2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
35 #include <sys/types.h>
38 #include <urcu/futex.h>
39 #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/compat/endian.h>
48 #include <common/compat/getenv.h>
49 #include <common/defaults.h>
50 #include <common/daemonize.h>
51 #include <common/futex.h>
52 #include <common/sessiond-comm/sessiond-comm.h>
53 #include <common/sessiond-comm/inet.h>
54 #include <common/sessiond-comm/relayd.h>
55 #include <common/uri.h>
56 #include <common/utils.h>
57 #include <common/config/session-config.h>
58 #include <urcu/rculist.h>
61 #include "ctf-trace.h"
64 #include "lttng-relayd.h"
66 #include "health-relayd.h"
67 #include "testpoint.h"
68 #include "viewer-stream.h"
71 #include "connection.h"
72 #include "tracefile-array.h"
74 static const char *help_msg
=
75 #ifdef LTTNG_EMBED_HELP
76 #include <lttng-relayd.8.h>
82 /* command line options */
83 char *opt_output_path
;
84 static int opt_daemon
, opt_background
;
87 * We need to wait for listener and live listener threads, as well as
88 * health check thread, before being ready to signal readiness.
90 #define NR_LTTNG_RELAY_READY 3
91 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
93 /* Size of receive buffer. */
94 #define RECV_DATA_BUFFER_SIZE 65536
96 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
97 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
99 static struct lttng_uri
*control_uri
;
100 static struct lttng_uri
*data_uri
;
101 static struct lttng_uri
*live_uri
;
103 const char *progname
;
105 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
106 static int tracing_group_name_override
;
108 const char * const config_section_name
= "relayd";
111 * Quit pipe for all threads. This permits a single cancellation point
112 * for all threads when receiving an event on the pipe.
114 int thread_quit_pipe
[2] = { -1, -1 };
117 * This pipe is used to inform the worker thread that a command is queued and
118 * ready to be processed.
120 static int relay_conn_pipe
[2] = { -1, -1 };
122 /* Shared between threads */
123 static int dispatch_thread_exit
;
125 static pthread_t listener_thread
;
126 static pthread_t dispatcher_thread
;
127 static pthread_t worker_thread
;
128 static pthread_t health_thread
;
131 * last_relay_stream_id_lock protects last_relay_stream_id increment
132 * atomicity on 32-bit architectures.
134 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
135 static uint64_t last_relay_stream_id
;
138 * Relay command queue.
140 * The relay_thread_listener and relay_thread_dispatcher communicate with this
143 static struct relay_conn_queue relay_conn_queue
;
145 /* buffer allocated at startup, used to store the trace data */
146 static char *data_buffer
;
147 static unsigned int data_buffer_size
;
149 /* Global relay stream hash table. */
150 struct lttng_ht
*relay_streams_ht
;
152 /* Global relay viewer stream hash table. */
153 struct lttng_ht
*viewer_streams_ht
;
155 /* Global relay sessions hash table. */
156 struct lttng_ht
*sessions_ht
;
158 /* Relayd health monitoring */
159 struct health_app
*health_relayd
;
161 static struct option long_options
[] = {
162 { "control-port", 1, 0, 'C', },
163 { "data-port", 1, 0, 'D', },
164 { "live-port", 1, 0, 'L', },
165 { "daemonize", 0, 0, 'd', },
166 { "background", 0, 0, 'b', },
167 { "group", 1, 0, 'g', },
168 { "help", 0, 0, 'h', },
169 { "output", 1, 0, 'o', },
170 { "verbose", 0, 0, 'v', },
171 { "config", 1, 0, 'f' },
172 { "version", 0, 0, 'V' },
176 static const char *config_ignore_options
[] = { "help", "config", "version" };
179 * Take an option from the getopt output and set it in the right variable to be
182 * Return 0 on success else a negative value.
184 static int set_option(int opt
, const char *arg
, const char *optname
)
190 fprintf(stderr
, "option %s", optname
);
192 fprintf(stderr
, " with arg %s\n", arg
);
196 if (lttng_is_setuid_setgid()) {
197 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
198 "-C, --control-port");
200 ret
= uri_parse(arg
, &control_uri
);
202 ERR("Invalid control URI specified");
205 if (control_uri
->port
== 0) {
206 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
211 if (lttng_is_setuid_setgid()) {
212 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
215 ret
= uri_parse(arg
, &data_uri
);
217 ERR("Invalid data URI specified");
220 if (data_uri
->port
== 0) {
221 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
226 if (lttng_is_setuid_setgid()) {
227 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
230 ret
= uri_parse(arg
, &live_uri
);
232 ERR("Invalid live URI specified");
235 if (live_uri
->port
== 0) {
236 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
247 if (lttng_is_setuid_setgid()) {
248 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
251 tracing_group_name
= strdup(arg
);
252 if (tracing_group_name
== NULL
) {
257 tracing_group_name_override
= 1;
261 ret
= utils_show_help(8, "lttng-relayd", help_msg
);
263 ERR("Cannot show --help for `lttng-relayd`");
268 fprintf(stdout
, "%s\n", VERSION
);
271 if (lttng_is_setuid_setgid()) {
272 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
275 ret
= asprintf(&opt_output_path
, "%s", arg
);
278 PERROR("asprintf opt_output_path");
284 /* Verbose level can increase using multiple -v */
286 lttng_opt_verbose
= config_parse_value(arg
);
288 /* Only 3 level of verbosity (-vvv). */
289 if (lttng_opt_verbose
< 3) {
290 lttng_opt_verbose
+= 1;
295 /* Unknown option or other error.
296 * Error is printed by getopt, just return */
309 * config_entry_handler_cb used to handle options read from a config file.
310 * See config_entry_handler_cb comment in common/config/session-config.h for the
311 * return value conventions.
313 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
317 if (!entry
|| !entry
->name
|| !entry
->value
) {
322 /* Check if the option is to be ignored */
323 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
324 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
329 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
330 /* Ignore if entry name is not fully matched. */
331 if (strcmp(entry
->name
, long_options
[i
].name
)) {
336 * If the option takes no argument on the command line,
337 * we have to check if the value is "true". We support
338 * non-zero numeric values, true, on and yes.
340 if (!long_options
[i
].has_arg
) {
341 ret
= config_parse_value(entry
->value
);
344 WARN("Invalid configuration value \"%s\" for option %s",
345 entry
->value
, entry
->name
);
347 /* False, skip boolean config option. */
352 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
356 WARN("Unrecognized option \"%s\" in daemon configuration file.",
363 static int set_options(int argc
, char **argv
)
365 int c
, ret
= 0, option_index
= 0, retval
= 0;
366 int orig_optopt
= optopt
, orig_optind
= optind
;
367 char *default_address
, *optstring
;
368 const char *config_path
= NULL
;
370 optstring
= utils_generate_optstring(long_options
,
371 sizeof(long_options
) / sizeof(struct option
));
377 /* Check for the --config option */
379 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
380 &option_index
)) != -1) {
384 } else if (c
!= 'f') {
388 if (lttng_is_setuid_setgid()) {
389 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
392 config_path
= utils_expand_path(optarg
);
394 ERR("Failed to resolve path: %s", optarg
);
399 ret
= config_get_section_entries(config_path
, config_section_name
,
400 config_entry_handler
, NULL
);
403 ERR("Invalid configuration option at line %i", ret
);
409 /* Reset getopt's global state */
410 optopt
= orig_optopt
;
411 optind
= orig_optind
;
413 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
418 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
425 /* assign default values */
426 if (control_uri
== NULL
) {
427 ret
= asprintf(&default_address
,
428 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
429 DEFAULT_NETWORK_CONTROL_PORT
);
431 PERROR("asprintf default data address");
436 ret
= uri_parse(default_address
, &control_uri
);
437 free(default_address
);
439 ERR("Invalid control URI specified");
444 if (data_uri
== NULL
) {
445 ret
= asprintf(&default_address
,
446 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
447 DEFAULT_NETWORK_DATA_PORT
);
449 PERROR("asprintf default data address");
454 ret
= uri_parse(default_address
, &data_uri
);
455 free(default_address
);
457 ERR("Invalid data URI specified");
462 if (live_uri
== NULL
) {
463 ret
= asprintf(&default_address
,
464 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
465 DEFAULT_NETWORK_VIEWER_PORT
);
467 PERROR("asprintf default viewer control address");
472 ret
= uri_parse(default_address
, &live_uri
);
473 free(default_address
);
475 ERR("Invalid viewer control URI specified");
486 static void print_global_objects(void)
488 rcu_register_thread();
490 print_viewer_streams();
491 print_relay_streams();
494 rcu_unregister_thread();
500 static void relayd_cleanup(void)
502 print_global_objects();
506 if (viewer_streams_ht
)
507 lttng_ht_destroy(viewer_streams_ht
);
508 if (relay_streams_ht
)
509 lttng_ht_destroy(relay_streams_ht
);
511 lttng_ht_destroy(sessions_ht
);
513 /* free the dynamically allocated opt_output_path */
514 free(opt_output_path
);
516 /* Close thread quit pipes */
517 utils_close_pipe(thread_quit_pipe
);
519 uri_free(control_uri
);
521 /* Live URI is freed in the live thread. */
523 if (tracing_group_name_override
) {
524 free((void *) tracing_group_name
);
529 * Write to writable pipe used to notify a thread.
531 static int notify_thread_pipe(int wpipe
)
535 ret
= lttng_write(wpipe
, "!", 1);
537 PERROR("write poll pipe");
545 static int notify_health_quit_pipe(int *pipe
)
549 ret
= lttng_write(pipe
[1], "4", 1);
551 PERROR("write relay health quit");
560 * Stop all relayd and relayd-live threads.
562 int lttng_relay_stop_threads(void)
566 /* Stopping all threads */
567 DBG("Terminating all threads");
568 if (notify_thread_pipe(thread_quit_pipe
[1])) {
569 ERR("write error on thread quit pipe");
573 if (notify_health_quit_pipe(health_quit_pipe
)) {
574 ERR("write error on health quit pipe");
577 /* Dispatch thread */
578 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
579 futex_nto1_wake(&relay_conn_queue
.futex
);
581 if (relayd_live_stop()) {
582 ERR("Error stopping live threads");
589 * Signal handler for the daemon
591 * Simply stop all worker threads, leaving main() return gracefully after
592 * joining all threads and calling cleanup().
594 static void sighandler(int sig
)
598 DBG("SIGINT caught");
599 if (lttng_relay_stop_threads()) {
600 ERR("Error stopping threads");
604 DBG("SIGTERM caught");
605 if (lttng_relay_stop_threads()) {
606 ERR("Error stopping threads");
610 CMM_STORE_SHARED(recv_child_signal
, 1);
618 * Setup signal handler for :
619 * SIGINT, SIGTERM, SIGPIPE
621 static int set_signal_handler(void)
627 if ((ret
= sigemptyset(&sigset
)) < 0) {
628 PERROR("sigemptyset");
635 sa
.sa_handler
= sighandler
;
636 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
641 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
646 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
651 sa
.sa_handler
= SIG_IGN
;
652 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
657 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
662 void lttng_relay_notify_ready(void)
664 /* Notify the parent of the fork() process that we are ready. */
665 if (opt_daemon
|| opt_background
) {
666 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
667 kill(child_ppid
, SIGUSR1
);
673 * Init thread quit pipe.
675 * Return -1 on error or 0 if all pipes are created.
677 static int init_thread_quit_pipe(void)
681 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
687 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
689 static int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
693 if (events
== NULL
|| size
== 0) {
698 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
704 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
716 * Check if the thread quit pipe was triggered.
718 * Return 1 if it was triggered else 0;
720 static int check_thread_quit_pipe(int fd
, uint32_t events
)
722 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
730 * Create and init socket from uri.
732 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
)
735 struct lttcomm_sock
*sock
= NULL
;
737 sock
= lttcomm_alloc_sock_from_uri(uri
);
739 ERR("Allocating socket");
743 ret
= lttcomm_create_sock(sock
);
747 DBG("Listening on sock %d", sock
->fd
);
749 ret
= sock
->ops
->bind(sock
);
754 ret
= sock
->ops
->listen(sock
, -1);
764 lttcomm_destroy_sock(sock
);
770 * This thread manages the listening for new connections on the network
772 static void *relay_thread_listener(void *data
)
774 int i
, ret
, pollfd
, err
= -1;
775 uint32_t revents
, nb_fd
;
776 struct lttng_poll_event events
;
777 struct lttcomm_sock
*control_sock
, *data_sock
;
779 DBG("[thread] Relay listener started");
781 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
783 health_code_update();
785 control_sock
= relay_socket_create(control_uri
);
787 goto error_sock_control
;
790 data_sock
= relay_socket_create(data_uri
);
792 goto error_sock_relay
;
796 * Pass 3 as size here for the thread quit pipe, control and
799 ret
= create_thread_poll_set(&events
, 3);
801 goto error_create_poll
;
804 /* Add the control socket */
805 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
810 /* Add the data socket */
811 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
816 lttng_relay_notify_ready();
818 if (testpoint(relayd_thread_listener
)) {
819 goto error_testpoint
;
823 health_code_update();
825 DBG("Listener accepting connections");
829 ret
= lttng_poll_wait(&events
, -1);
833 * Restart interrupted system call.
835 if (errno
== EINTR
) {
843 DBG("Relay new connection received");
844 for (i
= 0; i
< nb_fd
; i
++) {
845 health_code_update();
847 /* Fetch once the poll data */
848 revents
= LTTNG_POLL_GETEV(&events
, i
);
849 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
853 * No activity for this FD (poll
859 /* Thread quit pipe has been closed. Killing thread. */
860 ret
= check_thread_quit_pipe(pollfd
, revents
);
866 if (revents
& LPOLLIN
) {
868 * A new connection is requested, therefore a
869 * sessiond/consumerd connection is allocated in
870 * this thread, enqueued to a global queue and
871 * dequeued (and freed) in the worker thread.
874 struct relay_connection
*new_conn
;
875 struct lttcomm_sock
*newsock
;
876 enum connection_type type
;
878 if (pollfd
== data_sock
->fd
) {
880 newsock
= data_sock
->ops
->accept(data_sock
);
881 DBG("Relay data connection accepted, socket %d",
884 assert(pollfd
== control_sock
->fd
);
885 type
= RELAY_CONTROL
;
886 newsock
= control_sock
->ops
->accept(control_sock
);
887 DBG("Relay control connection accepted, socket %d",
891 PERROR("accepting sock");
895 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
898 PERROR("setsockopt inet");
899 lttcomm_destroy_sock(newsock
);
902 new_conn
= connection_create(newsock
, type
);
904 lttcomm_destroy_sock(newsock
);
908 /* Enqueue request for the dispatcher thread. */
909 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
913 * Wake the dispatch queue futex.
914 * Implicit memory barrier with the
915 * exchange in cds_wfcq_enqueue.
917 futex_nto1_wake(&relay_conn_queue
.futex
);
918 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
919 ERR("socket poll error");
922 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
932 lttng_poll_clean(&events
);
934 if (data_sock
->fd
>= 0) {
935 ret
= data_sock
->ops
->close(data_sock
);
940 lttcomm_destroy_sock(data_sock
);
942 if (control_sock
->fd
>= 0) {
943 ret
= control_sock
->ops
->close(control_sock
);
948 lttcomm_destroy_sock(control_sock
);
952 ERR("Health error occurred in %s", __func__
);
954 health_unregister(health_relayd
);
955 DBG("Relay listener thread cleanup complete");
956 lttng_relay_stop_threads();
961 * This thread manages the dispatching of the requests to worker threads
963 static void *relay_thread_dispatcher(void *data
)
967 struct cds_wfcq_node
*node
;
968 struct relay_connection
*new_conn
= NULL
;
970 DBG("[thread] Relay dispatcher started");
972 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
974 if (testpoint(relayd_thread_dispatcher
)) {
975 goto error_testpoint
;
978 health_code_update();
981 health_code_update();
983 /* Atomically prepare the queue futex */
984 futex_nto1_prepare(&relay_conn_queue
.futex
);
986 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
991 health_code_update();
993 /* Dequeue commands */
994 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
995 &relay_conn_queue
.tail
);
997 DBG("Woken up but nothing in the relay command queue");
998 /* Continue thread execution */
1001 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1003 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1006 * Inform worker thread of the new request. This
1007 * call is blocking so we can be assured that
1008 * the data will be read at some point in time
1009 * or wait to the end of the world :)
1011 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1013 PERROR("write connection pipe");
1014 connection_put(new_conn
);
1017 } while (node
!= NULL
);
1019 /* Futex wait on queue. Blocking call on futex() */
1020 health_poll_entry();
1021 futex_nto1_wait(&relay_conn_queue
.futex
);
1025 /* Normal exit, no error */
1032 ERR("Health error occurred in %s", __func__
);
1034 health_unregister(health_relayd
);
1035 DBG("Dispatch thread dying");
1036 lttng_relay_stop_threads();
1041 * Set index data from the control port to a given index object.
1043 static int set_index_control_data(struct relay_index
*index
,
1044 struct lttcomm_relayd_index
*data
,
1045 struct relay_connection
*conn
)
1047 struct ctf_packet_index index_data
;
1050 * The index on disk is encoded in big endian, so we don't need
1051 * to convert the data received on the network. The data_offset
1052 * value is NEVER modified here and is updated by the data
1055 index_data
.packet_size
= data
->packet_size
;
1056 index_data
.content_size
= data
->content_size
;
1057 index_data
.timestamp_begin
= data
->timestamp_begin
;
1058 index_data
.timestamp_end
= data
->timestamp_end
;
1059 index_data
.events_discarded
= data
->events_discarded
;
1060 index_data
.stream_id
= data
->stream_id
;
1062 if (conn
->minor
>= 8) {
1063 index
->index_data
.stream_instance_id
= data
->stream_instance_id
;
1064 index
->index_data
.packet_seq_num
= data
->packet_seq_num
;
1067 return relay_index_set_data(index
, &index_data
);
1071 * Handle the RELAYD_CREATE_SESSION command.
1073 * On success, send back the session id or else return a negative value.
1075 static int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
1076 struct relay_connection
*conn
)
1078 int ret
= 0, send_ret
;
1079 struct relay_session
*session
;
1080 struct lttcomm_relayd_status_session reply
;
1081 char session_name
[LTTNG_NAME_MAX
];
1082 char hostname
[LTTNG_HOST_NAME_MAX
];
1083 uint32_t live_timer
= 0;
1084 bool snapshot
= false;
1086 memset(session_name
, 0, LTTNG_NAME_MAX
);
1087 memset(hostname
, 0, LTTNG_HOST_NAME_MAX
);
1089 memset(&reply
, 0, sizeof(reply
));
1091 switch (conn
->minor
) {
1096 case 4: /* LTTng sessiond 2.4 */
1098 ret
= cmd_create_session_2_4(conn
, session_name
,
1099 hostname
, &live_timer
, &snapshot
);
1105 fprintf(stderr
, "name: %s\n", session_name
);
1106 session
= session_create(session_name
, hostname
, live_timer
,
1107 snapshot
, conn
->major
, conn
->minor
);
1112 assert(!conn
->session
);
1113 conn
->session
= session
;
1114 DBG("Created session %" PRIu64
, session
->id
);
1116 reply
.session_id
= htobe64(session
->id
);
1120 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1122 reply
.ret_code
= htobe32(LTTNG_OK
);
1125 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1127 ERR("Relayd sending session id");
1135 * When we have received all the streams and the metadata for a channel,
1136 * we make them visible to the viewer threads.
1138 static void publish_connection_local_streams(struct relay_connection
*conn
)
1140 struct relay_stream
*stream
;
1141 struct relay_session
*session
= conn
->session
;
1144 * We publish all streams belonging to a session atomically wrt
1147 pthread_mutex_lock(&session
->lock
);
1149 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1151 stream_publish(stream
);
1156 * Inform the viewer that there are new streams in the session.
1158 if (session
->viewer_attached
) {
1159 uatomic_set(&session
->new_streams
, 1);
1161 pthread_mutex_unlock(&session
->lock
);
1165 * relay_add_stream: allocate a new stream for a session
1167 static int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1168 struct relay_connection
*conn
)
1172 struct relay_session
*session
= conn
->session
;
1173 struct relay_stream
*stream
= NULL
;
1174 struct lttcomm_relayd_status_stream reply
;
1175 struct ctf_trace
*trace
= NULL
;
1176 uint64_t stream_handle
= -1ULL;
1177 char *path_name
= NULL
, *channel_name
= NULL
;
1178 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1180 if (!session
|| conn
->version_check_done
== 0) {
1181 ERR("Trying to add a stream before version check");
1183 goto end_no_session
;
1186 switch (session
->minor
) {
1187 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1188 ret
= cmd_recv_stream_2_1(conn
, &path_name
,
1191 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1192 case 3: /* LTTng sessiond 2.3. Allocates path_name and channel_name. */
1193 case 4: /* LTTng sessiond 2.4. Allocates path_name and channel_name. */
1194 case 5: /* LTTng sessiond 2.5. Allocates path_name and channel_name. */
1195 case 6: /* LTTng sessiond 2.6. Allocates path_name and channel_name. */
1196 case 7: /* LTTng sessiond 2.7. Allocates path_name and channel_name. */
1197 case 8: /* LTTng sessiond 2.8. Allocates path_name and channel_name. */
1198 case 9: /* LTTng sessiond 2.9. Allocates path_name and channel_name. */
1199 case 10: /* LTTng sessiond 2.10. Allocates path_name and channel_name. */
1200 ret
= cmd_recv_stream_2_2(conn
, &path_name
,
1201 &channel_name
, &tracefile_size
, &tracefile_count
);
1203 case 11: /* LTTng sessiond 2.11. Allocates path_name and channel_name. */
1205 ret
= cmd_recv_stream_2_11(conn
, &path_name
,
1206 &channel_name
, &tracefile_size
, &tracefile_count
,
1214 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1218 /* This stream here has one reference on the trace. */
1220 pthread_mutex_lock(&last_relay_stream_id_lock
);
1221 stream_handle
= ++last_relay_stream_id
;
1222 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1224 /* We pass ownership of path_name and channel_name. */
1225 stream
= stream_create(trace
, stream_handle
, path_name
,
1226 channel_name
, tracefile_size
, tracefile_count
);
1228 channel_name
= NULL
;
1231 * Streams are the owners of their trace. Reference to trace is
1232 * kept within stream_create().
1234 ctf_trace_put(trace
);
1237 memset(&reply
, 0, sizeof(reply
));
1238 reply
.handle
= htobe64(stream_handle
);
1240 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1242 reply
.ret_code
= htobe32(LTTNG_OK
);
1245 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1246 sizeof(struct lttcomm_relayd_status_stream
), 0);
1248 ERR("Relay sending stream id");
1249 ret
= (int) send_ret
;
1259 * relay_close_stream: close a specific stream
1261 static int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1262 struct relay_connection
*conn
)
1265 struct relay_session
*session
= conn
->session
;
1266 struct lttcomm_relayd_close_stream stream_info
;
1267 struct lttcomm_relayd_generic_reply reply
;
1268 struct relay_stream
*stream
;
1270 DBG("Close stream received");
1272 if (!session
|| conn
->version_check_done
== 0) {
1273 ERR("Trying to close a stream before version check");
1275 goto end_no_session
;
1278 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
1279 sizeof(struct lttcomm_relayd_close_stream
), 0);
1280 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1282 /* Orderly shutdown. Not necessary to print an error. */
1283 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1285 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1288 goto end_no_session
;
1291 stream
= stream_get_by_id(be64toh(stream_info
.stream_id
));
1298 * Set last_net_seq_num before the close flag. Required by data
1301 pthread_mutex_lock(&stream
->lock
);
1302 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1303 pthread_mutex_unlock(&stream
->lock
);
1306 * This is one of the conditions which may trigger a stream close
1307 * with the others being:
1308 * 1) A close command is received for a stream
1309 * 2) The control connection owning the stream is closed
1310 * 3) We have received all of the stream's data _after_ a close
1313 try_stream_close(stream
);
1314 if (stream
->is_metadata
) {
1315 struct relay_viewer_stream
*vstream
;
1317 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1319 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1321 * Since all the metadata has been sent to the
1322 * viewer and that we have a request to close
1323 * its stream, we can safely teardown the
1324 * corresponding metadata viewer stream.
1326 viewer_stream_put(vstream
);
1328 /* Put local reference. */
1329 viewer_stream_put(vstream
);
1335 memset(&reply
, 0, sizeof(reply
));
1337 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1339 reply
.ret_code
= htobe32(LTTNG_OK
);
1341 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1342 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1344 ERR("Relay sending stream id");
1353 * relay_reset_metadata: reset a metadata stream
1356 int relay_reset_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1357 struct relay_connection
*conn
)
1360 struct relay_session
*session
= conn
->session
;
1361 struct lttcomm_relayd_reset_metadata stream_info
;
1362 struct lttcomm_relayd_generic_reply reply
;
1363 struct relay_stream
*stream
;
1365 DBG("Reset metadata received");
1367 if (!session
|| conn
->version_check_done
== 0) {
1368 ERR("Trying to reset a metadata stream before version check");
1370 goto end_no_session
;
1373 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
1374 sizeof(struct lttcomm_relayd_reset_metadata
), 0);
1375 if (ret
< sizeof(struct lttcomm_relayd_reset_metadata
)) {
1377 /* Orderly shutdown. Not necessary to print an error. */
1378 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1380 ERR("Relay didn't receive valid reset_metadata struct "
1384 goto end_no_session
;
1386 DBG("Update metadata to version %" PRIu64
, be64toh(stream_info
.version
));
1388 /* Unsupported for live sessions for now. */
1389 if (session
->live_timer
!= 0) {
1394 stream
= stream_get_by_id(be64toh(stream_info
.stream_id
));
1399 pthread_mutex_lock(&stream
->lock
);
1400 if (!stream
->is_metadata
) {
1405 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
1406 0, 0, -1, -1, stream
->stream_fd
->fd
, NULL
,
1407 &stream
->stream_fd
->fd
);
1409 ERR("Failed to rotate metadata file %s of channel %s",
1410 stream
->path_name
, stream
->channel_name
);
1415 pthread_mutex_unlock(&stream
->lock
);
1419 memset(&reply
, 0, sizeof(reply
));
1421 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1423 reply
.ret_code
= htobe32(LTTNG_OK
);
1425 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1426 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1428 ERR("Relay sending reset metadata reply");
1437 * relay_unknown_command: send -1 if received unknown command
1439 static void relay_unknown_command(struct relay_connection
*conn
)
1441 struct lttcomm_relayd_generic_reply reply
;
1444 memset(&reply
, 0, sizeof(reply
));
1445 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1446 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1447 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1449 ERR("Relay sending unknown command");
1454 * relay_start: send an acknowledgment to the client to tell if we are
1455 * ready to receive data. We are ready if a session is established.
1457 static int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1458 struct relay_connection
*conn
)
1460 int ret
= htobe32(LTTNG_OK
);
1461 struct lttcomm_relayd_generic_reply reply
;
1462 struct relay_session
*session
= conn
->session
;
1465 DBG("Trying to start the streaming without a session established");
1466 ret
= htobe32(LTTNG_ERR_UNK
);
1469 memset(&reply
, 0, sizeof(reply
));
1470 reply
.ret_code
= ret
;
1471 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1472 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1474 ERR("Relay sending start ack");
1481 * Append padding to the file pointed by the file descriptor fd.
1483 static int write_padding_to_file(int fd
, uint32_t size
)
1492 zeros
= zmalloc(size
);
1493 if (zeros
== NULL
) {
1494 PERROR("zmalloc zeros for padding");
1499 ret
= lttng_write(fd
, zeros
, size
);
1501 PERROR("write padding to file");
1511 int rotate_index_file(struct relay_stream
*stream
)
1514 uint32_t major
, minor
;
1516 /* Put ref on previous index_file. */
1517 if (stream
->index_file
) {
1518 lttng_index_file_put(stream
->index_file
);
1519 stream
->index_file
= NULL
;
1521 major
= stream
->trace
->session
->major
;
1522 minor
= stream
->trace
->session
->minor
;
1523 stream
->index_file
= lttng_index_file_create(stream
->path_name
,
1524 stream
->channel_name
,
1525 -1, -1, stream
->tracefile_size
,
1526 tracefile_array_get_file_index_head(stream
->tfa
),
1527 lttng_to_index_major(major
, minor
),
1528 lttng_to_index_minor(major
, minor
));
1529 if (!stream
->index_file
) {
1541 int do_rotate_stream(struct relay_stream
*stream
)
1545 /* Perform the stream rotation. */
1546 ret
= utils_rotate_stream_file(stream
->path_name
,
1547 stream
->channel_name
, stream
->tracefile_size
,
1548 stream
->tracefile_count
, -1,
1549 -1, stream
->stream_fd
->fd
,
1550 NULL
, &stream
->stream_fd
->fd
);
1552 ERR("Rotating stream output file");
1555 stream
->tracefile_size_current
= 0;
1557 /* Rotate also the index if the stream is not a metadata stream. */
1558 if (!stream
->is_metadata
) {
1559 ret
= rotate_index_file(stream
);
1561 ERR("Failed to rotate index file");
1566 stream
->rotate_at_seq_num
= -1ULL;
1572 * Check if a stream should perform a rotation (for session rotation).
1573 * Must be called with the stream lock held.
1575 * Return 0 on success, a negative value on error.
1578 int check_rotate_stream(struct relay_stream
*stream
)
1582 /* No rotation expected */
1583 if (stream
->rotate_at_seq_num
== -1ULL) {
1588 if (stream
->prev_seq
< stream
->rotate_at_seq_num
) {
1589 DBG("Stream %" PRIu64
" no yet ready for rotation",
1590 stream
->stream_handle
);
1591 fprintf(stderr
, "Stream %" PRIu64
" no yet ready for rotation\n",
1592 stream
->stream_handle
);
1595 } else if (stream
->prev_seq
> stream
->rotate_at_seq_num
) {
1596 DBG("Rotation after too much data has been written in tracefile "
1597 "for stream %" PRIu64
", need to truncate before "
1598 "rotating", stream
->stream_handle
);
1599 fprintf(stderr
, "Rotation after too much data has been written in tracefile "
1600 "for stream %" PRIu64
", need to truncate before "
1601 "rotating\n", stream
->stream_handle
);
1604 DBG("Stream %" PRIu64
" ready for rotation", stream
->stream_handle
);
1605 fprintf(stderr
, "Stream %" PRIu64
" ready for rotation\n", stream
->stream_handle
);
1608 ret
= do_rotate_stream(stream
);
1615 * relay_recv_metadata: receive the metadata for the session.
1617 static int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1618 struct relay_connection
*conn
)
1622 struct relay_session
*session
= conn
->session
;
1623 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1624 struct relay_stream
*metadata_stream
;
1625 uint64_t data_size
, payload_size
;
1628 ERR("Metadata sent before version check");
1633 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1634 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1635 ERR("Incorrect data size");
1639 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1641 if (data_buffer_size
< data_size
) {
1642 /* In case the realloc fails, we can free the memory */
1645 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1646 if (!tmp_data_ptr
) {
1647 ERR("Allocating data buffer");
1652 data_buffer
= tmp_data_ptr
;
1653 data_buffer_size
= data_size
;
1655 memset(data_buffer
, 0, data_size
);
1656 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1657 size_ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
1658 if (size_ret
< 0 || size_ret
!= data_size
) {
1659 if (size_ret
== 0) {
1660 /* Orderly shutdown. Not necessary to print an error. */
1661 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1663 ERR("Relay didn't receive the whole metadata");
1668 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1670 metadata_stream
= stream_get_by_id(be64toh(metadata_struct
->stream_id
));
1671 if (!metadata_stream
) {
1676 pthread_mutex_lock(&metadata_stream
->lock
);
1678 size_ret
= lttng_write(metadata_stream
->stream_fd
->fd
, metadata_struct
->payload
,
1680 if (size_ret
< payload_size
) {
1681 ERR("Relay error writing metadata on file");
1686 size_ret
= write_padding_to_file(metadata_stream
->stream_fd
->fd
,
1687 be32toh(metadata_struct
->padding_size
));
1692 metadata_stream
->metadata_received
+=
1693 payload_size
+ be32toh(metadata_struct
->padding_size
);
1694 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1695 metadata_stream
->metadata_received
);
1697 ret
= check_rotate_stream(metadata_stream
);
1699 ERR("Check rotate stream");
1704 pthread_mutex_unlock(&metadata_stream
->lock
);
1705 stream_put(metadata_stream
);
1711 * relay_send_version: send relayd version number
1713 static int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1714 struct relay_connection
*conn
)
1717 struct lttcomm_relayd_version reply
, msg
;
1719 conn
->version_check_done
= 1;
1721 /* Get version from the other side. */
1722 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1723 if (ret
< 0 || ret
!= sizeof(msg
)) {
1725 /* Orderly shutdown. Not necessary to print an error. */
1726 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1728 ERR("Relay failed to receive the version values.");
1734 memset(&reply
, 0, sizeof(reply
));
1735 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1736 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1738 /* Major versions must be the same */
1739 if (reply
.major
!= be32toh(msg
.major
)) {
1740 DBG("Incompatible major versions (%u vs %u), deleting session",
1741 reply
.major
, be32toh(msg
.major
));
1742 connection_put(conn
);
1747 conn
->major
= reply
.major
;
1748 /* We adapt to the lowest compatible version */
1749 if (reply
.minor
<= be32toh(msg
.minor
)) {
1750 conn
->minor
= reply
.minor
;
1752 conn
->minor
= be32toh(msg
.minor
);
1755 reply
.major
= htobe32(reply
.major
);
1756 reply
.minor
= htobe32(reply
.minor
);
1757 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1758 sizeof(struct lttcomm_relayd_version
), 0);
1760 ERR("Relay sending version");
1763 DBG("Version check done using protocol %u.%u", conn
->major
,
1771 * Check for data pending for a given stream id from the session daemon.
1773 static int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1774 struct relay_connection
*conn
)
1776 struct relay_session
*session
= conn
->session
;
1777 struct lttcomm_relayd_data_pending msg
;
1778 struct lttcomm_relayd_generic_reply reply
;
1779 struct relay_stream
*stream
;
1781 uint64_t last_net_seq_num
, stream_id
;
1783 DBG("Data pending command received");
1785 if (!session
|| conn
->version_check_done
== 0) {
1786 ERR("Trying to check for data before version check");
1788 goto end_no_session
;
1791 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1792 if (ret
< sizeof(msg
)) {
1794 /* Orderly shutdown. Not necessary to print an error. */
1795 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1797 ERR("Relay didn't receive valid data_pending struct size : %d",
1801 goto end_no_session
;
1804 stream_id
= be64toh(msg
.stream_id
);
1805 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1807 stream
= stream_get_by_id(stream_id
);
1808 if (stream
== NULL
) {
1813 pthread_mutex_lock(&stream
->lock
);
1815 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1816 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1819 /* Avoid wrapping issue */
1820 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1821 /* Data has in fact been written and is NOT pending */
1824 /* Data still being streamed thus pending */
1828 stream
->data_pending_check_done
= true;
1829 pthread_mutex_unlock(&stream
->lock
);
1834 memset(&reply
, 0, sizeof(reply
));
1835 reply
.ret_code
= htobe32(ret
);
1836 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1838 ERR("Relay data pending ret code failed");
1846 * Wait for the control socket to reach a quiescent state.
1848 * Note that for now, when receiving this command from the session
1849 * daemon, this means that every subsequent commands or data received on
1850 * the control socket has been handled. So, this is why we simply return
1853 static int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1854 struct relay_connection
*conn
)
1858 struct relay_stream
*stream
;
1859 struct lttcomm_relayd_quiescent_control msg
;
1860 struct lttcomm_relayd_generic_reply reply
;
1862 DBG("Checking quiescent state on control socket");
1864 if (!conn
->session
|| conn
->version_check_done
== 0) {
1865 ERR("Trying to check for data before version check");
1867 goto end_no_session
;
1870 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1871 if (ret
< sizeof(msg
)) {
1873 /* Orderly shutdown. Not necessary to print an error. */
1874 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1876 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1880 goto end_no_session
;
1883 stream_id
= be64toh(msg
.stream_id
);
1884 stream
= stream_get_by_id(stream_id
);
1888 pthread_mutex_lock(&stream
->lock
);
1889 stream
->data_pending_check_done
= true;
1890 pthread_mutex_unlock(&stream
->lock
);
1891 DBG("Relay quiescent control pending flag set to %" PRIu64
, stream_id
);
1894 memset(&reply
, 0, sizeof(reply
));
1895 reply
.ret_code
= htobe32(LTTNG_OK
);
1896 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1898 ERR("Relay data quiescent control ret code failed");
1906 * Initialize a data pending command. This means that a consumer is about
1907 * to ask for data pending for each stream it holds. Simply iterate over
1908 * all streams of a session and set the data_pending_check_done flag.
1910 * This command returns to the client a LTTNG_OK code.
1912 static int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1913 struct relay_connection
*conn
)
1916 struct lttng_ht_iter iter
;
1917 struct lttcomm_relayd_begin_data_pending msg
;
1918 struct lttcomm_relayd_generic_reply reply
;
1919 struct relay_stream
*stream
;
1920 uint64_t session_id
;
1925 DBG("Init streams for data pending");
1927 if (!conn
->session
|| conn
->version_check_done
== 0) {
1928 ERR("Trying to check for data before version check");
1930 goto end_no_session
;
1933 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1934 if (ret
< sizeof(msg
)) {
1936 /* Orderly shutdown. Not necessary to print an error. */
1937 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1939 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1943 goto end_no_session
;
1946 session_id
= be64toh(msg
.session_id
);
1949 * Iterate over all streams to set the begin data pending flag.
1950 * For now, the streams are indexed by stream handle so we have
1951 * to iterate over all streams to find the one associated with
1952 * the right session_id.
1955 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1957 if (!stream_get(stream
)) {
1960 if (stream
->trace
->session
->id
== session_id
) {
1961 pthread_mutex_lock(&stream
->lock
);
1962 stream
->data_pending_check_done
= false;
1963 pthread_mutex_unlock(&stream
->lock
);
1964 DBG("Set begin data pending flag to stream %" PRIu64
,
1965 stream
->stream_handle
);
1971 memset(&reply
, 0, sizeof(reply
));
1972 /* All good, send back reply. */
1973 reply
.ret_code
= htobe32(LTTNG_OK
);
1975 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1977 ERR("Relay begin data pending send reply failed");
1985 * End data pending command. This will check, for a given session id, if
1986 * each stream associated with it has its data_pending_check_done flag
1987 * set. If not, this means that the client lost track of the stream but
1988 * the data is still being streamed on our side. In this case, we inform
1989 * the client that data is in flight.
1991 * Return to the client if there is data in flight or not with a ret_code.
1993 static int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1994 struct relay_connection
*conn
)
1997 struct lttng_ht_iter iter
;
1998 struct lttcomm_relayd_end_data_pending msg
;
1999 struct lttcomm_relayd_generic_reply reply
;
2000 struct relay_stream
*stream
;
2001 uint64_t session_id
;
2002 uint32_t is_data_inflight
= 0;
2004 DBG("End data pending command");
2006 if (!conn
->session
|| conn
->version_check_done
== 0) {
2007 ERR("Trying to check for data before version check");
2009 goto end_no_session
;
2012 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
2013 if (ret
< sizeof(msg
)) {
2015 /* Orderly shutdown. Not necessary to print an error. */
2016 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2018 ERR("Relay didn't receive valid end data_pending struct size: %d",
2022 goto end_no_session
;
2025 session_id
= be64toh(msg
.session_id
);
2028 * Iterate over all streams to see if the begin data pending
2032 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2034 if (!stream_get(stream
)) {
2037 if (stream
->trace
->session
->id
!= session_id
) {
2041 pthread_mutex_lock(&stream
->lock
);
2042 if (!stream
->data_pending_check_done
) {
2043 if (!stream
->closed
|| !(((int64_t) (stream
->prev_seq
- stream
->last_net_seq_num
)) >= 0)) {
2044 is_data_inflight
= 1;
2045 DBG("Data is still in flight for stream %" PRIu64
,
2046 stream
->stream_handle
);
2047 pthread_mutex_unlock(&stream
->lock
);
2052 pthread_mutex_unlock(&stream
->lock
);
2057 memset(&reply
, 0, sizeof(reply
));
2058 /* All good, send back reply. */
2059 reply
.ret_code
= htobe32(is_data_inflight
);
2061 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2063 ERR("Relay end data pending send reply failed");
2071 * Receive an index for a specific stream.
2073 * Return 0 on success else a negative value.
2075 static int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
2076 struct relay_connection
*conn
)
2079 struct relay_session
*session
= conn
->session
;
2080 struct lttcomm_relayd_index index_info
;
2081 struct relay_index
*index
;
2082 struct lttcomm_relayd_generic_reply reply
;
2083 struct relay_stream
*stream
;
2084 uint64_t net_seq_num
;
2089 DBG("Relay receiving index");
2091 if (!session
|| conn
->version_check_done
== 0) {
2092 ERR("Trying to close a stream before version check");
2094 goto end_no_session
;
2097 msg_len
= lttcomm_relayd_index_len(
2098 lttng_to_index_major(conn
->major
, conn
->minor
),
2099 lttng_to_index_minor(conn
->major
, conn
->minor
));
2100 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &index_info
,
2102 if (ret
< msg_len
) {
2104 /* Orderly shutdown. Not necessary to print an error. */
2105 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2107 ERR("Relay didn't receive valid index struct size : %d", ret
);
2110 goto end_no_session
;
2113 net_seq_num
= be64toh(index_info
.net_seq_num
);
2115 stream
= stream_get_by_id(be64toh(index_info
.relay_stream_id
));
2117 ERR("stream_get_by_id not found");
2121 pthread_mutex_lock(&stream
->lock
);
2123 /* Live beacon handling */
2124 if (index_info
.packet_size
== 0) {
2125 DBG("Received live beacon for stream %" PRIu64
,
2126 stream
->stream_handle
);
2129 * Only flag a stream inactive when it has already
2130 * received data and no indexes are in flight.
2132 if (stream
->index_received_seqcount
> 0
2133 && stream
->indexes_in_flight
== 0) {
2134 stream
->beacon_ts_end
=
2135 be64toh(index_info
.timestamp_end
);
2138 goto end_stream_put
;
2140 stream
->beacon_ts_end
= -1ULL;
2143 if (stream
->ctf_stream_id
== -1ULL) {
2144 stream
->ctf_stream_id
= be64toh(index_info
.stream_id
);
2146 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
2149 ERR("relay_index_get_by_id_or_create index NULL");
2150 goto end_stream_put
;
2152 if (set_index_control_data(index
, &index_info
, conn
)) {
2153 ERR("set_index_control_data error");
2154 relay_index_put(index
);
2156 goto end_stream_put
;
2158 ret
= relay_index_try_flush(index
);
2160 tracefile_array_commit_seq(stream
->tfa
);
2161 stream
->index_received_seqcount
++;
2162 } else if (ret
> 0) {
2166 ERR("relay_index_try_flush error %d", ret
);
2167 relay_index_put(index
);
2172 pthread_mutex_unlock(&stream
->lock
);
2177 memset(&reply
, 0, sizeof(reply
));
2179 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2181 reply
.ret_code
= htobe32(LTTNG_OK
);
2183 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2185 ERR("Relay sending close index id reply");
2194 * Receive the streams_sent message.
2196 * Return 0 on success else a negative value.
2198 static int relay_streams_sent(struct lttcomm_relayd_hdr
*recv_hdr
,
2199 struct relay_connection
*conn
)
2202 struct lttcomm_relayd_generic_reply reply
;
2206 DBG("Relay receiving streams_sent");
2208 if (!conn
->session
|| conn
->version_check_done
== 0) {
2209 ERR("Trying to close a stream before version check");
2211 goto end_no_session
;
2215 * Publish every pending stream in the connection recv list which are
2216 * now ready to be used by the viewer.
2218 publish_connection_local_streams(conn
);
2220 memset(&reply
, 0, sizeof(reply
));
2221 reply
.ret_code
= htobe32(LTTNG_OK
);
2222 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2224 ERR("Relay sending sent_stream reply");
2236 * relay_rotate_stream: rotate a stream to a new tracefile for the session
2237 * rotation feature (not the tracefile rotation feature).
2239 static int relay_rotate_session_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
2240 struct relay_connection
*conn
)
2243 struct relay_session
*session
= conn
->session
;
2244 struct lttcomm_relayd_rotate_stream stream_info
;
2245 struct lttcomm_relayd_generic_reply reply
;
2246 struct relay_stream
*stream
;
2249 DBG("Rotate stream received");
2251 if (!session
|| conn
->version_check_done
== 0) {
2252 ERR("Trying to rotate a stream before version check");
2254 goto end_no_session
;
2257 if (session
->minor
< 11) {
2258 ERR("Unsupported feature before 2.11");
2260 goto end_no_session
;
2263 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
2264 sizeof(stream_info
), 0);
2265 if (ret
< sizeof(stream_info
)) {
2267 /* Orderly shutdown. Not necessary to print an error. */
2268 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2270 ERR("Relay didn't receive valid rotate_stream struct size : %d", ret
);
2273 goto end_no_session
;
2276 stream
= stream_get_by_id(be64toh(stream_info
.stream_id
));
2282 len
= lttng_strnlen(stream_info
.new_pathname
,
2283 sizeof(stream_info
.new_pathname
));
2284 /* Ensure that NULL-terminated and fits in local filename length. */
2285 if (len
== sizeof(stream_info
.new_pathname
) || len
>= LTTNG_NAME_MAX
) {
2286 ret
= -ENAMETOOLONG
;
2287 ERR("Path name too long");
2291 pthread_mutex_lock(&stream
->lock
);
2293 /* Update the trace path (just the folder, the stream name does not change). */
2294 free(stream
->path_name
);
2295 stream
->path_name
= create_output_path(stream_info
.new_pathname
);
2296 if (!stream
->path_name
) {
2297 ERR("Failed to create a new output path");
2298 goto end_stream_unlock
;
2300 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
,
2303 ERR("relay creating output directory");
2306 stream
->chunk_id
= be64toh(stream_info
.new_chunk_id
);
2308 if (stream
->is_metadata
) {
2310 * The metadata stream is sent only over the control connection
2311 * so we know we have all the data to perform the stream
2314 ret
= do_rotate_stream(stream
);
2316 stream
->rotate_at_seq_num
= be64toh(stream_info
.rotate_at_seq_num
);
2317 ret
= check_rotate_stream(stream
);
2320 ERR("Check rotate stream");
2321 goto end_stream_unlock
;
2325 pthread_mutex_unlock(&stream
->lock
);
2328 memset(&reply
, 0, sizeof(reply
));
2330 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2332 reply
.ret_code
= htobe32(LTTNG_OK
);
2334 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2335 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2337 ERR("Relay sending stream id");
2346 * relay_rotate_rename: rename the trace folder after the rotation is
2347 * complete. We are not closing any fd here, just moving the folder, so it
2348 * works even if data is still in flight.
2350 static int relay_rotate_rename(struct lttcomm_relayd_hdr
*recv_hdr
,
2351 struct relay_connection
*conn
)
2354 struct relay_session
*session
= conn
->session
;
2355 struct lttcomm_relayd_rotate_rename stream_info
;
2356 struct lttcomm_relayd_generic_reply reply
;
2358 char *old
= NULL
, *new = NULL
;
2360 DBG("Rotate rename received");
2362 if (!session
|| conn
->version_check_done
== 0) {
2363 ERR("Trying to rename before version check");
2365 goto end_no_session
;
2368 if (session
->minor
< 11) {
2369 ERR("Unsupported feature before 2.11");
2371 goto end_no_session
;
2374 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
2375 sizeof(stream_info
), 0);
2376 if (ret
< sizeof(stream_info
)) {
2378 /* Orderly shutdown. Not necessary to print an error. */
2379 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2381 ERR("Relay didn't receive valid rotate_rename struct size : %d", ret
);
2384 goto end_no_session
;
2387 len
= lttng_strnlen(stream_info
.current_path
,
2388 sizeof(stream_info
.current_path
));
2389 /* Ensure that NULL-terminated and fits in local filename length. */
2390 if (len
== sizeof(stream_info
.current_path
) || len
>= LTTNG_NAME_MAX
) {
2391 ret
= -ENAMETOOLONG
;
2392 ERR("Path name too long");
2396 len
= lttng_strnlen(stream_info
.new_path
,
2397 sizeof(stream_info
.new_path
));
2398 /* Ensure that NULL-terminated and fits in local filename length. */
2399 if (len
== sizeof(stream_info
.new_path
) || len
>= LTTNG_NAME_MAX
) {
2400 ret
= -ENAMETOOLONG
;
2401 ERR("Path name too long");
2405 fprintf(stderr
, "Renaming %s to %s/\n", stream_info
.current_path
,
2406 stream_info
.new_path
);
2408 old
= create_output_path(stream_info
.current_path
);
2410 ERR("Failed to create current output path");
2415 new = create_output_path(stream_info
.new_path
);
2417 ERR("Failed to create new output path");
2422 ret
= utils_mkdir_recursive(new, S_IRWXU
| S_IRWXG
,
2425 ERR("relay creating output directory");
2429 ret
= rename(old
, new);
2430 if (ret
< 0 && errno
!= ENOENT
) {
2431 PERROR("Rename completed rotation chunk");
2437 memset(&reply
, 0, sizeof(reply
));
2439 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2441 reply
.ret_code
= htobe32(LTTNG_OK
);
2443 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2444 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2446 ERR("Relay sending stream id");
2457 int relay_rotate_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
2458 struct relay_connection
*conn
)
2460 struct relay_session
*session
= conn
->session
;
2461 struct lttcomm_relayd_rotate_pending msg
;
2462 struct lttcomm_relayd_generic_reply reply
;
2463 struct lttng_ht_iter iter
;
2464 struct relay_stream
*stream
;
2467 uint32_t rotate_pending
;
2469 DBG("Rotate pending command received");
2470 fprintf(stderr
, "Rotate pending command received\n");
2472 if (!session
|| conn
->version_check_done
== 0) {
2473 ERR("Trying to check for data before version check");
2475 goto end_no_session
;
2478 if (session
->minor
< 11) {
2479 ERR("Unsupported feature before 2.11");
2481 goto end_no_session
;
2484 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
2485 if (ret
< sizeof(msg
)) {
2487 /* Orderly shutdown. Not necessary to print an error. */
2488 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2490 ERR("Relay didn't receive valid rotate_pending struct size : %d",
2494 goto end_no_session
;
2497 chunk_id
= be64toh(msg
.chunk_id
);
2502 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2504 if (!stream_get(stream
)) {
2507 if (stream
->trace
->session
!= session
) {
2511 pthread_mutex_lock(&stream
->lock
);
2512 if (stream
->rotate_at_seq_num
!= -1ULL) {
2514 DBG("Stream %" PRIu64
" is still rotating",
2515 stream
->stream_handle
);
2516 } else if (stream
->chunk_id
< chunk_id
) {
2518 DBG("Stream %" PRIu64
" did not exist on the consumer "
2519 "when the last rotation started, but is"
2520 "still waiting for data before getting"
2522 stream
->stream_handle
);
2524 pthread_mutex_unlock(&stream
->lock
);
2526 if (rotate_pending
) {
2533 fprintf(stderr
, "Rotate pending done: %d\n", rotate_pending
);
2534 memset(&reply
, 0, sizeof(reply
));
2535 reply
.ret_code
= htobe32(rotate_pending
);
2536 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2538 ERR("Relay rotate pending ret code failed");
2546 * Process the commands received on the control socket
2548 static int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
2549 struct relay_connection
*conn
)
2553 switch (be32toh(recv_hdr
->cmd
)) {
2554 case RELAYD_CREATE_SESSION
:
2555 ret
= relay_create_session(recv_hdr
, conn
);
2557 case RELAYD_ADD_STREAM
:
2558 ret
= relay_add_stream(recv_hdr
, conn
);
2560 case RELAYD_START_DATA
:
2561 ret
= relay_start(recv_hdr
, conn
);
2563 case RELAYD_SEND_METADATA
:
2564 ret
= relay_recv_metadata(recv_hdr
, conn
);
2566 case RELAYD_VERSION
:
2567 ret
= relay_send_version(recv_hdr
, conn
);
2569 case RELAYD_CLOSE_STREAM
:
2570 ret
= relay_close_stream(recv_hdr
, conn
);
2572 case RELAYD_DATA_PENDING
:
2573 ret
= relay_data_pending(recv_hdr
, conn
);
2575 case RELAYD_QUIESCENT_CONTROL
:
2576 ret
= relay_quiescent_control(recv_hdr
, conn
);
2578 case RELAYD_BEGIN_DATA_PENDING
:
2579 ret
= relay_begin_data_pending(recv_hdr
, conn
);
2581 case RELAYD_END_DATA_PENDING
:
2582 ret
= relay_end_data_pending(recv_hdr
, conn
);
2584 case RELAYD_SEND_INDEX
:
2585 ret
= relay_recv_index(recv_hdr
, conn
);
2587 case RELAYD_STREAMS_SENT
:
2588 ret
= relay_streams_sent(recv_hdr
, conn
);
2590 case RELAYD_RESET_METADATA
:
2591 ret
= relay_reset_metadata(recv_hdr
, conn
);
2593 case RELAYD_ROTATE_STREAM
:
2594 ret
= relay_rotate_session_stream(recv_hdr
, conn
);
2596 case RELAYD_ROTATE_RENAME
:
2597 ret
= relay_rotate_rename(recv_hdr
, conn
);
2599 case RELAYD_ROTATE_PENDING
:
2600 ret
= relay_rotate_pending(recv_hdr
, conn
);
2602 case RELAYD_UPDATE_SYNC_INFO
:
2604 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
2605 relay_unknown_command(conn
);
2615 * Handle index for a data stream.
2617 * Called with the stream lock held.
2619 * Return 0 on success else a negative value.
2621 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2625 uint64_t data_offset
;
2626 struct relay_index
*index
;
2628 /* Get data offset because we are about to update the index. */
2629 data_offset
= htobe64(stream
->tracefile_size_current
);
2631 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
2632 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
2635 * Lookup for an existing index for that stream id/sequence
2636 * number. If it exists, the control thread has already received the
2637 * data for it, thus we need to write it to disk.
2639 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
2645 if (rotate_index
|| !stream
->index_file
) {
2646 ret
= rotate_index_file(stream
);
2648 ERR("Failed to rotate index");
2649 /* Put self-ref for this index due to error. */
2650 relay_index_put(index
);
2656 if (relay_index_set_file(index
, stream
->index_file
, data_offset
)) {
2658 /* Put self-ref for this index due to error. */
2659 relay_index_put(index
);
2664 ret
= relay_index_try_flush(index
);
2666 tracefile_array_commit_seq(stream
->tfa
);
2667 stream
->index_received_seqcount
++;
2668 } else if (ret
> 0) {
2672 /* Put self-ref for this index due to error. */
2673 relay_index_put(index
);
2682 * relay_process_data: Process the data received on the data socket
2684 static int relay_process_data(struct relay_connection
*conn
)
2686 int ret
= 0, rotate_index
= 0;
2688 struct relay_stream
*stream
;
2689 struct lttcomm_relayd_data_hdr data_hdr
;
2691 uint64_t net_seq_num
;
2693 struct relay_session
*session
;
2694 bool new_stream
= false, close_requested
= false;
2695 size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
2696 size_t recv_off
= 0;
2697 char data_buffer
[chunk_size
];
2699 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &data_hdr
,
2700 sizeof(struct lttcomm_relayd_data_hdr
), 0);
2703 /* Orderly shutdown. Not necessary to print an error. */
2704 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2706 ERR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2712 stream_id
= be64toh(data_hdr
.stream_id
);
2713 stream
= stream_get_by_id(stream_id
);
2715 ERR("relay_process_data: Cannot find stream %" PRIu64
, stream_id
);
2719 session
= stream
->trace
->session
;
2720 data_size
= be32toh(data_hdr
.data_size
);
2722 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2724 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2725 data_size
, stream_id
, net_seq_num
);
2727 pthread_mutex_lock(&stream
->lock
);
2729 /* Check if a rotation is needed. */
2730 if (stream
->tracefile_size
> 0 &&
2731 (stream
->tracefile_size_current
+ data_size
) >
2732 stream
->tracefile_size
) {
2733 uint64_t old_id
, new_id
;
2735 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
2736 tracefile_array_file_rotate(stream
->tfa
);
2738 /* new_id is updated by utils_rotate_stream_file. */
2741 ret
= utils_rotate_stream_file(stream
->path_name
,
2742 stream
->channel_name
, stream
->tracefile_size
,
2743 stream
->tracefile_count
, -1,
2744 -1, stream
->stream_fd
->fd
,
2745 &new_id
, &stream
->stream_fd
->fd
);
2747 ERR("Rotating stream output file");
2748 goto end_stream_unlock
;
2751 * Reset current size because we just performed a stream
2754 stream
->tracefile_size_current
= 0;
2759 * Index are handled in protocol version 2.4 and above. Also,
2760 * snapshot and index are NOT supported.
2762 if (session
->minor
>= 4 && !session
->snapshot
) {
2763 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2765 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2766 stream
->stream_handle
, net_seq_num
, ret
);
2767 goto end_stream_unlock
;
2771 for (recv_off
= 0; recv_off
< data_size
; recv_off
+= chunk_size
) {
2772 size_t recv_size
= min(data_size
- recv_off
, chunk_size
);
2774 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, recv_size
, 0);
2777 /* Orderly shutdown. Not necessary to print an error. */
2778 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2780 ERR("Socket %d error %d", conn
->sock
->fd
, ret
);
2783 goto end_stream_unlock
;
2786 /* Write data to stream output fd. */
2787 size_ret
= lttng_write(stream
->stream_fd
->fd
, data_buffer
,
2789 if (size_ret
< recv_size
) {
2790 ERR("Relay error writing data to file");
2792 goto end_stream_unlock
;
2795 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
2796 size_ret
, stream
->stream_handle
);
2799 ret
= write_padding_to_file(stream
->stream_fd
->fd
,
2800 be32toh(data_hdr
.padding_size
));
2802 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2803 stream
->stream_handle
, net_seq_num
, ret
);
2804 goto end_stream_unlock
;
2806 stream
->tracefile_size_current
+=
2807 data_size
+ be32toh(data_hdr
.padding_size
);
2808 if (stream
->prev_seq
== -1ULL) {
2812 stream
->prev_seq
= net_seq_num
;
2814 ret
= check_rotate_stream(stream
);
2816 ERR("Check rotate stream");
2817 goto end_stream_unlock
;
2821 close_requested
= stream
->close_requested
;
2822 pthread_mutex_unlock(&stream
->lock
);
2823 if (close_requested
) {
2824 try_stream_close(stream
);
2828 pthread_mutex_lock(&session
->lock
);
2829 uatomic_set(&session
->new_streams
, 1);
2830 pthread_mutex_unlock(&session
->lock
);
2837 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2841 (void) lttng_poll_del(events
, pollfd
);
2843 ret
= close(pollfd
);
2845 ERR("Closing pollfd %d", pollfd
);
2849 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
2850 int pollfd
, struct relay_connection
*conn
)
2852 const char *type_str
;
2854 switch (conn
->type
) {
2859 type_str
= "Control";
2861 case RELAY_VIEWER_COMMAND
:
2862 type_str
= "Viewer Command";
2864 case RELAY_VIEWER_NOTIFICATION
:
2865 type_str
= "Viewer Notification";
2868 type_str
= "Unknown";
2870 cleanup_connection_pollfd(events
, pollfd
);
2871 connection_put(conn
);
2872 DBG("%s connection closed with %d", type_str
, pollfd
);
2876 * This thread does the actual work
2878 static void *relay_thread_worker(void *data
)
2880 int ret
, err
= -1, last_seen_data_fd
= -1;
2882 struct lttng_poll_event events
;
2883 struct lttng_ht
*relay_connections_ht
;
2884 struct lttng_ht_iter iter
;
2885 struct lttcomm_relayd_hdr recv_hdr
;
2886 struct relay_connection
*destroy_conn
= NULL
;
2888 DBG("[thread] Relay worker started");
2890 rcu_register_thread();
2892 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2894 if (testpoint(relayd_thread_worker
)) {
2895 goto error_testpoint
;
2898 health_code_update();
2900 /* table of connections indexed on socket */
2901 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2902 if (!relay_connections_ht
) {
2903 goto relay_connections_ht_error
;
2906 ret
= create_thread_poll_set(&events
, 2);
2908 goto error_poll_create
;
2911 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2918 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2920 health_code_update();
2922 /* Infinite blocking call, waiting for transmission */
2923 DBG3("Relayd worker thread polling...");
2924 health_poll_entry();
2925 ret
= lttng_poll_wait(&events
, -1);
2929 * Restart interrupted system call.
2931 if (errno
== EINTR
) {
2940 * Process control. The control connection is
2941 * prioritized so we don't starve it with high
2942 * throughput tracing data on the data connection.
2944 for (i
= 0; i
< nb_fd
; i
++) {
2945 /* Fetch once the poll data */
2946 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2947 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2949 health_code_update();
2953 * No activity for this FD (poll
2959 /* Thread quit pipe has been closed. Killing thread. */
2960 ret
= check_thread_quit_pipe(pollfd
, revents
);
2966 /* Inspect the relay conn pipe for new connection */
2967 if (pollfd
== relay_conn_pipe
[0]) {
2968 if (revents
& LPOLLIN
) {
2969 struct relay_connection
*conn
;
2971 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
2975 lttng_poll_add(&events
, conn
->sock
->fd
,
2976 LPOLLIN
| LPOLLRDHUP
);
2977 connection_ht_add(relay_connections_ht
, conn
);
2978 DBG("Connection socket %d added", conn
->sock
->fd
);
2979 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2980 ERR("Relay connection pipe error");
2983 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2987 struct relay_connection
*ctrl_conn
;
2989 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
2990 /* If not found, there is a synchronization issue. */
2993 if (ctrl_conn
->type
== RELAY_DATA
) {
2994 if (revents
& LPOLLIN
) {
2996 * Flag the last seen data fd not deleted. It will be
2997 * used as the last seen fd if any fd gets deleted in
3000 last_notdel_data_fd
= pollfd
;
3002 goto put_ctrl_connection
;
3004 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3006 if (revents
& LPOLLIN
) {
3007 ret
= ctrl_conn
->sock
->ops
->recvmsg(ctrl_conn
->sock
,
3008 &recv_hdr
, sizeof(recv_hdr
), 0);
3010 /* Connection closed */
3011 relay_thread_close_connection(&events
, pollfd
,
3014 ret
= relay_process_control(&recv_hdr
, ctrl_conn
);
3016 /* Clear the session on error. */
3017 relay_thread_close_connection(&events
,
3022 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3023 relay_thread_close_connection(&events
,
3025 if (last_seen_data_fd
== pollfd
) {
3026 last_seen_data_fd
= last_notdel_data_fd
;
3029 ERR("Unexpected poll events %u for control sock %d",
3031 connection_put(ctrl_conn
);
3034 put_ctrl_connection
:
3035 connection_put(ctrl_conn
);
3040 * The last loop handled a control request, go back to poll to make
3041 * sure we prioritise the control socket.
3047 if (last_seen_data_fd
>= 0) {
3048 for (i
= 0; i
< nb_fd
; i
++) {
3049 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3051 health_code_update();
3053 if (last_seen_data_fd
== pollfd
) {
3060 /* Process data connection. */
3061 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3062 /* Fetch the poll data. */
3063 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3064 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3065 struct relay_connection
*data_conn
;
3067 health_code_update();
3070 /* No activity for this FD (poll implementation). */
3074 /* Skip the command pipe. It's handled in the first loop. */
3075 if (pollfd
== relay_conn_pipe
[0]) {
3079 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3081 /* Skip it. Might be removed before. */
3084 if (data_conn
->type
== RELAY_CONTROL
) {
3085 goto put_data_connection
;
3087 assert(data_conn
->type
== RELAY_DATA
);
3089 if (revents
& LPOLLIN
) {
3090 ret
= relay_process_data(data_conn
);
3091 /* Connection closed */
3093 relay_thread_close_connection(&events
, pollfd
,
3096 * Every goto restart call sets the last seen fd where
3097 * here we don't really care since we gracefully
3098 * continue the loop after the connection is deleted.
3101 /* Keep last seen port. */
3102 last_seen_data_fd
= pollfd
;
3103 connection_put(data_conn
);
3106 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3107 relay_thread_close_connection(&events
, pollfd
,
3110 ERR("Unknown poll events %u for data sock %d",
3113 put_data_connection
:
3114 connection_put(data_conn
);
3116 last_seen_data_fd
= -1;
3119 /* Normal exit, no error */
3124 /* Cleanup reamaining connection object. */
3126 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3129 health_code_update();
3131 if (session_abort(destroy_conn
->session
)) {
3136 * No need to grab another ref, because we own
3139 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
3144 lttng_poll_clean(&events
);
3146 lttng_ht_destroy(relay_connections_ht
);
3147 relay_connections_ht_error
:
3148 /* Close relay conn pipes */
3149 utils_close_pipe(relay_conn_pipe
);
3151 DBG("Thread exited with error");
3153 DBG("Worker thread cleanup complete");
3157 ERR("Health error occurred in %s", __func__
);
3159 health_unregister(health_relayd
);
3160 rcu_unregister_thread();
3161 lttng_relay_stop_threads();
3166 * Create the relay command pipe to wake thread_manage_apps.
3167 * Closed in cleanup().
3169 static int create_relay_conn_pipe(void)
3173 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
3181 int main(int argc
, char **argv
)
3183 int ret
= 0, retval
= 0;
3186 /* Parse arguments */
3188 if (set_options(argc
, argv
)) {
3193 if (set_signal_handler()) {
3198 /* Try to create directory if -o, --output is specified. */
3199 if (opt_output_path
) {
3200 if (*opt_output_path
!= '/') {
3201 ERR("Please specify an absolute path for -o, --output PATH");
3206 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
3209 ERR("Unable to create %s", opt_output_path
);
3216 if (opt_daemon
|| opt_background
) {
3219 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
3227 * We are in the child. Make sure all other file
3228 * descriptors are closed, in case we are called with
3229 * more opened file descriptors than the standard ones.
3231 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3236 /* Initialize thread health monitoring */
3237 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
3238 if (!health_relayd
) {
3239 PERROR("health_app_create error");
3241 goto exit_health_app_create
;
3244 /* Create thread quit pipe */
3245 if (init_thread_quit_pipe()) {
3247 goto exit_init_data
;
3250 /* Setup the thread apps communication pipe. */
3251 if (create_relay_conn_pipe()) {
3253 goto exit_init_data
;
3256 /* Init relay command queue. */
3257 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
3259 /* Initialize communication library */
3261 lttcomm_inet_init();
3263 /* tables of sessions indexed by session ID */
3264 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3267 goto exit_init_data
;
3270 /* tables of streams indexed by stream ID */
3271 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3272 if (!relay_streams_ht
) {
3274 goto exit_init_data
;
3277 /* tables of streams indexed by stream ID */
3278 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3279 if (!viewer_streams_ht
) {
3281 goto exit_init_data
;
3284 ret
= utils_create_pipe(health_quit_pipe
);
3287 goto exit_health_quit_pipe
;
3290 /* Create thread to manage the client socket */
3291 ret
= pthread_create(&health_thread
, default_pthread_attr(),
3292 thread_manage_health
, (void *) NULL
);
3295 PERROR("pthread_create health");
3297 goto exit_health_thread
;
3300 /* Setup the dispatcher thread */
3301 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
3302 relay_thread_dispatcher
, (void *) NULL
);
3305 PERROR("pthread_create dispatcher");
3307 goto exit_dispatcher_thread
;
3310 /* Setup the worker thread */
3311 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
3312 relay_thread_worker
, NULL
);
3315 PERROR("pthread_create worker");
3317 goto exit_worker_thread
;
3320 /* Setup the listener thread */
3321 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
3322 relay_thread_listener
, (void *) NULL
);
3325 PERROR("pthread_create listener");
3327 goto exit_listener_thread
;
3330 ret
= relayd_live_create(live_uri
);
3332 ERR("Starting live viewer threads");
3338 * This is where we start awaiting program completion (e.g. through
3339 * signal that asks threads to teardown).
3342 ret
= relayd_live_join();
3348 ret
= pthread_join(listener_thread
, &status
);
3351 PERROR("pthread_join listener_thread");
3355 exit_listener_thread
:
3356 ret
= pthread_join(worker_thread
, &status
);
3359 PERROR("pthread_join worker_thread");
3364 ret
= pthread_join(dispatcher_thread
, &status
);
3367 PERROR("pthread_join dispatcher_thread");
3370 exit_dispatcher_thread
:
3372 ret
= pthread_join(health_thread
, &status
);
3375 PERROR("pthread_join health_thread");
3380 utils_close_pipe(health_quit_pipe
);
3381 exit_health_quit_pipe
:
3384 health_app_destroy(health_relayd
);
3385 exit_health_app_create
:
3388 * Wait for all pending call_rcu work to complete before tearing
3389 * down data structures. call_rcu worker may be trying to
3390 * perform lookups in those structures.
3395 /* Ensure all prior call_rcu are done. */