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/daemonize.h>
49 #include <common/futex.h>
50 #include <common/sessiond-comm/sessiond-comm.h>
51 #include <common/sessiond-comm/inet.h>
52 #include <common/sessiond-comm/relayd.h>
53 #include <common/uri.h>
54 #include <common/utils.h>
55 #include <common/config/config.h>
58 #include "ctf-trace.h"
61 #include "lttng-relayd.h"
63 #include "health-relayd.h"
64 #include "testpoint.h"
66 /* command line options */
67 char *opt_output_path
;
68 static int opt_daemon
, opt_background
;
71 * We need to wait for listener and live listener threads, as well as
72 * health check thread, before being ready to signal readiness.
74 #define NR_LTTNG_RELAY_READY 3
75 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
76 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
77 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
79 static struct lttng_uri
*control_uri
;
80 static struct lttng_uri
*data_uri
;
81 static struct lttng_uri
*live_uri
;
85 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
86 static int tracing_group_name_override
;
88 const char * const config_section_name
= "relayd";
91 * Quit pipe for all threads. This permits a single cancellation point
92 * for all threads when receiving an event on the pipe.
94 int thread_quit_pipe
[2] = { -1, -1 };
97 * This pipe is used to inform the worker thread that a command is queued and
98 * ready to be processed.
100 static int relay_cmd_pipe
[2] = { -1, -1 };
102 /* Shared between threads */
103 static int dispatch_thread_exit
;
105 static pthread_t listener_thread
;
106 static pthread_t dispatcher_thread
;
107 static pthread_t worker_thread
;
108 static pthread_t health_thread
;
110 static uint64_t last_relay_stream_id
;
111 static uint64_t last_relay_session_id
;
114 * Relay command queue.
116 * The relay_thread_listener and relay_thread_dispatcher communicate with this
119 static struct relay_cmd_queue relay_cmd_queue
;
121 /* buffer allocated at startup, used to store the trace data */
122 static char *data_buffer
;
123 static unsigned int data_buffer_size
;
125 /* We need those values for the file/dir creation. */
126 static uid_t relayd_uid
;
127 static gid_t relayd_gid
;
129 /* Global relay stream hash table. */
130 struct lttng_ht
*relay_streams_ht
;
132 /* Global relay viewer stream hash table. */
133 struct lttng_ht
*viewer_streams_ht
;
135 /* Global hash table that stores relay index object. */
136 struct lttng_ht
*indexes_ht
;
138 /* Relayd health monitoring */
139 struct health_app
*health_relayd
;
141 static struct option long_options
[] = {
142 { "control-port", 1, 0, 'C', },
143 { "data-port", 1, 0, 'D', },
144 { "live-port", 1, 0, 'L', },
145 { "daemonize", 0, 0, 'd', },
146 { "background", 0, 0, 'b', },
147 { "group", 1, 0, 'g', },
148 { "help", 0, 0, 'h', },
149 { "output", 1, 0, 'o', },
150 { "verbose", 0, 0, 'v', },
151 { "config", 1, 0, 'f' },
155 static const char *config_ignore_options
[] = { "help", "config" };
158 * usage function on stderr
163 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
164 fprintf(stderr
, " -h, --help Display this usage.\n");
165 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
166 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
167 fprintf(stderr
, " -C, --control-port URL Control port listening.\n");
168 fprintf(stderr
, " -D, --data-port URL Data port listening.\n");
169 fprintf(stderr
, " -L, --live-port URL Live view port listening.\n");
170 fprintf(stderr
, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
171 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
172 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
173 fprintf(stderr
, " -f --config Load daemon configuration file\n");
177 * Take an option from the getopt output and set it in the right variable to be
180 * Return 0 on success else a negative value.
183 int set_option(int opt
, const char *arg
, const char *optname
)
189 fprintf(stderr
, "option %s", optname
);
191 fprintf(stderr
, " with arg %s\n", arg
);
195 ret
= uri_parse(arg
, &control_uri
);
197 ERR("Invalid control URI specified");
200 if (control_uri
->port
== 0) {
201 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
205 ret
= uri_parse(arg
, &data_uri
);
207 ERR("Invalid data URI specified");
210 if (data_uri
->port
== 0) {
211 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
215 ret
= uri_parse(arg
, &live_uri
);
217 ERR("Invalid live URI specified");
220 if (live_uri
->port
== 0) {
221 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
231 tracing_group_name
= strdup(arg
);
232 tracing_group_name_override
= 1;
238 ret
= asprintf(&opt_output_path
, "%s", arg
);
241 PERROR("asprintf opt_output_path");
246 /* Verbose level can increase using multiple -v */
248 lttng_opt_verbose
= config_parse_value(arg
);
250 lttng_opt_verbose
+= 1;
254 /* Unknown option or other error.
255 * Error is printed by getopt, just return */
268 * config_entry_handler_cb used to handle options read from a config file.
269 * See config_entry_handler_cb comment in common/config/config.h for the
270 * return value conventions.
273 int config_entry_handler(const struct config_entry
*entry
, void *unused
)
277 if (!entry
|| !entry
->name
|| !entry
->value
) {
282 /* Check if the option is to be ignored */
283 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
284 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
289 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
290 /* Ignore if entry name is not fully matched. */
291 if (strcmp(entry
->name
, long_options
[i
].name
)) {
296 * If the option takes no argument on the command line, we have to
297 * check if the value is "true". We support non-zero numeric values,
300 if (!long_options
[i
].has_arg
) {
301 ret
= config_parse_value(entry
->value
);
304 WARN("Invalid configuration value \"%s\" for option %s",
305 entry
->value
, entry
->name
);
307 /* False, skip boolean config option. */
312 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
316 WARN("Unrecognized option \"%s\" in daemon configuration file.",
324 int set_options(int argc
, char **argv
)
326 int c
, ret
= 0, option_index
= 0;
327 int orig_optopt
= optopt
, orig_optind
= optind
;
328 char *default_address
, *optstring
;
329 const char *config_path
= NULL
;
331 optstring
= utils_generate_optstring(long_options
,
332 sizeof(long_options
) / sizeof(struct option
));
338 /* Check for the --config option */
340 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
341 &option_index
)) != -1) {
345 } else if (c
!= 'f') {
349 config_path
= utils_expand_path(optarg
);
351 ERR("Failed to resolve path: %s", optarg
);
355 ret
= config_get_section_entries(config_path
, config_section_name
,
356 config_entry_handler
, NULL
);
359 ERR("Invalid configuration option at line %i", ret
);
365 /* Reset getopt's global state */
366 optopt
= orig_optopt
;
367 optind
= orig_optind
;
369 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
374 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
380 /* assign default values */
381 if (control_uri
== NULL
) {
382 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
383 DEFAULT_NETWORK_CONTROL_PORT
);
385 PERROR("asprintf default data address");
389 ret
= uri_parse(default_address
, &control_uri
);
390 free(default_address
);
392 ERR("Invalid control URI specified");
396 if (data_uri
== NULL
) {
397 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
398 DEFAULT_NETWORK_DATA_PORT
);
400 PERROR("asprintf default data address");
404 ret
= uri_parse(default_address
, &data_uri
);
405 free(default_address
);
407 ERR("Invalid data URI specified");
411 if (live_uri
== NULL
) {
412 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
413 DEFAULT_NETWORK_VIEWER_PORT
);
415 PERROR("asprintf default viewer control address");
419 ret
= uri_parse(default_address
, &live_uri
);
420 free(default_address
);
422 ERR("Invalid viewer control URI specified");
440 /* free the dynamically allocated opt_output_path */
441 free(opt_output_path
);
443 /* Close thread quit pipes */
444 utils_close_pipe(thread_quit_pipe
);
446 uri_free(control_uri
);
448 /* Live URI is freed in the live thread. */
450 if (tracing_group_name_override
) {
451 free((void *) tracing_group_name
);
456 * Write to writable pipe used to notify a thread.
459 int notify_thread_pipe(int wpipe
)
463 ret
= lttng_write(wpipe
, "!", 1);
465 PERROR("write poll pipe");
471 static void notify_health_quit_pipe(int *pipe
)
475 ret
= lttng_write(pipe
[1], "4", 1);
477 PERROR("write relay health quit");
482 * Stop all threads by closing the thread quit pipe.
485 void stop_threads(void)
489 /* Stopping all threads */
490 DBG("Terminating all threads");
491 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
493 ERR("write error on thread quit pipe");
496 notify_health_quit_pipe(health_quit_pipe
);
498 /* Dispatch thread */
499 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
500 futex_nto1_wake(&relay_cmd_queue
.futex
);
504 * Signal handler for the daemon
506 * Simply stop all worker threads, leaving main() return gracefully after
507 * joining all threads and calling cleanup().
510 void sighandler(int sig
)
514 DBG("SIGPIPE caught");
517 DBG("SIGINT caught");
521 DBG("SIGTERM caught");
525 CMM_STORE_SHARED(recv_child_signal
, 1);
533 * Setup signal handler for :
534 * SIGINT, SIGTERM, SIGPIPE
537 int set_signal_handler(void)
543 if ((ret
= sigemptyset(&sigset
)) < 0) {
544 PERROR("sigemptyset");
548 sa
.sa_handler
= sighandler
;
551 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
556 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
561 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
566 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
571 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
576 void lttng_relay_notify_ready(void)
578 /* Notify the parent of the fork() process that we are ready. */
579 if (opt_daemon
|| opt_background
) {
580 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
581 kill(child_ppid
, SIGUSR1
);
587 * Init thread quit pipe.
589 * Return -1 on error or 0 if all pipes are created.
592 int init_thread_quit_pipe(void)
596 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
602 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
605 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
609 if (events
== NULL
|| size
== 0) {
614 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
620 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
632 * Check if the thread quit pipe was triggered.
634 * Return 1 if it was triggered else 0;
637 int check_thread_quit_pipe(int fd
, uint32_t events
)
639 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
647 * Create and init socket from uri.
650 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
653 struct lttcomm_sock
*sock
= NULL
;
655 sock
= lttcomm_alloc_sock_from_uri(uri
);
657 ERR("Allocating socket");
661 ret
= lttcomm_create_sock(sock
);
665 DBG("Listening on sock %d", sock
->fd
);
667 ret
= sock
->ops
->bind(sock
);
672 ret
= sock
->ops
->listen(sock
, -1);
682 lttcomm_destroy_sock(sock
);
688 * Return nonzero if stream needs to be closed.
691 int close_stream_check(struct relay_stream
*stream
)
694 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
696 * We are about to close the stream so set the data pending flag to 1
697 * which will make the end data pending command skip the stream which
698 * is now closed and ready. Note that after proceeding to a file close,
699 * the written file is ready for reading.
701 stream
->data_pending_check_done
= 1;
708 * This thread manages the listening for new connections on the network
711 void *relay_thread_listener(void *data
)
713 int i
, ret
, pollfd
, err
= -1;
715 uint32_t revents
, nb_fd
;
716 struct lttng_poll_event events
;
717 struct lttcomm_sock
*control_sock
, *data_sock
;
719 DBG("[thread] Relay listener started");
721 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
723 health_code_update();
725 control_sock
= relay_init_sock(control_uri
);
727 goto error_sock_control
;
730 data_sock
= relay_init_sock(data_uri
);
732 goto error_sock_relay
;
736 * Pass 3 as size here for the thread quit pipe, control and data socket.
738 ret
= create_thread_poll_set(&events
, 3);
740 goto error_create_poll
;
743 /* Add the control socket */
744 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
749 /* Add the data socket */
750 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
755 lttng_relay_notify_ready();
757 if (testpoint(relayd_thread_listener
)) {
758 goto error_testpoint
;
762 health_code_update();
764 DBG("Listener accepting connections");
768 ret
= lttng_poll_wait(&events
, -1);
772 * Restart interrupted system call.
774 if (errno
== EINTR
) {
782 DBG("Relay new connection received");
783 for (i
= 0; i
< nb_fd
; i
++) {
784 health_code_update();
786 /* Fetch once the poll data */
787 revents
= LTTNG_POLL_GETEV(&events
, i
);
788 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
790 /* Thread quit pipe has been closed. Killing thread. */
791 ret
= check_thread_quit_pipe(pollfd
, revents
);
797 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
798 ERR("socket poll error");
800 } else if (revents
& LPOLLIN
) {
802 * Get allocated in this thread,
803 * enqueued to a global queue, dequeued
804 * and freed in the worker thread.
806 struct relay_command
*relay_cmd
;
807 struct lttcomm_sock
*newsock
;
809 relay_cmd
= zmalloc(sizeof(struct relay_command
));
810 if (relay_cmd
== NULL
) {
811 PERROR("relay command zmalloc");
815 if (pollfd
== data_sock
->fd
) {
816 newsock
= data_sock
->ops
->accept(data_sock
);
818 PERROR("accepting data sock");
822 relay_cmd
->type
= RELAY_DATA
;
823 DBG("Relay data connection accepted, socket %d", newsock
->fd
);
825 assert(pollfd
== control_sock
->fd
);
826 newsock
= control_sock
->ops
->accept(control_sock
);
828 PERROR("accepting control sock");
832 relay_cmd
->type
= RELAY_CONTROL
;
833 DBG("Relay control connection accepted, socket %d", newsock
->fd
);
835 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
,
838 PERROR("setsockopt inet");
839 lttcomm_destroy_sock(newsock
);
843 relay_cmd
->sock
= newsock
;
845 * Lock free enqueue the request.
847 cds_wfq_enqueue(&relay_cmd_queue
.queue
, &relay_cmd
->node
);
850 * Wake the dispatch queue futex. Implicit memory
851 * barrier with the exchange in cds_wfq_enqueue.
853 futex_nto1_wake(&relay_cmd_queue
.futex
);
862 lttng_poll_clean(&events
);
864 if (data_sock
->fd
>= 0) {
865 ret
= data_sock
->ops
->close(data_sock
);
870 lttcomm_destroy_sock(data_sock
);
872 if (control_sock
->fd
>= 0) {
873 ret
= control_sock
->ops
->close(control_sock
);
878 lttcomm_destroy_sock(control_sock
);
882 ERR("Health error occurred in %s", __func__
);
884 health_unregister(health_relayd
);
885 DBG("Relay listener thread cleanup complete");
891 * This thread manages the dispatching of the requests to worker threads
894 void *relay_thread_dispatcher(void *data
)
898 struct cds_wfq_node
*node
;
899 struct relay_command
*relay_cmd
= NULL
;
901 DBG("[thread] Relay dispatcher started");
903 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
905 if (testpoint(relayd_thread_dispatcher
)) {
906 goto error_testpoint
;
909 health_code_update();
911 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
912 health_code_update();
914 /* Atomically prepare the queue futex */
915 futex_nto1_prepare(&relay_cmd_queue
.futex
);
918 health_code_update();
920 /* Dequeue commands */
921 node
= cds_wfq_dequeue_blocking(&relay_cmd_queue
.queue
);
923 DBG("Woken up but nothing in the relay command queue");
924 /* Continue thread execution */
928 relay_cmd
= caa_container_of(node
, struct relay_command
, node
);
929 DBG("Dispatching request waiting on sock %d", relay_cmd
->sock
->fd
);
932 * Inform worker thread of the new request. This
933 * call is blocking so we can be assured that the data will be read
934 * at some point in time or wait to the end of the world :)
936 ret
= lttng_write(relay_cmd_pipe
[1], relay_cmd
,
937 sizeof(struct relay_command
));
939 if (ret
< sizeof(struct relay_command
)) {
940 PERROR("write cmd pipe");
943 } while (node
!= NULL
);
945 /* Futex wait on queue. Blocking call on futex() */
947 futex_nto1_wait(&relay_cmd_queue
.futex
);
951 /* Normal exit, no error */
958 ERR("Health error occurred in %s", __func__
);
960 health_unregister(health_relayd
);
961 DBG("Dispatch thread dying");
967 * Get stream from stream id.
968 * Need to be called with RCU read-side lock held.
970 struct relay_stream
*relay_stream_find_by_id(uint64_t stream_id
)
972 struct lttng_ht_node_ulong
*node
;
973 struct lttng_ht_iter iter
;
974 struct relay_stream
*ret
;
976 lttng_ht_lookup(relay_streams_ht
,
977 (void *)((unsigned long) stream_id
),
979 node
= lttng_ht_iter_get_node_ulong(&iter
);
981 DBG("Relay stream %" PRIu64
" not found", stream_id
);
986 ret
= caa_container_of(node
, struct relay_stream
, stream_n
);
993 void deferred_free_stream(struct rcu_head
*head
)
995 struct relay_stream
*stream
=
996 caa_container_of(head
, struct relay_stream
, rcu_node
);
998 free(stream
->path_name
);
999 free(stream
->channel_name
);
1004 void deferred_free_session(struct rcu_head
*head
)
1006 struct relay_session
*session
=
1007 caa_container_of(head
, struct relay_session
, rcu_node
);
1012 * Close a given stream. The stream is freed using a call RCU.
1014 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
1015 * BEFORE the stream lock MUST be acquired.
1017 static void destroy_stream(struct relay_stream
*stream
)
1020 struct relay_viewer_stream
*vstream
;
1021 struct lttng_ht_iter iter
;
1025 delret
= close(stream
->fd
);
1027 PERROR("close stream");
1030 if (stream
->index_fd
>= 0) {
1031 delret
= close(stream
->index_fd
);
1033 PERROR("close stream index_fd");
1037 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
);
1040 * Set the last good value into the viewer stream. This is done
1041 * right before the stream gets deleted from the hash table. The
1042 * lookup failure on the live thread side of a stream indicates
1043 * that the viewer stream index received value should be used.
1045 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
1046 vstream
->total_index_received
= stream
->total_index_received
;
1047 vstream
->tracefile_count_last
= stream
->tracefile_count_current
;
1048 vstream
->close_write_flag
= 1;
1049 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
1052 /* Cleanup index of that stream. */
1053 relay_index_destroy_by_stream_id(stream
->stream_handle
);
1055 iter
.iter
.node
= &stream
->stream_n
.node
;
1056 delret
= lttng_ht_del(relay_streams_ht
, &iter
);
1058 iter
.iter
.node
= &stream
->ctf_trace_node
.node
;
1059 delret
= lttng_ht_del(stream
->ctf_traces_ht
, &iter
);
1062 if (stream
->ctf_trace
) {
1063 ctf_trace_try_destroy(stream
->ctf_trace
);
1066 call_rcu(&stream
->rcu_node
, deferred_free_stream
);
1067 DBG("Closed tracefile %d from close stream", stream
->fd
);
1071 * relay_delete_session: Free all memory associated with a session and
1075 void relay_delete_session(struct relay_command
*cmd
,
1076 struct lttng_ht
*sessions_ht
)
1078 struct lttng_ht_iter iter
;
1079 struct lttng_ht_node_ulong
*node
;
1080 struct relay_stream
*stream
;
1083 if (!cmd
->session
) {
1087 DBG("Relay deleting session %" PRIu64
, cmd
->session
->id
);
1090 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, node
, node
) {
1091 node
= lttng_ht_iter_get_node_ulong(&iter
);
1095 stream
= caa_container_of(node
, struct relay_stream
, stream_n
);
1096 if (stream
->session
== cmd
->session
) {
1097 destroy_stream(stream
);
1098 cmd
->session
->stream_count
--;
1099 assert(cmd
->session
->stream_count
>= 0);
1103 /* Make this session not visible anymore. */
1104 iter
.iter
.node
= &cmd
->session
->session_n
.node
;
1105 ret
= lttng_ht_del(sessions_ht
, &iter
);
1107 call_rcu(&cmd
->session
->rcu_node
, deferred_free_session
);
1112 * Copy index data from the control port to a given index object.
1114 static void copy_index_control_data(struct relay_index
*index
,
1115 struct lttcomm_relayd_index
*data
)
1121 * The index on disk is encoded in big endian, so we don't need to convert
1122 * the data received on the network. The data_offset value is NEVER
1123 * modified here and is updated by the data thread.
1125 index
->index_data
.packet_size
= data
->packet_size
;
1126 index
->index_data
.content_size
= data
->content_size
;
1127 index
->index_data
.timestamp_begin
= data
->timestamp_begin
;
1128 index
->index_data
.timestamp_end
= data
->timestamp_end
;
1129 index
->index_data
.events_discarded
= data
->events_discarded
;
1130 index
->index_data
.stream_id
= data
->stream_id
;
1134 * Handle the RELAYD_CREATE_SESSION command.
1136 * On success, send back the session id or else return a negative value.
1139 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
1140 struct relay_command
*cmd
,
1141 struct lttng_ht
*sessions_ht
)
1143 int ret
= 0, send_ret
;
1144 struct relay_session
*session
;
1145 struct lttcomm_relayd_status_session reply
;
1150 memset(&reply
, 0, sizeof(reply
));
1152 session
= zmalloc(sizeof(struct relay_session
));
1153 if (session
== NULL
) {
1154 PERROR("relay session zmalloc");
1159 session
->id
= ++last_relay_session_id
;
1160 session
->sock
= cmd
->sock
;
1161 session
->minor
= cmd
->minor
;
1162 session
->major
= cmd
->major
;
1163 cmd
->session
= session
;
1165 reply
.session_id
= htobe64(session
->id
);
1167 switch (cmd
->minor
) {
1172 case 4: /* LTTng sessiond 2.4 */
1174 ret
= cmd_create_session_2_4(cmd
, session
);
1178 lttng_ht_node_init_ulong(&session
->session_n
,
1179 (unsigned long) session
->id
);
1180 lttng_ht_add_unique_ulong(sessions_ht
,
1181 &session
->session_n
);
1183 DBG("Created session %" PRIu64
, session
->id
);
1187 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1189 reply
.ret_code
= htobe32(LTTNG_OK
);
1192 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1194 ERR("Relayd sending session id");
1202 * When we have received all the streams and the metadata for a channel,
1203 * we make them visible to the viewer threads.
1206 void set_viewer_ready_flag(struct relay_command
*cmd
)
1208 struct relay_stream_recv_handle
*node
, *tmp_node
;
1210 cds_list_for_each_entry_safe(node
, tmp_node
, &cmd
->recv_head
, node
) {
1211 struct relay_stream
*stream
;
1214 stream
= relay_stream_find_by_id(node
->id
);
1217 * Stream is most probably being cleaned up by the data thread thus
1218 * simply continue to the next one.
1224 stream
->viewer_ready
= 1;
1227 /* Clean stream handle node. */
1228 cds_list_del(&node
->node
);
1236 * Add a recv handle node to the connection recv list with the given stream
1237 * handle. A new node is allocated thus must be freed when the node is deleted
1240 static void queue_stream_handle(uint64_t handle
, struct relay_command
*cmd
)
1242 struct relay_stream_recv_handle
*node
;
1246 node
= zmalloc(sizeof(*node
));
1248 PERROR("zmalloc queue stream handle");
1253 cds_list_add(&node
->node
, &cmd
->recv_head
);
1257 * relay_add_stream: allocate a new stream for a session
1260 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1261 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1263 struct relay_session
*session
= cmd
->session
;
1264 struct relay_stream
*stream
= NULL
;
1265 struct lttcomm_relayd_status_stream reply
;
1268 if (!session
|| cmd
->version_check_done
== 0) {
1269 ERR("Trying to add a stream before version check");
1271 goto end_no_session
;
1274 stream
= zmalloc(sizeof(struct relay_stream
));
1275 if (stream
== NULL
) {
1276 PERROR("relay stream zmalloc");
1278 goto end_no_session
;
1281 switch (cmd
->minor
) {
1282 case 1: /* LTTng sessiond 2.1 */
1283 ret
= cmd_recv_stream_2_1(cmd
, stream
);
1285 case 2: /* LTTng sessiond 2.2 */
1287 ret
= cmd_recv_stream_2_2(cmd
, stream
);
1291 goto err_free_stream
;
1295 stream
->stream_handle
= ++last_relay_stream_id
;
1296 stream
->prev_seq
= -1ULL;
1297 stream
->session
= session
;
1298 stream
->index_fd
= -1;
1299 stream
->read_index_fd
= -1;
1300 stream
->ctf_trace
= NULL
;
1301 pthread_mutex_init(&stream
->lock
, NULL
);
1303 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
);
1305 ERR("relay creating output directory");
1310 * No need to use run_as API here because whatever we receives, the relayd
1311 * uses its own credentials for the stream files.
1313 ret
= utils_create_stream_file(stream
->path_name
, stream
->channel_name
,
1314 stream
->tracefile_size
, 0, relayd_uid
, relayd_gid
, NULL
);
1316 ERR("Create output file");
1320 if (stream
->tracefile_size
) {
1321 DBG("Tracefile %s/%s_0 created", stream
->path_name
, stream
->channel_name
);
1323 DBG("Tracefile %s/%s created", stream
->path_name
, stream
->channel_name
);
1326 if (!strncmp(stream
->channel_name
, DEFAULT_METADATA_NAME
, NAME_MAX
)) {
1327 stream
->metadata_flag
= 1;
1329 * When we receive a new metadata stream, we create a new
1330 * ctf_trace and we assign this ctf_trace to all streams with
1333 * If later on we receive a new stream for the same ctf_trace,
1334 * we copy the information from the first hit in the HT to the
1337 stream
->ctf_trace
= ctf_trace_create();
1338 if (!stream
->ctf_trace
) {
1342 stream
->ctf_trace
->refcount
++;
1343 stream
->ctf_trace
->metadata_stream
= stream
;
1345 ctf_trace_assign(cmd
->ctf_traces_ht
, stream
);
1346 stream
->ctf_traces_ht
= cmd
->ctf_traces_ht
;
1349 * Add the stream handle in the recv list of the connection. Once the end
1350 * stream message is received, this list is emptied and streams are set
1351 * with the viewer ready flag.
1353 if (stream
->metadata_flag
) {
1354 stream
->viewer_ready
= 1;
1356 queue_stream_handle(stream
->stream_handle
, cmd
);
1359 lttng_ht_node_init_ulong(&stream
->stream_n
,
1360 (unsigned long) stream
->stream_handle
);
1361 lttng_ht_add_unique_ulong(relay_streams_ht
,
1364 lttng_ht_node_init_str(&stream
->ctf_trace_node
, stream
->path_name
);
1365 lttng_ht_add_str(cmd
->ctf_traces_ht
, &stream
->ctf_trace_node
);
1366 session
->stream_count
++;
1368 DBG("Relay new stream added %s with ID %" PRIu64
, stream
->channel_name
,
1369 stream
->stream_handle
);
1372 reply
.handle
= htobe64(stream
->stream_handle
);
1373 /* send the session id to the client or a negative return code on error */
1375 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1376 /* stream was not properly added to the ht, so free it */
1379 reply
.ret_code
= htobe32(LTTNG_OK
);
1382 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1383 sizeof(struct lttcomm_relayd_status_stream
), 0);
1385 ERR("Relay sending stream id");
1394 free(stream
->path_name
);
1395 free(stream
->channel_name
);
1401 * relay_close_stream: close a specific stream
1404 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1405 struct relay_command
*cmd
)
1408 struct relay_session
*session
= cmd
->session
;
1409 struct lttcomm_relayd_close_stream stream_info
;
1410 struct lttcomm_relayd_generic_reply reply
;
1411 struct relay_stream
*stream
;
1413 DBG("Close stream received");
1415 if (!session
|| cmd
->version_check_done
== 0) {
1416 ERR("Trying to close a stream before version check");
1418 goto end_no_session
;
1421 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &stream_info
,
1422 sizeof(struct lttcomm_relayd_close_stream
), 0);
1423 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1425 /* Orderly shutdown. Not necessary to print an error. */
1426 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1428 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1431 goto end_no_session
;
1435 stream
= relay_stream_find_by_id(be64toh(stream_info
.stream_id
));
1441 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1442 stream
->close_flag
= 1;
1443 session
->stream_count
--;
1444 assert(session
->stream_count
>= 0);
1446 if (close_stream_check(stream
)) {
1447 destroy_stream(stream
);
1454 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1456 reply
.ret_code
= htobe32(LTTNG_OK
);
1458 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1459 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1461 ERR("Relay sending stream id");
1470 * relay_unknown_command: send -1 if received unknown command
1473 void relay_unknown_command(struct relay_command
*cmd
)
1475 struct lttcomm_relayd_generic_reply reply
;
1478 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1479 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1480 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1482 ERR("Relay sending unknown command");
1487 * relay_start: send an acknowledgment to the client to tell if we are
1488 * ready to receive data. We are ready if a session is established.
1491 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1492 struct relay_command
*cmd
)
1494 int ret
= htobe32(LTTNG_OK
);
1495 struct lttcomm_relayd_generic_reply reply
;
1496 struct relay_session
*session
= cmd
->session
;
1499 DBG("Trying to start the streaming without a session established");
1500 ret
= htobe32(LTTNG_ERR_UNK
);
1503 reply
.ret_code
= ret
;
1504 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1505 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1507 ERR("Relay sending start ack");
1514 * Append padding to the file pointed by the file descriptor fd.
1516 static int write_padding_to_file(int fd
, uint32_t size
)
1525 zeros
= zmalloc(size
);
1526 if (zeros
== NULL
) {
1527 PERROR("zmalloc zeros for padding");
1532 ret
= lttng_write(fd
, zeros
, size
);
1534 PERROR("write padding to file");
1544 * relay_recv_metadata: receive the metada for the session.
1547 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1548 struct relay_command
*cmd
)
1550 int ret
= htobe32(LTTNG_OK
);
1552 struct relay_session
*session
= cmd
->session
;
1553 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1554 struct relay_stream
*metadata_stream
;
1555 uint64_t data_size
, payload_size
;
1558 ERR("Metadata sent before version check");
1563 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1564 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1565 ERR("Incorrect data size");
1569 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1571 if (data_buffer_size
< data_size
) {
1572 /* In case the realloc fails, we can free the memory */
1575 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1576 if (!tmp_data_ptr
) {
1577 ERR("Allocating data buffer");
1582 data_buffer
= tmp_data_ptr
;
1583 data_buffer_size
= data_size
;
1585 memset(data_buffer
, 0, data_size
);
1586 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1587 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
1588 if (ret
< 0 || ret
!= data_size
) {
1590 /* Orderly shutdown. Not necessary to print an error. */
1591 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1593 ERR("Relay didn't receive the whole metadata");
1598 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1601 metadata_stream
= relay_stream_find_by_id(
1602 be64toh(metadata_struct
->stream_id
));
1603 if (!metadata_stream
) {
1608 size_ret
= lttng_write(metadata_stream
->fd
, metadata_struct
->payload
,
1610 if (size_ret
< payload_size
) {
1611 ERR("Relay error writing metadata on file");
1616 ret
= write_padding_to_file(metadata_stream
->fd
,
1617 be32toh(metadata_struct
->padding_size
));
1621 metadata_stream
->ctf_trace
->metadata_received
+=
1622 payload_size
+ be32toh(metadata_struct
->padding_size
);
1624 DBG2("Relay metadata written");
1633 * relay_send_version: send relayd version number
1636 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1637 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1640 struct lttcomm_relayd_version reply
, msg
;
1644 cmd
->version_check_done
= 1;
1646 /* Get version from the other side. */
1647 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1648 if (ret
< 0 || ret
!= sizeof(msg
)) {
1650 /* Orderly shutdown. Not necessary to print an error. */
1651 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1653 ERR("Relay failed to receive the version values.");
1659 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1660 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1662 /* Major versions must be the same */
1663 if (reply
.major
!= be32toh(msg
.major
)) {
1664 DBG("Incompatible major versions (%u vs %u), deleting session",
1665 reply
.major
, be32toh(msg
.major
));
1666 relay_delete_session(cmd
, sessions_ht
);
1671 cmd
->major
= reply
.major
;
1672 /* We adapt to the lowest compatible version */
1673 if (reply
.minor
<= be32toh(msg
.minor
)) {
1674 cmd
->minor
= reply
.minor
;
1676 cmd
->minor
= be32toh(msg
.minor
);
1679 reply
.major
= htobe32(reply
.major
);
1680 reply
.minor
= htobe32(reply
.minor
);
1681 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1682 sizeof(struct lttcomm_relayd_version
), 0);
1684 ERR("Relay sending version");
1687 DBG("Version check done using protocol %u.%u", cmd
->major
,
1695 * Check for data pending for a given stream id from the session daemon.
1698 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1699 struct relay_command
*cmd
)
1701 struct relay_session
*session
= cmd
->session
;
1702 struct lttcomm_relayd_data_pending msg
;
1703 struct lttcomm_relayd_generic_reply reply
;
1704 struct relay_stream
*stream
;
1706 uint64_t last_net_seq_num
, stream_id
;
1708 DBG("Data pending command received");
1710 if (!session
|| cmd
->version_check_done
== 0) {
1711 ERR("Trying to check for data before version check");
1713 goto end_no_session
;
1716 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1717 if (ret
< sizeof(msg
)) {
1719 /* Orderly shutdown. Not necessary to print an error. */
1720 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1722 ERR("Relay didn't receive valid data_pending struct size : %d",
1726 goto end_no_session
;
1729 stream_id
= be64toh(msg
.stream_id
);
1730 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1733 stream
= relay_stream_find_by_id(stream_id
);
1734 if (stream
== NULL
) {
1739 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1740 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1743 /* Avoid wrapping issue */
1744 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1745 /* Data has in fact been written and is NOT pending */
1748 /* Data still being streamed thus pending */
1752 /* Pending check is now done. */
1753 stream
->data_pending_check_done
= 1;
1758 reply
.ret_code
= htobe32(ret
);
1759 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1761 ERR("Relay data pending ret code failed");
1769 * Wait for the control socket to reach a quiescent state.
1771 * Note that for now, when receiving this command from the session daemon, this
1772 * means that every subsequent commands or data received on the control socket
1773 * has been handled. So, this is why we simply return OK here.
1776 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1777 struct relay_command
*cmd
)
1781 struct relay_stream
*stream
;
1782 struct lttng_ht_iter iter
;
1783 struct lttcomm_relayd_quiescent_control msg
;
1784 struct lttcomm_relayd_generic_reply reply
;
1786 DBG("Checking quiescent state on control socket");
1788 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1789 ERR("Trying to check for data before version check");
1791 goto end_no_session
;
1794 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1795 if (ret
< sizeof(msg
)) {
1797 /* Orderly shutdown. Not necessary to print an error. */
1798 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1800 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1804 goto end_no_session
;
1807 stream_id
= be64toh(msg
.stream_id
);
1810 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1812 if (stream
->stream_handle
== stream_id
) {
1813 stream
->data_pending_check_done
= 1;
1814 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1821 reply
.ret_code
= htobe32(LTTNG_OK
);
1822 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1824 ERR("Relay data quiescent control ret code failed");
1832 * Initialize a data pending command. This means that a client is about to ask
1833 * for data pending for each stream he/she holds. Simply iterate over all
1834 * streams of a session and set the data_pending_check_done flag.
1836 * This command returns to the client a LTTNG_OK code.
1839 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1840 struct relay_command
*cmd
)
1843 struct lttng_ht_iter iter
;
1844 struct lttcomm_relayd_begin_data_pending msg
;
1845 struct lttcomm_relayd_generic_reply reply
;
1846 struct relay_stream
*stream
;
1847 uint64_t session_id
;
1852 DBG("Init streams for data pending");
1854 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1855 ERR("Trying to check for data before version check");
1857 goto end_no_session
;
1860 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1861 if (ret
< sizeof(msg
)) {
1863 /* Orderly shutdown. Not necessary to print an error. */
1864 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1866 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1870 goto end_no_session
;
1873 session_id
= be64toh(msg
.session_id
);
1876 * Iterate over all streams to set the begin data pending flag. For now, the
1877 * streams are indexed by stream handle so we have to iterate over all
1878 * streams to find the one associated with the right session_id.
1881 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1883 if (stream
->session
->id
== session_id
) {
1884 stream
->data_pending_check_done
= 0;
1885 DBG("Set begin data pending flag to stream %" PRIu64
,
1886 stream
->stream_handle
);
1891 /* All good, send back reply. */
1892 reply
.ret_code
= htobe32(LTTNG_OK
);
1894 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1896 ERR("Relay begin data pending send reply failed");
1904 * End data pending command. This will check, for a given session id, if each
1905 * stream associated with it has its data_pending_check_done flag set. If not,
1906 * this means that the client lost track of the stream but the data is still
1907 * being streamed on our side. In this case, we inform the client that data is
1910 * Return to the client if there is data in flight or not with a ret_code.
1913 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1914 struct relay_command
*cmd
)
1917 struct lttng_ht_iter iter
;
1918 struct lttcomm_relayd_end_data_pending msg
;
1919 struct lttcomm_relayd_generic_reply reply
;
1920 struct relay_stream
*stream
;
1921 uint64_t session_id
;
1922 uint32_t is_data_inflight
= 0;
1927 DBG("End data pending command");
1929 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1930 ERR("Trying to check for data before version check");
1932 goto end_no_session
;
1935 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1936 if (ret
< sizeof(msg
)) {
1938 /* Orderly shutdown. Not necessary to print an error. */
1939 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1941 ERR("Relay didn't receive valid end data_pending struct size: %d",
1945 goto end_no_session
;
1948 session_id
= be64toh(msg
.session_id
);
1950 /* Iterate over all streams to see if the begin data pending flag is set. */
1952 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1954 if (stream
->session
->id
== session_id
&&
1955 !stream
->data_pending_check_done
) {
1956 is_data_inflight
= 1;
1957 DBG("Data is still in flight for stream %" PRIu64
,
1958 stream
->stream_handle
);
1964 /* All good, send back reply. */
1965 reply
.ret_code
= htobe32(is_data_inflight
);
1967 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1969 ERR("Relay end data pending send reply failed");
1977 * Receive an index for a specific stream.
1979 * Return 0 on success else a negative value.
1982 int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1983 struct relay_command
*cmd
)
1985 int ret
, send_ret
, index_created
= 0;
1986 struct relay_session
*session
= cmd
->session
;
1987 struct lttcomm_relayd_index index_info
;
1988 struct relay_index
*index
, *wr_index
= NULL
;
1989 struct lttcomm_relayd_generic_reply reply
;
1990 struct relay_stream
*stream
;
1991 uint64_t net_seq_num
;
1995 DBG("Relay receiving index");
1997 if (!session
|| cmd
->version_check_done
== 0) {
1998 ERR("Trying to close a stream before version check");
2000 goto end_no_session
;
2003 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &index_info
,
2004 sizeof(index_info
), 0);
2005 if (ret
< sizeof(index_info
)) {
2007 /* Orderly shutdown. Not necessary to print an error. */
2008 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
2010 ERR("Relay didn't receive valid index struct size : %d", ret
);
2013 goto end_no_session
;
2016 net_seq_num
= be64toh(index_info
.net_seq_num
);
2019 stream
= relay_stream_find_by_id(be64toh(index_info
.relay_stream_id
));
2022 goto end_rcu_unlock
;
2025 /* Live beacon handling */
2026 if (index_info
.packet_size
== 0) {
2027 DBG("Received live beacon for stream %" PRIu64
, stream
->stream_handle
);
2030 * Only flag a stream inactive when it has already received data.
2032 if (stream
->total_index_received
> 0) {
2033 stream
->beacon_ts_end
= be64toh(index_info
.timestamp_end
);
2036 goto end_rcu_unlock
;
2038 stream
->beacon_ts_end
= -1ULL;
2041 index
= relay_index_find(stream
->stream_handle
, net_seq_num
);
2043 /* A successful creation will add the object to the HT. */
2044 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
2046 goto end_rcu_unlock
;
2051 copy_index_control_data(index
, &index_info
);
2053 if (index_created
) {
2055 * Try to add the relay index object to the hash table. If an object
2056 * already exist, destroy back the index created, set the data in this
2057 * object and write it on disk.
2059 relay_index_add(index
, &wr_index
);
2061 copy_index_control_data(wr_index
, &index_info
);
2065 /* The index already exists so write it on disk. */
2069 /* Do we have a writable ready index to write on disk. */
2071 /* Starting at 2.4, create the index file if none available. */
2072 if (cmd
->minor
>= 4 && stream
->index_fd
< 0) {
2073 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
2074 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
2075 stream
->tracefile_count_current
);
2077 goto end_rcu_unlock
;
2079 stream
->index_fd
= ret
;
2082 ret
= relay_index_write(wr_index
->fd
, wr_index
);
2084 goto end_rcu_unlock
;
2086 stream
->total_index_received
++;
2093 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2095 reply
.ret_code
= htobe32(LTTNG_OK
);
2097 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
2099 ERR("Relay sending close index id reply");
2108 * Receive the streams_sent message.
2110 * Return 0 on success else a negative value.
2113 int relay_streams_sent(struct lttcomm_relayd_hdr
*recv_hdr
,
2114 struct relay_command
*cmd
)
2117 struct lttcomm_relayd_generic_reply reply
;
2121 DBG("Relay receiving streams_sent");
2123 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
2124 ERR("Trying to close a stream before version check");
2126 goto end_no_session
;
2130 * Flag every pending stream in the connection recv list that they are
2131 * ready to be used by the viewer.
2133 set_viewer_ready_flag(cmd
);
2135 reply
.ret_code
= htobe32(LTTNG_OK
);
2136 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
2138 ERR("Relay sending sent_stream reply");
2150 * Process the commands received on the control socket
2153 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
2154 struct relay_command
*cmd
, struct relay_local_data
*ctx
)
2158 switch (be32toh(recv_hdr
->cmd
)) {
2159 case RELAYD_CREATE_SESSION
:
2160 ret
= relay_create_session(recv_hdr
, cmd
, ctx
->sessions_ht
);
2162 case RELAYD_ADD_STREAM
:
2163 ret
= relay_add_stream(recv_hdr
, cmd
, ctx
->sessions_ht
);
2165 case RELAYD_START_DATA
:
2166 ret
= relay_start(recv_hdr
, cmd
);
2168 case RELAYD_SEND_METADATA
:
2169 ret
= relay_recv_metadata(recv_hdr
, cmd
);
2171 case RELAYD_VERSION
:
2172 ret
= relay_send_version(recv_hdr
, cmd
, ctx
->sessions_ht
);
2174 case RELAYD_CLOSE_STREAM
:
2175 ret
= relay_close_stream(recv_hdr
, cmd
);
2177 case RELAYD_DATA_PENDING
:
2178 ret
= relay_data_pending(recv_hdr
, cmd
);
2180 case RELAYD_QUIESCENT_CONTROL
:
2181 ret
= relay_quiescent_control(recv_hdr
, cmd
);
2183 case RELAYD_BEGIN_DATA_PENDING
:
2184 ret
= relay_begin_data_pending(recv_hdr
, cmd
);
2186 case RELAYD_END_DATA_PENDING
:
2187 ret
= relay_end_data_pending(recv_hdr
, cmd
);
2189 case RELAYD_SEND_INDEX
:
2190 ret
= relay_recv_index(recv_hdr
, cmd
);
2192 case RELAYD_STREAMS_SENT
:
2193 ret
= relay_streams_sent(recv_hdr
, cmd
);
2195 case RELAYD_UPDATE_SYNC_INFO
:
2197 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
2198 relay_unknown_command(cmd
);
2208 * Handle index for a data stream.
2210 * RCU read side lock MUST be acquired.
2212 * Return 0 on success else a negative value.
2214 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2217 int ret
= 0, index_created
= 0;
2218 uint64_t stream_id
, data_offset
;
2219 struct relay_index
*index
, *wr_index
= NULL
;
2223 stream_id
= stream
->stream_handle
;
2224 /* Get data offset because we are about to update the index. */
2225 data_offset
= htobe64(stream
->tracefile_size_current
);
2228 * Lookup for an existing index for that stream id/sequence number. If on
2229 * exists, the control thread already received the data for it thus we need
2230 * to write it on disk.
2232 index
= relay_index_find(stream_id
, net_seq_num
);
2234 /* A successful creation will add the object to the HT. */
2235 index
= relay_index_create(stream_id
, net_seq_num
);
2243 if (rotate_index
|| stream
->index_fd
< 0) {
2244 index
->to_close_fd
= stream
->index_fd
;
2245 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
2246 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
2247 stream
->tracefile_count_current
);
2249 /* This will close the stream's index fd if one. */
2250 relay_index_free_safe(index
);
2253 stream
->index_fd
= ret
;
2255 index
->fd
= stream
->index_fd
;
2256 index
->index_data
.offset
= data_offset
;
2258 if (index_created
) {
2260 * Try to add the relay index object to the hash table. If an object
2261 * already exist, destroy back the index created and set the data.
2263 relay_index_add(index
, &wr_index
);
2265 /* Copy back data from the created index. */
2266 wr_index
->fd
= index
->fd
;
2267 wr_index
->to_close_fd
= index
->to_close_fd
;
2268 wr_index
->index_data
.offset
= data_offset
;
2272 /* The index already exists so write it on disk. */
2276 /* Do we have a writable ready index to write on disk. */
2278 ret
= relay_index_write(wr_index
->fd
, wr_index
);
2282 stream
->total_index_received
++;
2290 * relay_process_data: Process the data received on the data socket
2293 int relay_process_data(struct relay_command
*cmd
)
2295 int ret
= 0, rotate_index
= 0;
2297 struct relay_stream
*stream
;
2298 struct lttcomm_relayd_data_hdr data_hdr
;
2300 uint64_t net_seq_num
;
2303 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &data_hdr
,
2304 sizeof(struct lttcomm_relayd_data_hdr
), 0);
2307 /* Orderly shutdown. Not necessary to print an error. */
2308 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
2310 ERR("Unable to receive data header on sock %d", cmd
->sock
->fd
);
2316 stream_id
= be64toh(data_hdr
.stream_id
);
2319 stream
= relay_stream_find_by_id(stream_id
);
2322 goto end_rcu_unlock
;
2325 data_size
= be32toh(data_hdr
.data_size
);
2326 if (data_buffer_size
< data_size
) {
2329 tmp_data_ptr
= realloc(data_buffer
, data_size
);
2330 if (!tmp_data_ptr
) {
2331 ERR("Allocating data buffer");
2334 goto end_rcu_unlock
;
2336 data_buffer
= tmp_data_ptr
;
2337 data_buffer_size
= data_size
;
2339 memset(data_buffer
, 0, data_size
);
2341 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2343 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2344 data_size
, stream_id
, net_seq_num
);
2345 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
2348 /* Orderly shutdown. Not necessary to print an error. */
2349 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
2352 goto end_rcu_unlock
;
2355 /* Check if a rotation is needed. */
2356 if (stream
->tracefile_size
> 0 &&
2357 (stream
->tracefile_size_current
+ data_size
) >
2358 stream
->tracefile_size
) {
2359 struct relay_viewer_stream
*vstream
;
2362 new_id
= (stream
->tracefile_count_current
+ 1) %
2363 stream
->tracefile_count
;
2365 * When we wrap-around back to 0, we start overwriting old
2368 if (!stream
->tracefile_overwrite
&& new_id
== 0) {
2369 stream
->tracefile_overwrite
= 1;
2371 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
2372 if (stream
->tracefile_overwrite
) {
2373 stream
->oldest_tracefile_id
=
2374 (stream
->oldest_tracefile_id
+ 1) %
2375 stream
->tracefile_count
;
2377 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
);
2380 * The viewer is reading a file about to be
2381 * overwritten. Close the FDs it is
2382 * currently using and let it handle the fault.
2384 if (vstream
->tracefile_count_current
== new_id
) {
2385 pthread_mutex_lock(&vstream
->overwrite_lock
);
2386 vstream
->abort_flag
= 1;
2387 pthread_mutex_unlock(&vstream
->overwrite_lock
);
2388 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2389 stream
->channel_name
, new_id
);
2390 } else if (vstream
->tracefile_count_current
==
2391 stream
->tracefile_count_current
) {
2393 * The reader and writer were in the
2394 * same trace file, inform the viewer
2395 * that no new index will ever be added
2398 vstream
->close_write_flag
= 1;
2401 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
2402 stream
->tracefile_size
, stream
->tracefile_count
,
2403 relayd_uid
, relayd_gid
, stream
->fd
,
2404 &(stream
->tracefile_count_current
), &stream
->fd
);
2405 stream
->total_index_received
= 0;
2406 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
2408 ERR("Rotating stream output file");
2409 goto end_rcu_unlock
;
2411 /* Reset current size because we just perform a stream rotation. */
2412 stream
->tracefile_size_current
= 0;
2417 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2418 * index are NOT supported.
2420 if (stream
->session
->minor
>= 4 && !stream
->session
->snapshot
) {
2421 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2423 goto end_rcu_unlock
;
2427 /* Write data to stream output fd. */
2428 size_ret
= lttng_write(stream
->fd
, data_buffer
, data_size
);
2429 if (size_ret
< data_size
) {
2430 ERR("Relay error writing data to file");
2432 goto end_rcu_unlock
;
2435 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
2436 ret
, stream
->stream_handle
);
2438 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
2440 goto end_rcu_unlock
;
2442 stream
->tracefile_size_current
+= data_size
+ be32toh(data_hdr
.padding_size
);
2444 stream
->prev_seq
= net_seq_num
;
2446 /* Check if we need to close the FD */
2447 if (close_stream_check(stream
)) {
2448 destroy_stream(stream
);
2458 void relay_cleanup_poll_connection(struct lttng_poll_event
*events
, int pollfd
)
2462 lttng_poll_del(events
, pollfd
);
2464 ret
= close(pollfd
);
2466 ERR("Closing pollfd %d", pollfd
);
2471 int relay_add_connection(int fd
, struct lttng_poll_event
*events
,
2472 struct lttng_ht
*relay_connections_ht
)
2474 struct relay_command
*relay_connection
;
2477 relay_connection
= zmalloc(sizeof(struct relay_command
));
2478 if (relay_connection
== NULL
) {
2479 PERROR("Relay command zmalloc");
2482 ret
= lttng_read(fd
, relay_connection
, sizeof(struct relay_command
));
2483 if (ret
< sizeof(struct relay_command
)) {
2484 PERROR("read relay cmd pipe");
2487 CDS_INIT_LIST_HEAD(&relay_connection
->recv_head
);
2490 * Only used by the control side and the reference is copied inside each
2491 * stream from that connection. Thus a destroy HT must be done after every
2492 * stream has been destroyed.
2494 if (relay_connection
->type
== RELAY_CONTROL
) {
2495 relay_connection
->ctf_traces_ht
= lttng_ht_new(0,
2496 LTTNG_HT_TYPE_STRING
);
2497 if (!relay_connection
->ctf_traces_ht
) {
2502 lttng_ht_node_init_ulong(&relay_connection
->sock_n
,
2503 (unsigned long) relay_connection
->sock
->fd
);
2505 lttng_ht_add_unique_ulong(relay_connections_ht
,
2506 &relay_connection
->sock_n
);
2508 return lttng_poll_add(events
,
2509 relay_connection
->sock
->fd
,
2510 LPOLLIN
| LPOLLRDHUP
);
2513 free(relay_connection
);
2519 void deferred_free_connection(struct rcu_head
*head
)
2521 struct relay_command
*relay_connection
=
2522 caa_container_of(head
, struct relay_command
, rcu_node
);
2524 lttcomm_destroy_sock(relay_connection
->sock
);
2525 free(relay_connection
);
2529 void relay_del_connection(struct lttng_ht
*relay_connections_ht
,
2530 struct lttng_ht_iter
*iter
, struct relay_command
*relay_connection
,
2531 struct lttng_ht
*sessions_ht
)
2535 ret
= lttng_ht_del(relay_connections_ht
, iter
);
2538 if (relay_connection
->type
== RELAY_CONTROL
) {
2539 struct relay_stream_recv_handle
*node
, *tmp_node
;
2541 relay_delete_session(relay_connection
, sessions_ht
);
2542 lttng_ht_destroy(relay_connection
->ctf_traces_ht
);
2544 /* Clean up recv list. */
2545 cds_list_for_each_entry_safe(node
, tmp_node
,
2546 &relay_connection
->recv_head
, node
) {
2547 cds_list_del(&node
->node
);
2552 call_rcu(&relay_connection
->rcu_node
, deferred_free_connection
);
2556 * This thread does the actual work
2559 void *relay_thread_worker(void *data
)
2561 int ret
, err
= -1, last_seen_data_fd
= -1;
2563 struct relay_command
*relay_connection
;
2564 struct lttng_poll_event events
;
2565 struct lttng_ht
*relay_connections_ht
;
2566 struct lttng_ht_node_ulong
*node
;
2567 struct lttng_ht_iter iter
;
2568 struct lttcomm_relayd_hdr recv_hdr
;
2569 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
2570 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
2572 DBG("[thread] Relay worker started");
2574 rcu_register_thread();
2576 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2578 if (testpoint(relayd_thread_worker
)) {
2579 goto error_testpoint
;
2582 health_code_update();
2584 /* table of connections indexed on socket */
2585 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2586 if (!relay_connections_ht
) {
2587 goto relay_connections_ht_error
;
2590 /* Tables of received indexes indexed by index handle and net_seq_num. */
2591 indexes_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64
);
2593 goto indexes_ht_error
;
2596 ret
= create_thread_poll_set(&events
, 2);
2598 goto error_poll_create
;
2601 ret
= lttng_poll_add(&events
, relay_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2608 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2610 health_code_update();
2612 /* Infinite blocking call, waiting for transmission */
2613 DBG3("Relayd worker thread polling...");
2614 health_poll_entry();
2615 ret
= lttng_poll_wait(&events
, -1);
2619 * Restart interrupted system call.
2621 if (errno
== EINTR
) {
2630 * Process control. The control connection is prioritised so we don't
2631 * starve it with high throughout put tracing data on the data
2634 for (i
= 0; i
< nb_fd
; i
++) {
2635 /* Fetch once the poll data */
2636 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2637 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2639 health_code_update();
2641 /* Thread quit pipe has been closed. Killing thread. */
2642 ret
= check_thread_quit_pipe(pollfd
, revents
);
2648 /* Inspect the relay cmd pipe for new connection */
2649 if (pollfd
== relay_cmd_pipe
[0]) {
2650 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2651 ERR("Relay pipe error");
2653 } else if (revents
& LPOLLIN
) {
2654 DBG("Relay command received");
2655 ret
= relay_add_connection(relay_cmd_pipe
[0],
2656 &events
, relay_connections_ht
);
2661 } else if (revents
) {
2663 lttng_ht_lookup(relay_connections_ht
,
2664 (void *)((unsigned long) pollfd
),
2666 node
= lttng_ht_iter_get_node_ulong(&iter
);
2668 DBG2("Relay sock %d not found", pollfd
);
2672 relay_connection
= caa_container_of(node
,
2673 struct relay_command
, sock_n
);
2675 if (revents
& (LPOLLERR
)) {
2677 relay_cleanup_poll_connection(&events
, pollfd
);
2678 relay_del_connection(relay_connections_ht
,
2679 &iter
, relay_connection
, sessions_ht
);
2680 if (last_seen_data_fd
== pollfd
) {
2681 last_seen_data_fd
= last_notdel_data_fd
;
2683 } else if (revents
& (LPOLLHUP
| LPOLLRDHUP
)) {
2684 DBG("Socket %d hung up", pollfd
);
2685 relay_cleanup_poll_connection(&events
, pollfd
);
2686 relay_del_connection(relay_connections_ht
,
2687 &iter
, relay_connection
, sessions_ht
);
2688 if (last_seen_data_fd
== pollfd
) {
2689 last_seen_data_fd
= last_notdel_data_fd
;
2691 } else if (revents
& LPOLLIN
) {
2692 /* control socket */
2693 if (relay_connection
->type
== RELAY_CONTROL
) {
2694 ret
= relay_connection
->sock
->ops
->recvmsg(
2695 relay_connection
->sock
, &recv_hdr
,
2696 sizeof(struct lttcomm_relayd_hdr
), 0);
2697 /* connection closed */
2699 relay_cleanup_poll_connection(&events
, pollfd
);
2700 relay_del_connection(relay_connections_ht
,
2701 &iter
, relay_connection
, sessions_ht
);
2702 DBG("Control connection closed with %d", pollfd
);
2704 if (relay_connection
->session
) {
2705 DBG2("Relay worker receiving data for session : %" PRIu64
,
2706 relay_connection
->session
->id
);
2708 ret
= relay_process_control(&recv_hdr
,
2709 relay_connection
, relay_ctx
);
2711 /* Clear the session on error. */
2712 relay_cleanup_poll_connection(&events
, pollfd
);
2713 relay_del_connection(relay_connections_ht
,
2714 &iter
, relay_connection
, sessions_ht
);
2715 DBG("Connection closed with %d", pollfd
);
2721 * Flag the last seen data fd not deleted. It will be
2722 * used as the last seen fd if any fd gets deleted in
2725 last_notdel_data_fd
= pollfd
;
2733 * The last loop handled a control request, go back to poll to make
2734 * sure we prioritise the control socket.
2740 if (last_seen_data_fd
>= 0) {
2741 for (i
= 0; i
< nb_fd
; i
++) {
2742 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2744 health_code_update();
2746 if (last_seen_data_fd
== pollfd
) {
2753 /* Process data connection. */
2754 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2755 /* Fetch the poll data. */
2756 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2757 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2759 health_code_update();
2761 /* Skip the command pipe. It's handled in the first loop. */
2762 if (pollfd
== relay_cmd_pipe
[0]) {
2768 lttng_ht_lookup(relay_connections_ht
,
2769 (void *)((unsigned long) pollfd
),
2771 node
= lttng_ht_iter_get_node_ulong(&iter
);
2773 /* Skip it. Might be removed before. */
2777 relay_connection
= caa_container_of(node
,
2778 struct relay_command
, sock_n
);
2780 if (revents
& LPOLLIN
) {
2781 if (relay_connection
->type
!= RELAY_DATA
) {
2785 ret
= relay_process_data(relay_connection
);
2786 /* connection closed */
2788 relay_cleanup_poll_connection(&events
, pollfd
);
2789 relay_del_connection(relay_connections_ht
,
2790 &iter
, relay_connection
, sessions_ht
);
2791 DBG("Data connection closed with %d", pollfd
);
2793 * Every goto restart call sets the last seen fd where
2794 * here we don't really care since we gracefully
2795 * continue the loop after the connection is deleted.
2798 /* Keep last seen port. */
2799 last_seen_data_fd
= pollfd
;
2807 last_seen_data_fd
= -1;
2810 /* Normal exit, no error */
2815 lttng_poll_clean(&events
);
2817 /* empty the hash table and free the memory */
2819 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, node
, node
) {
2820 health_code_update();
2822 node
= lttng_ht_iter_get_node_ulong(&iter
);
2824 relay_connection
= caa_container_of(node
,
2825 struct relay_command
, sock_n
);
2826 relay_del_connection(relay_connections_ht
,
2827 &iter
, relay_connection
, sessions_ht
);
2832 lttng_ht_destroy(indexes_ht
);
2834 lttng_ht_destroy(relay_connections_ht
);
2835 relay_connections_ht_error
:
2836 /* Close relay cmd pipes */
2837 utils_close_pipe(relay_cmd_pipe
);
2839 DBG("Thread exited with error");
2841 DBG("Worker thread cleanup complete");
2846 ERR("Health error occurred in %s", __func__
);
2848 health_unregister(health_relayd
);
2849 rcu_unregister_thread();
2855 * Create the relay command pipe to wake thread_manage_apps.
2856 * Closed in cleanup().
2858 static int create_relay_cmd_pipe(void)
2862 ret
= utils_create_pipe_cloexec(relay_cmd_pipe
);
2870 int main(int argc
, char **argv
)
2874 struct relay_local_data
*relay_ctx
;
2876 /* Parse arguments */
2878 if ((ret
= set_options(argc
, argv
)) < 0) {
2882 if ((ret
= set_signal_handler()) < 0) {
2886 /* Try to create directory if -o, --output is specified. */
2887 if (opt_output_path
) {
2888 if (*opt_output_path
!= '/') {
2889 ERR("Please specify an absolute path for -o, --output PATH");
2893 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2895 ERR("Unable to create %s", opt_output_path
);
2901 if (opt_daemon
|| opt_background
) {
2904 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
2911 * We are in the child. Make sure all other file
2912 * descriptors are closed, in case we are called with
2913 * more opened file descriptors than the standard ones.
2915 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
2920 /* Create thread quit pipe */
2921 if ((ret
= init_thread_quit_pipe()) < 0) {
2925 /* We need those values for the file/dir creation. */
2926 relayd_uid
= getuid();
2927 relayd_gid
= getgid();
2929 /* Check if daemon is UID = 0 */
2930 if (relayd_uid
== 0) {
2931 if (control_uri
->port
< 1024 || data_uri
->port
< 1024 || live_uri
->port
< 1024) {
2932 ERR("Need to be root to use ports < 1024");
2938 /* Setup the thread apps communication pipe. */
2939 if ((ret
= create_relay_cmd_pipe()) < 0) {
2943 /* Init relay command queue. */
2944 cds_wfq_init(&relay_cmd_queue
.queue
);
2946 /* Set up max poll set size */
2947 lttng_poll_set_max_size();
2949 /* Initialize communication library */
2951 lttcomm_inet_init();
2953 relay_ctx
= zmalloc(sizeof(struct relay_local_data
));
2955 PERROR("relay_ctx");
2959 /* tables of sessions indexed by session ID */
2960 relay_ctx
->sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2961 if (!relay_ctx
->sessions_ht
) {
2962 goto exit_relay_ctx_sessions
;
2965 /* tables of streams indexed by stream ID */
2966 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2967 if (!relay_streams_ht
) {
2968 goto exit_relay_ctx_streams
;
2971 /* tables of streams indexed by stream ID */
2972 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2973 if (!viewer_streams_ht
) {
2974 goto exit_relay_ctx_viewer_streams
;
2977 /* Initialize thread health monitoring */
2978 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
2979 if (!health_relayd
) {
2980 PERROR("health_app_create error");
2981 goto exit_health_app_create
;
2984 ret
= utils_create_pipe(health_quit_pipe
);
2986 goto error_health_pipe
;
2989 /* Create thread to manage the client socket */
2990 ret
= pthread_create(&health_thread
, NULL
,
2991 thread_manage_health
, (void *) NULL
);
2993 PERROR("pthread_create health");
2997 /* Setup the dispatcher thread */
2998 ret
= pthread_create(&dispatcher_thread
, NULL
,
2999 relay_thread_dispatcher
, (void *) NULL
);
3001 PERROR("pthread_create dispatcher");
3002 goto exit_dispatcher
;
3005 /* Setup the worker thread */
3006 ret
= pthread_create(&worker_thread
, NULL
,
3007 relay_thread_worker
, (void *) relay_ctx
);
3009 PERROR("pthread_create worker");
3013 /* Setup the listener thread */
3014 ret
= pthread_create(&listener_thread
, NULL
,
3015 relay_thread_listener
, (void *) NULL
);
3017 PERROR("pthread_create listener");
3021 ret
= live_start_threads(live_uri
, relay_ctx
);
3023 ERR("Starting live viewer threads");
3028 ret
= pthread_join(listener_thread
, &status
);
3030 PERROR("pthread_join");
3031 goto error
; /* join error, exit without cleanup */
3035 ret
= pthread_join(worker_thread
, &status
);
3037 PERROR("pthread_join");
3038 goto error
; /* join error, exit without cleanup */
3042 ret
= pthread_join(dispatcher_thread
, &status
);
3044 PERROR("pthread_join");
3045 goto error
; /* join error, exit without cleanup */
3049 ret
= pthread_join(health_thread
, &status
);
3051 PERROR("pthread_join health thread");
3052 goto error
; /* join error, exit without cleanup */
3056 * Stop live threads only after joining other threads.
3058 live_stop_threads();
3061 utils_close_pipe(health_quit_pipe
);
3064 health_app_destroy(health_relayd
);
3066 exit_health_app_create
:
3067 lttng_ht_destroy(viewer_streams_ht
);
3069 exit_relay_ctx_viewer_streams
:
3070 lttng_ht_destroy(relay_streams_ht
);
3072 exit_relay_ctx_streams
:
3073 lttng_ht_destroy(relay_ctx
->sessions_ht
);
3075 exit_relay_ctx_sessions
: