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>
40 #include <urcu/rculist.h>
44 #include <lttng/lttng.h>
45 #include <common/common.h>
46 #include <common/compat/poll.h>
47 #include <common/compat/socket.h>
48 #include <common/compat/endian.h>
49 #include <common/compat/getenv.h>
50 #include <common/defaults.h>
51 #include <common/daemonize.h>
52 #include <common/futex.h>
53 #include <common/sessiond-comm/sessiond-comm.h>
54 #include <common/sessiond-comm/inet.h>
55 #include <common/sessiond-comm/relayd.h>
56 #include <common/uri.h>
57 #include <common/utils.h>
58 #include <common/align.h>
59 #include <common/config/session-config.h>
60 #include <common/dynamic-buffer.h>
61 #include <common/buffer-view.h>
62 #include <common/string-utils/format.h>
65 #include "ctf-trace.h"
68 #include "lttng-relayd.h"
70 #include "health-relayd.h"
71 #include "testpoint.h"
72 #include "viewer-stream.h"
75 #include "connection.h"
76 #include "tracefile-array.h"
77 #include "tcp_keep_alive.h"
78 #include "sessiond-trace-chunks.h"
80 static const char *help_msg
=
81 #ifdef LTTNG_EMBED_HELP
82 #include <lttng-relayd.8.h>
88 enum relay_connection_status
{
89 RELAY_CONNECTION_STATUS_OK
,
90 /* An error occurred while processing an event on the connection. */
91 RELAY_CONNECTION_STATUS_ERROR
,
92 /* Connection closed/shutdown cleanly. */
93 RELAY_CONNECTION_STATUS_CLOSED
,
96 /* command line options */
97 char *opt_output_path
;
98 static int opt_daemon
, opt_background
;
101 * We need to wait for listener and live listener threads, as well as
102 * health check thread, before being ready to signal readiness.
104 #define NR_LTTNG_RELAY_READY 3
105 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
107 /* Size of receive buffer. */
108 #define RECV_DATA_BUFFER_SIZE 65536
110 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
111 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
113 static struct lttng_uri
*control_uri
;
114 static struct lttng_uri
*data_uri
;
115 static struct lttng_uri
*live_uri
;
117 const char *progname
;
119 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
120 static int tracing_group_name_override
;
122 const char * const config_section_name
= "relayd";
125 * Quit pipe for all threads. This permits a single cancellation point
126 * for all threads when receiving an event on the pipe.
128 int thread_quit_pipe
[2] = { -1, -1 };
131 * This pipe is used to inform the worker thread that a command is queued and
132 * ready to be processed.
134 static int relay_conn_pipe
[2] = { -1, -1 };
136 /* Shared between threads */
137 static int dispatch_thread_exit
;
139 static pthread_t listener_thread
;
140 static pthread_t dispatcher_thread
;
141 static pthread_t worker_thread
;
142 static pthread_t health_thread
;
145 * last_relay_stream_id_lock protects last_relay_stream_id increment
146 * atomicity on 32-bit architectures.
148 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
149 static uint64_t last_relay_stream_id
;
152 * Relay command queue.
154 * The relay_thread_listener and relay_thread_dispatcher communicate with this
157 static struct relay_conn_queue relay_conn_queue
;
159 /* Global relay stream hash table. */
160 struct lttng_ht
*relay_streams_ht
;
162 /* Global relay viewer stream hash table. */
163 struct lttng_ht
*viewer_streams_ht
;
165 /* Global relay sessions hash table. */
166 struct lttng_ht
*sessions_ht
;
168 /* Relayd health monitoring */
169 struct health_app
*health_relayd
;
171 struct sessiond_trace_chunk_registry
*sessiond_trace_chunk_registry
;
173 static struct option long_options
[] = {
174 { "control-port", 1, 0, 'C', },
175 { "data-port", 1, 0, 'D', },
176 { "live-port", 1, 0, 'L', },
177 { "daemonize", 0, 0, 'd', },
178 { "background", 0, 0, 'b', },
179 { "group", 1, 0, 'g', },
180 { "help", 0, 0, 'h', },
181 { "output", 1, 0, 'o', },
182 { "verbose", 0, 0, 'v', },
183 { "config", 1, 0, 'f' },
184 { "version", 0, 0, 'V' },
188 static const char *config_ignore_options
[] = { "help", "config", "version" };
191 * Take an option from the getopt output and set it in the right variable to be
194 * Return 0 on success else a negative value.
196 static int set_option(int opt
, const char *arg
, const char *optname
)
202 fprintf(stderr
, "option %s", optname
);
204 fprintf(stderr
, " with arg %s\n", arg
);
208 if (lttng_is_setuid_setgid()) {
209 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
210 "-C, --control-port");
212 ret
= uri_parse(arg
, &control_uri
);
214 ERR("Invalid control URI specified");
217 if (control_uri
->port
== 0) {
218 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
223 if (lttng_is_setuid_setgid()) {
224 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
227 ret
= uri_parse(arg
, &data_uri
);
229 ERR("Invalid data URI specified");
232 if (data_uri
->port
== 0) {
233 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
238 if (lttng_is_setuid_setgid()) {
239 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
242 ret
= uri_parse(arg
, &live_uri
);
244 ERR("Invalid live URI specified");
247 if (live_uri
->port
== 0) {
248 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
259 if (lttng_is_setuid_setgid()) {
260 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
263 tracing_group_name
= strdup(arg
);
264 if (tracing_group_name
== NULL
) {
269 tracing_group_name_override
= 1;
273 ret
= utils_show_help(8, "lttng-relayd", help_msg
);
275 ERR("Cannot show --help for `lttng-relayd`");
280 fprintf(stdout
, "%s\n", VERSION
);
283 if (lttng_is_setuid_setgid()) {
284 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
287 ret
= asprintf(&opt_output_path
, "%s", arg
);
290 PERROR("asprintf opt_output_path");
296 /* Verbose level can increase using multiple -v */
298 lttng_opt_verbose
= config_parse_value(arg
);
300 /* Only 3 level of verbosity (-vvv). */
301 if (lttng_opt_verbose
< 3) {
302 lttng_opt_verbose
+= 1;
307 /* Unknown option or other error.
308 * Error is printed by getopt, just return */
321 * config_entry_handler_cb used to handle options read from a config file.
322 * See config_entry_handler_cb comment in common/config/session-config.h for the
323 * return value conventions.
325 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
329 if (!entry
|| !entry
->name
|| !entry
->value
) {
334 /* Check if the option is to be ignored */
335 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
336 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
341 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
342 /* Ignore if entry name is not fully matched. */
343 if (strcmp(entry
->name
, long_options
[i
].name
)) {
348 * If the option takes no argument on the command line,
349 * we have to check if the value is "true". We support
350 * non-zero numeric values, true, on and yes.
352 if (!long_options
[i
].has_arg
) {
353 ret
= config_parse_value(entry
->value
);
356 WARN("Invalid configuration value \"%s\" for option %s",
357 entry
->value
, entry
->name
);
359 /* False, skip boolean config option. */
364 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
368 WARN("Unrecognized option \"%s\" in daemon configuration file.",
375 static int set_options(int argc
, char **argv
)
377 int c
, ret
= 0, option_index
= 0, retval
= 0;
378 int orig_optopt
= optopt
, orig_optind
= optind
;
379 char *default_address
, *optstring
;
380 const char *config_path
= NULL
;
382 optstring
= utils_generate_optstring(long_options
,
383 sizeof(long_options
) / sizeof(struct option
));
389 /* Check for the --config option */
391 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
392 &option_index
)) != -1) {
396 } else if (c
!= 'f') {
400 if (lttng_is_setuid_setgid()) {
401 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
404 config_path
= utils_expand_path(optarg
);
406 ERR("Failed to resolve path: %s", optarg
);
411 ret
= config_get_section_entries(config_path
, config_section_name
,
412 config_entry_handler
, NULL
);
415 ERR("Invalid configuration option at line %i", ret
);
421 /* Reset getopt's global state */
422 optopt
= orig_optopt
;
423 optind
= orig_optind
;
425 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
430 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
437 /* assign default values */
438 if (control_uri
== NULL
) {
439 ret
= asprintf(&default_address
,
440 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
441 DEFAULT_NETWORK_CONTROL_PORT
);
443 PERROR("asprintf default data address");
448 ret
= uri_parse(default_address
, &control_uri
);
449 free(default_address
);
451 ERR("Invalid control URI specified");
456 if (data_uri
== NULL
) {
457 ret
= asprintf(&default_address
,
458 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
459 DEFAULT_NETWORK_DATA_PORT
);
461 PERROR("asprintf default data address");
466 ret
= uri_parse(default_address
, &data_uri
);
467 free(default_address
);
469 ERR("Invalid data URI specified");
474 if (live_uri
== NULL
) {
475 ret
= asprintf(&default_address
,
476 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
477 DEFAULT_NETWORK_VIEWER_PORT
);
479 PERROR("asprintf default viewer control address");
484 ret
= uri_parse(default_address
, &live_uri
);
485 free(default_address
);
487 ERR("Invalid viewer control URI specified");
498 static void print_global_objects(void)
500 rcu_register_thread();
502 print_viewer_streams();
503 print_relay_streams();
506 rcu_unregister_thread();
512 static void relayd_cleanup(void)
514 print_global_objects();
518 if (viewer_streams_ht
)
519 lttng_ht_destroy(viewer_streams_ht
);
520 if (relay_streams_ht
)
521 lttng_ht_destroy(relay_streams_ht
);
523 lttng_ht_destroy(sessions_ht
);
525 /* free the dynamically allocated opt_output_path */
526 free(opt_output_path
);
528 /* Close thread quit pipes */
529 utils_close_pipe(thread_quit_pipe
);
531 uri_free(control_uri
);
533 /* Live URI is freed in the live thread. */
535 if (tracing_group_name_override
) {
536 free((void *) tracing_group_name
);
541 * Write to writable pipe used to notify a thread.
543 static int notify_thread_pipe(int wpipe
)
547 ret
= lttng_write(wpipe
, "!", 1);
549 PERROR("write poll pipe");
557 static int notify_health_quit_pipe(int *pipe
)
561 ret
= lttng_write(pipe
[1], "4", 1);
563 PERROR("write relay health quit");
572 * Stop all relayd and relayd-live threads.
574 int lttng_relay_stop_threads(void)
578 /* Stopping all threads */
579 DBG("Terminating all threads");
580 if (notify_thread_pipe(thread_quit_pipe
[1])) {
581 ERR("write error on thread quit pipe");
585 if (notify_health_quit_pipe(health_quit_pipe
)) {
586 ERR("write error on health quit pipe");
589 /* Dispatch thread */
590 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
591 futex_nto1_wake(&relay_conn_queue
.futex
);
593 if (relayd_live_stop()) {
594 ERR("Error stopping live threads");
601 * Signal handler for the daemon
603 * Simply stop all worker threads, leaving main() return gracefully after
604 * joining all threads and calling cleanup().
606 static void sighandler(int sig
)
610 DBG("SIGINT caught");
611 if (lttng_relay_stop_threads()) {
612 ERR("Error stopping threads");
616 DBG("SIGTERM caught");
617 if (lttng_relay_stop_threads()) {
618 ERR("Error stopping threads");
622 CMM_STORE_SHARED(recv_child_signal
, 1);
630 * Setup signal handler for :
631 * SIGINT, SIGTERM, SIGPIPE
633 static int set_signal_handler(void)
639 if ((ret
= sigemptyset(&sigset
)) < 0) {
640 PERROR("sigemptyset");
647 sa
.sa_handler
= sighandler
;
648 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
653 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
658 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
663 sa
.sa_handler
= SIG_IGN
;
664 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
669 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
674 void lttng_relay_notify_ready(void)
676 /* Notify the parent of the fork() process that we are ready. */
677 if (opt_daemon
|| opt_background
) {
678 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
679 kill(child_ppid
, SIGUSR1
);
685 * Init thread quit pipe.
687 * Return -1 on error or 0 if all pipes are created.
689 static int init_thread_quit_pipe(void)
693 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
699 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
701 static int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
705 if (events
== NULL
|| size
== 0) {
710 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
716 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
728 * Check if the thread quit pipe was triggered.
730 * Return 1 if it was triggered else 0;
732 static int check_thread_quit_pipe(int fd
, uint32_t events
)
734 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
742 * Create and init socket from uri.
744 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
)
747 struct lttcomm_sock
*sock
= NULL
;
749 sock
= lttcomm_alloc_sock_from_uri(uri
);
751 ERR("Allocating socket");
755 ret
= lttcomm_create_sock(sock
);
759 DBG("Listening on sock %d", sock
->fd
);
761 ret
= sock
->ops
->bind(sock
);
763 PERROR("Failed to bind socket");
767 ret
= sock
->ops
->listen(sock
, -1);
777 lttcomm_destroy_sock(sock
);
783 * This thread manages the listening for new connections on the network
785 static void *relay_thread_listener(void *data
)
787 int i
, ret
, pollfd
, err
= -1;
788 uint32_t revents
, nb_fd
;
789 struct lttng_poll_event events
;
790 struct lttcomm_sock
*control_sock
, *data_sock
;
792 DBG("[thread] Relay listener started");
794 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
796 health_code_update();
798 control_sock
= relay_socket_create(control_uri
);
800 goto error_sock_control
;
803 data_sock
= relay_socket_create(data_uri
);
805 goto error_sock_relay
;
809 * Pass 3 as size here for the thread quit pipe, control and
812 ret
= create_thread_poll_set(&events
, 3);
814 goto error_create_poll
;
817 /* Add the control socket */
818 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
823 /* Add the data socket */
824 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
829 lttng_relay_notify_ready();
831 if (testpoint(relayd_thread_listener
)) {
832 goto error_testpoint
;
836 health_code_update();
838 DBG("Listener accepting connections");
842 ret
= lttng_poll_wait(&events
, -1);
846 * Restart interrupted system call.
848 if (errno
== EINTR
) {
856 DBG("Relay new connection received");
857 for (i
= 0; i
< nb_fd
; i
++) {
858 health_code_update();
860 /* Fetch once the poll data */
861 revents
= LTTNG_POLL_GETEV(&events
, i
);
862 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
864 /* Thread quit pipe has been closed. Killing thread. */
865 ret
= check_thread_quit_pipe(pollfd
, revents
);
871 if (revents
& LPOLLIN
) {
873 * A new connection is requested, therefore a
874 * sessiond/consumerd connection is allocated in
875 * this thread, enqueued to a global queue and
876 * dequeued (and freed) in the worker thread.
879 struct relay_connection
*new_conn
;
880 struct lttcomm_sock
*newsock
;
881 enum connection_type type
;
883 if (pollfd
== data_sock
->fd
) {
885 newsock
= data_sock
->ops
->accept(data_sock
);
886 DBG("Relay data connection accepted, socket %d",
889 assert(pollfd
== control_sock
->fd
);
890 type
= RELAY_CONTROL
;
891 newsock
= control_sock
->ops
->accept(control_sock
);
892 DBG("Relay control connection accepted, socket %d",
896 PERROR("accepting sock");
900 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
903 PERROR("setsockopt inet");
904 lttcomm_destroy_sock(newsock
);
908 ret
= socket_apply_keep_alive_config(newsock
->fd
);
910 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
912 lttcomm_destroy_sock(newsock
);
916 new_conn
= connection_create(newsock
, type
);
918 lttcomm_destroy_sock(newsock
);
922 /* Enqueue request for the dispatcher thread. */
923 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
927 * Wake the dispatch queue futex.
928 * Implicit memory barrier with the
929 * exchange in cds_wfcq_enqueue.
931 futex_nto1_wake(&relay_conn_queue
.futex
);
932 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
933 ERR("socket poll error");
936 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
946 lttng_poll_clean(&events
);
948 if (data_sock
->fd
>= 0) {
949 ret
= data_sock
->ops
->close(data_sock
);
954 lttcomm_destroy_sock(data_sock
);
956 if (control_sock
->fd
>= 0) {
957 ret
= control_sock
->ops
->close(control_sock
);
962 lttcomm_destroy_sock(control_sock
);
966 ERR("Health error occurred in %s", __func__
);
968 health_unregister(health_relayd
);
969 DBG("Relay listener thread cleanup complete");
970 lttng_relay_stop_threads();
975 * This thread manages the dispatching of the requests to worker threads
977 static void *relay_thread_dispatcher(void *data
)
981 struct cds_wfcq_node
*node
;
982 struct relay_connection
*new_conn
= NULL
;
984 DBG("[thread] Relay dispatcher started");
986 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
988 if (testpoint(relayd_thread_dispatcher
)) {
989 goto error_testpoint
;
992 health_code_update();
995 health_code_update();
997 /* Atomically prepare the queue futex */
998 futex_nto1_prepare(&relay_conn_queue
.futex
);
1000 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1005 health_code_update();
1007 /* Dequeue commands */
1008 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1009 &relay_conn_queue
.tail
);
1011 DBG("Woken up but nothing in the relay command queue");
1012 /* Continue thread execution */
1015 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1017 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1020 * Inform worker thread of the new request. This
1021 * call is blocking so we can be assured that
1022 * the data will be read at some point in time
1023 * or wait to the end of the world :)
1025 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1027 PERROR("write connection pipe");
1028 connection_put(new_conn
);
1031 } while (node
!= NULL
);
1033 /* Futex wait on queue. Blocking call on futex() */
1034 health_poll_entry();
1035 futex_nto1_wait(&relay_conn_queue
.futex
);
1039 /* Normal exit, no error */
1046 ERR("Health error occurred in %s", __func__
);
1048 health_unregister(health_relayd
);
1049 DBG("Dispatch thread dying");
1050 lttng_relay_stop_threads();
1054 static bool session_streams_have_index(const struct relay_session
*session
)
1056 return session
->minor
>= 4 && !session
->snapshot
;
1060 * Handle the RELAYD_CREATE_SESSION command.
1062 * On success, send back the session id or else return a negative value.
1064 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1065 struct relay_connection
*conn
,
1066 const struct lttng_buffer_view
*payload
)
1070 struct relay_session
*session
= NULL
;
1071 struct lttcomm_relayd_status_session reply
= {};
1072 char session_name
[LTTNG_NAME_MAX
] = {};
1073 char hostname
[LTTNG_HOST_NAME_MAX
] = {};
1074 uint32_t live_timer
= 0;
1075 bool snapshot
= false;
1076 /* Left nil for peers < 2.11. */
1077 lttng_uuid sessiond_uuid
= {};
1078 LTTNG_OPTIONAL(uint64_t) id_sessiond
= {};
1079 LTTNG_OPTIONAL(uint64_t) current_chunk_id
= {};
1080 LTTNG_OPTIONAL(time_t) creation_time
= {};
1082 if (conn
->minor
< 4) {
1083 /* From 2.1 to 2.3 */
1085 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1086 /* From 2.4 to 2.10 */
1087 ret
= cmd_create_session_2_4(payload
, session_name
,
1088 hostname
, &live_timer
, &snapshot
);
1090 bool has_current_chunk
;
1091 uint64_t current_chunk_id_value
;
1092 time_t creation_time_value
;
1093 uint64_t id_sessiond_value
;
1095 /* From 2.11 to ... */
1096 ret
= cmd_create_session_2_11(payload
, session_name
, hostname
,
1097 &live_timer
, &snapshot
, &id_sessiond_value
,
1098 sessiond_uuid
, &has_current_chunk
,
1099 ¤t_chunk_id_value
, &creation_time_value
);
1100 if (lttng_uuid_is_nil(sessiond_uuid
)) {
1101 /* The nil UUID is reserved for pre-2.11 clients. */
1102 ERR("Illegal nil UUID announced by peer in create session command");
1106 LTTNG_OPTIONAL_SET(&id_sessiond
, id_sessiond_value
);
1107 LTTNG_OPTIONAL_SET(&creation_time
, creation_time_value
);
1108 if (has_current_chunk
) {
1109 LTTNG_OPTIONAL_SET(¤t_chunk_id
,
1110 current_chunk_id_value
);
1118 session
= session_create(session_name
, hostname
, live_timer
,
1119 snapshot
, sessiond_uuid
,
1120 id_sessiond
.is_set
? &id_sessiond
.value
: NULL
,
1121 current_chunk_id
.is_set
? ¤t_chunk_id
.value
: NULL
,
1122 creation_time
.is_set
? &creation_time
.value
: NULL
,
1123 conn
->major
, conn
->minor
);
1128 assert(!conn
->session
);
1129 conn
->session
= session
;
1130 DBG("Created session %" PRIu64
, session
->id
);
1132 reply
.session_id
= htobe64(session
->id
);
1136 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1138 reply
.ret_code
= htobe32(LTTNG_OK
);
1141 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1142 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1143 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1147 if (ret
< 0 && session
) {
1148 session_put(session
);
1154 * When we have received all the streams and the metadata for a channel,
1155 * we make them visible to the viewer threads.
1157 static void publish_connection_local_streams(struct relay_connection
*conn
)
1159 struct relay_stream
*stream
;
1160 struct relay_session
*session
= conn
->session
;
1163 * We publish all streams belonging to a session atomically wrt
1166 pthread_mutex_lock(&session
->lock
);
1168 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1170 stream_publish(stream
);
1175 * Inform the viewer that there are new streams in the session.
1177 if (session
->viewer_attached
) {
1178 uatomic_set(&session
->new_streams
, 1);
1180 pthread_mutex_unlock(&session
->lock
);
1183 static int conform_channel_path(char *channel_path
)
1187 if (strstr("../", channel_path
)) {
1188 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1194 if (*channel_path
== '/') {
1195 const size_t len
= strlen(channel_path
);
1198 * Channel paths from peers prior to 2.11 are expressed as an
1199 * absolute path that is, in reality, relative to the relay
1200 * daemon's output directory. Remove the leading slash so it
1201 * is correctly interpreted as a relative path later on.
1203 * len (and not len - 1) is used to copy the trailing NULL.
1205 bcopy(channel_path
+ 1, channel_path
, len
);
1212 * relay_add_stream: allocate a new stream for a session
1214 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1215 struct relay_connection
*conn
,
1216 const struct lttng_buffer_view
*payload
)
1220 struct relay_session
*session
= conn
->session
;
1221 struct relay_stream
*stream
= NULL
;
1222 struct lttcomm_relayd_status_stream reply
;
1223 struct ctf_trace
*trace
= NULL
;
1224 uint64_t stream_handle
= -1ULL;
1225 char *path_name
= NULL
, *channel_name
= NULL
;
1226 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1227 LTTNG_OPTIONAL(uint64_t) stream_chunk_id
= {};
1229 if (!session
|| !conn
->version_check_done
) {
1230 ERR("Trying to add a stream before version check");
1232 goto end_no_session
;
1235 if (session
->minor
== 1) {
1237 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1239 } else if (session
->minor
> 1 && session
->minor
< 11) {
1240 /* From 2.2 to 2.10 */
1241 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1242 &channel_name
, &tracefile_size
, &tracefile_count
);
1244 /* From 2.11 to ... */
1245 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1246 &channel_name
, &tracefile_size
, &tracefile_count
,
1247 &stream_chunk_id
.value
);
1248 stream_chunk_id
.is_set
= true;
1255 if (conform_channel_path(path_name
)) {
1259 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1263 /* This stream here has one reference on the trace. */
1265 pthread_mutex_lock(&last_relay_stream_id_lock
);
1266 stream_handle
= ++last_relay_stream_id
;
1267 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1269 /* We pass ownership of path_name and channel_name. */
1270 stream
= stream_create(trace
, stream_handle
, path_name
,
1271 channel_name
, tracefile_size
, tracefile_count
);
1273 channel_name
= NULL
;
1276 * Streams are the owners of their trace. Reference to trace is
1277 * kept within stream_create().
1279 ctf_trace_put(trace
);
1282 memset(&reply
, 0, sizeof(reply
));
1283 reply
.handle
= htobe64(stream_handle
);
1285 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1287 reply
.ret_code
= htobe32(LTTNG_OK
);
1290 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1291 sizeof(struct lttcomm_relayd_status_stream
), 0);
1292 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1293 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1305 * relay_close_stream: close a specific stream
1307 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1308 struct relay_connection
*conn
,
1309 const struct lttng_buffer_view
*payload
)
1313 struct relay_session
*session
= conn
->session
;
1314 struct lttcomm_relayd_close_stream stream_info
;
1315 struct lttcomm_relayd_generic_reply reply
;
1316 struct relay_stream
*stream
;
1318 DBG("Close stream received");
1320 if (!session
|| !conn
->version_check_done
) {
1321 ERR("Trying to close a stream before version check");
1323 goto end_no_session
;
1326 if (payload
->size
< sizeof(stream_info
)) {
1327 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1328 sizeof(stream_info
), payload
->size
);
1330 goto end_no_session
;
1332 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1333 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1334 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1336 stream
= stream_get_by_id(stream_info
.stream_id
);
1343 * Set last_net_seq_num before the close flag. Required by data
1346 pthread_mutex_lock(&stream
->lock
);
1347 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1348 pthread_mutex_unlock(&stream
->lock
);
1351 * This is one of the conditions which may trigger a stream close
1352 * with the others being:
1353 * 1) A close command is received for a stream
1354 * 2) The control connection owning the stream is closed
1355 * 3) We have received all of the stream's data _after_ a close
1358 try_stream_close(stream
);
1359 if (stream
->is_metadata
) {
1360 struct relay_viewer_stream
*vstream
;
1362 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1364 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1366 * Since all the metadata has been sent to the
1367 * viewer and that we have a request to close
1368 * its stream, we can safely teardown the
1369 * corresponding metadata viewer stream.
1371 viewer_stream_put(vstream
);
1373 /* Put local reference. */
1374 viewer_stream_put(vstream
);
1381 memset(&reply
, 0, sizeof(reply
));
1383 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1385 reply
.ret_code
= htobe32(LTTNG_OK
);
1387 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1388 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1389 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1390 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1400 * relay_reset_metadata: reset a metadata stream
1403 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1404 struct relay_connection
*conn
,
1405 const struct lttng_buffer_view
*payload
)
1409 struct relay_session
*session
= conn
->session
;
1410 struct lttcomm_relayd_reset_metadata stream_info
;
1411 struct lttcomm_relayd_generic_reply reply
;
1412 struct relay_stream
*stream
;
1414 DBG("Reset metadata received");
1416 if (!session
|| !conn
->version_check_done
) {
1417 ERR("Trying to reset a metadata stream before version check");
1419 goto end_no_session
;
1422 if (payload
->size
< sizeof(stream_info
)) {
1423 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1424 sizeof(stream_info
), payload
->size
);
1426 goto end_no_session
;
1428 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1429 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1430 stream_info
.version
= be64toh(stream_info
.version
);
1432 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1434 /* Unsupported for live sessions for now. */
1435 if (session
->live_timer
!= 0) {
1440 stream
= stream_get_by_id(stream_info
.stream_id
);
1445 pthread_mutex_lock(&stream
->lock
);
1446 if (!stream
->is_metadata
) {
1451 ret
= stream_reset_file(stream
);
1453 ERR("Failed to reset metadata stream %" PRIu64
1454 ": stream_path = %s, channel = %s",
1455 stream
->stream_handle
, stream
->path_name
,
1456 stream
->channel_name
);
1460 pthread_mutex_unlock(&stream
->lock
);
1464 memset(&reply
, 0, sizeof(reply
));
1466 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1468 reply
.ret_code
= htobe32(LTTNG_OK
);
1470 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1471 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1472 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1473 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1483 * relay_unknown_command: send -1 if received unknown command
1485 static void relay_unknown_command(struct relay_connection
*conn
)
1487 struct lttcomm_relayd_generic_reply reply
;
1490 memset(&reply
, 0, sizeof(reply
));
1491 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1492 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1493 if (send_ret
< sizeof(reply
)) {
1494 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1499 * relay_start: send an acknowledgment to the client to tell if we are
1500 * ready to receive data. We are ready if a session is established.
1502 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1503 struct relay_connection
*conn
,
1504 const struct lttng_buffer_view
*payload
)
1508 struct lttcomm_relayd_generic_reply reply
;
1509 struct relay_session
*session
= conn
->session
;
1512 DBG("Trying to start the streaming without a session established");
1513 ret
= htobe32(LTTNG_ERR_UNK
);
1516 memset(&reply
, 0, sizeof(reply
));
1517 reply
.ret_code
= htobe32(LTTNG_OK
);
1518 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1520 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1521 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1530 * relay_recv_metadata: receive the metadata for the session.
1532 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1533 struct relay_connection
*conn
,
1534 const struct lttng_buffer_view
*payload
)
1537 struct relay_session
*session
= conn
->session
;
1538 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1539 struct relay_stream
*metadata_stream
;
1540 uint64_t metadata_payload_size
;
1541 struct lttng_buffer_view packet_view
;
1544 ERR("Metadata sent before version check");
1549 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1550 ERR("Incorrect data size");
1554 metadata_payload_size
= recv_hdr
->data_size
-
1555 sizeof(struct lttcomm_relayd_metadata_payload
);
1557 memcpy(&metadata_payload_header
, payload
->data
,
1558 sizeof(metadata_payload_header
));
1559 metadata_payload_header
.stream_id
= be64toh(
1560 metadata_payload_header
.stream_id
);
1561 metadata_payload_header
.padding_size
= be32toh(
1562 metadata_payload_header
.padding_size
);
1564 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1565 if (!metadata_stream
) {
1570 packet_view
= lttng_buffer_view_from_view(payload
,
1571 sizeof(metadata_payload_header
), metadata_payload_size
);
1572 if (!packet_view
.data
) {
1573 ERR("Invalid metadata packet length announced by header");
1578 pthread_mutex_lock(&metadata_stream
->lock
);
1579 ret
= stream_write(metadata_stream
, &packet_view
,
1580 metadata_payload_header
.padding_size
);
1581 pthread_mutex_unlock(&metadata_stream
->lock
);
1587 stream_put(metadata_stream
);
1593 * relay_send_version: send relayd version number
1595 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1596 struct relay_connection
*conn
,
1597 const struct lttng_buffer_view
*payload
)
1601 struct lttcomm_relayd_version reply
, msg
;
1602 bool compatible
= true;
1604 conn
->version_check_done
= true;
1606 /* Get version from the other side. */
1607 if (payload
->size
< sizeof(msg
)) {
1608 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1609 sizeof(msg
), payload
->size
);
1614 memcpy(&msg
, payload
->data
, sizeof(msg
));
1615 msg
.major
= be32toh(msg
.major
);
1616 msg
.minor
= be32toh(msg
.minor
);
1618 memset(&reply
, 0, sizeof(reply
));
1619 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1620 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1622 /* Major versions must be the same */
1623 if (reply
.major
!= msg
.major
) {
1624 DBG("Incompatible major versions (%u vs %u), deleting session",
1625 reply
.major
, msg
.major
);
1629 conn
->major
= reply
.major
;
1630 /* We adapt to the lowest compatible version */
1631 if (reply
.minor
<= msg
.minor
) {
1632 conn
->minor
= reply
.minor
;
1634 conn
->minor
= msg
.minor
;
1637 reply
.major
= htobe32(reply
.major
);
1638 reply
.minor
= htobe32(reply
.minor
);
1639 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1641 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1642 ERR("Failed to send \"send version\" command reply (ret = %zd)",
1655 DBG("Version check done using protocol %u.%u", conn
->major
,
1663 * Check for data pending for a given stream id from the session daemon.
1665 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1666 struct relay_connection
*conn
,
1667 const struct lttng_buffer_view
*payload
)
1669 struct relay_session
*session
= conn
->session
;
1670 struct lttcomm_relayd_data_pending msg
;
1671 struct lttcomm_relayd_generic_reply reply
;
1672 struct relay_stream
*stream
;
1675 uint64_t stream_seq
;
1677 DBG("Data pending command received");
1679 if (!session
|| !conn
->version_check_done
) {
1680 ERR("Trying to check for data before version check");
1682 goto end_no_session
;
1685 if (payload
->size
< sizeof(msg
)) {
1686 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
1687 sizeof(msg
), payload
->size
);
1689 goto end_no_session
;
1691 memcpy(&msg
, payload
->data
, sizeof(msg
));
1692 msg
.stream_id
= be64toh(msg
.stream_id
);
1693 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1695 stream
= stream_get_by_id(msg
.stream_id
);
1696 if (stream
== NULL
) {
1701 pthread_mutex_lock(&stream
->lock
);
1703 if (session_streams_have_index(session
)) {
1705 * Ensure that both the index and stream data have been
1706 * flushed up to the requested point.
1708 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
1710 stream_seq
= stream
->prev_data_seq
;
1712 DBG("Data pending for stream id %" PRIu64
": prev_data_seq %" PRIu64
1713 ", prev_index_seq %" PRIu64
1714 ", and last_seq %" PRIu64
, msg
.stream_id
,
1715 stream
->prev_data_seq
, stream
->prev_index_seq
,
1716 msg
.last_net_seq_num
);
1718 /* Avoid wrapping issue */
1719 if (((int64_t) (stream_seq
- msg
.last_net_seq_num
)) >= 0) {
1720 /* Data has in fact been written and is NOT pending */
1723 /* Data still being streamed thus pending */
1727 stream
->data_pending_check_done
= true;
1728 pthread_mutex_unlock(&stream
->lock
);
1733 memset(&reply
, 0, sizeof(reply
));
1734 reply
.ret_code
= htobe32(ret
);
1735 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1736 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1737 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
1747 * Wait for the control socket to reach a quiescent state.
1749 * Note that for now, when receiving this command from the session
1750 * daemon, this means that every subsequent commands or data received on
1751 * the control socket has been handled. So, this is why we simply return
1754 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
1755 struct relay_connection
*conn
,
1756 const struct lttng_buffer_view
*payload
)
1760 struct relay_stream
*stream
;
1761 struct lttcomm_relayd_quiescent_control msg
;
1762 struct lttcomm_relayd_generic_reply reply
;
1764 DBG("Checking quiescent state on control socket");
1766 if (!conn
->session
|| !conn
->version_check_done
) {
1767 ERR("Trying to check for data before version check");
1769 goto end_no_session
;
1772 if (payload
->size
< sizeof(msg
)) {
1773 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
1774 sizeof(msg
), payload
->size
);
1776 goto end_no_session
;
1778 memcpy(&msg
, payload
->data
, sizeof(msg
));
1779 msg
.stream_id
= be64toh(msg
.stream_id
);
1781 stream
= stream_get_by_id(msg
.stream_id
);
1785 pthread_mutex_lock(&stream
->lock
);
1786 stream
->data_pending_check_done
= true;
1787 pthread_mutex_unlock(&stream
->lock
);
1789 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
1792 memset(&reply
, 0, sizeof(reply
));
1793 reply
.ret_code
= htobe32(LTTNG_OK
);
1794 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1795 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1796 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
1808 * Initialize a data pending command. This means that a consumer is about
1809 * to ask for data pending for each stream it holds. Simply iterate over
1810 * all streams of a session and set the data_pending_check_done flag.
1812 * This command returns to the client a LTTNG_OK code.
1814 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1815 struct relay_connection
*conn
,
1816 const struct lttng_buffer_view
*payload
)
1820 struct lttng_ht_iter iter
;
1821 struct lttcomm_relayd_begin_data_pending msg
;
1822 struct lttcomm_relayd_generic_reply reply
;
1823 struct relay_stream
*stream
;
1828 DBG("Init streams for data pending");
1830 if (!conn
->session
|| !conn
->version_check_done
) {
1831 ERR("Trying to check for data before version check");
1833 goto end_no_session
;
1836 if (payload
->size
< sizeof(msg
)) {
1837 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
1838 sizeof(msg
), payload
->size
);
1840 goto end_no_session
;
1842 memcpy(&msg
, payload
->data
, sizeof(msg
));
1843 msg
.session_id
= be64toh(msg
.session_id
);
1846 * Iterate over all streams to set the begin data pending flag.
1847 * For now, the streams are indexed by stream handle so we have
1848 * to iterate over all streams to find the one associated with
1849 * the right session_id.
1852 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1854 if (!stream_get(stream
)) {
1857 if (stream
->trace
->session
->id
== msg
.session_id
) {
1858 pthread_mutex_lock(&stream
->lock
);
1859 stream
->data_pending_check_done
= false;
1860 pthread_mutex_unlock(&stream
->lock
);
1861 DBG("Set begin data pending flag to stream %" PRIu64
,
1862 stream
->stream_handle
);
1868 memset(&reply
, 0, sizeof(reply
));
1869 /* All good, send back reply. */
1870 reply
.ret_code
= htobe32(LTTNG_OK
);
1872 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1873 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1874 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
1886 * End data pending command. This will check, for a given session id, if
1887 * each stream associated with it has its data_pending_check_done flag
1888 * set. If not, this means that the client lost track of the stream but
1889 * the data is still being streamed on our side. In this case, we inform
1890 * the client that data is in flight.
1892 * Return to the client if there is data in flight or not with a ret_code.
1894 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1895 struct relay_connection
*conn
,
1896 const struct lttng_buffer_view
*payload
)
1900 struct lttng_ht_iter iter
;
1901 struct lttcomm_relayd_end_data_pending msg
;
1902 struct lttcomm_relayd_generic_reply reply
;
1903 struct relay_stream
*stream
;
1904 uint32_t is_data_inflight
= 0;
1906 DBG("End data pending command");
1908 if (!conn
->session
|| !conn
->version_check_done
) {
1909 ERR("Trying to check for data before version check");
1911 goto end_no_session
;
1914 if (payload
->size
< sizeof(msg
)) {
1915 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
1916 sizeof(msg
), payload
->size
);
1918 goto end_no_session
;
1920 memcpy(&msg
, payload
->data
, sizeof(msg
));
1921 msg
.session_id
= be64toh(msg
.session_id
);
1924 * Iterate over all streams to see if the begin data pending
1928 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1930 if (!stream_get(stream
)) {
1933 if (stream
->trace
->session
->id
!= msg
.session_id
) {
1937 pthread_mutex_lock(&stream
->lock
);
1938 if (!stream
->data_pending_check_done
) {
1939 uint64_t stream_seq
;
1941 if (session_streams_have_index(conn
->session
)) {
1943 * Ensure that both the index and stream data have been
1944 * flushed up to the requested point.
1946 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
1948 stream_seq
= stream
->prev_data_seq
;
1950 if (!stream
->closed
|| !(((int64_t) (stream_seq
- stream
->last_net_seq_num
)) >= 0)) {
1951 is_data_inflight
= 1;
1952 DBG("Data is still in flight for stream %" PRIu64
,
1953 stream
->stream_handle
);
1954 pthread_mutex_unlock(&stream
->lock
);
1959 pthread_mutex_unlock(&stream
->lock
);
1964 memset(&reply
, 0, sizeof(reply
));
1965 /* All good, send back reply. */
1966 reply
.ret_code
= htobe32(is_data_inflight
);
1968 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1969 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1970 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
1982 * Receive an index for a specific stream.
1984 * Return 0 on success else a negative value.
1986 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
1987 struct relay_connection
*conn
,
1988 const struct lttng_buffer_view
*payload
)
1992 struct relay_session
*session
= conn
->session
;
1993 struct lttcomm_relayd_index index_info
;
1994 struct lttcomm_relayd_generic_reply reply
;
1995 struct relay_stream
*stream
;
2000 DBG("Relay receiving index");
2002 if (!session
|| !conn
->version_check_done
) {
2003 ERR("Trying to close a stream before version check");
2005 goto end_no_session
;
2008 msg_len
= lttcomm_relayd_index_len(
2009 lttng_to_index_major(conn
->major
, conn
->minor
),
2010 lttng_to_index_minor(conn
->major
, conn
->minor
));
2011 if (payload
->size
< msg_len
) {
2012 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2013 msg_len
, payload
->size
);
2015 goto end_no_session
;
2017 memcpy(&index_info
, payload
->data
, msg_len
);
2018 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2019 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2020 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2021 index_info
.content_size
= be64toh(index_info
.content_size
);
2022 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2023 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2024 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2025 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2027 if (conn
->minor
>= 8) {
2028 index_info
.stream_instance_id
=
2029 be64toh(index_info
.stream_instance_id
);
2030 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2033 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2035 ERR("stream_get_by_id not found");
2040 pthread_mutex_lock(&stream
->lock
);
2041 ret
= stream_add_index(stream
, &index_info
);
2042 pthread_mutex_unlock(&stream
->lock
);
2044 goto end_stream_put
;
2050 memset(&reply
, 0, sizeof(reply
));
2052 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2054 reply
.ret_code
= htobe32(LTTNG_OK
);
2056 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2057 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2058 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2067 * Receive the streams_sent message.
2069 * Return 0 on success else a negative value.
2071 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2072 struct relay_connection
*conn
,
2073 const struct lttng_buffer_view
*payload
)
2077 struct lttcomm_relayd_generic_reply reply
;
2081 DBG("Relay receiving streams_sent");
2083 if (!conn
->session
|| !conn
->version_check_done
) {
2084 ERR("Trying to close a stream before version check");
2086 goto end_no_session
;
2090 * Publish every pending stream in the connection recv list which are
2091 * now ready to be used by the viewer.
2093 publish_connection_local_streams(conn
);
2095 memset(&reply
, 0, sizeof(reply
));
2096 reply
.ret_code
= htobe32(LTTNG_OK
);
2097 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2098 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2099 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2112 * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2113 * session rotation feature (not the tracefile rotation feature).
2115 static int relay_rotate_session_streams(
2116 const struct lttcomm_relayd_hdr
*recv_hdr
,
2117 struct relay_connection
*conn
,
2118 const struct lttng_buffer_view
*payload
)
2123 enum lttng_error_code reply_code
= LTTNG_ERR_UNK
;
2124 struct relay_session
*session
= conn
->session
;
2125 struct lttcomm_relayd_rotate_streams rotate_streams
;
2126 struct lttcomm_relayd_generic_reply reply
= {};
2127 struct relay_stream
*stream
= NULL
;
2128 const size_t header_len
= sizeof(struct lttcomm_relayd_rotate_streams
);
2129 struct lttng_trace_chunk
*next_trace_chunk
= NULL
;
2130 struct lttng_buffer_view stream_positions
;
2131 char chunk_id_buf
[MAX_INT_DEC_LEN(uint64_t)];
2132 const char *chunk_id_str
= "none";
2134 if (!session
|| !conn
->version_check_done
) {
2135 ERR("Trying to rotate a stream before version check");
2140 if (session
->major
== 2 && session
->minor
< 11) {
2141 ERR("Unsupported feature before 2.11");
2146 if (payload
->size
< header_len
) {
2147 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2148 header_len
, payload
->size
);
2153 memcpy(&rotate_streams
, payload
->data
, header_len
);
2155 /* Convert header to host endianness. */
2156 rotate_streams
= (typeof(rotate_streams
)) {
2157 .stream_count
= be32toh(rotate_streams
.stream_count
),
2158 .new_chunk_id
= (typeof(rotate_streams
.new_chunk_id
)) {
2159 .is_set
= !!rotate_streams
.new_chunk_id
.is_set
,
2160 .value
= be64toh(rotate_streams
.new_chunk_id
.value
),
2164 if (rotate_streams
.new_chunk_id
.is_set
) {
2166 * Retrieve the trace chunk the stream must transition to. As
2167 * per the protocol, this chunk should have been created
2168 * before this command is received.
2170 next_trace_chunk
= sessiond_trace_chunk_registry_get_chunk(
2171 sessiond_trace_chunk_registry
,
2172 session
->sessiond_uuid
, session
->id
,
2173 rotate_streams
.new_chunk_id
.value
);
2174 if (!next_trace_chunk
) {
2175 char uuid_str
[UUID_STR_LEN
];
2177 lttng_uuid_to_str(session
->sessiond_uuid
, uuid_str
);
2178 ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2179 ", trace_chunk_id = %" PRIu64
,
2180 uuid_str
, session
->id
,
2181 rotate_streams
.new_chunk_id
.value
);
2182 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2187 ret
= snprintf(chunk_id_buf
, sizeof(chunk_id_buf
), "%" PRIu64
,
2188 rotate_streams
.new_chunk_id
.value
);
2189 if (ret
< 0 || ret
>= sizeof(chunk_id_buf
)) {
2190 chunk_id_str
= "formatting error";
2192 chunk_id_str
= chunk_id_buf
;
2196 DBG("Rotate %" PRIu32
" streams of session \"%s\" to chunk \"%s\"",
2197 rotate_streams
.stream_count
, session
->session_name
,
2200 stream_positions
= lttng_buffer_view_from_view(payload
,
2201 sizeof(rotate_streams
), -1);
2202 if (!stream_positions
.data
||
2203 stream_positions
.size
<
2204 (rotate_streams
.stream_count
*
2205 sizeof(struct lttcomm_relayd_stream_rotation_position
))) {
2206 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2211 for (i
= 0; i
< rotate_streams
.stream_count
; i
++) {
2212 struct lttcomm_relayd_stream_rotation_position
*position_comm
=
2213 &((typeof(position_comm
)) stream_positions
.data
)[i
];
2214 const struct lttcomm_relayd_stream_rotation_position pos
= {
2215 .stream_id
= be64toh(position_comm
->stream_id
),
2216 .rotate_at_seq_num
= be64toh(
2217 position_comm
->rotate_at_seq_num
),
2220 stream
= stream_get_by_id(pos
.stream_id
);
2222 reply_code
= LTTNG_ERR_INVALID
;
2227 pthread_mutex_lock(&stream
->lock
);
2228 ret
= stream_set_pending_rotation(stream
, next_trace_chunk
,
2229 pos
.rotate_at_seq_num
);
2230 pthread_mutex_unlock(&stream
->lock
);
2232 reply_code
= LTTNG_ERR_FILE_CREATION_ERROR
;
2240 reply_code
= LTTNG_OK
;
2246 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2247 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2248 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2249 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2250 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2256 lttng_trace_chunk_put(next_trace_chunk
);
2260 static int init_session_output_directory_handle(struct relay_session
*session
,
2261 struct lttng_directory_handle
*handle
)
2264 /* hostname/session_name */
2265 char *session_directory
= NULL
;
2267 * base path + session_directory
2268 * e.g. /home/user/lttng-traces/hostname/session_name
2270 char *full_session_path
= NULL
;
2271 char creation_time_str
[16];
2272 struct tm
*timeinfo
;
2274 assert(session
->creation_time
.is_set
);
2275 timeinfo
= localtime(&session
->creation_time
.value
);
2280 strftime(creation_time_str
, sizeof(creation_time_str
), "%Y%m%d-%H%M%S",
2283 pthread_mutex_lock(&session
->lock
);
2284 ret
= asprintf(&session_directory
, "%s/%s-%s", session
->hostname
,
2285 session
->session_name
, creation_time_str
);
2286 pthread_mutex_unlock(&session
->lock
);
2288 PERROR("Failed to format session directory name");
2292 full_session_path
= create_output_path(session_directory
);
2293 if (!full_session_path
) {
2298 ret
= utils_mkdir_recursive(
2299 full_session_path
, S_IRWXU
| S_IRWXG
, -1, -1);
2301 ERR("Failed to create session output path \"%s\"",
2306 ret
= lttng_directory_handle_init(handle
, full_session_path
);
2311 free(session_directory
);
2312 free(full_session_path
);
2317 * relay_create_trace_chunk: create a new trace chunk
2319 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2320 struct relay_connection
*conn
,
2321 const struct lttng_buffer_view
*payload
)
2325 struct relay_session
*session
= conn
->session
;
2326 struct lttcomm_relayd_create_trace_chunk
*msg
;
2327 struct lttcomm_relayd_generic_reply reply
= {};
2328 struct lttng_buffer_view header_view
;
2329 struct lttng_buffer_view chunk_name_view
;
2330 struct lttng_trace_chunk
*chunk
= NULL
, *published_chunk
= NULL
;
2331 enum lttng_error_code reply_code
= LTTNG_OK
;
2332 enum lttng_trace_chunk_status chunk_status
;
2333 struct lttng_directory_handle session_output
;
2335 if (!session
|| !conn
->version_check_done
) {
2336 ERR("Trying to create a trace chunk before version check");
2341 if (session
->major
== 2 && session
->minor
< 11) {
2342 ERR("Chunk creation command is unsupported before 2.11");
2347 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2348 if (!header_view
.data
) {
2349 ERR("Failed to receive payload of chunk creation command");
2354 /* Convert to host endianness. */
2355 msg
= (typeof(msg
)) header_view
.data
;
2356 msg
->chunk_id
= be64toh(msg
->chunk_id
);
2357 msg
->creation_timestamp
= be64toh(msg
->creation_timestamp
);
2358 msg
->override_name_length
= be32toh(msg
->override_name_length
);
2360 chunk
= lttng_trace_chunk_create(
2361 msg
->chunk_id
, msg
->creation_timestamp
);
2363 ERR("Failed to create trace chunk in trace chunk creation command");
2365 reply_code
= LTTNG_ERR_NOMEM
;
2369 if (msg
->override_name_length
) {
2372 chunk_name_view
= lttng_buffer_view_from_view(payload
,
2374 msg
->override_name_length
);
2375 name
= chunk_name_view
.data
;
2376 if (!name
|| name
[msg
->override_name_length
- 1]) {
2377 ERR("Failed to receive payload of chunk creation command");
2379 reply_code
= LTTNG_ERR_INVALID
;
2383 chunk_status
= lttng_trace_chunk_override_name(
2384 chunk
, chunk_name_view
.data
);
2385 switch (chunk_status
) {
2386 case LTTNG_TRACE_CHUNK_STATUS_OK
:
2388 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT
:
2389 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2390 reply_code
= LTTNG_ERR_INVALID
;
2394 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2395 reply_code
= LTTNG_ERR_UNK
;
2401 ret
= init_session_output_directory_handle(
2402 conn
->session
, &session_output
);
2404 reply_code
= LTTNG_ERR_CREATE_DIR_FAIL
;
2408 chunk_status
= lttng_trace_chunk_set_credentials_current_user(chunk
);
2409 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2410 reply_code
= LTTNG_ERR_UNK
;
2415 chunk_status
= lttng_trace_chunk_set_as_owner(chunk
, &session_output
);
2416 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2417 reply_code
= LTTNG_ERR_UNK
;
2422 published_chunk
= sessiond_trace_chunk_registry_publish_chunk(
2423 sessiond_trace_chunk_registry
,
2424 conn
->session
->sessiond_uuid
,
2427 if (!published_chunk
) {
2428 char uuid_str
[UUID_STR_LEN
];
2430 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2431 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2436 reply_code
= LTTNG_ERR_NOMEM
;
2440 pthread_mutex_lock(&conn
->session
->lock
);
2441 if (conn
->session
->pending_closure_trace_chunk
) {
2443 * Invalid; this means a second create_trace_chunk command was
2444 * received before a close_trace_chunk.
2446 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2447 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2449 goto end_unlock_session
;
2451 conn
->session
->pending_closure_trace_chunk
=
2452 conn
->session
->current_trace_chunk
;
2453 conn
->session
->current_trace_chunk
= published_chunk
;
2454 published_chunk
= NULL
;
2456 pthread_mutex_unlock(&conn
->session
->lock
);
2458 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2459 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2461 sizeof(struct lttcomm_relayd_generic_reply
),
2463 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2464 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2469 lttng_trace_chunk_put(chunk
);
2470 lttng_trace_chunk_put(published_chunk
);
2471 lttng_directory_handle_fini(&session_output
);
2476 * relay_close_trace_chunk: close a trace chunk
2478 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2479 struct relay_connection
*conn
,
2480 const struct lttng_buffer_view
*payload
)
2484 struct relay_session
*session
= conn
->session
;
2485 struct lttcomm_relayd_close_trace_chunk
*msg
;
2486 struct lttcomm_relayd_generic_reply reply
= {};
2487 struct lttng_buffer_view header_view
;
2488 struct lttng_trace_chunk
*chunk
= NULL
;
2489 enum lttng_error_code reply_code
= LTTNG_OK
;
2490 enum lttng_trace_chunk_status chunk_status
;
2492 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type
) close_command
= {};
2493 time_t close_timestamp
;
2495 if (!session
|| !conn
->version_check_done
) {
2496 ERR("Trying to close a trace chunk before version check");
2501 if (session
->major
== 2 && session
->minor
< 11) {
2502 ERR("Chunk close command is unsupported before 2.11");
2507 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2508 if (!header_view
.data
) {
2509 ERR("Failed to receive payload of chunk close command");
2514 /* Convert to host endianness. */
2515 msg
= (typeof(msg
)) header_view
.data
;
2516 chunk_id
= be64toh(msg
->chunk_id
);
2517 close_timestamp
= (time_t) be64toh(msg
->close_timestamp
);
2518 close_command
= (typeof(close_command
)){
2519 .value
= be32toh(msg
->close_command
.value
),
2520 .is_set
= msg
->close_command
.is_set
,
2523 chunk
= sessiond_trace_chunk_registry_get_chunk(
2524 sessiond_trace_chunk_registry
,
2525 conn
->session
->sessiond_uuid
,
2529 char uuid_str
[UUID_STR_LEN
];
2531 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2532 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2537 reply_code
= LTTNG_ERR_NOMEM
;
2541 pthread_mutex_lock(&session
->lock
);
2542 if (session
->pending_closure_trace_chunk
&&
2543 session
->pending_closure_trace_chunk
!= chunk
) {
2544 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
2545 session
->session_name
);
2546 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2548 goto end_unlock_session
;
2551 chunk_status
= lttng_trace_chunk_set_close_timestamp(
2552 chunk
, close_timestamp
);
2553 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2554 ERR("Failed to set trace chunk close timestamp");
2556 reply_code
= LTTNG_ERR_UNK
;
2557 goto end_unlock_session
;
2560 if (close_command
.is_set
) {
2561 chunk_status
= lttng_trace_chunk_set_close_command(
2562 chunk
, close_command
.value
);
2563 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2565 reply_code
= LTTNG_ERR_INVALID
;
2566 goto end_unlock_session
;
2570 if (session
->current_trace_chunk
== chunk
) {
2572 * After a trace chunk close command, no new streams
2573 * referencing the chunk may be created. Hence, on the
2574 * event that no new trace chunk have been created for
2575 * the session, the reference to the current trace chunk
2576 * is released in order to allow it to be reclaimed when
2577 * the last stream releases its reference to it.
2579 lttng_trace_chunk_put(session
->current_trace_chunk
);
2580 session
->current_trace_chunk
= NULL
;
2582 lttng_trace_chunk_put(session
->pending_closure_trace_chunk
);
2583 session
->pending_closure_trace_chunk
= NULL
;
2585 pthread_mutex_unlock(&session
->lock
);
2588 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2589 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2591 sizeof(struct lttcomm_relayd_generic_reply
),
2593 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2594 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2599 lttng_trace_chunk_put(chunk
);
2604 * relay_trace_chunk_exists: check if a trace chunk exists
2606 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr
*recv_hdr
,
2607 struct relay_connection
*conn
,
2608 const struct lttng_buffer_view
*payload
)
2612 struct relay_session
*session
= conn
->session
;
2613 struct lttcomm_relayd_trace_chunk_exists
*msg
;
2614 struct lttcomm_relayd_trace_chunk_exists_reply reply
= {};
2615 struct lttng_buffer_view header_view
;
2616 struct lttng_trace_chunk
*chunk
= NULL
;
2619 if (!session
|| !conn
->version_check_done
) {
2620 ERR("Trying to close a trace chunk before version check");
2625 if (session
->major
== 2 && session
->minor
< 11) {
2626 ERR("Chunk close command is unsupported before 2.11");
2631 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2632 if (!header_view
.data
) {
2633 ERR("Failed to receive payload of chunk close command");
2638 /* Convert to host endianness. */
2639 msg
= (typeof(msg
)) header_view
.data
;
2640 chunk_id
= be64toh(msg
->chunk_id
);
2642 chunk
= sessiond_trace_chunk_registry_get_chunk(
2643 sessiond_trace_chunk_registry
,
2644 conn
->session
->sessiond_uuid
,
2648 reply
= (typeof(reply
)) {
2649 .generic
.ret_code
= htobe32((uint32_t) LTTNG_OK
),
2650 .trace_chunk_exists
= !!chunk
,
2652 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2653 &reply
, sizeof(reply
), 0);
2654 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2655 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2660 lttng_trace_chunk_put(chunk
);
2664 #define DBG_CMD(cmd_name, conn) \
2665 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2667 static int relay_process_control_command(struct relay_connection
*conn
,
2668 const struct lttcomm_relayd_hdr
*header
,
2669 const struct lttng_buffer_view
*payload
)
2673 switch (header
->cmd
) {
2674 case RELAYD_CREATE_SESSION
:
2675 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
2676 ret
= relay_create_session(header
, conn
, payload
);
2678 case RELAYD_ADD_STREAM
:
2679 DBG_CMD("RELAYD_ADD_STREAM", conn
);
2680 ret
= relay_add_stream(header
, conn
, payload
);
2682 case RELAYD_START_DATA
:
2683 DBG_CMD("RELAYD_START_DATA", conn
);
2684 ret
= relay_start(header
, conn
, payload
);
2686 case RELAYD_SEND_METADATA
:
2687 DBG_CMD("RELAYD_SEND_METADATA", conn
);
2688 ret
= relay_recv_metadata(header
, conn
, payload
);
2690 case RELAYD_VERSION
:
2691 DBG_CMD("RELAYD_VERSION", conn
);
2692 ret
= relay_send_version(header
, conn
, payload
);
2694 case RELAYD_CLOSE_STREAM
:
2695 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
2696 ret
= relay_close_stream(header
, conn
, payload
);
2698 case RELAYD_DATA_PENDING
:
2699 DBG_CMD("RELAYD_DATA_PENDING", conn
);
2700 ret
= relay_data_pending(header
, conn
, payload
);
2702 case RELAYD_QUIESCENT_CONTROL
:
2703 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
2704 ret
= relay_quiescent_control(header
, conn
, payload
);
2706 case RELAYD_BEGIN_DATA_PENDING
:
2707 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
2708 ret
= relay_begin_data_pending(header
, conn
, payload
);
2710 case RELAYD_END_DATA_PENDING
:
2711 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
2712 ret
= relay_end_data_pending(header
, conn
, payload
);
2714 case RELAYD_SEND_INDEX
:
2715 DBG_CMD("RELAYD_SEND_INDEX", conn
);
2716 ret
= relay_recv_index(header
, conn
, payload
);
2718 case RELAYD_STREAMS_SENT
:
2719 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
2720 ret
= relay_streams_sent(header
, conn
, payload
);
2722 case RELAYD_RESET_METADATA
:
2723 DBG_CMD("RELAYD_RESET_METADATA", conn
);
2724 ret
= relay_reset_metadata(header
, conn
, payload
);
2726 case RELAYD_ROTATE_STREAMS
:
2727 DBG_CMD("RELAYD_ROTATE_STREAMS", conn
);
2728 ret
= relay_rotate_session_streams(header
, conn
, payload
);
2730 case RELAYD_CREATE_TRACE_CHUNK
:
2731 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn
);
2732 ret
= relay_create_trace_chunk(header
, conn
, payload
);
2734 case RELAYD_CLOSE_TRACE_CHUNK
:
2735 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn
);
2736 ret
= relay_close_trace_chunk(header
, conn
, payload
);
2738 case RELAYD_TRACE_CHUNK_EXISTS
:
2739 DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn
);
2740 ret
= relay_trace_chunk_exists(header
, conn
, payload
);
2742 case RELAYD_UPDATE_SYNC_INFO
:
2744 ERR("Received unknown command (%u)", header
->cmd
);
2745 relay_unknown_command(conn
);
2754 static enum relay_connection_status
relay_process_control_receive_payload(
2755 struct relay_connection
*conn
)
2758 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2759 struct lttng_dynamic_buffer
*reception_buffer
=
2760 &conn
->protocol
.ctrl
.reception_buffer
;
2761 struct ctrl_connection_state_receive_payload
*state
=
2762 &conn
->protocol
.ctrl
.state
.receive_payload
;
2763 struct lttng_buffer_view payload_view
;
2765 if (state
->left_to_receive
== 0) {
2766 /* Short-circuit for payload-less commands. */
2767 goto reception_complete
;
2770 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2771 reception_buffer
->data
+ state
->received
,
2772 state
->left_to_receive
, MSG_DONTWAIT
);
2774 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2775 PERROR("Unable to receive command payload on sock %d",
2777 status
= RELAY_CONNECTION_STATUS_ERROR
;
2780 } else if (ret
== 0) {
2781 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2782 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2787 assert(ret
<= state
->left_to_receive
);
2789 state
->left_to_receive
-= ret
;
2790 state
->received
+= ret
;
2792 if (state
->left_to_receive
> 0) {
2794 * Can't transition to the protocol's next state, wait to
2795 * receive the rest of the header.
2797 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2798 state
->received
, state
->left_to_receive
,
2804 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
2805 conn
->sock
->fd
, state
->received
);
2807 * The payload required to process the command has been received.
2808 * A view to the reception buffer is forwarded to the various
2809 * commands and the state of the control is reset on success.
2811 * Commands are responsible for sending their reply to the peer.
2813 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
2815 ret
= relay_process_control_command(conn
,
2816 &state
->header
, &payload_view
);
2818 status
= RELAY_CONNECTION_STATUS_ERROR
;
2822 ret
= connection_reset_protocol_state(conn
);
2824 status
= RELAY_CONNECTION_STATUS_ERROR
;
2830 static enum relay_connection_status
relay_process_control_receive_header(
2831 struct relay_connection
*conn
)
2834 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2835 struct lttcomm_relayd_hdr header
;
2836 struct lttng_dynamic_buffer
*reception_buffer
=
2837 &conn
->protocol
.ctrl
.reception_buffer
;
2838 struct ctrl_connection_state_receive_header
*state
=
2839 &conn
->protocol
.ctrl
.state
.receive_header
;
2841 assert(state
->left_to_receive
!= 0);
2843 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2844 reception_buffer
->data
+ state
->received
,
2845 state
->left_to_receive
, MSG_DONTWAIT
);
2847 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2848 PERROR("Unable to receive control command header on sock %d",
2850 status
= RELAY_CONNECTION_STATUS_ERROR
;
2853 } else if (ret
== 0) {
2854 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2855 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2860 assert(ret
<= state
->left_to_receive
);
2862 state
->left_to_receive
-= ret
;
2863 state
->received
+= ret
;
2865 if (state
->left_to_receive
> 0) {
2867 * Can't transition to the protocol's next state, wait to
2868 * receive the rest of the header.
2870 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2871 state
->received
, state
->left_to_receive
,
2876 /* Transition to next state: receiving the command's payload. */
2877 conn
->protocol
.ctrl
.state_id
=
2878 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
2879 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
2880 header
.circuit_id
= be64toh(header
.circuit_id
);
2881 header
.data_size
= be64toh(header
.data_size
);
2882 header
.cmd
= be32toh(header
.cmd
);
2883 header
.cmd_version
= be32toh(header
.cmd_version
);
2884 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
2885 &header
, sizeof(header
));
2887 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
2888 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
2891 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
2892 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
2894 status
= RELAY_CONNECTION_STATUS_ERROR
;
2898 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
2900 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
2901 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
2904 status
= RELAY_CONNECTION_STATUS_ERROR
;
2908 if (header
.data_size
== 0) {
2910 * Manually invoke the next state as the poll loop
2911 * will not wake-up to allow us to proceed further.
2913 status
= relay_process_control_receive_payload(conn
);
2920 * Process the commands received on the control socket
2922 static enum relay_connection_status
relay_process_control(
2923 struct relay_connection
*conn
)
2925 enum relay_connection_status status
;
2927 switch (conn
->protocol
.ctrl
.state_id
) {
2928 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
2929 status
= relay_process_control_receive_header(conn
);
2931 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
2932 status
= relay_process_control_receive_payload(conn
);
2935 ERR("Unknown control connection protocol state encountered.");
2942 static enum relay_connection_status
relay_process_data_receive_header(
2943 struct relay_connection
*conn
)
2946 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2947 struct data_connection_state_receive_header
*state
=
2948 &conn
->protocol
.data
.state
.receive_header
;
2949 struct lttcomm_relayd_data_hdr header
;
2950 struct relay_stream
*stream
;
2952 assert(state
->left_to_receive
!= 0);
2954 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2955 state
->header_reception_buffer
+ state
->received
,
2956 state
->left_to_receive
, MSG_DONTWAIT
);
2958 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2959 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2960 status
= RELAY_CONNECTION_STATUS_ERROR
;
2963 } else if (ret
== 0) {
2964 /* Orderly shutdown. Not necessary to print an error. */
2965 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2966 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2971 assert(ret
<= state
->left_to_receive
);
2973 state
->left_to_receive
-= ret
;
2974 state
->received
+= ret
;
2976 if (state
->left_to_receive
> 0) {
2978 * Can't transition to the protocol's next state, wait to
2979 * receive the rest of the header.
2981 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2982 state
->received
, state
->left_to_receive
,
2987 /* Transition to next state: receiving the payload. */
2988 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
2990 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
2991 header
.circuit_id
= be64toh(header
.circuit_id
);
2992 header
.stream_id
= be64toh(header
.stream_id
);
2993 header
.data_size
= be32toh(header
.data_size
);
2994 header
.net_seq_num
= be64toh(header
.net_seq_num
);
2995 header
.padding_size
= be32toh(header
.padding_size
);
2996 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
2998 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3000 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3001 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3003 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3004 conn
->sock
->fd
, header
.circuit_id
,
3005 header
.stream_id
, header
.data_size
,
3006 header
.net_seq_num
, header
.padding_size
);
3008 stream
= stream_get_by_id(header
.stream_id
);
3010 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3012 /* Protocol error. */
3013 status
= RELAY_CONNECTION_STATUS_ERROR
;
3017 pthread_mutex_lock(&stream
->lock
);
3018 /* Prepare stream for the reception of a new packet. */
3019 ret
= stream_init_packet(stream
, header
.data_size
,
3020 &conn
->protocol
.data
.state
.receive_payload
.rotate_index
);
3021 pthread_mutex_unlock(&stream
->lock
);
3023 ERR("Failed to rotate stream output file");
3024 status
= RELAY_CONNECTION_STATUS_ERROR
;
3025 goto end_stream_unlock
;
3034 static enum relay_connection_status
relay_process_data_receive_payload(
3035 struct relay_connection
*conn
)
3038 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3039 struct relay_stream
*stream
;
3040 struct data_connection_state_receive_payload
*state
=
3041 &conn
->protocol
.data
.state
.receive_payload
;
3042 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3043 char data_buffer
[chunk_size
];
3044 bool partial_recv
= false;
3045 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3046 uint64_t left_to_receive
= state
->left_to_receive
;
3047 struct relay_session
*session
;
3049 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3050 state
->header
.stream_id
, state
->header
.net_seq_num
,
3051 state
->received
, left_to_receive
);
3053 stream
= stream_get_by_id(state
->header
.stream_id
);
3055 /* Protocol error. */
3056 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3057 state
->header
.stream_id
);
3058 status
= RELAY_CONNECTION_STATUS_ERROR
;
3062 pthread_mutex_lock(&stream
->lock
);
3063 session
= stream
->trace
->session
;
3064 if (!conn
->session
) {
3065 ret
= connection_set_session(conn
, session
);
3067 status
= RELAY_CONNECTION_STATUS_ERROR
;
3068 goto end_stream_unlock
;
3073 * The size of the "chunk" received on any iteration is bounded by:
3074 * - the data left to receive,
3075 * - the data immediately available on the socket,
3076 * - the on-stack data buffer
3078 while (left_to_receive
> 0 && !partial_recv
) {
3079 size_t recv_size
= min(left_to_receive
, chunk_size
);
3080 struct lttng_buffer_view packet_chunk
;
3082 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3083 recv_size
, MSG_DONTWAIT
);
3085 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3086 PERROR("Socket %d error", conn
->sock
->fd
);
3087 status
= RELAY_CONNECTION_STATUS_ERROR
;
3089 goto end_stream_unlock
;
3090 } else if (ret
== 0) {
3091 /* No more data ready to be consumed on socket. */
3092 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3093 state
->header
.stream_id
);
3094 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3096 } else if (ret
< (int) recv_size
) {
3098 * All the data available on the socket has been
3101 partial_recv
= true;
3105 packet_chunk
= lttng_buffer_view_init(data_buffer
,
3107 assert(packet_chunk
.data
);
3109 ret
= stream_write(stream
, &packet_chunk
, 0);
3111 ERR("Relay error writing data to file");
3112 status
= RELAY_CONNECTION_STATUS_ERROR
;
3113 goto end_stream_unlock
;
3116 left_to_receive
-= recv_size
;
3117 state
->received
+= recv_size
;
3118 state
->left_to_receive
= left_to_receive
;
3121 if (state
->left_to_receive
> 0) {
3123 * Did not receive all the data expected, wait for more data to
3124 * become available on the socket.
3126 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3127 state
->header
.stream_id
, state
->received
,
3128 state
->left_to_receive
);
3129 goto end_stream_unlock
;
3132 ret
= stream_write(stream
, NULL
, state
->header
.padding_size
);
3134 status
= RELAY_CONNECTION_STATUS_ERROR
;
3135 goto end_stream_unlock
;
3138 if (session_streams_have_index(session
)) {
3139 ret
= stream_update_index(stream
, state
->header
.net_seq_num
,
3140 state
->rotate_index
, &index_flushed
,
3141 state
->header
.data_size
+ state
->header
.padding_size
);
3143 ERR("Failed to update index: stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3144 stream
->stream_handle
,
3145 state
->header
.net_seq_num
, ret
);
3146 status
= RELAY_CONNECTION_STATUS_ERROR
;
3147 goto end_stream_unlock
;
3151 if (stream
->prev_data_seq
== -1ULL) {
3155 ret
= stream_complete_packet(stream
, state
->header
.data_size
+
3156 state
->header
.padding_size
, state
->header
.net_seq_num
,
3159 status
= RELAY_CONNECTION_STATUS_ERROR
;
3160 goto end_stream_unlock
;
3164 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3165 * contents of *state which are aliased (union) to the same location as
3166 * the new state. Don't use it beyond this point.
3168 connection_reset_protocol_state(conn
);
3172 close_requested
= stream
->close_requested
;
3173 pthread_mutex_unlock(&stream
->lock
);
3174 if (close_requested
&& left_to_receive
== 0) {
3175 try_stream_close(stream
);
3179 pthread_mutex_lock(&session
->lock
);
3180 uatomic_set(&session
->new_streams
, 1);
3181 pthread_mutex_unlock(&session
->lock
);
3190 * relay_process_data: Process the data received on the data socket
3192 static enum relay_connection_status
relay_process_data(
3193 struct relay_connection
*conn
)
3195 enum relay_connection_status status
;
3197 switch (conn
->protocol
.data
.state_id
) {
3198 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3199 status
= relay_process_data_receive_header(conn
);
3201 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3202 status
= relay_process_data_receive_payload(conn
);
3205 ERR("Unexpected data connection communication state.");
3212 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3216 (void) lttng_poll_del(events
, pollfd
);
3218 ret
= close(pollfd
);
3220 ERR("Closing pollfd %d", pollfd
);
3224 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3225 int pollfd
, struct relay_connection
*conn
)
3227 const char *type_str
;
3229 switch (conn
->type
) {
3234 type_str
= "Control";
3236 case RELAY_VIEWER_COMMAND
:
3237 type_str
= "Viewer Command";
3239 case RELAY_VIEWER_NOTIFICATION
:
3240 type_str
= "Viewer Notification";
3243 type_str
= "Unknown";
3245 cleanup_connection_pollfd(events
, pollfd
);
3246 connection_put(conn
);
3247 DBG("%s connection closed with %d", type_str
, pollfd
);
3251 * This thread does the actual work
3253 static void *relay_thread_worker(void *data
)
3255 int ret
, err
= -1, last_seen_data_fd
= -1;
3257 struct lttng_poll_event events
;
3258 struct lttng_ht
*relay_connections_ht
;
3259 struct lttng_ht_iter iter
;
3260 struct relay_connection
*destroy_conn
= NULL
;
3262 DBG("[thread] Relay worker started");
3264 rcu_register_thread();
3266 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3268 if (testpoint(relayd_thread_worker
)) {
3269 goto error_testpoint
;
3272 health_code_update();
3274 /* table of connections indexed on socket */
3275 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3276 if (!relay_connections_ht
) {
3277 goto relay_connections_ht_error
;
3280 ret
= create_thread_poll_set(&events
, 2);
3282 goto error_poll_create
;
3285 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3292 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3294 health_code_update();
3296 /* Infinite blocking call, waiting for transmission */
3297 DBG3("Relayd worker thread polling...");
3298 health_poll_entry();
3299 ret
= lttng_poll_wait(&events
, -1);
3303 * Restart interrupted system call.
3305 if (errno
== EINTR
) {
3314 * Process control. The control connection is
3315 * prioritized so we don't starve it with high
3316 * throughput tracing data on the data connection.
3318 for (i
= 0; i
< nb_fd
; i
++) {
3319 /* Fetch once the poll data */
3320 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3321 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3323 health_code_update();
3325 /* Thread quit pipe has been closed. Killing thread. */
3326 ret
= check_thread_quit_pipe(pollfd
, revents
);
3332 /* Inspect the relay conn pipe for new connection */
3333 if (pollfd
== relay_conn_pipe
[0]) {
3334 if (revents
& LPOLLIN
) {
3335 struct relay_connection
*conn
;
3337 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3341 lttng_poll_add(&events
, conn
->sock
->fd
,
3342 LPOLLIN
| LPOLLRDHUP
);
3343 connection_ht_add(relay_connections_ht
, conn
);
3344 DBG("Connection socket %d added", conn
->sock
->fd
);
3345 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3346 ERR("Relay connection pipe error");
3349 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3353 struct relay_connection
*ctrl_conn
;
3355 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3356 /* If not found, there is a synchronization issue. */
3359 if (ctrl_conn
->type
== RELAY_DATA
) {
3360 if (revents
& LPOLLIN
) {
3362 * Flag the last seen data fd not deleted. It will be
3363 * used as the last seen fd if any fd gets deleted in
3366 last_notdel_data_fd
= pollfd
;
3368 goto put_ctrl_connection
;
3370 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3372 if (revents
& LPOLLIN
) {
3373 enum relay_connection_status status
;
3375 status
= relay_process_control(ctrl_conn
);
3376 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3378 * On socket error flag the session as aborted to force
3379 * the cleanup of its stream otherwise it can leak
3380 * during the lifetime of the relayd.
3382 * This prevents situations in which streams can be
3383 * left opened because an index was received, the
3384 * control connection is closed, and the data
3385 * connection is closed (uncleanly) before the packet's
3388 * Since the control connection encountered an error,
3389 * it is okay to be conservative and close the
3390 * session right now as we can't rely on the protocol
3391 * being respected anymore.
3393 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3394 session_abort(ctrl_conn
->session
);
3397 /* Clear the connection on error or close. */
3398 relay_thread_close_connection(&events
,
3403 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3404 relay_thread_close_connection(&events
,
3406 if (last_seen_data_fd
== pollfd
) {
3407 last_seen_data_fd
= last_notdel_data_fd
;
3410 ERR("Unexpected poll events %u for control sock %d",
3412 connection_put(ctrl_conn
);
3415 put_ctrl_connection
:
3416 connection_put(ctrl_conn
);
3421 * The last loop handled a control request, go back to poll to make
3422 * sure we prioritise the control socket.
3428 if (last_seen_data_fd
>= 0) {
3429 for (i
= 0; i
< nb_fd
; i
++) {
3430 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3432 health_code_update();
3434 if (last_seen_data_fd
== pollfd
) {
3441 /* Process data connection. */
3442 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3443 /* Fetch the poll data. */
3444 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3445 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3446 struct relay_connection
*data_conn
;
3448 health_code_update();
3451 /* No activity for this FD (poll implementation). */
3455 /* Skip the command pipe. It's handled in the first loop. */
3456 if (pollfd
== relay_conn_pipe
[0]) {
3460 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3462 /* Skip it. Might be removed before. */
3465 if (data_conn
->type
== RELAY_CONTROL
) {
3466 goto put_data_connection
;
3468 assert(data_conn
->type
== RELAY_DATA
);
3470 if (revents
& LPOLLIN
) {
3471 enum relay_connection_status status
;
3473 status
= relay_process_data(data_conn
);
3474 /* Connection closed or error. */
3475 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3477 * On socket error flag the session as aborted to force
3478 * the cleanup of its stream otherwise it can leak
3479 * during the lifetime of the relayd.
3481 * This prevents situations in which streams can be
3482 * left opened because an index was received, the
3483 * control connection is closed, and the data
3484 * connection is closed (uncleanly) before the packet's
3487 * Since the data connection encountered an error,
3488 * it is okay to be conservative and close the
3489 * session right now as we can't rely on the protocol
3490 * being respected anymore.
3492 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3493 session_abort(data_conn
->session
);
3495 relay_thread_close_connection(&events
, pollfd
,
3498 * Every goto restart call sets the last seen fd where
3499 * here we don't really care since we gracefully
3500 * continue the loop after the connection is deleted.
3503 /* Keep last seen port. */
3504 last_seen_data_fd
= pollfd
;
3505 connection_put(data_conn
);
3508 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3509 relay_thread_close_connection(&events
, pollfd
,
3512 ERR("Unknown poll events %u for data sock %d",
3515 put_data_connection
:
3516 connection_put(data_conn
);
3518 last_seen_data_fd
= -1;
3521 /* Normal exit, no error */
3526 /* Cleanup remaining connection object. */
3528 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3531 health_code_update();
3533 session_abort(destroy_conn
->session
);
3536 * No need to grab another ref, because we own
3539 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
3544 lttng_poll_clean(&events
);
3546 lttng_ht_destroy(relay_connections_ht
);
3547 relay_connections_ht_error
:
3548 /* Close relay conn pipes */
3549 utils_close_pipe(relay_conn_pipe
);
3551 DBG("Thread exited with error");
3553 DBG("Worker thread cleanup complete");
3557 ERR("Health error occurred in %s", __func__
);
3559 health_unregister(health_relayd
);
3560 rcu_unregister_thread();
3561 lttng_relay_stop_threads();
3566 * Create the relay command pipe to wake thread_manage_apps.
3567 * Closed in cleanup().
3569 static int create_relay_conn_pipe(void)
3573 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
3581 int main(int argc
, char **argv
)
3583 int ret
= 0, retval
= 0;
3586 /* Parse arguments */
3588 if (set_options(argc
, argv
)) {
3593 if (set_signal_handler()) {
3598 /* Try to create directory if -o, --output is specified. */
3599 if (opt_output_path
) {
3600 if (*opt_output_path
!= '/') {
3601 ERR("Please specify an absolute path for -o, --output PATH");
3606 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
3609 ERR("Unable to create %s", opt_output_path
);
3616 if (opt_daemon
|| opt_background
) {
3619 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
3627 * We are in the child. Make sure all other file
3628 * descriptors are closed, in case we are called with
3629 * more opened file descriptors than the standard ones.
3631 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3636 sessiond_trace_chunk_registry
= sessiond_trace_chunk_registry_create();
3637 if (!sessiond_trace_chunk_registry
) {
3638 ERR("Failed to initialize session daemon trace chunk registry");
3640 goto exit_sessiond_trace_chunk_registry
;
3643 /* Initialize thread health monitoring */
3644 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
3645 if (!health_relayd
) {
3646 PERROR("health_app_create error");
3648 goto exit_health_app_create
;
3651 /* Create thread quit pipe */
3652 if (init_thread_quit_pipe()) {
3654 goto exit_init_data
;
3657 /* Setup the thread apps communication pipe. */
3658 if (create_relay_conn_pipe()) {
3660 goto exit_init_data
;
3663 /* Init relay command queue. */
3664 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
3666 /* Initialize communication library */
3668 lttcomm_inet_init();
3670 /* tables of sessions indexed by session ID */
3671 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3674 goto exit_init_data
;
3677 /* tables of streams indexed by stream ID */
3678 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3679 if (!relay_streams_ht
) {
3681 goto exit_init_data
;
3684 /* tables of streams indexed by stream ID */
3685 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3686 if (!viewer_streams_ht
) {
3688 goto exit_init_data
;
3691 ret
= utils_create_pipe(health_quit_pipe
);
3694 goto exit_health_quit_pipe
;
3697 /* Create thread to manage the client socket */
3698 ret
= pthread_create(&health_thread
, default_pthread_attr(),
3699 thread_manage_health
, (void *) NULL
);
3702 PERROR("pthread_create health");
3704 goto exit_health_thread
;
3707 /* Setup the dispatcher thread */
3708 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
3709 relay_thread_dispatcher
, (void *) NULL
);
3712 PERROR("pthread_create dispatcher");
3714 goto exit_dispatcher_thread
;
3717 /* Setup the worker thread */
3718 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
3719 relay_thread_worker
, NULL
);
3722 PERROR("pthread_create worker");
3724 goto exit_worker_thread
;
3727 /* Setup the listener thread */
3728 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
3729 relay_thread_listener
, (void *) NULL
);
3732 PERROR("pthread_create listener");
3734 goto exit_listener_thread
;
3737 ret
= relayd_live_create(live_uri
);
3739 ERR("Starting live viewer threads");
3745 * This is where we start awaiting program completion (e.g. through
3746 * signal that asks threads to teardown).
3749 ret
= relayd_live_join();
3755 ret
= pthread_join(listener_thread
, &status
);
3758 PERROR("pthread_join listener_thread");
3762 exit_listener_thread
:
3763 ret
= pthread_join(worker_thread
, &status
);
3766 PERROR("pthread_join worker_thread");
3771 ret
= pthread_join(dispatcher_thread
, &status
);
3774 PERROR("pthread_join dispatcher_thread");
3777 exit_dispatcher_thread
:
3779 ret
= pthread_join(health_thread
, &status
);
3782 PERROR("pthread_join health_thread");
3787 utils_close_pipe(health_quit_pipe
);
3788 exit_health_quit_pipe
:
3791 health_app_destroy(health_relayd
);
3792 sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry
);
3793 exit_health_app_create
:
3794 exit_sessiond_trace_chunk_registry
:
3797 * Wait for all pending call_rcu work to complete before tearing
3798 * down data structures. call_rcu worker may be trying to
3799 * perform lookups in those structures.
3804 /* Ensure all prior call_rcu are done. */