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 bool session_name_contains_creation_timestamp
= false;
1077 /* Left nil for peers < 2.11. */
1078 char base_path
[LTTNG_PATH_MAX
] = {};
1079 lttng_uuid sessiond_uuid
= {};
1080 LTTNG_OPTIONAL(uint64_t) id_sessiond
= {};
1081 LTTNG_OPTIONAL(uint64_t) current_chunk_id
= {};
1082 LTTNG_OPTIONAL(time_t) creation_time
= {};
1084 if (conn
->minor
< 4) {
1085 /* From 2.1 to 2.3 */
1087 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1088 /* From 2.4 to 2.10 */
1089 ret
= cmd_create_session_2_4(payload
, session_name
,
1090 hostname
, &live_timer
, &snapshot
);
1092 bool has_current_chunk
;
1093 uint64_t current_chunk_id_value
;
1094 time_t creation_time_value
;
1095 uint64_t id_sessiond_value
;
1097 /* From 2.11 to ... */
1098 ret
= cmd_create_session_2_11(payload
, session_name
, hostname
,
1099 base_path
, &live_timer
, &snapshot
, &id_sessiond_value
,
1100 sessiond_uuid
, &has_current_chunk
,
1101 ¤t_chunk_id_value
, &creation_time_value
,
1102 &session_name_contains_creation_timestamp
);
1103 if (lttng_uuid_is_nil(sessiond_uuid
)) {
1104 /* The nil UUID is reserved for pre-2.11 clients. */
1105 ERR("Illegal nil UUID announced by peer in create session command");
1109 LTTNG_OPTIONAL_SET(&id_sessiond
, id_sessiond_value
);
1110 LTTNG_OPTIONAL_SET(&creation_time
, creation_time_value
);
1111 if (has_current_chunk
) {
1112 LTTNG_OPTIONAL_SET(¤t_chunk_id
,
1113 current_chunk_id_value
);
1121 session
= session_create(session_name
, hostname
, base_path
, live_timer
,
1122 snapshot
, sessiond_uuid
,
1123 id_sessiond
.is_set
? &id_sessiond
.value
: NULL
,
1124 current_chunk_id
.is_set
? ¤t_chunk_id
.value
: NULL
,
1125 creation_time
.is_set
? &creation_time
.value
: NULL
,
1126 conn
->major
, conn
->minor
,
1127 session_name_contains_creation_timestamp
);
1132 assert(!conn
->session
);
1133 conn
->session
= session
;
1134 DBG("Created session %" PRIu64
, session
->id
);
1136 reply
.session_id
= htobe64(session
->id
);
1140 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1142 reply
.ret_code
= htobe32(LTTNG_OK
);
1145 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1146 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1147 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1151 if (ret
< 0 && session
) {
1152 session_put(session
);
1158 * When we have received all the streams and the metadata for a channel,
1159 * we make them visible to the viewer threads.
1161 static void publish_connection_local_streams(struct relay_connection
*conn
)
1163 struct relay_stream
*stream
;
1164 struct relay_session
*session
= conn
->session
;
1167 * We publish all streams belonging to a session atomically wrt
1170 pthread_mutex_lock(&session
->lock
);
1172 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1174 stream_publish(stream
);
1179 * Inform the viewer that there are new streams in the session.
1181 if (session
->viewer_attached
) {
1182 uatomic_set(&session
->new_streams
, 1);
1184 pthread_mutex_unlock(&session
->lock
);
1187 static int conform_channel_path(char *channel_path
)
1191 if (strstr("../", channel_path
)) {
1192 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1198 if (*channel_path
== '/') {
1199 const size_t len
= strlen(channel_path
);
1202 * Channel paths from peers prior to 2.11 are expressed as an
1203 * absolute path that is, in reality, relative to the relay
1204 * daemon's output directory. Remove the leading slash so it
1205 * is correctly interpreted as a relative path later on.
1207 * len (and not len - 1) is used to copy the trailing NULL.
1209 bcopy(channel_path
+ 1, channel_path
, len
);
1216 * relay_add_stream: allocate a new stream for a session
1218 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1219 struct relay_connection
*conn
,
1220 const struct lttng_buffer_view
*payload
)
1224 struct relay_session
*session
= conn
->session
;
1225 struct relay_stream
*stream
= NULL
;
1226 struct lttcomm_relayd_status_stream reply
;
1227 struct ctf_trace
*trace
= NULL
;
1228 uint64_t stream_handle
= -1ULL;
1229 char *path_name
= NULL
, *channel_name
= NULL
;
1230 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1231 LTTNG_OPTIONAL(uint64_t) stream_chunk_id
= {};
1233 if (!session
|| !conn
->version_check_done
) {
1234 ERR("Trying to add a stream before version check");
1236 goto end_no_session
;
1239 if (session
->minor
== 1) {
1241 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1243 } else if (session
->minor
> 1 && session
->minor
< 11) {
1244 /* From 2.2 to 2.10 */
1245 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1246 &channel_name
, &tracefile_size
, &tracefile_count
);
1248 /* From 2.11 to ... */
1249 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1250 &channel_name
, &tracefile_size
, &tracefile_count
,
1251 &stream_chunk_id
.value
);
1252 stream_chunk_id
.is_set
= true;
1259 if (conform_channel_path(path_name
)) {
1263 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1267 /* This stream here has one reference on the trace. */
1269 pthread_mutex_lock(&last_relay_stream_id_lock
);
1270 stream_handle
= ++last_relay_stream_id
;
1271 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1273 /* We pass ownership of path_name and channel_name. */
1274 stream
= stream_create(trace
, stream_handle
, path_name
,
1275 channel_name
, tracefile_size
, tracefile_count
);
1277 channel_name
= NULL
;
1280 * Streams are the owners of their trace. Reference to trace is
1281 * kept within stream_create().
1283 ctf_trace_put(trace
);
1286 memset(&reply
, 0, sizeof(reply
));
1287 reply
.handle
= htobe64(stream_handle
);
1289 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1291 reply
.ret_code
= htobe32(LTTNG_OK
);
1294 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1295 sizeof(struct lttcomm_relayd_status_stream
), 0);
1296 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1297 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1309 * relay_close_stream: close a specific stream
1311 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1312 struct relay_connection
*conn
,
1313 const struct lttng_buffer_view
*payload
)
1317 struct relay_session
*session
= conn
->session
;
1318 struct lttcomm_relayd_close_stream stream_info
;
1319 struct lttcomm_relayd_generic_reply reply
;
1320 struct relay_stream
*stream
;
1322 DBG("Close stream received");
1324 if (!session
|| !conn
->version_check_done
) {
1325 ERR("Trying to close a stream before version check");
1327 goto end_no_session
;
1330 if (payload
->size
< sizeof(stream_info
)) {
1331 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1332 sizeof(stream_info
), payload
->size
);
1334 goto end_no_session
;
1336 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1337 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1338 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1340 stream
= stream_get_by_id(stream_info
.stream_id
);
1347 * Set last_net_seq_num before the close flag. Required by data
1350 pthread_mutex_lock(&stream
->lock
);
1351 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1352 pthread_mutex_unlock(&stream
->lock
);
1355 * This is one of the conditions which may trigger a stream close
1356 * with the others being:
1357 * 1) A close command is received for a stream
1358 * 2) The control connection owning the stream is closed
1359 * 3) We have received all of the stream's data _after_ a close
1362 try_stream_close(stream
);
1363 if (stream
->is_metadata
) {
1364 struct relay_viewer_stream
*vstream
;
1366 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1368 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1370 * Since all the metadata has been sent to the
1371 * viewer and that we have a request to close
1372 * its stream, we can safely teardown the
1373 * corresponding metadata viewer stream.
1375 viewer_stream_put(vstream
);
1377 /* Put local reference. */
1378 viewer_stream_put(vstream
);
1385 memset(&reply
, 0, sizeof(reply
));
1387 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1389 reply
.ret_code
= htobe32(LTTNG_OK
);
1391 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1392 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1393 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1394 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1404 * relay_reset_metadata: reset a metadata stream
1407 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1408 struct relay_connection
*conn
,
1409 const struct lttng_buffer_view
*payload
)
1413 struct relay_session
*session
= conn
->session
;
1414 struct lttcomm_relayd_reset_metadata stream_info
;
1415 struct lttcomm_relayd_generic_reply reply
;
1416 struct relay_stream
*stream
;
1418 DBG("Reset metadata received");
1420 if (!session
|| !conn
->version_check_done
) {
1421 ERR("Trying to reset a metadata stream before version check");
1423 goto end_no_session
;
1426 if (payload
->size
< sizeof(stream_info
)) {
1427 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1428 sizeof(stream_info
), payload
->size
);
1430 goto end_no_session
;
1432 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1433 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1434 stream_info
.version
= be64toh(stream_info
.version
);
1436 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1438 /* Unsupported for live sessions for now. */
1439 if (session
->live_timer
!= 0) {
1444 stream
= stream_get_by_id(stream_info
.stream_id
);
1449 pthread_mutex_lock(&stream
->lock
);
1450 if (!stream
->is_metadata
) {
1455 ret
= stream_reset_file(stream
);
1457 ERR("Failed to reset metadata stream %" PRIu64
1458 ": stream_path = %s, channel = %s",
1459 stream
->stream_handle
, stream
->path_name
,
1460 stream
->channel_name
);
1464 pthread_mutex_unlock(&stream
->lock
);
1468 memset(&reply
, 0, sizeof(reply
));
1470 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1472 reply
.ret_code
= htobe32(LTTNG_OK
);
1474 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1475 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1476 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1477 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1487 * relay_unknown_command: send -1 if received unknown command
1489 static void relay_unknown_command(struct relay_connection
*conn
)
1491 struct lttcomm_relayd_generic_reply reply
;
1494 memset(&reply
, 0, sizeof(reply
));
1495 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1496 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1497 if (send_ret
< sizeof(reply
)) {
1498 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1503 * relay_start: send an acknowledgment to the client to tell if we are
1504 * ready to receive data. We are ready if a session is established.
1506 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1507 struct relay_connection
*conn
,
1508 const struct lttng_buffer_view
*payload
)
1512 struct lttcomm_relayd_generic_reply reply
;
1513 struct relay_session
*session
= conn
->session
;
1516 DBG("Trying to start the streaming without a session established");
1517 ret
= htobe32(LTTNG_ERR_UNK
);
1520 memset(&reply
, 0, sizeof(reply
));
1521 reply
.ret_code
= htobe32(LTTNG_OK
);
1522 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1524 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1525 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1534 * relay_recv_metadata: receive the metadata for the session.
1536 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1537 struct relay_connection
*conn
,
1538 const struct lttng_buffer_view
*payload
)
1541 struct relay_session
*session
= conn
->session
;
1542 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1543 struct relay_stream
*metadata_stream
;
1544 uint64_t metadata_payload_size
;
1545 struct lttng_buffer_view packet_view
;
1548 ERR("Metadata sent before version check");
1553 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1554 ERR("Incorrect data size");
1558 metadata_payload_size
= recv_hdr
->data_size
-
1559 sizeof(struct lttcomm_relayd_metadata_payload
);
1561 memcpy(&metadata_payload_header
, payload
->data
,
1562 sizeof(metadata_payload_header
));
1563 metadata_payload_header
.stream_id
= be64toh(
1564 metadata_payload_header
.stream_id
);
1565 metadata_payload_header
.padding_size
= be32toh(
1566 metadata_payload_header
.padding_size
);
1568 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1569 if (!metadata_stream
) {
1574 packet_view
= lttng_buffer_view_from_view(payload
,
1575 sizeof(metadata_payload_header
), metadata_payload_size
);
1576 if (!packet_view
.data
) {
1577 ERR("Invalid metadata packet length announced by header");
1582 pthread_mutex_lock(&metadata_stream
->lock
);
1583 ret
= stream_write(metadata_stream
, &packet_view
,
1584 metadata_payload_header
.padding_size
);
1585 pthread_mutex_unlock(&metadata_stream
->lock
);
1591 stream_put(metadata_stream
);
1597 * relay_send_version: send relayd version number
1599 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1600 struct relay_connection
*conn
,
1601 const struct lttng_buffer_view
*payload
)
1605 struct lttcomm_relayd_version reply
, msg
;
1606 bool compatible
= true;
1608 conn
->version_check_done
= true;
1610 /* Get version from the other side. */
1611 if (payload
->size
< sizeof(msg
)) {
1612 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1613 sizeof(msg
), payload
->size
);
1618 memcpy(&msg
, payload
->data
, sizeof(msg
));
1619 msg
.major
= be32toh(msg
.major
);
1620 msg
.minor
= be32toh(msg
.minor
);
1622 memset(&reply
, 0, sizeof(reply
));
1623 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1624 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1626 /* Major versions must be the same */
1627 if (reply
.major
!= msg
.major
) {
1628 DBG("Incompatible major versions (%u vs %u), deleting session",
1629 reply
.major
, msg
.major
);
1633 conn
->major
= reply
.major
;
1634 /* We adapt to the lowest compatible version */
1635 if (reply
.minor
<= msg
.minor
) {
1636 conn
->minor
= reply
.minor
;
1638 conn
->minor
= msg
.minor
;
1641 reply
.major
= htobe32(reply
.major
);
1642 reply
.minor
= htobe32(reply
.minor
);
1643 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1645 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1646 ERR("Failed to send \"send version\" command reply (ret = %zd)",
1659 DBG("Version check done using protocol %u.%u", conn
->major
,
1667 * Check for data pending for a given stream id from the session daemon.
1669 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1670 struct relay_connection
*conn
,
1671 const struct lttng_buffer_view
*payload
)
1673 struct relay_session
*session
= conn
->session
;
1674 struct lttcomm_relayd_data_pending msg
;
1675 struct lttcomm_relayd_generic_reply reply
;
1676 struct relay_stream
*stream
;
1679 uint64_t stream_seq
;
1681 DBG("Data pending command received");
1683 if (!session
|| !conn
->version_check_done
) {
1684 ERR("Trying to check for data before version check");
1686 goto end_no_session
;
1689 if (payload
->size
< sizeof(msg
)) {
1690 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
1691 sizeof(msg
), payload
->size
);
1693 goto end_no_session
;
1695 memcpy(&msg
, payload
->data
, sizeof(msg
));
1696 msg
.stream_id
= be64toh(msg
.stream_id
);
1697 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1699 stream
= stream_get_by_id(msg
.stream_id
);
1700 if (stream
== NULL
) {
1705 pthread_mutex_lock(&stream
->lock
);
1707 if (session_streams_have_index(session
)) {
1709 * Ensure that both the index and stream data have been
1710 * flushed up to the requested point.
1712 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
1714 stream_seq
= stream
->prev_data_seq
;
1716 DBG("Data pending for stream id %" PRIu64
": prev_data_seq %" PRIu64
1717 ", prev_index_seq %" PRIu64
1718 ", and last_seq %" PRIu64
, msg
.stream_id
,
1719 stream
->prev_data_seq
, stream
->prev_index_seq
,
1720 msg
.last_net_seq_num
);
1722 /* Avoid wrapping issue */
1723 if (((int64_t) (stream_seq
- msg
.last_net_seq_num
)) >= 0) {
1724 /* Data has in fact been written and is NOT pending */
1727 /* Data still being streamed thus pending */
1731 stream
->data_pending_check_done
= true;
1732 pthread_mutex_unlock(&stream
->lock
);
1737 memset(&reply
, 0, sizeof(reply
));
1738 reply
.ret_code
= htobe32(ret
);
1739 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1740 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1741 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
1751 * Wait for the control socket to reach a quiescent state.
1753 * Note that for now, when receiving this command from the session
1754 * daemon, this means that every subsequent commands or data received on
1755 * the control socket has been handled. So, this is why we simply return
1758 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
1759 struct relay_connection
*conn
,
1760 const struct lttng_buffer_view
*payload
)
1764 struct relay_stream
*stream
;
1765 struct lttcomm_relayd_quiescent_control msg
;
1766 struct lttcomm_relayd_generic_reply reply
;
1768 DBG("Checking quiescent state on control socket");
1770 if (!conn
->session
|| !conn
->version_check_done
) {
1771 ERR("Trying to check for data before version check");
1773 goto end_no_session
;
1776 if (payload
->size
< sizeof(msg
)) {
1777 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
1778 sizeof(msg
), payload
->size
);
1780 goto end_no_session
;
1782 memcpy(&msg
, payload
->data
, sizeof(msg
));
1783 msg
.stream_id
= be64toh(msg
.stream_id
);
1785 stream
= stream_get_by_id(msg
.stream_id
);
1789 pthread_mutex_lock(&stream
->lock
);
1790 stream
->data_pending_check_done
= true;
1791 pthread_mutex_unlock(&stream
->lock
);
1793 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
1796 memset(&reply
, 0, sizeof(reply
));
1797 reply
.ret_code
= htobe32(LTTNG_OK
);
1798 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1799 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1800 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
1812 * Initialize a data pending command. This means that a consumer is about
1813 * to ask for data pending for each stream it holds. Simply iterate over
1814 * all streams of a session and set the data_pending_check_done flag.
1816 * This command returns to the client a LTTNG_OK code.
1818 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1819 struct relay_connection
*conn
,
1820 const struct lttng_buffer_view
*payload
)
1824 struct lttng_ht_iter iter
;
1825 struct lttcomm_relayd_begin_data_pending msg
;
1826 struct lttcomm_relayd_generic_reply reply
;
1827 struct relay_stream
*stream
;
1832 DBG("Init streams for data pending");
1834 if (!conn
->session
|| !conn
->version_check_done
) {
1835 ERR("Trying to check for data before version check");
1837 goto end_no_session
;
1840 if (payload
->size
< sizeof(msg
)) {
1841 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
1842 sizeof(msg
), payload
->size
);
1844 goto end_no_session
;
1846 memcpy(&msg
, payload
->data
, sizeof(msg
));
1847 msg
.session_id
= be64toh(msg
.session_id
);
1850 * Iterate over all streams to set the begin data pending flag.
1851 * For now, the streams are indexed by stream handle so we have
1852 * to iterate over all streams to find the one associated with
1853 * the right session_id.
1856 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1858 if (!stream_get(stream
)) {
1861 if (stream
->trace
->session
->id
== msg
.session_id
) {
1862 pthread_mutex_lock(&stream
->lock
);
1863 stream
->data_pending_check_done
= false;
1864 pthread_mutex_unlock(&stream
->lock
);
1865 DBG("Set begin data pending flag to stream %" PRIu64
,
1866 stream
->stream_handle
);
1872 memset(&reply
, 0, sizeof(reply
));
1873 /* All good, send back reply. */
1874 reply
.ret_code
= htobe32(LTTNG_OK
);
1876 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1877 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1878 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
1890 * End data pending command. This will check, for a given session id, if
1891 * each stream associated with it has its data_pending_check_done flag
1892 * set. If not, this means that the client lost track of the stream but
1893 * the data is still being streamed on our side. In this case, we inform
1894 * the client that data is in flight.
1896 * Return to the client if there is data in flight or not with a ret_code.
1898 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1899 struct relay_connection
*conn
,
1900 const struct lttng_buffer_view
*payload
)
1904 struct lttng_ht_iter iter
;
1905 struct lttcomm_relayd_end_data_pending msg
;
1906 struct lttcomm_relayd_generic_reply reply
;
1907 struct relay_stream
*stream
;
1908 uint32_t is_data_inflight
= 0;
1910 DBG("End data pending command");
1912 if (!conn
->session
|| !conn
->version_check_done
) {
1913 ERR("Trying to check for data before version check");
1915 goto end_no_session
;
1918 if (payload
->size
< sizeof(msg
)) {
1919 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
1920 sizeof(msg
), payload
->size
);
1922 goto end_no_session
;
1924 memcpy(&msg
, payload
->data
, sizeof(msg
));
1925 msg
.session_id
= be64toh(msg
.session_id
);
1928 * Iterate over all streams to see if the begin data pending
1932 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1934 if (!stream_get(stream
)) {
1937 if (stream
->trace
->session
->id
!= msg
.session_id
) {
1941 pthread_mutex_lock(&stream
->lock
);
1942 if (!stream
->data_pending_check_done
) {
1943 uint64_t stream_seq
;
1945 if (session_streams_have_index(conn
->session
)) {
1947 * Ensure that both the index and stream data have been
1948 * flushed up to the requested point.
1950 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
1952 stream_seq
= stream
->prev_data_seq
;
1954 if (!stream
->closed
|| !(((int64_t) (stream_seq
- stream
->last_net_seq_num
)) >= 0)) {
1955 is_data_inflight
= 1;
1956 DBG("Data is still in flight for stream %" PRIu64
,
1957 stream
->stream_handle
);
1958 pthread_mutex_unlock(&stream
->lock
);
1963 pthread_mutex_unlock(&stream
->lock
);
1968 memset(&reply
, 0, sizeof(reply
));
1969 /* All good, send back reply. */
1970 reply
.ret_code
= htobe32(is_data_inflight
);
1972 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1973 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1974 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
1986 * Receive an index for a specific stream.
1988 * Return 0 on success else a negative value.
1990 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
1991 struct relay_connection
*conn
,
1992 const struct lttng_buffer_view
*payload
)
1996 struct relay_session
*session
= conn
->session
;
1997 struct lttcomm_relayd_index index_info
;
1998 struct lttcomm_relayd_generic_reply reply
;
1999 struct relay_stream
*stream
;
2004 DBG("Relay receiving index");
2006 if (!session
|| !conn
->version_check_done
) {
2007 ERR("Trying to close a stream before version check");
2009 goto end_no_session
;
2012 msg_len
= lttcomm_relayd_index_len(
2013 lttng_to_index_major(conn
->major
, conn
->minor
),
2014 lttng_to_index_minor(conn
->major
, conn
->minor
));
2015 if (payload
->size
< msg_len
) {
2016 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2017 msg_len
, payload
->size
);
2019 goto end_no_session
;
2021 memcpy(&index_info
, payload
->data
, msg_len
);
2022 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2023 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2024 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2025 index_info
.content_size
= be64toh(index_info
.content_size
);
2026 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2027 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2028 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2029 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2031 if (conn
->minor
>= 8) {
2032 index_info
.stream_instance_id
=
2033 be64toh(index_info
.stream_instance_id
);
2034 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2037 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2039 ERR("stream_get_by_id not found");
2044 pthread_mutex_lock(&stream
->lock
);
2045 ret
= stream_add_index(stream
, &index_info
);
2046 pthread_mutex_unlock(&stream
->lock
);
2048 goto end_stream_put
;
2054 memset(&reply
, 0, sizeof(reply
));
2056 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2058 reply
.ret_code
= htobe32(LTTNG_OK
);
2060 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2061 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2062 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2071 * Receive the streams_sent message.
2073 * Return 0 on success else a negative value.
2075 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2076 struct relay_connection
*conn
,
2077 const struct lttng_buffer_view
*payload
)
2081 struct lttcomm_relayd_generic_reply reply
;
2085 DBG("Relay receiving streams_sent");
2087 if (!conn
->session
|| !conn
->version_check_done
) {
2088 ERR("Trying to close a stream before version check");
2090 goto end_no_session
;
2094 * Publish every pending stream in the connection recv list which are
2095 * now ready to be used by the viewer.
2097 publish_connection_local_streams(conn
);
2099 memset(&reply
, 0, sizeof(reply
));
2100 reply
.ret_code
= htobe32(LTTNG_OK
);
2101 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2102 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2103 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2116 * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2117 * session rotation feature (not the tracefile rotation feature).
2119 static int relay_rotate_session_streams(
2120 const struct lttcomm_relayd_hdr
*recv_hdr
,
2121 struct relay_connection
*conn
,
2122 const struct lttng_buffer_view
*payload
)
2127 enum lttng_error_code reply_code
= LTTNG_ERR_UNK
;
2128 struct relay_session
*session
= conn
->session
;
2129 struct lttcomm_relayd_rotate_streams rotate_streams
;
2130 struct lttcomm_relayd_generic_reply reply
= {};
2131 struct relay_stream
*stream
= NULL
;
2132 const size_t header_len
= sizeof(struct lttcomm_relayd_rotate_streams
);
2133 struct lttng_trace_chunk
*next_trace_chunk
= NULL
;
2134 struct lttng_buffer_view stream_positions
;
2135 char chunk_id_buf
[MAX_INT_DEC_LEN(uint64_t)];
2136 const char *chunk_id_str
= "none";
2138 if (!session
|| !conn
->version_check_done
) {
2139 ERR("Trying to rotate a stream before version check");
2144 if (session
->major
== 2 && session
->minor
< 11) {
2145 ERR("Unsupported feature before 2.11");
2150 if (payload
->size
< header_len
) {
2151 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2152 header_len
, payload
->size
);
2157 memcpy(&rotate_streams
, payload
->data
, header_len
);
2159 /* Convert header to host endianness. */
2160 rotate_streams
= (typeof(rotate_streams
)) {
2161 .stream_count
= be32toh(rotate_streams
.stream_count
),
2162 .new_chunk_id
= (typeof(rotate_streams
.new_chunk_id
)) {
2163 .is_set
= !!rotate_streams
.new_chunk_id
.is_set
,
2164 .value
= be64toh(rotate_streams
.new_chunk_id
.value
),
2168 if (rotate_streams
.new_chunk_id
.is_set
) {
2170 * Retrieve the trace chunk the stream must transition to. As
2171 * per the protocol, this chunk should have been created
2172 * before this command is received.
2174 next_trace_chunk
= sessiond_trace_chunk_registry_get_chunk(
2175 sessiond_trace_chunk_registry
,
2176 session
->sessiond_uuid
, session
->id
,
2177 rotate_streams
.new_chunk_id
.value
);
2178 if (!next_trace_chunk
) {
2179 char uuid_str
[UUID_STR_LEN
];
2181 lttng_uuid_to_str(session
->sessiond_uuid
, uuid_str
);
2182 ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2183 ", trace_chunk_id = %" PRIu64
,
2184 uuid_str
, session
->id
,
2185 rotate_streams
.new_chunk_id
.value
);
2186 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2191 ret
= snprintf(chunk_id_buf
, sizeof(chunk_id_buf
), "%" PRIu64
,
2192 rotate_streams
.new_chunk_id
.value
);
2193 if (ret
< 0 || ret
>= sizeof(chunk_id_buf
)) {
2194 chunk_id_str
= "formatting error";
2196 chunk_id_str
= chunk_id_buf
;
2200 DBG("Rotate %" PRIu32
" streams of session \"%s\" to chunk \"%s\"",
2201 rotate_streams
.stream_count
, session
->session_name
,
2204 stream_positions
= lttng_buffer_view_from_view(payload
,
2205 sizeof(rotate_streams
), -1);
2206 if (!stream_positions
.data
||
2207 stream_positions
.size
<
2208 (rotate_streams
.stream_count
*
2209 sizeof(struct lttcomm_relayd_stream_rotation_position
))) {
2210 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2215 for (i
= 0; i
< rotate_streams
.stream_count
; i
++) {
2216 struct lttcomm_relayd_stream_rotation_position
*position_comm
=
2217 &((typeof(position_comm
)) stream_positions
.data
)[i
];
2218 const struct lttcomm_relayd_stream_rotation_position pos
= {
2219 .stream_id
= be64toh(position_comm
->stream_id
),
2220 .rotate_at_seq_num
= be64toh(
2221 position_comm
->rotate_at_seq_num
),
2224 stream
= stream_get_by_id(pos
.stream_id
);
2226 reply_code
= LTTNG_ERR_INVALID
;
2231 pthread_mutex_lock(&stream
->lock
);
2232 ret
= stream_set_pending_rotation(stream
, next_trace_chunk
,
2233 pos
.rotate_at_seq_num
);
2234 pthread_mutex_unlock(&stream
->lock
);
2236 reply_code
= LTTNG_ERR_FILE_CREATION_ERROR
;
2244 reply_code
= LTTNG_OK
;
2250 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2251 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2252 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2253 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2254 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2261 lttng_trace_chunk_put(next_trace_chunk
);
2265 static int init_session_output_directory_handle(struct relay_session
*session
,
2266 struct lttng_directory_handle
*handle
)
2270 * session_directory:
2272 * if base_path is \0'
2273 * hostname/session_name
2275 * hostname/base_path
2277 char *session_directory
= NULL
;
2279 * relayd_output_path/session_directory
2280 * e.g. /home/user/lttng-traces/hostname/session_name
2282 char *full_session_path
= NULL
;
2285 * If base path is set, it overrides the session name for the
2286 * session relative base path. No timestamp is appended if the
2287 * base path is overridden.
2289 * If the session name already contains the creation time (e.g.
2290 * auto-<timestamp>, don't append yet another timestamp after
2291 * the session name in the generated path.
2293 * Otherwise, generate the path with session_name-<timestamp>.
2295 if (session
->base_path
[0] != '\0') {
2296 pthread_mutex_lock(&session
->lock
);
2297 ret
= asprintf(&session_directory
, "%s/%s", session
->hostname
,
2298 session
->base_path
);
2299 pthread_mutex_unlock(&session
->lock
);
2300 } else if (session
->session_name_contains_creation_time
) {
2301 pthread_mutex_lock(&session
->lock
);
2302 ret
= asprintf(&session_directory
, "%s/%s", session
->hostname
,
2303 session
->session_name
);
2304 pthread_mutex_unlock(&session
->lock
);
2306 char session_creation_datetime
[16];
2307 size_t strftime_ret
;
2308 struct tm
*timeinfo
;
2309 time_t creation_time
;
2312 * The 2.11+ protocol guarantees that a creation time
2313 * is provided for a session. This would indicate a
2314 * protocol error or an improper use of this util.
2316 if (!session
->creation_time
.is_set
) {
2317 ERR("Creation time missing for session \"%s\" (protocol error)",
2318 session
->session_name
);
2322 creation_time
= LTTNG_OPTIONAL_GET(session
->creation_time
);
2324 timeinfo
= localtime(&creation_time
);
2326 ERR("Failed to get timeinfo while initializing session output directory handle");
2330 strftime_ret
= strftime(session_creation_datetime
,
2331 sizeof(session_creation_datetime
),
2332 "%Y%m%d-%H%M%S", timeinfo
);
2333 if (strftime_ret
== 0) {
2334 ERR("Failed to format session creation timestamp while initializing session output directory handle");
2338 pthread_mutex_lock(&session
->lock
);
2339 ret
= asprintf(&session_directory
, "%s/%s-%s",
2340 session
->hostname
, session
->session_name
,
2341 session_creation_datetime
);
2342 pthread_mutex_unlock(&session
->lock
);
2345 PERROR("Failed to format session directory name");
2349 full_session_path
= create_output_path(session_directory
);
2350 if (!full_session_path
) {
2355 ret
= utils_mkdir_recursive(
2356 full_session_path
, S_IRWXU
| S_IRWXG
, -1, -1);
2358 ERR("Failed to create session output path \"%s\"",
2363 ret
= lttng_directory_handle_init(handle
, full_session_path
);
2368 free(session_directory
);
2369 free(full_session_path
);
2374 * relay_create_trace_chunk: create a new trace chunk
2376 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2377 struct relay_connection
*conn
,
2378 const struct lttng_buffer_view
*payload
)
2382 struct relay_session
*session
= conn
->session
;
2383 struct lttcomm_relayd_create_trace_chunk
*msg
;
2384 struct lttcomm_relayd_generic_reply reply
= {};
2385 struct lttng_buffer_view header_view
;
2386 struct lttng_buffer_view chunk_name_view
;
2387 struct lttng_trace_chunk
*chunk
= NULL
, *published_chunk
= NULL
;
2388 enum lttng_error_code reply_code
= LTTNG_OK
;
2389 enum lttng_trace_chunk_status chunk_status
;
2390 struct lttng_directory_handle session_output
;
2392 if (!session
|| !conn
->version_check_done
) {
2393 ERR("Trying to create a trace chunk before version check");
2398 if (session
->major
== 2 && session
->minor
< 11) {
2399 ERR("Chunk creation command is unsupported before 2.11");
2404 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2405 if (!header_view
.data
) {
2406 ERR("Failed to receive payload of chunk creation command");
2411 /* Convert to host endianness. */
2412 msg
= (typeof(msg
)) header_view
.data
;
2413 msg
->chunk_id
= be64toh(msg
->chunk_id
);
2414 msg
->creation_timestamp
= be64toh(msg
->creation_timestamp
);
2415 msg
->override_name_length
= be32toh(msg
->override_name_length
);
2417 chunk
= lttng_trace_chunk_create(
2418 msg
->chunk_id
, msg
->creation_timestamp
);
2420 ERR("Failed to create trace chunk in trace chunk creation command");
2422 reply_code
= LTTNG_ERR_NOMEM
;
2426 if (msg
->override_name_length
) {
2429 chunk_name_view
= lttng_buffer_view_from_view(payload
,
2431 msg
->override_name_length
);
2432 name
= chunk_name_view
.data
;
2433 if (!name
|| name
[msg
->override_name_length
- 1]) {
2434 ERR("Failed to receive payload of chunk creation command");
2436 reply_code
= LTTNG_ERR_INVALID
;
2440 chunk_status
= lttng_trace_chunk_override_name(
2441 chunk
, chunk_name_view
.data
);
2442 switch (chunk_status
) {
2443 case LTTNG_TRACE_CHUNK_STATUS_OK
:
2445 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT
:
2446 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2447 reply_code
= LTTNG_ERR_INVALID
;
2451 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2452 reply_code
= LTTNG_ERR_UNK
;
2458 ret
= init_session_output_directory_handle(
2459 conn
->session
, &session_output
);
2461 reply_code
= LTTNG_ERR_CREATE_DIR_FAIL
;
2465 chunk_status
= lttng_trace_chunk_set_credentials_current_user(chunk
);
2466 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2467 reply_code
= LTTNG_ERR_UNK
;
2472 chunk_status
= lttng_trace_chunk_set_as_owner(chunk
, &session_output
);
2473 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2474 reply_code
= LTTNG_ERR_UNK
;
2479 published_chunk
= sessiond_trace_chunk_registry_publish_chunk(
2480 sessiond_trace_chunk_registry
,
2481 conn
->session
->sessiond_uuid
,
2484 if (!published_chunk
) {
2485 char uuid_str
[UUID_STR_LEN
];
2487 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2488 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2493 reply_code
= LTTNG_ERR_NOMEM
;
2497 pthread_mutex_lock(&conn
->session
->lock
);
2498 if (conn
->session
->pending_closure_trace_chunk
) {
2500 * Invalid; this means a second create_trace_chunk command was
2501 * received before a close_trace_chunk.
2503 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2504 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2506 goto end_unlock_session
;
2508 conn
->session
->pending_closure_trace_chunk
=
2509 conn
->session
->current_trace_chunk
;
2510 conn
->session
->current_trace_chunk
= published_chunk
;
2511 published_chunk
= NULL
;
2513 pthread_mutex_unlock(&conn
->session
->lock
);
2515 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2516 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2518 sizeof(struct lttcomm_relayd_generic_reply
),
2520 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2521 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2526 lttng_trace_chunk_put(chunk
);
2527 lttng_trace_chunk_put(published_chunk
);
2528 lttng_directory_handle_fini(&session_output
);
2533 * relay_close_trace_chunk: close a trace chunk
2535 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2536 struct relay_connection
*conn
,
2537 const struct lttng_buffer_view
*payload
)
2541 struct relay_session
*session
= conn
->session
;
2542 struct lttcomm_relayd_close_trace_chunk
*msg
;
2543 struct lttcomm_relayd_generic_reply reply
= {};
2544 struct lttng_buffer_view header_view
;
2545 struct lttng_trace_chunk
*chunk
= NULL
;
2546 enum lttng_error_code reply_code
= LTTNG_OK
;
2547 enum lttng_trace_chunk_status chunk_status
;
2549 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type
) close_command
= {};
2550 time_t close_timestamp
;
2552 if (!session
|| !conn
->version_check_done
) {
2553 ERR("Trying to close a trace chunk before version check");
2558 if (session
->major
== 2 && session
->minor
< 11) {
2559 ERR("Chunk close command is unsupported before 2.11");
2564 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2565 if (!header_view
.data
) {
2566 ERR("Failed to receive payload of chunk close command");
2571 /* Convert to host endianness. */
2572 msg
= (typeof(msg
)) header_view
.data
;
2573 chunk_id
= be64toh(msg
->chunk_id
);
2574 close_timestamp
= (time_t) be64toh(msg
->close_timestamp
);
2575 close_command
= (typeof(close_command
)){
2576 .value
= be32toh(msg
->close_command
.value
),
2577 .is_set
= msg
->close_command
.is_set
,
2580 chunk
= sessiond_trace_chunk_registry_get_chunk(
2581 sessiond_trace_chunk_registry
,
2582 conn
->session
->sessiond_uuid
,
2586 char uuid_str
[UUID_STR_LEN
];
2588 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2589 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2594 reply_code
= LTTNG_ERR_NOMEM
;
2598 pthread_mutex_lock(&session
->lock
);
2599 if (session
->pending_closure_trace_chunk
&&
2600 session
->pending_closure_trace_chunk
!= chunk
) {
2601 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
2602 session
->session_name
);
2603 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2605 goto end_unlock_session
;
2608 chunk_status
= lttng_trace_chunk_set_close_timestamp(
2609 chunk
, close_timestamp
);
2610 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2611 ERR("Failed to set trace chunk close timestamp");
2613 reply_code
= LTTNG_ERR_UNK
;
2614 goto end_unlock_session
;
2617 if (close_command
.is_set
) {
2618 chunk_status
= lttng_trace_chunk_set_close_command(
2619 chunk
, close_command
.value
);
2620 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2622 reply_code
= LTTNG_ERR_INVALID
;
2623 goto end_unlock_session
;
2627 if (session
->current_trace_chunk
== chunk
) {
2629 * After a trace chunk close command, no new streams
2630 * referencing the chunk may be created. Hence, on the
2631 * event that no new trace chunk have been created for
2632 * the session, the reference to the current trace chunk
2633 * is released in order to allow it to be reclaimed when
2634 * the last stream releases its reference to it.
2636 lttng_trace_chunk_put(session
->current_trace_chunk
);
2637 session
->current_trace_chunk
= NULL
;
2639 lttng_trace_chunk_put(session
->pending_closure_trace_chunk
);
2640 session
->pending_closure_trace_chunk
= NULL
;
2642 pthread_mutex_unlock(&session
->lock
);
2645 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2646 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2648 sizeof(struct lttcomm_relayd_generic_reply
),
2650 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2651 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2656 lttng_trace_chunk_put(chunk
);
2661 * relay_trace_chunk_exists: check if a trace chunk exists
2663 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr
*recv_hdr
,
2664 struct relay_connection
*conn
,
2665 const struct lttng_buffer_view
*payload
)
2669 struct relay_session
*session
= conn
->session
;
2670 struct lttcomm_relayd_trace_chunk_exists
*msg
;
2671 struct lttcomm_relayd_trace_chunk_exists_reply reply
= {};
2672 struct lttng_buffer_view header_view
;
2673 struct lttng_trace_chunk
*chunk
= NULL
;
2676 if (!session
|| !conn
->version_check_done
) {
2677 ERR("Trying to close a trace chunk before version check");
2682 if (session
->major
== 2 && session
->minor
< 11) {
2683 ERR("Chunk close command is unsupported before 2.11");
2688 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2689 if (!header_view
.data
) {
2690 ERR("Failed to receive payload of chunk close command");
2695 /* Convert to host endianness. */
2696 msg
= (typeof(msg
)) header_view
.data
;
2697 chunk_id
= be64toh(msg
->chunk_id
);
2699 chunk
= sessiond_trace_chunk_registry_get_chunk(
2700 sessiond_trace_chunk_registry
,
2701 conn
->session
->sessiond_uuid
,
2705 reply
= (typeof(reply
)) {
2706 .generic
.ret_code
= htobe32((uint32_t) LTTNG_OK
),
2707 .trace_chunk_exists
= !!chunk
,
2709 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2710 &reply
, sizeof(reply
), 0);
2711 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2712 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2717 lttng_trace_chunk_put(chunk
);
2721 #define DBG_CMD(cmd_name, conn) \
2722 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2724 static int relay_process_control_command(struct relay_connection
*conn
,
2725 const struct lttcomm_relayd_hdr
*header
,
2726 const struct lttng_buffer_view
*payload
)
2730 switch (header
->cmd
) {
2731 case RELAYD_CREATE_SESSION
:
2732 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
2733 ret
= relay_create_session(header
, conn
, payload
);
2735 case RELAYD_ADD_STREAM
:
2736 DBG_CMD("RELAYD_ADD_STREAM", conn
);
2737 ret
= relay_add_stream(header
, conn
, payload
);
2739 case RELAYD_START_DATA
:
2740 DBG_CMD("RELAYD_START_DATA", conn
);
2741 ret
= relay_start(header
, conn
, payload
);
2743 case RELAYD_SEND_METADATA
:
2744 DBG_CMD("RELAYD_SEND_METADATA", conn
);
2745 ret
= relay_recv_metadata(header
, conn
, payload
);
2747 case RELAYD_VERSION
:
2748 DBG_CMD("RELAYD_VERSION", conn
);
2749 ret
= relay_send_version(header
, conn
, payload
);
2751 case RELAYD_CLOSE_STREAM
:
2752 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
2753 ret
= relay_close_stream(header
, conn
, payload
);
2755 case RELAYD_DATA_PENDING
:
2756 DBG_CMD("RELAYD_DATA_PENDING", conn
);
2757 ret
= relay_data_pending(header
, conn
, payload
);
2759 case RELAYD_QUIESCENT_CONTROL
:
2760 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
2761 ret
= relay_quiescent_control(header
, conn
, payload
);
2763 case RELAYD_BEGIN_DATA_PENDING
:
2764 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
2765 ret
= relay_begin_data_pending(header
, conn
, payload
);
2767 case RELAYD_END_DATA_PENDING
:
2768 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
2769 ret
= relay_end_data_pending(header
, conn
, payload
);
2771 case RELAYD_SEND_INDEX
:
2772 DBG_CMD("RELAYD_SEND_INDEX", conn
);
2773 ret
= relay_recv_index(header
, conn
, payload
);
2775 case RELAYD_STREAMS_SENT
:
2776 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
2777 ret
= relay_streams_sent(header
, conn
, payload
);
2779 case RELAYD_RESET_METADATA
:
2780 DBG_CMD("RELAYD_RESET_METADATA", conn
);
2781 ret
= relay_reset_metadata(header
, conn
, payload
);
2783 case RELAYD_ROTATE_STREAMS
:
2784 DBG_CMD("RELAYD_ROTATE_STREAMS", conn
);
2785 ret
= relay_rotate_session_streams(header
, conn
, payload
);
2787 case RELAYD_CREATE_TRACE_CHUNK
:
2788 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn
);
2789 ret
= relay_create_trace_chunk(header
, conn
, payload
);
2791 case RELAYD_CLOSE_TRACE_CHUNK
:
2792 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn
);
2793 ret
= relay_close_trace_chunk(header
, conn
, payload
);
2795 case RELAYD_TRACE_CHUNK_EXISTS
:
2796 DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn
);
2797 ret
= relay_trace_chunk_exists(header
, conn
, payload
);
2799 case RELAYD_UPDATE_SYNC_INFO
:
2801 ERR("Received unknown command (%u)", header
->cmd
);
2802 relay_unknown_command(conn
);
2811 static enum relay_connection_status
relay_process_control_receive_payload(
2812 struct relay_connection
*conn
)
2815 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2816 struct lttng_dynamic_buffer
*reception_buffer
=
2817 &conn
->protocol
.ctrl
.reception_buffer
;
2818 struct ctrl_connection_state_receive_payload
*state
=
2819 &conn
->protocol
.ctrl
.state
.receive_payload
;
2820 struct lttng_buffer_view payload_view
;
2822 if (state
->left_to_receive
== 0) {
2823 /* Short-circuit for payload-less commands. */
2824 goto reception_complete
;
2827 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2828 reception_buffer
->data
+ state
->received
,
2829 state
->left_to_receive
, MSG_DONTWAIT
);
2831 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2832 PERROR("Unable to receive command payload on sock %d",
2834 status
= RELAY_CONNECTION_STATUS_ERROR
;
2837 } else if (ret
== 0) {
2838 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2839 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2844 assert(ret
<= state
->left_to_receive
);
2846 state
->left_to_receive
-= ret
;
2847 state
->received
+= ret
;
2849 if (state
->left_to_receive
> 0) {
2851 * Can't transition to the protocol's next state, wait to
2852 * receive the rest of the header.
2854 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2855 state
->received
, state
->left_to_receive
,
2861 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
2862 conn
->sock
->fd
, state
->received
);
2864 * The payload required to process the command has been received.
2865 * A view to the reception buffer is forwarded to the various
2866 * commands and the state of the control is reset on success.
2868 * Commands are responsible for sending their reply to the peer.
2870 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
2872 ret
= relay_process_control_command(conn
,
2873 &state
->header
, &payload_view
);
2875 status
= RELAY_CONNECTION_STATUS_ERROR
;
2879 ret
= connection_reset_protocol_state(conn
);
2881 status
= RELAY_CONNECTION_STATUS_ERROR
;
2887 static enum relay_connection_status
relay_process_control_receive_header(
2888 struct relay_connection
*conn
)
2891 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2892 struct lttcomm_relayd_hdr header
;
2893 struct lttng_dynamic_buffer
*reception_buffer
=
2894 &conn
->protocol
.ctrl
.reception_buffer
;
2895 struct ctrl_connection_state_receive_header
*state
=
2896 &conn
->protocol
.ctrl
.state
.receive_header
;
2898 assert(state
->left_to_receive
!= 0);
2900 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2901 reception_buffer
->data
+ state
->received
,
2902 state
->left_to_receive
, MSG_DONTWAIT
);
2904 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2905 PERROR("Unable to receive control command header on sock %d",
2907 status
= RELAY_CONNECTION_STATUS_ERROR
;
2910 } else if (ret
== 0) {
2911 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2912 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2917 assert(ret
<= state
->left_to_receive
);
2919 state
->left_to_receive
-= ret
;
2920 state
->received
+= ret
;
2922 if (state
->left_to_receive
> 0) {
2924 * Can't transition to the protocol's next state, wait to
2925 * receive the rest of the header.
2927 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2928 state
->received
, state
->left_to_receive
,
2933 /* Transition to next state: receiving the command's payload. */
2934 conn
->protocol
.ctrl
.state_id
=
2935 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
2936 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
2937 header
.circuit_id
= be64toh(header
.circuit_id
);
2938 header
.data_size
= be64toh(header
.data_size
);
2939 header
.cmd
= be32toh(header
.cmd
);
2940 header
.cmd_version
= be32toh(header
.cmd_version
);
2941 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
2942 &header
, sizeof(header
));
2944 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
2945 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
2948 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
2949 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
2951 status
= RELAY_CONNECTION_STATUS_ERROR
;
2955 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
2957 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
2958 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
2961 status
= RELAY_CONNECTION_STATUS_ERROR
;
2965 if (header
.data_size
== 0) {
2967 * Manually invoke the next state as the poll loop
2968 * will not wake-up to allow us to proceed further.
2970 status
= relay_process_control_receive_payload(conn
);
2977 * Process the commands received on the control socket
2979 static enum relay_connection_status
relay_process_control(
2980 struct relay_connection
*conn
)
2982 enum relay_connection_status status
;
2984 switch (conn
->protocol
.ctrl
.state_id
) {
2985 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
2986 status
= relay_process_control_receive_header(conn
);
2988 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
2989 status
= relay_process_control_receive_payload(conn
);
2992 ERR("Unknown control connection protocol state encountered.");
2999 static enum relay_connection_status
relay_process_data_receive_header(
3000 struct relay_connection
*conn
)
3003 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3004 struct data_connection_state_receive_header
*state
=
3005 &conn
->protocol
.data
.state
.receive_header
;
3006 struct lttcomm_relayd_data_hdr header
;
3007 struct relay_stream
*stream
;
3009 assert(state
->left_to_receive
!= 0);
3011 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3012 state
->header_reception_buffer
+ state
->received
,
3013 state
->left_to_receive
, MSG_DONTWAIT
);
3015 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3016 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
3017 status
= RELAY_CONNECTION_STATUS_ERROR
;
3020 } else if (ret
== 0) {
3021 /* Orderly shutdown. Not necessary to print an error. */
3022 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3023 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3028 assert(ret
<= state
->left_to_receive
);
3030 state
->left_to_receive
-= ret
;
3031 state
->received
+= ret
;
3033 if (state
->left_to_receive
> 0) {
3035 * Can't transition to the protocol's next state, wait to
3036 * receive the rest of the header.
3038 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3039 state
->received
, state
->left_to_receive
,
3044 /* Transition to next state: receiving the payload. */
3045 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3047 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3048 header
.circuit_id
= be64toh(header
.circuit_id
);
3049 header
.stream_id
= be64toh(header
.stream_id
);
3050 header
.data_size
= be32toh(header
.data_size
);
3051 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3052 header
.padding_size
= be32toh(header
.padding_size
);
3053 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3055 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3057 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3058 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3060 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3061 conn
->sock
->fd
, header
.circuit_id
,
3062 header
.stream_id
, header
.data_size
,
3063 header
.net_seq_num
, header
.padding_size
);
3065 stream
= stream_get_by_id(header
.stream_id
);
3067 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3069 /* Protocol error. */
3070 status
= RELAY_CONNECTION_STATUS_ERROR
;
3074 pthread_mutex_lock(&stream
->lock
);
3075 /* Prepare stream for the reception of a new packet. */
3076 ret
= stream_init_packet(stream
, header
.data_size
,
3077 &conn
->protocol
.data
.state
.receive_payload
.rotate_index
);
3078 pthread_mutex_unlock(&stream
->lock
);
3080 ERR("Failed to rotate stream output file");
3081 status
= RELAY_CONNECTION_STATUS_ERROR
;
3082 goto end_stream_unlock
;
3091 static enum relay_connection_status
relay_process_data_receive_payload(
3092 struct relay_connection
*conn
)
3095 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3096 struct relay_stream
*stream
;
3097 struct data_connection_state_receive_payload
*state
=
3098 &conn
->protocol
.data
.state
.receive_payload
;
3099 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3100 char data_buffer
[chunk_size
];
3101 bool partial_recv
= false;
3102 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3103 uint64_t left_to_receive
= state
->left_to_receive
;
3104 struct relay_session
*session
;
3106 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3107 state
->header
.stream_id
, state
->header
.net_seq_num
,
3108 state
->received
, left_to_receive
);
3110 stream
= stream_get_by_id(state
->header
.stream_id
);
3112 /* Protocol error. */
3113 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3114 state
->header
.stream_id
);
3115 status
= RELAY_CONNECTION_STATUS_ERROR
;
3119 pthread_mutex_lock(&stream
->lock
);
3120 session
= stream
->trace
->session
;
3121 if (!conn
->session
) {
3122 ret
= connection_set_session(conn
, session
);
3124 status
= RELAY_CONNECTION_STATUS_ERROR
;
3125 goto end_stream_unlock
;
3130 * The size of the "chunk" received on any iteration is bounded by:
3131 * - the data left to receive,
3132 * - the data immediately available on the socket,
3133 * - the on-stack data buffer
3135 while (left_to_receive
> 0 && !partial_recv
) {
3136 size_t recv_size
= min(left_to_receive
, chunk_size
);
3137 struct lttng_buffer_view packet_chunk
;
3139 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3140 recv_size
, MSG_DONTWAIT
);
3142 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3143 PERROR("Socket %d error", conn
->sock
->fd
);
3144 status
= RELAY_CONNECTION_STATUS_ERROR
;
3146 goto end_stream_unlock
;
3147 } else if (ret
== 0) {
3148 /* No more data ready to be consumed on socket. */
3149 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3150 state
->header
.stream_id
);
3151 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3153 } else if (ret
< (int) recv_size
) {
3155 * All the data available on the socket has been
3158 partial_recv
= true;
3162 packet_chunk
= lttng_buffer_view_init(data_buffer
,
3164 assert(packet_chunk
.data
);
3166 ret
= stream_write(stream
, &packet_chunk
, 0);
3168 ERR("Relay error writing data to file");
3169 status
= RELAY_CONNECTION_STATUS_ERROR
;
3170 goto end_stream_unlock
;
3173 left_to_receive
-= recv_size
;
3174 state
->received
+= recv_size
;
3175 state
->left_to_receive
= left_to_receive
;
3178 if (state
->left_to_receive
> 0) {
3180 * Did not receive all the data expected, wait for more data to
3181 * become available on the socket.
3183 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3184 state
->header
.stream_id
, state
->received
,
3185 state
->left_to_receive
);
3186 goto end_stream_unlock
;
3189 ret
= stream_write(stream
, NULL
, state
->header
.padding_size
);
3191 status
= RELAY_CONNECTION_STATUS_ERROR
;
3192 goto end_stream_unlock
;
3195 if (session_streams_have_index(session
)) {
3196 ret
= stream_update_index(stream
, state
->header
.net_seq_num
,
3197 state
->rotate_index
, &index_flushed
,
3198 state
->header
.data_size
+ state
->header
.padding_size
);
3200 ERR("Failed to update index: stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3201 stream
->stream_handle
,
3202 state
->header
.net_seq_num
, ret
);
3203 status
= RELAY_CONNECTION_STATUS_ERROR
;
3204 goto end_stream_unlock
;
3208 if (stream
->prev_data_seq
== -1ULL) {
3212 ret
= stream_complete_packet(stream
, state
->header
.data_size
+
3213 state
->header
.padding_size
, state
->header
.net_seq_num
,
3216 status
= RELAY_CONNECTION_STATUS_ERROR
;
3217 goto end_stream_unlock
;
3221 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3222 * contents of *state which are aliased (union) to the same location as
3223 * the new state. Don't use it beyond this point.
3225 connection_reset_protocol_state(conn
);
3229 close_requested
= stream
->close_requested
;
3230 pthread_mutex_unlock(&stream
->lock
);
3231 if (close_requested
&& left_to_receive
== 0) {
3232 try_stream_close(stream
);
3236 pthread_mutex_lock(&session
->lock
);
3237 uatomic_set(&session
->new_streams
, 1);
3238 pthread_mutex_unlock(&session
->lock
);
3247 * relay_process_data: Process the data received on the data socket
3249 static enum relay_connection_status
relay_process_data(
3250 struct relay_connection
*conn
)
3252 enum relay_connection_status status
;
3254 switch (conn
->protocol
.data
.state_id
) {
3255 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3256 status
= relay_process_data_receive_header(conn
);
3258 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3259 status
= relay_process_data_receive_payload(conn
);
3262 ERR("Unexpected data connection communication state.");
3269 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3273 (void) lttng_poll_del(events
, pollfd
);
3275 ret
= close(pollfd
);
3277 ERR("Closing pollfd %d", pollfd
);
3281 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3282 int pollfd
, struct relay_connection
*conn
)
3284 const char *type_str
;
3286 switch (conn
->type
) {
3291 type_str
= "Control";
3293 case RELAY_VIEWER_COMMAND
:
3294 type_str
= "Viewer Command";
3296 case RELAY_VIEWER_NOTIFICATION
:
3297 type_str
= "Viewer Notification";
3300 type_str
= "Unknown";
3302 cleanup_connection_pollfd(events
, pollfd
);
3303 connection_put(conn
);
3304 DBG("%s connection closed with %d", type_str
, pollfd
);
3308 * This thread does the actual work
3310 static void *relay_thread_worker(void *data
)
3312 int ret
, err
= -1, last_seen_data_fd
= -1;
3314 struct lttng_poll_event events
;
3315 struct lttng_ht
*relay_connections_ht
;
3316 struct lttng_ht_iter iter
;
3317 struct relay_connection
*destroy_conn
= NULL
;
3319 DBG("[thread] Relay worker started");
3321 rcu_register_thread();
3323 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3325 if (testpoint(relayd_thread_worker
)) {
3326 goto error_testpoint
;
3329 health_code_update();
3331 /* table of connections indexed on socket */
3332 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3333 if (!relay_connections_ht
) {
3334 goto relay_connections_ht_error
;
3337 ret
= create_thread_poll_set(&events
, 2);
3339 goto error_poll_create
;
3342 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3349 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3351 health_code_update();
3353 /* Infinite blocking call, waiting for transmission */
3354 DBG3("Relayd worker thread polling...");
3355 health_poll_entry();
3356 ret
= lttng_poll_wait(&events
, -1);
3360 * Restart interrupted system call.
3362 if (errno
== EINTR
) {
3371 * Process control. The control connection is
3372 * prioritized so we don't starve it with high
3373 * throughput tracing data on the data connection.
3375 for (i
= 0; i
< nb_fd
; i
++) {
3376 /* Fetch once the poll data */
3377 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3378 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3380 health_code_update();
3382 /* Thread quit pipe has been closed. Killing thread. */
3383 ret
= check_thread_quit_pipe(pollfd
, revents
);
3389 /* Inspect the relay conn pipe for new connection */
3390 if (pollfd
== relay_conn_pipe
[0]) {
3391 if (revents
& LPOLLIN
) {
3392 struct relay_connection
*conn
;
3394 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3398 lttng_poll_add(&events
, conn
->sock
->fd
,
3399 LPOLLIN
| LPOLLRDHUP
);
3400 connection_ht_add(relay_connections_ht
, conn
);
3401 DBG("Connection socket %d added", conn
->sock
->fd
);
3402 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3403 ERR("Relay connection pipe error");
3406 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3410 struct relay_connection
*ctrl_conn
;
3412 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3413 /* If not found, there is a synchronization issue. */
3416 if (ctrl_conn
->type
== RELAY_DATA
) {
3417 if (revents
& LPOLLIN
) {
3419 * Flag the last seen data fd not deleted. It will be
3420 * used as the last seen fd if any fd gets deleted in
3423 last_notdel_data_fd
= pollfd
;
3425 goto put_ctrl_connection
;
3427 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3429 if (revents
& LPOLLIN
) {
3430 enum relay_connection_status status
;
3432 status
= relay_process_control(ctrl_conn
);
3433 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3435 * On socket error flag the session as aborted to force
3436 * the cleanup of its stream otherwise it can leak
3437 * during the lifetime of the relayd.
3439 * This prevents situations in which streams can be
3440 * left opened because an index was received, the
3441 * control connection is closed, and the data
3442 * connection is closed (uncleanly) before the packet's
3445 * Since the control connection encountered an error,
3446 * it is okay to be conservative and close the
3447 * session right now as we can't rely on the protocol
3448 * being respected anymore.
3450 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3451 session_abort(ctrl_conn
->session
);
3454 /* Clear the connection on error or close. */
3455 relay_thread_close_connection(&events
,
3460 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3461 relay_thread_close_connection(&events
,
3463 if (last_seen_data_fd
== pollfd
) {
3464 last_seen_data_fd
= last_notdel_data_fd
;
3467 ERR("Unexpected poll events %u for control sock %d",
3469 connection_put(ctrl_conn
);
3472 put_ctrl_connection
:
3473 connection_put(ctrl_conn
);
3478 * The last loop handled a control request, go back to poll to make
3479 * sure we prioritise the control socket.
3485 if (last_seen_data_fd
>= 0) {
3486 for (i
= 0; i
< nb_fd
; i
++) {
3487 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3489 health_code_update();
3491 if (last_seen_data_fd
== pollfd
) {
3498 /* Process data connection. */
3499 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3500 /* Fetch the poll data. */
3501 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3502 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3503 struct relay_connection
*data_conn
;
3505 health_code_update();
3508 /* No activity for this FD (poll implementation). */
3512 /* Skip the command pipe. It's handled in the first loop. */
3513 if (pollfd
== relay_conn_pipe
[0]) {
3517 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3519 /* Skip it. Might be removed before. */
3522 if (data_conn
->type
== RELAY_CONTROL
) {
3523 goto put_data_connection
;
3525 assert(data_conn
->type
== RELAY_DATA
);
3527 if (revents
& LPOLLIN
) {
3528 enum relay_connection_status status
;
3530 status
= relay_process_data(data_conn
);
3531 /* Connection closed or error. */
3532 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3534 * On socket error flag the session as aborted to force
3535 * the cleanup of its stream otherwise it can leak
3536 * during the lifetime of the relayd.
3538 * This prevents situations in which streams can be
3539 * left opened because an index was received, the
3540 * control connection is closed, and the data
3541 * connection is closed (uncleanly) before the packet's
3544 * Since the data connection encountered an error,
3545 * it is okay to be conservative and close the
3546 * session right now as we can't rely on the protocol
3547 * being respected anymore.
3549 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3550 session_abort(data_conn
->session
);
3552 relay_thread_close_connection(&events
, pollfd
,
3555 * Every goto restart call sets the last seen fd where
3556 * here we don't really care since we gracefully
3557 * continue the loop after the connection is deleted.
3560 /* Keep last seen port. */
3561 last_seen_data_fd
= pollfd
;
3562 connection_put(data_conn
);
3565 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3566 relay_thread_close_connection(&events
, pollfd
,
3569 ERR("Unknown poll events %u for data sock %d",
3572 put_data_connection
:
3573 connection_put(data_conn
);
3575 last_seen_data_fd
= -1;
3578 /* Normal exit, no error */
3583 /* Cleanup remaining connection object. */
3585 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3588 health_code_update();
3590 session_abort(destroy_conn
->session
);
3593 * No need to grab another ref, because we own
3596 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
3601 lttng_poll_clean(&events
);
3603 lttng_ht_destroy(relay_connections_ht
);
3604 relay_connections_ht_error
:
3605 /* Close relay conn pipes */
3606 utils_close_pipe(relay_conn_pipe
);
3608 DBG("Thread exited with error");
3610 DBG("Worker thread cleanup complete");
3614 ERR("Health error occurred in %s", __func__
);
3616 health_unregister(health_relayd
);
3617 rcu_unregister_thread();
3618 lttng_relay_stop_threads();
3623 * Create the relay command pipe to wake thread_manage_apps.
3624 * Closed in cleanup().
3626 static int create_relay_conn_pipe(void)
3630 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
3638 int main(int argc
, char **argv
)
3640 int ret
= 0, retval
= 0;
3643 /* Parse arguments */
3645 if (set_options(argc
, argv
)) {
3650 if (set_signal_handler()) {
3655 /* Try to create directory if -o, --output is specified. */
3656 if (opt_output_path
) {
3657 if (*opt_output_path
!= '/') {
3658 ERR("Please specify an absolute path for -o, --output PATH");
3663 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
3666 ERR("Unable to create %s", opt_output_path
);
3673 if (opt_daemon
|| opt_background
) {
3676 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
3684 * We are in the child. Make sure all other file
3685 * descriptors are closed, in case we are called with
3686 * more opened file descriptors than the standard ones.
3688 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3693 sessiond_trace_chunk_registry
= sessiond_trace_chunk_registry_create();
3694 if (!sessiond_trace_chunk_registry
) {
3695 ERR("Failed to initialize session daemon trace chunk registry");
3697 goto exit_sessiond_trace_chunk_registry
;
3700 /* Initialize thread health monitoring */
3701 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
3702 if (!health_relayd
) {
3703 PERROR("health_app_create error");
3705 goto exit_health_app_create
;
3708 /* Create thread quit pipe */
3709 if (init_thread_quit_pipe()) {
3711 goto exit_init_data
;
3714 /* Setup the thread apps communication pipe. */
3715 if (create_relay_conn_pipe()) {
3717 goto exit_init_data
;
3720 /* Init relay command queue. */
3721 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
3723 /* Initialize communication library */
3725 lttcomm_inet_init();
3727 /* tables of sessions indexed by session ID */
3728 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3731 goto exit_init_data
;
3734 /* tables of streams indexed by stream ID */
3735 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3736 if (!relay_streams_ht
) {
3738 goto exit_init_data
;
3741 /* tables of streams indexed by stream ID */
3742 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3743 if (!viewer_streams_ht
) {
3745 goto exit_init_data
;
3748 ret
= utils_create_pipe(health_quit_pipe
);
3751 goto exit_health_quit_pipe
;
3754 /* Create thread to manage the client socket */
3755 ret
= pthread_create(&health_thread
, default_pthread_attr(),
3756 thread_manage_health
, (void *) NULL
);
3759 PERROR("pthread_create health");
3761 goto exit_health_thread
;
3764 /* Setup the dispatcher thread */
3765 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
3766 relay_thread_dispatcher
, (void *) NULL
);
3769 PERROR("pthread_create dispatcher");
3771 goto exit_dispatcher_thread
;
3774 /* Setup the worker thread */
3775 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
3776 relay_thread_worker
, NULL
);
3779 PERROR("pthread_create worker");
3781 goto exit_worker_thread
;
3784 /* Setup the listener thread */
3785 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
3786 relay_thread_listener
, (void *) NULL
);
3789 PERROR("pthread_create listener");
3791 goto exit_listener_thread
;
3794 ret
= relayd_live_create(live_uri
);
3796 ERR("Starting live viewer threads");
3802 * This is where we start awaiting program completion (e.g. through
3803 * signal that asks threads to teardown).
3806 ret
= relayd_live_join();
3812 ret
= pthread_join(listener_thread
, &status
);
3815 PERROR("pthread_join listener_thread");
3819 exit_listener_thread
:
3820 ret
= pthread_join(worker_thread
, &status
);
3823 PERROR("pthread_join worker_thread");
3828 ret
= pthread_join(dispatcher_thread
, &status
);
3831 PERROR("pthread_join dispatcher_thread");
3834 exit_dispatcher_thread
:
3836 ret
= pthread_join(health_thread
, &status
);
3839 PERROR("pthread_join health_thread");
3844 utils_close_pipe(health_quit_pipe
);
3845 exit_health_quit_pipe
:
3848 health_app_destroy(health_relayd
);
3849 sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry
);
3850 exit_health_app_create
:
3851 exit_sessiond_trace_chunk_registry
:
3854 * Wait for all pending call_rcu work to complete before tearing
3855 * down data structures. call_rcu worker may be trying to
3856 * perform lookups in those structures.
3861 /* Ensure all prior call_rcu are done. */