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 session
= session_create(session_name
, hostname
, live_timer
,
1106 snapshot
, conn
->major
, conn
->minor
);
1111 assert(!conn
->session
);
1112 conn
->session
= session
;
1113 DBG("Created session %" PRIu64
, session
->id
);
1115 reply
.session_id
= htobe64(session
->id
);
1119 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1121 reply
.ret_code
= htobe32(LTTNG_OK
);
1124 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1126 ERR("Relayd sending session id");
1134 * When we have received all the streams and the metadata for a channel,
1135 * we make them visible to the viewer threads.
1137 static void publish_connection_local_streams(struct relay_connection
*conn
)
1139 struct relay_stream
*stream
;
1140 struct relay_session
*session
= conn
->session
;
1143 * We publish all streams belonging to a session atomically wrt
1146 pthread_mutex_lock(&session
->lock
);
1148 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1150 stream_publish(stream
);
1155 * Inform the viewer that there are new streams in the session.
1157 if (session
->viewer_attached
) {
1158 uatomic_set(&session
->new_streams
, 1);
1160 pthread_mutex_unlock(&session
->lock
);
1164 * relay_add_stream: allocate a new stream for a session
1166 static int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1167 struct relay_connection
*conn
)
1171 struct relay_session
*session
= conn
->session
;
1172 struct relay_stream
*stream
= NULL
;
1173 struct lttcomm_relayd_status_stream reply
;
1174 struct ctf_trace
*trace
= NULL
;
1175 uint64_t stream_handle
= -1ULL;
1176 char *path_name
= NULL
, *channel_name
= NULL
;
1177 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1179 if (!session
|| conn
->version_check_done
== 0) {
1180 ERR("Trying to add a stream before version check");
1182 goto end_no_session
;
1185 switch (session
->minor
) {
1186 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1187 ret
= cmd_recv_stream_2_1(conn
, &path_name
,
1190 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1191 case 3: /* LTTng sessiond 2.3. Allocates path_name and channel_name. */
1192 case 4: /* LTTng sessiond 2.4. Allocates path_name and channel_name. */
1193 case 5: /* LTTng sessiond 2.5. Allocates path_name and channel_name. */
1194 case 6: /* LTTng sessiond 2.6. Allocates path_name and channel_name. */
1195 case 7: /* LTTng sessiond 2.7. Allocates path_name and channel_name. */
1196 case 8: /* LTTng sessiond 2.8. Allocates path_name and channel_name. */
1197 case 9: /* LTTng sessiond 2.9. Allocates path_name and channel_name. */
1198 case 10: /* LTTng sessiond 2.10. Allocates path_name and channel_name. */
1199 ret
= cmd_recv_stream_2_2(conn
, &path_name
,
1200 &channel_name
, &tracefile_size
, &tracefile_count
);
1202 case 11: /* LTTng sessiond 2.11. Allocates path_name and channel_name. */
1204 ret
= cmd_recv_stream_2_11(conn
, &path_name
,
1205 &channel_name
, &tracefile_size
, &tracefile_count
,
1213 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1217 /* This stream here has one reference on the trace. */
1219 pthread_mutex_lock(&last_relay_stream_id_lock
);
1220 stream_handle
= ++last_relay_stream_id
;
1221 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1223 /* We pass ownership of path_name and channel_name. */
1224 stream
= stream_create(trace
, stream_handle
, path_name
,
1225 channel_name
, tracefile_size
, tracefile_count
);
1227 channel_name
= NULL
;
1230 * Streams are the owners of their trace. Reference to trace is
1231 * kept within stream_create().
1233 ctf_trace_put(trace
);
1236 memset(&reply
, 0, sizeof(reply
));
1237 reply
.handle
= htobe64(stream_handle
);
1239 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1241 reply
.ret_code
= htobe32(LTTNG_OK
);
1244 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1245 sizeof(struct lttcomm_relayd_status_stream
), 0);
1247 ERR("Relay sending stream id");
1248 ret
= (int) send_ret
;
1258 * relay_close_stream: close a specific stream
1260 static int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1261 struct relay_connection
*conn
)
1264 struct relay_session
*session
= conn
->session
;
1265 struct lttcomm_relayd_close_stream stream_info
;
1266 struct lttcomm_relayd_generic_reply reply
;
1267 struct relay_stream
*stream
;
1269 DBG("Close stream received");
1271 if (!session
|| conn
->version_check_done
== 0) {
1272 ERR("Trying to close a stream before version check");
1274 goto end_no_session
;
1277 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
1278 sizeof(struct lttcomm_relayd_close_stream
), 0);
1279 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1281 /* Orderly shutdown. Not necessary to print an error. */
1282 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1284 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1287 goto end_no_session
;
1290 stream
= stream_get_by_id(be64toh(stream_info
.stream_id
));
1297 * Set last_net_seq_num before the close flag. Required by data
1300 pthread_mutex_lock(&stream
->lock
);
1301 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1302 pthread_mutex_unlock(&stream
->lock
);
1305 * This is one of the conditions which may trigger a stream close
1306 * with the others being:
1307 * 1) A close command is received for a stream
1308 * 2) The control connection owning the stream is closed
1309 * 3) We have received all of the stream's data _after_ a close
1312 try_stream_close(stream
);
1313 if (stream
->is_metadata
) {
1314 struct relay_viewer_stream
*vstream
;
1316 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1318 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1320 * Since all the metadata has been sent to the
1321 * viewer and that we have a request to close
1322 * its stream, we can safely teardown the
1323 * corresponding metadata viewer stream.
1325 viewer_stream_put(vstream
);
1327 /* Put local reference. */
1328 viewer_stream_put(vstream
);
1334 memset(&reply
, 0, sizeof(reply
));
1336 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1338 reply
.ret_code
= htobe32(LTTNG_OK
);
1340 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1341 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1343 ERR("Relay sending stream id");
1352 * relay_reset_metadata: reset a metadata stream
1355 int relay_reset_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1356 struct relay_connection
*conn
)
1359 struct relay_session
*session
= conn
->session
;
1360 struct lttcomm_relayd_reset_metadata stream_info
;
1361 struct lttcomm_relayd_generic_reply reply
;
1362 struct relay_stream
*stream
;
1364 DBG("Reset metadata received");
1366 if (!session
|| conn
->version_check_done
== 0) {
1367 ERR("Trying to reset a metadata stream before version check");
1369 goto end_no_session
;
1372 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
1373 sizeof(struct lttcomm_relayd_reset_metadata
), 0);
1374 if (ret
< sizeof(struct lttcomm_relayd_reset_metadata
)) {
1376 /* Orderly shutdown. Not necessary to print an error. */
1377 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1379 ERR("Relay didn't receive valid reset_metadata struct "
1383 goto end_no_session
;
1385 DBG("Update metadata to version %" PRIu64
, be64toh(stream_info
.version
));
1387 /* Unsupported for live sessions for now. */
1388 if (session
->live_timer
!= 0) {
1393 stream
= stream_get_by_id(be64toh(stream_info
.stream_id
));
1398 pthread_mutex_lock(&stream
->lock
);
1399 if (!stream
->is_metadata
) {
1404 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
1405 0, 0, -1, -1, stream
->stream_fd
->fd
, NULL
,
1406 &stream
->stream_fd
->fd
);
1408 ERR("Failed to rotate metadata file %s of channel %s",
1409 stream
->path_name
, stream
->channel_name
);
1414 pthread_mutex_unlock(&stream
->lock
);
1418 memset(&reply
, 0, sizeof(reply
));
1420 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1422 reply
.ret_code
= htobe32(LTTNG_OK
);
1424 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1425 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1427 ERR("Relay sending reset metadata reply");
1436 * relay_unknown_command: send -1 if received unknown command
1438 static void relay_unknown_command(struct relay_connection
*conn
)
1440 struct lttcomm_relayd_generic_reply reply
;
1443 memset(&reply
, 0, sizeof(reply
));
1444 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1445 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1446 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1448 ERR("Relay sending unknown command");
1453 * relay_start: send an acknowledgment to the client to tell if we are
1454 * ready to receive data. We are ready if a session is established.
1456 static int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1457 struct relay_connection
*conn
)
1459 int ret
= htobe32(LTTNG_OK
);
1460 struct lttcomm_relayd_generic_reply reply
;
1461 struct relay_session
*session
= conn
->session
;
1464 DBG("Trying to start the streaming without a session established");
1465 ret
= htobe32(LTTNG_ERR_UNK
);
1468 memset(&reply
, 0, sizeof(reply
));
1469 reply
.ret_code
= ret
;
1470 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1471 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1473 ERR("Relay sending start ack");
1480 * Append padding to the file pointed by the file descriptor fd.
1482 static int write_padding_to_file(int fd
, uint32_t size
)
1491 zeros
= zmalloc(size
);
1492 if (zeros
== NULL
) {
1493 PERROR("zmalloc zeros for padding");
1498 ret
= lttng_write(fd
, zeros
, size
);
1500 PERROR("write padding to file");
1510 int rotate_index_file(struct relay_stream
*stream
)
1513 uint32_t major
, minor
;
1515 /* Put ref on previous index_file. */
1516 if (stream
->index_file
) {
1517 lttng_index_file_put(stream
->index_file
);
1518 stream
->index_file
= NULL
;
1520 major
= stream
->trace
->session
->major
;
1521 minor
= stream
->trace
->session
->minor
;
1522 stream
->index_file
= lttng_index_file_create(stream
->path_name
,
1523 stream
->channel_name
,
1524 -1, -1, stream
->tracefile_size
,
1525 tracefile_array_get_file_index_head(stream
->tfa
),
1526 lttng_to_index_major(major
, minor
),
1527 lttng_to_index_minor(major
, minor
));
1528 if (!stream
->index_file
) {
1540 int do_rotate_stream(struct relay_stream
*stream
)
1544 /* Perform the stream rotation. */
1545 ret
= utils_rotate_stream_file(stream
->path_name
,
1546 stream
->channel_name
, stream
->tracefile_size
,
1547 stream
->tracefile_count
, -1,
1548 -1, stream
->stream_fd
->fd
,
1549 NULL
, &stream
->stream_fd
->fd
);
1551 ERR("Rotating stream output file");
1554 stream
->tracefile_size_current
= 0;
1556 /* Rotate also the index if the stream is not a metadata stream. */
1557 if (!stream
->is_metadata
) {
1558 ret
= rotate_index_file(stream
);
1560 ERR("Failed to rotate index file");
1565 stream
->rotate_at_seq_num
= -1ULL;
1571 * Check if a stream should perform a rotation (for session rotation).
1572 * Must be called with the stream lock held.
1574 * Return 0 on success, a negative value on error.
1577 int check_rotate_stream(struct relay_stream
*stream
)
1581 /* No rotation expected */
1582 if (stream
->rotate_at_seq_num
== -1ULL) {
1587 if (stream
->prev_seq
< stream
->rotate_at_seq_num
) {
1588 DBG("Stream %" PRIu64
" no yet ready for rotation",
1589 stream
->stream_handle
);
1592 } else if (stream
->prev_seq
> stream
->rotate_at_seq_num
) {
1593 DBG("Rotation after too much data has been written in tracefile "
1594 "for stream %" PRIu64
", need to truncate before "
1595 "rotating", stream
->stream_handle
);
1596 fprintf(stderr
, "Rotation after too much data has been written in tracefile "
1597 "for stream %" PRIu64
", need to truncate before "
1598 "rotating\n", stream
->stream_handle
);
1601 DBG("Stream %" PRIu64
" ready for rotation", stream
->stream_handle
);
1604 ret
= do_rotate_stream(stream
);
1611 * relay_recv_metadata: receive the metadata for the session.
1613 static int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1614 struct relay_connection
*conn
)
1618 struct relay_session
*session
= conn
->session
;
1619 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1620 struct relay_stream
*metadata_stream
;
1621 uint64_t data_size
, payload_size
;
1624 ERR("Metadata sent before version check");
1629 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1630 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1631 ERR("Incorrect data size");
1635 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1637 if (data_buffer_size
< data_size
) {
1638 /* In case the realloc fails, we can free the memory */
1641 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1642 if (!tmp_data_ptr
) {
1643 ERR("Allocating data buffer");
1648 data_buffer
= tmp_data_ptr
;
1649 data_buffer_size
= data_size
;
1651 memset(data_buffer
, 0, data_size
);
1652 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1653 size_ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
1654 if (size_ret
< 0 || size_ret
!= data_size
) {
1655 if (size_ret
== 0) {
1656 /* Orderly shutdown. Not necessary to print an error. */
1657 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1659 ERR("Relay didn't receive the whole metadata");
1664 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1666 metadata_stream
= stream_get_by_id(be64toh(metadata_struct
->stream_id
));
1667 if (!metadata_stream
) {
1672 pthread_mutex_lock(&metadata_stream
->lock
);
1674 size_ret
= lttng_write(metadata_stream
->stream_fd
->fd
, metadata_struct
->payload
,
1676 if (size_ret
< payload_size
) {
1677 ERR("Relay error writing metadata on file");
1682 size_ret
= write_padding_to_file(metadata_stream
->stream_fd
->fd
,
1683 be32toh(metadata_struct
->padding_size
));
1688 metadata_stream
->metadata_received
+=
1689 payload_size
+ be32toh(metadata_struct
->padding_size
);
1690 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1691 metadata_stream
->metadata_received
);
1693 ret
= check_rotate_stream(metadata_stream
);
1695 ERR("Check rotate stream");
1700 pthread_mutex_unlock(&metadata_stream
->lock
);
1701 stream_put(metadata_stream
);
1707 * relay_send_version: send relayd version number
1709 static int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1710 struct relay_connection
*conn
)
1713 struct lttcomm_relayd_version reply
, msg
;
1715 conn
->version_check_done
= 1;
1717 /* Get version from the other side. */
1718 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1719 if (ret
< 0 || ret
!= sizeof(msg
)) {
1721 /* Orderly shutdown. Not necessary to print an error. */
1722 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1724 ERR("Relay failed to receive the version values.");
1730 memset(&reply
, 0, sizeof(reply
));
1731 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1732 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1734 /* Major versions must be the same */
1735 if (reply
.major
!= be32toh(msg
.major
)) {
1736 DBG("Incompatible major versions (%u vs %u), deleting session",
1737 reply
.major
, be32toh(msg
.major
));
1738 connection_put(conn
);
1743 conn
->major
= reply
.major
;
1744 /* We adapt to the lowest compatible version */
1745 if (reply
.minor
<= be32toh(msg
.minor
)) {
1746 conn
->minor
= reply
.minor
;
1748 conn
->minor
= be32toh(msg
.minor
);
1751 reply
.major
= htobe32(reply
.major
);
1752 reply
.minor
= htobe32(reply
.minor
);
1753 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1754 sizeof(struct lttcomm_relayd_version
), 0);
1756 ERR("Relay sending version");
1759 DBG("Version check done using protocol %u.%u", conn
->major
,
1767 * Check for data pending for a given stream id from the session daemon.
1769 static int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1770 struct relay_connection
*conn
)
1772 struct relay_session
*session
= conn
->session
;
1773 struct lttcomm_relayd_data_pending msg
;
1774 struct lttcomm_relayd_generic_reply reply
;
1775 struct relay_stream
*stream
;
1777 uint64_t last_net_seq_num
, stream_id
;
1779 DBG("Data pending command received");
1781 if (!session
|| conn
->version_check_done
== 0) {
1782 ERR("Trying to check for data before version check");
1784 goto end_no_session
;
1787 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1788 if (ret
< sizeof(msg
)) {
1790 /* Orderly shutdown. Not necessary to print an error. */
1791 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1793 ERR("Relay didn't receive valid data_pending struct size : %d",
1797 goto end_no_session
;
1800 stream_id
= be64toh(msg
.stream_id
);
1801 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1803 stream
= stream_get_by_id(stream_id
);
1804 if (stream
== NULL
) {
1809 pthread_mutex_lock(&stream
->lock
);
1811 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1812 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1815 /* Avoid wrapping issue */
1816 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1817 /* Data has in fact been written and is NOT pending */
1820 /* Data still being streamed thus pending */
1824 stream
->data_pending_check_done
= true;
1825 pthread_mutex_unlock(&stream
->lock
);
1830 memset(&reply
, 0, sizeof(reply
));
1831 reply
.ret_code
= htobe32(ret
);
1832 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1834 ERR("Relay data pending ret code failed");
1842 * Wait for the control socket to reach a quiescent state.
1844 * Note that for now, when receiving this command from the session
1845 * daemon, this means that every subsequent commands or data received on
1846 * the control socket has been handled. So, this is why we simply return
1849 static int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1850 struct relay_connection
*conn
)
1854 struct relay_stream
*stream
;
1855 struct lttcomm_relayd_quiescent_control msg
;
1856 struct lttcomm_relayd_generic_reply reply
;
1858 DBG("Checking quiescent state on control socket");
1860 if (!conn
->session
|| conn
->version_check_done
== 0) {
1861 ERR("Trying to check for data before version check");
1863 goto end_no_session
;
1866 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1867 if (ret
< sizeof(msg
)) {
1869 /* Orderly shutdown. Not necessary to print an error. */
1870 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1872 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1876 goto end_no_session
;
1879 stream_id
= be64toh(msg
.stream_id
);
1880 stream
= stream_get_by_id(stream_id
);
1884 pthread_mutex_lock(&stream
->lock
);
1885 stream
->data_pending_check_done
= true;
1886 pthread_mutex_unlock(&stream
->lock
);
1887 DBG("Relay quiescent control pending flag set to %" PRIu64
, stream_id
);
1890 memset(&reply
, 0, sizeof(reply
));
1891 reply
.ret_code
= htobe32(LTTNG_OK
);
1892 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1894 ERR("Relay data quiescent control ret code failed");
1902 * Initialize a data pending command. This means that a consumer is about
1903 * to ask for data pending for each stream it holds. Simply iterate over
1904 * all streams of a session and set the data_pending_check_done flag.
1906 * This command returns to the client a LTTNG_OK code.
1908 static int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1909 struct relay_connection
*conn
)
1912 struct lttng_ht_iter iter
;
1913 struct lttcomm_relayd_begin_data_pending msg
;
1914 struct lttcomm_relayd_generic_reply reply
;
1915 struct relay_stream
*stream
;
1916 uint64_t session_id
;
1921 DBG("Init streams for data pending");
1923 if (!conn
->session
|| conn
->version_check_done
== 0) {
1924 ERR("Trying to check for data before version check");
1926 goto end_no_session
;
1929 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1930 if (ret
< sizeof(msg
)) {
1932 /* Orderly shutdown. Not necessary to print an error. */
1933 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1935 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1939 goto end_no_session
;
1942 session_id
= be64toh(msg
.session_id
);
1945 * Iterate over all streams to set the begin data pending flag.
1946 * For now, the streams are indexed by stream handle so we have
1947 * to iterate over all streams to find the one associated with
1948 * the right session_id.
1951 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1953 if (!stream_get(stream
)) {
1956 if (stream
->trace
->session
->id
== session_id
) {
1957 pthread_mutex_lock(&stream
->lock
);
1958 stream
->data_pending_check_done
= false;
1959 pthread_mutex_unlock(&stream
->lock
);
1960 DBG("Set begin data pending flag to stream %" PRIu64
,
1961 stream
->stream_handle
);
1967 memset(&reply
, 0, sizeof(reply
));
1968 /* All good, send back reply. */
1969 reply
.ret_code
= htobe32(LTTNG_OK
);
1971 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1973 ERR("Relay begin data pending send reply failed");
1981 * End data pending command. This will check, for a given session id, if
1982 * each stream associated with it has its data_pending_check_done flag
1983 * set. If not, this means that the client lost track of the stream but
1984 * the data is still being streamed on our side. In this case, we inform
1985 * the client that data is in flight.
1987 * Return to the client if there is data in flight or not with a ret_code.
1989 static int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1990 struct relay_connection
*conn
)
1993 struct lttng_ht_iter iter
;
1994 struct lttcomm_relayd_end_data_pending msg
;
1995 struct lttcomm_relayd_generic_reply reply
;
1996 struct relay_stream
*stream
;
1997 uint64_t session_id
;
1998 uint32_t is_data_inflight
= 0;
2000 DBG("End data pending command");
2002 if (!conn
->session
|| conn
->version_check_done
== 0) {
2003 ERR("Trying to check for data before version check");
2005 goto end_no_session
;
2008 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
2009 if (ret
< sizeof(msg
)) {
2011 /* Orderly shutdown. Not necessary to print an error. */
2012 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2014 ERR("Relay didn't receive valid end data_pending struct size: %d",
2018 goto end_no_session
;
2021 session_id
= be64toh(msg
.session_id
);
2024 * Iterate over all streams to see if the begin data pending
2028 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2030 if (!stream_get(stream
)) {
2033 if (stream
->trace
->session
->id
!= session_id
) {
2037 pthread_mutex_lock(&stream
->lock
);
2038 if (!stream
->data_pending_check_done
) {
2039 if (!stream
->closed
|| !(((int64_t) (stream
->prev_seq
- stream
->last_net_seq_num
)) >= 0)) {
2040 is_data_inflight
= 1;
2041 DBG("Data is still in flight for stream %" PRIu64
,
2042 stream
->stream_handle
);
2043 pthread_mutex_unlock(&stream
->lock
);
2048 pthread_mutex_unlock(&stream
->lock
);
2053 memset(&reply
, 0, sizeof(reply
));
2054 /* All good, send back reply. */
2055 reply
.ret_code
= htobe32(is_data_inflight
);
2057 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2059 ERR("Relay end data pending send reply failed");
2067 * Receive an index for a specific stream.
2069 * Return 0 on success else a negative value.
2071 static int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
2072 struct relay_connection
*conn
)
2075 struct relay_session
*session
= conn
->session
;
2076 struct lttcomm_relayd_index index_info
;
2077 struct relay_index
*index
;
2078 struct lttcomm_relayd_generic_reply reply
;
2079 struct relay_stream
*stream
;
2080 uint64_t net_seq_num
;
2085 DBG("Relay receiving index");
2087 if (!session
|| conn
->version_check_done
== 0) {
2088 ERR("Trying to close a stream before version check");
2090 goto end_no_session
;
2093 msg_len
= lttcomm_relayd_index_len(
2094 lttng_to_index_major(conn
->major
, conn
->minor
),
2095 lttng_to_index_minor(conn
->major
, conn
->minor
));
2096 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &index_info
,
2098 if (ret
< msg_len
) {
2100 /* Orderly shutdown. Not necessary to print an error. */
2101 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2103 ERR("Relay didn't receive valid index struct size : %d", ret
);
2106 goto end_no_session
;
2109 net_seq_num
= be64toh(index_info
.net_seq_num
);
2111 stream
= stream_get_by_id(be64toh(index_info
.relay_stream_id
));
2113 ERR("stream_get_by_id not found");
2117 pthread_mutex_lock(&stream
->lock
);
2119 /* Live beacon handling */
2120 if (index_info
.packet_size
== 0) {
2121 DBG("Received live beacon for stream %" PRIu64
,
2122 stream
->stream_handle
);
2125 * Only flag a stream inactive when it has already
2126 * received data and no indexes are in flight.
2128 if (stream
->index_received_seqcount
> 0
2129 && stream
->indexes_in_flight
== 0) {
2130 stream
->beacon_ts_end
=
2131 be64toh(index_info
.timestamp_end
);
2134 goto end_stream_put
;
2136 stream
->beacon_ts_end
= -1ULL;
2139 if (stream
->ctf_stream_id
== -1ULL) {
2140 stream
->ctf_stream_id
= be64toh(index_info
.stream_id
);
2142 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
2145 ERR("relay_index_get_by_id_or_create index NULL");
2146 goto end_stream_put
;
2148 if (set_index_control_data(index
, &index_info
, conn
)) {
2149 ERR("set_index_control_data error");
2150 relay_index_put(index
);
2152 goto end_stream_put
;
2154 ret
= relay_index_try_flush(index
);
2156 tracefile_array_commit_seq(stream
->tfa
);
2157 stream
->index_received_seqcount
++;
2158 } else if (ret
> 0) {
2162 ERR("relay_index_try_flush error %d", ret
);
2163 relay_index_put(index
);
2168 pthread_mutex_unlock(&stream
->lock
);
2173 memset(&reply
, 0, sizeof(reply
));
2175 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2177 reply
.ret_code
= htobe32(LTTNG_OK
);
2179 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2181 ERR("Relay sending close index id reply");
2190 * Receive the streams_sent message.
2192 * Return 0 on success else a negative value.
2194 static int relay_streams_sent(struct lttcomm_relayd_hdr
*recv_hdr
,
2195 struct relay_connection
*conn
)
2198 struct lttcomm_relayd_generic_reply reply
;
2202 DBG("Relay receiving streams_sent");
2204 if (!conn
->session
|| conn
->version_check_done
== 0) {
2205 ERR("Trying to close a stream before version check");
2207 goto end_no_session
;
2211 * Publish every pending stream in the connection recv list which are
2212 * now ready to be used by the viewer.
2214 publish_connection_local_streams(conn
);
2216 memset(&reply
, 0, sizeof(reply
));
2217 reply
.ret_code
= htobe32(LTTNG_OK
);
2218 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2220 ERR("Relay sending sent_stream reply");
2232 * relay_rotate_stream: rotate a stream to a new tracefile for the session
2233 * rotation feature (not the tracefile rotation feature).
2235 static int relay_rotate_session_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
2236 struct relay_connection
*conn
)
2239 struct relay_session
*session
= conn
->session
;
2240 struct lttcomm_relayd_rotate_stream stream_info
;
2241 struct lttcomm_relayd_generic_reply reply
;
2242 struct relay_stream
*stream
;
2245 DBG("Rotate stream received");
2247 if (!session
|| conn
->version_check_done
== 0) {
2248 ERR("Trying to rotate a stream before version check");
2250 goto end_no_session
;
2253 if (session
->minor
< 11) {
2254 ERR("Unsupported feature before 2.11");
2256 goto end_no_session
;
2259 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
2260 sizeof(stream_info
), 0);
2261 if (ret
< sizeof(stream_info
)) {
2263 /* Orderly shutdown. Not necessary to print an error. */
2264 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2266 ERR("Relay didn't receive valid rotate_stream struct size : %d", ret
);
2269 goto end_no_session
;
2272 stream
= stream_get_by_id(be64toh(stream_info
.stream_id
));
2278 len
= lttng_strnlen(stream_info
.new_pathname
,
2279 sizeof(stream_info
.new_pathname
));
2280 /* Ensure that NULL-terminated and fits in local filename length. */
2281 if (len
== sizeof(stream_info
.new_pathname
) || len
>= LTTNG_NAME_MAX
) {
2282 ret
= -ENAMETOOLONG
;
2283 ERR("Path name too long");
2287 pthread_mutex_lock(&stream
->lock
);
2289 /* Update the trace path (just the folder, the stream name does not change). */
2290 free(stream
->path_name
);
2291 stream
->path_name
= create_output_path(stream_info
.new_pathname
);
2292 if (!stream
->path_name
) {
2293 ERR("Failed to create a new output path");
2294 goto end_stream_unlock
;
2296 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
,
2299 ERR("relay creating output directory");
2302 stream
->chunk_id
= be64toh(stream_info
.new_chunk_id
);
2304 if (stream
->is_metadata
) {
2306 * The metadata stream is sent only over the control connection
2307 * so we know we have all the data to perform the stream
2310 ret
= do_rotate_stream(stream
);
2312 stream
->rotate_at_seq_num
= be64toh(stream_info
.rotate_at_seq_num
);
2313 ret
= check_rotate_stream(stream
);
2316 ERR("Check rotate stream");
2317 goto end_stream_unlock
;
2321 pthread_mutex_unlock(&stream
->lock
);
2324 memset(&reply
, 0, sizeof(reply
));
2326 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2328 reply
.ret_code
= htobe32(LTTNG_OK
);
2330 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2331 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2333 ERR("Relay sending stream id");
2342 * relay_rotate_rename: rename the trace folder after the rotation is
2343 * complete. We are not closing any fd here, just moving the folder, so it
2344 * works even if data is still in flight.
2346 static int relay_rotate_rename(struct lttcomm_relayd_hdr
*recv_hdr
,
2347 struct relay_connection
*conn
)
2350 struct relay_session
*session
= conn
->session
;
2351 struct lttcomm_relayd_rotate_rename stream_info
;
2352 struct lttcomm_relayd_generic_reply reply
;
2354 char *old
= NULL
, *new = NULL
;
2356 DBG("Rotate rename received");
2358 if (!session
|| conn
->version_check_done
== 0) {
2359 ERR("Trying to rename before version check");
2361 goto end_no_session
;
2364 if (session
->minor
< 11) {
2365 ERR("Unsupported feature before 2.11");
2367 goto end_no_session
;
2370 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
2371 sizeof(stream_info
), 0);
2372 if (ret
< sizeof(stream_info
)) {
2374 /* Orderly shutdown. Not necessary to print an error. */
2375 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2377 ERR("Relay didn't receive valid rotate_rename struct size : %d", ret
);
2380 goto end_no_session
;
2383 len
= lttng_strnlen(stream_info
.current_path
,
2384 sizeof(stream_info
.current_path
));
2385 /* Ensure that NULL-terminated and fits in local filename length. */
2386 if (len
== sizeof(stream_info
.current_path
) || len
>= LTTNG_NAME_MAX
) {
2387 ret
= -ENAMETOOLONG
;
2388 ERR("Path name too long");
2392 len
= lttng_strnlen(stream_info
.new_path
,
2393 sizeof(stream_info
.new_path
));
2394 /* Ensure that NULL-terminated and fits in local filename length. */
2395 if (len
== sizeof(stream_info
.new_path
) || len
>= LTTNG_NAME_MAX
) {
2396 ret
= -ENAMETOOLONG
;
2397 ERR("Path name too long");
2401 old
= create_output_path(stream_info
.current_path
);
2403 ERR("Failed to create current output path");
2408 new = create_output_path(stream_info
.new_path
);
2410 ERR("Failed to create new output path");
2415 ret
= utils_mkdir_recursive(new, S_IRWXU
| S_IRWXG
,
2418 ERR("relay creating output directory");
2422 ret
= rename(old
, new);
2423 if (ret
< 0 && errno
!= ENOENT
) {
2424 PERROR("Rename completed rotation chunk");
2430 memset(&reply
, 0, sizeof(reply
));
2432 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2434 reply
.ret_code
= htobe32(LTTNG_OK
);
2436 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2437 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2439 ERR("Relay sending stream id");
2450 int relay_rotate_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
2451 struct relay_connection
*conn
)
2453 struct relay_session
*session
= conn
->session
;
2454 struct lttcomm_relayd_rotate_pending msg
;
2455 struct lttcomm_relayd_generic_reply reply
;
2456 struct lttng_ht_iter iter
;
2457 struct relay_stream
*stream
;
2460 uint32_t rotate_pending
;
2462 DBG("Rotate pending command received");
2464 if (!session
|| conn
->version_check_done
== 0) {
2465 ERR("Trying to check for data before version check");
2467 goto end_no_session
;
2470 if (session
->minor
< 11) {
2471 ERR("Unsupported feature before 2.11");
2473 goto end_no_session
;
2476 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
2477 if (ret
< sizeof(msg
)) {
2479 /* Orderly shutdown. Not necessary to print an error. */
2480 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2482 ERR("Relay didn't receive valid rotate_pending struct size : %d",
2486 goto end_no_session
;
2489 chunk_id
= be64toh(msg
.chunk_id
);
2494 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2496 if (!stream_get(stream
)) {
2499 if (stream
->trace
->session
!= session
) {
2503 pthread_mutex_lock(&stream
->lock
);
2504 if (stream
->rotate_at_seq_num
!= -1ULL) {
2506 DBG("Stream %" PRIu64
" is still rotating",
2507 stream
->stream_handle
);
2508 } else if (stream
->chunk_id
< chunk_id
) {
2510 DBG("Stream %" PRIu64
" did not exist on the consumer "
2511 "when the last rotation started, but is"
2512 "still waiting for data before getting"
2514 stream
->stream_handle
);
2516 pthread_mutex_unlock(&stream
->lock
);
2518 if (rotate_pending
) {
2525 memset(&reply
, 0, sizeof(reply
));
2526 reply
.ret_code
= htobe32(rotate_pending
);
2527 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2529 ERR("Relay rotate pending ret code failed");
2537 * Process the commands received on the control socket
2539 static int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
2540 struct relay_connection
*conn
)
2544 switch (be32toh(recv_hdr
->cmd
)) {
2545 case RELAYD_CREATE_SESSION
:
2546 ret
= relay_create_session(recv_hdr
, conn
);
2548 case RELAYD_ADD_STREAM
:
2549 ret
= relay_add_stream(recv_hdr
, conn
);
2551 case RELAYD_START_DATA
:
2552 ret
= relay_start(recv_hdr
, conn
);
2554 case RELAYD_SEND_METADATA
:
2555 ret
= relay_recv_metadata(recv_hdr
, conn
);
2557 case RELAYD_VERSION
:
2558 ret
= relay_send_version(recv_hdr
, conn
);
2560 case RELAYD_CLOSE_STREAM
:
2561 ret
= relay_close_stream(recv_hdr
, conn
);
2563 case RELAYD_DATA_PENDING
:
2564 ret
= relay_data_pending(recv_hdr
, conn
);
2566 case RELAYD_QUIESCENT_CONTROL
:
2567 ret
= relay_quiescent_control(recv_hdr
, conn
);
2569 case RELAYD_BEGIN_DATA_PENDING
:
2570 ret
= relay_begin_data_pending(recv_hdr
, conn
);
2572 case RELAYD_END_DATA_PENDING
:
2573 ret
= relay_end_data_pending(recv_hdr
, conn
);
2575 case RELAYD_SEND_INDEX
:
2576 ret
= relay_recv_index(recv_hdr
, conn
);
2578 case RELAYD_STREAMS_SENT
:
2579 ret
= relay_streams_sent(recv_hdr
, conn
);
2581 case RELAYD_RESET_METADATA
:
2582 ret
= relay_reset_metadata(recv_hdr
, conn
);
2584 case RELAYD_ROTATE_STREAM
:
2585 ret
= relay_rotate_session_stream(recv_hdr
, conn
);
2587 case RELAYD_ROTATE_RENAME
:
2588 ret
= relay_rotate_rename(recv_hdr
, conn
);
2590 case RELAYD_ROTATE_PENDING
:
2591 ret
= relay_rotate_pending(recv_hdr
, conn
);
2593 case RELAYD_UPDATE_SYNC_INFO
:
2595 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
2596 relay_unknown_command(conn
);
2606 * Handle index for a data stream.
2608 * Called with the stream lock held.
2610 * Return 0 on success else a negative value.
2612 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2616 uint64_t data_offset
;
2617 struct relay_index
*index
;
2619 /* Get data offset because we are about to update the index. */
2620 data_offset
= htobe64(stream
->tracefile_size_current
);
2622 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
2623 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
2626 * Lookup for an existing index for that stream id/sequence
2627 * number. If it exists, the control thread has already received the
2628 * data for it, thus we need to write it to disk.
2630 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
2636 if (rotate_index
|| !stream
->index_file
) {
2637 ret
= rotate_index_file(stream
);
2639 ERR("Failed to rotate index");
2640 /* Put self-ref for this index due to error. */
2641 relay_index_put(index
);
2647 if (relay_index_set_file(index
, stream
->index_file
, data_offset
)) {
2649 /* Put self-ref for this index due to error. */
2650 relay_index_put(index
);
2655 ret
= relay_index_try_flush(index
);
2657 tracefile_array_commit_seq(stream
->tfa
);
2658 stream
->index_received_seqcount
++;
2659 } else if (ret
> 0) {
2663 /* Put self-ref for this index due to error. */
2664 relay_index_put(index
);
2673 * relay_process_data: Process the data received on the data socket
2675 static int relay_process_data(struct relay_connection
*conn
)
2677 int ret
= 0, rotate_index
= 0;
2679 struct relay_stream
*stream
;
2680 struct lttcomm_relayd_data_hdr data_hdr
;
2682 uint64_t net_seq_num
;
2684 struct relay_session
*session
;
2685 bool new_stream
= false, close_requested
= false;
2686 size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
2687 size_t recv_off
= 0;
2688 char data_buffer
[chunk_size
];
2690 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &data_hdr
,
2691 sizeof(struct lttcomm_relayd_data_hdr
), 0);
2694 /* Orderly shutdown. Not necessary to print an error. */
2695 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2697 ERR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2703 stream_id
= be64toh(data_hdr
.stream_id
);
2704 stream
= stream_get_by_id(stream_id
);
2706 ERR("relay_process_data: Cannot find stream %" PRIu64
, stream_id
);
2710 session
= stream
->trace
->session
;
2711 data_size
= be32toh(data_hdr
.data_size
);
2713 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2715 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2716 data_size
, stream_id
, net_seq_num
);
2718 pthread_mutex_lock(&stream
->lock
);
2720 /* Check if a rotation is needed. */
2721 if (stream
->tracefile_size
> 0 &&
2722 (stream
->tracefile_size_current
+ data_size
) >
2723 stream
->tracefile_size
) {
2724 uint64_t old_id
, new_id
;
2726 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
2727 tracefile_array_file_rotate(stream
->tfa
);
2729 /* new_id is updated by utils_rotate_stream_file. */
2732 ret
= utils_rotate_stream_file(stream
->path_name
,
2733 stream
->channel_name
, stream
->tracefile_size
,
2734 stream
->tracefile_count
, -1,
2735 -1, stream
->stream_fd
->fd
,
2736 &new_id
, &stream
->stream_fd
->fd
);
2738 ERR("Rotating stream output file");
2739 goto end_stream_unlock
;
2742 * Reset current size because we just performed a stream
2745 stream
->tracefile_size_current
= 0;
2750 * Index are handled in protocol version 2.4 and above. Also,
2751 * snapshot and index are NOT supported.
2753 if (session
->minor
>= 4 && !session
->snapshot
) {
2754 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2756 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2757 stream
->stream_handle
, net_seq_num
, ret
);
2758 goto end_stream_unlock
;
2762 for (recv_off
= 0; recv_off
< data_size
; recv_off
+= chunk_size
) {
2763 size_t recv_size
= min(data_size
- recv_off
, chunk_size
);
2765 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, recv_size
, 0);
2768 /* Orderly shutdown. Not necessary to print an error. */
2769 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2771 ERR("Socket %d error %d", conn
->sock
->fd
, ret
);
2774 goto end_stream_unlock
;
2777 /* Write data to stream output fd. */
2778 size_ret
= lttng_write(stream
->stream_fd
->fd
, data_buffer
,
2780 if (size_ret
< recv_size
) {
2781 ERR("Relay error writing data to file");
2783 goto end_stream_unlock
;
2786 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
2787 size_ret
, stream
->stream_handle
);
2790 ret
= write_padding_to_file(stream
->stream_fd
->fd
,
2791 be32toh(data_hdr
.padding_size
));
2793 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2794 stream
->stream_handle
, net_seq_num
, ret
);
2795 goto end_stream_unlock
;
2797 stream
->tracefile_size_current
+=
2798 data_size
+ be32toh(data_hdr
.padding_size
);
2799 if (stream
->prev_seq
== -1ULL) {
2803 stream
->prev_seq
= net_seq_num
;
2805 ret
= check_rotate_stream(stream
);
2807 ERR("Check rotate stream");
2808 goto end_stream_unlock
;
2812 close_requested
= stream
->close_requested
;
2813 pthread_mutex_unlock(&stream
->lock
);
2814 if (close_requested
) {
2815 try_stream_close(stream
);
2819 pthread_mutex_lock(&session
->lock
);
2820 uatomic_set(&session
->new_streams
, 1);
2821 pthread_mutex_unlock(&session
->lock
);
2828 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2832 (void) lttng_poll_del(events
, pollfd
);
2834 ret
= close(pollfd
);
2836 ERR("Closing pollfd %d", pollfd
);
2840 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
2841 int pollfd
, struct relay_connection
*conn
)
2843 const char *type_str
;
2845 switch (conn
->type
) {
2850 type_str
= "Control";
2852 case RELAY_VIEWER_COMMAND
:
2853 type_str
= "Viewer Command";
2855 case RELAY_VIEWER_NOTIFICATION
:
2856 type_str
= "Viewer Notification";
2859 type_str
= "Unknown";
2861 cleanup_connection_pollfd(events
, pollfd
);
2862 connection_put(conn
);
2863 DBG("%s connection closed with %d", type_str
, pollfd
);
2867 * This thread does the actual work
2869 static void *relay_thread_worker(void *data
)
2871 int ret
, err
= -1, last_seen_data_fd
= -1;
2873 struct lttng_poll_event events
;
2874 struct lttng_ht
*relay_connections_ht
;
2875 struct lttng_ht_iter iter
;
2876 struct lttcomm_relayd_hdr recv_hdr
;
2877 struct relay_connection
*destroy_conn
= NULL
;
2879 DBG("[thread] Relay worker started");
2881 rcu_register_thread();
2883 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2885 if (testpoint(relayd_thread_worker
)) {
2886 goto error_testpoint
;
2889 health_code_update();
2891 /* table of connections indexed on socket */
2892 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2893 if (!relay_connections_ht
) {
2894 goto relay_connections_ht_error
;
2897 ret
= create_thread_poll_set(&events
, 2);
2899 goto error_poll_create
;
2902 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2909 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2911 health_code_update();
2913 /* Infinite blocking call, waiting for transmission */
2914 DBG3("Relayd worker thread polling...");
2915 health_poll_entry();
2916 ret
= lttng_poll_wait(&events
, -1);
2920 * Restart interrupted system call.
2922 if (errno
== EINTR
) {
2931 * Process control. The control connection is
2932 * prioritized so we don't starve it with high
2933 * throughput tracing data on the data connection.
2935 for (i
= 0; i
< nb_fd
; i
++) {
2936 /* Fetch once the poll data */
2937 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2938 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2940 health_code_update();
2944 * No activity for this FD (poll
2950 /* Thread quit pipe has been closed. Killing thread. */
2951 ret
= check_thread_quit_pipe(pollfd
, revents
);
2957 /* Inspect the relay conn pipe for new connection */
2958 if (pollfd
== relay_conn_pipe
[0]) {
2959 if (revents
& LPOLLIN
) {
2960 struct relay_connection
*conn
;
2962 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
2966 lttng_poll_add(&events
, conn
->sock
->fd
,
2967 LPOLLIN
| LPOLLRDHUP
);
2968 connection_ht_add(relay_connections_ht
, conn
);
2969 DBG("Connection socket %d added", conn
->sock
->fd
);
2970 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2971 ERR("Relay connection pipe error");
2974 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2978 struct relay_connection
*ctrl_conn
;
2980 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
2981 /* If not found, there is a synchronization issue. */
2984 if (ctrl_conn
->type
== RELAY_DATA
) {
2985 if (revents
& LPOLLIN
) {
2987 * Flag the last seen data fd not deleted. It will be
2988 * used as the last seen fd if any fd gets deleted in
2991 last_notdel_data_fd
= pollfd
;
2993 goto put_ctrl_connection
;
2995 assert(ctrl_conn
->type
== RELAY_CONTROL
);
2997 if (revents
& LPOLLIN
) {
2998 ret
= ctrl_conn
->sock
->ops
->recvmsg(ctrl_conn
->sock
,
2999 &recv_hdr
, sizeof(recv_hdr
), 0);
3001 /* Connection closed */
3002 relay_thread_close_connection(&events
, pollfd
,
3005 ret
= relay_process_control(&recv_hdr
, ctrl_conn
);
3007 /* Clear the session on error. */
3008 relay_thread_close_connection(&events
,
3013 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3014 relay_thread_close_connection(&events
,
3016 if (last_seen_data_fd
== pollfd
) {
3017 last_seen_data_fd
= last_notdel_data_fd
;
3020 ERR("Unexpected poll events %u for control sock %d",
3022 connection_put(ctrl_conn
);
3025 put_ctrl_connection
:
3026 connection_put(ctrl_conn
);
3031 * The last loop handled a control request, go back to poll to make
3032 * sure we prioritise the control socket.
3038 if (last_seen_data_fd
>= 0) {
3039 for (i
= 0; i
< nb_fd
; i
++) {
3040 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3042 health_code_update();
3044 if (last_seen_data_fd
== pollfd
) {
3051 /* Process data connection. */
3052 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3053 /* Fetch the poll data. */
3054 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3055 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3056 struct relay_connection
*data_conn
;
3058 health_code_update();
3061 /* No activity for this FD (poll implementation). */
3065 /* Skip the command pipe. It's handled in the first loop. */
3066 if (pollfd
== relay_conn_pipe
[0]) {
3070 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3072 /* Skip it. Might be removed before. */
3075 if (data_conn
->type
== RELAY_CONTROL
) {
3076 goto put_data_connection
;
3078 assert(data_conn
->type
== RELAY_DATA
);
3080 if (revents
& LPOLLIN
) {
3081 ret
= relay_process_data(data_conn
);
3082 /* Connection closed */
3084 relay_thread_close_connection(&events
, pollfd
,
3087 * Every goto restart call sets the last seen fd where
3088 * here we don't really care since we gracefully
3089 * continue the loop after the connection is deleted.
3092 /* Keep last seen port. */
3093 last_seen_data_fd
= pollfd
;
3094 connection_put(data_conn
);
3097 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3098 relay_thread_close_connection(&events
, pollfd
,
3101 ERR("Unknown poll events %u for data sock %d",
3104 put_data_connection
:
3105 connection_put(data_conn
);
3107 last_seen_data_fd
= -1;
3110 /* Normal exit, no error */
3115 /* Cleanup reamaining connection object. */
3117 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3120 health_code_update();
3122 if (session_abort(destroy_conn
->session
)) {
3127 * No need to grab another ref, because we own
3130 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
3135 lttng_poll_clean(&events
);
3137 lttng_ht_destroy(relay_connections_ht
);
3138 relay_connections_ht_error
:
3139 /* Close relay conn pipes */
3140 utils_close_pipe(relay_conn_pipe
);
3142 DBG("Thread exited with error");
3144 DBG("Worker thread cleanup complete");
3148 ERR("Health error occurred in %s", __func__
);
3150 health_unregister(health_relayd
);
3151 rcu_unregister_thread();
3152 lttng_relay_stop_threads();
3157 * Create the relay command pipe to wake thread_manage_apps.
3158 * Closed in cleanup().
3160 static int create_relay_conn_pipe(void)
3164 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
3172 int main(int argc
, char **argv
)
3174 int ret
= 0, retval
= 0;
3177 /* Parse arguments */
3179 if (set_options(argc
, argv
)) {
3184 if (set_signal_handler()) {
3189 /* Try to create directory if -o, --output is specified. */
3190 if (opt_output_path
) {
3191 if (*opt_output_path
!= '/') {
3192 ERR("Please specify an absolute path for -o, --output PATH");
3197 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
3200 ERR("Unable to create %s", opt_output_path
);
3207 if (opt_daemon
|| opt_background
) {
3210 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
3218 * We are in the child. Make sure all other file
3219 * descriptors are closed, in case we are called with
3220 * more opened file descriptors than the standard ones.
3222 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3227 /* Initialize thread health monitoring */
3228 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
3229 if (!health_relayd
) {
3230 PERROR("health_app_create error");
3232 goto exit_health_app_create
;
3235 /* Create thread quit pipe */
3236 if (init_thread_quit_pipe()) {
3238 goto exit_init_data
;
3241 /* Setup the thread apps communication pipe. */
3242 if (create_relay_conn_pipe()) {
3244 goto exit_init_data
;
3247 /* Init relay command queue. */
3248 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
3250 /* Initialize communication library */
3252 lttcomm_inet_init();
3254 /* tables of sessions indexed by session ID */
3255 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3258 goto exit_init_data
;
3261 /* tables of streams indexed by stream ID */
3262 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3263 if (!relay_streams_ht
) {
3265 goto exit_init_data
;
3268 /* tables of streams indexed by stream ID */
3269 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3270 if (!viewer_streams_ht
) {
3272 goto exit_init_data
;
3275 ret
= utils_create_pipe(health_quit_pipe
);
3278 goto exit_health_quit_pipe
;
3281 /* Create thread to manage the client socket */
3282 ret
= pthread_create(&health_thread
, default_pthread_attr(),
3283 thread_manage_health
, (void *) NULL
);
3286 PERROR("pthread_create health");
3288 goto exit_health_thread
;
3291 /* Setup the dispatcher thread */
3292 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
3293 relay_thread_dispatcher
, (void *) NULL
);
3296 PERROR("pthread_create dispatcher");
3298 goto exit_dispatcher_thread
;
3301 /* Setup the worker thread */
3302 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
3303 relay_thread_worker
, NULL
);
3306 PERROR("pthread_create worker");
3308 goto exit_worker_thread
;
3311 /* Setup the listener thread */
3312 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
3313 relay_thread_listener
, (void *) NULL
);
3316 PERROR("pthread_create listener");
3318 goto exit_listener_thread
;
3321 ret
= relayd_live_create(live_uri
);
3323 ERR("Starting live viewer threads");
3329 * This is where we start awaiting program completion (e.g. through
3330 * signal that asks threads to teardown).
3333 ret
= relayd_live_join();
3339 ret
= pthread_join(listener_thread
, &status
);
3342 PERROR("pthread_join listener_thread");
3346 exit_listener_thread
:
3347 ret
= pthread_join(worker_thread
, &status
);
3350 PERROR("pthread_join worker_thread");
3355 ret
= pthread_join(dispatcher_thread
, &status
);
3358 PERROR("pthread_join dispatcher_thread");
3361 exit_dispatcher_thread
:
3363 ret
= pthread_join(health_thread
, &status
);
3366 PERROR("pthread_join health_thread");
3371 utils_close_pipe(health_quit_pipe
);
3372 exit_health_quit_pipe
:
3375 health_app_destroy(health_relayd
);
3376 exit_health_app_create
:
3379 * Wait for all pending call_rcu work to complete before tearing
3380 * down data structures. call_rcu worker may be trying to
3381 * perform lookups in those structures.
3386 /* Ensure all prior call_rcu are done. */