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/hashtable/hashtable.h>
51 #include <common/sessiond-comm/relayd.h>
52 #include <common/uri.h>
53 #include <common/utils.h>
55 #include "lttng-relayd.h"
57 /* command line options */
58 static int opt_daemon
;
59 static char *opt_output_path
;
60 static struct lttng_uri
*control_uri
;
61 static struct lttng_uri
*data_uri
;
64 static int is_root
; /* Set to 1 if the daemon is running as root */
67 * Quit pipe for all threads. This permits a single cancellation point
68 * for all threads when receiving an event on the pipe.
70 static int thread_quit_pipe
[2] = { -1, -1 };
73 * This pipe is used to inform the worker thread that a command is queued and
74 * ready to be processed.
76 static int relay_cmd_pipe
[2] = { -1, -1 };
78 /* Shared between threads */
79 static int dispatch_thread_exit
;
81 static pthread_t listener_thread
;
82 static pthread_t dispatcher_thread
;
83 static pthread_t worker_thread
;
85 static uint64_t last_relay_stream_id
;
86 static uint64_t last_relay_session_id
;
89 * Relay command queue.
91 * The relay_thread_listener and relay_thread_dispatcher communicate with this
94 static struct relay_cmd_queue relay_cmd_queue
;
96 /* buffer allocated at startup, used to store the trace data */
97 static char *data_buffer
;
98 static unsigned int data_buffer_size
;
101 * usage function on stderr
106 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
107 fprintf(stderr
, " -h, --help Display this usage.\n");
108 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
109 fprintf(stderr
, " -C, --control-port Control port listening (URI)\n");
110 fprintf(stderr
, " -D, --data-port Data port listening (URI)\n");
111 fprintf(stderr
, " -o, --output Output path for traces (PATH)\n");
112 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
116 int parse_args(int argc
, char **argv
)
120 char *default_address
;
122 static struct option long_options
[] = {
123 { "control-port", 1, 0, 'C', },
124 { "data-port", 1, 0, 'D', },
125 { "daemonize", 0, 0, 'd', },
126 { "help", 0, 0, 'h', },
127 { "output", 1, 0, 'o', },
128 { "verbose", 0, 0, 'v', },
133 int option_index
= 0;
134 c
= getopt_long(argc
, argv
, "dhv" "C:D:o:",
135 long_options
, &option_index
);
142 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
144 fprintf(stderr
, " with arg %s\n", optarg
);
148 ret
= uri_parse(optarg
, &control_uri
);
150 ERR("Invalid control URI specified");
153 if (control_uri
->port
== 0) {
154 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
158 ret
= uri_parse(optarg
, &data_uri
);
160 ERR("Invalid data URI specified");
163 if (data_uri
->port
== 0) {
164 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
174 ret
= asprintf(&opt_output_path
, "%s", optarg
);
176 PERROR("asprintf opt_output_path");
181 /* Verbose level can increase using multiple -v */
182 lttng_opt_verbose
+= 1;
185 /* Unknown option or other error.
186 * Error is printed by getopt, just return */
192 /* assign default values */
193 if (control_uri
== NULL
) {
194 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
195 DEFAULT_NETWORK_CONTROL_PORT
);
197 PERROR("asprintf default data address");
201 ret
= uri_parse(default_address
, &control_uri
);
202 free(default_address
);
204 ERR("Invalid control URI specified");
208 if (data_uri
== NULL
) {
209 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
210 DEFAULT_NETWORK_DATA_PORT
);
212 PERROR("asprintf default data address");
216 ret
= uri_parse(default_address
, &data_uri
);
217 free(default_address
);
219 ERR("Invalid data URI specified");
236 /* free the dynamically allocated opt_output_path */
237 free(opt_output_path
);
239 /* Close thread quit pipes */
240 utils_close_pipe(thread_quit_pipe
);
242 uri_free(control_uri
);
247 * Write to writable pipe used to notify a thread.
250 int notify_thread_pipe(int wpipe
)
255 ret
= write(wpipe
, "!", 1);
256 } while (ret
< 0 && errno
== EINTR
);
257 if (ret
< 0 || ret
!= 1) {
258 PERROR("write poll pipe");
265 * Stop all threads by closing the thread quit pipe.
268 void stop_threads(void)
272 /* Stopping all threads */
273 DBG("Terminating all threads");
274 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
276 ERR("write error on thread quit pipe");
279 /* Dispatch thread */
280 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
281 futex_nto1_wake(&relay_cmd_queue
.futex
);
285 * Signal handler for the daemon
287 * Simply stop all worker threads, leaving main() return gracefully after
288 * joining all threads and calling cleanup().
291 void sighandler(int sig
)
295 DBG("SIGPIPE caught");
298 DBG("SIGINT caught");
302 DBG("SIGTERM caught");
311 * Setup signal handler for :
312 * SIGINT, SIGTERM, SIGPIPE
315 int set_signal_handler(void)
321 if ((ret
= sigemptyset(&sigset
)) < 0) {
322 PERROR("sigemptyset");
326 sa
.sa_handler
= sighandler
;
329 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
334 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
339 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
344 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
350 * Init thread quit pipe.
352 * Return -1 on error or 0 if all pipes are created.
355 int init_thread_quit_pipe(void)
359 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
365 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
368 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
372 if (events
== NULL
|| size
== 0) {
377 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
383 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
395 * Check if the thread quit pipe was triggered.
397 * Return 1 if it was triggered else 0;
400 int check_thread_quit_pipe(int fd
, uint32_t events
)
402 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
410 * Create and init socket from uri.
413 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
416 struct lttcomm_sock
*sock
= NULL
;
418 sock
= lttcomm_alloc_sock_from_uri(uri
);
420 ERR("Allocating socket");
424 ret
= lttcomm_create_sock(sock
);
428 DBG("Listening on sock %d", sock
->fd
);
430 ret
= sock
->ops
->bind(sock
);
435 ret
= sock
->ops
->listen(sock
, -1);
445 lttcomm_destroy_sock(sock
);
451 * Return nonzero if stream needs to be closed.
454 int close_stream_check(struct relay_stream
*stream
)
457 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
459 * We are about to close the stream so set the data pending flag to 1
460 * which will make the end data pending command skip the stream which
461 * is now closed and ready. Note that after proceeding to a file close,
462 * the written file is ready for reading.
464 stream
->data_pending_check_done
= 1;
471 * This thread manages the listening for new connections on the network
474 void *relay_thread_listener(void *data
)
476 int i
, ret
, pollfd
, err
= -1;
478 uint32_t revents
, nb_fd
;
479 struct lttng_poll_event events
;
480 struct lttcomm_sock
*control_sock
, *data_sock
;
482 DBG("[thread] Relay listener started");
484 control_sock
= relay_init_sock(control_uri
);
486 goto error_sock_control
;
489 data_sock
= relay_init_sock(data_uri
);
491 goto error_sock_relay
;
495 * Pass 3 as size here for the thread quit pipe, control and data socket.
497 ret
= create_thread_poll_set(&events
, 3);
499 goto error_create_poll
;
502 /* Add the control socket */
503 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
508 /* Add the data socket */
509 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
515 DBG("Listener accepting connections");
518 ret
= lttng_poll_wait(&events
, -1);
521 * Restart interrupted system call.
523 if (errno
== EINTR
) {
531 DBG("Relay new connection received");
532 for (i
= 0; i
< nb_fd
; i
++) {
533 /* Fetch once the poll data */
534 revents
= LTTNG_POLL_GETEV(&events
, i
);
535 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
537 /* Thread quit pipe has been closed. Killing thread. */
538 ret
= check_thread_quit_pipe(pollfd
, revents
);
544 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
545 ERR("socket poll error");
547 } else if (revents
& LPOLLIN
) {
549 * Get allocated in this thread,
550 * enqueued to a global queue, dequeued
551 * and freed in the worker thread.
553 struct relay_command
*relay_cmd
;
554 struct lttcomm_sock
*newsock
;
556 relay_cmd
= zmalloc(sizeof(struct relay_command
));
557 if (relay_cmd
== NULL
) {
558 PERROR("relay command zmalloc");
562 if (pollfd
== data_sock
->fd
) {
563 newsock
= data_sock
->ops
->accept(data_sock
);
565 PERROR("accepting data sock");
569 relay_cmd
->type
= RELAY_DATA
;
570 DBG("Relay data connection accepted, socket %d", newsock
->fd
);
572 assert(pollfd
== control_sock
->fd
);
573 newsock
= control_sock
->ops
->accept(control_sock
);
575 PERROR("accepting control sock");
579 relay_cmd
->type
= RELAY_CONTROL
;
580 DBG("Relay control connection accepted, socket %d", newsock
->fd
);
582 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
,
585 PERROR("setsockopt inet");
586 lttcomm_destroy_sock(newsock
);
590 relay_cmd
->sock
= newsock
;
592 * Lock free enqueue the request.
594 cds_wfq_enqueue(&relay_cmd_queue
.queue
, &relay_cmd
->node
);
597 * Wake the dispatch queue futex. Implicit memory
598 * barrier with the exchange in cds_wfq_enqueue.
600 futex_nto1_wake(&relay_cmd_queue
.futex
);
608 lttng_poll_clean(&events
);
610 if (data_sock
->fd
>= 0) {
611 ret
= data_sock
->ops
->close(data_sock
);
616 lttcomm_destroy_sock(data_sock
);
618 if (control_sock
->fd
>= 0) {
619 ret
= control_sock
->ops
->close(control_sock
);
624 lttcomm_destroy_sock(control_sock
);
627 DBG("Thread exited with error");
629 DBG("Relay listener thread cleanup complete");
635 * This thread manages the dispatching of the requests to worker threads
638 void *relay_thread_dispatcher(void *data
)
641 struct cds_wfq_node
*node
;
642 struct relay_command
*relay_cmd
= NULL
;
644 DBG("[thread] Relay dispatcher started");
646 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
647 /* Atomically prepare the queue futex */
648 futex_nto1_prepare(&relay_cmd_queue
.futex
);
651 /* Dequeue commands */
652 node
= cds_wfq_dequeue_blocking(&relay_cmd_queue
.queue
);
654 DBG("Woken up but nothing in the relay command queue");
655 /* Continue thread execution */
659 relay_cmd
= caa_container_of(node
, struct relay_command
, node
);
660 DBG("Dispatching request waiting on sock %d", relay_cmd
->sock
->fd
);
663 * Inform worker thread of the new request. This
664 * call is blocking so we can be assured that the data will be read
665 * at some point in time or wait to the end of the world :)
668 ret
= write(relay_cmd_pipe
[1], relay_cmd
,
669 sizeof(struct relay_command
));
670 } while (ret
< 0 && errno
== EINTR
);
672 if (ret
< 0 || ret
!= sizeof(struct relay_command
)) {
673 PERROR("write cmd pipe");
676 } while (node
!= NULL
);
678 /* Futex wait on queue. Blocking call on futex() */
679 futex_nto1_wait(&relay_cmd_queue
.futex
);
683 DBG("Dispatch thread dying");
689 * config_get_default_path
691 * Returns the HOME directory path. Caller MUST NOT free(3) the return pointer.
694 char *config_get_default_path(void)
696 return getenv("HOME");
700 char *create_output_path_auto(char *path_name
)
703 char *traces_path
= NULL
;
704 char *alloc_path
= NULL
;
707 default_path
= config_get_default_path();
708 if (default_path
== NULL
) {
709 ERR("Home path not found.\n \
710 Please specify an output path using -o, --output PATH");
713 alloc_path
= strdup(default_path
);
714 if (alloc_path
== NULL
) {
715 PERROR("Path allocation");
718 ret
= asprintf(&traces_path
, "%s/" DEFAULT_TRACE_DIR_NAME
719 "/%s", alloc_path
, path_name
);
721 PERROR("asprintf trace dir name");
730 char *create_output_path_noauto(char *path_name
)
733 char *traces_path
= NULL
;
736 full_path
= utils_expand_path(opt_output_path
);
737 ret
= asprintf(&traces_path
, "%s/%s", full_path
, path_name
);
739 PERROR("asprintf trace dir name");
748 * create_output_path: create the output trace directory
751 char *create_output_path(char *path_name
)
753 if (opt_output_path
== NULL
) {
754 return create_output_path_auto(path_name
);
756 return create_output_path_noauto(path_name
);
761 * Get stream from stream id.
762 * Need to be called with RCU read-side lock held.
765 struct relay_stream
*relay_stream_from_stream_id(uint64_t stream_id
,
766 struct lttng_ht
*streams_ht
)
768 struct lttng_ht_node_ulong
*node
;
769 struct lttng_ht_iter iter
;
770 struct relay_stream
*ret
;
772 lttng_ht_lookup(streams_ht
,
773 (void *)((unsigned long) stream_id
),
775 node
= lttng_ht_iter_get_node_ulong(&iter
);
777 DBG("Relay stream %" PRIu64
" not found", stream_id
);
782 ret
= caa_container_of(node
, struct relay_stream
, stream_n
);
789 void deferred_free_stream(struct rcu_head
*head
)
791 struct relay_stream
*stream
=
792 caa_container_of(head
, struct relay_stream
, rcu_node
);
797 * relay_delete_session: Free all memory associated with a session and
801 void relay_delete_session(struct relay_command
*cmd
, struct lttng_ht
*streams_ht
)
803 struct lttng_ht_iter iter
;
804 struct lttng_ht_node_ulong
*node
;
805 struct relay_stream
*stream
;
812 DBG("Relay deleting session %" PRIu64
, cmd
->session
->id
);
815 cds_lfht_for_each_entry(streams_ht
->ht
, &iter
.iter
, node
, node
) {
816 node
= lttng_ht_iter_get_node_ulong(&iter
);
818 stream
= caa_container_of(node
,
819 struct relay_stream
, stream_n
);
820 if (stream
->session
== cmd
->session
) {
821 ret
= close(stream
->fd
);
823 PERROR("close stream fd on delete session");
825 ret
= lttng_ht_del(streams_ht
, &iter
);
827 call_rcu(&stream
->rcu_node
,
828 deferred_free_stream
);
838 * Handle the RELAYD_CREATE_SESSION command.
840 * On success, send back the session id or else return a negative value.
843 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
844 struct relay_command
*cmd
)
846 int ret
= 0, send_ret
;
847 struct relay_session
*session
;
848 struct lttcomm_relayd_status_session reply
;
853 memset(&reply
, 0, sizeof(reply
));
855 session
= zmalloc(sizeof(struct relay_session
));
856 if (session
== NULL
) {
857 PERROR("relay session zmalloc");
862 session
->id
= ++last_relay_session_id
;
863 session
->sock
= cmd
->sock
;
864 cmd
->session
= session
;
866 reply
.session_id
= htobe64(session
->id
);
868 DBG("Created session %" PRIu64
, session
->id
);
872 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
874 reply
.ret_code
= htobe32(LTTNG_OK
);
877 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
879 ERR("Relayd sending session id");
887 * relay_add_stream: allocate a new stream for a session
890 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
891 struct relay_command
*cmd
, struct lttng_ht
*streams_ht
)
893 struct relay_session
*session
= cmd
->session
;
894 struct lttcomm_relayd_add_stream stream_info
;
895 struct relay_stream
*stream
= NULL
;
896 struct lttcomm_relayd_status_stream reply
;
897 char *path
= NULL
, *root_path
= NULL
;
900 if (!session
|| cmd
->version_check_done
== 0) {
901 ERR("Trying to add a stream before version check");
906 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &stream_info
,
907 sizeof(struct lttcomm_relayd_add_stream
), 0);
908 if (ret
< sizeof(struct lttcomm_relayd_add_stream
)) {
910 /* Orderly shutdown. Not necessary to print an error. */
911 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
913 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
918 stream
= zmalloc(sizeof(struct relay_stream
));
919 if (stream
== NULL
) {
920 PERROR("relay stream zmalloc");
926 stream
->stream_handle
= ++last_relay_stream_id
;
927 stream
->prev_seq
= -1ULL;
928 stream
->session
= session
;
930 root_path
= create_output_path(stream_info
.pathname
);
935 ret
= utils_mkdir_recursive(root_path
, S_IRWXU
| S_IRWXG
);
937 ERR("relay creating output directory");
941 ret
= asprintf(&path
, "%s/%s", root_path
, stream_info
.channel_name
);
943 PERROR("asprintf stream path");
947 ret
= open(path
, O_WRONLY
|O_CREAT
|O_TRUNC
, S_IRWXU
|S_IRWXG
|S_IRWXO
);
949 PERROR("Relay creating trace file");
954 DBG("Tracefile %s created", path
);
956 lttng_ht_node_init_ulong(&stream
->stream_n
,
957 (unsigned long) stream
->stream_handle
);
958 lttng_ht_add_unique_ulong(streams_ht
,
961 DBG("Relay new stream added %s", stream_info
.channel_name
);
966 /* send the session id to the client or a negative return code on error */
968 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
970 reply
.ret_code
= htobe32(LTTNG_OK
);
972 reply
.handle
= htobe64(stream
->stream_handle
);
973 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
974 sizeof(struct lttcomm_relayd_status_stream
), 0);
976 ERR("Relay sending stream id");
986 * relay_close_stream: close a specific stream
989 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
990 struct relay_command
*cmd
, struct lttng_ht
*streams_ht
)
992 struct relay_session
*session
= cmd
->session
;
993 struct lttcomm_relayd_close_stream stream_info
;
994 struct lttcomm_relayd_generic_reply reply
;
995 struct relay_stream
*stream
;
997 struct lttng_ht_iter iter
;
999 DBG("Close stream received");
1001 if (!session
|| cmd
->version_check_done
== 0) {
1002 ERR("Trying to close a stream before version check");
1004 goto end_no_session
;
1007 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &stream_info
,
1008 sizeof(struct lttcomm_relayd_close_stream
), 0);
1009 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1011 /* Orderly shutdown. Not necessary to print an error. */
1012 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1014 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1017 goto end_no_session
;
1021 stream
= relay_stream_from_stream_id(be64toh(stream_info
.stream_id
),
1028 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1029 stream
->close_flag
= 1;
1031 if (close_stream_check(stream
)) {
1034 delret
= close(stream
->fd
);
1036 PERROR("close stream");
1038 iter
.iter
.node
= &stream
->stream_n
.node
;
1039 delret
= lttng_ht_del(streams_ht
, &iter
);
1041 call_rcu(&stream
->rcu_node
,
1042 deferred_free_stream
);
1043 DBG("Closed tracefile %d from close stream", stream
->fd
);
1050 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1052 reply
.ret_code
= htobe32(LTTNG_OK
);
1054 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1055 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1057 ERR("Relay sending stream id");
1066 * relay_unknown_command: send -1 if received unknown command
1069 void relay_unknown_command(struct relay_command
*cmd
)
1071 struct lttcomm_relayd_generic_reply reply
;
1074 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1075 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1076 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1078 ERR("Relay sending unknown command");
1083 * relay_start: send an acknowledgment to the client to tell if we are
1084 * ready to receive data. We are ready if a session is established.
1087 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1088 struct relay_command
*cmd
)
1090 int ret
= htobe32(LTTNG_OK
);
1091 struct lttcomm_relayd_generic_reply reply
;
1092 struct relay_session
*session
= cmd
->session
;
1095 DBG("Trying to start the streaming without a session established");
1096 ret
= htobe32(LTTNG_ERR_UNK
);
1099 reply
.ret_code
= ret
;
1100 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1101 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1103 ERR("Relay sending start ack");
1110 * Append padding to the file pointed by the file descriptor fd.
1112 static int write_padding_to_file(int fd
, uint32_t size
)
1121 zeros
= zmalloc(size
);
1122 if (zeros
== NULL
) {
1123 PERROR("zmalloc zeros for padding");
1129 ret
= write(fd
, zeros
, size
);
1130 } while (ret
< 0 && errno
== EINTR
);
1131 if (ret
< 0 || ret
!= size
) {
1132 PERROR("write padding to file");
1142 * relay_recv_metadata: receive the metada for the session.
1145 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1146 struct relay_command
*cmd
, struct lttng_ht
*streams_ht
)
1148 int ret
= htobe32(LTTNG_OK
);
1149 struct relay_session
*session
= cmd
->session
;
1150 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1151 struct relay_stream
*metadata_stream
;
1152 uint64_t data_size
, payload_size
;
1155 ERR("Metadata sent before version check");
1160 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1161 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1162 ERR("Incorrect data size");
1166 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1168 if (data_buffer_size
< data_size
) {
1169 /* In case the realloc fails, we can free the memory */
1172 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1173 if (!tmp_data_ptr
) {
1174 ERR("Allocating data buffer");
1179 data_buffer
= tmp_data_ptr
;
1180 data_buffer_size
= data_size
;
1182 memset(data_buffer
, 0, data_size
);
1183 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1184 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
1185 if (ret
< 0 || ret
!= data_size
) {
1187 /* Orderly shutdown. Not necessary to print an error. */
1188 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1190 ERR("Relay didn't receive the whole metadata");
1195 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1198 metadata_stream
= relay_stream_from_stream_id(
1199 be64toh(metadata_struct
->stream_id
), streams_ht
);
1200 if (!metadata_stream
) {
1206 ret
= write(metadata_stream
->fd
, metadata_struct
->payload
,
1208 } while (ret
< 0 && errno
== EINTR
);
1209 if (ret
< 0 || ret
!= payload_size
) {
1210 ERR("Relay error writing metadata on file");
1215 ret
= write_padding_to_file(metadata_stream
->fd
,
1216 be32toh(metadata_struct
->padding_size
));
1221 DBG2("Relay metadata written");
1230 * relay_send_version: send relayd version number
1233 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1234 struct relay_command
*cmd
)
1237 struct lttcomm_relayd_version reply
, msg
;
1241 cmd
->version_check_done
= 1;
1243 /* Get version from the other side. */
1244 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1245 if (ret
< 0 || ret
!= sizeof(msg
)) {
1247 /* Orderly shutdown. Not necessary to print an error. */
1248 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1250 ERR("Relay failed to receive the version values.");
1257 * For now, we just ignore the received version but after 2.1 stable
1258 * release, a check must be done to see if we either adapt to the other
1259 * side version (which MUST be lower than us) or keep the latest data
1260 * structure considering that the other side will adapt.
1263 ret
= sscanf(VERSION
, "%10u.%10u", &reply
.major
, &reply
.minor
);
1265 ERR("Error in scanning version");
1269 reply
.major
= htobe32(reply
.major
);
1270 reply
.minor
= htobe32(reply
.minor
);
1271 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1272 sizeof(struct lttcomm_relayd_version
), 0);
1274 ERR("Relay sending version");
1276 DBG("Version check done (%u.%u)", be32toh(reply
.major
),
1277 be32toh(reply
.minor
));
1284 * Check for data pending for a given stream id from the session daemon.
1287 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1288 struct relay_command
*cmd
, struct lttng_ht
*streams_ht
)
1290 struct relay_session
*session
= cmd
->session
;
1291 struct lttcomm_relayd_data_pending msg
;
1292 struct lttcomm_relayd_generic_reply reply
;
1293 struct relay_stream
*stream
;
1295 uint64_t last_net_seq_num
, stream_id
;
1297 DBG("Data pending command received");
1299 if (!session
|| cmd
->version_check_done
== 0) {
1300 ERR("Trying to check for data before version check");
1302 goto end_no_session
;
1305 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1306 if (ret
< sizeof(msg
)) {
1308 /* Orderly shutdown. Not necessary to print an error. */
1309 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1311 ERR("Relay didn't receive valid data_pending struct size : %d",
1315 goto end_no_session
;
1318 stream_id
= be64toh(msg
.stream_id
);
1319 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1322 stream
= relay_stream_from_stream_id(stream_id
, streams_ht
);
1323 if (stream
== NULL
) {
1328 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1329 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1332 /* Avoid wrapping issue */
1333 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1334 /* Data has in fact been written and is NOT pending */
1337 /* Data still being streamed thus pending */
1341 /* Pending check is now done. */
1342 stream
->data_pending_check_done
= 1;
1347 reply
.ret_code
= htobe32(ret
);
1348 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1350 ERR("Relay data pending ret code failed");
1358 * Wait for the control socket to reach a quiescent state.
1360 * Note that for now, when receiving this command from the session daemon, this
1361 * means that every subsequent commands or data received on the control socket
1362 * has been handled. So, this is why we simply return OK here.
1365 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1366 struct relay_command
*cmd
, struct lttng_ht
*streams_ht
)
1370 struct relay_stream
*stream
;
1371 struct lttng_ht_iter iter
;
1372 struct lttcomm_relayd_quiescent_control msg
;
1373 struct lttcomm_relayd_generic_reply reply
;
1375 DBG("Checking quiescent state on control socket");
1377 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1378 ERR("Trying to check for data before version check");
1380 goto end_no_session
;
1383 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1384 if (ret
< sizeof(msg
)) {
1386 /* Orderly shutdown. Not necessary to print an error. */
1387 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1389 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1393 goto end_no_session
;
1396 stream_id
= be64toh(msg
.stream_id
);
1399 cds_lfht_for_each_entry(streams_ht
->ht
, &iter
.iter
, stream
, stream_n
.node
) {
1400 if (stream
->stream_handle
== stream_id
) {
1401 stream
->data_pending_check_done
= 1;
1402 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1409 reply
.ret_code
= htobe32(LTTNG_OK
);
1410 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1412 ERR("Relay data quiescent control ret code failed");
1420 * Initialize a data pending command. This means that a client is about to ask
1421 * for data pending for each stream he/she holds. Simply iterate over all
1422 * streams of a session and set the data_pending_check_done flag.
1424 * This command returns to the client a LTTNG_OK code.
1427 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1428 struct relay_command
*cmd
, struct lttng_ht
*streams_ht
)
1431 struct lttng_ht_iter iter
;
1432 struct lttcomm_relayd_begin_data_pending msg
;
1433 struct lttcomm_relayd_generic_reply reply
;
1434 struct relay_stream
*stream
;
1435 uint64_t session_id
;
1441 DBG("Init streams for data pending");
1443 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1444 ERR("Trying to check for data before version check");
1446 goto end_no_session
;
1449 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1450 if (ret
< sizeof(msg
)) {
1452 /* Orderly shutdown. Not necessary to print an error. */
1453 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1455 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1459 goto end_no_session
;
1462 session_id
= be64toh(msg
.session_id
);
1465 * Iterate over all streams to set the begin data pending flag. For now, the
1466 * streams are indexed by stream handle so we have to iterate over all
1467 * streams to find the one associated with the right session_id.
1470 cds_lfht_for_each_entry(streams_ht
->ht
, &iter
.iter
, stream
, stream_n
.node
) {
1471 if (stream
->session
->id
== session_id
) {
1472 stream
->data_pending_check_done
= 0;
1473 DBG("Set begin data pending flag to stream %" PRIu64
,
1474 stream
->stream_handle
);
1479 /* All good, send back reply. */
1480 reply
.ret_code
= htobe32(LTTNG_OK
);
1482 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1484 ERR("Relay begin data pending send reply failed");
1492 * End data pending command. This will check, for a given session id, if each
1493 * stream associated with it has its data_pending_check_done flag set. If not,
1494 * this means that the client lost track of the stream but the data is still
1495 * being streamed on our side. In this case, we inform the client that data is
1498 * Return to the client if there is data in flight or not with a ret_code.
1501 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1502 struct relay_command
*cmd
, struct lttng_ht
*streams_ht
)
1505 struct lttng_ht_iter iter
;
1506 struct lttcomm_relayd_end_data_pending msg
;
1507 struct lttcomm_relayd_generic_reply reply
;
1508 struct relay_stream
*stream
;
1509 uint64_t session_id
;
1510 uint32_t is_data_inflight
= 0;
1516 DBG("End data pending command");
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 end data_pending struct size: %d",
1534 goto end_no_session
;
1537 session_id
= be64toh(msg
.session_id
);
1539 /* Iterate over all streams to see if the begin data pending flag is set. */
1541 cds_lfht_for_each_entry(streams_ht
->ht
, &iter
.iter
, stream
, stream_n
.node
) {
1542 if (stream
->session
->id
== session_id
&&
1543 !stream
->data_pending_check_done
) {
1544 is_data_inflight
= 1;
1545 DBG("Data is still in flight for stream %" PRIu64
,
1546 stream
->stream_handle
);
1552 /* All good, send back reply. */
1553 reply
.ret_code
= htobe32(is_data_inflight
);
1555 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1557 ERR("Relay end data pending send reply failed");
1565 * relay_process_control: Process the commands received on the control socket
1568 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1569 struct relay_command
*cmd
, struct lttng_ht
*streams_ht
)
1573 switch (be32toh(recv_hdr
->cmd
)) {
1574 case RELAYD_CREATE_SESSION
:
1575 ret
= relay_create_session(recv_hdr
, cmd
);
1577 case RELAYD_ADD_STREAM
:
1578 ret
= relay_add_stream(recv_hdr
, cmd
, streams_ht
);
1580 case RELAYD_START_DATA
:
1581 ret
= relay_start(recv_hdr
, cmd
);
1583 case RELAYD_SEND_METADATA
:
1584 ret
= relay_recv_metadata(recv_hdr
, cmd
, streams_ht
);
1586 case RELAYD_VERSION
:
1587 ret
= relay_send_version(recv_hdr
, cmd
);
1589 case RELAYD_CLOSE_STREAM
:
1590 ret
= relay_close_stream(recv_hdr
, cmd
, streams_ht
);
1592 case RELAYD_DATA_PENDING
:
1593 ret
= relay_data_pending(recv_hdr
, cmd
, streams_ht
);
1595 case RELAYD_QUIESCENT_CONTROL
:
1596 ret
= relay_quiescent_control(recv_hdr
, cmd
, streams_ht
);
1598 case RELAYD_BEGIN_DATA_PENDING
:
1599 ret
= relay_begin_data_pending(recv_hdr
, cmd
, streams_ht
);
1601 case RELAYD_END_DATA_PENDING
:
1602 ret
= relay_end_data_pending(recv_hdr
, cmd
, streams_ht
);
1604 case RELAYD_UPDATE_SYNC_INFO
:
1606 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
1607 relay_unknown_command(cmd
);
1617 * relay_process_data: Process the data received on the data socket
1620 int relay_process_data(struct relay_command
*cmd
, struct lttng_ht
*streams_ht
)
1623 struct relay_stream
*stream
;
1624 struct lttcomm_relayd_data_hdr data_hdr
;
1626 uint64_t net_seq_num
;
1629 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &data_hdr
,
1630 sizeof(struct lttcomm_relayd_data_hdr
), 0);
1633 /* Orderly shutdown. Not necessary to print an error. */
1634 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1636 ERR("Unable to receive data header on sock %d", cmd
->sock
->fd
);
1642 stream_id
= be64toh(data_hdr
.stream_id
);
1645 stream
= relay_stream_from_stream_id(stream_id
, streams_ht
);
1651 data_size
= be32toh(data_hdr
.data_size
);
1652 if (data_buffer_size
< data_size
) {
1655 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1656 if (!tmp_data_ptr
) {
1657 ERR("Allocating data buffer");
1662 data_buffer
= tmp_data_ptr
;
1663 data_buffer_size
= data_size
;
1665 memset(data_buffer
, 0, data_size
);
1667 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
1669 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
1670 data_size
, stream_id
, net_seq_num
);
1671 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
1674 /* Orderly shutdown. Not necessary to print an error. */
1675 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1682 ret
= write(stream
->fd
, data_buffer
, data_size
);
1683 } while (ret
< 0 && errno
== EINTR
);
1684 if (ret
< 0 || ret
!= data_size
) {
1685 ERR("Relay error writing data to file");
1690 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
1691 ret
, stream
->stream_handle
);
1693 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
1698 stream
->prev_seq
= net_seq_num
;
1700 /* Check if we need to close the FD */
1701 if (close_stream_check(stream
)) {
1703 struct lttng_ht_iter iter
;
1705 cret
= close(stream
->fd
);
1707 PERROR("close stream process data");
1709 iter
.iter
.node
= &stream
->stream_n
.node
;
1710 ret
= lttng_ht_del(streams_ht
, &iter
);
1712 call_rcu(&stream
->rcu_node
,
1713 deferred_free_stream
);
1714 DBG("Closed tracefile %d after recv data", stream
->fd
);
1724 void relay_cleanup_poll_connection(struct lttng_poll_event
*events
, int pollfd
)
1728 lttng_poll_del(events
, pollfd
);
1730 ret
= close(pollfd
);
1732 ERR("Closing pollfd %d", pollfd
);
1737 int relay_add_connection(int fd
, struct lttng_poll_event
*events
,
1738 struct lttng_ht
*relay_connections_ht
)
1740 struct relay_command
*relay_connection
;
1743 relay_connection
= zmalloc(sizeof(struct relay_command
));
1744 if (relay_connection
== NULL
) {
1745 PERROR("Relay command zmalloc");
1749 ret
= read(fd
, relay_connection
, sizeof(struct relay_command
));
1750 } while (ret
< 0 && errno
== EINTR
);
1751 if (ret
< 0 || ret
< sizeof(struct relay_command
)) {
1752 PERROR("read relay cmd pipe");
1756 lttng_ht_node_init_ulong(&relay_connection
->sock_n
,
1757 (unsigned long) relay_connection
->sock
->fd
);
1759 lttng_ht_add_unique_ulong(relay_connections_ht
,
1760 &relay_connection
->sock_n
);
1762 return lttng_poll_add(events
,
1763 relay_connection
->sock
->fd
,
1764 LPOLLIN
| LPOLLRDHUP
);
1767 free(relay_connection
);
1773 void deferred_free_connection(struct rcu_head
*head
)
1775 struct relay_command
*relay_connection
=
1776 caa_container_of(head
, struct relay_command
, rcu_node
);
1778 lttcomm_destroy_sock(relay_connection
->sock
);
1779 free(relay_connection
);
1783 void relay_del_connection(struct lttng_ht
*relay_connections_ht
,
1784 struct lttng_ht
*streams_ht
, struct lttng_ht_iter
*iter
,
1785 struct relay_command
*relay_connection
)
1789 ret
= lttng_ht_del(relay_connections_ht
, iter
);
1791 if (relay_connection
->type
== RELAY_CONTROL
) {
1792 relay_delete_session(relay_connection
, streams_ht
);
1795 call_rcu(&relay_connection
->rcu_node
,
1796 deferred_free_connection
);
1800 * This thread does the actual work
1803 void *relay_thread_worker(void *data
)
1805 int ret
, err
= -1, last_seen_data_fd
= -1;
1807 struct relay_command
*relay_connection
;
1808 struct lttng_poll_event events
;
1809 struct lttng_ht
*relay_connections_ht
;
1810 struct lttng_ht_node_ulong
*node
;
1811 struct lttng_ht_iter iter
;
1812 struct lttng_ht
*streams_ht
;
1813 struct lttcomm_relayd_hdr recv_hdr
;
1815 DBG("[thread] Relay worker started");
1817 rcu_register_thread();
1819 /* table of connections indexed on socket */
1820 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
1821 if (!relay_connections_ht
) {
1822 goto relay_connections_ht_error
;
1825 /* tables of streams indexed by stream ID */
1826 streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
1828 goto streams_ht_error
;
1831 ret
= create_thread_poll_set(&events
, 2);
1833 goto error_poll_create
;
1836 ret
= lttng_poll_add(&events
, relay_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1843 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
1845 /* Infinite blocking call, waiting for transmission */
1846 DBG3("Relayd worker thread polling...");
1847 ret
= lttng_poll_wait(&events
, -1);
1850 * Restart interrupted system call.
1852 if (errno
== EINTR
) {
1861 * Process control. The control connection is prioritised so we don't
1862 * starve it with high throughout put tracing data on the data
1865 for (i
= 0; i
< nb_fd
; i
++) {
1866 /* Fetch once the poll data */
1867 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1868 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1870 /* Thread quit pipe has been closed. Killing thread. */
1871 ret
= check_thread_quit_pipe(pollfd
, revents
);
1877 /* Inspect the relay cmd pipe for new connection */
1878 if (pollfd
== relay_cmd_pipe
[0]) {
1879 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1880 ERR("Relay pipe error");
1882 } else if (revents
& LPOLLIN
) {
1883 DBG("Relay command received");
1884 ret
= relay_add_connection(relay_cmd_pipe
[0],
1885 &events
, relay_connections_ht
);
1890 } else if (revents
) {
1892 lttng_ht_lookup(relay_connections_ht
,
1893 (void *)((unsigned long) pollfd
),
1895 node
= lttng_ht_iter_get_node_ulong(&iter
);
1897 DBG2("Relay sock %d not found", pollfd
);
1901 relay_connection
= caa_container_of(node
,
1902 struct relay_command
, sock_n
);
1904 if (revents
& (LPOLLERR
)) {
1906 relay_cleanup_poll_connection(&events
, pollfd
);
1907 relay_del_connection(relay_connections_ht
,
1910 if (last_seen_data_fd
== pollfd
) {
1911 last_seen_data_fd
= last_notdel_data_fd
;
1913 } else if (revents
& (LPOLLHUP
| LPOLLRDHUP
)) {
1914 DBG("Socket %d hung up", pollfd
);
1915 relay_cleanup_poll_connection(&events
, pollfd
);
1916 relay_del_connection(relay_connections_ht
,
1919 if (last_seen_data_fd
== pollfd
) {
1920 last_seen_data_fd
= last_notdel_data_fd
;
1922 } else if (revents
& LPOLLIN
) {
1923 /* control socket */
1924 if (relay_connection
->type
== RELAY_CONTROL
) {
1925 ret
= relay_connection
->sock
->ops
->recvmsg(
1926 relay_connection
->sock
, &recv_hdr
,
1927 sizeof(struct lttcomm_relayd_hdr
), 0);
1928 /* connection closed */
1930 relay_cleanup_poll_connection(&events
, pollfd
);
1931 relay_del_connection(relay_connections_ht
,
1934 DBG("Control connection closed with %d", pollfd
);
1936 if (relay_connection
->session
) {
1937 DBG2("Relay worker receiving data for session : %" PRIu64
,
1938 relay_connection
->session
->id
);
1940 ret
= relay_process_control(&recv_hdr
,
1944 /* Clear the session on error. */
1945 relay_cleanup_poll_connection(&events
, pollfd
);
1946 relay_del_connection(relay_connections_ht
,
1949 DBG("Connection closed with %d", pollfd
);
1955 * Flag the last seen data fd not deleted. It will be
1956 * used as the last seen fd if any fd gets deleted in
1959 last_notdel_data_fd
= pollfd
;
1967 * The last loop handled a control request, go back to poll to make
1968 * sure we prioritise the control socket.
1974 if (last_seen_data_fd
>= 0) {
1975 for (i
= 0; i
< nb_fd
; i
++) {
1976 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1977 if (last_seen_data_fd
== pollfd
) {
1984 /* Process data connection. */
1985 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
1986 /* Fetch the poll data. */
1987 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1988 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1990 /* Skip the command pipe. It's handled in the first loop. */
1991 if (pollfd
== relay_cmd_pipe
[0]) {
1997 lttng_ht_lookup(relay_connections_ht
,
1998 (void *)((unsigned long) pollfd
),
2000 node
= lttng_ht_iter_get_node_ulong(&iter
);
2002 /* Skip it. Might be removed before. */
2006 relay_connection
= caa_container_of(node
,
2007 struct relay_command
, sock_n
);
2009 if (revents
& LPOLLIN
) {
2010 if (relay_connection
->type
!= RELAY_DATA
) {
2014 ret
= relay_process_data(relay_connection
, streams_ht
);
2015 /* connection closed */
2017 relay_cleanup_poll_connection(&events
, pollfd
);
2018 relay_del_connection(relay_connections_ht
,
2021 DBG("Data connection closed with %d", pollfd
);
2023 * Every goto restart call sets the last seen fd where
2024 * here we don't really care since we gracefully
2025 * continue the loop after the connection is deleted.
2028 /* Keep last seen port. */
2029 last_seen_data_fd
= pollfd
;
2037 last_seen_data_fd
= -1;
2042 lttng_poll_clean(&events
);
2044 /* empty the hash table and free the memory */
2046 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, node
, node
) {
2047 node
= lttng_ht_iter_get_node_ulong(&iter
);
2049 relay_connection
= caa_container_of(node
,
2050 struct relay_command
, sock_n
);
2051 relay_del_connection(relay_connections_ht
,
2058 lttng_ht_destroy(streams_ht
);
2060 lttng_ht_destroy(relay_connections_ht
);
2061 relay_connections_ht_error
:
2062 /* Close relay cmd pipes */
2063 utils_close_pipe(relay_cmd_pipe
);
2065 DBG("Thread exited with error");
2067 DBG("Worker thread cleanup complete");
2070 rcu_unregister_thread();
2075 * Create the relay command pipe to wake thread_manage_apps.
2076 * Closed in cleanup().
2078 static int create_relay_cmd_pipe(void)
2082 ret
= utils_create_pipe_cloexec(relay_cmd_pipe
);
2090 int main(int argc
, char **argv
)
2095 /* Create thread quit pipe */
2096 if ((ret
= init_thread_quit_pipe()) < 0) {
2100 /* Parse arguments */
2102 if ((ret
= parse_args(argc
, argv
)) < 0) {
2106 if ((ret
= set_signal_handler()) < 0) {
2110 /* Try to create directory if -o, --output is specified. */
2111 if (opt_output_path
) {
2112 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2114 ERR("Unable to create %s", opt_output_path
);
2128 /* Check if daemon is UID = 0 */
2129 is_root
= !getuid();
2132 if (control_uri
->port
< 1024 || data_uri
->port
< 1024) {
2133 ERR("Need to be root to use ports < 1024");
2139 /* Setup the thread apps communication pipe. */
2140 if ((ret
= create_relay_cmd_pipe()) < 0) {
2144 /* Init relay command queue. */
2145 cds_wfq_init(&relay_cmd_queue
.queue
);
2147 /* Set up max poll set size */
2148 lttng_poll_set_max_size();
2150 /* Setup the dispatcher thread */
2151 ret
= pthread_create(&dispatcher_thread
, NULL
,
2152 relay_thread_dispatcher
, (void *) NULL
);
2154 PERROR("pthread_create dispatcher");
2155 goto exit_dispatcher
;
2158 /* Setup the worker thread */
2159 ret
= pthread_create(&worker_thread
, NULL
,
2160 relay_thread_worker
, (void *) NULL
);
2162 PERROR("pthread_create worker");
2166 /* Setup the listener thread */
2167 ret
= pthread_create(&listener_thread
, NULL
,
2168 relay_thread_listener
, (void *) NULL
);
2170 PERROR("pthread_create listener");
2175 ret
= pthread_join(listener_thread
, &status
);
2177 PERROR("pthread_join");
2178 goto error
; /* join error, exit without cleanup */
2182 ret
= pthread_join(worker_thread
, &status
);
2184 PERROR("pthread_join");
2185 goto error
; /* join error, exit without cleanup */
2189 ret
= pthread_join(dispatcher_thread
, &status
);
2191 PERROR("pthread_join");
2192 goto error
; /* join error, exit without cleanup */