2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
35 #include <sys/types.h>
38 #include <urcu/futex.h>
39 #include <urcu/uatomic.h>
40 #include <urcu/rculist.h>
44 #include <lttng/lttng.h>
45 #include <common/common.h>
46 #include <common/compat/poll.h>
47 #include <common/compat/socket.h>
48 #include <common/compat/endian.h>
49 #include <common/compat/getenv.h>
50 #include <common/defaults.h>
51 #include <common/daemonize.h>
52 #include <common/futex.h>
53 #include <common/sessiond-comm/sessiond-comm.h>
54 #include <common/sessiond-comm/inet.h>
55 #include <common/sessiond-comm/relayd.h>
56 #include <common/uri.h>
57 #include <common/utils.h>
58 #include <common/align.h>
59 #include <common/config/session-config.h>
60 #include <common/dynamic-buffer.h>
61 #include <common/buffer-view.h>
62 #include <common/string-utils/format.h>
65 #include "ctf-trace.h"
68 #include "lttng-relayd.h"
70 #include "health-relayd.h"
71 #include "testpoint.h"
72 #include "viewer-stream.h"
75 #include "connection.h"
76 #include "tracefile-array.h"
77 #include "tcp_keep_alive.h"
78 #include "sessiond-trace-chunks.h"
80 static const char *help_msg
=
81 #ifdef LTTNG_EMBED_HELP
82 #include <lttng-relayd.8.h>
88 enum relay_connection_status
{
89 RELAY_CONNECTION_STATUS_OK
,
90 /* An error occurred while processing an event on the connection. */
91 RELAY_CONNECTION_STATUS_ERROR
,
92 /* Connection closed/shutdown cleanly. */
93 RELAY_CONNECTION_STATUS_CLOSED
,
96 /* command line options */
97 char *opt_output_path
;
98 static int opt_daemon
, opt_background
;
101 * We need to wait for listener and live listener threads, as well as
102 * health check thread, before being ready to signal readiness.
104 #define NR_LTTNG_RELAY_READY 3
105 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
107 /* Size of receive buffer. */
108 #define RECV_DATA_BUFFER_SIZE 65536
110 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
111 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
113 static struct lttng_uri
*control_uri
;
114 static struct lttng_uri
*data_uri
;
115 static struct lttng_uri
*live_uri
;
117 const char *progname
;
119 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
120 static int tracing_group_name_override
;
122 const char * const config_section_name
= "relayd";
125 * Quit pipe for all threads. This permits a single cancellation point
126 * for all threads when receiving an event on the pipe.
128 int thread_quit_pipe
[2] = { -1, -1 };
131 * This pipe is used to inform the worker thread that a command is queued and
132 * ready to be processed.
134 static int relay_conn_pipe
[2] = { -1, -1 };
136 /* Shared between threads */
137 static int dispatch_thread_exit
;
139 static pthread_t listener_thread
;
140 static pthread_t dispatcher_thread
;
141 static pthread_t worker_thread
;
142 static pthread_t health_thread
;
145 * last_relay_stream_id_lock protects last_relay_stream_id increment
146 * atomicity on 32-bit architectures.
148 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
149 static uint64_t last_relay_stream_id
;
152 * Relay command queue.
154 * The relay_thread_listener and relay_thread_dispatcher communicate with this
157 static struct relay_conn_queue relay_conn_queue
;
159 /* Global relay stream hash table. */
160 struct lttng_ht
*relay_streams_ht
;
162 /* Global relay viewer stream hash table. */
163 struct lttng_ht
*viewer_streams_ht
;
165 /* Global relay sessions hash table. */
166 struct lttng_ht
*sessions_ht
;
168 /* Relayd health monitoring */
169 struct health_app
*health_relayd
;
171 struct sessiond_trace_chunk_registry
*sessiond_trace_chunk_registry
;
173 static struct option long_options
[] = {
174 { "control-port", 1, 0, 'C', },
175 { "data-port", 1, 0, 'D', },
176 { "live-port", 1, 0, 'L', },
177 { "daemonize", 0, 0, 'd', },
178 { "background", 0, 0, 'b', },
179 { "group", 1, 0, 'g', },
180 { "help", 0, 0, 'h', },
181 { "output", 1, 0, 'o', },
182 { "verbose", 0, 0, 'v', },
183 { "config", 1, 0, 'f' },
184 { "version", 0, 0, 'V' },
188 static const char *config_ignore_options
[] = { "help", "config", "version" };
191 * Take an option from the getopt output and set it in the right variable to be
194 * Return 0 on success else a negative value.
196 static int set_option(int opt
, const char *arg
, const char *optname
)
202 fprintf(stderr
, "option %s", optname
);
204 fprintf(stderr
, " with arg %s\n", arg
);
208 if (lttng_is_setuid_setgid()) {
209 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
210 "-C, --control-port");
212 ret
= uri_parse(arg
, &control_uri
);
214 ERR("Invalid control URI specified");
217 if (control_uri
->port
== 0) {
218 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
223 if (lttng_is_setuid_setgid()) {
224 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
227 ret
= uri_parse(arg
, &data_uri
);
229 ERR("Invalid data URI specified");
232 if (data_uri
->port
== 0) {
233 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
238 if (lttng_is_setuid_setgid()) {
239 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
242 ret
= uri_parse(arg
, &live_uri
);
244 ERR("Invalid live URI specified");
247 if (live_uri
->port
== 0) {
248 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
259 if (lttng_is_setuid_setgid()) {
260 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
263 tracing_group_name
= strdup(arg
);
264 if (tracing_group_name
== NULL
) {
269 tracing_group_name_override
= 1;
273 ret
= utils_show_help(8, "lttng-relayd", help_msg
);
275 ERR("Cannot show --help for `lttng-relayd`");
280 fprintf(stdout
, "%s\n", VERSION
);
283 if (lttng_is_setuid_setgid()) {
284 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
287 ret
= asprintf(&opt_output_path
, "%s", arg
);
290 PERROR("asprintf opt_output_path");
296 /* Verbose level can increase using multiple -v */
298 lttng_opt_verbose
= config_parse_value(arg
);
300 /* Only 3 level of verbosity (-vvv). */
301 if (lttng_opt_verbose
< 3) {
302 lttng_opt_verbose
+= 1;
307 /* Unknown option or other error.
308 * Error is printed by getopt, just return */
321 * config_entry_handler_cb used to handle options read from a config file.
322 * See config_entry_handler_cb comment in common/config/session-config.h for the
323 * return value conventions.
325 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
329 if (!entry
|| !entry
->name
|| !entry
->value
) {
334 /* Check if the option is to be ignored */
335 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
336 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
341 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
342 /* Ignore if entry name is not fully matched. */
343 if (strcmp(entry
->name
, long_options
[i
].name
)) {
348 * If the option takes no argument on the command line,
349 * we have to check if the value is "true". We support
350 * non-zero numeric values, true, on and yes.
352 if (!long_options
[i
].has_arg
) {
353 ret
= config_parse_value(entry
->value
);
356 WARN("Invalid configuration value \"%s\" for option %s",
357 entry
->value
, entry
->name
);
359 /* False, skip boolean config option. */
364 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
368 WARN("Unrecognized option \"%s\" in daemon configuration file.",
375 static int set_options(int argc
, char **argv
)
377 int c
, ret
= 0, option_index
= 0, retval
= 0;
378 int orig_optopt
= optopt
, orig_optind
= optind
;
379 char *default_address
, *optstring
;
380 const char *config_path
= NULL
;
382 optstring
= utils_generate_optstring(long_options
,
383 sizeof(long_options
) / sizeof(struct option
));
389 /* Check for the --config option */
391 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
392 &option_index
)) != -1) {
396 } else if (c
!= 'f') {
400 if (lttng_is_setuid_setgid()) {
401 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
404 config_path
= utils_expand_path(optarg
);
406 ERR("Failed to resolve path: %s", optarg
);
411 ret
= config_get_section_entries(config_path
, config_section_name
,
412 config_entry_handler
, NULL
);
415 ERR("Invalid configuration option at line %i", ret
);
421 /* Reset getopt's global state */
422 optopt
= orig_optopt
;
423 optind
= orig_optind
;
425 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
430 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
437 /* assign default values */
438 if (control_uri
== NULL
) {
439 ret
= asprintf(&default_address
,
440 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
441 DEFAULT_NETWORK_CONTROL_PORT
);
443 PERROR("asprintf default data address");
448 ret
= uri_parse(default_address
, &control_uri
);
449 free(default_address
);
451 ERR("Invalid control URI specified");
456 if (data_uri
== NULL
) {
457 ret
= asprintf(&default_address
,
458 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
459 DEFAULT_NETWORK_DATA_PORT
);
461 PERROR("asprintf default data address");
466 ret
= uri_parse(default_address
, &data_uri
);
467 free(default_address
);
469 ERR("Invalid data URI specified");
474 if (live_uri
== NULL
) {
475 ret
= asprintf(&default_address
,
476 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
477 DEFAULT_NETWORK_VIEWER_PORT
);
479 PERROR("asprintf default viewer control address");
484 ret
= uri_parse(default_address
, &live_uri
);
485 free(default_address
);
487 ERR("Invalid viewer control URI specified");
498 static void print_global_objects(void)
500 rcu_register_thread();
502 print_viewer_streams();
503 print_relay_streams();
506 rcu_unregister_thread();
512 static void relayd_cleanup(void)
514 print_global_objects();
518 if (viewer_streams_ht
)
519 lttng_ht_destroy(viewer_streams_ht
);
520 if (relay_streams_ht
)
521 lttng_ht_destroy(relay_streams_ht
);
523 lttng_ht_destroy(sessions_ht
);
525 /* free the dynamically allocated opt_output_path */
526 free(opt_output_path
);
528 /* Close thread quit pipes */
529 utils_close_pipe(thread_quit_pipe
);
531 uri_free(control_uri
);
533 /* Live URI is freed in the live thread. */
535 if (tracing_group_name_override
) {
536 free((void *) tracing_group_name
);
541 * Write to writable pipe used to notify a thread.
543 static int notify_thread_pipe(int wpipe
)
547 ret
= lttng_write(wpipe
, "!", 1);
549 PERROR("write poll pipe");
557 static int notify_health_quit_pipe(int *pipe
)
561 ret
= lttng_write(pipe
[1], "4", 1);
563 PERROR("write relay health quit");
572 * Stop all relayd and relayd-live threads.
574 int lttng_relay_stop_threads(void)
578 /* Stopping all threads */
579 DBG("Terminating all threads");
580 if (notify_thread_pipe(thread_quit_pipe
[1])) {
581 ERR("write error on thread quit pipe");
585 if (notify_health_quit_pipe(health_quit_pipe
)) {
586 ERR("write error on health quit pipe");
589 /* Dispatch thread */
590 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
591 futex_nto1_wake(&relay_conn_queue
.futex
);
593 if (relayd_live_stop()) {
594 ERR("Error stopping live threads");
601 * Signal handler for the daemon
603 * Simply stop all worker threads, leaving main() return gracefully after
604 * joining all threads and calling cleanup().
606 static void sighandler(int sig
)
610 DBG("SIGINT caught");
611 if (lttng_relay_stop_threads()) {
612 ERR("Error stopping threads");
616 DBG("SIGTERM caught");
617 if (lttng_relay_stop_threads()) {
618 ERR("Error stopping threads");
622 CMM_STORE_SHARED(recv_child_signal
, 1);
630 * Setup signal handler for :
631 * SIGINT, SIGTERM, SIGPIPE
633 static int set_signal_handler(void)
639 if ((ret
= sigemptyset(&sigset
)) < 0) {
640 PERROR("sigemptyset");
647 sa
.sa_handler
= sighandler
;
648 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
653 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
658 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
663 sa
.sa_handler
= SIG_IGN
;
664 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
669 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
674 void lttng_relay_notify_ready(void)
676 /* Notify the parent of the fork() process that we are ready. */
677 if (opt_daemon
|| opt_background
) {
678 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
679 kill(child_ppid
, SIGUSR1
);
685 * Init thread quit pipe.
687 * Return -1 on error or 0 if all pipes are created.
689 static int init_thread_quit_pipe(void)
693 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
699 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
701 static int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
705 if (events
== NULL
|| size
== 0) {
710 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
716 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
728 * Check if the thread quit pipe was triggered.
730 * Return 1 if it was triggered else 0;
732 static int check_thread_quit_pipe(int fd
, uint32_t events
)
734 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
742 * Create and init socket from uri.
744 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
)
747 struct lttcomm_sock
*sock
= NULL
;
749 sock
= lttcomm_alloc_sock_from_uri(uri
);
751 ERR("Allocating socket");
755 ret
= lttcomm_create_sock(sock
);
759 DBG("Listening on sock %d", sock
->fd
);
761 ret
= sock
->ops
->bind(sock
);
763 PERROR("Failed to bind socket");
767 ret
= sock
->ops
->listen(sock
, -1);
777 lttcomm_destroy_sock(sock
);
783 * This thread manages the listening for new connections on the network
785 static void *relay_thread_listener(void *data
)
787 int i
, ret
, pollfd
, err
= -1;
788 uint32_t revents
, nb_fd
;
789 struct lttng_poll_event events
;
790 struct lttcomm_sock
*control_sock
, *data_sock
;
792 DBG("[thread] Relay listener started");
794 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
796 health_code_update();
798 control_sock
= relay_socket_create(control_uri
);
800 goto error_sock_control
;
803 data_sock
= relay_socket_create(data_uri
);
805 goto error_sock_relay
;
809 * Pass 3 as size here for the thread quit pipe, control and
812 ret
= create_thread_poll_set(&events
, 3);
814 goto error_create_poll
;
817 /* Add the control socket */
818 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
823 /* Add the data socket */
824 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
829 lttng_relay_notify_ready();
831 if (testpoint(relayd_thread_listener
)) {
832 goto error_testpoint
;
836 health_code_update();
838 DBG("Listener accepting connections");
842 ret
= lttng_poll_wait(&events
, -1);
846 * Restart interrupted system call.
848 if (errno
== EINTR
) {
856 DBG("Relay new connection received");
857 for (i
= 0; i
< nb_fd
; i
++) {
858 health_code_update();
860 /* Fetch once the poll data */
861 revents
= LTTNG_POLL_GETEV(&events
, i
);
862 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
864 /* Thread quit pipe has been closed. Killing thread. */
865 ret
= check_thread_quit_pipe(pollfd
, revents
);
871 if (revents
& LPOLLIN
) {
873 * A new connection is requested, therefore a
874 * sessiond/consumerd connection is allocated in
875 * this thread, enqueued to a global queue and
876 * dequeued (and freed) in the worker thread.
879 struct relay_connection
*new_conn
;
880 struct lttcomm_sock
*newsock
;
881 enum connection_type type
;
883 if (pollfd
== data_sock
->fd
) {
885 newsock
= data_sock
->ops
->accept(data_sock
);
886 DBG("Relay data connection accepted, socket %d",
889 assert(pollfd
== control_sock
->fd
);
890 type
= RELAY_CONTROL
;
891 newsock
= control_sock
->ops
->accept(control_sock
);
892 DBG("Relay control connection accepted, socket %d",
896 PERROR("accepting sock");
900 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
903 PERROR("setsockopt inet");
904 lttcomm_destroy_sock(newsock
);
908 ret
= socket_apply_keep_alive_config(newsock
->fd
);
910 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
912 lttcomm_destroy_sock(newsock
);
916 new_conn
= connection_create(newsock
, type
);
918 lttcomm_destroy_sock(newsock
);
922 /* Enqueue request for the dispatcher thread. */
923 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
927 * Wake the dispatch queue futex.
928 * Implicit memory barrier with the
929 * exchange in cds_wfcq_enqueue.
931 futex_nto1_wake(&relay_conn_queue
.futex
);
932 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
933 ERR("socket poll error");
936 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
946 lttng_poll_clean(&events
);
948 if (data_sock
->fd
>= 0) {
949 ret
= data_sock
->ops
->close(data_sock
);
954 lttcomm_destroy_sock(data_sock
);
956 if (control_sock
->fd
>= 0) {
957 ret
= control_sock
->ops
->close(control_sock
);
962 lttcomm_destroy_sock(control_sock
);
966 ERR("Health error occurred in %s", __func__
);
968 health_unregister(health_relayd
);
969 DBG("Relay listener thread cleanup complete");
970 lttng_relay_stop_threads();
975 * This thread manages the dispatching of the requests to worker threads
977 static void *relay_thread_dispatcher(void *data
)
981 struct cds_wfcq_node
*node
;
982 struct relay_connection
*new_conn
= NULL
;
984 DBG("[thread] Relay dispatcher started");
986 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
988 if (testpoint(relayd_thread_dispatcher
)) {
989 goto error_testpoint
;
992 health_code_update();
995 health_code_update();
997 /* Atomically prepare the queue futex */
998 futex_nto1_prepare(&relay_conn_queue
.futex
);
1000 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1005 health_code_update();
1007 /* Dequeue commands */
1008 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1009 &relay_conn_queue
.tail
);
1011 DBG("Woken up but nothing in the relay command queue");
1012 /* Continue thread execution */
1015 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1017 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1020 * Inform worker thread of the new request. This
1021 * call is blocking so we can be assured that
1022 * the data will be read at some point in time
1023 * or wait to the end of the world :)
1025 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1027 PERROR("write connection pipe");
1028 connection_put(new_conn
);
1031 } while (node
!= NULL
);
1033 /* Futex wait on queue. Blocking call on futex() */
1034 health_poll_entry();
1035 futex_nto1_wait(&relay_conn_queue
.futex
);
1039 /* Normal exit, no error */
1046 ERR("Health error occurred in %s", __func__
);
1048 health_unregister(health_relayd
);
1049 DBG("Dispatch thread dying");
1050 lttng_relay_stop_threads();
1054 static bool session_streams_have_index(const struct relay_session
*session
)
1056 return session
->minor
>= 4 && !session
->snapshot
;
1060 * Handle the RELAYD_CREATE_SESSION command.
1062 * On success, send back the session id or else return a negative value.
1064 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1065 struct relay_connection
*conn
,
1066 const struct lttng_buffer_view
*payload
)
1070 struct relay_session
*session
= NULL
;
1071 struct lttcomm_relayd_status_session reply
= {};
1072 char session_name
[LTTNG_NAME_MAX
] = {};
1073 char hostname
[LTTNG_HOST_NAME_MAX
] = {};
1074 uint32_t live_timer
= 0;
1075 bool snapshot
= false;
1076 /* Left nil for peers < 2.11. */
1077 char base_path
[LTTNG_PATH_MAX
] = {};
1078 lttng_uuid sessiond_uuid
= {};
1079 LTTNG_OPTIONAL(uint64_t) id_sessiond
= {};
1080 LTTNG_OPTIONAL(uint64_t) current_chunk_id
= {};
1081 LTTNG_OPTIONAL(time_t) creation_time
= {};
1083 if (conn
->minor
< 4) {
1084 /* From 2.1 to 2.3 */
1086 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1087 /* From 2.4 to 2.10 */
1088 ret
= cmd_create_session_2_4(payload
, session_name
,
1089 hostname
, &live_timer
, &snapshot
);
1091 bool has_current_chunk
;
1092 uint64_t current_chunk_id_value
;
1093 time_t creation_time_value
;
1094 uint64_t id_sessiond_value
;
1096 /* From 2.11 to ... */
1097 ret
= cmd_create_session_2_11(payload
, session_name
, hostname
,
1098 base_path
, &live_timer
, &snapshot
, &id_sessiond_value
,
1099 sessiond_uuid
, &has_current_chunk
,
1100 ¤t_chunk_id_value
, &creation_time_value
);
1101 if (lttng_uuid_is_nil(sessiond_uuid
)) {
1102 /* The nil UUID is reserved for pre-2.11 clients. */
1103 ERR("Illegal nil UUID announced by peer in create session command");
1107 LTTNG_OPTIONAL_SET(&id_sessiond
, id_sessiond_value
);
1108 LTTNG_OPTIONAL_SET(&creation_time
, creation_time_value
);
1109 if (has_current_chunk
) {
1110 LTTNG_OPTIONAL_SET(¤t_chunk_id
,
1111 current_chunk_id_value
);
1119 session
= session_create(session_name
, hostname
, base_path
, live_timer
,
1120 snapshot
, sessiond_uuid
,
1121 id_sessiond
.is_set
? &id_sessiond
.value
: NULL
,
1122 current_chunk_id
.is_set
? ¤t_chunk_id
.value
: NULL
,
1123 creation_time
.is_set
? &creation_time
.value
: NULL
,
1124 conn
->major
, conn
->minor
);
1129 assert(!conn
->session
);
1130 conn
->session
= session
;
1131 DBG("Created session %" PRIu64
, session
->id
);
1133 reply
.session_id
= htobe64(session
->id
);
1137 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1139 reply
.ret_code
= htobe32(LTTNG_OK
);
1142 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1143 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1144 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1148 if (ret
< 0 && session
) {
1149 session_put(session
);
1155 * When we have received all the streams and the metadata for a channel,
1156 * we make them visible to the viewer threads.
1158 static void publish_connection_local_streams(struct relay_connection
*conn
)
1160 struct relay_stream
*stream
;
1161 struct relay_session
*session
= conn
->session
;
1164 * We publish all streams belonging to a session atomically wrt
1167 pthread_mutex_lock(&session
->lock
);
1169 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1171 stream_publish(stream
);
1176 * Inform the viewer that there are new streams in the session.
1178 if (session
->viewer_attached
) {
1179 uatomic_set(&session
->new_streams
, 1);
1181 pthread_mutex_unlock(&session
->lock
);
1184 static int conform_channel_path(char *channel_path
)
1188 if (strstr("../", channel_path
)) {
1189 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1195 if (*channel_path
== '/') {
1196 const size_t len
= strlen(channel_path
);
1199 * Channel paths from peers prior to 2.11 are expressed as an
1200 * absolute path that is, in reality, relative to the relay
1201 * daemon's output directory. Remove the leading slash so it
1202 * is correctly interpreted as a relative path later on.
1204 * len (and not len - 1) is used to copy the trailing NULL.
1206 bcopy(channel_path
+ 1, channel_path
, len
);
1213 * relay_add_stream: allocate a new stream for a session
1215 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1216 struct relay_connection
*conn
,
1217 const struct lttng_buffer_view
*payload
)
1221 struct relay_session
*session
= conn
->session
;
1222 struct relay_stream
*stream
= NULL
;
1223 struct lttcomm_relayd_status_stream reply
;
1224 struct ctf_trace
*trace
= NULL
;
1225 uint64_t stream_handle
= -1ULL;
1226 char *path_name
= NULL
, *channel_name
= NULL
;
1227 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1228 LTTNG_OPTIONAL(uint64_t) stream_chunk_id
= {};
1230 if (!session
|| !conn
->version_check_done
) {
1231 ERR("Trying to add a stream before version check");
1233 goto end_no_session
;
1236 if (session
->minor
== 1) {
1238 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1240 } else if (session
->minor
> 1 && session
->minor
< 11) {
1241 /* From 2.2 to 2.10 */
1242 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1243 &channel_name
, &tracefile_size
, &tracefile_count
);
1245 /* From 2.11 to ... */
1246 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1247 &channel_name
, &tracefile_size
, &tracefile_count
,
1248 &stream_chunk_id
.value
);
1249 stream_chunk_id
.is_set
= true;
1256 if (conform_channel_path(path_name
)) {
1260 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1264 /* This stream here has one reference on the trace. */
1266 pthread_mutex_lock(&last_relay_stream_id_lock
);
1267 stream_handle
= ++last_relay_stream_id
;
1268 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1270 /* We pass ownership of path_name and channel_name. */
1271 stream
= stream_create(trace
, stream_handle
, path_name
,
1272 channel_name
, tracefile_size
, tracefile_count
);
1274 channel_name
= NULL
;
1277 * Streams are the owners of their trace. Reference to trace is
1278 * kept within stream_create().
1280 ctf_trace_put(trace
);
1283 memset(&reply
, 0, sizeof(reply
));
1284 reply
.handle
= htobe64(stream_handle
);
1286 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1288 reply
.ret_code
= htobe32(LTTNG_OK
);
1291 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1292 sizeof(struct lttcomm_relayd_status_stream
), 0);
1293 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1294 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1306 * relay_close_stream: close a specific stream
1308 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1309 struct relay_connection
*conn
,
1310 const struct lttng_buffer_view
*payload
)
1314 struct relay_session
*session
= conn
->session
;
1315 struct lttcomm_relayd_close_stream stream_info
;
1316 struct lttcomm_relayd_generic_reply reply
;
1317 struct relay_stream
*stream
;
1319 DBG("Close stream received");
1321 if (!session
|| !conn
->version_check_done
) {
1322 ERR("Trying to close a stream before version check");
1324 goto end_no_session
;
1327 if (payload
->size
< sizeof(stream_info
)) {
1328 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1329 sizeof(stream_info
), payload
->size
);
1331 goto end_no_session
;
1333 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1334 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1335 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1337 stream
= stream_get_by_id(stream_info
.stream_id
);
1344 * Set last_net_seq_num before the close flag. Required by data
1347 pthread_mutex_lock(&stream
->lock
);
1348 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1349 pthread_mutex_unlock(&stream
->lock
);
1352 * This is one of the conditions which may trigger a stream close
1353 * with the others being:
1354 * 1) A close command is received for a stream
1355 * 2) The control connection owning the stream is closed
1356 * 3) We have received all of the stream's data _after_ a close
1359 try_stream_close(stream
);
1360 if (stream
->is_metadata
) {
1361 struct relay_viewer_stream
*vstream
;
1363 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1365 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1367 * Since all the metadata has been sent to the
1368 * viewer and that we have a request to close
1369 * its stream, we can safely teardown the
1370 * corresponding metadata viewer stream.
1372 viewer_stream_put(vstream
);
1374 /* Put local reference. */
1375 viewer_stream_put(vstream
);
1382 memset(&reply
, 0, sizeof(reply
));
1384 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1386 reply
.ret_code
= htobe32(LTTNG_OK
);
1388 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1389 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1390 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1391 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1401 * relay_reset_metadata: reset a metadata stream
1404 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1405 struct relay_connection
*conn
,
1406 const struct lttng_buffer_view
*payload
)
1410 struct relay_session
*session
= conn
->session
;
1411 struct lttcomm_relayd_reset_metadata stream_info
;
1412 struct lttcomm_relayd_generic_reply reply
;
1413 struct relay_stream
*stream
;
1415 DBG("Reset metadata received");
1417 if (!session
|| !conn
->version_check_done
) {
1418 ERR("Trying to reset a metadata stream before version check");
1420 goto end_no_session
;
1423 if (payload
->size
< sizeof(stream_info
)) {
1424 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1425 sizeof(stream_info
), payload
->size
);
1427 goto end_no_session
;
1429 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1430 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1431 stream_info
.version
= be64toh(stream_info
.version
);
1433 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1435 /* Unsupported for live sessions for now. */
1436 if (session
->live_timer
!= 0) {
1441 stream
= stream_get_by_id(stream_info
.stream_id
);
1446 pthread_mutex_lock(&stream
->lock
);
1447 if (!stream
->is_metadata
) {
1452 ret
= stream_reset_file(stream
);
1454 ERR("Failed to reset metadata stream %" PRIu64
1455 ": stream_path = %s, channel = %s",
1456 stream
->stream_handle
, stream
->path_name
,
1457 stream
->channel_name
);
1461 pthread_mutex_unlock(&stream
->lock
);
1465 memset(&reply
, 0, sizeof(reply
));
1467 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1469 reply
.ret_code
= htobe32(LTTNG_OK
);
1471 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1472 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1473 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1474 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1484 * relay_unknown_command: send -1 if received unknown command
1486 static void relay_unknown_command(struct relay_connection
*conn
)
1488 struct lttcomm_relayd_generic_reply reply
;
1491 memset(&reply
, 0, sizeof(reply
));
1492 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1493 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1494 if (send_ret
< sizeof(reply
)) {
1495 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1500 * relay_start: send an acknowledgment to the client to tell if we are
1501 * ready to receive data. We are ready if a session is established.
1503 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1504 struct relay_connection
*conn
,
1505 const struct lttng_buffer_view
*payload
)
1509 struct lttcomm_relayd_generic_reply reply
;
1510 struct relay_session
*session
= conn
->session
;
1513 DBG("Trying to start the streaming without a session established");
1514 ret
= htobe32(LTTNG_ERR_UNK
);
1517 memset(&reply
, 0, sizeof(reply
));
1518 reply
.ret_code
= htobe32(LTTNG_OK
);
1519 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1521 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1522 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1531 * relay_recv_metadata: receive the metadata for the session.
1533 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1534 struct relay_connection
*conn
,
1535 const struct lttng_buffer_view
*payload
)
1538 struct relay_session
*session
= conn
->session
;
1539 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1540 struct relay_stream
*metadata_stream
;
1541 uint64_t metadata_payload_size
;
1542 struct lttng_buffer_view packet_view
;
1545 ERR("Metadata sent before version check");
1550 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1551 ERR("Incorrect data size");
1555 metadata_payload_size
= recv_hdr
->data_size
-
1556 sizeof(struct lttcomm_relayd_metadata_payload
);
1558 memcpy(&metadata_payload_header
, payload
->data
,
1559 sizeof(metadata_payload_header
));
1560 metadata_payload_header
.stream_id
= be64toh(
1561 metadata_payload_header
.stream_id
);
1562 metadata_payload_header
.padding_size
= be32toh(
1563 metadata_payload_header
.padding_size
);
1565 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1566 if (!metadata_stream
) {
1571 packet_view
= lttng_buffer_view_from_view(payload
,
1572 sizeof(metadata_payload_header
), metadata_payload_size
);
1573 if (!packet_view
.data
) {
1574 ERR("Invalid metadata packet length announced by header");
1579 pthread_mutex_lock(&metadata_stream
->lock
);
1580 ret
= stream_write(metadata_stream
, &packet_view
,
1581 metadata_payload_header
.padding_size
);
1582 pthread_mutex_unlock(&metadata_stream
->lock
);
1588 stream_put(metadata_stream
);
1594 * relay_send_version: send relayd version number
1596 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1597 struct relay_connection
*conn
,
1598 const struct lttng_buffer_view
*payload
)
1602 struct lttcomm_relayd_version reply
, msg
;
1603 bool compatible
= true;
1605 conn
->version_check_done
= true;
1607 /* Get version from the other side. */
1608 if (payload
->size
< sizeof(msg
)) {
1609 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1610 sizeof(msg
), payload
->size
);
1615 memcpy(&msg
, payload
->data
, sizeof(msg
));
1616 msg
.major
= be32toh(msg
.major
);
1617 msg
.minor
= be32toh(msg
.minor
);
1619 memset(&reply
, 0, sizeof(reply
));
1620 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1621 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1623 /* Major versions must be the same */
1624 if (reply
.major
!= msg
.major
) {
1625 DBG("Incompatible major versions (%u vs %u), deleting session",
1626 reply
.major
, msg
.major
);
1630 conn
->major
= reply
.major
;
1631 /* We adapt to the lowest compatible version */
1632 if (reply
.minor
<= msg
.minor
) {
1633 conn
->minor
= reply
.minor
;
1635 conn
->minor
= msg
.minor
;
1638 reply
.major
= htobe32(reply
.major
);
1639 reply
.minor
= htobe32(reply
.minor
);
1640 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1642 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1643 ERR("Failed to send \"send version\" command reply (ret = %zd)",
1656 DBG("Version check done using protocol %u.%u", conn
->major
,
1664 * Check for data pending for a given stream id from the session daemon.
1666 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1667 struct relay_connection
*conn
,
1668 const struct lttng_buffer_view
*payload
)
1670 struct relay_session
*session
= conn
->session
;
1671 struct lttcomm_relayd_data_pending msg
;
1672 struct lttcomm_relayd_generic_reply reply
;
1673 struct relay_stream
*stream
;
1676 uint64_t stream_seq
;
1678 DBG("Data pending command received");
1680 if (!session
|| !conn
->version_check_done
) {
1681 ERR("Trying to check for data before version check");
1683 goto end_no_session
;
1686 if (payload
->size
< sizeof(msg
)) {
1687 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
1688 sizeof(msg
), payload
->size
);
1690 goto end_no_session
;
1692 memcpy(&msg
, payload
->data
, sizeof(msg
));
1693 msg
.stream_id
= be64toh(msg
.stream_id
);
1694 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1696 stream
= stream_get_by_id(msg
.stream_id
);
1697 if (stream
== NULL
) {
1702 pthread_mutex_lock(&stream
->lock
);
1704 if (session_streams_have_index(session
)) {
1706 * Ensure that both the index and stream data have been
1707 * flushed up to the requested point.
1709 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
1711 stream_seq
= stream
->prev_data_seq
;
1713 DBG("Data pending for stream id %" PRIu64
": prev_data_seq %" PRIu64
1714 ", prev_index_seq %" PRIu64
1715 ", and last_seq %" PRIu64
, msg
.stream_id
,
1716 stream
->prev_data_seq
, stream
->prev_index_seq
,
1717 msg
.last_net_seq_num
);
1719 /* Avoid wrapping issue */
1720 if (((int64_t) (stream_seq
- msg
.last_net_seq_num
)) >= 0) {
1721 /* Data has in fact been written and is NOT pending */
1724 /* Data still being streamed thus pending */
1728 stream
->data_pending_check_done
= true;
1729 pthread_mutex_unlock(&stream
->lock
);
1734 memset(&reply
, 0, sizeof(reply
));
1735 reply
.ret_code
= htobe32(ret
);
1736 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1737 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1738 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
1748 * Wait for the control socket to reach a quiescent state.
1750 * Note that for now, when receiving this command from the session
1751 * daemon, this means that every subsequent commands or data received on
1752 * the control socket has been handled. So, this is why we simply return
1755 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
1756 struct relay_connection
*conn
,
1757 const struct lttng_buffer_view
*payload
)
1761 struct relay_stream
*stream
;
1762 struct lttcomm_relayd_quiescent_control msg
;
1763 struct lttcomm_relayd_generic_reply reply
;
1765 DBG("Checking quiescent state on control socket");
1767 if (!conn
->session
|| !conn
->version_check_done
) {
1768 ERR("Trying to check for data before version check");
1770 goto end_no_session
;
1773 if (payload
->size
< sizeof(msg
)) {
1774 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
1775 sizeof(msg
), payload
->size
);
1777 goto end_no_session
;
1779 memcpy(&msg
, payload
->data
, sizeof(msg
));
1780 msg
.stream_id
= be64toh(msg
.stream_id
);
1782 stream
= stream_get_by_id(msg
.stream_id
);
1786 pthread_mutex_lock(&stream
->lock
);
1787 stream
->data_pending_check_done
= true;
1788 pthread_mutex_unlock(&stream
->lock
);
1790 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
1793 memset(&reply
, 0, sizeof(reply
));
1794 reply
.ret_code
= htobe32(LTTNG_OK
);
1795 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1796 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1797 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
1809 * Initialize a data pending command. This means that a consumer is about
1810 * to ask for data pending for each stream it holds. Simply iterate over
1811 * all streams of a session and set the data_pending_check_done flag.
1813 * This command returns to the client a LTTNG_OK code.
1815 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1816 struct relay_connection
*conn
,
1817 const struct lttng_buffer_view
*payload
)
1821 struct lttng_ht_iter iter
;
1822 struct lttcomm_relayd_begin_data_pending msg
;
1823 struct lttcomm_relayd_generic_reply reply
;
1824 struct relay_stream
*stream
;
1829 DBG("Init streams for data pending");
1831 if (!conn
->session
|| !conn
->version_check_done
) {
1832 ERR("Trying to check for data before version check");
1834 goto end_no_session
;
1837 if (payload
->size
< sizeof(msg
)) {
1838 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
1839 sizeof(msg
), payload
->size
);
1841 goto end_no_session
;
1843 memcpy(&msg
, payload
->data
, sizeof(msg
));
1844 msg
.session_id
= be64toh(msg
.session_id
);
1847 * Iterate over all streams to set the begin data pending flag.
1848 * For now, the streams are indexed by stream handle so we have
1849 * to iterate over all streams to find the one associated with
1850 * the right session_id.
1853 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1855 if (!stream_get(stream
)) {
1858 if (stream
->trace
->session
->id
== msg
.session_id
) {
1859 pthread_mutex_lock(&stream
->lock
);
1860 stream
->data_pending_check_done
= false;
1861 pthread_mutex_unlock(&stream
->lock
);
1862 DBG("Set begin data pending flag to stream %" PRIu64
,
1863 stream
->stream_handle
);
1869 memset(&reply
, 0, sizeof(reply
));
1870 /* All good, send back reply. */
1871 reply
.ret_code
= htobe32(LTTNG_OK
);
1873 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1874 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1875 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
1887 * End data pending command. This will check, for a given session id, if
1888 * each stream associated with it has its data_pending_check_done flag
1889 * set. If not, this means that the client lost track of the stream but
1890 * the data is still being streamed on our side. In this case, we inform
1891 * the client that data is in flight.
1893 * Return to the client if there is data in flight or not with a ret_code.
1895 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1896 struct relay_connection
*conn
,
1897 const struct lttng_buffer_view
*payload
)
1901 struct lttng_ht_iter iter
;
1902 struct lttcomm_relayd_end_data_pending msg
;
1903 struct lttcomm_relayd_generic_reply reply
;
1904 struct relay_stream
*stream
;
1905 uint32_t is_data_inflight
= 0;
1907 DBG("End data pending command");
1909 if (!conn
->session
|| !conn
->version_check_done
) {
1910 ERR("Trying to check for data before version check");
1912 goto end_no_session
;
1915 if (payload
->size
< sizeof(msg
)) {
1916 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
1917 sizeof(msg
), payload
->size
);
1919 goto end_no_session
;
1921 memcpy(&msg
, payload
->data
, sizeof(msg
));
1922 msg
.session_id
= be64toh(msg
.session_id
);
1925 * Iterate over all streams to see if the begin data pending
1929 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1931 if (!stream_get(stream
)) {
1934 if (stream
->trace
->session
->id
!= msg
.session_id
) {
1938 pthread_mutex_lock(&stream
->lock
);
1939 if (!stream
->data_pending_check_done
) {
1940 uint64_t stream_seq
;
1942 if (session_streams_have_index(conn
->session
)) {
1944 * Ensure that both the index and stream data have been
1945 * flushed up to the requested point.
1947 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
1949 stream_seq
= stream
->prev_data_seq
;
1951 if (!stream
->closed
|| !(((int64_t) (stream_seq
- stream
->last_net_seq_num
)) >= 0)) {
1952 is_data_inflight
= 1;
1953 DBG("Data is still in flight for stream %" PRIu64
,
1954 stream
->stream_handle
);
1955 pthread_mutex_unlock(&stream
->lock
);
1960 pthread_mutex_unlock(&stream
->lock
);
1965 memset(&reply
, 0, sizeof(reply
));
1966 /* All good, send back reply. */
1967 reply
.ret_code
= htobe32(is_data_inflight
);
1969 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1970 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1971 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
1983 * Receive an index for a specific stream.
1985 * Return 0 on success else a negative value.
1987 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
1988 struct relay_connection
*conn
,
1989 const struct lttng_buffer_view
*payload
)
1993 struct relay_session
*session
= conn
->session
;
1994 struct lttcomm_relayd_index index_info
;
1995 struct lttcomm_relayd_generic_reply reply
;
1996 struct relay_stream
*stream
;
2001 DBG("Relay receiving index");
2003 if (!session
|| !conn
->version_check_done
) {
2004 ERR("Trying to close a stream before version check");
2006 goto end_no_session
;
2009 msg_len
= lttcomm_relayd_index_len(
2010 lttng_to_index_major(conn
->major
, conn
->minor
),
2011 lttng_to_index_minor(conn
->major
, conn
->minor
));
2012 if (payload
->size
< msg_len
) {
2013 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2014 msg_len
, payload
->size
);
2016 goto end_no_session
;
2018 memcpy(&index_info
, payload
->data
, msg_len
);
2019 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2020 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2021 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2022 index_info
.content_size
= be64toh(index_info
.content_size
);
2023 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2024 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2025 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2026 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2028 if (conn
->minor
>= 8) {
2029 index_info
.stream_instance_id
=
2030 be64toh(index_info
.stream_instance_id
);
2031 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2034 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2036 ERR("stream_get_by_id not found");
2041 pthread_mutex_lock(&stream
->lock
);
2042 ret
= stream_add_index(stream
, &index_info
);
2043 pthread_mutex_unlock(&stream
->lock
);
2045 goto end_stream_put
;
2051 memset(&reply
, 0, sizeof(reply
));
2053 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2055 reply
.ret_code
= htobe32(LTTNG_OK
);
2057 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2058 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2059 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2068 * Receive the streams_sent message.
2070 * Return 0 on success else a negative value.
2072 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2073 struct relay_connection
*conn
,
2074 const struct lttng_buffer_view
*payload
)
2078 struct lttcomm_relayd_generic_reply reply
;
2082 DBG("Relay receiving streams_sent");
2084 if (!conn
->session
|| !conn
->version_check_done
) {
2085 ERR("Trying to close a stream before version check");
2087 goto end_no_session
;
2091 * Publish every pending stream in the connection recv list which are
2092 * now ready to be used by the viewer.
2094 publish_connection_local_streams(conn
);
2096 memset(&reply
, 0, sizeof(reply
));
2097 reply
.ret_code
= htobe32(LTTNG_OK
);
2098 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2099 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2100 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2113 * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2114 * session rotation feature (not the tracefile rotation feature).
2116 static int relay_rotate_session_streams(
2117 const struct lttcomm_relayd_hdr
*recv_hdr
,
2118 struct relay_connection
*conn
,
2119 const struct lttng_buffer_view
*payload
)
2124 enum lttng_error_code reply_code
= LTTNG_ERR_UNK
;
2125 struct relay_session
*session
= conn
->session
;
2126 struct lttcomm_relayd_rotate_streams rotate_streams
;
2127 struct lttcomm_relayd_generic_reply reply
= {};
2128 struct relay_stream
*stream
= NULL
;
2129 const size_t header_len
= sizeof(struct lttcomm_relayd_rotate_streams
);
2130 struct lttng_trace_chunk
*next_trace_chunk
= NULL
;
2131 struct lttng_buffer_view stream_positions
;
2132 char chunk_id_buf
[MAX_INT_DEC_LEN(uint64_t)];
2133 const char *chunk_id_str
= "none";
2135 if (!session
|| !conn
->version_check_done
) {
2136 ERR("Trying to rotate a stream before version check");
2141 if (session
->major
== 2 && session
->minor
< 11) {
2142 ERR("Unsupported feature before 2.11");
2147 if (payload
->size
< header_len
) {
2148 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2149 header_len
, payload
->size
);
2154 memcpy(&rotate_streams
, payload
->data
, header_len
);
2156 /* Convert header to host endianness. */
2157 rotate_streams
= (typeof(rotate_streams
)) {
2158 .stream_count
= be32toh(rotate_streams
.stream_count
),
2159 .new_chunk_id
= (typeof(rotate_streams
.new_chunk_id
)) {
2160 .is_set
= !!rotate_streams
.new_chunk_id
.is_set
,
2161 .value
= be64toh(rotate_streams
.new_chunk_id
.value
),
2165 if (rotate_streams
.new_chunk_id
.is_set
) {
2167 * Retrieve the trace chunk the stream must transition to. As
2168 * per the protocol, this chunk should have been created
2169 * before this command is received.
2171 next_trace_chunk
= sessiond_trace_chunk_registry_get_chunk(
2172 sessiond_trace_chunk_registry
,
2173 session
->sessiond_uuid
, session
->id
,
2174 rotate_streams
.new_chunk_id
.value
);
2175 if (!next_trace_chunk
) {
2176 char uuid_str
[UUID_STR_LEN
];
2178 lttng_uuid_to_str(session
->sessiond_uuid
, uuid_str
);
2179 ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2180 ", trace_chunk_id = %" PRIu64
,
2181 uuid_str
, session
->id
,
2182 rotate_streams
.new_chunk_id
.value
);
2183 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2188 ret
= snprintf(chunk_id_buf
, sizeof(chunk_id_buf
), "%" PRIu64
,
2189 rotate_streams
.new_chunk_id
.value
);
2190 if (ret
< 0 || ret
>= sizeof(chunk_id_buf
)) {
2191 chunk_id_str
= "formatting error";
2193 chunk_id_str
= chunk_id_buf
;
2197 DBG("Rotate %" PRIu32
" streams of session \"%s\" to chunk \"%s\"",
2198 rotate_streams
.stream_count
, session
->session_name
,
2201 stream_positions
= lttng_buffer_view_from_view(payload
,
2202 sizeof(rotate_streams
), -1);
2203 if (!stream_positions
.data
||
2204 stream_positions
.size
<
2205 (rotate_streams
.stream_count
*
2206 sizeof(struct lttcomm_relayd_stream_rotation_position
))) {
2207 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2212 for (i
= 0; i
< rotate_streams
.stream_count
; i
++) {
2213 struct lttcomm_relayd_stream_rotation_position
*position_comm
=
2214 &((typeof(position_comm
)) stream_positions
.data
)[i
];
2215 const struct lttcomm_relayd_stream_rotation_position pos
= {
2216 .stream_id
= be64toh(position_comm
->stream_id
),
2217 .rotate_at_seq_num
= be64toh(
2218 position_comm
->rotate_at_seq_num
),
2221 stream
= stream_get_by_id(pos
.stream_id
);
2223 reply_code
= LTTNG_ERR_INVALID
;
2228 pthread_mutex_lock(&stream
->lock
);
2229 ret
= stream_set_pending_rotation(stream
, next_trace_chunk
,
2230 pos
.rotate_at_seq_num
);
2231 pthread_mutex_unlock(&stream
->lock
);
2233 reply_code
= LTTNG_ERR_FILE_CREATION_ERROR
;
2241 reply_code
= LTTNG_OK
;
2247 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2248 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2249 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2250 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2251 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2258 lttng_trace_chunk_put(next_trace_chunk
);
2262 static int init_session_output_directory_handle(struct relay_session
*session
,
2263 struct lttng_directory_handle
*handle
)
2267 * session_directory:
2269 * if base_path is NULL
2270 * hostname/session_name
2272 * hostname/base_path
2274 char *session_directory
= NULL
;
2276 * base path + session_directory
2277 * e.g. /home/user/lttng-traces/hostname/session_name
2279 char *full_session_path
= NULL
;
2282 * If base path is set, it overrides the session name for the
2283 * session relative base path. No timestamp is appended if the
2284 * base path is overridden.
2286 if (session
->base_path
[0] == '\0') {
2287 char creation_time_str
[16];
2288 struct tm
*timeinfo
;
2290 assert(session
->creation_time
.is_set
);
2291 timeinfo
= localtime(&session
->creation_time
.value
);
2296 strftime(creation_time_str
, sizeof(creation_time_str
), "%Y%m%d-%H%M%S",
2299 pthread_mutex_lock(&session
->lock
);
2300 ret
= asprintf(&session_directory
, "%s/%s-%s", session
->hostname
,
2301 session
->session_name
, creation_time_str
);
2302 pthread_mutex_unlock(&session
->lock
);
2304 pthread_mutex_lock(&session
->lock
);
2305 ret
= asprintf(&session_directory
, "%s/%s", session
->hostname
,
2306 session
->base_path
);
2307 pthread_mutex_unlock(&session
->lock
);
2310 PERROR("Failed to format session directory name");
2314 full_session_path
= create_output_path(session_directory
);
2315 if (!full_session_path
) {
2320 ret
= utils_mkdir_recursive(
2321 full_session_path
, S_IRWXU
| S_IRWXG
, -1, -1);
2323 ERR("Failed to create session output path \"%s\"",
2328 ret
= lttng_directory_handle_init(handle
, full_session_path
);
2333 free(session_directory
);
2334 free(full_session_path
);
2339 * relay_create_trace_chunk: create a new trace chunk
2341 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2342 struct relay_connection
*conn
,
2343 const struct lttng_buffer_view
*payload
)
2347 struct relay_session
*session
= conn
->session
;
2348 struct lttcomm_relayd_create_trace_chunk
*msg
;
2349 struct lttcomm_relayd_generic_reply reply
= {};
2350 struct lttng_buffer_view header_view
;
2351 struct lttng_buffer_view chunk_name_view
;
2352 struct lttng_trace_chunk
*chunk
= NULL
, *published_chunk
= NULL
;
2353 enum lttng_error_code reply_code
= LTTNG_OK
;
2354 enum lttng_trace_chunk_status chunk_status
;
2355 struct lttng_directory_handle session_output
;
2357 if (!session
|| !conn
->version_check_done
) {
2358 ERR("Trying to create a trace chunk before version check");
2363 if (session
->major
== 2 && session
->minor
< 11) {
2364 ERR("Chunk creation command is unsupported before 2.11");
2369 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2370 if (!header_view
.data
) {
2371 ERR("Failed to receive payload of chunk creation command");
2376 /* Convert to host endianness. */
2377 msg
= (typeof(msg
)) header_view
.data
;
2378 msg
->chunk_id
= be64toh(msg
->chunk_id
);
2379 msg
->creation_timestamp
= be64toh(msg
->creation_timestamp
);
2380 msg
->override_name_length
= be32toh(msg
->override_name_length
);
2382 chunk
= lttng_trace_chunk_create(
2383 msg
->chunk_id
, msg
->creation_timestamp
);
2385 ERR("Failed to create trace chunk in trace chunk creation command");
2387 reply_code
= LTTNG_ERR_NOMEM
;
2391 if (msg
->override_name_length
) {
2394 chunk_name_view
= lttng_buffer_view_from_view(payload
,
2396 msg
->override_name_length
);
2397 name
= chunk_name_view
.data
;
2398 if (!name
|| name
[msg
->override_name_length
- 1]) {
2399 ERR("Failed to receive payload of chunk creation command");
2401 reply_code
= LTTNG_ERR_INVALID
;
2405 chunk_status
= lttng_trace_chunk_override_name(
2406 chunk
, chunk_name_view
.data
);
2407 switch (chunk_status
) {
2408 case LTTNG_TRACE_CHUNK_STATUS_OK
:
2410 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT
:
2411 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2412 reply_code
= LTTNG_ERR_INVALID
;
2416 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2417 reply_code
= LTTNG_ERR_UNK
;
2423 ret
= init_session_output_directory_handle(
2424 conn
->session
, &session_output
);
2426 reply_code
= LTTNG_ERR_CREATE_DIR_FAIL
;
2430 chunk_status
= lttng_trace_chunk_set_credentials_current_user(chunk
);
2431 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2432 reply_code
= LTTNG_ERR_UNK
;
2437 chunk_status
= lttng_trace_chunk_set_as_owner(chunk
, &session_output
);
2438 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2439 reply_code
= LTTNG_ERR_UNK
;
2444 published_chunk
= sessiond_trace_chunk_registry_publish_chunk(
2445 sessiond_trace_chunk_registry
,
2446 conn
->session
->sessiond_uuid
,
2449 if (!published_chunk
) {
2450 char uuid_str
[UUID_STR_LEN
];
2452 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2453 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2458 reply_code
= LTTNG_ERR_NOMEM
;
2462 pthread_mutex_lock(&conn
->session
->lock
);
2463 if (conn
->session
->pending_closure_trace_chunk
) {
2465 * Invalid; this means a second create_trace_chunk command was
2466 * received before a close_trace_chunk.
2468 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2469 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2471 goto end_unlock_session
;
2473 conn
->session
->pending_closure_trace_chunk
=
2474 conn
->session
->current_trace_chunk
;
2475 conn
->session
->current_trace_chunk
= published_chunk
;
2476 published_chunk
= NULL
;
2478 pthread_mutex_unlock(&conn
->session
->lock
);
2480 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2481 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2483 sizeof(struct lttcomm_relayd_generic_reply
),
2485 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2486 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2491 lttng_trace_chunk_put(chunk
);
2492 lttng_trace_chunk_put(published_chunk
);
2493 lttng_directory_handle_fini(&session_output
);
2498 * relay_close_trace_chunk: close a trace chunk
2500 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2501 struct relay_connection
*conn
,
2502 const struct lttng_buffer_view
*payload
)
2506 struct relay_session
*session
= conn
->session
;
2507 struct lttcomm_relayd_close_trace_chunk
*msg
;
2508 struct lttcomm_relayd_generic_reply reply
= {};
2509 struct lttng_buffer_view header_view
;
2510 struct lttng_trace_chunk
*chunk
= NULL
;
2511 enum lttng_error_code reply_code
= LTTNG_OK
;
2512 enum lttng_trace_chunk_status chunk_status
;
2514 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type
) close_command
= {};
2515 time_t close_timestamp
;
2517 if (!session
|| !conn
->version_check_done
) {
2518 ERR("Trying to close a trace chunk before version check");
2523 if (session
->major
== 2 && session
->minor
< 11) {
2524 ERR("Chunk close command is unsupported before 2.11");
2529 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2530 if (!header_view
.data
) {
2531 ERR("Failed to receive payload of chunk close command");
2536 /* Convert to host endianness. */
2537 msg
= (typeof(msg
)) header_view
.data
;
2538 chunk_id
= be64toh(msg
->chunk_id
);
2539 close_timestamp
= (time_t) be64toh(msg
->close_timestamp
);
2540 close_command
= (typeof(close_command
)){
2541 .value
= be32toh(msg
->close_command
.value
),
2542 .is_set
= msg
->close_command
.is_set
,
2545 chunk
= sessiond_trace_chunk_registry_get_chunk(
2546 sessiond_trace_chunk_registry
,
2547 conn
->session
->sessiond_uuid
,
2551 char uuid_str
[UUID_STR_LEN
];
2553 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2554 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2559 reply_code
= LTTNG_ERR_NOMEM
;
2563 pthread_mutex_lock(&session
->lock
);
2564 if (session
->pending_closure_trace_chunk
&&
2565 session
->pending_closure_trace_chunk
!= chunk
) {
2566 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
2567 session
->session_name
);
2568 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2570 goto end_unlock_session
;
2573 chunk_status
= lttng_trace_chunk_set_close_timestamp(
2574 chunk
, close_timestamp
);
2575 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2576 ERR("Failed to set trace chunk close timestamp");
2578 reply_code
= LTTNG_ERR_UNK
;
2579 goto end_unlock_session
;
2582 if (close_command
.is_set
) {
2583 chunk_status
= lttng_trace_chunk_set_close_command(
2584 chunk
, close_command
.value
);
2585 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2587 reply_code
= LTTNG_ERR_INVALID
;
2588 goto end_unlock_session
;
2592 if (session
->current_trace_chunk
== chunk
) {
2594 * After a trace chunk close command, no new streams
2595 * referencing the chunk may be created. Hence, on the
2596 * event that no new trace chunk have been created for
2597 * the session, the reference to the current trace chunk
2598 * is released in order to allow it to be reclaimed when
2599 * the last stream releases its reference to it.
2601 lttng_trace_chunk_put(session
->current_trace_chunk
);
2602 session
->current_trace_chunk
= NULL
;
2604 lttng_trace_chunk_put(session
->pending_closure_trace_chunk
);
2605 session
->pending_closure_trace_chunk
= NULL
;
2607 pthread_mutex_unlock(&session
->lock
);
2610 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2611 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2613 sizeof(struct lttcomm_relayd_generic_reply
),
2615 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2616 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2621 lttng_trace_chunk_put(chunk
);
2626 * relay_trace_chunk_exists: check if a trace chunk exists
2628 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr
*recv_hdr
,
2629 struct relay_connection
*conn
,
2630 const struct lttng_buffer_view
*payload
)
2634 struct relay_session
*session
= conn
->session
;
2635 struct lttcomm_relayd_trace_chunk_exists
*msg
;
2636 struct lttcomm_relayd_trace_chunk_exists_reply reply
= {};
2637 struct lttng_buffer_view header_view
;
2638 struct lttng_trace_chunk
*chunk
= NULL
;
2641 if (!session
|| !conn
->version_check_done
) {
2642 ERR("Trying to close a trace chunk before version check");
2647 if (session
->major
== 2 && session
->minor
< 11) {
2648 ERR("Chunk close command is unsupported before 2.11");
2653 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2654 if (!header_view
.data
) {
2655 ERR("Failed to receive payload of chunk close command");
2660 /* Convert to host endianness. */
2661 msg
= (typeof(msg
)) header_view
.data
;
2662 chunk_id
= be64toh(msg
->chunk_id
);
2664 chunk
= sessiond_trace_chunk_registry_get_chunk(
2665 sessiond_trace_chunk_registry
,
2666 conn
->session
->sessiond_uuid
,
2670 reply
= (typeof(reply
)) {
2671 .generic
.ret_code
= htobe32((uint32_t) LTTNG_OK
),
2672 .trace_chunk_exists
= !!chunk
,
2674 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2675 &reply
, sizeof(reply
), 0);
2676 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2677 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2682 lttng_trace_chunk_put(chunk
);
2686 #define DBG_CMD(cmd_name, conn) \
2687 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2689 static int relay_process_control_command(struct relay_connection
*conn
,
2690 const struct lttcomm_relayd_hdr
*header
,
2691 const struct lttng_buffer_view
*payload
)
2695 switch (header
->cmd
) {
2696 case RELAYD_CREATE_SESSION
:
2697 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
2698 ret
= relay_create_session(header
, conn
, payload
);
2700 case RELAYD_ADD_STREAM
:
2701 DBG_CMD("RELAYD_ADD_STREAM", conn
);
2702 ret
= relay_add_stream(header
, conn
, payload
);
2704 case RELAYD_START_DATA
:
2705 DBG_CMD("RELAYD_START_DATA", conn
);
2706 ret
= relay_start(header
, conn
, payload
);
2708 case RELAYD_SEND_METADATA
:
2709 DBG_CMD("RELAYD_SEND_METADATA", conn
);
2710 ret
= relay_recv_metadata(header
, conn
, payload
);
2712 case RELAYD_VERSION
:
2713 DBG_CMD("RELAYD_VERSION", conn
);
2714 ret
= relay_send_version(header
, conn
, payload
);
2716 case RELAYD_CLOSE_STREAM
:
2717 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
2718 ret
= relay_close_stream(header
, conn
, payload
);
2720 case RELAYD_DATA_PENDING
:
2721 DBG_CMD("RELAYD_DATA_PENDING", conn
);
2722 ret
= relay_data_pending(header
, conn
, payload
);
2724 case RELAYD_QUIESCENT_CONTROL
:
2725 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
2726 ret
= relay_quiescent_control(header
, conn
, payload
);
2728 case RELAYD_BEGIN_DATA_PENDING
:
2729 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
2730 ret
= relay_begin_data_pending(header
, conn
, payload
);
2732 case RELAYD_END_DATA_PENDING
:
2733 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
2734 ret
= relay_end_data_pending(header
, conn
, payload
);
2736 case RELAYD_SEND_INDEX
:
2737 DBG_CMD("RELAYD_SEND_INDEX", conn
);
2738 ret
= relay_recv_index(header
, conn
, payload
);
2740 case RELAYD_STREAMS_SENT
:
2741 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
2742 ret
= relay_streams_sent(header
, conn
, payload
);
2744 case RELAYD_RESET_METADATA
:
2745 DBG_CMD("RELAYD_RESET_METADATA", conn
);
2746 ret
= relay_reset_metadata(header
, conn
, payload
);
2748 case RELAYD_ROTATE_STREAMS
:
2749 DBG_CMD("RELAYD_ROTATE_STREAMS", conn
);
2750 ret
= relay_rotate_session_streams(header
, conn
, payload
);
2752 case RELAYD_CREATE_TRACE_CHUNK
:
2753 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn
);
2754 ret
= relay_create_trace_chunk(header
, conn
, payload
);
2756 case RELAYD_CLOSE_TRACE_CHUNK
:
2757 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn
);
2758 ret
= relay_close_trace_chunk(header
, conn
, payload
);
2760 case RELAYD_TRACE_CHUNK_EXISTS
:
2761 DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn
);
2762 ret
= relay_trace_chunk_exists(header
, conn
, payload
);
2764 case RELAYD_UPDATE_SYNC_INFO
:
2766 ERR("Received unknown command (%u)", header
->cmd
);
2767 relay_unknown_command(conn
);
2776 static enum relay_connection_status
relay_process_control_receive_payload(
2777 struct relay_connection
*conn
)
2780 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2781 struct lttng_dynamic_buffer
*reception_buffer
=
2782 &conn
->protocol
.ctrl
.reception_buffer
;
2783 struct ctrl_connection_state_receive_payload
*state
=
2784 &conn
->protocol
.ctrl
.state
.receive_payload
;
2785 struct lttng_buffer_view payload_view
;
2787 if (state
->left_to_receive
== 0) {
2788 /* Short-circuit for payload-less commands. */
2789 goto reception_complete
;
2792 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2793 reception_buffer
->data
+ state
->received
,
2794 state
->left_to_receive
, MSG_DONTWAIT
);
2796 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2797 PERROR("Unable to receive command payload on sock %d",
2799 status
= RELAY_CONNECTION_STATUS_ERROR
;
2802 } else if (ret
== 0) {
2803 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2804 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2809 assert(ret
<= state
->left_to_receive
);
2811 state
->left_to_receive
-= ret
;
2812 state
->received
+= ret
;
2814 if (state
->left_to_receive
> 0) {
2816 * Can't transition to the protocol's next state, wait to
2817 * receive the rest of the header.
2819 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2820 state
->received
, state
->left_to_receive
,
2826 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
2827 conn
->sock
->fd
, state
->received
);
2829 * The payload required to process the command has been received.
2830 * A view to the reception buffer is forwarded to the various
2831 * commands and the state of the control is reset on success.
2833 * Commands are responsible for sending their reply to the peer.
2835 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
2837 ret
= relay_process_control_command(conn
,
2838 &state
->header
, &payload_view
);
2840 status
= RELAY_CONNECTION_STATUS_ERROR
;
2844 ret
= connection_reset_protocol_state(conn
);
2846 status
= RELAY_CONNECTION_STATUS_ERROR
;
2852 static enum relay_connection_status
relay_process_control_receive_header(
2853 struct relay_connection
*conn
)
2856 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2857 struct lttcomm_relayd_hdr header
;
2858 struct lttng_dynamic_buffer
*reception_buffer
=
2859 &conn
->protocol
.ctrl
.reception_buffer
;
2860 struct ctrl_connection_state_receive_header
*state
=
2861 &conn
->protocol
.ctrl
.state
.receive_header
;
2863 assert(state
->left_to_receive
!= 0);
2865 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2866 reception_buffer
->data
+ state
->received
,
2867 state
->left_to_receive
, MSG_DONTWAIT
);
2869 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2870 PERROR("Unable to receive control command header on sock %d",
2872 status
= RELAY_CONNECTION_STATUS_ERROR
;
2875 } else if (ret
== 0) {
2876 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2877 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2882 assert(ret
<= state
->left_to_receive
);
2884 state
->left_to_receive
-= ret
;
2885 state
->received
+= ret
;
2887 if (state
->left_to_receive
> 0) {
2889 * Can't transition to the protocol's next state, wait to
2890 * receive the rest of the header.
2892 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2893 state
->received
, state
->left_to_receive
,
2898 /* Transition to next state: receiving the command's payload. */
2899 conn
->protocol
.ctrl
.state_id
=
2900 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
2901 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
2902 header
.circuit_id
= be64toh(header
.circuit_id
);
2903 header
.data_size
= be64toh(header
.data_size
);
2904 header
.cmd
= be32toh(header
.cmd
);
2905 header
.cmd_version
= be32toh(header
.cmd_version
);
2906 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
2907 &header
, sizeof(header
));
2909 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
2910 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
2913 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
2914 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
2916 status
= RELAY_CONNECTION_STATUS_ERROR
;
2920 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
2922 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
2923 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
2926 status
= RELAY_CONNECTION_STATUS_ERROR
;
2930 if (header
.data_size
== 0) {
2932 * Manually invoke the next state as the poll loop
2933 * will not wake-up to allow us to proceed further.
2935 status
= relay_process_control_receive_payload(conn
);
2942 * Process the commands received on the control socket
2944 static enum relay_connection_status
relay_process_control(
2945 struct relay_connection
*conn
)
2947 enum relay_connection_status status
;
2949 switch (conn
->protocol
.ctrl
.state_id
) {
2950 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
2951 status
= relay_process_control_receive_header(conn
);
2953 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
2954 status
= relay_process_control_receive_payload(conn
);
2957 ERR("Unknown control connection protocol state encountered.");
2964 static enum relay_connection_status
relay_process_data_receive_header(
2965 struct relay_connection
*conn
)
2968 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2969 struct data_connection_state_receive_header
*state
=
2970 &conn
->protocol
.data
.state
.receive_header
;
2971 struct lttcomm_relayd_data_hdr header
;
2972 struct relay_stream
*stream
;
2974 assert(state
->left_to_receive
!= 0);
2976 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2977 state
->header_reception_buffer
+ state
->received
,
2978 state
->left_to_receive
, MSG_DONTWAIT
);
2980 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2981 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2982 status
= RELAY_CONNECTION_STATUS_ERROR
;
2985 } else if (ret
== 0) {
2986 /* Orderly shutdown. Not necessary to print an error. */
2987 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2988 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2993 assert(ret
<= state
->left_to_receive
);
2995 state
->left_to_receive
-= ret
;
2996 state
->received
+= ret
;
2998 if (state
->left_to_receive
> 0) {
3000 * Can't transition to the protocol's next state, wait to
3001 * receive the rest of the header.
3003 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3004 state
->received
, state
->left_to_receive
,
3009 /* Transition to next state: receiving the payload. */
3010 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3012 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3013 header
.circuit_id
= be64toh(header
.circuit_id
);
3014 header
.stream_id
= be64toh(header
.stream_id
);
3015 header
.data_size
= be32toh(header
.data_size
);
3016 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3017 header
.padding_size
= be32toh(header
.padding_size
);
3018 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3020 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3022 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3023 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3025 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3026 conn
->sock
->fd
, header
.circuit_id
,
3027 header
.stream_id
, header
.data_size
,
3028 header
.net_seq_num
, header
.padding_size
);
3030 stream
= stream_get_by_id(header
.stream_id
);
3032 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3034 /* Protocol error. */
3035 status
= RELAY_CONNECTION_STATUS_ERROR
;
3039 pthread_mutex_lock(&stream
->lock
);
3040 /* Prepare stream for the reception of a new packet. */
3041 ret
= stream_init_packet(stream
, header
.data_size
,
3042 &conn
->protocol
.data
.state
.receive_payload
.rotate_index
);
3043 pthread_mutex_unlock(&stream
->lock
);
3045 ERR("Failed to rotate stream output file");
3046 status
= RELAY_CONNECTION_STATUS_ERROR
;
3047 goto end_stream_unlock
;
3056 static enum relay_connection_status
relay_process_data_receive_payload(
3057 struct relay_connection
*conn
)
3060 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3061 struct relay_stream
*stream
;
3062 struct data_connection_state_receive_payload
*state
=
3063 &conn
->protocol
.data
.state
.receive_payload
;
3064 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3065 char data_buffer
[chunk_size
];
3066 bool partial_recv
= false;
3067 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3068 uint64_t left_to_receive
= state
->left_to_receive
;
3069 struct relay_session
*session
;
3071 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3072 state
->header
.stream_id
, state
->header
.net_seq_num
,
3073 state
->received
, left_to_receive
);
3075 stream
= stream_get_by_id(state
->header
.stream_id
);
3077 /* Protocol error. */
3078 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3079 state
->header
.stream_id
);
3080 status
= RELAY_CONNECTION_STATUS_ERROR
;
3084 pthread_mutex_lock(&stream
->lock
);
3085 session
= stream
->trace
->session
;
3086 if (!conn
->session
) {
3087 ret
= connection_set_session(conn
, session
);
3089 status
= RELAY_CONNECTION_STATUS_ERROR
;
3090 goto end_stream_unlock
;
3095 * The size of the "chunk" received on any iteration is bounded by:
3096 * - the data left to receive,
3097 * - the data immediately available on the socket,
3098 * - the on-stack data buffer
3100 while (left_to_receive
> 0 && !partial_recv
) {
3101 size_t recv_size
= min(left_to_receive
, chunk_size
);
3102 struct lttng_buffer_view packet_chunk
;
3104 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3105 recv_size
, MSG_DONTWAIT
);
3107 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3108 PERROR("Socket %d error", conn
->sock
->fd
);
3109 status
= RELAY_CONNECTION_STATUS_ERROR
;
3111 goto end_stream_unlock
;
3112 } else if (ret
== 0) {
3113 /* No more data ready to be consumed on socket. */
3114 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3115 state
->header
.stream_id
);
3116 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3118 } else if (ret
< (int) recv_size
) {
3120 * All the data available on the socket has been
3123 partial_recv
= true;
3127 packet_chunk
= lttng_buffer_view_init(data_buffer
,
3129 assert(packet_chunk
.data
);
3131 ret
= stream_write(stream
, &packet_chunk
, 0);
3133 ERR("Relay error writing data to file");
3134 status
= RELAY_CONNECTION_STATUS_ERROR
;
3135 goto end_stream_unlock
;
3138 left_to_receive
-= recv_size
;
3139 state
->received
+= recv_size
;
3140 state
->left_to_receive
= left_to_receive
;
3143 if (state
->left_to_receive
> 0) {
3145 * Did not receive all the data expected, wait for more data to
3146 * become available on the socket.
3148 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3149 state
->header
.stream_id
, state
->received
,
3150 state
->left_to_receive
);
3151 goto end_stream_unlock
;
3154 ret
= stream_write(stream
, NULL
, state
->header
.padding_size
);
3156 status
= RELAY_CONNECTION_STATUS_ERROR
;
3157 goto end_stream_unlock
;
3160 if (session_streams_have_index(session
)) {
3161 ret
= stream_update_index(stream
, state
->header
.net_seq_num
,
3162 state
->rotate_index
, &index_flushed
,
3163 state
->header
.data_size
+ state
->header
.padding_size
);
3165 ERR("Failed to update index: stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3166 stream
->stream_handle
,
3167 state
->header
.net_seq_num
, ret
);
3168 status
= RELAY_CONNECTION_STATUS_ERROR
;
3169 goto end_stream_unlock
;
3173 if (stream
->prev_data_seq
== -1ULL) {
3177 ret
= stream_complete_packet(stream
, state
->header
.data_size
+
3178 state
->header
.padding_size
, state
->header
.net_seq_num
,
3181 status
= RELAY_CONNECTION_STATUS_ERROR
;
3182 goto end_stream_unlock
;
3186 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3187 * contents of *state which are aliased (union) to the same location as
3188 * the new state. Don't use it beyond this point.
3190 connection_reset_protocol_state(conn
);
3194 close_requested
= stream
->close_requested
;
3195 pthread_mutex_unlock(&stream
->lock
);
3196 if (close_requested
&& left_to_receive
== 0) {
3197 try_stream_close(stream
);
3201 pthread_mutex_lock(&session
->lock
);
3202 uatomic_set(&session
->new_streams
, 1);
3203 pthread_mutex_unlock(&session
->lock
);
3212 * relay_process_data: Process the data received on the data socket
3214 static enum relay_connection_status
relay_process_data(
3215 struct relay_connection
*conn
)
3217 enum relay_connection_status status
;
3219 switch (conn
->protocol
.data
.state_id
) {
3220 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3221 status
= relay_process_data_receive_header(conn
);
3223 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3224 status
= relay_process_data_receive_payload(conn
);
3227 ERR("Unexpected data connection communication state.");
3234 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3238 (void) lttng_poll_del(events
, pollfd
);
3240 ret
= close(pollfd
);
3242 ERR("Closing pollfd %d", pollfd
);
3246 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3247 int pollfd
, struct relay_connection
*conn
)
3249 const char *type_str
;
3251 switch (conn
->type
) {
3256 type_str
= "Control";
3258 case RELAY_VIEWER_COMMAND
:
3259 type_str
= "Viewer Command";
3261 case RELAY_VIEWER_NOTIFICATION
:
3262 type_str
= "Viewer Notification";
3265 type_str
= "Unknown";
3267 cleanup_connection_pollfd(events
, pollfd
);
3268 connection_put(conn
);
3269 DBG("%s connection closed with %d", type_str
, pollfd
);
3273 * This thread does the actual work
3275 static void *relay_thread_worker(void *data
)
3277 int ret
, err
= -1, last_seen_data_fd
= -1;
3279 struct lttng_poll_event events
;
3280 struct lttng_ht
*relay_connections_ht
;
3281 struct lttng_ht_iter iter
;
3282 struct relay_connection
*destroy_conn
= NULL
;
3284 DBG("[thread] Relay worker started");
3286 rcu_register_thread();
3288 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3290 if (testpoint(relayd_thread_worker
)) {
3291 goto error_testpoint
;
3294 health_code_update();
3296 /* table of connections indexed on socket */
3297 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3298 if (!relay_connections_ht
) {
3299 goto relay_connections_ht_error
;
3302 ret
= create_thread_poll_set(&events
, 2);
3304 goto error_poll_create
;
3307 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3314 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3316 health_code_update();
3318 /* Infinite blocking call, waiting for transmission */
3319 DBG3("Relayd worker thread polling...");
3320 health_poll_entry();
3321 ret
= lttng_poll_wait(&events
, -1);
3325 * Restart interrupted system call.
3327 if (errno
== EINTR
) {
3336 * Process control. The control connection is
3337 * prioritized so we don't starve it with high
3338 * throughput tracing data on the data connection.
3340 for (i
= 0; i
< nb_fd
; i
++) {
3341 /* Fetch once the poll data */
3342 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3343 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3345 health_code_update();
3347 /* Thread quit pipe has been closed. Killing thread. */
3348 ret
= check_thread_quit_pipe(pollfd
, revents
);
3354 /* Inspect the relay conn pipe for new connection */
3355 if (pollfd
== relay_conn_pipe
[0]) {
3356 if (revents
& LPOLLIN
) {
3357 struct relay_connection
*conn
;
3359 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3363 lttng_poll_add(&events
, conn
->sock
->fd
,
3364 LPOLLIN
| LPOLLRDHUP
);
3365 connection_ht_add(relay_connections_ht
, conn
);
3366 DBG("Connection socket %d added", conn
->sock
->fd
);
3367 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3368 ERR("Relay connection pipe error");
3371 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3375 struct relay_connection
*ctrl_conn
;
3377 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3378 /* If not found, there is a synchronization issue. */
3381 if (ctrl_conn
->type
== RELAY_DATA
) {
3382 if (revents
& LPOLLIN
) {
3384 * Flag the last seen data fd not deleted. It will be
3385 * used as the last seen fd if any fd gets deleted in
3388 last_notdel_data_fd
= pollfd
;
3390 goto put_ctrl_connection
;
3392 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3394 if (revents
& LPOLLIN
) {
3395 enum relay_connection_status status
;
3397 status
= relay_process_control(ctrl_conn
);
3398 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3400 * On socket error flag the session as aborted to force
3401 * the cleanup of its stream otherwise it can leak
3402 * during the lifetime of the relayd.
3404 * This prevents situations in which streams can be
3405 * left opened because an index was received, the
3406 * control connection is closed, and the data
3407 * connection is closed (uncleanly) before the packet's
3410 * Since the control connection encountered an error,
3411 * it is okay to be conservative and close the
3412 * session right now as we can't rely on the protocol
3413 * being respected anymore.
3415 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3416 session_abort(ctrl_conn
->session
);
3419 /* Clear the connection on error or close. */
3420 relay_thread_close_connection(&events
,
3425 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3426 relay_thread_close_connection(&events
,
3428 if (last_seen_data_fd
== pollfd
) {
3429 last_seen_data_fd
= last_notdel_data_fd
;
3432 ERR("Unexpected poll events %u for control sock %d",
3434 connection_put(ctrl_conn
);
3437 put_ctrl_connection
:
3438 connection_put(ctrl_conn
);
3443 * The last loop handled a control request, go back to poll to make
3444 * sure we prioritise the control socket.
3450 if (last_seen_data_fd
>= 0) {
3451 for (i
= 0; i
< nb_fd
; i
++) {
3452 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3454 health_code_update();
3456 if (last_seen_data_fd
== pollfd
) {
3463 /* Process data connection. */
3464 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3465 /* Fetch the poll data. */
3466 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3467 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3468 struct relay_connection
*data_conn
;
3470 health_code_update();
3473 /* No activity for this FD (poll implementation). */
3477 /* Skip the command pipe. It's handled in the first loop. */
3478 if (pollfd
== relay_conn_pipe
[0]) {
3482 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3484 /* Skip it. Might be removed before. */
3487 if (data_conn
->type
== RELAY_CONTROL
) {
3488 goto put_data_connection
;
3490 assert(data_conn
->type
== RELAY_DATA
);
3492 if (revents
& LPOLLIN
) {
3493 enum relay_connection_status status
;
3495 status
= relay_process_data(data_conn
);
3496 /* Connection closed or error. */
3497 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3499 * On socket error flag the session as aborted to force
3500 * the cleanup of its stream otherwise it can leak
3501 * during the lifetime of the relayd.
3503 * This prevents situations in which streams can be
3504 * left opened because an index was received, the
3505 * control connection is closed, and the data
3506 * connection is closed (uncleanly) before the packet's
3509 * Since the data connection encountered an error,
3510 * it is okay to be conservative and close the
3511 * session right now as we can't rely on the protocol
3512 * being respected anymore.
3514 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3515 session_abort(data_conn
->session
);
3517 relay_thread_close_connection(&events
, pollfd
,
3520 * Every goto restart call sets the last seen fd where
3521 * here we don't really care since we gracefully
3522 * continue the loop after the connection is deleted.
3525 /* Keep last seen port. */
3526 last_seen_data_fd
= pollfd
;
3527 connection_put(data_conn
);
3530 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3531 relay_thread_close_connection(&events
, pollfd
,
3534 ERR("Unknown poll events %u for data sock %d",
3537 put_data_connection
:
3538 connection_put(data_conn
);
3540 last_seen_data_fd
= -1;
3543 /* Normal exit, no error */
3548 /* Cleanup remaining connection object. */
3550 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3553 health_code_update();
3555 session_abort(destroy_conn
->session
);
3558 * No need to grab another ref, because we own
3561 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
3566 lttng_poll_clean(&events
);
3568 lttng_ht_destroy(relay_connections_ht
);
3569 relay_connections_ht_error
:
3570 /* Close relay conn pipes */
3571 utils_close_pipe(relay_conn_pipe
);
3573 DBG("Thread exited with error");
3575 DBG("Worker thread cleanup complete");
3579 ERR("Health error occurred in %s", __func__
);
3581 health_unregister(health_relayd
);
3582 rcu_unregister_thread();
3583 lttng_relay_stop_threads();
3588 * Create the relay command pipe to wake thread_manage_apps.
3589 * Closed in cleanup().
3591 static int create_relay_conn_pipe(void)
3595 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
3603 int main(int argc
, char **argv
)
3605 int ret
= 0, retval
= 0;
3608 /* Parse arguments */
3610 if (set_options(argc
, argv
)) {
3615 if (set_signal_handler()) {
3620 /* Try to create directory if -o, --output is specified. */
3621 if (opt_output_path
) {
3622 if (*opt_output_path
!= '/') {
3623 ERR("Please specify an absolute path for -o, --output PATH");
3628 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
3631 ERR("Unable to create %s", opt_output_path
);
3638 if (opt_daemon
|| opt_background
) {
3641 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
3649 * We are in the child. Make sure all other file
3650 * descriptors are closed, in case we are called with
3651 * more opened file descriptors than the standard ones.
3653 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3658 sessiond_trace_chunk_registry
= sessiond_trace_chunk_registry_create();
3659 if (!sessiond_trace_chunk_registry
) {
3660 ERR("Failed to initialize session daemon trace chunk registry");
3662 goto exit_sessiond_trace_chunk_registry
;
3665 /* Initialize thread health monitoring */
3666 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
3667 if (!health_relayd
) {
3668 PERROR("health_app_create error");
3670 goto exit_health_app_create
;
3673 /* Create thread quit pipe */
3674 if (init_thread_quit_pipe()) {
3676 goto exit_init_data
;
3679 /* Setup the thread apps communication pipe. */
3680 if (create_relay_conn_pipe()) {
3682 goto exit_init_data
;
3685 /* Init relay command queue. */
3686 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
3688 /* Initialize communication library */
3690 lttcomm_inet_init();
3692 /* tables of sessions indexed by session ID */
3693 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3696 goto exit_init_data
;
3699 /* tables of streams indexed by stream ID */
3700 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3701 if (!relay_streams_ht
) {
3703 goto exit_init_data
;
3706 /* tables of streams indexed by stream ID */
3707 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3708 if (!viewer_streams_ht
) {
3710 goto exit_init_data
;
3713 ret
= utils_create_pipe(health_quit_pipe
);
3716 goto exit_health_quit_pipe
;
3719 /* Create thread to manage the client socket */
3720 ret
= pthread_create(&health_thread
, default_pthread_attr(),
3721 thread_manage_health
, (void *) NULL
);
3724 PERROR("pthread_create health");
3726 goto exit_health_thread
;
3729 /* Setup the dispatcher thread */
3730 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
3731 relay_thread_dispatcher
, (void *) NULL
);
3734 PERROR("pthread_create dispatcher");
3736 goto exit_dispatcher_thread
;
3739 /* Setup the worker thread */
3740 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
3741 relay_thread_worker
, NULL
);
3744 PERROR("pthread_create worker");
3746 goto exit_worker_thread
;
3749 /* Setup the listener thread */
3750 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
3751 relay_thread_listener
, (void *) NULL
);
3754 PERROR("pthread_create listener");
3756 goto exit_listener_thread
;
3759 ret
= relayd_live_create(live_uri
);
3761 ERR("Starting live viewer threads");
3767 * This is where we start awaiting program completion (e.g. through
3768 * signal that asks threads to teardown).
3771 ret
= relayd_live_join();
3777 ret
= pthread_join(listener_thread
, &status
);
3780 PERROR("pthread_join listener_thread");
3784 exit_listener_thread
:
3785 ret
= pthread_join(worker_thread
, &status
);
3788 PERROR("pthread_join worker_thread");
3793 ret
= pthread_join(dispatcher_thread
, &status
);
3796 PERROR("pthread_join dispatcher_thread");
3799 exit_dispatcher_thread
:
3801 ret
= pthread_join(health_thread
, &status
);
3804 PERROR("pthread_join health_thread");
3809 utils_close_pipe(health_quit_pipe
);
3810 exit_health_quit_pipe
:
3813 health_app_destroy(health_relayd
);
3814 sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry
);
3815 exit_health_app_create
:
3816 exit_sessiond_trace_chunk_registry
:
3819 * Wait for all pending call_rcu work to complete before tearing
3820 * down data structures. call_rcu worker may be trying to
3821 * perform lookups in those structures.
3826 /* Ensure all prior call_rcu are done. */