {
440 ret
= fd_tracker_util_poll_create(the_fd_tracker
,
441 name
, events
, 1, LTTNG_CLOEXEC
);
444 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
456 * Check if the thread quit pipe was triggered.
458 * Return 1 if it was triggered else 0;
461 int check_thread_quit_pipe(int fd
, uint32_t events
)
463 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
471 int create_sock(void *data
, int *out_fd
)
474 struct lttcomm_sock
*sock
= data
;
476 ret
= lttcomm_create_sock(sock
);
487 int close_sock(void *data
, int *in_fd
)
489 struct lttcomm_sock
*sock
= data
;
491 return sock
->ops
->close(sock
);
494 static int accept_sock(void *data
, int *out_fd
)
497 /* Socks is an array of in_sock, out_sock. */
498 struct lttcomm_sock
**socks
= data
;
499 struct lttcomm_sock
*in_sock
= socks
[0];
501 socks
[1] = in_sock
->ops
->accept(in_sock
);
506 *out_fd
= socks
[1]->fd
;
512 struct lttcomm_sock
*accept_live_sock(struct lttcomm_sock
*listening_sock
,
516 struct lttcomm_sock
*socks
[2] = { listening_sock
, NULL
};
517 struct lttcomm_sock
*new_sock
= NULL
;
519 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &out_fd
,
520 (const char **) &name
, 1, accept_sock
, &socks
);
525 DBG("%s accepted, socket %d", name
, new_sock
->fd
);
531 * Create and init socket from uri.
534 struct lttcomm_sock
*init_socket(struct lttng_uri
*uri
, const char *name
)
537 struct lttcomm_sock
*sock
= NULL
;
538 char uri_str
[LTTNG_PATH_MAX
];
539 char *formated_name
= NULL
;
541 sock
= lttcomm_alloc_sock_from_uri(uri
);
543 ERR("Allocating socket");
548 * Don't fail to create the socket if the name can't be built as it is
549 * only used for debugging purposes.
551 ret
= uri_to_str_url(uri
, uri_str
, sizeof(uri_str
));
552 uri_str
[sizeof(uri_str
) - 1] = '\0';
554 ret
= asprintf(&formated_name
, "%s socket @ %s", name
,
557 formated_name
= NULL
;
561 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &sock_fd
,
562 (const char **) (formated_name
? &formated_name
: NULL
),
563 1, create_sock
, sock
);
565 DBG("Listening on %s socket %d", name
, sock
->fd
);
567 ret
= sock
->ops
->bind(sock
);
569 PERROR("Failed to bind lttng-live socket");
573 ret
= sock
->ops
->listen(sock
, -1);
583 lttcomm_destroy_sock(sock
);
589 * This thread manages the listening for new connections on the network
592 void *thread_listener(void *data
)
594 int i
, ret
, pollfd
, err
= -1;
595 uint32_t revents
, nb_fd
;
596 struct lttng_poll_event events
;
597 struct lttcomm_sock
*live_control_sock
;
599 DBG("[thread] Relay live listener started");
601 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_LISTENER
);
603 health_code_update();
605 live_control_sock
= init_socket(live_uri
, "Live listener");
606 if (!live_control_sock
) {
607 goto error_sock_control
;
610 /* Pass 2 as size here for the thread quit pipe and control sockets. */
611 ret
= create_named_thread_poll_set(&events
, 2,
612 "Live listener thread epoll");
614 goto error_create_poll
;
617 /* Add the control socket */
618 ret
= lttng_poll_add(&events
, live_control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
623 lttng_relay_notify_ready();
625 if (testpoint(relayd_thread_live_listener
)) {
626 goto error_testpoint
;
630 health_code_update();
632 DBG("Listener accepting live viewers connections");
636 ret
= lttng_poll_wait(&events
, -1);
640 * Restart interrupted system call.
642 if (errno
== EINTR
) {
649 DBG("Relay new viewer connection received");
650 for (i
= 0; i
< nb_fd
; i
++) {
651 health_code_update();
653 /* Fetch once the poll data */
654 revents
= LTTNG_POLL_GETEV(&events
, i
);
655 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
657 /* Thread quit pipe has been closed. Killing thread. */
658 ret
= check_thread_quit_pipe(pollfd
, revents
);
664 if (revents
& LPOLLIN
) {
666 * A new connection is requested, therefore a
667 * viewer connection is allocated in this
668 * thread, enqueued to a global queue and
669 * dequeued (and freed) in the worker thread.
672 struct relay_connection
*new_conn
;
673 struct lttcomm_sock
*newsock
;
675 newsock
= accept_live_sock(live_control_sock
,
676 "Live socket to client");
678 PERROR("accepting control sock");
681 DBG("Relay viewer connection accepted socket %d", newsock
->fd
);
683 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
686 PERROR("setsockopt inet");
687 lttcomm_destroy_sock(newsock
);
690 new_conn
= connection_create(newsock
, RELAY_CONNECTION_UNKNOWN
);
692 lttcomm_destroy_sock(newsock
);
695 /* Ownership assumed by the connection. */
698 /* Enqueue request for the dispatcher thread. */
699 cds_wfcq_enqueue(&viewer_conn_queue
.head
, &viewer_conn_queue
.tail
,
703 * Wake the dispatch queue futex.
704 * Implicit memory barrier with the
705 * exchange in cds_wfcq_enqueue.
707 futex_nto1_wake(&viewer_conn_queue
.futex
);
708 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
709 ERR("socket poll error");
712 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
722 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
724 if (live_control_sock
->fd
>= 0) {
725 int sock_fd
= live_control_sock
->fd
;
727 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
728 &sock_fd
, 1, close_sock
,
733 live_control_sock
->fd
= -1;
735 lttcomm_destroy_sock(live_control_sock
);
739 DBG("Live viewer listener thread exited with error");
741 health_unregister(health_relayd
);
742 DBG("Live viewer listener thread cleanup complete");
743 if (lttng_relay_stop_threads()) {
744 ERR("Error stopping threads");
750 * This thread manages the dispatching of the requests to worker threads
753 void *thread_dispatcher(void *data
)
757 struct cds_wfcq_node
*node
;
758 struct relay_connection
*conn
= NULL
;
760 DBG("[thread] Live viewer relay dispatcher started");
762 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER
);
764 if (testpoint(relayd_thread_live_dispatcher
)) {
765 goto error_testpoint
;
768 health_code_update();
771 health_code_update();
773 /* Atomically prepare the queue futex */
774 futex_nto1_prepare(&viewer_conn_queue
.futex
);
776 if (CMM_LOAD_SHARED(live_dispatch_thread_exit
)) {
781 health_code_update();
783 /* Dequeue commands */
784 node
= cds_wfcq_dequeue_blocking(&viewer_conn_queue
.head
,
785 &viewer_conn_queue
.tail
);
787 DBG("Woken up but nothing in the live-viewer "
788 "relay command queue");
789 /* Continue thread execution */
792 conn
= caa_container_of(node
, struct relay_connection
, qnode
);
793 DBG("Dispatching viewer request waiting on sock %d",
797 * Inform worker thread of the new request. This
798 * call is blocking so we can be assured that
799 * the data will be read at some point in time
800 * or wait to the end of the world :)
802 ret
= lttng_write(live_conn_pipe
[1], &conn
, sizeof(conn
));
804 PERROR("write conn pipe");
805 connection_put(conn
);
808 } while (node
!= NULL
);
810 /* Futex wait on queue. Blocking call on futex() */
812 futex_nto1_wait(&viewer_conn_queue
.futex
);
816 /* Normal exit, no error */
823 ERR("Health error occurred in %s", __func__
);
825 health_unregister(health_relayd
);
826 DBG("Live viewer dispatch thread dying");
827 if (lttng_relay_stop_threads()) {
828 ERR("Error stopping threads");
834 * Establish connection with the viewer and check the versions.
836 * Return 0 on success or else negative value.
839 int viewer_connect(struct relay_connection
*conn
)
842 struct lttng_viewer_connect reply
, msg
;
844 conn
->version_check_done
= 1;
846 health_code_update();
848 DBG("Viewer is establishing a connection to the relayd.");
850 ret
= recv_request(conn
->sock
, &msg
, sizeof(msg
));
855 health_code_update();
857 memset(&reply
, 0, sizeof(reply
));
858 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
859 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
861 /* Major versions must be the same */
862 if (reply
.major
!= be32toh(msg
.major
)) {
863 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
864 reply
.major
, be32toh(msg
.major
));
869 conn
->major
= reply
.major
;
870 /* We adapt to the lowest compatible version */
871 if (reply
.minor
<= be32toh(msg
.minor
)) {
872 conn
->minor
= reply
.minor
;
874 conn
->minor
= be32toh(msg
.minor
);
877 if (be32toh(msg
.type
) == LTTNG_VIEWER_CLIENT_COMMAND
) {
878 conn
->type
= RELAY_VIEWER_COMMAND
;
879 } else if (be32toh(msg
.type
) == LTTNG_VIEWER_CLIENT_NOTIFICATION
) {
880 conn
->type
= RELAY_VIEWER_NOTIFICATION
;
882 ERR("Unknown connection type : %u", be32toh(msg
.type
));
887 reply
.major
= htobe32(reply
.major
);
888 reply
.minor
= htobe32(reply
.minor
);
889 if (conn
->type
== RELAY_VIEWER_COMMAND
) {
891 * Increment outside of htobe64 macro, because the argument can
892 * be used more than once within the macro, and thus the
893 * operation may be undefined.
895 pthread_mutex_lock(&last_relay_viewer_session_id_lock
);
896 last_relay_viewer_session_id
++;
897 pthread_mutex_unlock(&last_relay_viewer_session_id_lock
);
898 reply
.viewer_session_id
= htobe64(last_relay_viewer_session_id
);
901 health_code_update();
903 ret
= send_response(conn
->sock
, &reply
, sizeof(reply
));
908 health_code_update();
910 DBG("Version check done using protocol %u.%u", conn
->major
, conn
->minor
);
918 * Send the viewer the list of current sessions.
919 * We need to create a copy of the hash table content because otherwise
920 * we cannot assume the number of entries stays the same between getting
921 * the number of HT elements and iteration over the HT.
923 * Return 0 on success or else a negative value.
926 int viewer_list_sessions(struct relay_connection
*conn
)
929 struct lttng_viewer_list_sessions session_list
;
930 struct lttng_ht_iter iter
;
931 struct relay_session
*session
;
932 struct lttng_viewer_session
*send_session_buf
= NULL
;
933 uint32_t buf_count
= SESSION_BUF_DEFAULT_COUNT
;
936 DBG("List sessions received");
938 send_session_buf
= zmalloc(SESSION_BUF_DEFAULT_COUNT
* sizeof(*send_session_buf
));
939 if (!send_session_buf
) {
944 cds_lfht_for_each_entry(sessions_ht
->ht
, &iter
.iter
, session
,
946 struct lttng_viewer_session
*send_session
;
948 health_code_update();
950 pthread_mutex_lock(&session
->lock
);
951 if (session
->connection_closed
) {
952 /* Skip closed session */
955 if (!session
->current_trace_chunk
) {
957 * Skip un-attachable session. It is either
958 * being destroyed or has not had a trace
959 * chunk created against it yet.
964 if (count
>= buf_count
) {
965 struct lttng_viewer_session
*newbuf
;
966 uint32_t new_buf_count
= buf_count
<< 1;
968 newbuf
= realloc(send_session_buf
,
969 new_buf_count
* sizeof(*send_session_buf
));
974 send_session_buf
= newbuf
;
975 buf_count
= new_buf_count
;
977 send_session
= &send_session_buf
[count
];
978 if (lttng_strncpy(send_session
->session_name
,
979 session
->session_name
,
980 sizeof(send_session
->session_name
))) {
984 if (lttng_strncpy(send_session
->hostname
, session
->hostname
,
985 sizeof(send_session
->hostname
))) {
989 send_session
->id
= htobe64(session
->id
);
990 send_session
->live_timer
= htobe32(session
->live_timer
);
991 if (session
->viewer_attached
) {
992 send_session
->clients
= htobe32(1);
994 send_session
->clients
= htobe32(0);
996 send_session
->streams
= htobe32(session
->stream_count
);
999 pthread_mutex_unlock(&session
->lock
);
1002 pthread_mutex_unlock(&session
->lock
);
1010 session_list
.sessions_count
= htobe32(count
);
1012 health_code_update();
1014 ret
= send_response(conn
->sock
, &session_list
, sizeof(session_list
));
1019 health_code_update();
1021 ret
= send_response(conn
->sock
, send_session_buf
,
1022 count
* sizeof(*send_session_buf
));
1026 health_code_update();
1030 free(send_session_buf
);
1035 * Send the viewer the list of current streams.
1038 int viewer_get_new_streams(struct relay_connection
*conn
)
1040 int ret
, send_streams
= 0;
1041 uint32_t nb_created
= 0, nb_unsent
= 0, nb_streams
= 0, nb_total
= 0;
1042 struct lttng_viewer_new_streams_request request
;
1043 struct lttng_viewer_new_streams_response response
;
1044 struct relay_session
*session
= NULL
;
1045 uint64_t session_id
;
1046 bool closed
= false;
1050 DBG("Get new streams received");
1052 health_code_update();
1054 /* Receive the request from the connected client. */
1055 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
1059 session_id
= be64toh(request
.session_id
);
1061 health_code_update();
1063 memset(&response
, 0, sizeof(response
));
1065 session
= session_get_by_id(session_id
);
1067 DBG("Relay session %" PRIu64
" not found", session_id
);
1068 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR
);
1072 if (!viewer_session_is_attached(conn
->viewer_session
, session
)) {
1073 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR
);
1077 pthread_mutex_lock(&session
->lock
);
1078 ret
= make_viewer_streams(session
,
1079 conn
->viewer_session
->current_trace_chunk
,
1080 LTTNG_VIEWER_SEEK_LAST
, &nb_total
, &nb_unsent
,
1081 &nb_created
, &closed
);
1083 goto error_unlock_session
;
1086 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_OK
);
1088 /* Only send back the newly created streams with the unsent ones. */
1089 nb_streams
= nb_created
+ nb_unsent
;
1090 response
.streams_count
= htobe32(nb_streams
);
1093 * If the session is closed, HUP when there are no more streams
1096 if (closed
&& nb_total
== 0) {
1098 response
.streams_count
= 0;
1099 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP
);
1100 goto send_reply_unlock
;
1103 pthread_mutex_unlock(&session
->lock
);
1106 health_code_update();
1107 ret
= send_response(conn
->sock
, &response
, sizeof(response
));
1109 goto end_put_session
;
1111 health_code_update();
1114 * Unknown or empty session, just return gracefully, the viewer
1115 * knows what is happening.
1117 if (!send_streams
|| !nb_streams
) {
1119 goto end_put_session
;
1123 * Send stream and *DON'T* ignore the sent flag so every viewer
1124 * streams that were not sent from that point will be sent to
1127 ret
= send_viewer_streams(conn
->sock
, session_id
, 0);
1129 goto end_put_session
;
1134 session_put(session
);
1138 error_unlock_session
:
1139 pthread_mutex_unlock(&session
->lock
);
1140 session_put(session
);
1145 * Send the viewer the list of current sessions.
1148 int viewer_attach_session(struct relay_connection
*conn
)
1150 int send_streams
= 0;
1152 uint32_t nb_streams
= 0;
1153 enum lttng_viewer_seek seek_type
;
1154 struct lttng_viewer_attach_session_request request
;
1155 struct lttng_viewer_attach_session_response response
;
1156 struct relay_session
*session
= NULL
;
1157 enum lttng_viewer_attach_return_code viewer_attach_status
;
1158 bool closed
= false;
1159 uint64_t session_id
;
1163 health_code_update();
1165 /* Receive the request from the connected client. */
1166 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
1171 session_id
= be64toh(request
.session_id
);
1172 health_code_update();
1174 memset(&response
, 0, sizeof(response
));
1176 if (!conn
->viewer_session
) {
1177 DBG("Client trying to attach before creating a live viewer session");
1178 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION
);
1182 session
= session_get_by_id(session_id
);
1184 DBG("Relay session %" PRIu64
" not found", session_id
);
1185 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_UNK
);
1188 DBG("Attach session ID %" PRIu64
" received", session_id
);
1190 pthread_mutex_lock(&session
->lock
);
1191 if (!session
->current_trace_chunk
) {
1193 * Session is either being destroyed or it never had a trace
1194 * chunk created against it.
1196 DBG("Session requested by live client has no current trace chunk, returning unknown session");
1197 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_UNK
);
1200 if (session
->live_timer
== 0) {
1201 DBG("Not live session");
1202 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE
);
1207 viewer_attach_status
= viewer_session_attach(conn
->viewer_session
,
1209 if (viewer_attach_status
!= LTTNG_VIEWER_ATTACH_OK
) {
1210 response
.status
= htobe32(viewer_attach_status
);
1214 switch (be32toh(request
.seek
)) {
1215 case LTTNG_VIEWER_SEEK_BEGINNING
:
1216 case LTTNG_VIEWER_SEEK_LAST
:
1217 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_OK
);
1218 seek_type
= be32toh(request
.seek
);
1221 ERR("Wrong seek parameter");
1222 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR
);
1227 ret
= make_viewer_streams(session
,
1228 conn
->viewer_session
->current_trace_chunk
, seek_type
,
1229 &nb_streams
, NULL
, NULL
, &closed
);
1231 goto end_put_session
;
1233 pthread_mutex_unlock(&session
->lock
);
1234 session_put(session
);
1237 response
.streams_count
= htobe32(nb_streams
);
1239 * If the session is closed when the viewer is attaching, it
1240 * means some of the streams may have been concurrently removed,
1241 * so we don't allow the viewer to attach, even if there are
1242 * streams available.
1246 response
.streams_count
= 0;
1247 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_UNK
);
1252 health_code_update();
1253 ret
= send_response(conn
->sock
, &response
, sizeof(response
));
1255 goto end_put_session
;
1257 health_code_update();
1260 * Unknown or empty session, just return gracefully, the viewer
1261 * knows what is happening.
1263 if (!send_streams
|| !nb_streams
) {
1265 goto end_put_session
;
1268 /* Send stream and ignore the sent flag. */
1269 ret
= send_viewer_streams(conn
->sock
, session_id
, 1);
1271 goto end_put_session
;
1276 pthread_mutex_unlock(&session
->lock
);
1277 session_put(session
);
1284 * Open the index file if needed for the given vstream.
1286 * If an index file is successfully opened, the vstream will set it as its
1287 * current index file.
1289 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1291 * Called with rstream lock held.
1293 static int try_open_index(struct relay_viewer_stream
*vstream
,
1294 struct relay_stream
*rstream
)
1297 const uint32_t connection_major
= rstream
->trace
->session
->major
;
1298 const uint32_t connection_minor
= rstream
->trace
->session
->minor
;
1299 enum lttng_trace_chunk_status chunk_status
;
1301 if (vstream
->index_file
) {
1306 * First time, we open the index file and at least one index is ready.
1308 if (rstream
->index_received_seqcount
== 0) {
1312 chunk_status
= lttng_index_file_create_from_trace_chunk_read_only(
1313 vstream
->stream_file
.trace_chunk
, rstream
->path_name
,
1314 rstream
->channel_name
, rstream
->tracefile_size
,
1315 vstream
->current_tracefile_id
,
1316 lttng_to_index_major(connection_major
, connection_minor
),
1317 lttng_to_index_minor(connection_major
, connection_minor
),
1318 true, &vstream
->index_file
);
1319 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
1320 if (chunk_status
== LTTNG_TRACE_CHUNK_STATUS_NO_FILE
) {
1332 * Check the status of the index for the given stream. This function
1333 * updates the index structure if needed and can put (close) the vstream
1334 * in the HUP situation.
1336 * Return 0 means that we can proceed with the index. A value of 1 means
1337 * that the index has been updated and is ready to be sent to the
1338 * client. A negative value indicates an error that can't be handled.
1340 * Called with rstream lock held.
1342 static int check_index_status(struct relay_viewer_stream
*vstream
,
1343 struct relay_stream
*rstream
, struct ctf_trace
*trace
,
1344 struct lttng_viewer_index
*index
)
1348 DBG("Check index status: index_received_seqcount %" PRIu64
" "
1349 "index_sent_seqcount %" PRIu64
" "
1350 "for stream %" PRIu64
,
1351 rstream
->index_received_seqcount
,
1352 vstream
->index_sent_seqcount
,
1353 vstream
->stream
->stream_handle
);
1354 if ((trace
->session
->connection_closed
|| rstream
->closed
)
1355 && rstream
->index_received_seqcount
1356 == vstream
->index_sent_seqcount
) {
1358 * Last index sent and session connection or relay
1359 * stream are closed.
1361 index
->status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1363 } else if (rstream
->beacon_ts_end
!= -1ULL &&
1364 (rstream
->index_received_seqcount
== 0 ||
1365 (vstream
->index_sent_seqcount
!= 0 &&
1366 rstream
->index_received_seqcount
1367 <= vstream
->index_sent_seqcount
))) {
1369 * We've received a synchronization beacon and the last index
1370 * available has been sent, the index for now is inactive.
1372 * In this case, we have received a beacon which allows us to
1373 * inform the client of a time interval during which we can
1374 * guarantee that there are no events to read (and never will
1377 * The sent seqcount can grow higher than receive seqcount on
1378 * clear because the rotation performed by clear will push
1379 * the index_sent_seqcount ahead (see
1380 * viewer_stream_sync_tracefile_array_tail) and skip over
1381 * packet sequence numbers.
1383 index
->status
= htobe32(LTTNG_VIEWER_INDEX_INACTIVE
);
1384 index
->timestamp_end
= htobe64(rstream
->beacon_ts_end
);
1385 index
->stream_id
= htobe64(rstream
->ctf_stream_id
);
1386 DBG("Check index status: inactive with beacon, for stream %" PRIu64
,
1387 vstream
->stream
->stream_handle
);
1389 } else if (rstream
->index_received_seqcount
== 0 ||
1390 (vstream
->index_sent_seqcount
!= 0 &&
1391 rstream
->index_received_seqcount
1392 <= vstream
->index_sent_seqcount
)) {
1394 * This checks whether received <= sent seqcount. In
1395 * this case, we have not received a beacon. Therefore,
1396 * we can only ask the client to retry later.
1398 * The sent seqcount can grow higher than receive seqcount on
1399 * clear because the rotation performed by clear will push
1400 * the index_sent_seqcount ahead (see
1401 * viewer_stream_sync_tracefile_array_tail) and skip over
1402 * packet sequence numbers.
1404 index
->status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1405 DBG("Check index status: retry for stream %" PRIu64
,
1406 vstream
->stream
->stream_handle
);
1408 } else if (!tracefile_array_seq_in_file(rstream
->tfa
,
1409 vstream
->current_tracefile_id
,
1410 vstream
->index_sent_seqcount
)) {
1412 * The next index we want to send cannot be read either
1413 * because we need to perform a rotation, or due to
1414 * the producer having overwritten its trace file.
1416 DBG("Viewer stream %" PRIu64
" rotation",
1417 vstream
->stream
->stream_handle
);
1418 ret
= viewer_stream_rotate(vstream
);
1420 /* EOF across entire stream. */
1421 index
->status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1425 * If we have been pushed due to overwrite, it
1426 * necessarily means there is data that can be read in
1427 * the stream. If we rotated because we reached the end
1428 * of a tracefile, it means the following tracefile
1429 * needs to contain at least one index, else we would
1430 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1431 * viewer. The updated index_sent_seqcount needs to
1432 * point to a readable index entry now.
1434 * In the case where we "rotate" on a single file, we
1435 * can end up in a case where the requested index is
1436 * still unavailable.
1438 if (rstream
->tracefile_count
== 1 &&
1439 !tracefile_array_seq_in_file(
1441 vstream
->current_tracefile_id
,
1442 vstream
->index_sent_seqcount
)) {
1443 index
->status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1444 DBG("Check index status: retry: "
1445 "tracefile array sequence number %" PRIu64
1446 " not in file for stream %" PRIu64
,
1447 vstream
->index_sent_seqcount
,
1448 vstream
->stream
->stream_handle
);
1451 assert(tracefile_array_seq_in_file(rstream
->tfa
,
1452 vstream
->current_tracefile_id
,
1453 vstream
->index_sent_seqcount
));
1455 /* ret == 0 means successful so we continue. */
1460 viewer_stream_put(vstream
);
1466 * Send the next index for a stream.
1468 * Return 0 on success or else a negative value.
1471 int viewer_get_next_index(struct relay_connection
*conn
)
1474 struct lttng_viewer_get_next_index request_index
;
1475 struct lttng_viewer_index viewer_index
;
1476 struct ctf_packet_index packet_index
;
1477 struct relay_viewer_stream
*vstream
= NULL
;
1478 struct relay_stream
*rstream
= NULL
;
1479 struct ctf_trace
*ctf_trace
= NULL
;
1480 struct relay_viewer_stream
*metadata_viewer_stream
= NULL
;
1484 DBG("Viewer get next index");
1486 memset(&viewer_index
, 0, sizeof(viewer_index
));
1487 health_code_update();
1489 ret
= recv_request(conn
->sock
, &request_index
, sizeof(request_index
));
1493 health_code_update();
1495 vstream
= viewer_stream_get_by_id(be64toh(request_index
.stream_id
));
1497 DBG("Client requested index of unknown stream id %" PRIu64
,
1498 (uint64_t) be64toh(request_index
.stream_id
));
1499 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1503 /* Use back. ref. Protected by refcounts. */
1504 rstream
= vstream
->stream
;
1505 ctf_trace
= rstream
->trace
;
1507 /* metadata_viewer_stream may be NULL. */
1508 metadata_viewer_stream
=
1509 ctf_trace_get_viewer_metadata_stream(ctf_trace
);
1511 pthread_mutex_lock(&rstream
->lock
);
1514 * The viewer should not ask for index on metadata stream.
1516 if (rstream
->is_metadata
) {
1517 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1521 if (rstream
->ongoing_rotation
.is_set
) {
1522 /* Rotation is ongoing, try again later. */
1523 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1527 if (rstream
->trace
->session
->ongoing_rotation
) {
1528 /* Rotation is ongoing, try again later. */
1529 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1533 if (rstream
->trace_chunk
) {
1534 uint64_t rchunk_id
, vchunk_id
;
1537 * If the relay stream is not yet closed, ensure the viewer
1538 * chunk matches the relay chunk after clear.
1540 if (lttng_trace_chunk_get_id(rstream
->trace_chunk
,
1541 &rchunk_id
) != LTTNG_TRACE_CHUNK_STATUS_OK
) {
1542 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1545 if (lttng_trace_chunk_get_id(
1546 conn
->viewer_session
->current_trace_chunk
,
1547 &vchunk_id
) != LTTNG_TRACE_CHUNK_STATUS_OK
) {
1548 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1552 if (rchunk_id
!= vchunk_id
) {
1553 DBG("Relay and viewer chunk ids differ: "
1554 "rchunk_id %" PRIu64
" vchunk_id %" PRIu64
,
1555 rchunk_id
, vchunk_id
);
1557 lttng_trace_chunk_put(
1558 conn
->viewer_session
->current_trace_chunk
);
1559 conn
->viewer_session
->current_trace_chunk
= NULL
;
1560 ret
= viewer_session_set_trace_chunk_copy(
1561 conn
->viewer_session
,
1562 rstream
->trace_chunk
);
1564 viewer_index
.status
=
1565 htobe32(LTTNG_VIEWER_INDEX_ERR
);
1570 if (conn
->viewer_session
->current_trace_chunk
!=
1571 vstream
->stream_file
.trace_chunk
) {
1572 bool acquired_reference
;
1574 DBG("Viewer session and viewer stream chunk differ: "
1575 "vsession chunk %p vstream chunk %p",
1576 conn
->viewer_session
->current_trace_chunk
,
1577 vstream
->stream_file
.trace_chunk
);
1578 lttng_trace_chunk_put(vstream
->stream_file
.trace_chunk
);
1579 acquired_reference
= lttng_trace_chunk_get(conn
->viewer_session
->current_trace_chunk
);
1580 assert(acquired_reference
);
1581 vstream
->stream_file
.trace_chunk
=
1582 conn
->viewer_session
->current_trace_chunk
;
1583 viewer_stream_sync_tracefile_array_tail(vstream
);
1584 viewer_stream_close_files(vstream
);
1587 ret
= check_index_status(vstream
, rstream
, ctf_trace
, &viewer_index
);
1590 } else if (ret
== 1) {
1592 * We have no index to send and check_index_status has populated
1593 * viewer_index's status.
1597 /* At this point, ret is 0 thus we will be able to read the index. */
1600 /* Try to open an index if one is needed for that stream. */
1601 ret
= try_open_index(vstream
, rstream
);
1602 if (ret
== -ENOENT
) {
1603 if (rstream
->closed
) {
1604 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1607 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1612 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1617 * vstream->stream_fd may be NULL if it has been closed by
1618 * tracefile rotation, or if we are at the beginning of the
1619 * stream. We open the data stream file here to protect against
1620 * overwrite caused by tracefile rotation (in association with
1621 * unlink performed before overwrite).
1623 if (!vstream
->stream_file
.handle
) {
1624 char file_path
[LTTNG_PATH_MAX
];
1625 enum lttng_trace_chunk_status status
;
1626 struct fs_handle
*fs_handle
;
1628 ret
= utils_stream_file_path(rstream
->path_name
,
1629 rstream
->channel_name
, rstream
->tracefile_size
,
1630 vstream
->current_tracefile_id
, NULL
, file_path
,
1637 * It is possible the the file we are trying to open is
1638 * missing if the stream has been closed (application exits with
1639 * per-pid buffers) and a clear command has been performed.
1641 status
= lttng_trace_chunk_open_fs_handle(
1642 vstream
->stream_file
.trace_chunk
,
1643 file_path
, O_RDONLY
, 0, &fs_handle
, true);
1644 if (status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
1645 if (status
== LTTNG_TRACE_CHUNK_STATUS_NO_FILE
&&
1647 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1650 PERROR("Failed to open trace file for viewer stream");
1653 vstream
->stream_file
.handle
= fs_handle
;
1656 ret
= check_new_streams(conn
);
1658 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1660 } else if (ret
== 1) {
1661 viewer_index
.flags
|= LTTNG_VIEWER_FLAG_NEW_STREAM
;
1664 ret
= lttng_index_file_read(vstream
->index_file
, &packet_index
);
1666 ERR("Relay error reading index file");
1667 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1670 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_OK
);
1671 vstream
->index_sent_seqcount
++;
1675 * Indexes are stored in big endian, no need to switch before sending.
1677 DBG("Sending viewer index for stream %" PRIu64
" offset %" PRIu64
,
1678 rstream
->stream_handle
,
1679 (uint64_t) be64toh(packet_index
.offset
));
1680 viewer_index
.offset
= packet_index
.offset
;
1681 viewer_index
.packet_size
= packet_index
.packet_size
;
1682 viewer_index
.content_size
= packet_index
.content_size
;
1683 viewer_index
.timestamp_begin
= packet_index
.timestamp_begin
;
1684 viewer_index
.timestamp_end
= packet_index
.timestamp_end
;
1685 viewer_index
.events_discarded
= packet_index
.events_discarded
;
1686 viewer_index
.stream_id
= packet_index
.stream_id
;
1690 pthread_mutex_unlock(&rstream
->lock
);
1693 if (metadata_viewer_stream
) {
1694 pthread_mutex_lock(&metadata_viewer_stream
->stream
->lock
);
1695 DBG("get next index metadata check: recv %" PRIu64
1697 metadata_viewer_stream
->stream
->metadata_received
,
1698 metadata_viewer_stream
->metadata_sent
);
1699 if (!metadata_viewer_stream
->stream
->metadata_received
||
1700 metadata_viewer_stream
->stream
->metadata_received
>
1701 metadata_viewer_stream
->metadata_sent
) {
1702 viewer_index
.flags
|= LTTNG_VIEWER_FLAG_NEW_METADATA
;
1704 pthread_mutex_unlock(&metadata_viewer_stream
->stream
->lock
);
1707 viewer_index
.flags
= htobe32(viewer_index
.flags
);
1708 health_code_update();
1710 ret
= send_response(conn
->sock
, &viewer_index
, sizeof(viewer_index
));
1714 health_code_update();
1717 DBG("Index %" PRIu64
" for stream %" PRIu64
" sent",
1718 vstream
->index_sent_seqcount
,
1719 vstream
->stream
->stream_handle
);
1722 if (metadata_viewer_stream
) {
1723 viewer_stream_put(metadata_viewer_stream
);
1726 viewer_stream_put(vstream
);
1731 pthread_mutex_unlock(&rstream
->lock
);
1732 if (metadata_viewer_stream
) {
1733 viewer_stream_put(metadata_viewer_stream
);
1735 viewer_stream_put(vstream
);
1740 * Send the next index for a stream
1742 * Return 0 on success or else a negative value.
1745 int viewer_get_packet(struct relay_connection
*conn
)
1750 struct lttng_viewer_get_packet get_packet_info
;
1751 struct lttng_viewer_trace_packet reply_header
;
1752 struct relay_viewer_stream
*vstream
= NULL
;
1753 uint32_t reply_size
= sizeof(reply_header
);
1754 uint32_t packet_data_len
= 0;
1758 DBG2("Relay get data packet");
1760 health_code_update();
1762 ret
= recv_request(conn
->sock
, &get_packet_info
,
1763 sizeof(get_packet_info
));
1767 health_code_update();
1769 /* From this point on, the error label can be reached. */
1770 memset(&reply_header
, 0, sizeof(reply_header
));
1771 stream_id
= (uint64_t) be64toh(get_packet_info
.stream_id
);
1773 vstream
= viewer_stream_get_by_id(stream_id
);
1775 DBG("Client requested packet of unknown stream id %" PRIu64
,
1777 reply_header
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_ERR
);
1778 goto send_reply_nolock
;
1780 packet_data_len
= be32toh(get_packet_info
.len
);
1781 reply_size
+= packet_data_len
;
1784 reply
= zmalloc(reply_size
);
1786 PERROR("packet reply zmalloc");
1787 reply_size
= sizeof(reply_header
);
1791 pthread_mutex_lock(&vstream
->stream
->lock
);
1792 lseek_ret
= fs_handle_seek(vstream
->stream_file
.handle
,
1793 be64toh(get_packet_info
.offset
), SEEK_SET
);
1794 if (lseek_ret
< 0) {
1795 PERROR("Failed to seek file system handle of viewer stream %" PRIu64
1796 " to offset %" PRIu64
,
1798 (uint64_t) be64toh(get_packet_info
.offset
));
1801 read_len
= fs_handle_read(vstream
->stream_file
.handle
,
1802 reply
+ sizeof(reply_header
), packet_data_len
);
1803 if (read_len
< packet_data_len
) {
1804 PERROR("Failed to read from file system handle of viewer stream id %" PRIu64
1805 ", offset: %" PRIu64
,
1807 (uint64_t) be64toh(get_packet_info
.offset
));
1810 reply_header
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_OK
);
1811 reply_header
.len
= htobe32(packet_data_len
);
1815 reply_header
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_ERR
);
1819 pthread_mutex_unlock(&vstream
->stream
->lock
);
1823 health_code_update();
1826 memcpy(reply
, &reply_header
, sizeof(reply_header
));
1827 ret
= send_response(conn
->sock
, reply
, reply_size
);
1829 /* No reply to send. */
1830 ret
= send_response(conn
->sock
, &reply_header
,
1834 health_code_update();
1836 PERROR("sendmsg of packet data failed");
1840 DBG("Sent %u bytes for stream %" PRIu64
, reply_size
, stream_id
);
1846 viewer_stream_put(vstream
);
1852 * Send the session's metadata
1854 * Return 0 on success else a negative value.
1857 int viewer_get_metadata(struct relay_connection
*conn
)
1864 struct lttng_viewer_get_metadata request
;
1865 struct lttng_viewer_metadata_packet reply
;
1866 struct relay_viewer_stream
*vstream
= NULL
;
1870 DBG("Relay get metadata");
1872 health_code_update();
1874 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
1878 health_code_update();
1880 memset(&reply
, 0, sizeof(reply
));
1882 vstream
= viewer_stream_get_by_id(be64toh(request
.stream_id
));
1885 * The metadata stream can be closed by a CLOSE command
1886 * just before we attach. It can also be closed by
1887 * per-pid tracing during tracing. Therefore, it is
1888 * possible that we cannot find this viewer stream.
1889 * Reply back to the client with an error if we cannot
1892 DBG("Client requested metadata of unknown stream id %" PRIu64
,
1893 (uint64_t) be64toh(request
.stream_id
));
1894 reply
.status
= htobe32(LTTNG_VIEWER_METADATA_ERR
);
1897 pthread_mutex_lock(&vstream
->stream
->lock
);
1898 if (!vstream
->stream
->is_metadata
) {
1899 ERR("Invalid metadata stream");
1903 if (vstream
->metadata_sent
>= vstream
->stream
->metadata_received
) {
1905 * The live viewers expect to receive a NO_NEW_METADATA
1906 * status before a stream disappears, otherwise they abort the
1907 * entire live connection when receiving an error status.
1909 * Clear feature resets the metadata_sent to 0 until the
1910 * same metadata is received again.
1912 reply
.status
= htobe32(LTTNG_VIEWER_NO_NEW_METADATA
);
1914 * The live viewer considers a closed 0 byte metadata stream as
1917 if (vstream
->metadata_sent
> 0) {
1918 vstream
->stream
->no_new_metadata_notified
= true;
1919 if (vstream
->stream
->closed
) {
1920 /* Release ownership for the viewer metadata stream. */
1921 viewer_stream_put(vstream
);
1927 len
= vstream
->stream
->metadata_received
- vstream
->metadata_sent
;
1929 /* first time, we open the metadata file */
1930 if (!vstream
->stream_file
.handle
) {
1931 struct fs_handle
*fs_handle
;
1932 char file_path
[LTTNG_PATH_MAX
];
1933 enum lttng_trace_chunk_status status
;
1934 struct relay_stream
*rstream
= vstream
->stream
;
1936 ret
= utils_stream_file_path(rstream
->path_name
,
1937 rstream
->channel_name
, rstream
->tracefile_size
,
1938 vstream
->current_tracefile_id
, NULL
, file_path
,
1945 * It is possible the the metadata file we are trying to open is
1946 * missing if the stream has been closed (application exits with
1947 * per-pid buffers) and a clear command has been performed.
1949 status
= lttng_trace_chunk_open_fs_handle(
1950 vstream
->stream_file
.trace_chunk
,
1951 file_path
, O_RDONLY
, 0, &fs_handle
, true);
1952 if (status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
1953 if (status
== LTTNG_TRACE_CHUNK_STATUS_NO_FILE
) {
1954 reply
.status
= htobe32(LTTNG_VIEWER_NO_NEW_METADATA
);
1956 if (vstream
->stream
->closed
) {
1957 viewer_stream_put(vstream
);
1961 PERROR("Failed to open metadata file for viewer stream");
1964 vstream
->stream_file
.handle
= fs_handle
;
1967 reply
.len
= htobe64(len
);
1968 data
= zmalloc(len
);
1970 PERROR("viewer metadata zmalloc");
1974 fd
= fs_handle_get_fd(vstream
->stream_file
.handle
);
1976 ERR("Failed to restore viewer stream file system handle");
1979 read_len
= lttng_read(fd
, data
, len
);
1980 fs_handle_put_fd(vstream
->stream_file
.handle
);
1982 if (read_len
< len
) {
1983 PERROR("Relay reading metadata file");
1986 vstream
->metadata_sent
+= read_len
;
1987 reply
.status
= htobe32(LTTNG_VIEWER_METADATA_OK
);
1992 reply
.status
= htobe32(LTTNG_VIEWER_METADATA_ERR
);
1995 health_code_update();
1997 pthread_mutex_unlock(&vstream
->stream
->lock
);
1999 ret
= send_response(conn
->sock
, &reply
, sizeof(reply
));
2003 health_code_update();
2006 ret
= send_response(conn
->sock
, data
, len
);
2012 DBG("Sent %" PRIu64
" bytes of metadata for stream %" PRIu64
, len
,
2013 (uint64_t) be64toh(request
.stream_id
));
2015 DBG("Metadata sent");
2021 viewer_stream_put(vstream
);
2027 * Create a viewer session.
2029 * Return 0 on success or else a negative value.
2032 int viewer_create_session(struct relay_connection
*conn
)
2035 struct lttng_viewer_create_session_response resp
;
2037 DBG("Viewer create session received");
2039 memset(&resp
, 0, sizeof(resp
));
2040 resp
.status
= htobe32(LTTNG_VIEWER_CREATE_SESSION_OK
);
2041 conn
->viewer_session
= viewer_session_create();
2042 if (!conn
->viewer_session
) {
2043 ERR("Allocation viewer session");
2044 resp
.status
= htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR
);
2049 health_code_update();
2050 ret
= send_response(conn
->sock
, &resp
, sizeof(resp
));
2054 health_code_update();
2062 * Detach a viewer session.
2064 * Return 0 on success or else a negative value.
2067 int viewer_detach_session(struct relay_connection
*conn
)
2070 struct lttng_viewer_detach_session_response response
;
2071 struct lttng_viewer_detach_session_request request
;
2072 struct relay_session
*session
= NULL
;
2073 uint64_t viewer_session_to_close
;
2075 DBG("Viewer detach session received");
2079 health_code_update();
2081 /* Receive the request from the connected client. */
2082 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
2086 viewer_session_to_close
= be64toh(request
.session_id
);
2088 if (!conn
->viewer_session
) {
2089 DBG("Client trying to detach before creating a live viewer session");
2090 response
.status
= htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR
);
2094 health_code_update();
2096 memset(&response
, 0, sizeof(response
));
2097 DBG("Detaching from session ID %" PRIu64
, viewer_session_to_close
);
2099 session
= session_get_by_id(be64toh(request
.session_id
));
2101 DBG("Relay session %" PRIu64
" not found",
2102 (uint64_t) be64toh(request
.session_id
));
2103 response
.status
= htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK
);
2107 ret
= viewer_session_is_attached(conn
->viewer_session
, session
);
2109 DBG("Not attached to this session");
2110 response
.status
= htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR
);
2111 goto send_reply_put
;
2114 viewer_session_close_one_session(conn
->viewer_session
, session
);
2115 response
.status
= htobe32(LTTNG_VIEWER_DETACH_SESSION_OK
);
2116 DBG("Session %" PRIu64
" detached.", viewer_session_to_close
);
2119 session_put(session
);
2122 health_code_update();
2123 ret
= send_response(conn
->sock
, &response
, sizeof(response
));
2127 health_code_update();
2135 * live_relay_unknown_command: send -1 if received unknown command
2138 void live_relay_unknown_command(struct relay_connection
*conn
)
2140 struct lttcomm_relayd_generic_reply reply
;
2142 memset(&reply
, 0, sizeof(reply
));
2143 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2144 (void) send_response(conn
->sock
, &reply
, sizeof(reply
));
2148 * Process the commands received on the control socket
2151 int process_control(struct lttng_viewer_cmd
*recv_hdr
,
2152 struct relay_connection
*conn
)
2157 msg_value
= be32toh(recv_hdr
->cmd
);
2160 * Make sure we've done the version check before any command other then a
2161 * new client connection.
2163 if (msg_value
!= LTTNG_VIEWER_CONNECT
&& !conn
->version_check_done
) {
2164 ERR("Viewer conn value %" PRIu32
" before version check", msg_value
);
2169 switch (msg_value
) {
2170 case LTTNG_VIEWER_CONNECT
:
2171 ret
= viewer_connect(conn
);
2173 case LTTNG_VIEWER_LIST_SESSIONS
:
2174 ret
= viewer_list_sessions(conn
);
2176 case LTTNG_VIEWER_ATTACH_SESSION
:
2177 ret
= viewer_attach_session(conn
);
2179 case LTTNG_VIEWER_GET_NEXT_INDEX
:
2180 ret
= viewer_get_next_index(conn
);
2182 case LTTNG_VIEWER_GET_PACKET
:
2183 ret
= viewer_get_packet(conn
);
2185 case LTTNG_VIEWER_GET_METADATA
:
2186 ret
= viewer_get_metadata(conn
);
2188 case LTTNG_VIEWER_GET_NEW_STREAMS
:
2189 ret
= viewer_get_new_streams(conn
);
2191 case LTTNG_VIEWER_CREATE_SESSION
:
2192 ret
= viewer_create_session(conn
);
2194 case LTTNG_VIEWER_DETACH_SESSION
:
2195 ret
= viewer_detach_session(conn
);
2198 ERR("Received unknown viewer command (%u)",
2199 be32toh(recv_hdr
->cmd
));
2200 live_relay_unknown_command(conn
);
2210 void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2214 (void) lttng_poll_del(events
, pollfd
);
2216 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
, &pollfd
, 1,
2217 fd_tracker_util_close_fd
, NULL
);
2219 ERR("Closing pollfd %d", pollfd
);
2224 * This thread does the actual work
2227 void *thread_worker(void *data
)
2231 struct lttng_poll_event events
;
2232 struct lttng_ht
*viewer_connections_ht
;
2233 struct lttng_ht_iter iter
;
2234 struct lttng_viewer_cmd recv_hdr
;
2235 struct relay_connection
*destroy_conn
;
2237 DBG("[thread] Live viewer relay worker started");
2239 rcu_register_thread();
2241 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_WORKER
);
2243 if (testpoint(relayd_thread_live_worker
)) {
2244 goto error_testpoint
;
2247 /* table of connections indexed on socket */
2248 viewer_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2249 if (!viewer_connections_ht
) {
2250 goto viewer_connections_ht_error
;
2253 ret
= create_named_thread_poll_set(&events
, 2,
2254 "Live viewer worker thread epoll");
2256 goto error_poll_create
;
2259 ret
= lttng_poll_add(&events
, live_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2268 health_code_update();
2270 /* Infinite blocking call, waiting for transmission */
2271 DBG3("Relayd live viewer worker thread polling...");
2272 health_poll_entry();
2273 ret
= lttng_poll_wait(&events
, -1);
2277 * Restart interrupted system call.
2279 if (errno
== EINTR
) {
2288 * Process control. The control connection is prioritised so we don't
2289 * starve it with high throughput tracing data on the data
2292 for (i
= 0; i
< nb_fd
; i
++) {
2293 /* Fetch once the poll data */
2294 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2295 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2297 health_code_update();
2299 /* Thread quit pipe has been closed. Killing thread. */
2300 ret
= check_thread_quit_pipe(pollfd
, revents
);
2306 /* Inspect the relay conn pipe for new connection. */
2307 if (pollfd
== live_conn_pipe
[0]) {
2308 if (revents
& LPOLLIN
) {
2309 struct relay_connection
*conn
;
2311 ret
= lttng_read(live_conn_pipe
[0],
2312 &conn
, sizeof(conn
));
2316 ret
= lttng_poll_add(&events
,
2318 LPOLLIN
| LPOLLRDHUP
);
2320 ERR("Failed to add new live connection file descriptor to poll set");
2323 connection_ht_add(viewer_connections_ht
, conn
);
2324 DBG("Connection socket %d added to poll", conn
->sock
->fd
);
2325 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2326 ERR("Relay live pipe error");
2329 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2333 /* Connection activity. */
2334 struct relay_connection
*conn
;
2336 conn
= connection_get_by_sock(viewer_connections_ht
, pollfd
);
2341 if (revents
& LPOLLIN
) {
2342 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &recv_hdr
,
2343 sizeof(recv_hdr
), 0);
2345 /* Connection closed. */
2346 cleanup_connection_pollfd(&events
, pollfd
);
2347 /* Put "create" ownership reference. */
2348 connection_put(conn
);
2349 DBG("Viewer control conn closed with %d", pollfd
);
2351 ret
= process_control(&recv_hdr
, conn
);
2353 /* Clear the session on error. */
2354 cleanup_connection_pollfd(&events
, pollfd
);
2355 /* Put "create" ownership reference. */
2356 connection_put(conn
);
2357 DBG("Viewer connection closed with %d", pollfd
);
2360 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2361 cleanup_connection_pollfd(&events
, pollfd
);
2362 /* Put "create" ownership reference. */
2363 connection_put(conn
);
2365 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2366 connection_put(conn
);
2369 /* Put local "get_by_sock" reference. */
2370 connection_put(conn
);
2377 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
2379 /* Cleanup remaining connection object. */
2381 cds_lfht_for_each_entry(viewer_connections_ht
->ht
, &iter
.iter
,
2384 health_code_update();
2385 connection_put(destroy_conn
);
2389 lttng_ht_destroy(viewer_connections_ht
);
2390 viewer_connections_ht_error
:
2391 /* Close relay conn pipes */
2392 (void) fd_tracker_util_pipe_close(the_fd_tracker
, live_conn_pipe
);
2394 DBG("Viewer worker thread exited with error");
2396 DBG("Viewer worker thread cleanup complete");
2400 ERR("Health error occurred in %s", __func__
);
2402 health_unregister(health_relayd
);
2403 if (lttng_relay_stop_threads()) {
2404 ERR("Error stopping threads");
2406 rcu_unregister_thread();
2411 * Create the relay command pipe to wake thread_manage_apps.
2412 * Closed in cleanup().
2414 static int create_conn_pipe(void)
2416 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
2417 "Live connection pipe", live_conn_pipe
);
2420 int relayd_live_join(void)
2422 int ret
, retval
= 0;
2425 ret
= pthread_join(live_listener_thread
, &status
);
2428 PERROR("pthread_join live listener");
2432 ret
= pthread_join(live_worker_thread
, &status
);
2435 PERROR("pthread_join live worker");
2439 ret
= pthread_join(live_dispatcher_thread
, &status
);
2442 PERROR("pthread_join live dispatcher");
2446 cleanup_relayd_live();
2454 int relayd_live_create(struct lttng_uri
*uri
)
2456 int ret
= 0, retval
= 0;
2462 goto exit_init_data
;
2466 /* Check if daemon is UID = 0 */
2467 is_root
= !getuid();
2470 if (live_uri
->port
< 1024) {
2471 ERR("Need to be root to use ports < 1024");
2473 goto exit_init_data
;
2477 /* Setup the thread apps communication pipe. */
2478 if (create_conn_pipe()) {
2480 goto exit_init_data
;
2483 /* Init relay command queue. */
2484 cds_wfcq_init(&viewer_conn_queue
.head
, &viewer_conn_queue
.tail
);
2486 /* Set up max poll set size */
2487 if (lttng_poll_set_max_size()) {
2489 goto exit_init_data
;
2492 /* Setup the dispatcher thread */
2493 ret
= pthread_create(&live_dispatcher_thread
, default_pthread_attr(),
2494 thread_dispatcher
, (void *) NULL
);
2497 PERROR("pthread_create viewer dispatcher");
2499 goto exit_dispatcher_thread
;
2502 /* Setup the worker thread */
2503 ret
= pthread_create(&live_worker_thread
, default_pthread_attr(),
2504 thread_worker
, NULL
);
2507 PERROR("pthread_create viewer worker");
2509 goto exit_worker_thread
;
2512 /* Setup the listener thread */
2513 ret
= pthread_create(&live_listener_thread
, default_pthread_attr(),
2514 thread_listener
, (void *) NULL
);
2517 PERROR("pthread_create viewer listener");
2519 goto exit_listener_thread
;
2523 * All OK, started all threads.
2528 * Join on the live_listener_thread should anything be added after
2529 * the live_listener thread's creation.
2532 exit_listener_thread
:
2534 ret
= pthread_join(live_worker_thread
, &status
);
2537 PERROR("pthread_join live worker");
2542 ret
= pthread_join(live_dispatcher_thread
, &status
);
2545 PERROR("pthread_join live dispatcher");
2548 exit_dispatcher_thread
:
2551 cleanup_relayd_live();