2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
33 #include <sys/types.h>
36 #include <urcu/futex.h>
37 #include <urcu/uatomic.h>
42 #include <lttng/lttng.h>
43 #include <common/common.h>
44 #include <common/compat/poll.h>
45 #include <common/compat/socket.h>
46 #include <common/defaults.h>
47 #include <common/futex.h>
48 #include <common/sessiond-comm/sessiond-comm.h>
49 #include <common/sessiond-comm/inet.h>
50 #include <common/sessiond-comm/relayd.h>
51 #include <common/uri.h>
52 #include <common/utils.h>
55 #include "ctf-trace.h"
58 #include "lttng-relayd.h"
61 /* command line options */
62 char *opt_output_path
;
63 static int opt_daemon
;
64 static struct lttng_uri
*control_uri
;
65 static struct lttng_uri
*data_uri
;
66 static struct lttng_uri
*live_uri
;
71 * Quit pipe for all threads. This permits a single cancellation point
72 * for all threads when receiving an event on the pipe.
74 static int thread_quit_pipe
[2] = { -1, -1 };
77 * This pipe is used to inform the worker thread that a command is queued and
78 * ready to be processed.
80 static int relay_cmd_pipe
[2] = { -1, -1 };
82 /* Shared between threads */
83 static int dispatch_thread_exit
;
85 static pthread_t listener_thread
;
86 static pthread_t dispatcher_thread
;
87 static pthread_t worker_thread
;
89 static uint64_t last_relay_stream_id
;
90 static uint64_t last_relay_session_id
;
93 * Relay command queue.
95 * The relay_thread_listener and relay_thread_dispatcher communicate with this
98 static struct relay_cmd_queue relay_cmd_queue
;
100 /* buffer allocated at startup, used to store the trace data */
101 static char *data_buffer
;
102 static unsigned int data_buffer_size
;
104 /* Global hash table that stores relay index object. */
105 static struct lttng_ht
*indexes_ht
;
107 /* We need those values for the file/dir creation. */
108 static uid_t relayd_uid
;
109 static gid_t relayd_gid
;
111 /* Global relay stream hash table. */
112 struct lttng_ht
*relay_streams_ht
;
115 * usage function on stderr
120 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
121 fprintf(stderr
, " -h, --help Display this usage.\n");
122 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
123 fprintf(stderr
, " -C, --control-port URL Control port listening.\n");
124 fprintf(stderr
, " -D, --data-port URL Data port listening.\n");
125 fprintf(stderr
, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
126 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
130 int parse_args(int argc
, char **argv
)
134 char *default_address
;
136 static struct option long_options
[] = {
137 { "control-port", 1, 0, 'C', },
138 { "data-port", 1, 0, 'D', },
139 { "daemonize", 0, 0, 'd', },
140 { "help", 0, 0, 'h', },
141 { "output", 1, 0, 'o', },
142 { "verbose", 0, 0, 'v', },
147 int option_index
= 0;
148 c
= getopt_long(argc
, argv
, "dhv" "C:D:o:",
149 long_options
, &option_index
);
156 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
158 fprintf(stderr
, " with arg %s\n", optarg
);
162 ret
= uri_parse(optarg
, &control_uri
);
164 ERR("Invalid control URI specified");
167 if (control_uri
->port
== 0) {
168 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
172 ret
= uri_parse(optarg
, &data_uri
);
174 ERR("Invalid data URI specified");
177 if (data_uri
->port
== 0) {
178 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
188 ret
= asprintf(&opt_output_path
, "%s", optarg
);
190 PERROR("asprintf opt_output_path");
195 /* Verbose level can increase using multiple -v */
196 lttng_opt_verbose
+= 1;
199 /* Unknown option or other error.
200 * Error is printed by getopt, just return */
206 /* assign default values */
207 if (control_uri
== NULL
) {
208 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
209 DEFAULT_NETWORK_CONTROL_PORT
);
211 PERROR("asprintf default data address");
215 ret
= uri_parse(default_address
, &control_uri
);
216 free(default_address
);
218 ERR("Invalid control URI specified");
222 if (data_uri
== NULL
) {
223 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
224 DEFAULT_NETWORK_DATA_PORT
);
226 PERROR("asprintf default data address");
230 ret
= uri_parse(default_address
, &data_uri
);
231 free(default_address
);
233 ERR("Invalid data URI specified");
237 if (live_uri
== NULL
) {
238 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
239 DEFAULT_NETWORK_VIEWER_PORT
);
241 PERROR("asprintf default viewer control address");
245 ret
= uri_parse(default_address
, &live_uri
);
246 free(default_address
);
248 ERR("Invalid viewer control URI specified");
265 /* free the dynamically allocated opt_output_path */
266 free(opt_output_path
);
268 /* Close thread quit pipes */
269 utils_close_pipe(thread_quit_pipe
);
271 uri_free(control_uri
);
276 * Write to writable pipe used to notify a thread.
279 int notify_thread_pipe(int wpipe
)
284 ret
= write(wpipe
, "!", 1);
285 } while (ret
< 0 && errno
== EINTR
);
286 if (ret
< 0 || ret
!= 1) {
287 PERROR("write poll pipe");
294 * Stop all threads by closing the thread quit pipe.
297 void stop_threads(void)
301 /* Stopping all threads */
302 DBG("Terminating all threads");
303 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
305 ERR("write error on thread quit pipe");
308 /* Dispatch thread */
309 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
310 futex_nto1_wake(&relay_cmd_queue
.futex
);
314 * Signal handler for the daemon
316 * Simply stop all worker threads, leaving main() return gracefully after
317 * joining all threads and calling cleanup().
320 void sighandler(int sig
)
324 DBG("SIGPIPE caught");
327 DBG("SIGINT caught");
331 DBG("SIGTERM caught");
340 * Setup signal handler for :
341 * SIGINT, SIGTERM, SIGPIPE
344 int set_signal_handler(void)
350 if ((ret
= sigemptyset(&sigset
)) < 0) {
351 PERROR("sigemptyset");
355 sa
.sa_handler
= sighandler
;
358 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
363 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
368 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
373 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
379 * Init thread quit pipe.
381 * Return -1 on error or 0 if all pipes are created.
384 int init_thread_quit_pipe(void)
388 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
394 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
397 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
401 if (events
== NULL
|| size
== 0) {
406 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
412 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
424 * Check if the thread quit pipe was triggered.
426 * Return 1 if it was triggered else 0;
429 int check_thread_quit_pipe(int fd
, uint32_t events
)
431 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
439 * Create and init socket from uri.
442 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
445 struct lttcomm_sock
*sock
= NULL
;
447 sock
= lttcomm_alloc_sock_from_uri(uri
);
449 ERR("Allocating socket");
453 ret
= lttcomm_create_sock(sock
);
457 DBG("Listening on sock %d", sock
->fd
);
459 ret
= sock
->ops
->bind(sock
);
464 ret
= sock
->ops
->listen(sock
, -1);
474 lttcomm_destroy_sock(sock
);
480 * Return nonzero if stream needs to be closed.
483 int close_stream_check(struct relay_stream
*stream
)
486 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
488 * We are about to close the stream so set the data pending flag to 1
489 * which will make the end data pending command skip the stream which
490 * is now closed and ready. Note that after proceeding to a file close,
491 * the written file is ready for reading.
493 stream
->data_pending_check_done
= 1;
500 * This thread manages the listening for new connections on the network
503 void *relay_thread_listener(void *data
)
505 int i
, ret
, pollfd
, err
= -1;
507 uint32_t revents
, nb_fd
;
508 struct lttng_poll_event events
;
509 struct lttcomm_sock
*control_sock
, *data_sock
;
511 DBG("[thread] Relay listener started");
513 control_sock
= relay_init_sock(control_uri
);
515 goto error_sock_control
;
518 data_sock
= relay_init_sock(data_uri
);
520 goto error_sock_relay
;
524 * Pass 3 as size here for the thread quit pipe, control and data socket.
526 ret
= create_thread_poll_set(&events
, 3);
528 goto error_create_poll
;
531 /* Add the control socket */
532 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
537 /* Add the data socket */
538 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
544 DBG("Listener accepting connections");
547 ret
= lttng_poll_wait(&events
, -1);
550 * Restart interrupted system call.
552 if (errno
== EINTR
) {
560 DBG("Relay new connection received");
561 for (i
= 0; i
< nb_fd
; i
++) {
562 /* Fetch once the poll data */
563 revents
= LTTNG_POLL_GETEV(&events
, i
);
564 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
566 /* Thread quit pipe has been closed. Killing thread. */
567 ret
= check_thread_quit_pipe(pollfd
, revents
);
573 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
574 ERR("socket poll error");
576 } else if (revents
& LPOLLIN
) {
578 * Get allocated in this thread,
579 * enqueued to a global queue, dequeued
580 * and freed in the worker thread.
582 struct relay_command
*relay_cmd
;
583 struct lttcomm_sock
*newsock
;
585 relay_cmd
= zmalloc(sizeof(struct relay_command
));
586 if (relay_cmd
== NULL
) {
587 PERROR("relay command zmalloc");
591 if (pollfd
== data_sock
->fd
) {
592 newsock
= data_sock
->ops
->accept(data_sock
);
594 PERROR("accepting data sock");
598 relay_cmd
->type
= RELAY_DATA
;
599 DBG("Relay data connection accepted, socket %d", newsock
->fd
);
601 assert(pollfd
== control_sock
->fd
);
602 newsock
= control_sock
->ops
->accept(control_sock
);
604 PERROR("accepting control sock");
608 relay_cmd
->type
= RELAY_CONTROL
;
609 DBG("Relay control connection accepted, socket %d", newsock
->fd
);
611 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
,
614 PERROR("setsockopt inet");
615 lttcomm_destroy_sock(newsock
);
619 relay_cmd
->sock
= newsock
;
621 * Lock free enqueue the request.
623 cds_wfq_enqueue(&relay_cmd_queue
.queue
, &relay_cmd
->node
);
626 * Wake the dispatch queue futex. Implicit memory
627 * barrier with the exchange in cds_wfq_enqueue.
629 futex_nto1_wake(&relay_cmd_queue
.futex
);
637 lttng_poll_clean(&events
);
639 if (data_sock
->fd
>= 0) {
640 ret
= data_sock
->ops
->close(data_sock
);
645 lttcomm_destroy_sock(data_sock
);
647 if (control_sock
->fd
>= 0) {
648 ret
= control_sock
->ops
->close(control_sock
);
653 lttcomm_destroy_sock(control_sock
);
656 DBG("Thread exited with error");
658 DBG("Relay listener thread cleanup complete");
664 * This thread manages the dispatching of the requests to worker threads
667 void *relay_thread_dispatcher(void *data
)
670 struct cds_wfq_node
*node
;
671 struct relay_command
*relay_cmd
= NULL
;
673 DBG("[thread] Relay dispatcher started");
675 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
676 /* Atomically prepare the queue futex */
677 futex_nto1_prepare(&relay_cmd_queue
.futex
);
680 /* Dequeue commands */
681 node
= cds_wfq_dequeue_blocking(&relay_cmd_queue
.queue
);
683 DBG("Woken up but nothing in the relay command queue");
684 /* Continue thread execution */
688 relay_cmd
= caa_container_of(node
, struct relay_command
, node
);
689 DBG("Dispatching request waiting on sock %d", relay_cmd
->sock
->fd
);
692 * Inform worker thread of the new request. This
693 * call is blocking so we can be assured that the data will be read
694 * at some point in time or wait to the end of the world :)
697 ret
= write(relay_cmd_pipe
[1], relay_cmd
,
698 sizeof(struct relay_command
));
699 } while (ret
< 0 && errno
== EINTR
);
701 if (ret
< 0 || ret
!= sizeof(struct relay_command
)) {
702 PERROR("write cmd pipe");
705 } while (node
!= NULL
);
707 /* Futex wait on queue. Blocking call on futex() */
708 futex_nto1_wait(&relay_cmd_queue
.futex
);
712 DBG("Dispatch thread dying");
718 * Get stream from stream id.
719 * Need to be called with RCU read-side lock held.
721 struct relay_stream
*relay_stream_find_by_id(uint64_t stream_id
)
723 struct lttng_ht_node_ulong
*node
;
724 struct lttng_ht_iter iter
;
725 struct relay_stream
*ret
;
727 lttng_ht_lookup(relay_streams_ht
,
728 (void *)((unsigned long) stream_id
),
730 node
= lttng_ht_iter_get_node_ulong(&iter
);
732 DBG("Relay stream %" PRIu64
" not found", stream_id
);
737 ret
= caa_container_of(node
, struct relay_stream
, stream_n
);
744 void deferred_free_stream(struct rcu_head
*head
)
746 struct relay_stream
*stream
=
747 caa_container_of(head
, struct relay_stream
, rcu_node
);
749 ctf_trace_try_destroy(stream
->ctf_trace
);
751 free(stream
->path_name
);
752 free(stream
->channel_name
);
757 void deferred_free_session(struct rcu_head
*head
)
759 struct relay_session
*session
=
760 caa_container_of(head
, struct relay_session
, rcu_node
);
765 * relay_delete_session: Free all memory associated with a session and
769 void relay_delete_session(struct relay_command
*cmd
,
770 struct lttng_ht
*sessions_ht
)
772 struct lttng_ht_iter iter
;
773 struct lttng_ht_node_ulong
*node
;
774 struct relay_stream
*stream
;
781 DBG("Relay deleting session %" PRIu64
, cmd
->session
->id
);
784 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, node
, node
) {
785 node
= lttng_ht_iter_get_node_ulong(&iter
);
787 stream
= caa_container_of(node
,
788 struct relay_stream
, stream_n
);
789 if (stream
->session
== cmd
->session
) {
790 ret
= close(stream
->fd
);
792 PERROR("close stream fd on delete session");
794 ret
= lttng_ht_del(relay_streams_ht
, &iter
);
796 call_rcu(&stream
->rcu_node
,
797 deferred_free_stream
);
799 /* Cleanup index of that stream. */
800 relay_index_destroy_by_stream_id(stream
->stream_handle
,
804 iter
.iter
.node
= &cmd
->session
->session_n
.node
;
805 ret
= lttng_ht_del(sessions_ht
, &iter
);
807 call_rcu(&cmd
->session
->rcu_node
,
808 deferred_free_session
);
813 * Copy index data from the control port to a given index object.
815 static void copy_index_control_data(struct relay_index
*index
,
816 struct lttcomm_relayd_index
*data
)
822 * The index on disk is encoded in big endian, so we don't need to convert
823 * the data received on the network. The data_offset value is NEVER
824 * modified here and is updated by the data thread.
826 index
->index_data
.packet_size
= data
->packet_size
;
827 index
->index_data
.content_size
= data
->content_size
;
828 index
->index_data
.timestamp_begin
= data
->timestamp_begin
;
829 index
->index_data
.timestamp_end
= data
->timestamp_end
;
830 index
->index_data
.events_discarded
= data
->events_discarded
;
831 index
->index_data
.stream_id
= data
->stream_id
;
835 * Handle the RELAYD_CREATE_SESSION command.
837 * On success, send back the session id or else return a negative value.
840 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
841 struct relay_command
*cmd
,
842 struct lttng_ht
*sessions_ht
)
844 int ret
= 0, send_ret
;
845 struct relay_session
*session
;
846 struct lttcomm_relayd_status_session reply
;
851 memset(&reply
, 0, sizeof(reply
));
853 session
= zmalloc(sizeof(struct relay_session
));
854 if (session
== NULL
) {
855 PERROR("relay session zmalloc");
860 session
->id
= ++last_relay_session_id
;
861 session
->sock
= cmd
->sock
;
862 cmd
->session
= session
;
864 reply
.session_id
= htobe64(session
->id
);
866 switch (cmd
->minor
) {
867 case 4: /* LTTng sessiond 2.4 */
869 ret
= cmd_create_session_2_4(cmd
, session
);
873 lttng_ht_node_init_ulong(&session
->session_n
,
874 (unsigned long) session
->id
);
875 lttng_ht_add_unique_ulong(sessions_ht
,
876 &session
->session_n
);
878 DBG("Created session %" PRIu64
, session
->id
);
882 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
884 reply
.ret_code
= htobe32(LTTNG_OK
);
887 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
889 ERR("Relayd sending session id");
897 * relay_add_stream: allocate a new stream for a session
900 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
901 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
903 struct relay_session
*session
= cmd
->session
;
904 struct relay_stream
*stream
= NULL
;
905 struct lttcomm_relayd_status_stream reply
;
908 if (!session
|| cmd
->version_check_done
== 0) {
909 ERR("Trying to add a stream before version check");
914 stream
= zmalloc(sizeof(struct relay_stream
));
915 if (stream
== NULL
) {
916 PERROR("relay stream zmalloc");
921 switch (cmd
->minor
) {
922 case 1: /* LTTng sessiond 2.1 */
923 ret
= cmd_recv_stream_2_1(cmd
, stream
);
925 case 2: /* LTTng sessiond 2.2 */
927 ret
= cmd_recv_stream_2_2(cmd
, stream
);
931 goto err_free_stream
;
935 stream
->stream_handle
= ++last_relay_stream_id
;
936 stream
->prev_seq
= -1ULL;
937 stream
->session
= session
;
938 stream
->index_fd
= -1;
939 stream
->read_index_fd
= -1;
940 stream
->ctf_trace
= NULL
;
941 pthread_mutex_init(&stream
->lock
, NULL
);
943 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
);
945 ERR("relay creating output directory");
950 * No need to use run_as API here because whatever we receives, the relayd
951 * uses its own credentials for the stream files.
953 ret
= utils_create_stream_file(stream
->path_name
, stream
->channel_name
,
954 stream
->tracefile_size
, 0, relayd_uid
, relayd_gid
, NULL
);
956 ERR("Create output file");
960 if (stream
->tracefile_size
) {
961 DBG("Tracefile %s/%s_0 created", stream
->path_name
, stream
->channel_name
);
963 DBG("Tracefile %s/%s created", stream
->path_name
, stream
->channel_name
);
966 if (!strncmp(stream
->channel_name
, DEFAULT_METADATA_NAME
, NAME_MAX
)) {
967 stream
->metadata_flag
= 1;
969 * When we receive a new metadata stream, we create a new
970 * ctf_trace and we assign this ctf_trace to all streams with
973 * If later on we receive a new stream for the same ctf_trace,
974 * we copy the information from the first hit in the HT to the
977 stream
->ctf_trace
= ctf_trace_create();
978 if (!stream
->ctf_trace
) {
982 stream
->ctf_trace
->refcount
++;
983 stream
->ctf_trace
->metadata_stream
= stream
;
985 ctf_trace_assign(cmd
->ctf_traces_ht
, stream
);
987 lttng_ht_node_init_ulong(&stream
->stream_n
,
988 (unsigned long) stream
->stream_handle
);
989 lttng_ht_add_unique_ulong(relay_streams_ht
,
992 lttng_ht_node_init_str(&stream
->ctf_trace_node
, stream
->path_name
);
993 lttng_ht_add_str(cmd
->ctf_traces_ht
, &stream
->ctf_trace_node
);
995 DBG("Relay new stream added %s with ID %" PRIu64
, stream
->channel_name
,
996 stream
->stream_handle
);
999 reply
.handle
= htobe64(stream
->stream_handle
);
1000 /* send the session id to the client or a negative return code on error */
1002 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1003 /* stream was not properly added to the ht, so free it */
1006 reply
.ret_code
= htobe32(LTTNG_OK
);
1009 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1010 sizeof(struct lttcomm_relayd_status_stream
), 0);
1012 ERR("Relay sending stream id");
1021 free(stream
->path_name
);
1022 free(stream
->channel_name
);
1028 * relay_close_stream: close a specific stream
1031 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1032 struct relay_command
*cmd
, struct lttng_ht
*viewer_streams_ht
)
1034 struct relay_session
*session
= cmd
->session
;
1035 struct lttcomm_relayd_close_stream stream_info
;
1036 struct lttcomm_relayd_generic_reply reply
;
1037 struct relay_stream
*stream
;
1039 struct lttng_ht_iter iter
;
1041 DBG("Close stream received");
1043 if (!session
|| cmd
->version_check_done
== 0) {
1044 ERR("Trying to close a stream before version check");
1046 goto end_no_session
;
1049 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &stream_info
,
1050 sizeof(struct lttcomm_relayd_close_stream
), 0);
1051 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1053 /* Orderly shutdown. Not necessary to print an error. */
1054 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1056 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1059 goto end_no_session
;
1063 stream
= relay_stream_find_by_id(be64toh(stream_info
.stream_id
));
1069 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1070 stream
->close_flag
= 1;
1072 if (close_stream_check(stream
)) {
1074 struct relay_viewer_stream
*vstream
;
1076 delret
= close(stream
->fd
);
1078 PERROR("close stream");
1081 if (stream
->index_fd
>= 0) {
1082 delret
= close(stream
->index_fd
);
1084 PERROR("close stream index_fd");
1088 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
,
1092 * Set the last good value into the viewer stream. This is done
1093 * right before the stream gets deleted from the hash table. The
1094 * lookup failure on the live thread side of a stream indicates
1095 * that the viewer stream index received value should be used.
1097 vstream
->total_index_received
= stream
->total_index_received
;
1100 iter
.iter
.node
= &stream
->stream_n
.node
;
1101 delret
= lttng_ht_del(relay_streams_ht
, &iter
);
1103 iter
.iter
.node
= &stream
->ctf_trace_node
.node
;
1104 delret
= lttng_ht_del(cmd
->ctf_traces_ht
, &iter
);
1106 call_rcu(&stream
->rcu_node
,
1107 deferred_free_stream
);
1108 DBG("Closed tracefile %d from close stream", stream
->fd
);
1115 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1117 reply
.ret_code
= htobe32(LTTNG_OK
);
1119 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1120 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1122 ERR("Relay sending stream id");
1131 * relay_unknown_command: send -1 if received unknown command
1134 void relay_unknown_command(struct relay_command
*cmd
)
1136 struct lttcomm_relayd_generic_reply reply
;
1139 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1140 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1141 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1143 ERR("Relay sending unknown command");
1148 * relay_start: send an acknowledgment to the client to tell if we are
1149 * ready to receive data. We are ready if a session is established.
1152 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1153 struct relay_command
*cmd
)
1155 int ret
= htobe32(LTTNG_OK
);
1156 struct lttcomm_relayd_generic_reply reply
;
1157 struct relay_session
*session
= cmd
->session
;
1160 DBG("Trying to start the streaming without a session established");
1161 ret
= htobe32(LTTNG_ERR_UNK
);
1164 reply
.ret_code
= ret
;
1165 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1166 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1168 ERR("Relay sending start ack");
1175 * Append padding to the file pointed by the file descriptor fd.
1177 static int write_padding_to_file(int fd
, uint32_t size
)
1186 zeros
= zmalloc(size
);
1187 if (zeros
== NULL
) {
1188 PERROR("zmalloc zeros for padding");
1194 ret
= write(fd
, zeros
, size
);
1195 } while (ret
< 0 && errno
== EINTR
);
1196 if (ret
< 0 || ret
!= size
) {
1197 PERROR("write padding to file");
1207 * relay_recv_metadata: receive the metada for the session.
1210 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1211 struct relay_command
*cmd
)
1213 int ret
= htobe32(LTTNG_OK
);
1214 struct relay_session
*session
= cmd
->session
;
1215 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1216 struct relay_stream
*metadata_stream
;
1217 uint64_t data_size
, payload_size
;
1220 ERR("Metadata sent before version check");
1225 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1226 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1227 ERR("Incorrect data size");
1231 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1233 if (data_buffer_size
< data_size
) {
1234 /* In case the realloc fails, we can free the memory */
1237 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1238 if (!tmp_data_ptr
) {
1239 ERR("Allocating data buffer");
1244 data_buffer
= tmp_data_ptr
;
1245 data_buffer_size
= data_size
;
1247 memset(data_buffer
, 0, data_size
);
1248 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1249 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
1250 if (ret
< 0 || ret
!= data_size
) {
1252 /* Orderly shutdown. Not necessary to print an error. */
1253 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1255 ERR("Relay didn't receive the whole metadata");
1260 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1263 metadata_stream
= relay_stream_find_by_id(
1264 be64toh(metadata_struct
->stream_id
));
1265 if (!metadata_stream
) {
1271 ret
= write(metadata_stream
->fd
, metadata_struct
->payload
,
1273 } while (ret
< 0 && errno
== EINTR
);
1274 if (ret
< 0 || ret
!= payload_size
) {
1275 ERR("Relay error writing metadata on file");
1280 ret
= write_padding_to_file(metadata_stream
->fd
,
1281 be32toh(metadata_struct
->padding_size
));
1285 metadata_stream
->ctf_trace
->metadata_received
+=
1286 payload_size
+ be32toh(metadata_struct
->padding_size
);
1288 DBG2("Relay metadata written");
1297 * relay_send_version: send relayd version number
1300 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1301 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1304 struct lttcomm_relayd_version reply
, msg
;
1308 cmd
->version_check_done
= 1;
1310 /* Get version from the other side. */
1311 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1312 if (ret
< 0 || ret
!= sizeof(msg
)) {
1314 /* Orderly shutdown. Not necessary to print an error. */
1315 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1317 ERR("Relay failed to receive the version values.");
1323 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1324 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1326 /* Major versions must be the same */
1327 if (reply
.major
!= be32toh(msg
.major
)) {
1328 DBG("Incompatible major versions (%u vs %u), deleting session",
1329 reply
.major
, be32toh(msg
.major
));
1330 relay_delete_session(cmd
, sessions_ht
);
1335 cmd
->major
= reply
.major
;
1336 /* We adapt to the lowest compatible version */
1337 if (reply
.minor
<= be32toh(msg
.minor
)) {
1338 cmd
->minor
= reply
.minor
;
1340 cmd
->minor
= be32toh(msg
.minor
);
1343 reply
.major
= htobe32(reply
.major
);
1344 reply
.minor
= htobe32(reply
.minor
);
1345 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1346 sizeof(struct lttcomm_relayd_version
), 0);
1348 ERR("Relay sending version");
1351 DBG("Version check done using protocol %u.%u", cmd
->major
,
1359 * Check for data pending for a given stream id from the session daemon.
1362 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1363 struct relay_command
*cmd
)
1365 struct relay_session
*session
= cmd
->session
;
1366 struct lttcomm_relayd_data_pending msg
;
1367 struct lttcomm_relayd_generic_reply reply
;
1368 struct relay_stream
*stream
;
1370 uint64_t last_net_seq_num
, stream_id
;
1372 DBG("Data pending command received");
1374 if (!session
|| cmd
->version_check_done
== 0) {
1375 ERR("Trying to check for data before version check");
1377 goto end_no_session
;
1380 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1381 if (ret
< sizeof(msg
)) {
1383 /* Orderly shutdown. Not necessary to print an error. */
1384 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1386 ERR("Relay didn't receive valid data_pending struct size : %d",
1390 goto end_no_session
;
1393 stream_id
= be64toh(msg
.stream_id
);
1394 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1397 stream
= relay_stream_find_by_id(stream_id
);
1398 if (stream
== NULL
) {
1403 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1404 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1407 /* Avoid wrapping issue */
1408 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1409 /* Data has in fact been written and is NOT pending */
1412 /* Data still being streamed thus pending */
1416 /* Pending check is now done. */
1417 stream
->data_pending_check_done
= 1;
1422 reply
.ret_code
= htobe32(ret
);
1423 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1425 ERR("Relay data pending ret code failed");
1433 * Wait for the control socket to reach a quiescent state.
1435 * Note that for now, when receiving this command from the session daemon, this
1436 * means that every subsequent commands or data received on the control socket
1437 * has been handled. So, this is why we simply return OK here.
1440 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1441 struct relay_command
*cmd
)
1445 struct relay_stream
*stream
;
1446 struct lttng_ht_iter iter
;
1447 struct lttcomm_relayd_quiescent_control msg
;
1448 struct lttcomm_relayd_generic_reply reply
;
1450 DBG("Checking quiescent state on control socket");
1452 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1453 ERR("Trying to check for data before version check");
1455 goto end_no_session
;
1458 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1459 if (ret
< sizeof(msg
)) {
1461 /* Orderly shutdown. Not necessary to print an error. */
1462 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1464 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1468 goto end_no_session
;
1471 stream_id
= be64toh(msg
.stream_id
);
1474 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1476 if (stream
->stream_handle
== stream_id
) {
1477 stream
->data_pending_check_done
= 1;
1478 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1485 reply
.ret_code
= htobe32(LTTNG_OK
);
1486 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1488 ERR("Relay data quiescent control ret code failed");
1496 * Initialize a data pending command. This means that a client is about to ask
1497 * for data pending for each stream he/she holds. Simply iterate over all
1498 * streams of a session and set the data_pending_check_done flag.
1500 * This command returns to the client a LTTNG_OK code.
1503 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1504 struct relay_command
*cmd
)
1507 struct lttng_ht_iter iter
;
1508 struct lttcomm_relayd_begin_data_pending msg
;
1509 struct lttcomm_relayd_generic_reply reply
;
1510 struct relay_stream
*stream
;
1511 uint64_t session_id
;
1516 DBG("Init streams for data pending");
1518 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1519 ERR("Trying to check for data before version check");
1521 goto end_no_session
;
1524 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1525 if (ret
< sizeof(msg
)) {
1527 /* Orderly shutdown. Not necessary to print an error. */
1528 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1530 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1534 goto end_no_session
;
1537 session_id
= be64toh(msg
.session_id
);
1540 * Iterate over all streams to set the begin data pending flag. For now, the
1541 * streams are indexed by stream handle so we have to iterate over all
1542 * streams to find the one associated with the right session_id.
1545 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1547 if (stream
->session
->id
== session_id
) {
1548 stream
->data_pending_check_done
= 0;
1549 DBG("Set begin data pending flag to stream %" PRIu64
,
1550 stream
->stream_handle
);
1555 /* All good, send back reply. */
1556 reply
.ret_code
= htobe32(LTTNG_OK
);
1558 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1560 ERR("Relay begin data pending send reply failed");
1568 * End data pending command. This will check, for a given session id, if each
1569 * stream associated with it has its data_pending_check_done flag set. If not,
1570 * this means that the client lost track of the stream but the data is still
1571 * being streamed on our side. In this case, we inform the client that data is
1574 * Return to the client if there is data in flight or not with a ret_code.
1577 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1578 struct relay_command
*cmd
)
1581 struct lttng_ht_iter iter
;
1582 struct lttcomm_relayd_end_data_pending msg
;
1583 struct lttcomm_relayd_generic_reply reply
;
1584 struct relay_stream
*stream
;
1585 uint64_t session_id
;
1586 uint32_t is_data_inflight
= 0;
1591 DBG("End data pending command");
1593 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1594 ERR("Trying to check for data before version check");
1596 goto end_no_session
;
1599 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1600 if (ret
< sizeof(msg
)) {
1602 /* Orderly shutdown. Not necessary to print an error. */
1603 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1605 ERR("Relay didn't receive valid end data_pending struct size: %d",
1609 goto end_no_session
;
1612 session_id
= be64toh(msg
.session_id
);
1614 /* Iterate over all streams to see if the begin data pending flag is set. */
1616 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1618 if (stream
->session
->id
== session_id
&&
1619 !stream
->data_pending_check_done
) {
1620 is_data_inflight
= 1;
1621 DBG("Data is still in flight for stream %" PRIu64
,
1622 stream
->stream_handle
);
1628 /* All good, send back reply. */
1629 reply
.ret_code
= htobe32(is_data_inflight
);
1631 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1633 ERR("Relay end data pending send reply failed");
1641 * Receive an index for a specific stream.
1643 * Return 0 on success else a negative value.
1646 int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1647 struct relay_command
*cmd
, struct lttng_ht
*indexes_ht
)
1649 int ret
, send_ret
, index_created
= 0;
1650 struct relay_session
*session
= cmd
->session
;
1651 struct lttcomm_relayd_index index_info
;
1652 struct relay_index
*index
, *wr_index
= NULL
;
1653 struct lttcomm_relayd_generic_reply reply
;
1654 struct relay_stream
*stream
;
1655 uint64_t net_seq_num
;
1660 DBG("Relay receiving index");
1662 if (!session
|| cmd
->version_check_done
== 0) {
1663 ERR("Trying to close a stream before version check");
1665 goto end_no_session
;
1668 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &index_info
,
1669 sizeof(index_info
), 0);
1670 if (ret
< sizeof(index_info
)) {
1672 /* Orderly shutdown. Not necessary to print an error. */
1673 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1675 ERR("Relay didn't receive valid index struct size : %d", ret
);
1678 goto end_no_session
;
1681 net_seq_num
= be64toh(index_info
.net_seq_num
);
1684 stream
= relay_stream_find_by_id(be64toh(index_info
.relay_stream_id
));
1687 goto end_rcu_unlock
;
1690 /* Live beacon handling */
1691 if (index_info
.packet_size
== 0) {
1692 DBG("Received live beacon for stream %" PRIu64
, stream
->stream_handle
);
1695 * Only flag a stream inactive when it has already received data.
1697 if (stream
->total_index_received
> 0) {
1698 stream
->beacon_ts_end
= be64toh(index_info
.timestamp_end
);
1701 goto end_rcu_unlock
;
1703 stream
->beacon_ts_end
= -1ULL;
1706 index
= relay_index_find(stream
->stream_handle
, net_seq_num
, indexes_ht
);
1708 /* A successful creation will add the object to the HT. */
1709 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
1711 goto end_rcu_unlock
;
1716 copy_index_control_data(index
, &index_info
);
1718 if (index_created
) {
1720 * Try to add the relay index object to the hash table. If an object
1721 * already exist, destroy back the index created, set the data in this
1722 * object and write it on disk.
1724 relay_index_add(index
, indexes_ht
, &wr_index
);
1726 copy_index_control_data(wr_index
, &index_info
);
1730 /* The index already exists so write it on disk. */
1734 /* Do we have a writable ready index to write on disk. */
1736 /* Starting at 2.4, create the index file if none available. */
1737 if (cmd
->minor
>= 4 && stream
->index_fd
< 0) {
1738 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
1739 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
1740 stream
->tracefile_count_current
);
1742 goto end_rcu_unlock
;
1744 stream
->index_fd
= ret
;
1747 ret
= relay_index_write(wr_index
->fd
, wr_index
, indexes_ht
);
1749 goto end_rcu_unlock
;
1751 stream
->total_index_received
++;
1758 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1760 reply
.ret_code
= htobe32(LTTNG_OK
);
1762 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1764 ERR("Relay sending close index id reply");
1773 * Process the commands received on the control socket
1776 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1777 struct relay_command
*cmd
, struct relay_local_data
*ctx
)
1781 switch (be32toh(recv_hdr
->cmd
)) {
1782 case RELAYD_CREATE_SESSION
:
1783 ret
= relay_create_session(recv_hdr
, cmd
, ctx
->sessions_ht
);
1785 case RELAYD_ADD_STREAM
:
1786 ret
= relay_add_stream(recv_hdr
, cmd
, ctx
->sessions_ht
);
1788 case RELAYD_START_DATA
:
1789 ret
= relay_start(recv_hdr
, cmd
);
1791 case RELAYD_SEND_METADATA
:
1792 ret
= relay_recv_metadata(recv_hdr
, cmd
);
1794 case RELAYD_VERSION
:
1795 ret
= relay_send_version(recv_hdr
, cmd
, ctx
->sessions_ht
);
1797 case RELAYD_CLOSE_STREAM
:
1798 ret
= relay_close_stream(recv_hdr
, cmd
, ctx
->viewer_streams_ht
);
1800 case RELAYD_DATA_PENDING
:
1801 ret
= relay_data_pending(recv_hdr
, cmd
);
1803 case RELAYD_QUIESCENT_CONTROL
:
1804 ret
= relay_quiescent_control(recv_hdr
, cmd
);
1806 case RELAYD_BEGIN_DATA_PENDING
:
1807 ret
= relay_begin_data_pending(recv_hdr
, cmd
);
1809 case RELAYD_END_DATA_PENDING
:
1810 ret
= relay_end_data_pending(recv_hdr
, cmd
);
1812 case RELAYD_SEND_INDEX
:
1813 ret
= relay_recv_index(recv_hdr
, cmd
, indexes_ht
);
1815 case RELAYD_UPDATE_SYNC_INFO
:
1817 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
1818 relay_unknown_command(cmd
);
1828 * relay_process_data: Process the data received on the data socket
1831 int relay_process_data(struct relay_command
*cmd
,
1832 struct lttng_ht
*indexes_ht
)
1834 int ret
= 0, rotate_index
= 0, index_created
= 0;
1835 struct relay_stream
*stream
;
1836 struct relay_index
*index
, *wr_index
= NULL
;
1837 struct lttcomm_relayd_data_hdr data_hdr
;
1838 uint64_t stream_id
, data_offset
;
1839 uint64_t net_seq_num
;
1842 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &data_hdr
,
1843 sizeof(struct lttcomm_relayd_data_hdr
), 0);
1846 /* Orderly shutdown. Not necessary to print an error. */
1847 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1849 ERR("Unable to receive data header on sock %d", cmd
->sock
->fd
);
1855 stream_id
= be64toh(data_hdr
.stream_id
);
1858 stream
= relay_stream_find_by_id(stream_id
);
1861 goto end_rcu_unlock
;
1864 data_size
= be32toh(data_hdr
.data_size
);
1865 if (data_buffer_size
< data_size
) {
1868 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1869 if (!tmp_data_ptr
) {
1870 ERR("Allocating data buffer");
1873 goto end_rcu_unlock
;
1875 data_buffer
= tmp_data_ptr
;
1876 data_buffer_size
= data_size
;
1878 memset(data_buffer
, 0, data_size
);
1880 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
1882 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
1883 data_size
, stream_id
, net_seq_num
);
1884 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
1887 /* Orderly shutdown. Not necessary to print an error. */
1888 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1891 goto end_rcu_unlock
;
1894 /* Check if a rotation is needed. */
1895 if (stream
->tracefile_size
> 0 &&
1896 (stream
->tracefile_size_current
+ data_size
) >
1897 stream
->tracefile_size
) {
1898 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
1899 stream
->tracefile_size
, stream
->tracefile_count
,
1900 relayd_uid
, relayd_gid
, stream
->fd
,
1901 &(stream
->tracefile_count_current
), &stream
->fd
);
1903 ERR("Rotating stream output file");
1904 goto end_rcu_unlock
;
1906 /* Reset current size because we just perform a stream rotation. */
1907 stream
->tracefile_size_current
= 0;
1911 /* Get data offset because we are about to update the index. */
1912 data_offset
= htobe64(stream
->tracefile_size_current
);
1915 * Lookup for an existing index for that stream id/sequence number. If on
1916 * exists, the control thread already received the data for it thus we need
1917 * to write it on disk.
1919 index
= relay_index_find(stream_id
, net_seq_num
, indexes_ht
);
1921 /* A successful creation will add the object to the HT. */
1922 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
1924 goto end_rcu_unlock
;
1929 if (rotate_index
|| stream
->index_fd
< 0) {
1930 index
->to_close_fd
= stream
->index_fd
;
1931 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
1932 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
1933 stream
->tracefile_count_current
);
1935 /* This will close the stream's index fd if one. */
1936 relay_index_free_safe(index
);
1937 goto end_rcu_unlock
;
1939 stream
->index_fd
= ret
;
1941 index
->fd
= stream
->index_fd
;
1942 index
->index_data
.offset
= data_offset
;
1944 if (index_created
) {
1946 * Try to add the relay index object to the hash table. If an object
1947 * already exist, destroy back the index created and set the data.
1949 relay_index_add(index
, indexes_ht
, &wr_index
);
1951 /* Copy back data from the created index. */
1952 wr_index
->fd
= index
->fd
;
1953 wr_index
->to_close_fd
= index
->to_close_fd
;
1954 wr_index
->index_data
.offset
= data_offset
;
1958 /* The index already exists so write it on disk. */
1962 /* Do we have a writable ready index to write on disk. */
1964 /* Starting at 2.4, create the index file if none available. */
1965 if (cmd
->minor
>= 4 && stream
->index_fd
< 0) {
1966 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
1967 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
1968 stream
->tracefile_count_current
);
1970 goto end_rcu_unlock
;
1972 stream
->index_fd
= ret
;
1975 ret
= relay_index_write(wr_index
->fd
, wr_index
, indexes_ht
);
1977 goto end_rcu_unlock
;
1979 stream
->total_index_received
++;
1983 ret
= write(stream
->fd
, data_buffer
, data_size
);
1984 } while (ret
< 0 && errno
== EINTR
);
1985 if (ret
< 0 || ret
!= data_size
) {
1986 ERR("Relay error writing data to file");
1988 goto end_rcu_unlock
;
1991 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
1992 ret
, stream
->stream_handle
);
1994 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
1996 goto end_rcu_unlock
;
1998 stream
->tracefile_size_current
+= data_size
+ be32toh(data_hdr
.padding_size
);
2000 stream
->prev_seq
= net_seq_num
;
2002 /* Check if we need to close the FD */
2003 if (close_stream_check(stream
)) {
2005 struct lttng_ht_iter iter
;
2007 cret
= close(stream
->fd
);
2009 PERROR("close stream process data");
2012 cret
= close(stream
->index_fd
);
2014 PERROR("close stream index_fd");
2016 iter
.iter
.node
= &stream
->stream_n
.node
;
2017 ret
= lttng_ht_del(relay_streams_ht
, &iter
);
2019 call_rcu(&stream
->rcu_node
,
2020 deferred_free_stream
);
2021 DBG("Closed tracefile %d after recv data", stream
->fd
);
2031 void relay_cleanup_poll_connection(struct lttng_poll_event
*events
, int pollfd
)
2035 lttng_poll_del(events
, pollfd
);
2037 ret
= close(pollfd
);
2039 ERR("Closing pollfd %d", pollfd
);
2044 int relay_add_connection(int fd
, struct lttng_poll_event
*events
,
2045 struct lttng_ht
*relay_connections_ht
)
2047 struct relay_command
*relay_connection
;
2050 relay_connection
= zmalloc(sizeof(struct relay_command
));
2051 if (relay_connection
== NULL
) {
2052 PERROR("Relay command zmalloc");
2056 ret
= read(fd
, relay_connection
, sizeof(struct relay_command
));
2057 } while (ret
< 0 && errno
== EINTR
);
2058 if (ret
< 0 || ret
< sizeof(struct relay_command
)) {
2059 PERROR("read relay cmd pipe");
2063 relay_connection
->ctf_traces_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_STRING
);
2064 if (!relay_connection
->ctf_traces_ht
) {
2068 lttng_ht_node_init_ulong(&relay_connection
->sock_n
,
2069 (unsigned long) relay_connection
->sock
->fd
);
2071 lttng_ht_add_unique_ulong(relay_connections_ht
,
2072 &relay_connection
->sock_n
);
2074 return lttng_poll_add(events
,
2075 relay_connection
->sock
->fd
,
2076 LPOLLIN
| LPOLLRDHUP
);
2079 free(relay_connection
);
2085 void deferred_free_connection(struct rcu_head
*head
)
2087 struct relay_command
*relay_connection
=
2088 caa_container_of(head
, struct relay_command
, rcu_node
);
2090 lttng_ht_destroy(relay_connection
->ctf_traces_ht
);
2091 lttcomm_destroy_sock(relay_connection
->sock
);
2092 free(relay_connection
);
2096 void relay_del_connection(struct lttng_ht
*relay_connections_ht
,
2097 struct lttng_ht_iter
*iter
, struct relay_command
*relay_connection
,
2098 struct lttng_ht
*sessions_ht
)
2102 ret
= lttng_ht_del(relay_connections_ht
, iter
);
2104 if (relay_connection
->type
== RELAY_CONTROL
) {
2105 relay_delete_session(relay_connection
, sessions_ht
);
2108 call_rcu(&relay_connection
->rcu_node
,
2109 deferred_free_connection
);
2113 * This thread does the actual work
2116 void *relay_thread_worker(void *data
)
2118 int ret
, err
= -1, last_seen_data_fd
= -1;
2120 struct relay_command
*relay_connection
;
2121 struct lttng_poll_event events
;
2122 struct lttng_ht
*relay_connections_ht
;
2123 struct lttng_ht_node_ulong
*node
;
2124 struct lttng_ht_iter iter
;
2125 struct lttcomm_relayd_hdr recv_hdr
;
2126 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
2127 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
2129 DBG("[thread] Relay worker started");
2131 rcu_register_thread();
2133 /* table of connections indexed on socket */
2134 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2135 if (!relay_connections_ht
) {
2136 goto relay_connections_ht_error
;
2139 /* Tables of received indexes indexed by index handle and net_seq_num. */
2140 indexes_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64
);
2142 goto indexes_ht_error
;
2145 ret
= create_thread_poll_set(&events
, 2);
2147 goto error_poll_create
;
2150 ret
= lttng_poll_add(&events
, relay_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2157 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2159 /* Infinite blocking call, waiting for transmission */
2160 DBG3("Relayd worker thread polling...");
2161 ret
= lttng_poll_wait(&events
, -1);
2164 * Restart interrupted system call.
2166 if (errno
== EINTR
) {
2175 * Process control. The control connection is prioritised so we don't
2176 * starve it with high throughout put tracing data on the data
2179 for (i
= 0; i
< nb_fd
; i
++) {
2180 /* Fetch once the poll data */
2181 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2182 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2184 /* Thread quit pipe has been closed. Killing thread. */
2185 ret
= check_thread_quit_pipe(pollfd
, revents
);
2191 /* Inspect the relay cmd pipe for new connection */
2192 if (pollfd
== relay_cmd_pipe
[0]) {
2193 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2194 ERR("Relay pipe error");
2196 } else if (revents
& LPOLLIN
) {
2197 DBG("Relay command received");
2198 ret
= relay_add_connection(relay_cmd_pipe
[0],
2199 &events
, relay_connections_ht
);
2204 } else if (revents
) {
2206 lttng_ht_lookup(relay_connections_ht
,
2207 (void *)((unsigned long) pollfd
),
2209 node
= lttng_ht_iter_get_node_ulong(&iter
);
2211 DBG2("Relay sock %d not found", pollfd
);
2215 relay_connection
= caa_container_of(node
,
2216 struct relay_command
, sock_n
);
2218 if (revents
& (LPOLLERR
)) {
2220 relay_cleanup_poll_connection(&events
, pollfd
);
2221 relay_del_connection(relay_connections_ht
,
2222 &iter
, relay_connection
, sessions_ht
);
2223 if (last_seen_data_fd
== pollfd
) {
2224 last_seen_data_fd
= last_notdel_data_fd
;
2226 } else if (revents
& (LPOLLHUP
| LPOLLRDHUP
)) {
2227 DBG("Socket %d hung up", pollfd
);
2228 relay_cleanup_poll_connection(&events
, pollfd
);
2229 relay_del_connection(relay_connections_ht
,
2230 &iter
, relay_connection
, sessions_ht
);
2231 if (last_seen_data_fd
== pollfd
) {
2232 last_seen_data_fd
= last_notdel_data_fd
;
2234 } else if (revents
& LPOLLIN
) {
2235 /* control socket */
2236 if (relay_connection
->type
== RELAY_CONTROL
) {
2237 ret
= relay_connection
->sock
->ops
->recvmsg(
2238 relay_connection
->sock
, &recv_hdr
,
2239 sizeof(struct lttcomm_relayd_hdr
), 0);
2240 /* connection closed */
2242 relay_cleanup_poll_connection(&events
, pollfd
);
2243 relay_del_connection(relay_connections_ht
,
2244 &iter
, relay_connection
, sessions_ht
);
2245 DBG("Control connection closed with %d", pollfd
);
2247 if (relay_connection
->session
) {
2248 DBG2("Relay worker receiving data for session : %" PRIu64
,
2249 relay_connection
->session
->id
);
2251 ret
= relay_process_control(&recv_hdr
,
2252 relay_connection
, relay_ctx
);
2254 /* Clear the session on error. */
2255 relay_cleanup_poll_connection(&events
, pollfd
);
2256 relay_del_connection(relay_connections_ht
,
2257 &iter
, relay_connection
, sessions_ht
);
2258 DBG("Connection closed with %d", pollfd
);
2264 * Flag the last seen data fd not deleted. It will be
2265 * used as the last seen fd if any fd gets deleted in
2268 last_notdel_data_fd
= pollfd
;
2276 * The last loop handled a control request, go back to poll to make
2277 * sure we prioritise the control socket.
2283 if (last_seen_data_fd
>= 0) {
2284 for (i
= 0; i
< nb_fd
; i
++) {
2285 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2286 if (last_seen_data_fd
== pollfd
) {
2293 /* Process data connection. */
2294 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2295 /* Fetch the poll data. */
2296 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2297 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2299 /* Skip the command pipe. It's handled in the first loop. */
2300 if (pollfd
== relay_cmd_pipe
[0]) {
2306 lttng_ht_lookup(relay_connections_ht
,
2307 (void *)((unsigned long) pollfd
),
2309 node
= lttng_ht_iter_get_node_ulong(&iter
);
2311 /* Skip it. Might be removed before. */
2315 relay_connection
= caa_container_of(node
,
2316 struct relay_command
, sock_n
);
2318 if (revents
& LPOLLIN
) {
2319 if (relay_connection
->type
!= RELAY_DATA
) {
2323 ret
= relay_process_data(relay_connection
, indexes_ht
);
2324 /* connection closed */
2326 relay_cleanup_poll_connection(&events
, pollfd
);
2327 relay_del_connection(relay_connections_ht
,
2328 &iter
, relay_connection
, sessions_ht
);
2329 DBG("Data connection closed with %d", pollfd
);
2331 * Every goto restart call sets the last seen fd where
2332 * here we don't really care since we gracefully
2333 * continue the loop after the connection is deleted.
2336 /* Keep last seen port. */
2337 last_seen_data_fd
= pollfd
;
2345 last_seen_data_fd
= -1;
2350 lttng_poll_clean(&events
);
2352 /* empty the hash table and free the memory */
2354 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, node
, node
) {
2355 node
= lttng_ht_iter_get_node_ulong(&iter
);
2357 relay_connection
= caa_container_of(node
,
2358 struct relay_command
, sock_n
);
2359 relay_del_connection(relay_connections_ht
,
2360 &iter
, relay_connection
, sessions_ht
);
2365 lttng_ht_destroy(indexes_ht
);
2367 lttng_ht_destroy(relay_connections_ht
);
2368 relay_connections_ht_error
:
2369 /* Close relay cmd pipes */
2370 utils_close_pipe(relay_cmd_pipe
);
2372 DBG("Thread exited with error");
2374 DBG("Worker thread cleanup complete");
2377 rcu_unregister_thread();
2382 * Create the relay command pipe to wake thread_manage_apps.
2383 * Closed in cleanup().
2385 static int create_relay_cmd_pipe(void)
2389 ret
= utils_create_pipe_cloexec(relay_cmd_pipe
);
2397 int main(int argc
, char **argv
)
2401 struct relay_local_data
*relay_ctx
;
2403 /* Create thread quit pipe */
2404 if ((ret
= init_thread_quit_pipe()) < 0) {
2408 /* Parse arguments */
2410 if ((ret
= parse_args(argc
, argv
)) < 0) {
2414 if ((ret
= set_signal_handler()) < 0) {
2418 /* Try to create directory if -o, --output is specified. */
2419 if (opt_output_path
) {
2420 if (*opt_output_path
!= '/') {
2421 ERR("Please specify an absolute path for -o, --output PATH");
2425 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2427 ERR("Unable to create %s", opt_output_path
);
2441 /* We need those values for the file/dir creation. */
2442 relayd_uid
= getuid();
2443 relayd_gid
= getgid();
2445 /* Check if daemon is UID = 0 */
2446 if (relayd_uid
== 0) {
2447 if (control_uri
->port
< 1024 || data_uri
->port
< 1024) {
2448 ERR("Need to be root to use ports < 1024");
2454 /* Setup the thread apps communication pipe. */
2455 if ((ret
= create_relay_cmd_pipe()) < 0) {
2459 /* Init relay command queue. */
2460 cds_wfq_init(&relay_cmd_queue
.queue
);
2462 /* Set up max poll set size */
2463 lttng_poll_set_max_size();
2465 /* Initialize communication library */
2468 relay_ctx
= zmalloc(sizeof(struct relay_local_data
));
2470 PERROR("relay_ctx");
2474 /* tables of sessions indexed by session ID */
2475 relay_ctx
->sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2476 if (!relay_ctx
->sessions_ht
) {
2477 goto exit_relay_ctx_sessions
;
2480 /* tables of streams indexed by stream ID */
2481 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2482 if (!relay_streams_ht
) {
2483 goto exit_relay_ctx_streams
;
2486 /* tables of streams indexed by stream ID */
2487 relay_ctx
->viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2488 if (!relay_ctx
->viewer_streams_ht
) {
2489 goto exit_relay_ctx_viewer_streams
;
2492 /* Setup the dispatcher thread */
2493 ret
= pthread_create(&dispatcher_thread
, NULL
,
2494 relay_thread_dispatcher
, (void *) NULL
);
2496 PERROR("pthread_create dispatcher");
2497 goto exit_dispatcher
;
2500 /* Setup the worker thread */
2501 ret
= pthread_create(&worker_thread
, NULL
,
2502 relay_thread_worker
, (void *) relay_ctx
);
2504 PERROR("pthread_create worker");
2508 /* Setup the listener thread */
2509 ret
= pthread_create(&listener_thread
, NULL
,
2510 relay_thread_listener
, (void *) NULL
);
2512 PERROR("pthread_create listener");
2516 ret
= live_start_threads(live_uri
, relay_ctx
);
2518 ERR("Starting live viewer threads");
2522 ret
= pthread_join(listener_thread
, &status
);
2524 PERROR("pthread_join");
2525 goto error
; /* join error, exit without cleanup */
2529 ret
= pthread_join(worker_thread
, &status
);
2531 PERROR("pthread_join");
2532 goto error
; /* join error, exit without cleanup */
2536 ret
= pthread_join(dispatcher_thread
, &status
);
2538 PERROR("pthread_join");
2539 goto error
; /* join error, exit without cleanup */
2541 lttng_ht_destroy(relay_ctx
->viewer_streams_ht
);
2543 exit_relay_ctx_viewer_streams
:
2544 lttng_ht_destroy(relay_streams_ht
);
2546 exit_relay_ctx_streams
:
2547 lttng_ht_destroy(relay_ctx
->sessions_ht
);
2549 exit_relay_ctx_sessions
:
2553 live_stop_threads();