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
;
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 { "group", 1, 0, 'g', },
135 { "help", 0, 0, 'h', },
136 { "output", 1, 0, 'o', },
137 { "verbose", 0, 0, 'v', },
138 { "config", 1, 0, 'f' },
142 static const char *config_ignore_options
[] = { "help", "config" };
145 * usage function on stderr
150 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
151 fprintf(stderr
, " -h, --help Display this usage.\n");
152 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
153 fprintf(stderr
, " -C, --control-port URL Control port listening.\n");
154 fprintf(stderr
, " -D, --data-port URL Data port listening.\n");
155 fprintf(stderr
, " -L, --live-port URL Live view port listening.\n");
156 fprintf(stderr
, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
157 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
158 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
159 fprintf(stderr
, " -f --config Load daemon configuration file\n");
163 * Take an option from the getopt output and set it in the right variable to be
166 * Return 0 on success else a negative value.
169 int set_option(int opt
, const char *arg
, const char *optname
)
175 fprintf(stderr
, "option %s", optname
);
177 fprintf(stderr
, " with arg %s\n", arg
);
181 ret
= uri_parse(arg
, &control_uri
);
183 ERR("Invalid control URI specified");
186 if (control_uri
->port
== 0) {
187 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
191 ret
= uri_parse(arg
, &data_uri
);
193 ERR("Invalid data URI specified");
196 if (data_uri
->port
== 0) {
197 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
201 ret
= uri_parse(arg
, &live_uri
);
203 ERR("Invalid live URI specified");
206 if (live_uri
->port
== 0) {
207 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
214 tracing_group_name
= strdup(arg
);
215 tracing_group_name_override
= 1;
221 ret
= asprintf(&opt_output_path
, "%s", arg
);
224 PERROR("asprintf opt_output_path");
229 /* Verbose level can increase using multiple -v */
231 lttng_opt_verbose
= config_parse_value(arg
);
233 lttng_opt_verbose
+= 1;
237 /* Unknown option or other error.
238 * Error is printed by getopt, just return */
251 * config_entry_handler_cb used to handle options read from a config file.
252 * See config_entry_handler_cb comment in common/config/config.h for the
253 * return value conventions.
256 int config_entry_handler(const struct config_entry
*entry
, void *unused
)
260 if (!entry
|| !entry
->name
|| !entry
->value
) {
265 /* Check if the option is to be ignored */
266 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
267 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
272 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
273 /* Ignore if entry name is not fully matched. */
274 if (strcmp(entry
->name
, long_options
[i
].name
)) {
279 * If the option takes no argument on the command line, we have to
280 * check if the value is "true". We support non-zero numeric values,
283 if (!long_options
[i
].has_arg
) {
284 ret
= config_parse_value(entry
->value
);
287 WARN("Invalid configuration value \"%s\" for option %s",
288 entry
->value
, entry
->name
);
290 /* False, skip boolean config option. */
295 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
299 WARN("Unrecognized option \"%s\" in daemon configuration file.",
307 int set_options(int argc
, char **argv
)
309 int c
, ret
= 0, option_index
= 0;
310 int orig_optopt
= optopt
, orig_optind
= optind
;
311 char *default_address
, *optstring
;
312 const char *config_path
= NULL
;
314 optstring
= utils_generate_optstring(long_options
,
315 sizeof(long_options
) / sizeof(struct option
));
321 /* Check for the --config option */
323 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
324 &option_index
)) != -1) {
328 } else if (c
!= 'f') {
332 config_path
= utils_expand_path(optarg
);
334 ERR("Failed to resolve path: %s", optarg
);
338 ret
= config_get_section_entries(config_path
, config_section_name
,
339 config_entry_handler
, NULL
);
342 ERR("Invalid configuration option at line %i", ret
);
348 /* Reset getopt's global state */
349 optopt
= orig_optopt
;
350 optind
= orig_optind
;
352 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
357 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
363 /* assign default values */
364 if (control_uri
== NULL
) {
365 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
366 DEFAULT_NETWORK_CONTROL_PORT
);
368 PERROR("asprintf default data address");
372 ret
= uri_parse(default_address
, &control_uri
);
373 free(default_address
);
375 ERR("Invalid control URI specified");
379 if (data_uri
== NULL
) {
380 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
381 DEFAULT_NETWORK_DATA_PORT
);
383 PERROR("asprintf default data address");
387 ret
= uri_parse(default_address
, &data_uri
);
388 free(default_address
);
390 ERR("Invalid data URI specified");
394 if (live_uri
== NULL
) {
395 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
396 DEFAULT_NETWORK_VIEWER_PORT
);
398 PERROR("asprintf default viewer control address");
402 ret
= uri_parse(default_address
, &live_uri
);
403 free(default_address
);
405 ERR("Invalid viewer control URI specified");
423 /* free the dynamically allocated opt_output_path */
424 free(opt_output_path
);
426 /* Close thread quit pipes */
427 utils_close_pipe(thread_quit_pipe
);
429 uri_free(control_uri
);
431 /* Live URI is freed in the live thread. */
433 if (tracing_group_name_override
) {
434 free((void *) tracing_group_name
);
439 * Write to writable pipe used to notify a thread.
442 int notify_thread_pipe(int wpipe
)
446 ret
= lttng_write(wpipe
, "!", 1);
448 PERROR("write poll pipe");
454 static void notify_health_quit_pipe(int *pipe
)
458 ret
= lttng_write(pipe
[1], "4", 1);
460 PERROR("write relay health quit");
465 * Stop all threads by closing the thread quit pipe.
468 void stop_threads(void)
472 /* Stopping all threads */
473 DBG("Terminating all threads");
474 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
476 ERR("write error on thread quit pipe");
479 notify_health_quit_pipe(health_quit_pipe
);
481 /* Dispatch thread */
482 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
483 futex_nto1_wake(&relay_cmd_queue
.futex
);
487 * Signal handler for the daemon
489 * Simply stop all worker threads, leaving main() return gracefully after
490 * joining all threads and calling cleanup().
493 void sighandler(int sig
)
497 DBG("SIGPIPE caught");
500 DBG("SIGINT caught");
504 DBG("SIGTERM caught");
513 * Setup signal handler for :
514 * SIGINT, SIGTERM, SIGPIPE
517 int set_signal_handler(void)
523 if ((ret
= sigemptyset(&sigset
)) < 0) {
524 PERROR("sigemptyset");
528 sa
.sa_handler
= sighandler
;
531 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
536 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
541 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
546 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
552 * Init thread quit pipe.
554 * Return -1 on error or 0 if all pipes are created.
557 int init_thread_quit_pipe(void)
561 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
567 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
570 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
574 if (events
== NULL
|| size
== 0) {
579 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
585 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
597 * Check if the thread quit pipe was triggered.
599 * Return 1 if it was triggered else 0;
602 int check_thread_quit_pipe(int fd
, uint32_t events
)
604 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
612 * Create and init socket from uri.
615 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
618 struct lttcomm_sock
*sock
= NULL
;
620 sock
= lttcomm_alloc_sock_from_uri(uri
);
622 ERR("Allocating socket");
626 ret
= lttcomm_create_sock(sock
);
630 DBG("Listening on sock %d", sock
->fd
);
632 ret
= sock
->ops
->bind(sock
);
637 ret
= sock
->ops
->listen(sock
, -1);
647 lttcomm_destroy_sock(sock
);
653 * Return nonzero if stream needs to be closed.
656 int close_stream_check(struct relay_stream
*stream
)
659 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
661 * We are about to close the stream so set the data pending flag to 1
662 * which will make the end data pending command skip the stream which
663 * is now closed and ready. Note that after proceeding to a file close,
664 * the written file is ready for reading.
666 stream
->data_pending_check_done
= 1;
673 * This thread manages the listening for new connections on the network
676 void *relay_thread_listener(void *data
)
678 int i
, ret
, pollfd
, err
= -1;
680 uint32_t revents
, nb_fd
;
681 struct lttng_poll_event events
;
682 struct lttcomm_sock
*control_sock
, *data_sock
;
684 DBG("[thread] Relay listener started");
686 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
688 health_code_update();
690 control_sock
= relay_init_sock(control_uri
);
692 goto error_sock_control
;
695 data_sock
= relay_init_sock(data_uri
);
697 goto error_sock_relay
;
701 * Pass 3 as size here for the thread quit pipe, control and data socket.
703 ret
= create_thread_poll_set(&events
, 3);
705 goto error_create_poll
;
708 /* Add the control socket */
709 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
714 /* Add the data socket */
715 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
721 health_code_update();
723 DBG("Listener accepting connections");
727 ret
= lttng_poll_wait(&events
, -1);
731 * Restart interrupted system call.
733 if (errno
== EINTR
) {
741 DBG("Relay new connection received");
742 for (i
= 0; i
< nb_fd
; i
++) {
743 health_code_update();
745 /* Fetch once the poll data */
746 revents
= LTTNG_POLL_GETEV(&events
, i
);
747 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
749 /* Thread quit pipe has been closed. Killing thread. */
750 ret
= check_thread_quit_pipe(pollfd
, revents
);
756 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
757 ERR("socket poll error");
759 } else if (revents
& LPOLLIN
) {
761 * Get allocated in this thread,
762 * enqueued to a global queue, dequeued
763 * and freed in the worker thread.
765 struct relay_command
*relay_cmd
;
766 struct lttcomm_sock
*newsock
;
768 relay_cmd
= zmalloc(sizeof(struct relay_command
));
769 if (relay_cmd
== NULL
) {
770 PERROR("relay command zmalloc");
774 if (pollfd
== data_sock
->fd
) {
775 newsock
= data_sock
->ops
->accept(data_sock
);
777 PERROR("accepting data sock");
781 relay_cmd
->type
= RELAY_DATA
;
782 DBG("Relay data connection accepted, socket %d", newsock
->fd
);
784 assert(pollfd
== control_sock
->fd
);
785 newsock
= control_sock
->ops
->accept(control_sock
);
787 PERROR("accepting control sock");
791 relay_cmd
->type
= RELAY_CONTROL
;
792 DBG("Relay control connection accepted, socket %d", newsock
->fd
);
794 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
,
797 PERROR("setsockopt inet");
798 lttcomm_destroy_sock(newsock
);
802 relay_cmd
->sock
= newsock
;
804 * Lock free enqueue the request.
806 cds_wfq_enqueue(&relay_cmd_queue
.queue
, &relay_cmd
->node
);
809 * Wake the dispatch queue futex. Implicit memory
810 * barrier with the exchange in cds_wfq_enqueue.
812 futex_nto1_wake(&relay_cmd_queue
.futex
);
820 lttng_poll_clean(&events
);
822 if (data_sock
->fd
>= 0) {
823 ret
= data_sock
->ops
->close(data_sock
);
828 lttcomm_destroy_sock(data_sock
);
830 if (control_sock
->fd
>= 0) {
831 ret
= control_sock
->ops
->close(control_sock
);
836 lttcomm_destroy_sock(control_sock
);
840 ERR("Health error occurred in %s", __func__
);
842 health_unregister(health_relayd
);
843 DBG("Relay listener thread cleanup complete");
849 * This thread manages the dispatching of the requests to worker threads
852 void *relay_thread_dispatcher(void *data
)
856 struct cds_wfq_node
*node
;
857 struct relay_command
*relay_cmd
= NULL
;
859 DBG("[thread] Relay dispatcher started");
861 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
863 health_code_update();
865 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
866 health_code_update();
868 /* Atomically prepare the queue futex */
869 futex_nto1_prepare(&relay_cmd_queue
.futex
);
872 health_code_update();
874 /* Dequeue commands */
875 node
= cds_wfq_dequeue_blocking(&relay_cmd_queue
.queue
);
877 DBG("Woken up but nothing in the relay command queue");
878 /* Continue thread execution */
882 relay_cmd
= caa_container_of(node
, struct relay_command
, node
);
883 DBG("Dispatching request waiting on sock %d", relay_cmd
->sock
->fd
);
886 * Inform worker thread of the new request. This
887 * call is blocking so we can be assured that the data will be read
888 * at some point in time or wait to the end of the world :)
890 ret
= lttng_write(relay_cmd_pipe
[1], relay_cmd
,
891 sizeof(struct relay_command
));
893 if (ret
< sizeof(struct relay_command
)) {
894 PERROR("write cmd pipe");
897 } while (node
!= NULL
);
899 /* Futex wait on queue. Blocking call on futex() */
901 futex_nto1_wait(&relay_cmd_queue
.futex
);
905 /* Normal exit, no error */
911 ERR("Health error occurred in %s", __func__
);
913 health_unregister(health_relayd
);
914 DBG("Dispatch thread dying");
920 * Get stream from stream id.
921 * Need to be called with RCU read-side lock held.
923 struct relay_stream
*relay_stream_find_by_id(uint64_t stream_id
)
925 struct lttng_ht_node_ulong
*node
;
926 struct lttng_ht_iter iter
;
927 struct relay_stream
*ret
;
929 lttng_ht_lookup(relay_streams_ht
,
930 (void *)((unsigned long) stream_id
),
932 node
= lttng_ht_iter_get_node_ulong(&iter
);
934 DBG("Relay stream %" PRIu64
" not found", stream_id
);
939 ret
= caa_container_of(node
, struct relay_stream
, stream_n
);
946 void deferred_free_stream(struct rcu_head
*head
)
948 struct relay_stream
*stream
=
949 caa_container_of(head
, struct relay_stream
, rcu_node
);
951 free(stream
->path_name
);
952 free(stream
->channel_name
);
957 void deferred_free_session(struct rcu_head
*head
)
959 struct relay_session
*session
=
960 caa_container_of(head
, struct relay_session
, rcu_node
);
965 * Close a given stream. The stream is freed using a call RCU.
967 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
968 * BEFORE the stream lock MUST be acquired.
970 static void destroy_stream(struct relay_stream
*stream
)
973 struct relay_viewer_stream
*vstream
;
974 struct lttng_ht_iter iter
;
978 delret
= close(stream
->fd
);
980 PERROR("close stream");
983 if (stream
->index_fd
>= 0) {
984 delret
= close(stream
->index_fd
);
986 PERROR("close stream index_fd");
990 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
);
993 * Set the last good value into the viewer stream. This is done
994 * right before the stream gets deleted from the hash table. The
995 * lookup failure on the live thread side of a stream indicates
996 * that the viewer stream index received value should be used.
998 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
999 vstream
->total_index_received
= stream
->total_index_received
;
1000 vstream
->tracefile_count_last
= stream
->tracefile_count_current
;
1001 vstream
->close_write_flag
= 1;
1002 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
1005 /* Cleanup index of that stream. */
1006 relay_index_destroy_by_stream_id(stream
->stream_handle
);
1008 iter
.iter
.node
= &stream
->stream_n
.node
;
1009 delret
= lttng_ht_del(relay_streams_ht
, &iter
);
1011 iter
.iter
.node
= &stream
->ctf_trace_node
.node
;
1012 delret
= lttng_ht_del(stream
->ctf_traces_ht
, &iter
);
1015 if (stream
->ctf_trace
) {
1016 ctf_trace_try_destroy(stream
->ctf_trace
);
1019 call_rcu(&stream
->rcu_node
, deferred_free_stream
);
1020 DBG("Closed tracefile %d from close stream", stream
->fd
);
1024 * relay_delete_session: Free all memory associated with a session and
1028 void relay_delete_session(struct relay_command
*cmd
,
1029 struct lttng_ht
*sessions_ht
)
1031 struct lttng_ht_iter iter
;
1032 struct lttng_ht_node_ulong
*node
;
1033 struct relay_stream
*stream
;
1036 if (!cmd
->session
) {
1040 DBG("Relay deleting session %" PRIu64
, cmd
->session
->id
);
1043 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, node
, node
) {
1044 node
= lttng_ht_iter_get_node_ulong(&iter
);
1048 stream
= caa_container_of(node
, struct relay_stream
, stream_n
);
1049 if (stream
->session
== cmd
->session
) {
1050 destroy_stream(stream
);
1051 cmd
->session
->stream_count
--;
1052 assert(cmd
->session
->stream_count
>= 0);
1056 /* Make this session not visible anymore. */
1057 iter
.iter
.node
= &cmd
->session
->session_n
.node
;
1058 ret
= lttng_ht_del(sessions_ht
, &iter
);
1060 call_rcu(&cmd
->session
->rcu_node
, deferred_free_session
);
1065 * Copy index data from the control port to a given index object.
1067 static void copy_index_control_data(struct relay_index
*index
,
1068 struct lttcomm_relayd_index
*data
)
1074 * The index on disk is encoded in big endian, so we don't need to convert
1075 * the data received on the network. The data_offset value is NEVER
1076 * modified here and is updated by the data thread.
1078 index
->index_data
.packet_size
= data
->packet_size
;
1079 index
->index_data
.content_size
= data
->content_size
;
1080 index
->index_data
.timestamp_begin
= data
->timestamp_begin
;
1081 index
->index_data
.timestamp_end
= data
->timestamp_end
;
1082 index
->index_data
.events_discarded
= data
->events_discarded
;
1083 index
->index_data
.stream_id
= data
->stream_id
;
1087 * Handle the RELAYD_CREATE_SESSION command.
1089 * On success, send back the session id or else return a negative value.
1092 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
1093 struct relay_command
*cmd
,
1094 struct lttng_ht
*sessions_ht
)
1096 int ret
= 0, send_ret
;
1097 struct relay_session
*session
;
1098 struct lttcomm_relayd_status_session reply
;
1103 memset(&reply
, 0, sizeof(reply
));
1105 session
= zmalloc(sizeof(struct relay_session
));
1106 if (session
== NULL
) {
1107 PERROR("relay session zmalloc");
1112 session
->id
= ++last_relay_session_id
;
1113 session
->sock
= cmd
->sock
;
1114 session
->minor
= cmd
->minor
;
1115 session
->major
= cmd
->major
;
1116 cmd
->session
= session
;
1118 reply
.session_id
= htobe64(session
->id
);
1120 switch (cmd
->minor
) {
1121 case 4: /* LTTng sessiond 2.4 */
1123 ret
= cmd_create_session_2_4(cmd
, session
);
1127 lttng_ht_node_init_ulong(&session
->session_n
,
1128 (unsigned long) session
->id
);
1129 lttng_ht_add_unique_ulong(sessions_ht
,
1130 &session
->session_n
);
1132 DBG("Created session %" PRIu64
, session
->id
);
1136 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1138 reply
.ret_code
= htobe32(LTTNG_OK
);
1141 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1143 ERR("Relayd sending session id");
1151 * When we have received all the streams and the metadata for a channel,
1152 * we make them visible to the viewer threads.
1155 void set_viewer_ready_flag(struct relay_command
*cmd
)
1157 struct relay_stream_recv_handle
*node
, *tmp_node
;
1159 cds_list_for_each_entry_safe(node
, tmp_node
, &cmd
->recv_head
, node
) {
1160 struct relay_stream
*stream
;
1163 stream
= relay_stream_find_by_id(node
->id
);
1166 * Stream is most probably being cleaned up by the data thread thus
1167 * simply continue to the next one.
1173 stream
->viewer_ready
= 1;
1176 /* Clean stream handle node. */
1177 cds_list_del(&node
->node
);
1185 * Add a recv handle node to the connection recv list with the given stream
1186 * handle. A new node is allocated thus must be freed when the node is deleted
1189 static void queue_stream_handle(uint64_t handle
, struct relay_command
*cmd
)
1191 struct relay_stream_recv_handle
*node
;
1195 node
= zmalloc(sizeof(*node
));
1197 PERROR("zmalloc queue stream handle");
1202 cds_list_add(&node
->node
, &cmd
->recv_head
);
1206 * relay_add_stream: allocate a new stream for a session
1209 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1210 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1212 struct relay_session
*session
= cmd
->session
;
1213 struct relay_stream
*stream
= NULL
;
1214 struct lttcomm_relayd_status_stream reply
;
1217 if (!session
|| cmd
->version_check_done
== 0) {
1218 ERR("Trying to add a stream before version check");
1220 goto end_no_session
;
1223 stream
= zmalloc(sizeof(struct relay_stream
));
1224 if (stream
== NULL
) {
1225 PERROR("relay stream zmalloc");
1227 goto end_no_session
;
1230 switch (cmd
->minor
) {
1231 case 1: /* LTTng sessiond 2.1 */
1232 ret
= cmd_recv_stream_2_1(cmd
, stream
);
1234 case 2: /* LTTng sessiond 2.2 */
1236 ret
= cmd_recv_stream_2_2(cmd
, stream
);
1240 goto err_free_stream
;
1244 stream
->stream_handle
= ++last_relay_stream_id
;
1245 stream
->prev_seq
= -1ULL;
1246 stream
->session
= session
;
1247 stream
->index_fd
= -1;
1248 stream
->read_index_fd
= -1;
1249 stream
->ctf_trace
= NULL
;
1250 pthread_mutex_init(&stream
->lock
, NULL
);
1252 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
);
1254 ERR("relay creating output directory");
1259 * No need to use run_as API here because whatever we receives, the relayd
1260 * uses its own credentials for the stream files.
1262 ret
= utils_create_stream_file(stream
->path_name
, stream
->channel_name
,
1263 stream
->tracefile_size
, 0, relayd_uid
, relayd_gid
, NULL
);
1265 ERR("Create output file");
1269 if (stream
->tracefile_size
) {
1270 DBG("Tracefile %s/%s_0 created", stream
->path_name
, stream
->channel_name
);
1272 DBG("Tracefile %s/%s created", stream
->path_name
, stream
->channel_name
);
1275 if (!strncmp(stream
->channel_name
, DEFAULT_METADATA_NAME
, NAME_MAX
)) {
1276 stream
->metadata_flag
= 1;
1278 * When we receive a new metadata stream, we create a new
1279 * ctf_trace and we assign this ctf_trace to all streams with
1282 * If later on we receive a new stream for the same ctf_trace,
1283 * we copy the information from the first hit in the HT to the
1286 stream
->ctf_trace
= ctf_trace_create();
1287 if (!stream
->ctf_trace
) {
1291 stream
->ctf_trace
->refcount
++;
1292 stream
->ctf_trace
->metadata_stream
= stream
;
1294 ctf_trace_assign(cmd
->ctf_traces_ht
, stream
);
1295 stream
->ctf_traces_ht
= cmd
->ctf_traces_ht
;
1298 * Add the stream handle in the recv list of the connection. Once the end
1299 * stream message is received, this list is emptied and streams are set
1300 * with the viewer ready flag.
1302 if (stream
->metadata_flag
) {
1303 stream
->viewer_ready
= 1;
1305 queue_stream_handle(stream
->stream_handle
, cmd
);
1308 lttng_ht_node_init_ulong(&stream
->stream_n
,
1309 (unsigned long) stream
->stream_handle
);
1310 lttng_ht_add_unique_ulong(relay_streams_ht
,
1313 lttng_ht_node_init_str(&stream
->ctf_trace_node
, stream
->path_name
);
1314 lttng_ht_add_str(cmd
->ctf_traces_ht
, &stream
->ctf_trace_node
);
1315 session
->stream_count
++;
1317 DBG("Relay new stream added %s with ID %" PRIu64
, stream
->channel_name
,
1318 stream
->stream_handle
);
1321 reply
.handle
= htobe64(stream
->stream_handle
);
1322 /* send the session id to the client or a negative return code on error */
1324 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1325 /* stream was not properly added to the ht, so free it */
1328 reply
.ret_code
= htobe32(LTTNG_OK
);
1331 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1332 sizeof(struct lttcomm_relayd_status_stream
), 0);
1334 ERR("Relay sending stream id");
1343 free(stream
->path_name
);
1344 free(stream
->channel_name
);
1350 * relay_close_stream: close a specific stream
1353 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1354 struct relay_command
*cmd
)
1357 struct relay_session
*session
= cmd
->session
;
1358 struct lttcomm_relayd_close_stream stream_info
;
1359 struct lttcomm_relayd_generic_reply reply
;
1360 struct relay_stream
*stream
;
1362 DBG("Close stream received");
1364 if (!session
|| cmd
->version_check_done
== 0) {
1365 ERR("Trying to close a stream before version check");
1367 goto end_no_session
;
1370 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &stream_info
,
1371 sizeof(struct lttcomm_relayd_close_stream
), 0);
1372 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1374 /* Orderly shutdown. Not necessary to print an error. */
1375 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1377 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1380 goto end_no_session
;
1384 stream
= relay_stream_find_by_id(be64toh(stream_info
.stream_id
));
1390 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1391 stream
->close_flag
= 1;
1392 session
->stream_count
--;
1393 assert(session
->stream_count
>= 0);
1395 if (close_stream_check(stream
)) {
1396 destroy_stream(stream
);
1403 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1405 reply
.ret_code
= htobe32(LTTNG_OK
);
1407 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1408 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1410 ERR("Relay sending stream id");
1419 * relay_unknown_command: send -1 if received unknown command
1422 void relay_unknown_command(struct relay_command
*cmd
)
1424 struct lttcomm_relayd_generic_reply reply
;
1427 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1428 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1429 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1431 ERR("Relay sending unknown command");
1436 * relay_start: send an acknowledgment to the client to tell if we are
1437 * ready to receive data. We are ready if a session is established.
1440 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1441 struct relay_command
*cmd
)
1443 int ret
= htobe32(LTTNG_OK
);
1444 struct lttcomm_relayd_generic_reply reply
;
1445 struct relay_session
*session
= cmd
->session
;
1448 DBG("Trying to start the streaming without a session established");
1449 ret
= htobe32(LTTNG_ERR_UNK
);
1452 reply
.ret_code
= ret
;
1453 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1454 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1456 ERR("Relay sending start ack");
1463 * Append padding to the file pointed by the file descriptor fd.
1465 static int write_padding_to_file(int fd
, uint32_t size
)
1474 zeros
= zmalloc(size
);
1475 if (zeros
== NULL
) {
1476 PERROR("zmalloc zeros for padding");
1481 ret
= lttng_write(fd
, zeros
, size
);
1483 PERROR("write padding to file");
1493 * relay_recv_metadata: receive the metada for the session.
1496 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1497 struct relay_command
*cmd
)
1499 int ret
= htobe32(LTTNG_OK
);
1501 struct relay_session
*session
= cmd
->session
;
1502 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1503 struct relay_stream
*metadata_stream
;
1504 uint64_t data_size
, payload_size
;
1507 ERR("Metadata sent before version check");
1512 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1513 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1514 ERR("Incorrect data size");
1518 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1520 if (data_buffer_size
< data_size
) {
1521 /* In case the realloc fails, we can free the memory */
1524 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1525 if (!tmp_data_ptr
) {
1526 ERR("Allocating data buffer");
1531 data_buffer
= tmp_data_ptr
;
1532 data_buffer_size
= data_size
;
1534 memset(data_buffer
, 0, data_size
);
1535 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1536 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
1537 if (ret
< 0 || ret
!= data_size
) {
1539 /* Orderly shutdown. Not necessary to print an error. */
1540 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1542 ERR("Relay didn't receive the whole metadata");
1547 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1550 metadata_stream
= relay_stream_find_by_id(
1551 be64toh(metadata_struct
->stream_id
));
1552 if (!metadata_stream
) {
1557 size_ret
= lttng_write(metadata_stream
->fd
, metadata_struct
->payload
,
1559 if (size_ret
< payload_size
) {
1560 ERR("Relay error writing metadata on file");
1565 ret
= write_padding_to_file(metadata_stream
->fd
,
1566 be32toh(metadata_struct
->padding_size
));
1570 metadata_stream
->ctf_trace
->metadata_received
+=
1571 payload_size
+ be32toh(metadata_struct
->padding_size
);
1573 DBG2("Relay metadata written");
1582 * relay_send_version: send relayd version number
1585 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1586 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1589 struct lttcomm_relayd_version reply
, msg
;
1593 cmd
->version_check_done
= 1;
1595 /* Get version from the other side. */
1596 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1597 if (ret
< 0 || ret
!= sizeof(msg
)) {
1599 /* Orderly shutdown. Not necessary to print an error. */
1600 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1602 ERR("Relay failed to receive the version values.");
1608 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1609 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1611 /* Major versions must be the same */
1612 if (reply
.major
!= be32toh(msg
.major
)) {
1613 DBG("Incompatible major versions (%u vs %u), deleting session",
1614 reply
.major
, be32toh(msg
.major
));
1615 relay_delete_session(cmd
, sessions_ht
);
1620 cmd
->major
= reply
.major
;
1621 /* We adapt to the lowest compatible version */
1622 if (reply
.minor
<= be32toh(msg
.minor
)) {
1623 cmd
->minor
= reply
.minor
;
1625 cmd
->minor
= be32toh(msg
.minor
);
1628 reply
.major
= htobe32(reply
.major
);
1629 reply
.minor
= htobe32(reply
.minor
);
1630 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1631 sizeof(struct lttcomm_relayd_version
), 0);
1633 ERR("Relay sending version");
1636 DBG("Version check done using protocol %u.%u", cmd
->major
,
1644 * Check for data pending for a given stream id from the session daemon.
1647 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1648 struct relay_command
*cmd
)
1650 struct relay_session
*session
= cmd
->session
;
1651 struct lttcomm_relayd_data_pending msg
;
1652 struct lttcomm_relayd_generic_reply reply
;
1653 struct relay_stream
*stream
;
1655 uint64_t last_net_seq_num
, stream_id
;
1657 DBG("Data pending command received");
1659 if (!session
|| cmd
->version_check_done
== 0) {
1660 ERR("Trying to check for data before version check");
1662 goto end_no_session
;
1665 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1666 if (ret
< sizeof(msg
)) {
1668 /* Orderly shutdown. Not necessary to print an error. */
1669 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1671 ERR("Relay didn't receive valid data_pending struct size : %d",
1675 goto end_no_session
;
1678 stream_id
= be64toh(msg
.stream_id
);
1679 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1682 stream
= relay_stream_find_by_id(stream_id
);
1683 if (stream
== NULL
) {
1688 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1689 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1692 /* Avoid wrapping issue */
1693 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1694 /* Data has in fact been written and is NOT pending */
1697 /* Data still being streamed thus pending */
1701 /* Pending check is now done. */
1702 stream
->data_pending_check_done
= 1;
1707 reply
.ret_code
= htobe32(ret
);
1708 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1710 ERR("Relay data pending ret code failed");
1718 * Wait for the control socket to reach a quiescent state.
1720 * Note that for now, when receiving this command from the session daemon, this
1721 * means that every subsequent commands or data received on the control socket
1722 * has been handled. So, this is why we simply return OK here.
1725 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1726 struct relay_command
*cmd
)
1730 struct relay_stream
*stream
;
1731 struct lttng_ht_iter iter
;
1732 struct lttcomm_relayd_quiescent_control msg
;
1733 struct lttcomm_relayd_generic_reply reply
;
1735 DBG("Checking quiescent state on control socket");
1737 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1738 ERR("Trying to check for data before version check");
1740 goto end_no_session
;
1743 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1744 if (ret
< sizeof(msg
)) {
1746 /* Orderly shutdown. Not necessary to print an error. */
1747 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1749 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1753 goto end_no_session
;
1756 stream_id
= be64toh(msg
.stream_id
);
1759 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1761 if (stream
->stream_handle
== stream_id
) {
1762 stream
->data_pending_check_done
= 1;
1763 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1770 reply
.ret_code
= htobe32(LTTNG_OK
);
1771 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1773 ERR("Relay data quiescent control ret code failed");
1781 * Initialize a data pending command. This means that a client is about to ask
1782 * for data pending for each stream he/she holds. Simply iterate over all
1783 * streams of a session and set the data_pending_check_done flag.
1785 * This command returns to the client a LTTNG_OK code.
1788 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1789 struct relay_command
*cmd
)
1792 struct lttng_ht_iter iter
;
1793 struct lttcomm_relayd_begin_data_pending msg
;
1794 struct lttcomm_relayd_generic_reply reply
;
1795 struct relay_stream
*stream
;
1796 uint64_t session_id
;
1801 DBG("Init streams for data pending");
1803 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1804 ERR("Trying to check for data before version check");
1806 goto end_no_session
;
1809 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1810 if (ret
< sizeof(msg
)) {
1812 /* Orderly shutdown. Not necessary to print an error. */
1813 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1815 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1819 goto end_no_session
;
1822 session_id
= be64toh(msg
.session_id
);
1825 * Iterate over all streams to set the begin data pending flag. For now, the
1826 * streams are indexed by stream handle so we have to iterate over all
1827 * streams to find the one associated with the right session_id.
1830 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1832 if (stream
->session
->id
== session_id
) {
1833 stream
->data_pending_check_done
= 0;
1834 DBG("Set begin data pending flag to stream %" PRIu64
,
1835 stream
->stream_handle
);
1840 /* All good, send back reply. */
1841 reply
.ret_code
= htobe32(LTTNG_OK
);
1843 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1845 ERR("Relay begin data pending send reply failed");
1853 * End data pending command. This will check, for a given session id, if each
1854 * stream associated with it has its data_pending_check_done flag set. If not,
1855 * this means that the client lost track of the stream but the data is still
1856 * being streamed on our side. In this case, we inform the client that data is
1859 * Return to the client if there is data in flight or not with a ret_code.
1862 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1863 struct relay_command
*cmd
)
1866 struct lttng_ht_iter iter
;
1867 struct lttcomm_relayd_end_data_pending msg
;
1868 struct lttcomm_relayd_generic_reply reply
;
1869 struct relay_stream
*stream
;
1870 uint64_t session_id
;
1871 uint32_t is_data_inflight
= 0;
1876 DBG("End data pending command");
1878 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1879 ERR("Trying to check for data before version check");
1881 goto end_no_session
;
1884 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1885 if (ret
< sizeof(msg
)) {
1887 /* Orderly shutdown. Not necessary to print an error. */
1888 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1890 ERR("Relay didn't receive valid end data_pending struct size: %d",
1894 goto end_no_session
;
1897 session_id
= be64toh(msg
.session_id
);
1899 /* Iterate over all streams to see if the begin data pending flag is set. */
1901 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1903 if (stream
->session
->id
== session_id
&&
1904 !stream
->data_pending_check_done
) {
1905 is_data_inflight
= 1;
1906 DBG("Data is still in flight for stream %" PRIu64
,
1907 stream
->stream_handle
);
1913 /* All good, send back reply. */
1914 reply
.ret_code
= htobe32(is_data_inflight
);
1916 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1918 ERR("Relay end data pending send reply failed");
1926 * Receive an index for a specific stream.
1928 * Return 0 on success else a negative value.
1931 int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1932 struct relay_command
*cmd
)
1934 int ret
, send_ret
, index_created
= 0;
1935 struct relay_session
*session
= cmd
->session
;
1936 struct lttcomm_relayd_index index_info
;
1937 struct relay_index
*index
, *wr_index
= NULL
;
1938 struct lttcomm_relayd_generic_reply reply
;
1939 struct relay_stream
*stream
;
1940 uint64_t net_seq_num
;
1944 DBG("Relay receiving index");
1946 if (!session
|| cmd
->version_check_done
== 0) {
1947 ERR("Trying to close a stream before version check");
1949 goto end_no_session
;
1952 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &index_info
,
1953 sizeof(index_info
), 0);
1954 if (ret
< sizeof(index_info
)) {
1956 /* Orderly shutdown. Not necessary to print an error. */
1957 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1959 ERR("Relay didn't receive valid index struct size : %d", ret
);
1962 goto end_no_session
;
1965 net_seq_num
= be64toh(index_info
.net_seq_num
);
1968 stream
= relay_stream_find_by_id(be64toh(index_info
.relay_stream_id
));
1971 goto end_rcu_unlock
;
1974 /* Live beacon handling */
1975 if (index_info
.packet_size
== 0) {
1976 DBG("Received live beacon for stream %" PRIu64
, stream
->stream_handle
);
1979 * Only flag a stream inactive when it has already received data.
1981 if (stream
->total_index_received
> 0) {
1982 stream
->beacon_ts_end
= be64toh(index_info
.timestamp_end
);
1985 goto end_rcu_unlock
;
1987 stream
->beacon_ts_end
= -1ULL;
1990 index
= relay_index_find(stream
->stream_handle
, net_seq_num
);
1992 /* A successful creation will add the object to the HT. */
1993 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
1995 goto end_rcu_unlock
;
2000 copy_index_control_data(index
, &index_info
);
2002 if (index_created
) {
2004 * Try to add the relay index object to the hash table. If an object
2005 * already exist, destroy back the index created, set the data in this
2006 * object and write it on disk.
2008 relay_index_add(index
, &wr_index
);
2010 copy_index_control_data(wr_index
, &index_info
);
2014 /* The index already exists so write it on disk. */
2018 /* Do we have a writable ready index to write on disk. */
2020 /* Starting at 2.4, create the index file if none available. */
2021 if (cmd
->minor
>= 4 && stream
->index_fd
< 0) {
2022 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
2023 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
2024 stream
->tracefile_count_current
);
2026 goto end_rcu_unlock
;
2028 stream
->index_fd
= ret
;
2031 ret
= relay_index_write(wr_index
->fd
, wr_index
);
2033 goto end_rcu_unlock
;
2035 stream
->total_index_received
++;
2042 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2044 reply
.ret_code
= htobe32(LTTNG_OK
);
2046 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
2048 ERR("Relay sending close index id reply");
2057 * Receive the streams_sent message.
2059 * Return 0 on success else a negative value.
2062 int relay_streams_sent(struct lttcomm_relayd_hdr
*recv_hdr
,
2063 struct relay_command
*cmd
)
2066 struct lttcomm_relayd_generic_reply reply
;
2070 DBG("Relay receiving streams_sent");
2072 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
2073 ERR("Trying to close a stream before version check");
2075 goto end_no_session
;
2079 * Flag every pending stream in the connection recv list that they are
2080 * ready to be used by the viewer.
2082 set_viewer_ready_flag(cmd
);
2084 reply
.ret_code
= htobe32(LTTNG_OK
);
2085 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
2087 ERR("Relay sending sent_stream reply");
2099 * Process the commands received on the control socket
2102 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
2103 struct relay_command
*cmd
, struct relay_local_data
*ctx
)
2107 switch (be32toh(recv_hdr
->cmd
)) {
2108 case RELAYD_CREATE_SESSION
:
2109 ret
= relay_create_session(recv_hdr
, cmd
, ctx
->sessions_ht
);
2111 case RELAYD_ADD_STREAM
:
2112 ret
= relay_add_stream(recv_hdr
, cmd
, ctx
->sessions_ht
);
2114 case RELAYD_START_DATA
:
2115 ret
= relay_start(recv_hdr
, cmd
);
2117 case RELAYD_SEND_METADATA
:
2118 ret
= relay_recv_metadata(recv_hdr
, cmd
);
2120 case RELAYD_VERSION
:
2121 ret
= relay_send_version(recv_hdr
, cmd
, ctx
->sessions_ht
);
2123 case RELAYD_CLOSE_STREAM
:
2124 ret
= relay_close_stream(recv_hdr
, cmd
);
2126 case RELAYD_DATA_PENDING
:
2127 ret
= relay_data_pending(recv_hdr
, cmd
);
2129 case RELAYD_QUIESCENT_CONTROL
:
2130 ret
= relay_quiescent_control(recv_hdr
, cmd
);
2132 case RELAYD_BEGIN_DATA_PENDING
:
2133 ret
= relay_begin_data_pending(recv_hdr
, cmd
);
2135 case RELAYD_END_DATA_PENDING
:
2136 ret
= relay_end_data_pending(recv_hdr
, cmd
);
2138 case RELAYD_SEND_INDEX
:
2139 ret
= relay_recv_index(recv_hdr
, cmd
);
2141 case RELAYD_STREAMS_SENT
:
2142 ret
= relay_streams_sent(recv_hdr
, cmd
);
2144 case RELAYD_UPDATE_SYNC_INFO
:
2146 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
2147 relay_unknown_command(cmd
);
2157 * Handle index for a data stream.
2159 * RCU read side lock MUST be acquired.
2161 * Return 0 on success else a negative value.
2163 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2166 int ret
= 0, index_created
= 0;
2167 uint64_t stream_id
, data_offset
;
2168 struct relay_index
*index
, *wr_index
= NULL
;
2172 stream_id
= stream
->stream_handle
;
2173 /* Get data offset because we are about to update the index. */
2174 data_offset
= htobe64(stream
->tracefile_size_current
);
2177 * Lookup for an existing index for that stream id/sequence number. If on
2178 * exists, the control thread already received the data for it thus we need
2179 * to write it on disk.
2181 index
= relay_index_find(stream_id
, net_seq_num
);
2183 /* A successful creation will add the object to the HT. */
2184 index
= relay_index_create(stream_id
, net_seq_num
);
2192 if (rotate_index
|| stream
->index_fd
< 0) {
2193 index
->to_close_fd
= stream
->index_fd
;
2194 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
2195 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
2196 stream
->tracefile_count_current
);
2198 /* This will close the stream's index fd if one. */
2199 relay_index_free_safe(index
);
2202 stream
->index_fd
= ret
;
2204 index
->fd
= stream
->index_fd
;
2205 index
->index_data
.offset
= data_offset
;
2207 if (index_created
) {
2209 * Try to add the relay index object to the hash table. If an object
2210 * already exist, destroy back the index created and set the data.
2212 relay_index_add(index
, &wr_index
);
2214 /* Copy back data from the created index. */
2215 wr_index
->fd
= index
->fd
;
2216 wr_index
->to_close_fd
= index
->to_close_fd
;
2217 wr_index
->index_data
.offset
= data_offset
;
2221 /* The index already exists so write it on disk. */
2225 /* Do we have a writable ready index to write on disk. */
2227 ret
= relay_index_write(wr_index
->fd
, wr_index
);
2231 stream
->total_index_received
++;
2239 * relay_process_data: Process the data received on the data socket
2242 int relay_process_data(struct relay_command
*cmd
)
2244 int ret
= 0, rotate_index
= 0;
2246 struct relay_stream
*stream
;
2247 struct lttcomm_relayd_data_hdr data_hdr
;
2249 uint64_t net_seq_num
;
2252 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &data_hdr
,
2253 sizeof(struct lttcomm_relayd_data_hdr
), 0);
2256 /* Orderly shutdown. Not necessary to print an error. */
2257 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
2259 ERR("Unable to receive data header on sock %d", cmd
->sock
->fd
);
2265 stream_id
= be64toh(data_hdr
.stream_id
);
2268 stream
= relay_stream_find_by_id(stream_id
);
2271 goto end_rcu_unlock
;
2274 data_size
= be32toh(data_hdr
.data_size
);
2275 if (data_buffer_size
< data_size
) {
2278 tmp_data_ptr
= realloc(data_buffer
, data_size
);
2279 if (!tmp_data_ptr
) {
2280 ERR("Allocating data buffer");
2283 goto end_rcu_unlock
;
2285 data_buffer
= tmp_data_ptr
;
2286 data_buffer_size
= data_size
;
2288 memset(data_buffer
, 0, data_size
);
2290 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2292 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2293 data_size
, stream_id
, net_seq_num
);
2294 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
2297 /* Orderly shutdown. Not necessary to print an error. */
2298 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
2301 goto end_rcu_unlock
;
2304 /* Check if a rotation is needed. */
2305 if (stream
->tracefile_size
> 0 &&
2306 (stream
->tracefile_size_current
+ data_size
) >
2307 stream
->tracefile_size
) {
2308 struct relay_viewer_stream
*vstream
;
2311 new_id
= (stream
->tracefile_count_current
+ 1) %
2312 stream
->tracefile_count
;
2314 * When we wrap-around back to 0, we start overwriting old
2317 if (!stream
->tracefile_overwrite
&& new_id
== 0) {
2318 stream
->tracefile_overwrite
= 1;
2320 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
2321 if (stream
->tracefile_overwrite
) {
2322 stream
->oldest_tracefile_id
=
2323 (stream
->oldest_tracefile_id
+ 1) %
2324 stream
->tracefile_count
;
2326 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
);
2329 * The viewer is reading a file about to be
2330 * overwritten. Close the FDs it is
2331 * currently using and let it handle the fault.
2333 if (vstream
->tracefile_count_current
== new_id
) {
2334 pthread_mutex_lock(&vstream
->overwrite_lock
);
2335 vstream
->abort_flag
= 1;
2336 pthread_mutex_unlock(&vstream
->overwrite_lock
);
2337 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2338 stream
->channel_name
, new_id
);
2339 } else if (vstream
->tracefile_count_current
==
2340 stream
->tracefile_count_current
) {
2342 * The reader and writer were in the
2343 * same trace file, inform the viewer
2344 * that no new index will ever be added
2347 vstream
->close_write_flag
= 1;
2350 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
2351 stream
->tracefile_size
, stream
->tracefile_count
,
2352 relayd_uid
, relayd_gid
, stream
->fd
,
2353 &(stream
->tracefile_count_current
), &stream
->fd
);
2354 stream
->total_index_received
= 0;
2355 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
2357 ERR("Rotating stream output file");
2358 goto end_rcu_unlock
;
2360 /* Reset current size because we just perform a stream rotation. */
2361 stream
->tracefile_size_current
= 0;
2366 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2367 * index are NOT supported.
2369 if (stream
->session
->minor
>= 4 && !stream
->session
->snapshot
) {
2370 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2372 goto end_rcu_unlock
;
2376 /* Write data to stream output fd. */
2377 size_ret
= lttng_write(stream
->fd
, data_buffer
, data_size
);
2378 if (size_ret
< data_size
) {
2379 ERR("Relay error writing data to file");
2381 goto end_rcu_unlock
;
2384 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
2385 ret
, stream
->stream_handle
);
2387 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
2389 goto end_rcu_unlock
;
2391 stream
->tracefile_size_current
+= data_size
+ be32toh(data_hdr
.padding_size
);
2393 stream
->prev_seq
= net_seq_num
;
2395 /* Check if we need to close the FD */
2396 if (close_stream_check(stream
)) {
2397 destroy_stream(stream
);
2407 void relay_cleanup_poll_connection(struct lttng_poll_event
*events
, int pollfd
)
2411 lttng_poll_del(events
, pollfd
);
2413 ret
= close(pollfd
);
2415 ERR("Closing pollfd %d", pollfd
);
2420 int relay_add_connection(int fd
, struct lttng_poll_event
*events
,
2421 struct lttng_ht
*relay_connections_ht
)
2423 struct relay_command
*relay_connection
;
2426 relay_connection
= zmalloc(sizeof(struct relay_command
));
2427 if (relay_connection
== NULL
) {
2428 PERROR("Relay command zmalloc");
2431 ret
= lttng_read(fd
, relay_connection
, sizeof(struct relay_command
));
2432 if (ret
< sizeof(struct relay_command
)) {
2433 PERROR("read relay cmd pipe");
2436 CDS_INIT_LIST_HEAD(&relay_connection
->recv_head
);
2439 * Only used by the control side and the reference is copied inside each
2440 * stream from that connection. Thus a destroy HT must be done after every
2441 * stream has been destroyed.
2443 if (relay_connection
->type
== RELAY_CONTROL
) {
2444 relay_connection
->ctf_traces_ht
= lttng_ht_new(0,
2445 LTTNG_HT_TYPE_STRING
);
2446 if (!relay_connection
->ctf_traces_ht
) {
2451 lttng_ht_node_init_ulong(&relay_connection
->sock_n
,
2452 (unsigned long) relay_connection
->sock
->fd
);
2454 lttng_ht_add_unique_ulong(relay_connections_ht
,
2455 &relay_connection
->sock_n
);
2457 return lttng_poll_add(events
,
2458 relay_connection
->sock
->fd
,
2459 LPOLLIN
| LPOLLRDHUP
);
2462 free(relay_connection
);
2468 void deferred_free_connection(struct rcu_head
*head
)
2470 struct relay_command
*relay_connection
=
2471 caa_container_of(head
, struct relay_command
, rcu_node
);
2473 lttcomm_destroy_sock(relay_connection
->sock
);
2474 free(relay_connection
);
2478 void relay_del_connection(struct lttng_ht
*relay_connections_ht
,
2479 struct lttng_ht_iter
*iter
, struct relay_command
*relay_connection
,
2480 struct lttng_ht
*sessions_ht
)
2484 ret
= lttng_ht_del(relay_connections_ht
, iter
);
2487 if (relay_connection
->type
== RELAY_CONTROL
) {
2488 struct relay_stream_recv_handle
*node
, *tmp_node
;
2490 relay_delete_session(relay_connection
, sessions_ht
);
2491 lttng_ht_destroy(relay_connection
->ctf_traces_ht
);
2493 /* Clean up recv list. */
2494 cds_list_for_each_entry_safe(node
, tmp_node
,
2495 &relay_connection
->recv_head
, node
) {
2496 cds_list_del(&node
->node
);
2501 call_rcu(&relay_connection
->rcu_node
, deferred_free_connection
);
2505 * This thread does the actual work
2508 void *relay_thread_worker(void *data
)
2510 int ret
, err
= -1, last_seen_data_fd
= -1;
2512 struct relay_command
*relay_connection
;
2513 struct lttng_poll_event events
;
2514 struct lttng_ht
*relay_connections_ht
;
2515 struct lttng_ht_node_ulong
*node
;
2516 struct lttng_ht_iter iter
;
2517 struct lttcomm_relayd_hdr recv_hdr
;
2518 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
2519 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
2521 DBG("[thread] Relay worker started");
2523 rcu_register_thread();
2525 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2527 health_code_update();
2529 /* table of connections indexed on socket */
2530 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2531 if (!relay_connections_ht
) {
2532 goto relay_connections_ht_error
;
2535 /* Tables of received indexes indexed by index handle and net_seq_num. */
2536 indexes_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64
);
2538 goto indexes_ht_error
;
2541 ret
= create_thread_poll_set(&events
, 2);
2543 goto error_poll_create
;
2546 ret
= lttng_poll_add(&events
, relay_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2553 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2555 health_code_update();
2557 /* Infinite blocking call, waiting for transmission */
2558 DBG3("Relayd worker thread polling...");
2559 health_poll_entry();
2560 ret
= lttng_poll_wait(&events
, -1);
2564 * Restart interrupted system call.
2566 if (errno
== EINTR
) {
2575 * Process control. The control connection is prioritised so we don't
2576 * starve it with high throughout put tracing data on the data
2579 for (i
= 0; i
< nb_fd
; i
++) {
2580 /* Fetch once the poll data */
2581 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2582 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2584 health_code_update();
2586 /* Thread quit pipe has been closed. Killing thread. */
2587 ret
= check_thread_quit_pipe(pollfd
, revents
);
2593 /* Inspect the relay cmd pipe for new connection */
2594 if (pollfd
== relay_cmd_pipe
[0]) {
2595 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2596 ERR("Relay pipe error");
2598 } else if (revents
& LPOLLIN
) {
2599 DBG("Relay command received");
2600 ret
= relay_add_connection(relay_cmd_pipe
[0],
2601 &events
, relay_connections_ht
);
2606 } else if (revents
) {
2608 lttng_ht_lookup(relay_connections_ht
,
2609 (void *)((unsigned long) pollfd
),
2611 node
= lttng_ht_iter_get_node_ulong(&iter
);
2613 DBG2("Relay sock %d not found", pollfd
);
2617 relay_connection
= caa_container_of(node
,
2618 struct relay_command
, sock_n
);
2620 if (revents
& (LPOLLERR
)) {
2622 relay_cleanup_poll_connection(&events
, pollfd
);
2623 relay_del_connection(relay_connections_ht
,
2624 &iter
, relay_connection
, sessions_ht
);
2625 if (last_seen_data_fd
== pollfd
) {
2626 last_seen_data_fd
= last_notdel_data_fd
;
2628 } else if (revents
& (LPOLLHUP
| LPOLLRDHUP
)) {
2629 DBG("Socket %d hung up", pollfd
);
2630 relay_cleanup_poll_connection(&events
, pollfd
);
2631 relay_del_connection(relay_connections_ht
,
2632 &iter
, relay_connection
, sessions_ht
);
2633 if (last_seen_data_fd
== pollfd
) {
2634 last_seen_data_fd
= last_notdel_data_fd
;
2636 } else if (revents
& LPOLLIN
) {
2637 /* control socket */
2638 if (relay_connection
->type
== RELAY_CONTROL
) {
2639 ret
= relay_connection
->sock
->ops
->recvmsg(
2640 relay_connection
->sock
, &recv_hdr
,
2641 sizeof(struct lttcomm_relayd_hdr
), 0);
2642 /* connection closed */
2644 relay_cleanup_poll_connection(&events
, pollfd
);
2645 relay_del_connection(relay_connections_ht
,
2646 &iter
, relay_connection
, sessions_ht
);
2647 DBG("Control connection closed with %d", pollfd
);
2649 if (relay_connection
->session
) {
2650 DBG2("Relay worker receiving data for session : %" PRIu64
,
2651 relay_connection
->session
->id
);
2653 ret
= relay_process_control(&recv_hdr
,
2654 relay_connection
, relay_ctx
);
2656 /* Clear the session on error. */
2657 relay_cleanup_poll_connection(&events
, pollfd
);
2658 relay_del_connection(relay_connections_ht
,
2659 &iter
, relay_connection
, sessions_ht
);
2660 DBG("Connection closed with %d", pollfd
);
2666 * Flag the last seen data fd not deleted. It will be
2667 * used as the last seen fd if any fd gets deleted in
2670 last_notdel_data_fd
= pollfd
;
2678 * The last loop handled a control request, go back to poll to make
2679 * sure we prioritise the control socket.
2685 if (last_seen_data_fd
>= 0) {
2686 for (i
= 0; i
< nb_fd
; i
++) {
2687 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2689 health_code_update();
2691 if (last_seen_data_fd
== pollfd
) {
2698 /* Process data connection. */
2699 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2700 /* Fetch the poll data. */
2701 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2702 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2704 health_code_update();
2706 /* Skip the command pipe. It's handled in the first loop. */
2707 if (pollfd
== relay_cmd_pipe
[0]) {
2713 lttng_ht_lookup(relay_connections_ht
,
2714 (void *)((unsigned long) pollfd
),
2716 node
= lttng_ht_iter_get_node_ulong(&iter
);
2718 /* Skip it. Might be removed before. */
2722 relay_connection
= caa_container_of(node
,
2723 struct relay_command
, sock_n
);
2725 if (revents
& LPOLLIN
) {
2726 if (relay_connection
->type
!= RELAY_DATA
) {
2730 ret
= relay_process_data(relay_connection
);
2731 /* connection closed */
2733 relay_cleanup_poll_connection(&events
, pollfd
);
2734 relay_del_connection(relay_connections_ht
,
2735 &iter
, relay_connection
, sessions_ht
);
2736 DBG("Data connection closed with %d", pollfd
);
2738 * Every goto restart call sets the last seen fd where
2739 * here we don't really care since we gracefully
2740 * continue the loop after the connection is deleted.
2743 /* Keep last seen port. */
2744 last_seen_data_fd
= pollfd
;
2752 last_seen_data_fd
= -1;
2755 /* Normal exit, no error */
2760 lttng_poll_clean(&events
);
2762 /* empty the hash table and free the memory */
2764 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, node
, node
) {
2765 health_code_update();
2767 node
= lttng_ht_iter_get_node_ulong(&iter
);
2769 relay_connection
= caa_container_of(node
,
2770 struct relay_command
, sock_n
);
2771 relay_del_connection(relay_connections_ht
,
2772 &iter
, relay_connection
, sessions_ht
);
2777 lttng_ht_destroy(indexes_ht
);
2779 lttng_ht_destroy(relay_connections_ht
);
2780 relay_connections_ht_error
:
2781 /* Close relay cmd pipes */
2782 utils_close_pipe(relay_cmd_pipe
);
2784 DBG("Thread exited with error");
2786 DBG("Worker thread cleanup complete");
2790 ERR("Health error occurred in %s", __func__
);
2792 health_unregister(health_relayd
);
2793 rcu_unregister_thread();
2799 * Create the relay command pipe to wake thread_manage_apps.
2800 * Closed in cleanup().
2802 static int create_relay_cmd_pipe(void)
2806 ret
= utils_create_pipe_cloexec(relay_cmd_pipe
);
2814 int main(int argc
, char **argv
)
2818 struct relay_local_data
*relay_ctx
;
2820 /* Create thread quit pipe */
2821 if ((ret
= init_thread_quit_pipe()) < 0) {
2825 /* Parse arguments */
2827 if ((ret
= set_options(argc
, argv
)) < 0) {
2831 if ((ret
= set_signal_handler()) < 0) {
2835 /* Try to create directory if -o, --output is specified. */
2836 if (opt_output_path
) {
2837 if (*opt_output_path
!= '/') {
2838 ERR("Please specify an absolute path for -o, --output PATH");
2842 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2844 ERR("Unable to create %s", opt_output_path
);
2858 /* We need those values for the file/dir creation. */
2859 relayd_uid
= getuid();
2860 relayd_gid
= getgid();
2862 /* Check if daemon is UID = 0 */
2863 if (relayd_uid
== 0) {
2864 if (control_uri
->port
< 1024 || data_uri
->port
< 1024 || live_uri
->port
< 1024) {
2865 ERR("Need to be root to use ports < 1024");
2871 /* Setup the thread apps communication pipe. */
2872 if ((ret
= create_relay_cmd_pipe()) < 0) {
2876 /* Init relay command queue. */
2877 cds_wfq_init(&relay_cmd_queue
.queue
);
2879 /* Set up max poll set size */
2880 lttng_poll_set_max_size();
2882 /* Initialize communication library */
2885 relay_ctx
= zmalloc(sizeof(struct relay_local_data
));
2887 PERROR("relay_ctx");
2891 /* tables of sessions indexed by session ID */
2892 relay_ctx
->sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2893 if (!relay_ctx
->sessions_ht
) {
2894 goto exit_relay_ctx_sessions
;
2897 /* tables of streams indexed by stream ID */
2898 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2899 if (!relay_streams_ht
) {
2900 goto exit_relay_ctx_streams
;
2903 /* tables of streams indexed by stream ID */
2904 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2905 if (!viewer_streams_ht
) {
2906 goto exit_relay_ctx_viewer_streams
;
2909 /* Initialize thread health monitoring */
2910 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
2911 if (!health_relayd
) {
2912 PERROR("health_app_create error");
2913 goto exit_health_app_create
;
2916 ret
= utils_create_pipe(health_quit_pipe
);
2918 goto error_health_pipe
;
2921 /* Create thread to manage the client socket */
2922 ret
= pthread_create(&health_thread
, NULL
,
2923 thread_manage_health
, (void *) NULL
);
2925 PERROR("pthread_create health");
2929 /* Setup the dispatcher thread */
2930 ret
= pthread_create(&dispatcher_thread
, NULL
,
2931 relay_thread_dispatcher
, (void *) NULL
);
2933 PERROR("pthread_create dispatcher");
2934 goto exit_dispatcher
;
2937 /* Setup the worker thread */
2938 ret
= pthread_create(&worker_thread
, NULL
,
2939 relay_thread_worker
, (void *) relay_ctx
);
2941 PERROR("pthread_create worker");
2945 /* Setup the listener thread */
2946 ret
= pthread_create(&listener_thread
, NULL
,
2947 relay_thread_listener
, (void *) NULL
);
2949 PERROR("pthread_create listener");
2953 ret
= live_start_threads(live_uri
, relay_ctx
, thread_quit_pipe
);
2955 ERR("Starting live viewer threads");
2960 ret
= pthread_join(listener_thread
, &status
);
2962 PERROR("pthread_join");
2963 goto error
; /* join error, exit without cleanup */
2967 ret
= pthread_join(worker_thread
, &status
);
2969 PERROR("pthread_join");
2970 goto error
; /* join error, exit without cleanup */
2974 ret
= pthread_join(dispatcher_thread
, &status
);
2976 PERROR("pthread_join");
2977 goto error
; /* join error, exit without cleanup */
2981 ret
= pthread_join(health_thread
, &status
);
2983 PERROR("pthread_join health thread");
2984 goto error
; /* join error, exit without cleanup */
2988 * Stop live threads only after joining other threads.
2990 live_stop_threads();
2993 utils_close_pipe(health_quit_pipe
);
2996 health_app_destroy(health_relayd
);
2998 exit_health_app_create
:
2999 lttng_ht_destroy(viewer_streams_ht
);
3001 exit_relay_ctx_viewer_streams
:
3002 lttng_ht_destroy(relay_streams_ht
);
3004 exit_relay_ctx_streams
:
3005 lttng_ht_destroy(relay_ctx
->sessions_ht
);
3007 exit_relay_ctx_sessions
: