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>
37 #include <sys/resource.h>
39 #include <urcu/futex.h>
40 #include <urcu/uatomic.h>
45 #include <lttng/lttng.h>
46 #include <common/common.h>
47 #include <common/compat/poll.h>
48 #include <common/compat/socket.h>
49 #include <common/compat/endian.h>
50 #include <common/compat/getenv.h>
51 #include <common/defaults.h>
52 #include <common/daemonize.h>
53 #include <common/futex.h>
54 #include <common/sessiond-comm/sessiond-comm.h>
55 #include <common/sessiond-comm/inet.h>
56 #include <common/sessiond-comm/relayd.h>
57 #include <common/uri.h>
58 #include <common/utils.h>
59 #include <common/config/session-config.h>
60 #include <common/dynamic-buffer.h>
61 #include <common/buffer-view.h>
62 #include <common/fd-tracker/utils.h>
63 #include <urcu/rculist.h>
67 #include "ctf-trace.h"
70 #include "lttng-relayd.h"
72 #include "health-relayd.h"
73 #include "testpoint.h"
74 #include "viewer-stream.h"
77 #include "connection.h"
78 #include "tracefile-array.h"
79 #include "tcp_keep_alive.h"
81 enum relay_connection_status
{
82 RELAY_CONNECTION_STATUS_OK
,
83 /* An error occured while processing an event on the connection. */
84 RELAY_CONNECTION_STATUS_ERROR
,
85 /* Connection closed/shutdown cleanly. */
86 RELAY_CONNECTION_STATUS_CLOSED
,
89 /* command line options */
90 char *opt_output_path
, *opt_working_directory
;
91 static int opt_daemon
, opt_background
, opt_print_version
;
92 int opt_group_output_by_session
;
93 int opt_group_output_by_host
;
96 * We need to wait for listener and live listener threads, as well as
97 * health check thread, before being ready to signal readiness.
99 #define NR_LTTNG_RELAY_READY 3
100 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
102 /* Size of receive buffer. */
103 #define RECV_DATA_BUFFER_SIZE 65536
105 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
106 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
108 static struct lttng_uri
*control_uri
;
109 static struct lttng_uri
*data_uri
;
110 static struct lttng_uri
*live_uri
;
112 const char *progname
;
114 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
115 static int tracing_group_name_override
;
117 const char * const config_section_name
= "relayd";
120 * Quit pipe for all threads. This permits a single cancellation point
121 * for all threads when receiving an event on the pipe.
123 int thread_quit_pipe
[2] = { -1, -1 };
126 * This pipe is used to inform the worker thread that a command is queued and
127 * ready to be processed.
129 static int relay_conn_pipe
[2] = { -1, -1 };
131 /* Shared between threads */
132 static int dispatch_thread_exit
;
134 static pthread_t listener_thread
;
135 static pthread_t dispatcher_thread
;
136 static pthread_t worker_thread
;
137 static pthread_t health_thread
;
140 * last_relay_stream_id_lock protects last_relay_stream_id increment
141 * atomicity on 32-bit architectures.
143 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
144 static uint64_t last_relay_stream_id
;
147 * Relay command queue.
149 * The relay_thread_listener and relay_thread_dispatcher communicate with this
152 static struct relay_conn_queue relay_conn_queue
;
154 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
155 static unsigned int lttng_opt_fd_cap
;
157 /* Global relay stream hash table. */
158 struct lttng_ht
*relay_streams_ht
;
160 /* Global relay viewer stream hash table. */
161 struct lttng_ht
*viewer_streams_ht
;
163 /* Global relay sessions hash table. */
164 struct lttng_ht
*sessions_ht
;
166 /* Relayd health monitoring */
167 struct health_app
*health_relayd
;
169 /* Global fd tracker. */
170 struct fd_tracker
*the_fd_tracker
;
172 static struct option long_options
[] = {
173 { "control-port", 1, 0, 'C', },
174 { "data-port", 1, 0, 'D', },
175 { "live-port", 1, 0, 'L', },
176 { "daemonize", 0, 0, 'd', },
177 { "background", 0, 0, 'b', },
178 { "group", 1, 0, 'g', },
179 { "fd-cap", 1, 0, '\0', },
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' },
185 { "working-directory", 1, 0, 'w', },
186 { "group-output-by-session", 0, 0, 's', },
187 { "group-output-by-host", 0, 0, 'p', },
191 static const char *config_ignore_options
[] = { "help", "config", "version" };
193 static void print_version(void) {
194 fprintf(stdout
, "%s\n", VERSION
);
197 static void relayd_config_log(void)
199 DBG("LTTng-relayd " VERSION
" - " VERSION_NAME
"%s%s",
200 GIT_VERSION
[0] == '\0' ? "" : " - " GIT_VERSION
,
201 EXTRA_VERSION_NAME
[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME
);
202 if (EXTRA_VERSION_DESCRIPTION
[0] != '\0') {
203 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION
"\n");
205 if (EXTRA_VERSION_PATCHES
[0] != '\0') {
206 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES
"\n");
211 * Take an option from the getopt output and set it in the right variable to be
214 * Return 0 on success else a negative value.
216 static int set_option(int opt
, const char *arg
, const char *optname
)
222 if (!strcmp(optname
, "fd-cap")) {
226 v
= strtoul(arg
, NULL
, 0);
227 if (errno
!= 0 || !isdigit(arg
[0])) {
228 ERR("Wrong value in --fd-cap parameter: %s", arg
);
232 if (v
< DEFAULT_RELAYD_MINIMAL_FD_CAP
) {
233 ERR("File descriptor cap must be set to at least %d",
234 DEFAULT_RELAYD_MINIMAL_FD_CAP
);
237 ERR("File descriptor cap overflow in --fd-cap parameter: %s", arg
);
241 lttng_opt_fd_cap
= (unsigned int) v
;
242 DBG3("File descriptor cap set to %u", lttng_opt_fd_cap
);
245 fprintf(stderr
, "unknown option %s", optname
);
247 fprintf(stderr
, " with arg %s\n", arg
);
252 if (lttng_is_setuid_setgid()) {
253 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
254 "-C, --control-port");
256 ret
= uri_parse(arg
, &control_uri
);
258 ERR("Invalid control URI specified");
261 if (control_uri
->port
== 0) {
262 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
267 if (lttng_is_setuid_setgid()) {
268 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
271 ret
= uri_parse(arg
, &data_uri
);
273 ERR("Invalid data URI specified");
276 if (data_uri
->port
== 0) {
277 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
282 if (lttng_is_setuid_setgid()) {
283 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
286 ret
= uri_parse(arg
, &live_uri
);
288 ERR("Invalid live URI specified");
291 if (live_uri
->port
== 0) {
292 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
303 if (lttng_is_setuid_setgid()) {
304 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
307 tracing_group_name
= strdup(arg
);
308 if (tracing_group_name
== NULL
) {
313 tracing_group_name_override
= 1;
317 ret
= utils_show_man_page(8, "lttng-relayd");
319 ERR("Cannot view man page lttng-relayd(8)");
324 opt_print_version
= 1;
327 if (lttng_is_setuid_setgid()) {
328 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
331 ret
= asprintf(&opt_output_path
, "%s", arg
);
334 PERROR("asprintf opt_output_path");
340 if (lttng_is_setuid_setgid()) {
341 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
342 "-w, --working-directory");
344 ret
= asprintf(&opt_working_directory
, "%s", arg
);
347 PERROR("asprintf working_directory");
354 /* Verbose level can increase using multiple -v */
356 lttng_opt_verbose
= config_parse_value(arg
);
358 /* Only 3 level of verbosity (-vvv). */
359 if (lttng_opt_verbose
< 3) {
360 lttng_opt_verbose
+= 1;
365 if (opt_group_output_by_host
) {
366 ERR("Cannot set --group-output-by-session, --group-output-by-host already defined");
369 opt_group_output_by_session
= 1;
372 if (opt_group_output_by_session
) {
373 ERR("Cannot set --group-output-by-host, --group-output-by-session already defined");
376 opt_group_output_by_host
= 1;
379 /* Unknown option or other error.
380 * Error is printed by getopt, just return */
393 * config_entry_handler_cb used to handle options read from a config file.
394 * See config_entry_handler_cb comment in common/config/session-config.h for the
395 * return value conventions.
397 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
401 if (!entry
|| !entry
->name
|| !entry
->value
) {
406 /* Check if the option is to be ignored */
407 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
408 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
413 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
414 /* Ignore if entry name is not fully matched. */
415 if (strcmp(entry
->name
, long_options
[i
].name
)) {
420 * If the option takes no argument on the command line,
421 * we have to check if the value is "true". We support
422 * non-zero numeric values, true, on and yes.
424 if (!long_options
[i
].has_arg
) {
425 ret
= config_parse_value(entry
->value
);
428 WARN("Invalid configuration value \"%s\" for option %s",
429 entry
->value
, entry
->name
);
431 /* False, skip boolean config option. */
436 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
440 WARN("Unrecognized option \"%s\" in daemon configuration file.",
447 static void parse_env_options(void)
451 value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV
);
453 opt_working_directory
= value
;
457 static int set_options(int argc
, char **argv
)
459 int c
, ret
= 0, option_index
= 0, retval
= 0;
460 int orig_optopt
= optopt
, orig_optind
= optind
;
461 char *default_address
, *optstring
;
462 const char *config_path
= NULL
;
464 optstring
= utils_generate_optstring(long_options
,
465 sizeof(long_options
) / sizeof(struct option
));
471 /* Check for the --config option */
473 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
474 &option_index
)) != -1) {
478 } else if (c
!= 'f') {
482 if (lttng_is_setuid_setgid()) {
483 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
486 config_path
= utils_expand_path(optarg
);
488 ERR("Failed to resolve path: %s", optarg
);
493 ret
= config_get_section_entries(config_path
, config_section_name
,
494 config_entry_handler
, NULL
);
497 ERR("Invalid configuration option at line %i", ret
);
503 /* Reset getopt's global state */
504 optopt
= orig_optopt
;
505 optind
= orig_optind
;
507 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
512 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
519 /* assign default values */
520 if (control_uri
== NULL
) {
521 ret
= asprintf(&default_address
,
522 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
523 DEFAULT_NETWORK_CONTROL_PORT
);
525 PERROR("asprintf default data address");
530 ret
= uri_parse(default_address
, &control_uri
);
531 free(default_address
);
533 ERR("Invalid control URI specified");
538 if (data_uri
== NULL
) {
539 ret
= asprintf(&default_address
,
540 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
541 DEFAULT_NETWORK_DATA_PORT
);
543 PERROR("asprintf default data address");
548 ret
= uri_parse(default_address
, &data_uri
);
549 free(default_address
);
551 ERR("Invalid data URI specified");
556 if (live_uri
== NULL
) {
557 ret
= asprintf(&default_address
,
558 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
559 DEFAULT_NETWORK_VIEWER_PORT
);
561 PERROR("asprintf default viewer control address");
566 ret
= uri_parse(default_address
, &live_uri
);
567 free(default_address
);
569 ERR("Invalid viewer control URI specified");
574 if (lttng_opt_fd_cap
== 0) {
576 struct rlimit rlimit
;
578 ret
= getrlimit(RLIMIT_NOFILE
, &rlimit
);
580 PERROR("Failed to get file descriptor limit");
584 lttng_opt_fd_cap
= rlimit
.rlim_cur
;
587 if (!opt_group_output_by_session
&& !opt_group_output_by_host
) {
588 /* Group by host by default */
589 opt_group_output_by_host
= 1;
597 static void print_global_objects(void)
599 print_viewer_streams();
600 print_relay_streams();
607 static void relayd_cleanup(void)
609 print_global_objects();
613 if (viewer_streams_ht
)
614 lttng_ht_destroy(viewer_streams_ht
);
615 if (relay_streams_ht
)
616 lttng_ht_destroy(relay_streams_ht
);
618 lttng_ht_destroy(sessions_ht
);
620 /* free the dynamically allocated opt_output_path */
621 free(opt_output_path
);
623 /* Close thread quit pipes */
624 (void) fd_tracker_util_pipe_close(the_fd_tracker
, thread_quit_pipe
);
626 uri_free(control_uri
);
628 /* Live URI is freed in the live thread. */
630 if (tracing_group_name_override
) {
631 free((void *) tracing_group_name
);
636 * Write to writable pipe used to notify a thread.
638 static int notify_thread_pipe(int wpipe
)
642 ret
= lttng_write(wpipe
, "!", 1);
644 PERROR("write poll pipe");
652 static int notify_health_quit_pipe(int *pipe
)
656 ret
= lttng_write(pipe
[1], "4", 1);
658 PERROR("write relay health quit");
667 * Stop all relayd and relayd-live threads.
669 int lttng_relay_stop_threads(void)
673 /* Stopping all threads */
674 DBG("Terminating all threads");
675 if (notify_thread_pipe(thread_quit_pipe
[1])) {
676 ERR("write error on thread quit pipe");
680 if (notify_health_quit_pipe(health_quit_pipe
)) {
681 ERR("write error on health quit pipe");
684 /* Dispatch thread */
685 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
686 futex_nto1_wake(&relay_conn_queue
.futex
);
688 if (relayd_live_stop()) {
689 ERR("Error stopping live threads");
696 * Signal handler for the daemon
698 * Simply stop all worker threads, leaving main() return gracefully after
699 * joining all threads and calling cleanup().
701 static void sighandler(int sig
)
705 DBG("SIGINT caught");
706 if (lttng_relay_stop_threads()) {
707 ERR("Error stopping threads");
711 DBG("SIGTERM caught");
712 if (lttng_relay_stop_threads()) {
713 ERR("Error stopping threads");
717 CMM_STORE_SHARED(recv_child_signal
, 1);
725 * Setup signal handler for :
726 * SIGINT, SIGTERM, SIGPIPE
728 static int set_signal_handler(void)
734 if ((ret
= sigemptyset(&sigset
)) < 0) {
735 PERROR("sigemptyset");
742 sa
.sa_handler
= sighandler
;
743 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
748 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
753 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
758 sa
.sa_handler
= SIG_IGN
;
759 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
764 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
769 void lttng_relay_notify_ready(void)
771 /* Notify the parent of the fork() process that we are ready. */
772 if (opt_daemon
|| opt_background
) {
773 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
774 kill(child_ppid
, SIGUSR1
);
780 * Init thread quit pipe.
782 * Return -1 on error or 0 if all pipes are created.
784 static int init_thread_quit_pipe(void)
786 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
787 "Quit pipe", thread_quit_pipe
);
791 * Init health quit pipe.
793 * Return -1 on error or 0 if all pipes are created.
795 static int init_health_quit_pipe(void)
797 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
798 "Health quit pipe", health_quit_pipe
);
802 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
804 static int create_named_thread_poll_set(struct lttng_poll_event
*events
,
805 int size
, const char *name
)
809 if (events
== NULL
|| size
== 0) {
814 ret
= fd_tracker_util_poll_create(the_fd_tracker
,
815 name
, events
, 1, LTTNG_CLOEXEC
);
818 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
830 * Check if the thread quit pipe was triggered.
832 * Return 1 if it was triggered else 0;
834 static int check_thread_quit_pipe(int fd
, uint32_t events
)
836 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
843 static int create_sock(void *data
, int *out_fd
)
846 struct lttcomm_sock
*sock
= data
;
848 ret
= lttcomm_create_sock(sock
);
858 static int close_sock(void *data
, int *in_fd
)
860 struct lttcomm_sock
*sock
= data
;
862 return sock
->ops
->close(sock
);
865 static int accept_sock(void *data
, int *out_fd
)
868 /* Socks is an array of in_sock, out_sock. */
869 struct lttcomm_sock
**socks
= data
;
870 struct lttcomm_sock
*in_sock
= socks
[0];
872 socks
[1] = in_sock
->ops
->accept(in_sock
);
877 *out_fd
= socks
[1]->fd
;
883 * Create and init socket from uri.
885 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
,
889 struct lttcomm_sock
*sock
= NULL
;
890 char uri_str
[PATH_MAX
];
891 char *formated_name
= NULL
;
893 sock
= lttcomm_alloc_sock_from_uri(uri
);
895 ERR("Allocating socket");
900 * Don't fail to create the socket if the name can't be built as it is
901 * only used for debugging purposes.
903 ret
= uri_to_str_url(uri
, uri_str
, sizeof(uri_str
));
904 uri_str
[sizeof(uri_str
) - 1] = '\0';
906 ret
= asprintf(&formated_name
, "%s socket @ %s", name
,
909 formated_name
= NULL
;
913 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &sock_fd
,
914 (const char **) (formated_name
? &formated_name
: NULL
),
915 1, create_sock
, sock
);
917 DBG("Listening on %s socket %d", name
, sock
->fd
);
919 ret
= sock
->ops
->bind(sock
);
924 ret
= sock
->ops
->listen(sock
, -1);
934 lttcomm_destroy_sock(sock
);
940 struct lttcomm_sock
*accept_relayd_sock(struct lttcomm_sock
*listening_sock
,
944 struct lttcomm_sock
*socks
[2] = { listening_sock
, NULL
};
945 struct lttcomm_sock
*new_sock
= NULL
;
947 ret
= fd_tracker_open_unsuspendable_fd(
948 the_fd_tracker
, &out_fd
,
949 (const char **) &name
,
950 1, accept_sock
, &socks
);
955 DBG("%s accepted, socket %d", name
, new_sock
->fd
);
961 * This thread manages the listening for new connections on the network
963 static void *relay_thread_listener(void *data
)
965 int i
, ret
, pollfd
, err
= -1;
966 uint32_t revents
, nb_fd
;
967 struct lttng_poll_event events
;
968 struct lttcomm_sock
*control_sock
, *data_sock
;
970 DBG("[thread] Relay listener started");
972 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
974 health_code_update();
976 control_sock
= relay_socket_create(control_uri
, "Control listener");
978 goto error_sock_control
;
981 data_sock
= relay_socket_create(data_uri
, "Data listener");
983 goto error_sock_relay
;
987 * Pass 3 as size here for the thread quit pipe, control and
990 ret
= create_named_thread_poll_set(&events
, 3, "Listener thread epoll");
992 goto error_create_poll
;
995 /* Add the control socket */
996 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1001 /* Add the data socket */
1002 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1004 goto error_poll_add
;
1007 lttng_relay_notify_ready();
1009 if (testpoint(relayd_thread_listener
)) {
1010 goto error_testpoint
;
1014 health_code_update();
1016 DBG("Listener accepting connections");
1019 health_poll_entry();
1020 ret
= lttng_poll_wait(&events
, -1);
1024 * Restart interrupted system call.
1026 if (errno
== EINTR
) {
1034 DBG("Relay new connection received");
1035 for (i
= 0; i
< nb_fd
; i
++) {
1036 health_code_update();
1038 /* Fetch once the poll data */
1039 revents
= LTTNG_POLL_GETEV(&events
, i
);
1040 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1044 * No activity for this FD (poll
1050 /* Thread quit pipe has been closed. Killing thread. */
1051 ret
= check_thread_quit_pipe(pollfd
, revents
);
1057 if (revents
& LPOLLIN
) {
1059 * A new connection is requested, therefore a
1060 * sessiond/consumerd connection is allocated in
1061 * this thread, enqueued to a global queue and
1062 * dequeued (and freed) in the worker thread.
1065 struct relay_connection
*new_conn
;
1066 struct lttcomm_sock
*newsock
= NULL
;
1067 enum connection_type type
;
1069 if (pollfd
== data_sock
->fd
) {
1071 newsock
= accept_relayd_sock(data_sock
,
1072 "Data socket to relayd");
1074 assert(pollfd
== control_sock
->fd
);
1075 type
= RELAY_CONTROL
;
1076 newsock
= accept_relayd_sock(control_sock
,
1077 "Control socket to relayd");
1080 PERROR("accepting sock");
1084 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
1087 PERROR("setsockopt inet");
1088 lttcomm_destroy_sock(newsock
);
1092 ret
= socket_apply_keep_alive_config(newsock
->fd
);
1094 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1096 lttcomm_destroy_sock(newsock
);
1100 new_conn
= connection_create(newsock
, type
);
1102 lttcomm_destroy_sock(newsock
);
1106 /* Enqueue request for the dispatcher thread. */
1107 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
1111 * Wake the dispatch queue futex.
1112 * Implicit memory barrier with the
1113 * exchange in cds_wfcq_enqueue.
1115 futex_nto1_wake(&relay_conn_queue
.futex
);
1116 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1117 ERR("socket poll error");
1120 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1130 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
1132 if (data_sock
->fd
>= 0) {
1133 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1134 &data_sock
->fd
, 1, close_sock
,
1140 lttcomm_destroy_sock(data_sock
);
1142 if (control_sock
->fd
>= 0) {
1143 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1144 &control_sock
->fd
, 1, close_sock
,
1150 lttcomm_destroy_sock(control_sock
);
1154 ERR("Health error occurred in %s", __func__
);
1156 health_unregister(health_relayd
);
1157 DBG("Relay listener thread cleanup complete");
1158 lttng_relay_stop_threads();
1163 * This thread manages the dispatching of the requests to worker threads
1165 static void *relay_thread_dispatcher(void *data
)
1169 struct cds_wfcq_node
*node
;
1170 struct relay_connection
*new_conn
= NULL
;
1172 DBG("[thread] Relay dispatcher started");
1174 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
1176 if (testpoint(relayd_thread_dispatcher
)) {
1177 goto error_testpoint
;
1180 health_code_update();
1183 health_code_update();
1185 /* Atomically prepare the queue futex */
1186 futex_nto1_prepare(&relay_conn_queue
.futex
);
1188 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1193 health_code_update();
1195 /* Dequeue commands */
1196 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1197 &relay_conn_queue
.tail
);
1199 DBG("Woken up but nothing in the relay command queue");
1200 /* Continue thread execution */
1203 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1205 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1208 * Inform worker thread of the new request. This
1209 * call is blocking so we can be assured that
1210 * the data will be read at some point in time
1211 * or wait to the end of the world :)
1213 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1215 PERROR("write connection pipe");
1216 connection_put(new_conn
);
1219 } while (node
!= NULL
);
1221 /* Futex wait on queue. Blocking call on futex() */
1222 health_poll_entry();
1223 futex_nto1_wait(&relay_conn_queue
.futex
);
1227 /* Normal exit, no error */
1234 ERR("Health error occurred in %s", __func__
);
1236 health_unregister(health_relayd
);
1237 DBG("Dispatch thread dying");
1238 lttng_relay_stop_threads();
1243 * Set index data from the control port to a given index object.
1245 static int set_index_control_data(struct relay_index
*index
,
1246 struct lttcomm_relayd_index
*data
,
1247 struct relay_connection
*conn
)
1249 struct ctf_packet_index index_data
;
1252 * The index on disk is encoded in big endian.
1254 index_data
.packet_size
= htobe64(data
->packet_size
);
1255 index_data
.content_size
= htobe64(data
->content_size
);
1256 index_data
.timestamp_begin
= htobe64(data
->timestamp_begin
);
1257 index_data
.timestamp_end
= htobe64(data
->timestamp_end
);
1258 index_data
.events_discarded
= htobe64(data
->events_discarded
);
1259 index_data
.stream_id
= htobe64(data
->stream_id
);
1261 if (conn
->minor
>= 8) {
1262 index
->index_data
.stream_instance_id
= htobe64(data
->stream_instance_id
);
1263 index
->index_data
.packet_seq_num
= htobe64(data
->packet_seq_num
);
1266 return relay_index_set_data(index
, &index_data
);
1270 * Handle the RELAYD_CREATE_SESSION command.
1272 * On success, send back the session id or else return a negative value.
1274 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1275 struct relay_connection
*conn
,
1276 const struct lttng_buffer_view
*payload
)
1280 struct relay_session
*session
;
1281 struct lttcomm_relayd_status_session reply
;
1282 char session_name
[LTTNG_NAME_MAX
];
1283 char hostname
[LTTNG_HOST_NAME_MAX
];
1284 uint32_t live_timer
= 0;
1285 bool snapshot
= false;
1287 memset(session_name
, 0, LTTNG_NAME_MAX
);
1288 memset(hostname
, 0, LTTNG_HOST_NAME_MAX
);
1290 memset(&reply
, 0, sizeof(reply
));
1292 switch (conn
->minor
) {
1297 case 4: /* LTTng sessiond 2.4 */
1299 ret
= cmd_create_session_2_4(payload
, session_name
,
1300 hostname
, &live_timer
, &snapshot
);
1306 session
= session_create(session_name
, hostname
, live_timer
,
1307 snapshot
, conn
->major
, conn
->minor
);
1312 assert(!conn
->session
);
1313 conn
->session
= session
;
1314 DBG("Created session %" PRIu64
, session
->id
);
1316 reply
.session_id
= htobe64(session
->id
);
1320 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1322 reply
.ret_code
= htobe32(LTTNG_OK
);
1325 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1326 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1327 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1336 * When we have received all the streams and the metadata for a channel,
1337 * we make them visible to the viewer threads.
1339 static void publish_connection_local_streams(struct relay_connection
*conn
)
1341 struct relay_stream
*stream
;
1342 struct relay_session
*session
= conn
->session
;
1345 * We publish all streams belonging to a session atomically wrt
1348 pthread_mutex_lock(&session
->lock
);
1350 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1352 stream_publish(stream
);
1357 * Inform the viewer that there are new streams in the session.
1359 if (session
->viewer_attached
) {
1360 uatomic_set(&session
->new_streams
, 1);
1362 pthread_mutex_unlock(&session
->lock
);
1366 * relay_add_stream: allocate a new stream for a session
1368 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1369 struct relay_connection
*conn
,
1370 const struct lttng_buffer_view
*payload
)
1374 struct relay_session
*session
= conn
->session
;
1375 struct relay_stream
*stream
= NULL
;
1376 struct lttcomm_relayd_status_stream reply
;
1377 struct ctf_trace
*trace
= NULL
;
1378 uint64_t stream_handle
= -1ULL;
1379 char *path_name
= NULL
, *channel_name
= NULL
;
1380 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1382 if (!session
|| !conn
->version_check_done
) {
1383 ERR("Trying to add a stream before version check");
1385 goto end_no_session
;
1388 switch (session
->minor
) {
1389 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1390 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1391 &channel_name
, session
);
1393 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1395 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1396 &channel_name
, &tracefile_size
, &tracefile_count
,
1404 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1408 /* This stream here has one reference on the trace. */
1410 pthread_mutex_lock(&last_relay_stream_id_lock
);
1411 stream_handle
= ++last_relay_stream_id
;
1412 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1414 /* We pass ownership of path_name and channel_name. */
1415 stream
= stream_create(trace
, stream_handle
, path_name
,
1416 channel_name
, tracefile_size
, tracefile_count
);
1418 channel_name
= NULL
;
1421 * Streams are the owners of their trace. Reference to trace is
1422 * kept within stream_create().
1424 ctf_trace_put(trace
);
1427 memset(&reply
, 0, sizeof(reply
));
1428 reply
.handle
= htobe64(stream_handle
);
1430 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1432 reply
.ret_code
= htobe32(LTTNG_OK
);
1435 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1436 sizeof(struct lttcomm_relayd_status_stream
), 0);
1437 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1438 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1450 * relay_close_stream: close a specific stream
1452 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1453 struct relay_connection
*conn
,
1454 const struct lttng_buffer_view
*payload
)
1458 struct relay_session
*session
= conn
->session
;
1459 struct lttcomm_relayd_close_stream stream_info
;
1460 struct lttcomm_relayd_generic_reply reply
;
1461 struct relay_stream
*stream
;
1463 DBG("Close stream received");
1465 if (!session
|| !conn
->version_check_done
) {
1466 ERR("Trying to close a stream before version check");
1468 goto end_no_session
;
1471 if (payload
->size
< sizeof(stream_info
)) {
1472 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1473 sizeof(stream_info
), payload
->size
);
1475 goto end_no_session
;
1477 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1478 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1479 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1481 stream
= stream_get_by_id(stream_info
.stream_id
);
1488 * Set last_net_seq_num before the close flag. Required by data
1491 pthread_mutex_lock(&stream
->lock
);
1492 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1493 pthread_mutex_unlock(&stream
->lock
);
1496 * This is one of the conditions which may trigger a stream close
1497 * with the others being:
1498 * 1) A close command is received for a stream
1499 * 2) The control connection owning the stream is closed
1500 * 3) We have received all of the stream's data _after_ a close
1503 try_stream_close(stream
);
1504 if (stream
->is_metadata
) {
1505 struct relay_viewer_stream
*vstream
;
1507 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1509 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1511 * Since all the metadata has been sent to the
1512 * viewer and that we have a request to close
1513 * its stream, we can safely teardown the
1514 * corresponding metadata viewer stream.
1516 viewer_stream_put(vstream
);
1518 /* Put local reference. */
1519 viewer_stream_put(vstream
);
1526 memset(&reply
, 0, sizeof(reply
));
1528 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1530 reply
.ret_code
= htobe32(LTTNG_OK
);
1532 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1533 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1534 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1535 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1545 * relay_reset_metadata: reset a metadata stream
1548 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1549 struct relay_connection
*conn
,
1550 const struct lttng_buffer_view
*payload
)
1554 struct relay_session
*session
= conn
->session
;
1555 struct lttcomm_relayd_reset_metadata stream_info
;
1556 struct lttcomm_relayd_generic_reply reply
;
1557 struct relay_stream
*stream
;
1559 DBG("Reset metadata received");
1561 if (!session
|| !conn
->version_check_done
) {
1562 ERR("Trying to reset a metadata stream before version check");
1564 goto end_no_session
;
1567 if (payload
->size
< sizeof(stream_info
)) {
1568 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1569 sizeof(stream_info
), payload
->size
);
1571 goto end_no_session
;
1573 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1574 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1575 stream_info
.version
= be64toh(stream_info
.version
);
1577 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1579 /* Unsupported for live sessions for now. */
1580 if (session
->live_timer
!= 0) {
1585 stream
= stream_get_by_id(stream_info
.stream_id
);
1590 pthread_mutex_lock(&stream
->lock
);
1591 if (!stream
->is_metadata
) {
1596 ret
= stream_fd_rotate(stream
->stream_fd
,
1597 stream
->path_name
, stream
->channel_name
, 0, 0, NULL
);
1599 ERR("Failed to rotate metadata file %s of channel %s",
1600 stream
->path_name
, stream
->channel_name
);
1605 pthread_mutex_unlock(&stream
->lock
);
1609 memset(&reply
, 0, sizeof(reply
));
1611 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1613 reply
.ret_code
= htobe32(LTTNG_OK
);
1615 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1616 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1617 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1618 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1628 * relay_unknown_command: send -1 if received unknown command
1630 static void relay_unknown_command(struct relay_connection
*conn
)
1632 struct lttcomm_relayd_generic_reply reply
;
1635 memset(&reply
, 0, sizeof(reply
));
1636 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1637 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1638 if (send_ret
< sizeof(reply
)) {
1639 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1644 * relay_start: send an acknowledgment to the client to tell if we are
1645 * ready to receive data. We are ready if a session is established.
1647 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1648 struct relay_connection
*conn
,
1649 const struct lttng_buffer_view
*payload
)
1653 struct lttcomm_relayd_generic_reply reply
;
1654 struct relay_session
*session
= conn
->session
;
1657 DBG("Trying to start the streaming without a session established");
1658 ret
= htobe32(LTTNG_ERR_UNK
);
1661 memset(&reply
, 0, sizeof(reply
));
1662 reply
.ret_code
= htobe32(LTTNG_OK
);
1663 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1665 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1666 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1675 * Append padding to the file pointed by the file descriptor fd.
1677 static int write_padding_to_file(int fd
, uint32_t size
)
1686 zeros
= zmalloc(size
);
1687 if (zeros
== NULL
) {
1688 PERROR("zmalloc zeros for padding");
1693 ret
= lttng_write(fd
, zeros
, size
);
1695 PERROR("write padding to file");
1705 * relay_recv_metadata: receive the metadata for the session.
1707 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1708 struct relay_connection
*conn
,
1709 const struct lttng_buffer_view
*payload
)
1713 struct relay_session
*session
= conn
->session
;
1714 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1715 struct relay_stream
*metadata_stream
;
1716 uint64_t metadata_payload_size
;
1717 int metadata_fd
= -1;
1720 ERR("Metadata sent before version check");
1725 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1726 ERR("Incorrect data size");
1730 metadata_payload_size
= recv_hdr
->data_size
-
1731 sizeof(struct lttcomm_relayd_metadata_payload
);
1733 memcpy(&metadata_payload_header
, payload
->data
,
1734 sizeof(metadata_payload_header
));
1735 metadata_payload_header
.stream_id
= be64toh(
1736 metadata_payload_header
.stream_id
);
1737 metadata_payload_header
.padding_size
= be32toh(
1738 metadata_payload_header
.padding_size
);
1740 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1741 if (!metadata_stream
) {
1746 pthread_mutex_lock(&metadata_stream
->lock
);
1748 metadata_fd
= stream_fd_get_fd(metadata_stream
->stream_fd
);
1749 if (metadata_fd
< 0) {
1752 size_ret
= lttng_write(metadata_fd
,
1753 payload
->data
+ sizeof(metadata_payload_header
),
1754 metadata_payload_size
);
1755 if (size_ret
< metadata_payload_size
) {
1756 ERR("Relay error writing metadata on file");
1761 size_ret
= write_padding_to_file(metadata_fd
,
1762 metadata_payload_header
.padding_size
);
1763 if (size_ret
< (int64_t) metadata_payload_header
.padding_size
) {
1768 metadata_stream
->metadata_received
+=
1769 metadata_payload_size
+ metadata_payload_header
.padding_size
;
1770 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1771 metadata_stream
->metadata_received
);
1774 stream_fd_put_fd(metadata_stream
->stream_fd
);
1776 pthread_mutex_unlock(&metadata_stream
->lock
);
1777 stream_put(metadata_stream
);
1783 * relay_send_version: send relayd version number
1785 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1786 struct relay_connection
*conn
,
1787 const struct lttng_buffer_view
*payload
)
1791 struct lttcomm_relayd_version reply
, msg
;
1792 bool compatible
= true;
1794 conn
->version_check_done
= true;
1796 /* Get version from the other side. */
1797 if (payload
->size
< sizeof(msg
)) {
1798 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1799 sizeof(msg
), payload
->size
);
1804 memcpy(&msg
, payload
->data
, sizeof(msg
));
1805 msg
.major
= be32toh(msg
.major
);
1806 msg
.minor
= be32toh(msg
.minor
);
1808 memset(&reply
, 0, sizeof(reply
));
1809 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1810 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1812 /* Major versions must be the same */
1813 if (reply
.major
!= msg
.major
) {
1814 DBG("Incompatible major versions (%u vs %u), deleting session",
1815 reply
.major
, msg
.major
);
1819 conn
->major
= reply
.major
;
1820 /* We adapt to the lowest compatible version */
1821 if (reply
.minor
<= msg
.minor
) {
1822 conn
->minor
= reply
.minor
;
1824 conn
->minor
= msg
.minor
;
1827 reply
.major
= htobe32(reply
.major
);
1828 reply
.minor
= htobe32(reply
.minor
);
1829 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1831 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1832 ERR("Failed to send \"send version\" command reply (ret = %zd)",
1845 DBG("Version check done using protocol %u.%u", conn
->major
,
1853 * Check for data pending for a given stream id from the session daemon.
1855 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1856 struct relay_connection
*conn
,
1857 const struct lttng_buffer_view
*payload
)
1859 struct relay_session
*session
= conn
->session
;
1860 struct lttcomm_relayd_data_pending msg
;
1861 struct lttcomm_relayd_generic_reply reply
;
1862 struct relay_stream
*stream
;
1866 DBG("Data pending command received");
1868 if (!session
|| !conn
->version_check_done
) {
1869 ERR("Trying to check for data before version check");
1871 goto end_no_session
;
1874 if (payload
->size
< sizeof(msg
)) {
1875 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
1876 sizeof(msg
), payload
->size
);
1878 goto end_no_session
;
1880 memcpy(&msg
, payload
->data
, sizeof(msg
));
1881 msg
.stream_id
= be64toh(msg
.stream_id
);
1882 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1884 stream
= stream_get_by_id(msg
.stream_id
);
1885 if (stream
== NULL
) {
1890 pthread_mutex_lock(&stream
->lock
);
1892 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1893 " and last_seq %" PRIu64
, msg
.stream_id
,
1894 stream
->prev_seq
, msg
.last_net_seq_num
);
1896 /* Avoid wrapping issue */
1897 if (((int64_t) (stream
->prev_seq
- msg
.last_net_seq_num
)) >= 0) {
1898 /* Data has in fact been written and is NOT pending */
1901 /* Data still being streamed thus pending */
1905 stream
->data_pending_check_done
= true;
1906 pthread_mutex_unlock(&stream
->lock
);
1911 memset(&reply
, 0, sizeof(reply
));
1912 reply
.ret_code
= htobe32(ret
);
1913 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1914 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1915 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
1925 * Wait for the control socket to reach a quiescent state.
1927 * Note that for now, when receiving this command from the session
1928 * daemon, this means that every subsequent commands or data received on
1929 * the control socket has been handled. So, this is why we simply return
1932 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
1933 struct relay_connection
*conn
,
1934 const struct lttng_buffer_view
*payload
)
1938 struct relay_stream
*stream
;
1939 struct lttcomm_relayd_quiescent_control msg
;
1940 struct lttcomm_relayd_generic_reply reply
;
1942 DBG("Checking quiescent state on control socket");
1944 if (!conn
->session
|| !conn
->version_check_done
) {
1945 ERR("Trying to check for data before version check");
1947 goto end_no_session
;
1950 if (payload
->size
< sizeof(msg
)) {
1951 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
1952 sizeof(msg
), payload
->size
);
1954 goto end_no_session
;
1956 memcpy(&msg
, payload
->data
, sizeof(msg
));
1957 msg
.stream_id
= be64toh(msg
.stream_id
);
1959 stream
= stream_get_by_id(msg
.stream_id
);
1963 pthread_mutex_lock(&stream
->lock
);
1964 stream
->data_pending_check_done
= true;
1965 pthread_mutex_unlock(&stream
->lock
);
1967 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
1970 memset(&reply
, 0, sizeof(reply
));
1971 reply
.ret_code
= htobe32(LTTNG_OK
);
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 \"quiescent control\" command reply (ret = %zd)",
1986 * Initialize a data pending command. This means that a consumer is about
1987 * to ask for data pending for each stream it holds. Simply iterate over
1988 * all streams of a session and set the data_pending_check_done flag.
1990 * This command returns to the client a LTTNG_OK code.
1992 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1993 struct relay_connection
*conn
,
1994 const struct lttng_buffer_view
*payload
)
1998 struct lttng_ht_iter iter
;
1999 struct lttcomm_relayd_begin_data_pending msg
;
2000 struct lttcomm_relayd_generic_reply reply
;
2001 struct relay_stream
*stream
;
2006 DBG("Init streams for data pending");
2008 if (!conn
->session
|| !conn
->version_check_done
) {
2009 ERR("Trying to check for data before version check");
2011 goto end_no_session
;
2014 if (payload
->size
< sizeof(msg
)) {
2015 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2016 sizeof(msg
), payload
->size
);
2018 goto end_no_session
;
2020 memcpy(&msg
, payload
->data
, sizeof(msg
));
2021 msg
.session_id
= be64toh(msg
.session_id
);
2024 * Iterate over all streams to set the begin data pending flag.
2025 * For now, the streams are indexed by stream handle so we have
2026 * to iterate over all streams to find the one associated with
2027 * the right session_id.
2030 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2032 if (!stream_get(stream
)) {
2035 if (stream
->trace
->session
->id
== msg
.session_id
) {
2036 pthread_mutex_lock(&stream
->lock
);
2037 stream
->data_pending_check_done
= false;
2038 pthread_mutex_unlock(&stream
->lock
);
2039 DBG("Set begin data pending flag to stream %" PRIu64
,
2040 stream
->stream_handle
);
2046 memset(&reply
, 0, sizeof(reply
));
2047 /* All good, send back reply. */
2048 reply
.ret_code
= htobe32(LTTNG_OK
);
2050 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2051 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2052 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2064 * End data pending command. This will check, for a given session id, if
2065 * each stream associated with it has its data_pending_check_done flag
2066 * set. If not, this means that the client lost track of the stream but
2067 * the data is still being streamed on our side. In this case, we inform
2068 * the client that data is in flight.
2070 * Return to the client if there is data in flight or not with a ret_code.
2072 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2073 struct relay_connection
*conn
,
2074 const struct lttng_buffer_view
*payload
)
2078 struct lttng_ht_iter iter
;
2079 struct lttcomm_relayd_end_data_pending msg
;
2080 struct lttcomm_relayd_generic_reply reply
;
2081 struct relay_stream
*stream
;
2082 uint32_t is_data_inflight
= 0;
2084 DBG("End data pending command");
2086 if (!conn
->session
|| !conn
->version_check_done
) {
2087 ERR("Trying to check for data before version check");
2089 goto end_no_session
;
2092 if (payload
->size
< sizeof(msg
)) {
2093 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2094 sizeof(msg
), payload
->size
);
2096 goto end_no_session
;
2098 memcpy(&msg
, payload
->data
, sizeof(msg
));
2099 msg
.session_id
= be64toh(msg
.session_id
);
2102 * Iterate over all streams to see if the begin data pending
2106 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2108 if (!stream_get(stream
)) {
2111 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2115 pthread_mutex_lock(&stream
->lock
);
2116 if (!stream
->data_pending_check_done
) {
2117 if (!stream
->closed
|| !(((int64_t) (stream
->prev_seq
- stream
->last_net_seq_num
)) >= 0)) {
2118 is_data_inflight
= 1;
2119 DBG("Data is still in flight for stream %" PRIu64
,
2120 stream
->stream_handle
);
2121 pthread_mutex_unlock(&stream
->lock
);
2126 pthread_mutex_unlock(&stream
->lock
);
2131 memset(&reply
, 0, sizeof(reply
));
2132 /* All good, send back reply. */
2133 reply
.ret_code
= htobe32(is_data_inflight
);
2135 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2136 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2137 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2149 * Receive an index for a specific stream.
2151 * Return 0 on success else a negative value.
2153 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2154 struct relay_connection
*conn
,
2155 const struct lttng_buffer_view
*payload
)
2159 struct relay_session
*session
= conn
->session
;
2160 struct lttcomm_relayd_index index_info
;
2161 struct relay_index
*index
;
2162 struct lttcomm_relayd_generic_reply reply
;
2163 struct relay_stream
*stream
;
2168 DBG("Relay receiving index");
2170 if (!session
|| !conn
->version_check_done
) {
2171 ERR("Trying to close a stream before version check");
2173 goto end_no_session
;
2176 msg_len
= lttcomm_relayd_index_len(
2177 lttng_to_index_major(conn
->major
, conn
->minor
),
2178 lttng_to_index_minor(conn
->major
, conn
->minor
));
2179 if (payload
->size
< msg_len
) {
2180 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2181 msg_len
, payload
->size
);
2183 goto end_no_session
;
2185 memcpy(&index_info
, payload
->data
, msg_len
);
2186 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2187 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2188 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2189 index_info
.content_size
= be64toh(index_info
.content_size
);
2190 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2191 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2192 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2193 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2195 if (conn
->minor
>= 8) {
2196 index_info
.stream_instance_id
=
2197 be64toh(index_info
.stream_instance_id
);
2198 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2201 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2203 ERR("stream_get_by_id not found");
2207 pthread_mutex_lock(&stream
->lock
);
2209 /* Live beacon handling */
2210 if (index_info
.packet_size
== 0) {
2211 DBG("Received live beacon for stream %" PRIu64
,
2212 stream
->stream_handle
);
2215 * Only flag a stream inactive when it has already
2216 * received data and no indexes are in flight.
2218 if (stream
->index_received_seqcount
> 0
2219 && stream
->indexes_in_flight
== 0) {
2220 stream
->beacon_ts_end
= index_info
.timestamp_end
;
2223 goto end_stream_put
;
2225 stream
->beacon_ts_end
= -1ULL;
2228 if (stream
->ctf_stream_id
== -1ULL) {
2229 stream
->ctf_stream_id
= index_info
.stream_id
;
2231 index
= relay_index_get_by_id_or_create(stream
, index_info
.net_seq_num
);
2234 ERR("relay_index_get_by_id_or_create index NULL");
2235 goto end_stream_put
;
2237 if (set_index_control_data(index
, &index_info
, conn
)) {
2238 ERR("set_index_control_data error");
2239 relay_index_put(index
);
2241 goto end_stream_put
;
2243 ret
= relay_index_try_flush(index
);
2245 tracefile_array_commit_seq(stream
->tfa
);
2246 stream
->index_received_seqcount
++;
2247 } else if (ret
> 0) {
2251 ERR("relay_index_try_flush error %d", ret
);
2252 relay_index_put(index
);
2257 pthread_mutex_unlock(&stream
->lock
);
2262 memset(&reply
, 0, sizeof(reply
));
2264 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2266 reply
.ret_code
= htobe32(LTTNG_OK
);
2268 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2269 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2270 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2279 * Receive the streams_sent message.
2281 * Return 0 on success else a negative value.
2283 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2284 struct relay_connection
*conn
,
2285 const struct lttng_buffer_view
*payload
)
2289 struct lttcomm_relayd_generic_reply reply
;
2293 DBG("Relay receiving streams_sent");
2295 if (!conn
->session
|| !conn
->version_check_done
) {
2296 ERR("Trying to close a stream before version check");
2298 goto end_no_session
;
2302 * Publish every pending stream in the connection recv list which are
2303 * now ready to be used by the viewer.
2305 publish_connection_local_streams(conn
);
2307 memset(&reply
, 0, sizeof(reply
));
2308 reply
.ret_code
= htobe32(LTTNG_OK
);
2309 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2310 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2311 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2323 #define DBG_CMD(cmd_name, conn) \
2324 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2326 static int relay_process_control_command(struct relay_connection
*conn
,
2327 const struct lttcomm_relayd_hdr
*header
,
2328 const struct lttng_buffer_view
*payload
)
2332 switch (header
->cmd
) {
2333 case RELAYD_CREATE_SESSION
:
2334 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
2335 ret
= relay_create_session(header
, conn
, payload
);
2337 case RELAYD_ADD_STREAM
:
2338 DBG_CMD("RELAYD_ADD_STREAM", conn
);
2339 ret
= relay_add_stream(header
, conn
, payload
);
2341 case RELAYD_START_DATA
:
2342 DBG_CMD("RELAYD_START_DATA", conn
);
2343 ret
= relay_start(header
, conn
, payload
);
2345 case RELAYD_SEND_METADATA
:
2346 DBG_CMD("RELAYD_SEND_METADATA", conn
);
2347 ret
= relay_recv_metadata(header
, conn
, payload
);
2349 case RELAYD_VERSION
:
2350 DBG_CMD("RELAYD_VERSION", conn
);
2351 ret
= relay_send_version(header
, conn
, payload
);
2353 case RELAYD_CLOSE_STREAM
:
2354 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
2355 ret
= relay_close_stream(header
, conn
, payload
);
2357 case RELAYD_DATA_PENDING
:
2358 DBG_CMD("RELAYD_DATA_PENDING", conn
);
2359 ret
= relay_data_pending(header
, conn
, payload
);
2361 case RELAYD_QUIESCENT_CONTROL
:
2362 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
2363 ret
= relay_quiescent_control(header
, conn
, payload
);
2365 case RELAYD_BEGIN_DATA_PENDING
:
2366 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
2367 ret
= relay_begin_data_pending(header
, conn
, payload
);
2369 case RELAYD_END_DATA_PENDING
:
2370 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
2371 ret
= relay_end_data_pending(header
, conn
, payload
);
2373 case RELAYD_SEND_INDEX
:
2374 DBG_CMD("RELAYD_SEND_INDEX", conn
);
2375 ret
= relay_recv_index(header
, conn
, payload
);
2377 case RELAYD_STREAMS_SENT
:
2378 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
2379 ret
= relay_streams_sent(header
, conn
, payload
);
2381 case RELAYD_RESET_METADATA
:
2382 DBG_CMD("RELAYD_RESET_METADATA", conn
);
2383 ret
= relay_reset_metadata(header
, conn
, payload
);
2385 case RELAYD_UPDATE_SYNC_INFO
:
2387 ERR("Received unknown command (%u)", header
->cmd
);
2388 relay_unknown_command(conn
);
2397 static enum relay_connection_status
relay_process_control_receive_payload(
2398 struct relay_connection
*conn
)
2401 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2402 struct lttng_dynamic_buffer
*reception_buffer
=
2403 &conn
->protocol
.ctrl
.reception_buffer
;
2404 struct ctrl_connection_state_receive_payload
*state
=
2405 &conn
->protocol
.ctrl
.state
.receive_payload
;
2406 struct lttng_buffer_view payload_view
;
2408 if (state
->left_to_receive
== 0) {
2409 /* Short-circuit for payload-less commands. */
2410 goto reception_complete
;
2412 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2413 reception_buffer
->data
+ state
->received
,
2414 state
->left_to_receive
, MSG_DONTWAIT
);
2416 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2417 PERROR("Unable to receive command payload on sock %d",
2419 status
= RELAY_CONNECTION_STATUS_ERROR
;
2422 } else if (ret
== 0) {
2423 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2424 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2429 assert(ret
<= state
->left_to_receive
);
2431 state
->left_to_receive
-= ret
;
2432 state
->received
+= ret
;
2434 if (state
->left_to_receive
> 0) {
2436 * Can't transition to the protocol's next state, wait to
2437 * receive the rest of the header.
2439 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2440 state
->received
, state
->left_to_receive
,
2446 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
2447 conn
->sock
->fd
, state
->received
);
2449 * The payload required to process the command has been received.
2450 * A view to the reception buffer is forwarded to the various
2451 * commands and the state of the control is reset on success.
2453 * Commands are responsible for sending their reply to the peer.
2455 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
2457 ret
= relay_process_control_command(conn
,
2458 &state
->header
, &payload_view
);
2460 status
= RELAY_CONNECTION_STATUS_ERROR
;
2464 ret
= connection_reset_protocol_state(conn
);
2466 status
= RELAY_CONNECTION_STATUS_ERROR
;
2472 static enum relay_connection_status
relay_process_control_receive_header(
2473 struct relay_connection
*conn
)
2476 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2477 struct lttcomm_relayd_hdr header
;
2478 struct lttng_dynamic_buffer
*reception_buffer
=
2479 &conn
->protocol
.ctrl
.reception_buffer
;
2480 struct ctrl_connection_state_receive_header
*state
=
2481 &conn
->protocol
.ctrl
.state
.receive_header
;
2483 assert(state
->left_to_receive
!= 0);
2485 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2486 reception_buffer
->data
+ state
->received
,
2487 state
->left_to_receive
, MSG_DONTWAIT
);
2489 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2490 PERROR("Unable to receive control command header on sock %d",
2492 status
= RELAY_CONNECTION_STATUS_ERROR
;
2495 } else if (ret
== 0) {
2496 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2497 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2502 assert(ret
<= state
->left_to_receive
);
2504 state
->left_to_receive
-= ret
;
2505 state
->received
+= ret
;
2507 if (state
->left_to_receive
> 0) {
2509 * Can't transition to the protocol's next state, wait to
2510 * receive the rest of the header.
2512 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2513 state
->received
, state
->left_to_receive
,
2518 /* Transition to next state: receiving the command's payload. */
2519 conn
->protocol
.ctrl
.state_id
=
2520 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
2521 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
2522 header
.circuit_id
= be64toh(header
.circuit_id
);
2523 header
.data_size
= be64toh(header
.data_size
);
2524 header
.cmd
= be32toh(header
.cmd
);
2525 header
.cmd_version
= be32toh(header
.cmd_version
);
2526 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
2527 &header
, sizeof(header
));
2529 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
2530 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
2533 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
2534 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
2536 status
= RELAY_CONNECTION_STATUS_ERROR
;
2540 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
2542 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
2543 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
2546 status
= RELAY_CONNECTION_STATUS_ERROR
;
2550 if (header
.data_size
== 0) {
2552 * Manually invoke the next state as the poll loop
2553 * will not wake-up to allow us to proceed further.
2555 status
= relay_process_control_receive_payload(conn
);
2562 * Process the commands received on the control socket
2564 static enum relay_connection_status
relay_process_control(
2565 struct relay_connection
*conn
)
2567 enum relay_connection_status status
;
2569 switch (conn
->protocol
.ctrl
.state_id
) {
2570 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
2571 status
= relay_process_control_receive_header(conn
);
2573 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
2574 status
= relay_process_control_receive_payload(conn
);
2577 ERR("Unknown control connection protocol state encountered.");
2585 * Handle index for a data stream.
2587 * Called with the stream lock held.
2589 * Return 0 on success else a negative value.
2591 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2595 uint64_t data_offset
;
2596 struct relay_index
*index
;
2598 /* Get data offset because we are about to update the index. */
2599 data_offset
= htobe64(stream
->tracefile_size_current
);
2601 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
2602 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
2605 * Lookup for an existing index for that stream id/sequence
2606 * number. If it exists, the control thread has already received the
2607 * data for it, thus we need to write it to disk.
2609 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
2615 if (rotate_index
|| !stream
->index_file
) {
2616 uint32_t major
, minor
;
2618 /* Put ref on previous index_file. */
2619 if (stream
->index_file
) {
2620 lttng_index_file_put(stream
->index_file
);
2621 stream
->index_file
= NULL
;
2623 major
= stream
->trace
->session
->major
;
2624 minor
= stream
->trace
->session
->minor
;
2625 stream
->index_file
= lttng_index_file_create(stream
->path_name
,
2626 stream
->channel_name
,
2627 -1, -1, stream
->tracefile_size
,
2628 tracefile_array_get_file_index_head(stream
->tfa
),
2629 lttng_to_index_major(major
, minor
),
2630 lttng_to_index_minor(major
, minor
));
2631 if (!stream
->index_file
) {
2633 /* Put self-ref for this index due to error. */
2634 relay_index_put(index
);
2640 if (relay_index_set_file(index
, stream
->index_file
, data_offset
)) {
2642 /* Put self-ref for this index due to error. */
2643 relay_index_put(index
);
2648 ret
= relay_index_try_flush(index
);
2650 tracefile_array_commit_seq(stream
->tfa
);
2651 stream
->index_received_seqcount
++;
2652 } else if (ret
> 0) {
2656 /* Put self-ref for this index due to error. */
2657 relay_index_put(index
);
2665 static enum relay_connection_status
relay_process_data_receive_header(
2666 struct relay_connection
*conn
)
2669 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2670 struct data_connection_state_receive_header
*state
=
2671 &conn
->protocol
.data
.state
.receive_header
;
2672 struct lttcomm_relayd_data_hdr header
;
2673 struct relay_stream
*stream
;
2675 assert(state
->left_to_receive
!= 0);
2677 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2678 state
->header_reception_buffer
+ state
->received
,
2679 state
->left_to_receive
, MSG_DONTWAIT
);
2681 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2682 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2683 status
= RELAY_CONNECTION_STATUS_ERROR
;
2686 } else if (ret
== 0) {
2687 /* Orderly shutdown. Not necessary to print an error. */
2688 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2689 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2694 assert(ret
<= state
->left_to_receive
);
2696 state
->left_to_receive
-= ret
;
2697 state
->received
+= ret
;
2699 if (state
->left_to_receive
> 0) {
2701 * Can't transition to the protocol's next state, wait to
2702 * receive the rest of the header.
2704 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2705 state
->received
, state
->left_to_receive
,
2711 /* Transition to next state: receiving the payload. */
2712 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
2714 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
2715 header
.circuit_id
= be64toh(header
.circuit_id
);
2716 header
.stream_id
= be64toh(header
.stream_id
);
2717 header
.data_size
= be32toh(header
.data_size
);
2718 header
.net_seq_num
= be64toh(header
.net_seq_num
);
2719 header
.padding_size
= be32toh(header
.padding_size
);
2720 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
2722 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
2724 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
2725 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
2727 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
2728 conn
->sock
->fd
, header
.circuit_id
,
2729 header
.stream_id
, header
.data_size
,
2730 header
.net_seq_num
, header
.padding_size
);
2732 stream
= stream_get_by_id(header
.stream_id
);
2734 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
2736 /* Protocol error. */
2737 status
= RELAY_CONNECTION_STATUS_ERROR
;
2741 pthread_mutex_lock(&stream
->lock
);
2743 /* Check if a rotation is needed. */
2744 if (stream
->tracefile_size
> 0 &&
2745 (stream
->tracefile_size_current
+ header
.data_size
) >
2746 stream
->tracefile_size
) {
2747 uint64_t old_id
, new_id
;
2749 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
2750 tracefile_array_file_rotate(stream
->tfa
);
2752 /* new_id is updated by utils_rotate_stream_file. */
2755 ret
= stream_fd_rotate(stream
->stream_fd
, stream
->path_name
,
2756 stream
->channel_name
, stream
->tracefile_size
,
2757 stream
->tracefile_count
, &new_id
);
2759 ERR("Failed to rotate stream output file");
2760 status
= RELAY_CONNECTION_STATUS_ERROR
;
2761 goto end_stream_unlock
;
2765 * Reset current size because we just performed a stream
2768 stream
->tracefile_size_current
= 0;
2769 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= true;
2774 pthread_mutex_unlock(&stream
->lock
);
2780 static enum relay_connection_status
relay_process_data_receive_payload(
2781 struct relay_connection
*conn
)
2784 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2785 struct relay_stream
*stream
;
2786 struct data_connection_state_receive_payload
*state
=
2787 &conn
->protocol
.data
.state
.receive_payload
;
2788 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
2789 char data_buffer
[chunk_size
];
2790 bool partial_recv
= false;
2791 bool new_stream
= false, close_requested
= false;
2792 uint64_t left_to_receive
= state
->left_to_receive
;
2793 struct relay_session
*session
;
2796 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
2797 state
->header
.stream_id
, state
->header
.net_seq_num
,
2798 state
->received
, left_to_receive
);
2800 stream
= stream_get_by_id(state
->header
.stream_id
);
2802 /* Protocol error. */
2803 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
2804 state
->header
.stream_id
);
2805 status
= RELAY_CONNECTION_STATUS_ERROR
;
2809 pthread_mutex_lock(&stream
->lock
);
2810 session
= stream
->trace
->session
;
2811 if (!conn
->session
) {
2812 ret
= connection_set_session(conn
, session
);
2814 status
= RELAY_CONNECTION_STATUS_ERROR
;
2815 goto end_stream_unlock
;
2819 stream_fd
= stream_fd_get_fd(stream
->stream_fd
);
2820 if (stream_fd
< 0) {
2821 status
= RELAY_CONNECTION_STATUS_ERROR
;
2822 goto end_stream_unlock
;
2826 * The size of the "chunk" received on any iteration is bounded by:
2827 * - the data left to receive,
2828 * - the data immediately available on the socket,
2829 * - the on-stack data buffer
2831 while (left_to_receive
> 0 && !partial_recv
) {
2833 size_t recv_size
= min(left_to_receive
, chunk_size
);
2835 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
2836 recv_size
, MSG_DONTWAIT
);
2838 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2839 PERROR("Socket %d error", conn
->sock
->fd
);
2840 status
= RELAY_CONNECTION_STATUS_ERROR
;
2843 } else if (ret
== 0) {
2844 /* No more data ready to be consumed on socket. */
2845 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
2846 state
->header
.stream_id
);
2847 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2849 } else if (ret
< (int) recv_size
) {
2851 * All the data available on the socket has been
2854 partial_recv
= true;
2859 /* Write data to stream output fd. */
2860 write_ret
= lttng_write(stream_fd
, data_buffer
,
2862 if (write_ret
< (ssize_t
) recv_size
) {
2863 ERR("Relay error writing data to file");
2864 status
= RELAY_CONNECTION_STATUS_ERROR
;
2868 left_to_receive
-= recv_size
;
2869 state
->received
+= recv_size
;
2870 state
->left_to_receive
= left_to_receive
;
2872 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
2873 write_ret
, stream
->stream_handle
);
2876 if (state
->left_to_receive
> 0) {
2878 * Did not receive all the data expected, wait for more data to
2879 * become available on the socket.
2881 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
2882 state
->header
.stream_id
, state
->received
,
2883 state
->left_to_receive
);
2887 ret
= write_padding_to_file(stream_fd
,
2888 state
->header
.padding_size
);
2889 if ((int64_t) ret
< (int64_t) state
->header
.padding_size
) {
2890 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2891 stream
->stream_handle
,
2892 state
->header
.net_seq_num
, ret
);
2893 status
= RELAY_CONNECTION_STATUS_ERROR
;
2898 if (session
->minor
>= 4 && !session
->snapshot
) {
2899 ret
= handle_index_data(stream
, state
->header
.net_seq_num
,
2900 state
->rotate_index
);
2902 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2903 stream
->stream_handle
,
2904 state
->header
.net_seq_num
, ret
);
2905 status
= RELAY_CONNECTION_STATUS_ERROR
;
2910 stream
->tracefile_size_current
+= state
->header
.data_size
+
2911 state
->header
.padding_size
;
2913 if (stream
->prev_seq
== -1ULL) {
2917 stream
->prev_seq
= state
->header
.net_seq_num
;
2920 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
2921 * contents of *state which are aliased (union) to the same location as
2922 * the new state. Don't use it beyond this point.
2924 connection_reset_protocol_state(conn
);
2928 stream_fd_put_fd(stream
->stream_fd
);
2930 close_requested
= stream
->close_requested
;
2931 pthread_mutex_unlock(&stream
->lock
);
2932 if (close_requested
&& left_to_receive
== 0) {
2933 try_stream_close(stream
);
2937 pthread_mutex_lock(&session
->lock
);
2938 uatomic_set(&session
->new_streams
, 1);
2939 pthread_mutex_unlock(&session
->lock
);
2948 * relay_process_data: Process the data received on the data socket
2950 static enum relay_connection_status
relay_process_data(
2951 struct relay_connection
*conn
)
2953 enum relay_connection_status status
;
2955 switch (conn
->protocol
.data
.state_id
) {
2956 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
2957 status
= relay_process_data_receive_header(conn
);
2959 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
2960 status
= relay_process_data_receive_payload(conn
);
2963 ERR("Unexpected data connection communication state.");
2970 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2974 (void) lttng_poll_del(events
, pollfd
);
2976 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
, &pollfd
, 1,
2977 fd_tracker_util_close_fd
, NULL
);
2979 ERR("Closing pollfd %d", pollfd
);
2983 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
2984 int pollfd
, struct relay_connection
*conn
)
2986 const char *type_str
;
2988 switch (conn
->type
) {
2993 type_str
= "Control";
2995 case RELAY_VIEWER_COMMAND
:
2996 type_str
= "Viewer Command";
2998 case RELAY_VIEWER_NOTIFICATION
:
2999 type_str
= "Viewer Notification";
3002 type_str
= "Unknown";
3004 cleanup_connection_pollfd(events
, pollfd
);
3005 connection_put(conn
);
3006 DBG("%s connection closed with %d", type_str
, pollfd
);
3010 * This thread does the actual work
3012 static void *relay_thread_worker(void *data
)
3014 int ret
, err
= -1, last_seen_data_fd
= -1;
3016 struct lttng_poll_event events
;
3017 struct lttng_ht
*relay_connections_ht
;
3018 struct lttng_ht_iter iter
;
3019 struct relay_connection
*destroy_conn
= NULL
;
3021 DBG("[thread] Relay worker started");
3023 rcu_register_thread();
3025 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3027 if (testpoint(relayd_thread_worker
)) {
3028 goto error_testpoint
;
3031 health_code_update();
3033 /* table of connections indexed on socket */
3034 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3035 if (!relay_connections_ht
) {
3036 goto relay_connections_ht_error
;
3039 ret
= create_named_thread_poll_set(&events
, 2, "Worker thread epoll");
3041 goto error_poll_create
;
3044 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3051 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3053 health_code_update();
3055 /* Infinite blocking call, waiting for transmission */
3056 DBG3("Relayd worker thread polling...");
3057 health_poll_entry();
3058 ret
= lttng_poll_wait(&events
, -1);
3062 * Restart interrupted system call.
3064 if (errno
== EINTR
) {
3073 * Process control. The control connection is
3074 * prioritized so we don't starve it with high
3075 * throughput tracing data on the data connection.
3077 for (i
= 0; i
< nb_fd
; i
++) {
3078 /* Fetch once the poll data */
3079 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3080 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3082 health_code_update();
3086 * No activity for this FD (poll
3092 /* Thread quit pipe has been closed. Killing thread. */
3093 ret
= check_thread_quit_pipe(pollfd
, revents
);
3099 /* Inspect the relay conn pipe for new connection */
3100 if (pollfd
== relay_conn_pipe
[0]) {
3101 if (revents
& LPOLLIN
) {
3102 struct relay_connection
*conn
;
3104 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3108 lttng_poll_add(&events
, conn
->sock
->fd
,
3109 LPOLLIN
| LPOLLRDHUP
);
3110 connection_ht_add(relay_connections_ht
, conn
);
3111 DBG("Connection socket %d added", conn
->sock
->fd
);
3112 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3113 ERR("Relay connection pipe error");
3116 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3120 struct relay_connection
*ctrl_conn
;
3122 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3123 /* If not found, there is a synchronization issue. */
3126 if (ctrl_conn
->type
== RELAY_DATA
) {
3127 if (revents
& LPOLLIN
) {
3129 * Flag the last seen data fd not deleted. It will be
3130 * used as the last seen fd if any fd gets deleted in
3133 last_notdel_data_fd
= pollfd
;
3135 goto put_ctrl_connection
;
3137 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3139 if (revents
& LPOLLIN
) {
3140 enum relay_connection_status status
;
3142 status
= relay_process_control(ctrl_conn
);
3143 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3145 * On socket error flag the session as aborted to force
3146 * the cleanup of its stream otherwise it can leak
3147 * during the lifetime of the relayd.
3149 * This prevents situations in which streams can be
3150 * left opened because an index was received, the
3151 * control connection is closed, and the data
3152 * connection is closed (uncleanly) before the packet's
3155 * Since the control connection encountered an error,
3156 * it is okay to be conservative and close the
3157 * session right now as we can't rely on the protocol
3158 * being respected anymore.
3160 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3161 session_abort(ctrl_conn
->session
);
3164 /* Clear the connection on error or close. */
3165 relay_thread_close_connection(&events
,
3170 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3171 relay_thread_close_connection(&events
,
3173 if (last_seen_data_fd
== pollfd
) {
3174 last_seen_data_fd
= last_notdel_data_fd
;
3177 ERR("Unexpected poll events %u for control sock %d",
3179 connection_put(ctrl_conn
);
3182 put_ctrl_connection
:
3183 connection_put(ctrl_conn
);
3188 * The last loop handled a control request, go back to poll to make
3189 * sure we prioritise the control socket.
3195 if (last_seen_data_fd
>= 0) {
3196 for (i
= 0; i
< nb_fd
; i
++) {
3197 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3199 health_code_update();
3201 if (last_seen_data_fd
== pollfd
) {
3208 /* Process data connection. */
3209 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3210 /* Fetch the poll data. */
3211 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3212 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3213 struct relay_connection
*data_conn
;
3215 health_code_update();
3218 /* No activity for this FD (poll implementation). */
3222 /* Skip the command pipe. It's handled in the first loop. */
3223 if (pollfd
== relay_conn_pipe
[0]) {
3227 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3229 /* Skip it. Might be removed before. */
3232 if (data_conn
->type
== RELAY_CONTROL
) {
3233 goto put_data_connection
;
3235 assert(data_conn
->type
== RELAY_DATA
);
3237 if (revents
& LPOLLIN
) {
3238 enum relay_connection_status status
;
3240 status
= relay_process_data(data_conn
);
3241 /* Connection closed or error. */
3242 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3244 * On socket error flag the session as aborted to force
3245 * the cleanup of its stream otherwise it can leak
3246 * during the lifetime of the relayd.
3248 * This prevents situations in which streams can be
3249 * left opened because an index was received, the
3250 * control connection is closed, and the data
3251 * connection is closed (uncleanly) before the packet's
3254 * Since the data connection encountered an error,
3255 * it is okay to be conservative and close the
3256 * session right now as we can't rely on the protocol
3257 * being respected anymore.
3259 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3260 session_abort(data_conn
->session
);
3262 relay_thread_close_connection(&events
, pollfd
,
3265 * Every goto restart call sets the last seen fd where
3266 * here we don't really care since we gracefully
3267 * continue the loop after the connection is deleted.
3270 /* Keep last seen port. */
3271 last_seen_data_fd
= pollfd
;
3272 connection_put(data_conn
);
3275 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3276 relay_thread_close_connection(&events
, pollfd
,
3279 ERR("Unknown poll events %u for data sock %d",
3282 put_data_connection
:
3283 connection_put(data_conn
);
3285 last_seen_data_fd
= -1;
3288 /* Normal exit, no error */
3293 /* Cleanup reamaining connection object. */
3295 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3298 health_code_update();
3300 session_abort(destroy_conn
->session
);
3303 * No need to grab another ref, because we own
3306 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
3311 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
3313 lttng_ht_destroy(relay_connections_ht
);
3314 relay_connections_ht_error
:
3315 /* Close relay conn pipes */
3316 (void) fd_tracker_util_pipe_close(the_fd_tracker
,
3319 DBG("Thread exited with error");
3321 DBG("Worker thread cleanup complete");
3325 ERR("Health error occurred in %s", __func__
);
3327 health_unregister(health_relayd
);
3328 rcu_unregister_thread();
3329 lttng_relay_stop_threads();
3334 * Create the relay command pipe to wake thread_manage_apps.
3335 * Closed in cleanup().
3337 static int create_relay_conn_pipe(void)
3339 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
3340 "Relayd connection pipe", relay_conn_pipe
);
3344 int stdio_open(void *data
, int *fds
)
3346 fds
[0] = fileno(stdout
);
3347 fds
[1] = fileno(stderr
);
3352 int noop_close(void *data
, int *fds
)
3358 int track_stdio(void)
3361 const char *names
[] = { "stdout", "stderr" };
3363 return fd_tracker_open_unsuspendable_fd(the_fd_tracker
, fds
,
3364 names
, 2, stdio_open
, NULL
);
3368 void untrack_stdio(void)
3370 int fds
[] = { fileno(stdout
), fileno(stderr
) };
3373 * noop_close is used since we don't really want to close
3374 * the stdio output fds; we merely want to stop tracking them.
3376 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
3377 fds
, 2, noop_close
, NULL
);
3383 int main(int argc
, char **argv
)
3385 int ret
= 0, retval
= 0;
3388 /* Parse environment variables */
3389 parse_env_options();
3393 * Command line arguments overwrite environment.
3396 if (set_options(argc
, argv
)) {
3401 if (set_signal_handler()) {
3406 relayd_config_log();
3408 if (opt_print_version
) {
3414 ret
= fclose(stdin
);
3416 PERROR("Failed to close stdin");
3419 /* Try to create directory if -o, --output is specified. */
3420 if (opt_output_path
) {
3421 if (*opt_output_path
!= '/') {
3422 ERR("Please specify an absolute path for -o, --output PATH");
3427 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
3430 ERR("Unable to create %s", opt_output_path
);
3437 if (opt_daemon
|| opt_background
) {
3438 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
3446 if (opt_working_directory
) {
3447 ret
= utils_change_working_dir(opt_working_directory
);
3449 ERR("Changing working directory");
3454 * The RCU thread registration (and use, through the fd-tracker's
3455 * creation) is done after the daemonization to allow us to not
3456 * deal with liburcu's fork() management as the call RCU needs to
3459 rcu_register_thread();
3461 the_fd_tracker
= fd_tracker_create(lttng_opt_fd_cap
);
3462 if (!the_fd_tracker
) {
3467 ret
= track_stdio();
3473 /* Initialize thread health monitoring */
3474 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
3475 if (!health_relayd
) {
3476 PERROR("health_app_create error");
3478 goto exit_health_app_create
;
3481 /* Create thread quit pipe */
3482 if (init_thread_quit_pipe()) {
3484 goto exit_init_data
;
3487 /* Setup the thread apps communication pipe. */
3488 if (create_relay_conn_pipe()) {
3490 goto exit_init_data
;
3493 /* Init relay command queue. */
3494 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
3496 /* Initialize communication library */
3498 lttcomm_inet_init();
3500 /* tables of sessions indexed by session ID */
3501 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3504 goto exit_init_data
;
3507 /* tables of streams indexed by stream ID */
3508 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3509 if (!relay_streams_ht
) {
3511 goto exit_init_data
;
3514 /* tables of streams indexed by stream ID */
3515 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3516 if (!viewer_streams_ht
) {
3518 goto exit_init_data
;
3521 ret
= init_health_quit_pipe();
3524 goto exit_health_quit_pipe
;
3527 /* Create thread to manage the client socket */
3528 ret
= pthread_create(&health_thread
, default_pthread_attr(),
3529 thread_manage_health
, (void *) NULL
);
3532 PERROR("pthread_create health");
3534 goto exit_health_thread
;
3537 /* Setup the dispatcher thread */
3538 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
3539 relay_thread_dispatcher
, (void *) NULL
);
3542 PERROR("pthread_create dispatcher");
3544 goto exit_dispatcher_thread
;
3547 /* Setup the worker thread */
3548 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
3549 relay_thread_worker
, NULL
);
3552 PERROR("pthread_create worker");
3554 goto exit_worker_thread
;
3557 /* Setup the listener thread */
3558 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
3559 relay_thread_listener
, (void *) NULL
);
3562 PERROR("pthread_create listener");
3564 goto exit_listener_thread
;
3567 ret
= relayd_live_create(live_uri
);
3569 ERR("Starting live viewer threads");
3575 * This is where we start awaiting program completion (e.g. through
3576 * signal that asks threads to teardown).
3579 ret
= relayd_live_join();
3585 ret
= pthread_join(listener_thread
, &status
);
3588 PERROR("pthread_join listener_thread");
3592 exit_listener_thread
:
3593 ret
= pthread_join(worker_thread
, &status
);
3596 PERROR("pthread_join worker_thread");
3601 ret
= pthread_join(dispatcher_thread
, &status
);
3604 PERROR("pthread_join dispatcher_thread");
3607 exit_dispatcher_thread
:
3609 ret
= pthread_join(health_thread
, &status
);
3612 PERROR("pthread_join health_thread");
3617 (void) fd_tracker_util_pipe_close(the_fd_tracker
, health_quit_pipe
);
3618 exit_health_quit_pipe
:
3621 health_app_destroy(health_relayd
);
3622 exit_health_app_create
:
3625 * Wait for all pending call_rcu work to complete before tearing
3626 * down data structures. call_rcu worker may be trying to
3627 * perform lookups in those structures.
3632 /* Ensure all prior call_rcu are done. */
3637 * fd_tracker_destroy() will log the contents of the fd-tracker
3638 * if a leak is detected.
3640 fd_tracker_destroy(the_fd_tracker
);
3641 rcu_unregister_thread();