2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
34 #include <sys/types.h>
37 #include <urcu/futex.h>
38 #include <urcu/uatomic.h>
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/defaults.h>
48 #include <common/futex.h>
49 #include <common/sessiond-comm/sessiond-comm.h>
50 #include <common/sessiond-comm/inet.h>
51 #include <common/sessiond-comm/relayd.h>
52 #include <common/uri.h>
53 #include <common/utils.h>
54 #include <common/config/config.h>
57 #include "ctf-trace.h"
60 #include "lttng-relayd.h"
62 #include "health-relayd.h"
64 /* command line options */
65 char *opt_output_path
;
66 static int opt_daemon
, opt_background
;
67 static struct lttng_uri
*control_uri
;
68 static struct lttng_uri
*data_uri
;
69 static struct lttng_uri
*live_uri
;
73 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
74 static int tracing_group_name_override
;
76 const char * const config_section_name
= "relayd";
79 * Quit pipe for all threads. This permits a single cancellation point
80 * for all threads when receiving an event on the pipe.
82 static int thread_quit_pipe
[2] = { -1, -1 };
85 * This pipe is used to inform the worker thread that a command is queued and
86 * ready to be processed.
88 static int relay_cmd_pipe
[2] = { -1, -1 };
90 /* Shared between threads */
91 static int dispatch_thread_exit
;
93 static pthread_t listener_thread
;
94 static pthread_t dispatcher_thread
;
95 static pthread_t worker_thread
;
96 static pthread_t health_thread
;
98 static uint64_t last_relay_stream_id
;
99 static uint64_t last_relay_session_id
;
102 * Relay command queue.
104 * The relay_thread_listener and relay_thread_dispatcher communicate with this
107 static struct relay_cmd_queue relay_cmd_queue
;
109 /* buffer allocated at startup, used to store the trace data */
110 static char *data_buffer
;
111 static unsigned int data_buffer_size
;
113 /* We need those values for the file/dir creation. */
114 static uid_t relayd_uid
;
115 static gid_t relayd_gid
;
117 /* Global relay stream hash table. */
118 struct lttng_ht
*relay_streams_ht
;
120 /* Global relay viewer stream hash table. */
121 struct lttng_ht
*viewer_streams_ht
;
123 /* Global hash table that stores relay index object. */
124 struct lttng_ht
*indexes_ht
;
126 /* Relayd health monitoring */
127 struct health_app
*health_relayd
;
129 static struct option long_options
[] = {
130 { "control-port", 1, 0, 'C', },
131 { "data-port", 1, 0, 'D', },
132 { "live-port", 1, 0, 'L', },
133 { "daemonize", 0, 0, 'd', },
134 { "background", 0, 0, 'b', },
135 { "group", 1, 0, 'g', },
136 { "help", 0, 0, 'h', },
137 { "output", 1, 0, 'o', },
138 { "verbose", 0, 0, 'v', },
139 { "config", 1, 0, 'f' },
143 static const char *config_ignore_options
[] = { "help", "config" };
146 * usage function on stderr
151 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
152 fprintf(stderr
, " -h, --help Display this usage.\n");
153 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
154 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
155 fprintf(stderr
, " -C, --control-port URL Control port listening.\n");
156 fprintf(stderr
, " -D, --data-port URL Data port listening.\n");
157 fprintf(stderr
, " -L, --live-port URL Live view port listening.\n");
158 fprintf(stderr
, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
159 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
160 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
161 fprintf(stderr
, " -f --config Load daemon configuration file\n");
165 * Take an option from the getopt output and set it in the right variable to be
168 * Return 0 on success else a negative value.
171 int set_option(int opt
, const char *arg
, const char *optname
)
177 fprintf(stderr
, "option %s", optname
);
179 fprintf(stderr
, " with arg %s\n", arg
);
183 ret
= uri_parse(arg
, &control_uri
);
185 ERR("Invalid control URI specified");
188 if (control_uri
->port
== 0) {
189 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
193 ret
= uri_parse(arg
, &data_uri
);
195 ERR("Invalid data URI specified");
198 if (data_uri
->port
== 0) {
199 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
203 ret
= uri_parse(arg
, &live_uri
);
205 ERR("Invalid live URI specified");
208 if (live_uri
->port
== 0) {
209 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
219 tracing_group_name
= strdup(arg
);
220 tracing_group_name_override
= 1;
226 ret
= asprintf(&opt_output_path
, "%s", arg
);
229 PERROR("asprintf opt_output_path");
234 /* Verbose level can increase using multiple -v */
236 lttng_opt_verbose
= config_parse_value(arg
);
238 lttng_opt_verbose
+= 1;
242 /* Unknown option or other error.
243 * Error is printed by getopt, just return */
256 * config_entry_handler_cb used to handle options read from a config file.
257 * See config_entry_handler_cb comment in common/config/config.h for the
258 * return value conventions.
261 int config_entry_handler(const struct config_entry
*entry
, void *unused
)
265 if (!entry
|| !entry
->name
|| !entry
->value
) {
270 /* Check if the option is to be ignored */
271 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
272 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
277 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
278 /* Ignore if entry name is not fully matched. */
279 if (strcmp(entry
->name
, long_options
[i
].name
)) {
284 * If the option takes no argument on the command line, we have to
285 * check if the value is "true". We support non-zero numeric values,
288 if (!long_options
[i
].has_arg
) {
289 ret
= config_parse_value(entry
->value
);
292 WARN("Invalid configuration value \"%s\" for option %s",
293 entry
->value
, entry
->name
);
295 /* False, skip boolean config option. */
300 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
304 WARN("Unrecognized option \"%s\" in daemon configuration file.",
312 int set_options(int argc
, char **argv
)
314 int c
, ret
= 0, option_index
= 0;
315 int orig_optopt
= optopt
, orig_optind
= optind
;
316 char *default_address
, *optstring
;
317 const char *config_path
= NULL
;
319 optstring
= utils_generate_optstring(long_options
,
320 sizeof(long_options
) / sizeof(struct option
));
326 /* Check for the --config option */
328 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
329 &option_index
)) != -1) {
333 } else if (c
!= 'f') {
337 config_path
= utils_expand_path(optarg
);
339 ERR("Failed to resolve path: %s", optarg
);
343 ret
= config_get_section_entries(config_path
, config_section_name
,
344 config_entry_handler
, NULL
);
347 ERR("Invalid configuration option at line %i", ret
);
353 /* Reset getopt's global state */
354 optopt
= orig_optopt
;
355 optind
= orig_optind
;
357 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
362 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
368 /* assign default values */
369 if (control_uri
== NULL
) {
370 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
371 DEFAULT_NETWORK_CONTROL_PORT
);
373 PERROR("asprintf default data address");
377 ret
= uri_parse(default_address
, &control_uri
);
378 free(default_address
);
380 ERR("Invalid control URI specified");
384 if (data_uri
== NULL
) {
385 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
386 DEFAULT_NETWORK_DATA_PORT
);
388 PERROR("asprintf default data address");
392 ret
= uri_parse(default_address
, &data_uri
);
393 free(default_address
);
395 ERR("Invalid data URI specified");
399 if (live_uri
== NULL
) {
400 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
401 DEFAULT_NETWORK_VIEWER_PORT
);
403 PERROR("asprintf default viewer control address");
407 ret
= uri_parse(default_address
, &live_uri
);
408 free(default_address
);
410 ERR("Invalid viewer control URI specified");
428 /* free the dynamically allocated opt_output_path */
429 free(opt_output_path
);
431 /* Close thread quit pipes */
432 utils_close_pipe(thread_quit_pipe
);
434 uri_free(control_uri
);
436 /* Live URI is freed in the live thread. */
438 if (tracing_group_name_override
) {
439 free((void *) tracing_group_name
);
444 * Write to writable pipe used to notify a thread.
447 int notify_thread_pipe(int wpipe
)
451 ret
= lttng_write(wpipe
, "!", 1);
453 PERROR("write poll pipe");
459 static void notify_health_quit_pipe(int *pipe
)
463 ret
= lttng_write(pipe
[1], "4", 1);
465 PERROR("write relay health quit");
470 * Stop all threads by closing the thread quit pipe.
473 void stop_threads(void)
477 /* Stopping all threads */
478 DBG("Terminating all threads");
479 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
481 ERR("write error on thread quit pipe");
484 notify_health_quit_pipe(health_quit_pipe
);
486 /* Dispatch thread */
487 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
488 futex_nto1_wake(&relay_cmd_queue
.futex
);
492 * Signal handler for the daemon
494 * Simply stop all worker threads, leaving main() return gracefully after
495 * joining all threads and calling cleanup().
498 void sighandler(int sig
)
502 DBG("SIGPIPE caught");
505 DBG("SIGINT caught");
509 DBG("SIGTERM caught");
518 * Setup signal handler for :
519 * SIGINT, SIGTERM, SIGPIPE
522 int set_signal_handler(void)
528 if ((ret
= sigemptyset(&sigset
)) < 0) {
529 PERROR("sigemptyset");
533 sa
.sa_handler
= sighandler
;
536 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
541 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
546 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
551 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
557 * Init thread quit pipe.
559 * Return -1 on error or 0 if all pipes are created.
562 int init_thread_quit_pipe(void)
566 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
572 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
575 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
579 if (events
== NULL
|| size
== 0) {
584 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
590 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
602 * Check if the thread quit pipe was triggered.
604 * Return 1 if it was triggered else 0;
607 int check_thread_quit_pipe(int fd
, uint32_t events
)
609 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
617 * Create and init socket from uri.
620 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
623 struct lttcomm_sock
*sock
= NULL
;
625 sock
= lttcomm_alloc_sock_from_uri(uri
);
627 ERR("Allocating socket");
631 ret
= lttcomm_create_sock(sock
);
635 DBG("Listening on sock %d", sock
->fd
);
637 ret
= sock
->ops
->bind(sock
);
642 ret
= sock
->ops
->listen(sock
, -1);
652 lttcomm_destroy_sock(sock
);
658 * Return nonzero if stream needs to be closed.
661 int close_stream_check(struct relay_stream
*stream
)
664 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
666 * We are about to close the stream so set the data pending flag to 1
667 * which will make the end data pending command skip the stream which
668 * is now closed and ready. Note that after proceeding to a file close,
669 * the written file is ready for reading.
671 stream
->data_pending_check_done
= 1;
678 * This thread manages the listening for new connections on the network
681 void *relay_thread_listener(void *data
)
683 int i
, ret
, pollfd
, err
= -1;
685 uint32_t revents
, nb_fd
;
686 struct lttng_poll_event events
;
687 struct lttcomm_sock
*control_sock
, *data_sock
;
689 DBG("[thread] Relay listener started");
691 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
693 health_code_update();
695 control_sock
= relay_init_sock(control_uri
);
697 goto error_sock_control
;
700 data_sock
= relay_init_sock(data_uri
);
702 goto error_sock_relay
;
706 * Pass 3 as size here for the thread quit pipe, control and data socket.
708 ret
= create_thread_poll_set(&events
, 3);
710 goto error_create_poll
;
713 /* Add the control socket */
714 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
719 /* Add the data socket */
720 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
726 health_code_update();
728 DBG("Listener accepting connections");
732 ret
= lttng_poll_wait(&events
, -1);
736 * Restart interrupted system call.
738 if (errno
== EINTR
) {
746 DBG("Relay new connection received");
747 for (i
= 0; i
< nb_fd
; i
++) {
748 health_code_update();
750 /* Fetch once the poll data */
751 revents
= LTTNG_POLL_GETEV(&events
, i
);
752 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
754 /* Thread quit pipe has been closed. Killing thread. */
755 ret
= check_thread_quit_pipe(pollfd
, revents
);
761 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
762 ERR("socket poll error");
764 } else if (revents
& LPOLLIN
) {
766 * Get allocated in this thread,
767 * enqueued to a global queue, dequeued
768 * and freed in the worker thread.
770 struct relay_command
*relay_cmd
;
771 struct lttcomm_sock
*newsock
;
773 relay_cmd
= zmalloc(sizeof(struct relay_command
));
774 if (relay_cmd
== NULL
) {
775 PERROR("relay command zmalloc");
779 if (pollfd
== data_sock
->fd
) {
780 newsock
= data_sock
->ops
->accept(data_sock
);
782 PERROR("accepting data sock");
786 relay_cmd
->type
= RELAY_DATA
;
787 DBG("Relay data connection accepted, socket %d", newsock
->fd
);
789 assert(pollfd
== control_sock
->fd
);
790 newsock
= control_sock
->ops
->accept(control_sock
);
792 PERROR("accepting control sock");
796 relay_cmd
->type
= RELAY_CONTROL
;
797 DBG("Relay control connection accepted, socket %d", newsock
->fd
);
799 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
,
802 PERROR("setsockopt inet");
803 lttcomm_destroy_sock(newsock
);
807 relay_cmd
->sock
= newsock
;
809 * Lock free enqueue the request.
811 cds_wfq_enqueue(&relay_cmd_queue
.queue
, &relay_cmd
->node
);
814 * Wake the dispatch queue futex. Implicit memory
815 * barrier with the exchange in cds_wfq_enqueue.
817 futex_nto1_wake(&relay_cmd_queue
.futex
);
825 lttng_poll_clean(&events
);
827 if (data_sock
->fd
>= 0) {
828 ret
= data_sock
->ops
->close(data_sock
);
833 lttcomm_destroy_sock(data_sock
);
835 if (control_sock
->fd
>= 0) {
836 ret
= control_sock
->ops
->close(control_sock
);
841 lttcomm_destroy_sock(control_sock
);
845 ERR("Health error occurred in %s", __func__
);
847 health_unregister(health_relayd
);
848 DBG("Relay listener thread cleanup complete");
854 * This thread manages the dispatching of the requests to worker threads
857 void *relay_thread_dispatcher(void *data
)
861 struct cds_wfq_node
*node
;
862 struct relay_command
*relay_cmd
= NULL
;
864 DBG("[thread] Relay dispatcher started");
866 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
868 health_code_update();
870 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
871 health_code_update();
873 /* Atomically prepare the queue futex */
874 futex_nto1_prepare(&relay_cmd_queue
.futex
);
877 health_code_update();
879 /* Dequeue commands */
880 node
= cds_wfq_dequeue_blocking(&relay_cmd_queue
.queue
);
882 DBG("Woken up but nothing in the relay command queue");
883 /* Continue thread execution */
887 relay_cmd
= caa_container_of(node
, struct relay_command
, node
);
888 DBG("Dispatching request waiting on sock %d", relay_cmd
->sock
->fd
);
891 * Inform worker thread of the new request. This
892 * call is blocking so we can be assured that the data will be read
893 * at some point in time or wait to the end of the world :)
895 ret
= lttng_write(relay_cmd_pipe
[1], relay_cmd
,
896 sizeof(struct relay_command
));
898 if (ret
< sizeof(struct relay_command
)) {
899 PERROR("write cmd pipe");
902 } while (node
!= NULL
);
904 /* Futex wait on queue. Blocking call on futex() */
906 futex_nto1_wait(&relay_cmd_queue
.futex
);
910 /* Normal exit, no error */
916 ERR("Health error occurred in %s", __func__
);
918 health_unregister(health_relayd
);
919 DBG("Dispatch thread dying");
925 * Get stream from stream id.
926 * Need to be called with RCU read-side lock held.
928 struct relay_stream
*relay_stream_find_by_id(uint64_t stream_id
)
930 struct lttng_ht_node_ulong
*node
;
931 struct lttng_ht_iter iter
;
932 struct relay_stream
*ret
;
934 lttng_ht_lookup(relay_streams_ht
,
935 (void *)((unsigned long) stream_id
),
937 node
= lttng_ht_iter_get_node_ulong(&iter
);
939 DBG("Relay stream %" PRIu64
" not found", stream_id
);
944 ret
= caa_container_of(node
, struct relay_stream
, stream_n
);
951 void deferred_free_stream(struct rcu_head
*head
)
953 struct relay_stream
*stream
=
954 caa_container_of(head
, struct relay_stream
, rcu_node
);
956 free(stream
->path_name
);
957 free(stream
->channel_name
);
962 void deferred_free_session(struct rcu_head
*head
)
964 struct relay_session
*session
=
965 caa_container_of(head
, struct relay_session
, rcu_node
);
970 * Close a given stream. The stream is freed using a call RCU.
972 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
973 * BEFORE the stream lock MUST be acquired.
975 static void destroy_stream(struct relay_stream
*stream
)
978 struct relay_viewer_stream
*vstream
;
979 struct lttng_ht_iter iter
;
983 delret
= close(stream
->fd
);
985 PERROR("close stream");
988 if (stream
->index_fd
>= 0) {
989 delret
= close(stream
->index_fd
);
991 PERROR("close stream index_fd");
995 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
);
998 * Set the last good value into the viewer stream. This is done
999 * right before the stream gets deleted from the hash table. The
1000 * lookup failure on the live thread side of a stream indicates
1001 * that the viewer stream index received value should be used.
1003 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
1004 vstream
->total_index_received
= stream
->total_index_received
;
1005 vstream
->tracefile_count_last
= stream
->tracefile_count_current
;
1006 vstream
->close_write_flag
= 1;
1007 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
1010 /* Cleanup index of that stream. */
1011 relay_index_destroy_by_stream_id(stream
->stream_handle
);
1013 iter
.iter
.node
= &stream
->stream_n
.node
;
1014 delret
= lttng_ht_del(relay_streams_ht
, &iter
);
1016 iter
.iter
.node
= &stream
->ctf_trace_node
.node
;
1017 delret
= lttng_ht_del(stream
->ctf_traces_ht
, &iter
);
1020 if (stream
->ctf_trace
) {
1021 ctf_trace_try_destroy(stream
->ctf_trace
);
1024 call_rcu(&stream
->rcu_node
, deferred_free_stream
);
1025 DBG("Closed tracefile %d from close stream", stream
->fd
);
1029 * relay_delete_session: Free all memory associated with a session and
1033 void relay_delete_session(struct relay_command
*cmd
,
1034 struct lttng_ht
*sessions_ht
)
1036 struct lttng_ht_iter iter
;
1037 struct lttng_ht_node_ulong
*node
;
1038 struct relay_stream
*stream
;
1041 if (!cmd
->session
) {
1045 DBG("Relay deleting session %" PRIu64
, cmd
->session
->id
);
1048 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, node
, node
) {
1049 node
= lttng_ht_iter_get_node_ulong(&iter
);
1053 stream
= caa_container_of(node
, struct relay_stream
, stream_n
);
1054 if (stream
->session
== cmd
->session
) {
1055 destroy_stream(stream
);
1056 cmd
->session
->stream_count
--;
1057 assert(cmd
->session
->stream_count
>= 0);
1061 /* Make this session not visible anymore. */
1062 iter
.iter
.node
= &cmd
->session
->session_n
.node
;
1063 ret
= lttng_ht_del(sessions_ht
, &iter
);
1065 call_rcu(&cmd
->session
->rcu_node
, deferred_free_session
);
1070 * Copy index data from the control port to a given index object.
1072 static void copy_index_control_data(struct relay_index
*index
,
1073 struct lttcomm_relayd_index
*data
)
1079 * The index on disk is encoded in big endian, so we don't need to convert
1080 * the data received on the network. The data_offset value is NEVER
1081 * modified here and is updated by the data thread.
1083 index
->index_data
.packet_size
= data
->packet_size
;
1084 index
->index_data
.content_size
= data
->content_size
;
1085 index
->index_data
.timestamp_begin
= data
->timestamp_begin
;
1086 index
->index_data
.timestamp_end
= data
->timestamp_end
;
1087 index
->index_data
.events_discarded
= data
->events_discarded
;
1088 index
->index_data
.stream_id
= data
->stream_id
;
1092 * Handle the RELAYD_CREATE_SESSION command.
1094 * On success, send back the session id or else return a negative value.
1097 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
1098 struct relay_command
*cmd
,
1099 struct lttng_ht
*sessions_ht
)
1101 int ret
= 0, send_ret
;
1102 struct relay_session
*session
;
1103 struct lttcomm_relayd_status_session reply
;
1108 memset(&reply
, 0, sizeof(reply
));
1110 session
= zmalloc(sizeof(struct relay_session
));
1111 if (session
== NULL
) {
1112 PERROR("relay session zmalloc");
1117 session
->id
= ++last_relay_session_id
;
1118 session
->sock
= cmd
->sock
;
1119 session
->minor
= cmd
->minor
;
1120 session
->major
= cmd
->major
;
1121 cmd
->session
= session
;
1123 reply
.session_id
= htobe64(session
->id
);
1125 switch (cmd
->minor
) {
1126 case 4: /* LTTng sessiond 2.4 */
1128 ret
= cmd_create_session_2_4(cmd
, session
);
1132 lttng_ht_node_init_ulong(&session
->session_n
,
1133 (unsigned long) session
->id
);
1134 lttng_ht_add_unique_ulong(sessions_ht
,
1135 &session
->session_n
);
1137 DBG("Created session %" PRIu64
, session
->id
);
1141 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1143 reply
.ret_code
= htobe32(LTTNG_OK
);
1146 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1148 ERR("Relayd sending session id");
1156 * When we have received all the streams and the metadata for a channel,
1157 * we make them visible to the viewer threads.
1160 void set_viewer_ready_flag(struct relay_command
*cmd
)
1162 struct relay_stream_recv_handle
*node
, *tmp_node
;
1164 cds_list_for_each_entry_safe(node
, tmp_node
, &cmd
->recv_head
, node
) {
1165 struct relay_stream
*stream
;
1168 stream
= relay_stream_find_by_id(node
->id
);
1171 * Stream is most probably being cleaned up by the data thread thus
1172 * simply continue to the next one.
1178 stream
->viewer_ready
= 1;
1181 /* Clean stream handle node. */
1182 cds_list_del(&node
->node
);
1190 * Add a recv handle node to the connection recv list with the given stream
1191 * handle. A new node is allocated thus must be freed when the node is deleted
1194 static void queue_stream_handle(uint64_t handle
, struct relay_command
*cmd
)
1196 struct relay_stream_recv_handle
*node
;
1200 node
= zmalloc(sizeof(*node
));
1202 PERROR("zmalloc queue stream handle");
1207 cds_list_add(&node
->node
, &cmd
->recv_head
);
1211 * relay_add_stream: allocate a new stream for a session
1214 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1215 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1217 struct relay_session
*session
= cmd
->session
;
1218 struct relay_stream
*stream
= NULL
;
1219 struct lttcomm_relayd_status_stream reply
;
1222 if (!session
|| cmd
->version_check_done
== 0) {
1223 ERR("Trying to add a stream before version check");
1225 goto end_no_session
;
1228 stream
= zmalloc(sizeof(struct relay_stream
));
1229 if (stream
== NULL
) {
1230 PERROR("relay stream zmalloc");
1232 goto end_no_session
;
1235 switch (cmd
->minor
) {
1236 case 1: /* LTTng sessiond 2.1 */
1237 ret
= cmd_recv_stream_2_1(cmd
, stream
);
1239 case 2: /* LTTng sessiond 2.2 */
1241 ret
= cmd_recv_stream_2_2(cmd
, stream
);
1245 goto err_free_stream
;
1249 stream
->stream_handle
= ++last_relay_stream_id
;
1250 stream
->prev_seq
= -1ULL;
1251 stream
->session
= session
;
1252 stream
->index_fd
= -1;
1253 stream
->read_index_fd
= -1;
1254 stream
->ctf_trace
= NULL
;
1255 pthread_mutex_init(&stream
->lock
, NULL
);
1257 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
);
1259 ERR("relay creating output directory");
1264 * No need to use run_as API here because whatever we receives, the relayd
1265 * uses its own credentials for the stream files.
1267 ret
= utils_create_stream_file(stream
->path_name
, stream
->channel_name
,
1268 stream
->tracefile_size
, 0, relayd_uid
, relayd_gid
, NULL
);
1270 ERR("Create output file");
1274 if (stream
->tracefile_size
) {
1275 DBG("Tracefile %s/%s_0 created", stream
->path_name
, stream
->channel_name
);
1277 DBG("Tracefile %s/%s created", stream
->path_name
, stream
->channel_name
);
1280 if (!strncmp(stream
->channel_name
, DEFAULT_METADATA_NAME
, NAME_MAX
)) {
1281 stream
->metadata_flag
= 1;
1283 * When we receive a new metadata stream, we create a new
1284 * ctf_trace and we assign this ctf_trace to all streams with
1287 * If later on we receive a new stream for the same ctf_trace,
1288 * we copy the information from the first hit in the HT to the
1291 stream
->ctf_trace
= ctf_trace_create();
1292 if (!stream
->ctf_trace
) {
1296 stream
->ctf_trace
->refcount
++;
1297 stream
->ctf_trace
->metadata_stream
= stream
;
1299 ctf_trace_assign(cmd
->ctf_traces_ht
, stream
);
1300 stream
->ctf_traces_ht
= cmd
->ctf_traces_ht
;
1303 * Add the stream handle in the recv list of the connection. Once the end
1304 * stream message is received, this list is emptied and streams are set
1305 * with the viewer ready flag.
1307 if (stream
->metadata_flag
) {
1308 stream
->viewer_ready
= 1;
1310 queue_stream_handle(stream
->stream_handle
, cmd
);
1313 lttng_ht_node_init_ulong(&stream
->stream_n
,
1314 (unsigned long) stream
->stream_handle
);
1315 lttng_ht_add_unique_ulong(relay_streams_ht
,
1318 lttng_ht_node_init_str(&stream
->ctf_trace_node
, stream
->path_name
);
1319 lttng_ht_add_str(cmd
->ctf_traces_ht
, &stream
->ctf_trace_node
);
1320 session
->stream_count
++;
1322 DBG("Relay new stream added %s with ID %" PRIu64
, stream
->channel_name
,
1323 stream
->stream_handle
);
1326 reply
.handle
= htobe64(stream
->stream_handle
);
1327 /* send the session id to the client or a negative return code on error */
1329 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1330 /* stream was not properly added to the ht, so free it */
1333 reply
.ret_code
= htobe32(LTTNG_OK
);
1336 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1337 sizeof(struct lttcomm_relayd_status_stream
), 0);
1339 ERR("Relay sending stream id");
1348 free(stream
->path_name
);
1349 free(stream
->channel_name
);
1355 * relay_close_stream: close a specific stream
1358 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1359 struct relay_command
*cmd
)
1362 struct relay_session
*session
= cmd
->session
;
1363 struct lttcomm_relayd_close_stream stream_info
;
1364 struct lttcomm_relayd_generic_reply reply
;
1365 struct relay_stream
*stream
;
1367 DBG("Close stream received");
1369 if (!session
|| cmd
->version_check_done
== 0) {
1370 ERR("Trying to close a stream before version check");
1372 goto end_no_session
;
1375 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &stream_info
,
1376 sizeof(struct lttcomm_relayd_close_stream
), 0);
1377 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1379 /* Orderly shutdown. Not necessary to print an error. */
1380 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1382 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1385 goto end_no_session
;
1389 stream
= relay_stream_find_by_id(be64toh(stream_info
.stream_id
));
1395 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1396 stream
->close_flag
= 1;
1397 session
->stream_count
--;
1398 assert(session
->stream_count
>= 0);
1400 if (close_stream_check(stream
)) {
1401 destroy_stream(stream
);
1408 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1410 reply
.ret_code
= htobe32(LTTNG_OK
);
1412 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1413 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1415 ERR("Relay sending stream id");
1424 * relay_unknown_command: send -1 if received unknown command
1427 void relay_unknown_command(struct relay_command
*cmd
)
1429 struct lttcomm_relayd_generic_reply reply
;
1432 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1433 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1434 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1436 ERR("Relay sending unknown command");
1441 * relay_start: send an acknowledgment to the client to tell if we are
1442 * ready to receive data. We are ready if a session is established.
1445 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1446 struct relay_command
*cmd
)
1448 int ret
= htobe32(LTTNG_OK
);
1449 struct lttcomm_relayd_generic_reply reply
;
1450 struct relay_session
*session
= cmd
->session
;
1453 DBG("Trying to start the streaming without a session established");
1454 ret
= htobe32(LTTNG_ERR_UNK
);
1457 reply
.ret_code
= ret
;
1458 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1459 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1461 ERR("Relay sending start ack");
1468 * Append padding to the file pointed by the file descriptor fd.
1470 static int write_padding_to_file(int fd
, uint32_t size
)
1479 zeros
= zmalloc(size
);
1480 if (zeros
== NULL
) {
1481 PERROR("zmalloc zeros for padding");
1486 ret
= lttng_write(fd
, zeros
, size
);
1488 PERROR("write padding to file");
1498 * relay_recv_metadata: receive the metada for the session.
1501 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1502 struct relay_command
*cmd
)
1504 int ret
= htobe32(LTTNG_OK
);
1506 struct relay_session
*session
= cmd
->session
;
1507 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1508 struct relay_stream
*metadata_stream
;
1509 uint64_t data_size
, payload_size
;
1512 ERR("Metadata sent before version check");
1517 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1518 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1519 ERR("Incorrect data size");
1523 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1525 if (data_buffer_size
< data_size
) {
1526 /* In case the realloc fails, we can free the memory */
1529 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1530 if (!tmp_data_ptr
) {
1531 ERR("Allocating data buffer");
1536 data_buffer
= tmp_data_ptr
;
1537 data_buffer_size
= data_size
;
1539 memset(data_buffer
, 0, data_size
);
1540 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1541 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
1542 if (ret
< 0 || ret
!= data_size
) {
1544 /* Orderly shutdown. Not necessary to print an error. */
1545 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1547 ERR("Relay didn't receive the whole metadata");
1552 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1555 metadata_stream
= relay_stream_find_by_id(
1556 be64toh(metadata_struct
->stream_id
));
1557 if (!metadata_stream
) {
1562 size_ret
= lttng_write(metadata_stream
->fd
, metadata_struct
->payload
,
1564 if (size_ret
< payload_size
) {
1565 ERR("Relay error writing metadata on file");
1570 ret
= write_padding_to_file(metadata_stream
->fd
,
1571 be32toh(metadata_struct
->padding_size
));
1575 metadata_stream
->ctf_trace
->metadata_received
+=
1576 payload_size
+ be32toh(metadata_struct
->padding_size
);
1578 DBG2("Relay metadata written");
1587 * relay_send_version: send relayd version number
1590 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1591 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1594 struct lttcomm_relayd_version reply
, msg
;
1598 cmd
->version_check_done
= 1;
1600 /* Get version from the other side. */
1601 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1602 if (ret
< 0 || ret
!= sizeof(msg
)) {
1604 /* Orderly shutdown. Not necessary to print an error. */
1605 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1607 ERR("Relay failed to receive the version values.");
1613 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1614 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1616 /* Major versions must be the same */
1617 if (reply
.major
!= be32toh(msg
.major
)) {
1618 DBG("Incompatible major versions (%u vs %u), deleting session",
1619 reply
.major
, be32toh(msg
.major
));
1620 relay_delete_session(cmd
, sessions_ht
);
1625 cmd
->major
= reply
.major
;
1626 /* We adapt to the lowest compatible version */
1627 if (reply
.minor
<= be32toh(msg
.minor
)) {
1628 cmd
->minor
= reply
.minor
;
1630 cmd
->minor
= be32toh(msg
.minor
);
1633 reply
.major
= htobe32(reply
.major
);
1634 reply
.minor
= htobe32(reply
.minor
);
1635 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1636 sizeof(struct lttcomm_relayd_version
), 0);
1638 ERR("Relay sending version");
1641 DBG("Version check done using protocol %u.%u", cmd
->major
,
1649 * Check for data pending for a given stream id from the session daemon.
1652 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1653 struct relay_command
*cmd
)
1655 struct relay_session
*session
= cmd
->session
;
1656 struct lttcomm_relayd_data_pending msg
;
1657 struct lttcomm_relayd_generic_reply reply
;
1658 struct relay_stream
*stream
;
1660 uint64_t last_net_seq_num
, stream_id
;
1662 DBG("Data pending command received");
1664 if (!session
|| cmd
->version_check_done
== 0) {
1665 ERR("Trying to check for data before version check");
1667 goto end_no_session
;
1670 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1671 if (ret
< sizeof(msg
)) {
1673 /* Orderly shutdown. Not necessary to print an error. */
1674 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1676 ERR("Relay didn't receive valid data_pending struct size : %d",
1680 goto end_no_session
;
1683 stream_id
= be64toh(msg
.stream_id
);
1684 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1687 stream
= relay_stream_find_by_id(stream_id
);
1688 if (stream
== NULL
) {
1693 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1694 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1697 /* Avoid wrapping issue */
1698 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1699 /* Data has in fact been written and is NOT pending */
1702 /* Data still being streamed thus pending */
1706 /* Pending check is now done. */
1707 stream
->data_pending_check_done
= 1;
1712 reply
.ret_code
= htobe32(ret
);
1713 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1715 ERR("Relay data pending ret code failed");
1723 * Wait for the control socket to reach a quiescent state.
1725 * Note that for now, when receiving this command from the session daemon, this
1726 * means that every subsequent commands or data received on the control socket
1727 * has been handled. So, this is why we simply return OK here.
1730 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1731 struct relay_command
*cmd
)
1735 struct relay_stream
*stream
;
1736 struct lttng_ht_iter iter
;
1737 struct lttcomm_relayd_quiescent_control msg
;
1738 struct lttcomm_relayd_generic_reply reply
;
1740 DBG("Checking quiescent state on control socket");
1742 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1743 ERR("Trying to check for data before version check");
1745 goto end_no_session
;
1748 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1749 if (ret
< sizeof(msg
)) {
1751 /* Orderly shutdown. Not necessary to print an error. */
1752 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1754 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1758 goto end_no_session
;
1761 stream_id
= be64toh(msg
.stream_id
);
1764 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1766 if (stream
->stream_handle
== stream_id
) {
1767 stream
->data_pending_check_done
= 1;
1768 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1775 reply
.ret_code
= htobe32(LTTNG_OK
);
1776 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1778 ERR("Relay data quiescent control ret code failed");
1786 * Initialize a data pending command. This means that a client is about to ask
1787 * for data pending for each stream he/she holds. Simply iterate over all
1788 * streams of a session and set the data_pending_check_done flag.
1790 * This command returns to the client a LTTNG_OK code.
1793 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1794 struct relay_command
*cmd
)
1797 struct lttng_ht_iter iter
;
1798 struct lttcomm_relayd_begin_data_pending msg
;
1799 struct lttcomm_relayd_generic_reply reply
;
1800 struct relay_stream
*stream
;
1801 uint64_t session_id
;
1806 DBG("Init streams for data pending");
1808 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1809 ERR("Trying to check for data before version check");
1811 goto end_no_session
;
1814 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1815 if (ret
< sizeof(msg
)) {
1817 /* Orderly shutdown. Not necessary to print an error. */
1818 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1820 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1824 goto end_no_session
;
1827 session_id
= be64toh(msg
.session_id
);
1830 * Iterate over all streams to set the begin data pending flag. For now, the
1831 * streams are indexed by stream handle so we have to iterate over all
1832 * streams to find the one associated with the right session_id.
1835 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1837 if (stream
->session
->id
== session_id
) {
1838 stream
->data_pending_check_done
= 0;
1839 DBG("Set begin data pending flag to stream %" PRIu64
,
1840 stream
->stream_handle
);
1845 /* All good, send back reply. */
1846 reply
.ret_code
= htobe32(LTTNG_OK
);
1848 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1850 ERR("Relay begin data pending send reply failed");
1858 * End data pending command. This will check, for a given session id, if each
1859 * stream associated with it has its data_pending_check_done flag set. If not,
1860 * this means that the client lost track of the stream but the data is still
1861 * being streamed on our side. In this case, we inform the client that data is
1864 * Return to the client if there is data in flight or not with a ret_code.
1867 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1868 struct relay_command
*cmd
)
1871 struct lttng_ht_iter iter
;
1872 struct lttcomm_relayd_end_data_pending msg
;
1873 struct lttcomm_relayd_generic_reply reply
;
1874 struct relay_stream
*stream
;
1875 uint64_t session_id
;
1876 uint32_t is_data_inflight
= 0;
1881 DBG("End data pending command");
1883 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1884 ERR("Trying to check for data before version check");
1886 goto end_no_session
;
1889 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1890 if (ret
< sizeof(msg
)) {
1892 /* Orderly shutdown. Not necessary to print an error. */
1893 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1895 ERR("Relay didn't receive valid end data_pending struct size: %d",
1899 goto end_no_session
;
1902 session_id
= be64toh(msg
.session_id
);
1904 /* Iterate over all streams to see if the begin data pending flag is set. */
1906 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1908 if (stream
->session
->id
== session_id
&&
1909 !stream
->data_pending_check_done
) {
1910 is_data_inflight
= 1;
1911 DBG("Data is still in flight for stream %" PRIu64
,
1912 stream
->stream_handle
);
1918 /* All good, send back reply. */
1919 reply
.ret_code
= htobe32(is_data_inflight
);
1921 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1923 ERR("Relay end data pending send reply failed");
1931 * Receive an index for a specific stream.
1933 * Return 0 on success else a negative value.
1936 int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1937 struct relay_command
*cmd
)
1939 int ret
, send_ret
, index_created
= 0;
1940 struct relay_session
*session
= cmd
->session
;
1941 struct lttcomm_relayd_index index_info
;
1942 struct relay_index
*index
, *wr_index
= NULL
;
1943 struct lttcomm_relayd_generic_reply reply
;
1944 struct relay_stream
*stream
;
1945 uint64_t net_seq_num
;
1949 DBG("Relay receiving index");
1951 if (!session
|| cmd
->version_check_done
== 0) {
1952 ERR("Trying to close a stream before version check");
1954 goto end_no_session
;
1957 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &index_info
,
1958 sizeof(index_info
), 0);
1959 if (ret
< sizeof(index_info
)) {
1961 /* Orderly shutdown. Not necessary to print an error. */
1962 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1964 ERR("Relay didn't receive valid index struct size : %d", ret
);
1967 goto end_no_session
;
1970 net_seq_num
= be64toh(index_info
.net_seq_num
);
1973 stream
= relay_stream_find_by_id(be64toh(index_info
.relay_stream_id
));
1976 goto end_rcu_unlock
;
1979 /* Live beacon handling */
1980 if (index_info
.packet_size
== 0) {
1981 DBG("Received live beacon for stream %" PRIu64
, stream
->stream_handle
);
1984 * Only flag a stream inactive when it has already received data.
1986 if (stream
->total_index_received
> 0) {
1987 stream
->beacon_ts_end
= be64toh(index_info
.timestamp_end
);
1990 goto end_rcu_unlock
;
1992 stream
->beacon_ts_end
= -1ULL;
1995 index
= relay_index_find(stream
->stream_handle
, net_seq_num
);
1997 /* A successful creation will add the object to the HT. */
1998 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
2000 goto end_rcu_unlock
;
2005 copy_index_control_data(index
, &index_info
);
2007 if (index_created
) {
2009 * Try to add the relay index object to the hash table. If an object
2010 * already exist, destroy back the index created, set the data in this
2011 * object and write it on disk.
2013 relay_index_add(index
, &wr_index
);
2015 copy_index_control_data(wr_index
, &index_info
);
2019 /* The index already exists so write it on disk. */
2023 /* Do we have a writable ready index to write on disk. */
2025 /* Starting at 2.4, create the index file if none available. */
2026 if (cmd
->minor
>= 4 && stream
->index_fd
< 0) {
2027 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
2028 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
2029 stream
->tracefile_count_current
);
2031 goto end_rcu_unlock
;
2033 stream
->index_fd
= ret
;
2036 ret
= relay_index_write(wr_index
->fd
, wr_index
);
2038 goto end_rcu_unlock
;
2040 stream
->total_index_received
++;
2047 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2049 reply
.ret_code
= htobe32(LTTNG_OK
);
2051 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
2053 ERR("Relay sending close index id reply");
2062 * Receive the streams_sent message.
2064 * Return 0 on success else a negative value.
2067 int relay_streams_sent(struct lttcomm_relayd_hdr
*recv_hdr
,
2068 struct relay_command
*cmd
)
2071 struct lttcomm_relayd_generic_reply reply
;
2075 DBG("Relay receiving streams_sent");
2077 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
2078 ERR("Trying to close a stream before version check");
2080 goto end_no_session
;
2084 * Flag every pending stream in the connection recv list that they are
2085 * ready to be used by the viewer.
2087 set_viewer_ready_flag(cmd
);
2089 reply
.ret_code
= htobe32(LTTNG_OK
);
2090 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
2092 ERR("Relay sending sent_stream reply");
2104 * Process the commands received on the control socket
2107 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
2108 struct relay_command
*cmd
, struct relay_local_data
*ctx
)
2112 switch (be32toh(recv_hdr
->cmd
)) {
2113 case RELAYD_CREATE_SESSION
:
2114 ret
= relay_create_session(recv_hdr
, cmd
, ctx
->sessions_ht
);
2116 case RELAYD_ADD_STREAM
:
2117 ret
= relay_add_stream(recv_hdr
, cmd
, ctx
->sessions_ht
);
2119 case RELAYD_START_DATA
:
2120 ret
= relay_start(recv_hdr
, cmd
);
2122 case RELAYD_SEND_METADATA
:
2123 ret
= relay_recv_metadata(recv_hdr
, cmd
);
2125 case RELAYD_VERSION
:
2126 ret
= relay_send_version(recv_hdr
, cmd
, ctx
->sessions_ht
);
2128 case RELAYD_CLOSE_STREAM
:
2129 ret
= relay_close_stream(recv_hdr
, cmd
);
2131 case RELAYD_DATA_PENDING
:
2132 ret
= relay_data_pending(recv_hdr
, cmd
);
2134 case RELAYD_QUIESCENT_CONTROL
:
2135 ret
= relay_quiescent_control(recv_hdr
, cmd
);
2137 case RELAYD_BEGIN_DATA_PENDING
:
2138 ret
= relay_begin_data_pending(recv_hdr
, cmd
);
2140 case RELAYD_END_DATA_PENDING
:
2141 ret
= relay_end_data_pending(recv_hdr
, cmd
);
2143 case RELAYD_SEND_INDEX
:
2144 ret
= relay_recv_index(recv_hdr
, cmd
);
2146 case RELAYD_STREAMS_SENT
:
2147 ret
= relay_streams_sent(recv_hdr
, cmd
);
2149 case RELAYD_UPDATE_SYNC_INFO
:
2151 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
2152 relay_unknown_command(cmd
);
2162 * Handle index for a data stream.
2164 * RCU read side lock MUST be acquired.
2166 * Return 0 on success else a negative value.
2168 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2171 int ret
= 0, index_created
= 0;
2172 uint64_t stream_id
, data_offset
;
2173 struct relay_index
*index
, *wr_index
= NULL
;
2177 stream_id
= stream
->stream_handle
;
2178 /* Get data offset because we are about to update the index. */
2179 data_offset
= htobe64(stream
->tracefile_size_current
);
2182 * Lookup for an existing index for that stream id/sequence number. If on
2183 * exists, the control thread already received the data for it thus we need
2184 * to write it on disk.
2186 index
= relay_index_find(stream_id
, net_seq_num
);
2188 /* A successful creation will add the object to the HT. */
2189 index
= relay_index_create(stream_id
, net_seq_num
);
2197 if (rotate_index
|| stream
->index_fd
< 0) {
2198 index
->to_close_fd
= stream
->index_fd
;
2199 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
2200 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
2201 stream
->tracefile_count_current
);
2203 /* This will close the stream's index fd if one. */
2204 relay_index_free_safe(index
);
2207 stream
->index_fd
= ret
;
2209 index
->fd
= stream
->index_fd
;
2210 index
->index_data
.offset
= data_offset
;
2212 if (index_created
) {
2214 * Try to add the relay index object to the hash table. If an object
2215 * already exist, destroy back the index created and set the data.
2217 relay_index_add(index
, &wr_index
);
2219 /* Copy back data from the created index. */
2220 wr_index
->fd
= index
->fd
;
2221 wr_index
->to_close_fd
= index
->to_close_fd
;
2222 wr_index
->index_data
.offset
= data_offset
;
2226 /* The index already exists so write it on disk. */
2230 /* Do we have a writable ready index to write on disk. */
2232 ret
= relay_index_write(wr_index
->fd
, wr_index
);
2236 stream
->total_index_received
++;
2244 * relay_process_data: Process the data received on the data socket
2247 int relay_process_data(struct relay_command
*cmd
)
2249 int ret
= 0, rotate_index
= 0;
2251 struct relay_stream
*stream
;
2252 struct lttcomm_relayd_data_hdr data_hdr
;
2254 uint64_t net_seq_num
;
2257 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &data_hdr
,
2258 sizeof(struct lttcomm_relayd_data_hdr
), 0);
2261 /* Orderly shutdown. Not necessary to print an error. */
2262 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
2264 ERR("Unable to receive data header on sock %d", cmd
->sock
->fd
);
2270 stream_id
= be64toh(data_hdr
.stream_id
);
2273 stream
= relay_stream_find_by_id(stream_id
);
2276 goto end_rcu_unlock
;
2279 data_size
= be32toh(data_hdr
.data_size
);
2280 if (data_buffer_size
< data_size
) {
2283 tmp_data_ptr
= realloc(data_buffer
, data_size
);
2284 if (!tmp_data_ptr
) {
2285 ERR("Allocating data buffer");
2288 goto end_rcu_unlock
;
2290 data_buffer
= tmp_data_ptr
;
2291 data_buffer_size
= data_size
;
2293 memset(data_buffer
, 0, data_size
);
2295 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2297 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2298 data_size
, stream_id
, net_seq_num
);
2299 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
2302 /* Orderly shutdown. Not necessary to print an error. */
2303 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
2306 goto end_rcu_unlock
;
2309 /* Check if a rotation is needed. */
2310 if (stream
->tracefile_size
> 0 &&
2311 (stream
->tracefile_size_current
+ data_size
) >
2312 stream
->tracefile_size
) {
2313 struct relay_viewer_stream
*vstream
;
2316 new_id
= (stream
->tracefile_count_current
+ 1) %
2317 stream
->tracefile_count
;
2319 * When we wrap-around back to 0, we start overwriting old
2322 if (!stream
->tracefile_overwrite
&& new_id
== 0) {
2323 stream
->tracefile_overwrite
= 1;
2325 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
2326 if (stream
->tracefile_overwrite
) {
2327 stream
->oldest_tracefile_id
=
2328 (stream
->oldest_tracefile_id
+ 1) %
2329 stream
->tracefile_count
;
2331 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
);
2334 * The viewer is reading a file about to be
2335 * overwritten. Close the FDs it is
2336 * currently using and let it handle the fault.
2338 if (vstream
->tracefile_count_current
== new_id
) {
2339 pthread_mutex_lock(&vstream
->overwrite_lock
);
2340 vstream
->abort_flag
= 1;
2341 pthread_mutex_unlock(&vstream
->overwrite_lock
);
2342 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2343 stream
->channel_name
, new_id
);
2344 } else if (vstream
->tracefile_count_current
==
2345 stream
->tracefile_count_current
) {
2347 * The reader and writer were in the
2348 * same trace file, inform the viewer
2349 * that no new index will ever be added
2352 vstream
->close_write_flag
= 1;
2355 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
2356 stream
->tracefile_size
, stream
->tracefile_count
,
2357 relayd_uid
, relayd_gid
, stream
->fd
,
2358 &(stream
->tracefile_count_current
), &stream
->fd
);
2359 stream
->total_index_received
= 0;
2360 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
2362 ERR("Rotating stream output file");
2363 goto end_rcu_unlock
;
2365 /* Reset current size because we just perform a stream rotation. */
2366 stream
->tracefile_size_current
= 0;
2371 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2372 * index are NOT supported.
2374 if (stream
->session
->minor
>= 4 && !stream
->session
->snapshot
) {
2375 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2377 goto end_rcu_unlock
;
2381 /* Write data to stream output fd. */
2382 size_ret
= lttng_write(stream
->fd
, data_buffer
, data_size
);
2383 if (size_ret
< data_size
) {
2384 ERR("Relay error writing data to file");
2386 goto end_rcu_unlock
;
2389 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
2390 ret
, stream
->stream_handle
);
2392 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
2394 goto end_rcu_unlock
;
2396 stream
->tracefile_size_current
+= data_size
+ be32toh(data_hdr
.padding_size
);
2398 stream
->prev_seq
= net_seq_num
;
2400 /* Check if we need to close the FD */
2401 if (close_stream_check(stream
)) {
2402 destroy_stream(stream
);
2412 void relay_cleanup_poll_connection(struct lttng_poll_event
*events
, int pollfd
)
2416 lttng_poll_del(events
, pollfd
);
2418 ret
= close(pollfd
);
2420 ERR("Closing pollfd %d", pollfd
);
2425 int relay_add_connection(int fd
, struct lttng_poll_event
*events
,
2426 struct lttng_ht
*relay_connections_ht
)
2428 struct relay_command
*relay_connection
;
2431 relay_connection
= zmalloc(sizeof(struct relay_command
));
2432 if (relay_connection
== NULL
) {
2433 PERROR("Relay command zmalloc");
2436 ret
= lttng_read(fd
, relay_connection
, sizeof(struct relay_command
));
2437 if (ret
< sizeof(struct relay_command
)) {
2438 PERROR("read relay cmd pipe");
2441 CDS_INIT_LIST_HEAD(&relay_connection
->recv_head
);
2444 * Only used by the control side and the reference is copied inside each
2445 * stream from that connection. Thus a destroy HT must be done after every
2446 * stream has been destroyed.
2448 if (relay_connection
->type
== RELAY_CONTROL
) {
2449 relay_connection
->ctf_traces_ht
= lttng_ht_new(0,
2450 LTTNG_HT_TYPE_STRING
);
2451 if (!relay_connection
->ctf_traces_ht
) {
2456 lttng_ht_node_init_ulong(&relay_connection
->sock_n
,
2457 (unsigned long) relay_connection
->sock
->fd
);
2459 lttng_ht_add_unique_ulong(relay_connections_ht
,
2460 &relay_connection
->sock_n
);
2462 return lttng_poll_add(events
,
2463 relay_connection
->sock
->fd
,
2464 LPOLLIN
| LPOLLRDHUP
);
2467 free(relay_connection
);
2473 void deferred_free_connection(struct rcu_head
*head
)
2475 struct relay_command
*relay_connection
=
2476 caa_container_of(head
, struct relay_command
, rcu_node
);
2478 lttcomm_destroy_sock(relay_connection
->sock
);
2479 free(relay_connection
);
2483 void relay_del_connection(struct lttng_ht
*relay_connections_ht
,
2484 struct lttng_ht_iter
*iter
, struct relay_command
*relay_connection
,
2485 struct lttng_ht
*sessions_ht
)
2489 ret
= lttng_ht_del(relay_connections_ht
, iter
);
2492 if (relay_connection
->type
== RELAY_CONTROL
) {
2493 struct relay_stream_recv_handle
*node
, *tmp_node
;
2495 relay_delete_session(relay_connection
, sessions_ht
);
2496 lttng_ht_destroy(relay_connection
->ctf_traces_ht
);
2498 /* Clean up recv list. */
2499 cds_list_for_each_entry_safe(node
, tmp_node
,
2500 &relay_connection
->recv_head
, node
) {
2501 cds_list_del(&node
->node
);
2506 call_rcu(&relay_connection
->rcu_node
, deferred_free_connection
);
2510 * This thread does the actual work
2513 void *relay_thread_worker(void *data
)
2515 int ret
, err
= -1, last_seen_data_fd
= -1;
2517 struct relay_command
*relay_connection
;
2518 struct lttng_poll_event events
;
2519 struct lttng_ht
*relay_connections_ht
;
2520 struct lttng_ht_node_ulong
*node
;
2521 struct lttng_ht_iter iter
;
2522 struct lttcomm_relayd_hdr recv_hdr
;
2523 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
2524 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
2526 DBG("[thread] Relay worker started");
2528 rcu_register_thread();
2530 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2532 health_code_update();
2534 /* table of connections indexed on socket */
2535 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2536 if (!relay_connections_ht
) {
2537 goto relay_connections_ht_error
;
2540 /* Tables of received indexes indexed by index handle and net_seq_num. */
2541 indexes_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64
);
2543 goto indexes_ht_error
;
2546 ret
= create_thread_poll_set(&events
, 2);
2548 goto error_poll_create
;
2551 ret
= lttng_poll_add(&events
, relay_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2558 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2560 health_code_update();
2562 /* Infinite blocking call, waiting for transmission */
2563 DBG3("Relayd worker thread polling...");
2564 health_poll_entry();
2565 ret
= lttng_poll_wait(&events
, -1);
2569 * Restart interrupted system call.
2571 if (errno
== EINTR
) {
2580 * Process control. The control connection is prioritised so we don't
2581 * starve it with high throughout put tracing data on the data
2584 for (i
= 0; i
< nb_fd
; i
++) {
2585 /* Fetch once the poll data */
2586 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2587 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2589 health_code_update();
2591 /* Thread quit pipe has been closed. Killing thread. */
2592 ret
= check_thread_quit_pipe(pollfd
, revents
);
2598 /* Inspect the relay cmd pipe for new connection */
2599 if (pollfd
== relay_cmd_pipe
[0]) {
2600 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2601 ERR("Relay pipe error");
2603 } else if (revents
& LPOLLIN
) {
2604 DBG("Relay command received");
2605 ret
= relay_add_connection(relay_cmd_pipe
[0],
2606 &events
, relay_connections_ht
);
2611 } else if (revents
) {
2613 lttng_ht_lookup(relay_connections_ht
,
2614 (void *)((unsigned long) pollfd
),
2616 node
= lttng_ht_iter_get_node_ulong(&iter
);
2618 DBG2("Relay sock %d not found", pollfd
);
2622 relay_connection
= caa_container_of(node
,
2623 struct relay_command
, sock_n
);
2625 if (revents
& (LPOLLERR
)) {
2627 relay_cleanup_poll_connection(&events
, pollfd
);
2628 relay_del_connection(relay_connections_ht
,
2629 &iter
, relay_connection
, sessions_ht
);
2630 if (last_seen_data_fd
== pollfd
) {
2631 last_seen_data_fd
= last_notdel_data_fd
;
2633 } else if (revents
& (LPOLLHUP
| LPOLLRDHUP
)) {
2634 DBG("Socket %d hung up", pollfd
);
2635 relay_cleanup_poll_connection(&events
, pollfd
);
2636 relay_del_connection(relay_connections_ht
,
2637 &iter
, relay_connection
, sessions_ht
);
2638 if (last_seen_data_fd
== pollfd
) {
2639 last_seen_data_fd
= last_notdel_data_fd
;
2641 } else if (revents
& LPOLLIN
) {
2642 /* control socket */
2643 if (relay_connection
->type
== RELAY_CONTROL
) {
2644 ret
= relay_connection
->sock
->ops
->recvmsg(
2645 relay_connection
->sock
, &recv_hdr
,
2646 sizeof(struct lttcomm_relayd_hdr
), 0);
2647 /* connection closed */
2649 relay_cleanup_poll_connection(&events
, pollfd
);
2650 relay_del_connection(relay_connections_ht
,
2651 &iter
, relay_connection
, sessions_ht
);
2652 DBG("Control connection closed with %d", pollfd
);
2654 if (relay_connection
->session
) {
2655 DBG2("Relay worker receiving data for session : %" PRIu64
,
2656 relay_connection
->session
->id
);
2658 ret
= relay_process_control(&recv_hdr
,
2659 relay_connection
, relay_ctx
);
2661 /* Clear the session on error. */
2662 relay_cleanup_poll_connection(&events
, pollfd
);
2663 relay_del_connection(relay_connections_ht
,
2664 &iter
, relay_connection
, sessions_ht
);
2665 DBG("Connection closed with %d", pollfd
);
2671 * Flag the last seen data fd not deleted. It will be
2672 * used as the last seen fd if any fd gets deleted in
2675 last_notdel_data_fd
= pollfd
;
2683 * The last loop handled a control request, go back to poll to make
2684 * sure we prioritise the control socket.
2690 if (last_seen_data_fd
>= 0) {
2691 for (i
= 0; i
< nb_fd
; i
++) {
2692 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2694 health_code_update();
2696 if (last_seen_data_fd
== pollfd
) {
2703 /* Process data connection. */
2704 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2705 /* Fetch the poll data. */
2706 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2707 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2709 health_code_update();
2711 /* Skip the command pipe. It's handled in the first loop. */
2712 if (pollfd
== relay_cmd_pipe
[0]) {
2718 lttng_ht_lookup(relay_connections_ht
,
2719 (void *)((unsigned long) pollfd
),
2721 node
= lttng_ht_iter_get_node_ulong(&iter
);
2723 /* Skip it. Might be removed before. */
2727 relay_connection
= caa_container_of(node
,
2728 struct relay_command
, sock_n
);
2730 if (revents
& LPOLLIN
) {
2731 if (relay_connection
->type
!= RELAY_DATA
) {
2735 ret
= relay_process_data(relay_connection
);
2736 /* connection closed */
2738 relay_cleanup_poll_connection(&events
, pollfd
);
2739 relay_del_connection(relay_connections_ht
,
2740 &iter
, relay_connection
, sessions_ht
);
2741 DBG("Data connection closed with %d", pollfd
);
2743 * Every goto restart call sets the last seen fd where
2744 * here we don't really care since we gracefully
2745 * continue the loop after the connection is deleted.
2748 /* Keep last seen port. */
2749 last_seen_data_fd
= pollfd
;
2757 last_seen_data_fd
= -1;
2760 /* Normal exit, no error */
2765 lttng_poll_clean(&events
);
2767 /* empty the hash table and free the memory */
2769 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, node
, node
) {
2770 health_code_update();
2772 node
= lttng_ht_iter_get_node_ulong(&iter
);
2774 relay_connection
= caa_container_of(node
,
2775 struct relay_command
, sock_n
);
2776 relay_del_connection(relay_connections_ht
,
2777 &iter
, relay_connection
, sessions_ht
);
2782 lttng_ht_destroy(indexes_ht
);
2784 lttng_ht_destroy(relay_connections_ht
);
2785 relay_connections_ht_error
:
2786 /* Close relay cmd pipes */
2787 utils_close_pipe(relay_cmd_pipe
);
2789 DBG("Thread exited with error");
2791 DBG("Worker thread cleanup complete");
2795 ERR("Health error occurred in %s", __func__
);
2797 health_unregister(health_relayd
);
2798 rcu_unregister_thread();
2804 * Create the relay command pipe to wake thread_manage_apps.
2805 * Closed in cleanup().
2807 static int create_relay_cmd_pipe(void)
2811 ret
= utils_create_pipe_cloexec(relay_cmd_pipe
);
2819 int main(int argc
, char **argv
)
2823 struct relay_local_data
*relay_ctx
;
2825 /* Create thread quit pipe */
2826 if ((ret
= init_thread_quit_pipe()) < 0) {
2830 /* Parse arguments */
2832 if ((ret
= set_options(argc
, argv
)) < 0) {
2836 if ((ret
= set_signal_handler()) < 0) {
2840 /* Try to create directory if -o, --output is specified. */
2841 if (opt_output_path
) {
2842 if (*opt_output_path
!= '/') {
2843 ERR("Please specify an absolute path for -o, --output PATH");
2847 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2849 ERR("Unable to create %s", opt_output_path
);
2855 if (opt_daemon
|| opt_background
) {
2856 ret
= daemon(0, opt_background
);
2863 /* We need those values for the file/dir creation. */
2864 relayd_uid
= getuid();
2865 relayd_gid
= getgid();
2867 /* Check if daemon is UID = 0 */
2868 if (relayd_uid
== 0) {
2869 if (control_uri
->port
< 1024 || data_uri
->port
< 1024 || live_uri
->port
< 1024) {
2870 ERR("Need to be root to use ports < 1024");
2876 /* Setup the thread apps communication pipe. */
2877 if ((ret
= create_relay_cmd_pipe()) < 0) {
2881 /* Init relay command queue. */
2882 cds_wfq_init(&relay_cmd_queue
.queue
);
2884 /* Set up max poll set size */
2885 lttng_poll_set_max_size();
2887 /* Initialize communication library */
2889 lttcomm_inet_init();
2891 relay_ctx
= zmalloc(sizeof(struct relay_local_data
));
2893 PERROR("relay_ctx");
2897 /* tables of sessions indexed by session ID */
2898 relay_ctx
->sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2899 if (!relay_ctx
->sessions_ht
) {
2900 goto exit_relay_ctx_sessions
;
2903 /* tables of streams indexed by stream ID */
2904 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2905 if (!relay_streams_ht
) {
2906 goto exit_relay_ctx_streams
;
2909 /* tables of streams indexed by stream ID */
2910 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2911 if (!viewer_streams_ht
) {
2912 goto exit_relay_ctx_viewer_streams
;
2915 /* Initialize thread health monitoring */
2916 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
2917 if (!health_relayd
) {
2918 PERROR("health_app_create error");
2919 goto exit_health_app_create
;
2922 ret
= utils_create_pipe(health_quit_pipe
);
2924 goto error_health_pipe
;
2927 /* Create thread to manage the client socket */
2928 ret
= pthread_create(&health_thread
, NULL
,
2929 thread_manage_health
, (void *) NULL
);
2931 PERROR("pthread_create health");
2935 /* Setup the dispatcher thread */
2936 ret
= pthread_create(&dispatcher_thread
, NULL
,
2937 relay_thread_dispatcher
, (void *) NULL
);
2939 PERROR("pthread_create dispatcher");
2940 goto exit_dispatcher
;
2943 /* Setup the worker thread */
2944 ret
= pthread_create(&worker_thread
, NULL
,
2945 relay_thread_worker
, (void *) relay_ctx
);
2947 PERROR("pthread_create worker");
2951 /* Setup the listener thread */
2952 ret
= pthread_create(&listener_thread
, NULL
,
2953 relay_thread_listener
, (void *) NULL
);
2955 PERROR("pthread_create listener");
2959 ret
= live_start_threads(live_uri
, relay_ctx
, thread_quit_pipe
);
2961 ERR("Starting live viewer threads");
2966 ret
= pthread_join(listener_thread
, &status
);
2968 PERROR("pthread_join");
2969 goto error
; /* join error, exit without cleanup */
2973 ret
= pthread_join(worker_thread
, &status
);
2975 PERROR("pthread_join");
2976 goto error
; /* join error, exit without cleanup */
2980 ret
= pthread_join(dispatcher_thread
, &status
);
2982 PERROR("pthread_join");
2983 goto error
; /* join error, exit without cleanup */
2987 ret
= pthread_join(health_thread
, &status
);
2989 PERROR("pthread_join health thread");
2990 goto error
; /* join error, exit without cleanup */
2994 * Stop live threads only after joining other threads.
2996 live_stop_threads();
2999 utils_close_pipe(health_quit_pipe
);
3002 health_app_destroy(health_relayd
);
3004 exit_health_app_create
:
3005 lttng_ht_destroy(viewer_streams_ht
);
3007 exit_relay_ctx_viewer_streams
:
3008 lttng_ht_destroy(relay_streams_ht
);
3010 exit_relay_ctx_streams
:
3011 lttng_ht_destroy(relay_ctx
->sessions_ht
);
3013 exit_relay_ctx_sessions
: