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>
41 #include <urcu/rculist.h>
47 #include <lttng/lttng.h>
48 #include <common/common.h>
49 #include <common/compat/poll.h>
50 #include <common/compat/socket.h>
51 #include <common/compat/endian.h>
52 #include <common/compat/getenv.h>
53 #include <common/defaults.h>
54 #include <common/daemonize.h>
55 #include <common/futex.h>
56 #include <common/sessiond-comm/sessiond-comm.h>
57 #include <common/sessiond-comm/inet.h>
58 #include <common/sessiond-comm/relayd.h>
59 #include <common/uri.h>
60 #include <common/utils.h>
61 #include <common/align.h>
62 #include <common/config/session-config.h>
63 #include <common/dynamic-buffer.h>
64 #include <common/buffer-view.h>
65 #include <common/string-utils/format.h>
66 #include <common/fd-tracker/fd-tracker.h>
67 #include <common/fd-tracker/utils.h>
69 #include "backward-compatibility-group-by.h"
71 #include "connection.h"
72 #include "ctf-trace.h"
73 #include "health-relayd.h"
76 #include "lttng-relayd.h"
78 #include "sessiond-trace-chunks.h"
80 #include "tcp_keep_alive.h"
81 #include "testpoint.h"
82 #include "tracefile-array.h"
85 #include "viewer-stream.h"
87 static const char *help_msg
=
88 #ifdef LTTNG_EMBED_HELP
89 #include <lttng-relayd.8.h>
95 enum relay_connection_status
{
96 RELAY_CONNECTION_STATUS_OK
,
97 /* An error occurred while processing an event on the connection. */
98 RELAY_CONNECTION_STATUS_ERROR
,
99 /* Connection closed/shutdown cleanly. */
100 RELAY_CONNECTION_STATUS_CLOSED
,
103 /* command line options */
104 char *opt_output_path
, *opt_working_directory
;
105 static int opt_daemon
, opt_background
, opt_print_version
, opt_allow_clear
= 1;
106 enum relay_group_output_by opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
;
109 * We need to wait for listener and live listener threads, as well as
110 * health check thread, before being ready to signal readiness.
112 #define NR_LTTNG_RELAY_READY 3
113 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
115 /* Size of receive buffer. */
116 #define RECV_DATA_BUFFER_SIZE 65536
118 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
119 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
121 static struct lttng_uri
*control_uri
;
122 static struct lttng_uri
*data_uri
;
123 static struct lttng_uri
*live_uri
;
125 const char *progname
;
127 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
128 static int tracing_group_name_override
;
130 const char * const config_section_name
= "relayd";
133 * Quit pipe for all threads. This permits a single cancellation point
134 * for all threads when receiving an event on the pipe.
136 int thread_quit_pipe
[2] = { -1, -1 };
139 * This pipe is used to inform the worker thread that a command is queued and
140 * ready to be processed.
142 static int relay_conn_pipe
[2] = { -1, -1 };
144 /* Shared between threads */
145 static int dispatch_thread_exit
;
147 static pthread_t listener_thread
;
148 static pthread_t dispatcher_thread
;
149 static pthread_t worker_thread
;
150 static pthread_t health_thread
;
153 * last_relay_stream_id_lock protects last_relay_stream_id increment
154 * atomicity on 32-bit architectures.
156 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
157 static uint64_t last_relay_stream_id
;
160 * Relay command queue.
162 * The relay_thread_listener and relay_thread_dispatcher communicate with this
165 static struct relay_conn_queue relay_conn_queue
;
167 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
168 static unsigned int lttng_opt_fd_pool_size
= -1;
170 /* Global relay stream hash table. */
171 struct lttng_ht
*relay_streams_ht
;
173 /* Global relay viewer stream hash table. */
174 struct lttng_ht
*viewer_streams_ht
;
176 /* Global relay sessions hash table. */
177 struct lttng_ht
*sessions_ht
;
179 /* Relayd health monitoring */
180 struct health_app
*health_relayd
;
182 struct sessiond_trace_chunk_registry
*sessiond_trace_chunk_registry
;
184 /* Global fd tracker. */
185 struct fd_tracker
*the_fd_tracker
;
187 static struct option long_options
[] = {
188 { "control-port", 1, 0, 'C', },
189 { "data-port", 1, 0, 'D', },
190 { "live-port", 1, 0, 'L', },
191 { "daemonize", 0, 0, 'd', },
192 { "background", 0, 0, 'b', },
193 { "group", 1, 0, 'g', },
194 { "fd-pool-size", 1, 0, '\0', },
195 { "help", 0, 0, 'h', },
196 { "output", 1, 0, 'o', },
197 { "verbose", 0, 0, 'v', },
198 { "config", 1, 0, 'f' },
199 { "version", 0, 0, 'V' },
200 { "working-directory", 1, 0, 'w', },
201 { "group-output-by-session", 0, 0, 's', },
202 { "group-output-by-host", 0, 0, 'p', },
203 { "disallow-clear", 0, 0, 'x' },
207 static const char *config_ignore_options
[] = { "help", "config", "version" };
209 static void print_version(void) {
210 fprintf(stdout
, "%s\n", VERSION
);
213 static void relayd_config_log(void)
215 DBG("LTTng-relayd " VERSION
" - " VERSION_NAME
"%s%s",
216 GIT_VERSION
[0] == '\0' ? "" : " - " GIT_VERSION
,
217 EXTRA_VERSION_NAME
[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME
);
218 if (EXTRA_VERSION_DESCRIPTION
[0] != '\0') {
219 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION
"\n");
221 if (EXTRA_VERSION_PATCHES
[0] != '\0') {
222 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES
"\n");
227 * Take an option from the getopt output and set it in the right variable to be
230 * Return 0 on success else a negative value.
232 static int set_option(int opt
, const char *arg
, const char *optname
)
238 if (!strcmp(optname
, "fd-pool-size")) {
242 v
= strtoul(arg
, NULL
, 0);
243 if (errno
!= 0 || !isdigit(arg
[0])) {
244 ERR("Wrong value in --fd-pool-size parameter: %s", arg
);
249 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s", arg
);
253 lttng_opt_fd_pool_size
= (unsigned int) v
;
255 fprintf(stderr
, "unknown option %s", optname
);
257 fprintf(stderr
, " with arg %s\n", arg
);
262 if (lttng_is_setuid_setgid()) {
263 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
264 "-C, --control-port");
266 ret
= uri_parse(arg
, &control_uri
);
268 ERR("Invalid control URI specified");
271 if (control_uri
->port
== 0) {
272 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
277 if (lttng_is_setuid_setgid()) {
278 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
281 ret
= uri_parse(arg
, &data_uri
);
283 ERR("Invalid data URI specified");
286 if (data_uri
->port
== 0) {
287 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
292 if (lttng_is_setuid_setgid()) {
293 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
296 ret
= uri_parse(arg
, &live_uri
);
298 ERR("Invalid live URI specified");
301 if (live_uri
->port
== 0) {
302 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
313 if (lttng_is_setuid_setgid()) {
314 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
317 tracing_group_name
= strdup(arg
);
318 if (tracing_group_name
== NULL
) {
323 tracing_group_name_override
= 1;
327 ret
= utils_show_help(8, "lttng-relayd", help_msg
);
329 ERR("Cannot show --help for `lttng-relayd`");
334 opt_print_version
= 1;
337 if (lttng_is_setuid_setgid()) {
338 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
341 ret
= asprintf(&opt_output_path
, "%s", arg
);
344 PERROR("asprintf opt_output_path");
350 if (lttng_is_setuid_setgid()) {
351 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
352 "-w, --working-directory");
354 ret
= asprintf(&opt_working_directory
, "%s", arg
);
357 PERROR("asprintf opt_working_directory");
364 /* Verbose level can increase using multiple -v */
366 lttng_opt_verbose
= config_parse_value(arg
);
368 /* Only 3 level of verbosity (-vvv). */
369 if (lttng_opt_verbose
< 3) {
370 lttng_opt_verbose
+= 1;
375 if (opt_group_output_by
!= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
376 ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
379 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_SESSION
;
382 if (opt_group_output_by
!= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
383 ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
386 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_HOST
;
393 /* Unknown option or other error.
394 * Error is printed by getopt, just return */
407 * config_entry_handler_cb used to handle options read from a config file.
408 * See config_entry_handler_cb comment in common/config/session-config.h for the
409 * return value conventions.
411 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
415 if (!entry
|| !entry
->name
|| !entry
->value
) {
420 /* Check if the option is to be ignored */
421 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
422 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
427 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
428 /* Ignore if entry name is not fully matched. */
429 if (strcmp(entry
->name
, long_options
[i
].name
)) {
434 * If the option takes no argument on the command line,
435 * we have to check if the value is "true". We support
436 * non-zero numeric values, true, on and yes.
438 if (!long_options
[i
].has_arg
) {
439 ret
= config_parse_value(entry
->value
);
442 WARN("Invalid configuration value \"%s\" for option %s",
443 entry
->value
, entry
->name
);
445 /* False, skip boolean config option. */
450 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
454 WARN("Unrecognized option \"%s\" in daemon configuration file.",
461 static int parse_env_options(void)
466 value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV
);
468 opt_working_directory
= strdup(value
);
469 if (!opt_working_directory
) {
470 ERR("Failed to allocate working directory string (\"%s\")",
478 static int set_fd_pool_size(void)
481 struct rlimit rlimit
;
483 ret
= getrlimit(RLIMIT_NOFILE
, &rlimit
);
485 PERROR("Failed to get file descriptor limit");
490 DBG("File descriptor count limits are %" PRIu64
" (soft) and %" PRIu64
" (hard)",
491 (uint64_t) rlimit
.rlim_cur
,
492 (uint64_t) rlimit
.rlim_max
);
493 if (lttng_opt_fd_pool_size
== -1) {
494 /* Use default value (soft limit - reserve). */
495 if (rlimit
.rlim_cur
< DEFAULT_RELAYD_MIN_FD_POOL_SIZE
) {
496 ERR("The process' file number limit is too low (%" PRIu64
"). The process' file number limit must be set to at least %i.",
497 (uint64_t) rlimit
.rlim_cur
, DEFAULT_RELAYD_MIN_FD_POOL_SIZE
);
501 lttng_opt_fd_pool_size
= rlimit
.rlim_cur
-
502 DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
;
506 if (lttng_opt_fd_pool_size
< DEFAULT_RELAYD_MIN_FD_POOL_SIZE
) {
507 ERR("File descriptor pool size must be set to at least %d",
508 DEFAULT_RELAYD_MIN_FD_POOL_SIZE
);
513 if (lttng_opt_fd_pool_size
> rlimit
.rlim_cur
) {
514 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64
").",
515 lttng_opt_fd_pool_size
, (uint64_t) rlimit
.rlim_cur
);
520 DBG("File descriptor pool size argument (%u) adjusted to %u to accomodate transient fd uses",
521 lttng_opt_fd_pool_size
,
522 lttng_opt_fd_pool_size
- DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
);
523 lttng_opt_fd_pool_size
-= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
;
528 static int set_options(int argc
, char **argv
)
530 int c
, ret
= 0, option_index
= 0, retval
= 0;
531 int orig_optopt
= optopt
, orig_optind
= optind
;
532 char *default_address
, *optstring
;
533 const char *config_path
= NULL
;
535 optstring
= utils_generate_optstring(long_options
,
536 sizeof(long_options
) / sizeof(struct option
));
542 /* Check for the --config option */
544 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
545 &option_index
)) != -1) {
549 } else if (c
!= 'f') {
553 if (lttng_is_setuid_setgid()) {
554 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
557 config_path
= utils_expand_path(optarg
);
559 ERR("Failed to resolve path: %s", optarg
);
564 ret
= config_get_section_entries(config_path
, config_section_name
,
565 config_entry_handler
, NULL
);
568 ERR("Invalid configuration option at line %i", ret
);
574 /* Reset getopt's global state */
575 optopt
= orig_optopt
;
576 optind
= orig_optind
;
578 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
583 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
590 /* assign default values */
591 if (control_uri
== NULL
) {
592 ret
= asprintf(&default_address
,
593 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
594 DEFAULT_NETWORK_CONTROL_PORT
);
596 PERROR("asprintf default data address");
601 ret
= uri_parse(default_address
, &control_uri
);
602 free(default_address
);
604 ERR("Invalid control URI specified");
609 if (data_uri
== NULL
) {
610 ret
= asprintf(&default_address
,
611 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
612 DEFAULT_NETWORK_DATA_PORT
);
614 PERROR("asprintf default data address");
619 ret
= uri_parse(default_address
, &data_uri
);
620 free(default_address
);
622 ERR("Invalid data URI specified");
627 if (live_uri
== NULL
) {
628 ret
= asprintf(&default_address
,
629 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
630 DEFAULT_NETWORK_VIEWER_PORT
);
632 PERROR("asprintf default viewer control address");
637 ret
= uri_parse(default_address
, &live_uri
);
638 free(default_address
);
640 ERR("Invalid viewer control URI specified");
645 ret
= set_fd_pool_size();
651 if (opt_group_output_by
== RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
652 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_HOST
;
654 if (opt_allow_clear
) {
655 /* Check if env variable exists. */
656 const char *value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV
);
658 ret
= config_parse_value(value
);
660 ERR("Invalid value for %s specified", DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV
);
664 opt_allow_clear
= !ret
;
673 static void print_global_objects(void)
675 print_viewer_streams();
676 print_relay_streams();
680 static int noop_close(void *data
, int *fds
)
685 static void untrack_stdio(void)
687 int fds
[] = { fileno(stdout
), fileno(stderr
) };
690 * noop_close is used since we don't really want to close
691 * the stdio output fds; we merely want to stop tracking them.
693 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
694 fds
, 2, noop_close
, NULL
);
700 static void relayd_cleanup(void)
702 print_global_objects();
706 if (viewer_streams_ht
)
707 lttng_ht_destroy(viewer_streams_ht
);
708 if (relay_streams_ht
)
709 lttng_ht_destroy(relay_streams_ht
);
711 lttng_ht_destroy(sessions_ht
);
713 free(opt_output_path
);
714 free(opt_working_directory
);
717 health_app_destroy(health_relayd
);
719 /* Close thread quit pipes */
720 if (health_quit_pipe
[0] != -1) {
721 (void) fd_tracker_util_pipe_close(
722 the_fd_tracker
, health_quit_pipe
);
724 if (thread_quit_pipe
[0] != -1) {
725 (void) fd_tracker_util_pipe_close(
726 the_fd_tracker
, thread_quit_pipe
);
728 if (sessiond_trace_chunk_registry
) {
729 sessiond_trace_chunk_registry_destroy(
730 sessiond_trace_chunk_registry
);
732 if (the_fd_tracker
) {
735 * fd_tracker_destroy() will log the contents of the fd-tracker
736 * if a leak is detected.
738 fd_tracker_destroy(the_fd_tracker
);
741 uri_free(control_uri
);
743 /* Live URI is freed in the live thread. */
745 if (tracing_group_name_override
) {
746 free((void *) tracing_group_name
);
751 * Write to writable pipe used to notify a thread.
753 static int notify_thread_pipe(int wpipe
)
757 ret
= lttng_write(wpipe
, "!", 1);
759 PERROR("write poll pipe");
767 static int notify_health_quit_pipe(int *pipe
)
771 ret
= lttng_write(pipe
[1], "4", 1);
773 PERROR("write relay health quit");
782 * Stop all relayd and relayd-live threads.
784 int lttng_relay_stop_threads(void)
788 /* Stopping all threads */
789 DBG("Terminating all threads");
790 if (notify_thread_pipe(thread_quit_pipe
[1])) {
791 ERR("write error on thread quit pipe");
795 if (notify_health_quit_pipe(health_quit_pipe
)) {
796 ERR("write error on health quit pipe");
799 /* Dispatch thread */
800 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
801 futex_nto1_wake(&relay_conn_queue
.futex
);
803 if (relayd_live_stop()) {
804 ERR("Error stopping live threads");
811 * Signal handler for the daemon
813 * Simply stop all worker threads, leaving main() return gracefully after
814 * joining all threads and calling cleanup().
816 static void sighandler(int sig
)
820 DBG("SIGINT caught");
821 if (lttng_relay_stop_threads()) {
822 ERR("Error stopping threads");
826 DBG("SIGTERM caught");
827 if (lttng_relay_stop_threads()) {
828 ERR("Error stopping threads");
832 CMM_STORE_SHARED(recv_child_signal
, 1);
840 * Setup signal handler for :
841 * SIGINT, SIGTERM, SIGPIPE
843 static int set_signal_handler(void)
849 if ((ret
= sigemptyset(&sigset
)) < 0) {
850 PERROR("sigemptyset");
857 sa
.sa_handler
= sighandler
;
858 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
863 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
868 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
873 sa
.sa_handler
= SIG_IGN
;
874 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
879 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
884 void lttng_relay_notify_ready(void)
886 /* Notify the parent of the fork() process that we are ready. */
887 if (opt_daemon
|| opt_background
) {
888 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
889 kill(child_ppid
, SIGUSR1
);
895 * Init thread quit pipe.
897 * Return -1 on error or 0 if all pipes are created.
899 static int init_thread_quit_pipe(void)
901 return fd_tracker_util_pipe_open_cloexec(
902 the_fd_tracker
, "Quit pipe", thread_quit_pipe
);
906 * Init health quit pipe.
908 * Return -1 on error or 0 if all pipes are created.
910 static int init_health_quit_pipe(void)
912 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
913 "Health quit pipe", health_quit_pipe
);
917 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
919 static int create_named_thread_poll_set(struct lttng_poll_event
*events
,
920 int size
, const char *name
)
924 if (events
== NULL
|| size
== 0) {
929 ret
= fd_tracker_util_poll_create(the_fd_tracker
,
930 name
, events
, 1, LTTNG_CLOEXEC
);
933 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
945 * Check if the thread quit pipe was triggered.
947 * Return 1 if it was triggered else 0;
949 static int check_thread_quit_pipe(int fd
, uint32_t events
)
951 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
958 static int create_sock(void *data
, int *out_fd
)
961 struct lttcomm_sock
*sock
= data
;
963 ret
= lttcomm_create_sock(sock
);
973 static int close_sock(void *data
, int *in_fd
)
975 struct lttcomm_sock
*sock
= data
;
977 return sock
->ops
->close(sock
);
980 static int accept_sock(void *data
, int *out_fd
)
983 /* Socks is an array of in_sock, out_sock. */
984 struct lttcomm_sock
**socks
= data
;
985 struct lttcomm_sock
*in_sock
= socks
[0];
987 socks
[1] = in_sock
->ops
->accept(in_sock
);
992 *out_fd
= socks
[1]->fd
;
998 * Create and init socket from uri.
1000 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
,
1004 struct lttcomm_sock
*sock
= NULL
;
1005 char uri_str
[PATH_MAX
];
1006 char *formated_name
= NULL
;
1008 sock
= lttcomm_alloc_sock_from_uri(uri
);
1010 ERR("Allocating socket");
1015 * Don't fail to create the socket if the name can't be built as it is
1016 * only used for debugging purposes.
1018 ret
= uri_to_str_url(uri
, uri_str
, sizeof(uri_str
));
1019 uri_str
[sizeof(uri_str
) - 1] = '\0';
1021 ret
= asprintf(&formated_name
, "%s socket @ %s", name
,
1024 formated_name
= NULL
;
1028 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &sock_fd
,
1029 (const char **) (formated_name
? &formated_name
: NULL
),
1030 1, create_sock
, sock
);
1031 free(formated_name
);
1032 DBG("Listening on %s socket %d", name
, sock
->fd
);
1034 ret
= sock
->ops
->bind(sock
);
1036 PERROR("Failed to bind socket");
1040 ret
= sock
->ops
->listen(sock
, -1);
1050 lttcomm_destroy_sock(sock
);
1056 struct lttcomm_sock
*accept_relayd_sock(struct lttcomm_sock
*listening_sock
,
1060 struct lttcomm_sock
*socks
[2] = { listening_sock
, NULL
};
1061 struct lttcomm_sock
*new_sock
= NULL
;
1063 ret
= fd_tracker_open_unsuspendable_fd(
1064 the_fd_tracker
, &out_fd
,
1065 (const char **) &name
,
1066 1, accept_sock
, &socks
);
1070 new_sock
= socks
[1];
1071 DBG("%s accepted, socket %d", name
, new_sock
->fd
);
1077 * This thread manages the listening for new connections on the network
1079 static void *relay_thread_listener(void *data
)
1081 int i
, ret
, pollfd
, err
= -1;
1082 uint32_t revents
, nb_fd
;
1083 struct lttng_poll_event events
;
1084 struct lttcomm_sock
*control_sock
, *data_sock
;
1086 DBG("[thread] Relay listener started");
1088 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
1090 health_code_update();
1092 control_sock
= relay_socket_create(control_uri
, "Control listener");
1093 if (!control_sock
) {
1094 goto error_sock_control
;
1097 data_sock
= relay_socket_create(data_uri
, "Data listener");
1099 goto error_sock_relay
;
1103 * Pass 3 as size here for the thread quit pipe, control and
1106 ret
= create_named_thread_poll_set(&events
, 3, "Listener thread epoll");
1108 goto error_create_poll
;
1111 /* Add the control socket */
1112 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1114 goto error_poll_add
;
1117 /* Add the data socket */
1118 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1120 goto error_poll_add
;
1123 lttng_relay_notify_ready();
1125 if (testpoint(relayd_thread_listener
)) {
1126 goto error_testpoint
;
1130 health_code_update();
1132 DBG("Listener accepting connections");
1135 health_poll_entry();
1136 ret
= lttng_poll_wait(&events
, -1);
1140 * Restart interrupted system call.
1142 if (errno
== EINTR
) {
1150 DBG("Relay new connection received");
1151 for (i
= 0; i
< nb_fd
; i
++) {
1152 health_code_update();
1154 /* Fetch once the poll data */
1155 revents
= LTTNG_POLL_GETEV(&events
, i
);
1156 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1158 /* Thread quit pipe has been closed. Killing thread. */
1159 ret
= check_thread_quit_pipe(pollfd
, revents
);
1165 if (revents
& LPOLLIN
) {
1167 * A new connection is requested, therefore a
1168 * sessiond/consumerd connection is allocated in
1169 * this thread, enqueued to a global queue and
1170 * dequeued (and freed) in the worker thread.
1173 struct relay_connection
*new_conn
;
1174 struct lttcomm_sock
*newsock
= NULL
;
1175 enum connection_type type
;
1177 if (pollfd
== data_sock
->fd
) {
1179 newsock
= accept_relayd_sock(data_sock
,
1180 "Data socket to relayd");
1182 assert(pollfd
== control_sock
->fd
);
1183 type
= RELAY_CONTROL
;
1184 newsock
= accept_relayd_sock(control_sock
,
1185 "Control socket to relayd");
1188 PERROR("accepting sock");
1192 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
1195 PERROR("setsockopt inet");
1196 lttcomm_destroy_sock(newsock
);
1200 ret
= socket_apply_keep_alive_config(newsock
->fd
);
1202 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1204 lttcomm_destroy_sock(newsock
);
1208 new_conn
= connection_create(newsock
, type
);
1210 lttcomm_destroy_sock(newsock
);
1214 /* Enqueue request for the dispatcher thread. */
1215 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
1219 * Wake the dispatch queue futex.
1220 * Implicit memory barrier with the
1221 * exchange in cds_wfcq_enqueue.
1223 futex_nto1_wake(&relay_conn_queue
.futex
);
1224 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1225 ERR("socket poll error");
1228 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1238 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
1240 if (data_sock
->fd
>= 0) {
1241 int data_sock_fd
= data_sock
->fd
;
1243 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1244 &data_sock_fd
, 1, close_sock
,
1247 PERROR("Failed to close the data listener socket file descriptor");
1251 lttcomm_destroy_sock(data_sock
);
1253 if (control_sock
->fd
>= 0) {
1254 int control_sock_fd
= control_sock
->fd
;
1256 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1257 &control_sock_fd
, 1, close_sock
,
1260 PERROR("Failed to close the control listener socket file descriptor");
1262 control_sock
->fd
= -1;
1264 lttcomm_destroy_sock(control_sock
);
1268 ERR("Health error occurred in %s", __func__
);
1270 health_unregister(health_relayd
);
1271 DBG("Relay listener thread cleanup complete");
1272 lttng_relay_stop_threads();
1277 * This thread manages the dispatching of the requests to worker threads
1279 static void *relay_thread_dispatcher(void *data
)
1283 struct cds_wfcq_node
*node
;
1284 struct relay_connection
*new_conn
= NULL
;
1286 DBG("[thread] Relay dispatcher started");
1288 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
1290 if (testpoint(relayd_thread_dispatcher
)) {
1291 goto error_testpoint
;
1294 health_code_update();
1297 health_code_update();
1299 /* Atomically prepare the queue futex */
1300 futex_nto1_prepare(&relay_conn_queue
.futex
);
1302 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1307 health_code_update();
1309 /* Dequeue commands */
1310 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1311 &relay_conn_queue
.tail
);
1313 DBG("Woken up but nothing in the relay command queue");
1314 /* Continue thread execution */
1317 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1319 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1322 * Inform worker thread of the new request. This
1323 * call is blocking so we can be assured that
1324 * the data will be read at some point in time
1325 * or wait to the end of the world :)
1327 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1329 PERROR("write connection pipe");
1330 connection_put(new_conn
);
1333 } while (node
!= NULL
);
1335 /* Futex wait on queue. Blocking call on futex() */
1336 health_poll_entry();
1337 futex_nto1_wait(&relay_conn_queue
.futex
);
1341 /* Normal exit, no error */
1348 ERR("Health error occurred in %s", __func__
);
1350 health_unregister(health_relayd
);
1351 DBG("Dispatch thread dying");
1352 lttng_relay_stop_threads();
1356 static bool session_streams_have_index(const struct relay_session
*session
)
1358 return session
->minor
>= 4 && !session
->snapshot
;
1362 * Handle the RELAYD_CREATE_SESSION command.
1364 * On success, send back the session id or else return a negative value.
1366 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1367 struct relay_connection
*conn
,
1368 const struct lttng_buffer_view
*payload
)
1372 struct relay_session
*session
= NULL
;
1373 struct lttcomm_relayd_create_session_reply_2_11 reply
= {};
1374 char session_name
[LTTNG_NAME_MAX
] = {};
1375 char hostname
[LTTNG_HOST_NAME_MAX
] = {};
1376 uint32_t live_timer
= 0;
1377 bool snapshot
= false;
1378 bool session_name_contains_creation_timestamp
= false;
1379 /* Left nil for peers < 2.11. */
1380 char base_path
[LTTNG_PATH_MAX
] = {};
1381 lttng_uuid sessiond_uuid
= {};
1382 LTTNG_OPTIONAL(uint64_t) id_sessiond
= {};
1383 LTTNG_OPTIONAL(uint64_t) current_chunk_id
= {};
1384 LTTNG_OPTIONAL(time_t) creation_time
= {};
1385 struct lttng_dynamic_buffer reply_payload
;
1387 lttng_dynamic_buffer_init(&reply_payload
);
1389 if (conn
->minor
< 4) {
1390 /* From 2.1 to 2.3 */
1392 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1393 /* From 2.4 to 2.10 */
1394 ret
= cmd_create_session_2_4(payload
, session_name
,
1395 hostname
, &live_timer
, &snapshot
);
1397 bool has_current_chunk
;
1398 uint64_t current_chunk_id_value
;
1399 time_t creation_time_value
;
1400 uint64_t id_sessiond_value
;
1402 /* From 2.11 to ... */
1403 ret
= cmd_create_session_2_11(payload
, session_name
, hostname
,
1404 base_path
, &live_timer
, &snapshot
, &id_sessiond_value
,
1405 sessiond_uuid
, &has_current_chunk
,
1406 ¤t_chunk_id_value
, &creation_time_value
,
1407 &session_name_contains_creation_timestamp
);
1408 if (lttng_uuid_is_nil(sessiond_uuid
)) {
1409 /* The nil UUID is reserved for pre-2.11 clients. */
1410 ERR("Illegal nil UUID announced by peer in create session command");
1414 LTTNG_OPTIONAL_SET(&id_sessiond
, id_sessiond_value
);
1415 LTTNG_OPTIONAL_SET(&creation_time
, creation_time_value
);
1416 if (has_current_chunk
) {
1417 LTTNG_OPTIONAL_SET(¤t_chunk_id
,
1418 current_chunk_id_value
);
1426 session
= session_create(session_name
, hostname
, base_path
, live_timer
,
1427 snapshot
, sessiond_uuid
,
1428 id_sessiond
.is_set
? &id_sessiond
.value
: NULL
,
1429 current_chunk_id
.is_set
? ¤t_chunk_id
.value
: NULL
,
1430 creation_time
.is_set
? &creation_time
.value
: NULL
,
1431 conn
->major
, conn
->minor
,
1432 session_name_contains_creation_timestamp
);
1437 assert(!conn
->session
);
1438 conn
->session
= session
;
1439 DBG("Created session %" PRIu64
, session
->id
);
1441 reply
.generic
.session_id
= htobe64(session
->id
);
1445 reply
.generic
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1447 reply
.generic
.ret_code
= htobe32(LTTNG_OK
);
1450 if (conn
->minor
< 11) {
1451 /* From 2.1 to 2.10 */
1452 ret
= lttng_dynamic_buffer_append(&reply_payload
,
1453 &reply
.generic
, sizeof(reply
.generic
));
1455 ERR("Failed to append \"create session\" command reply header to payload buffer");
1460 const uint32_t output_path_length
=
1461 session
? strlen(session
->output_path
) + 1 : 0;
1463 reply
.output_path_length
= htobe32(output_path_length
);
1464 ret
= lttng_dynamic_buffer_append(
1465 &reply_payload
, &reply
, sizeof(reply
));
1467 ERR("Failed to append \"create session\" command reply header to payload buffer");
1471 if (output_path_length
) {
1472 ret
= lttng_dynamic_buffer_append(&reply_payload
,
1473 session
->output_path
,
1474 output_path_length
);
1476 ERR("Failed to append \"create session\" command reply path to payload buffer");
1482 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, reply_payload
.data
,
1483 reply_payload
.size
, 0);
1484 if (send_ret
< (ssize_t
) reply_payload
.size
) {
1485 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1486 reply_payload
.size
, send_ret
);
1490 if (ret
< 0 && session
) {
1491 session_put(session
);
1493 lttng_dynamic_buffer_reset(&reply_payload
);
1498 * When we have received all the streams and the metadata for a channel,
1499 * we make them visible to the viewer threads.
1501 static void publish_connection_local_streams(struct relay_connection
*conn
)
1503 struct relay_stream
*stream
;
1504 struct relay_session
*session
= conn
->session
;
1507 * We publish all streams belonging to a session atomically wrt
1510 pthread_mutex_lock(&session
->lock
);
1512 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1514 stream_publish(stream
);
1519 * Inform the viewer that there are new streams in the session.
1521 if (session
->viewer_attached
) {
1522 uatomic_set(&session
->new_streams
, 1);
1524 pthread_mutex_unlock(&session
->lock
);
1527 static int conform_channel_path(char *channel_path
)
1531 if (strstr("../", channel_path
)) {
1532 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1538 if (*channel_path
== '/') {
1539 const size_t len
= strlen(channel_path
);
1542 * Channel paths from peers prior to 2.11 are expressed as an
1543 * absolute path that is, in reality, relative to the relay
1544 * daemon's output directory. Remove the leading slash so it
1545 * is correctly interpreted as a relative path later on.
1547 * len (and not len - 1) is used to copy the trailing NULL.
1549 bcopy(channel_path
+ 1, channel_path
, len
);
1556 * relay_add_stream: allocate a new stream for a session
1558 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1559 struct relay_connection
*conn
,
1560 const struct lttng_buffer_view
*payload
)
1564 struct relay_session
*session
= conn
->session
;
1565 struct relay_stream
*stream
= NULL
;
1566 struct lttcomm_relayd_status_stream reply
;
1567 struct ctf_trace
*trace
= NULL
;
1568 uint64_t stream_handle
= -1ULL;
1569 char *path_name
= NULL
, *channel_name
= NULL
;
1570 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1571 LTTNG_OPTIONAL(uint64_t) stream_chunk_id
= {};
1573 if (!session
|| !conn
->version_check_done
) {
1574 ERR("Trying to add a stream before version check");
1576 goto end_no_session
;
1579 if (session
->minor
== 1) {
1581 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1583 } else if (session
->minor
> 1 && session
->minor
< 11) {
1584 /* From 2.2 to 2.10 */
1585 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1586 &channel_name
, &tracefile_size
, &tracefile_count
);
1588 /* From 2.11 to ... */
1589 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1590 &channel_name
, &tracefile_size
, &tracefile_count
,
1591 &stream_chunk_id
.value
);
1592 stream_chunk_id
.is_set
= true;
1599 if (conform_channel_path(path_name
)) {
1604 * Backward compatibility for --group-output-by-session.
1605 * Prior to lttng 2.11, the complete path is passed by the stream.
1606 * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1607 * >=2.11 the chunk is responsible for the output path. When dealing
1608 * with producer < 2.11 the chunk output_path is the root output path
1609 * and the stream carries the complete path (path_name).
1610 * To support --group-output-by-session with older producer (<2.11), we
1611 * need to craft the path based on the stream path.
1613 if (opt_group_output_by
== RELAYD_GROUP_OUTPUT_BY_SESSION
) {
1614 if (conn
->minor
< 4) {
1616 * From 2.1 to 2.3, the session_name is not passed on
1617 * the RELAYD_CREATE_SESSION command. The session name
1618 * is necessary to detect the presence of a base_path
1619 * inside the stream path. Without it we cannot perform
1620 * a valid group-output-by-session transformation.
1622 WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1623 " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1624 session
->id
, path_name
);
1625 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1626 char *group_by_session_path_name
;
1628 assert(session
->session_name
[0] != '\0');
1630 group_by_session_path_name
=
1631 backward_compat_group_by_session(
1633 session
->session_name
);
1634 if (!group_by_session_path_name
) {
1635 ERR("Failed to apply group by session to stream of session %" PRIu64
,
1640 DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1641 path_name
, group_by_session_path_name
);
1644 path_name
= group_by_session_path_name
;
1648 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1653 /* This stream here has one reference on the trace. */
1654 pthread_mutex_lock(&last_relay_stream_id_lock
);
1655 stream_handle
= ++last_relay_stream_id
;
1656 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1658 /* We pass ownership of path_name and channel_name. */
1659 stream
= stream_create(trace
, stream_handle
, path_name
,
1660 channel_name
, tracefile_size
, tracefile_count
);
1662 channel_name
= NULL
;
1665 * Streams are the owners of their trace. Reference to trace is
1666 * kept within stream_create().
1668 ctf_trace_put(trace
);
1671 memset(&reply
, 0, sizeof(reply
));
1672 reply
.handle
= htobe64(stream_handle
);
1674 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1676 reply
.ret_code
= htobe32(LTTNG_OK
);
1679 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1680 sizeof(struct lttcomm_relayd_status_stream
), 0);
1681 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1682 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1694 * relay_close_stream: close a specific stream
1696 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1697 struct relay_connection
*conn
,
1698 const struct lttng_buffer_view
*payload
)
1702 struct relay_session
*session
= conn
->session
;
1703 struct lttcomm_relayd_close_stream stream_info
;
1704 struct lttcomm_relayd_generic_reply reply
;
1705 struct relay_stream
*stream
;
1707 DBG("Close stream received");
1709 if (!session
|| !conn
->version_check_done
) {
1710 ERR("Trying to close a stream before version check");
1712 goto end_no_session
;
1715 if (payload
->size
< sizeof(stream_info
)) {
1716 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1717 sizeof(stream_info
), payload
->size
);
1719 goto end_no_session
;
1721 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1722 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1723 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1725 stream
= stream_get_by_id(stream_info
.stream_id
);
1732 * Set last_net_seq_num before the close flag. Required by data
1735 pthread_mutex_lock(&stream
->lock
);
1736 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1737 pthread_mutex_unlock(&stream
->lock
);
1740 * This is one of the conditions which may trigger a stream close
1741 * with the others being:
1742 * 1) A close command is received for a stream
1743 * 2) The control connection owning the stream is closed
1744 * 3) We have received all of the stream's data _after_ a close
1747 try_stream_close(stream
);
1748 if (stream
->is_metadata
) {
1749 struct relay_viewer_stream
*vstream
;
1751 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1753 if (stream
->no_new_metadata_notified
) {
1755 * Since all the metadata has been sent to the
1756 * viewer and that we have a request to close
1757 * its stream, we can safely teardown the
1758 * corresponding metadata viewer stream.
1760 viewer_stream_put(vstream
);
1762 /* Put local reference. */
1763 viewer_stream_put(vstream
);
1770 memset(&reply
, 0, sizeof(reply
));
1772 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1774 reply
.ret_code
= htobe32(LTTNG_OK
);
1776 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1777 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1778 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1779 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1789 * relay_reset_metadata: reset a metadata stream
1792 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1793 struct relay_connection
*conn
,
1794 const struct lttng_buffer_view
*payload
)
1798 struct relay_session
*session
= conn
->session
;
1799 struct lttcomm_relayd_reset_metadata stream_info
;
1800 struct lttcomm_relayd_generic_reply reply
;
1801 struct relay_stream
*stream
;
1803 DBG("Reset metadata received");
1805 if (!session
|| !conn
->version_check_done
) {
1806 ERR("Trying to reset a metadata stream before version check");
1808 goto end_no_session
;
1811 if (payload
->size
< sizeof(stream_info
)) {
1812 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1813 sizeof(stream_info
), payload
->size
);
1815 goto end_no_session
;
1817 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1818 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1819 stream_info
.version
= be64toh(stream_info
.version
);
1821 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1823 /* Unsupported for live sessions for now. */
1824 if (session
->live_timer
!= 0) {
1829 stream
= stream_get_by_id(stream_info
.stream_id
);
1834 pthread_mutex_lock(&stream
->lock
);
1835 if (!stream
->is_metadata
) {
1840 ret
= stream_reset_file(stream
);
1842 ERR("Failed to reset metadata stream %" PRIu64
1843 ": stream_path = %s, channel = %s",
1844 stream
->stream_handle
, stream
->path_name
,
1845 stream
->channel_name
);
1849 pthread_mutex_unlock(&stream
->lock
);
1853 memset(&reply
, 0, sizeof(reply
));
1855 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1857 reply
.ret_code
= htobe32(LTTNG_OK
);
1859 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1860 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1861 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1862 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1872 * relay_unknown_command: send -1 if received unknown command
1874 static void relay_unknown_command(struct relay_connection
*conn
)
1876 struct lttcomm_relayd_generic_reply reply
;
1879 memset(&reply
, 0, sizeof(reply
));
1880 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1881 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1882 if (send_ret
< sizeof(reply
)) {
1883 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1888 * relay_start: send an acknowledgment to the client to tell if we are
1889 * ready to receive data. We are ready if a session is established.
1891 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1892 struct relay_connection
*conn
,
1893 const struct lttng_buffer_view
*payload
)
1897 struct lttcomm_relayd_generic_reply reply
;
1898 struct relay_session
*session
= conn
->session
;
1901 DBG("Trying to start the streaming without a session established");
1902 ret
= htobe32(LTTNG_ERR_UNK
);
1905 memset(&reply
, 0, sizeof(reply
));
1906 reply
.ret_code
= htobe32(LTTNG_OK
);
1907 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1909 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1910 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1919 * relay_recv_metadata: receive the metadata for the session.
1921 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1922 struct relay_connection
*conn
,
1923 const struct lttng_buffer_view
*payload
)
1926 struct relay_session
*session
= conn
->session
;
1927 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1928 struct relay_stream
*metadata_stream
;
1929 uint64_t metadata_payload_size
;
1930 struct lttng_buffer_view packet_view
;
1933 ERR("Metadata sent before version check");
1938 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1939 ERR("Incorrect data size");
1943 metadata_payload_size
= recv_hdr
->data_size
-
1944 sizeof(struct lttcomm_relayd_metadata_payload
);
1946 memcpy(&metadata_payload_header
, payload
->data
,
1947 sizeof(metadata_payload_header
));
1948 metadata_payload_header
.stream_id
= be64toh(
1949 metadata_payload_header
.stream_id
);
1950 metadata_payload_header
.padding_size
= be32toh(
1951 metadata_payload_header
.padding_size
);
1953 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1954 if (!metadata_stream
) {
1959 packet_view
= lttng_buffer_view_from_view(payload
,
1960 sizeof(metadata_payload_header
), metadata_payload_size
);
1961 if (!packet_view
.data
) {
1962 ERR("Invalid metadata packet length announced by header");
1967 pthread_mutex_lock(&metadata_stream
->lock
);
1968 ret
= stream_write(metadata_stream
, &packet_view
,
1969 metadata_payload_header
.padding_size
);
1970 pthread_mutex_unlock(&metadata_stream
->lock
);
1976 stream_put(metadata_stream
);
1982 * relay_send_version: send relayd version number
1984 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1985 struct relay_connection
*conn
,
1986 const struct lttng_buffer_view
*payload
)
1990 struct lttcomm_relayd_version reply
, msg
;
1991 bool compatible
= true;
1993 conn
->version_check_done
= true;
1995 /* Get version from the other side. */
1996 if (payload
->size
< sizeof(msg
)) {
1997 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1998 sizeof(msg
), payload
->size
);
2003 memcpy(&msg
, payload
->data
, sizeof(msg
));
2004 msg
.major
= be32toh(msg
.major
);
2005 msg
.minor
= be32toh(msg
.minor
);
2007 memset(&reply
, 0, sizeof(reply
));
2008 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
2009 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
2011 /* Major versions must be the same */
2012 if (reply
.major
!= msg
.major
) {
2013 DBG("Incompatible major versions (%u vs %u), deleting session",
2014 reply
.major
, msg
.major
);
2018 conn
->major
= reply
.major
;
2019 /* We adapt to the lowest compatible version */
2020 if (reply
.minor
<= msg
.minor
) {
2021 conn
->minor
= reply
.minor
;
2023 conn
->minor
= msg
.minor
;
2026 reply
.major
= htobe32(reply
.major
);
2027 reply
.minor
= htobe32(reply
.minor
);
2028 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2030 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2031 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2044 DBG("Version check done using protocol %u.%u", conn
->major
,
2052 * Check for data pending for a given stream id from the session daemon.
2054 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2055 struct relay_connection
*conn
,
2056 const struct lttng_buffer_view
*payload
)
2058 struct relay_session
*session
= conn
->session
;
2059 struct lttcomm_relayd_data_pending msg
;
2060 struct lttcomm_relayd_generic_reply reply
;
2061 struct relay_stream
*stream
;
2064 uint64_t stream_seq
;
2066 DBG("Data pending command received");
2068 if (!session
|| !conn
->version_check_done
) {
2069 ERR("Trying to check for data before version check");
2071 goto end_no_session
;
2074 if (payload
->size
< sizeof(msg
)) {
2075 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2076 sizeof(msg
), payload
->size
);
2078 goto end_no_session
;
2080 memcpy(&msg
, payload
->data
, sizeof(msg
));
2081 msg
.stream_id
= be64toh(msg
.stream_id
);
2082 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
2084 stream
= stream_get_by_id(msg
.stream_id
);
2085 if (stream
== NULL
) {
2090 pthread_mutex_lock(&stream
->lock
);
2092 if (session_streams_have_index(session
)) {
2094 * Ensure that both the index and stream data have been
2095 * flushed up to the requested point.
2097 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2099 stream_seq
= stream
->prev_data_seq
;
2101 DBG("Data pending for stream id %" PRIu64
": prev_data_seq %" PRIu64
2102 ", prev_index_seq %" PRIu64
2103 ", and last_seq %" PRIu64
, msg
.stream_id
,
2104 stream
->prev_data_seq
, stream
->prev_index_seq
,
2105 msg
.last_net_seq_num
);
2107 /* Avoid wrapping issue */
2108 if (((int64_t) (stream_seq
- msg
.last_net_seq_num
)) >= 0) {
2109 /* Data has in fact been written and is NOT pending */
2112 /* Data still being streamed thus pending */
2116 stream
->data_pending_check_done
= true;
2117 pthread_mutex_unlock(&stream
->lock
);
2122 memset(&reply
, 0, sizeof(reply
));
2123 reply
.ret_code
= htobe32(ret
);
2124 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2125 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2126 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2136 * Wait for the control socket to reach a quiescent state.
2138 * Note that for now, when receiving this command from the session
2139 * daemon, this means that every subsequent commands or data received on
2140 * the control socket has been handled. So, this is why we simply return
2143 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
2144 struct relay_connection
*conn
,
2145 const struct lttng_buffer_view
*payload
)
2149 struct relay_stream
*stream
;
2150 struct lttcomm_relayd_quiescent_control msg
;
2151 struct lttcomm_relayd_generic_reply reply
;
2153 DBG("Checking quiescent state on control socket");
2155 if (!conn
->session
|| !conn
->version_check_done
) {
2156 ERR("Trying to check for data before version check");
2158 goto end_no_session
;
2161 if (payload
->size
< sizeof(msg
)) {
2162 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2163 sizeof(msg
), payload
->size
);
2165 goto end_no_session
;
2167 memcpy(&msg
, payload
->data
, sizeof(msg
));
2168 msg
.stream_id
= be64toh(msg
.stream_id
);
2170 stream
= stream_get_by_id(msg
.stream_id
);
2174 pthread_mutex_lock(&stream
->lock
);
2175 stream
->data_pending_check_done
= true;
2176 pthread_mutex_unlock(&stream
->lock
);
2178 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
2181 memset(&reply
, 0, sizeof(reply
));
2182 reply
.ret_code
= htobe32(LTTNG_OK
);
2183 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2184 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2185 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2197 * Initialize a data pending command. This means that a consumer is about
2198 * to ask for data pending for each stream it holds. Simply iterate over
2199 * all streams of a session and set the data_pending_check_done flag.
2201 * This command returns to the client a LTTNG_OK code.
2203 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2204 struct relay_connection
*conn
,
2205 const struct lttng_buffer_view
*payload
)
2209 struct lttng_ht_iter iter
;
2210 struct lttcomm_relayd_begin_data_pending msg
;
2211 struct lttcomm_relayd_generic_reply reply
;
2212 struct relay_stream
*stream
;
2217 DBG("Init streams for data pending");
2219 if (!conn
->session
|| !conn
->version_check_done
) {
2220 ERR("Trying to check for data before version check");
2222 goto end_no_session
;
2225 if (payload
->size
< sizeof(msg
)) {
2226 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2227 sizeof(msg
), payload
->size
);
2229 goto end_no_session
;
2231 memcpy(&msg
, payload
->data
, sizeof(msg
));
2232 msg
.session_id
= be64toh(msg
.session_id
);
2235 * Iterate over all streams to set the begin data pending flag.
2236 * For now, the streams are indexed by stream handle so we have
2237 * to iterate over all streams to find the one associated with
2238 * the right session_id.
2241 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2243 if (!stream_get(stream
)) {
2246 if (stream
->trace
->session
->id
== msg
.session_id
) {
2247 pthread_mutex_lock(&stream
->lock
);
2248 stream
->data_pending_check_done
= false;
2249 pthread_mutex_unlock(&stream
->lock
);
2250 DBG("Set begin data pending flag to stream %" PRIu64
,
2251 stream
->stream_handle
);
2257 memset(&reply
, 0, sizeof(reply
));
2258 /* All good, send back reply. */
2259 reply
.ret_code
= htobe32(LTTNG_OK
);
2261 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2262 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2263 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2275 * End data pending command. This will check, for a given session id, if
2276 * each stream associated with it has its data_pending_check_done flag
2277 * set. If not, this means that the client lost track of the stream but
2278 * the data is still being streamed on our side. In this case, we inform
2279 * the client that data is in flight.
2281 * Return to the client if there is data in flight or not with a ret_code.
2283 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2284 struct relay_connection
*conn
,
2285 const struct lttng_buffer_view
*payload
)
2289 struct lttng_ht_iter iter
;
2290 struct lttcomm_relayd_end_data_pending msg
;
2291 struct lttcomm_relayd_generic_reply reply
;
2292 struct relay_stream
*stream
;
2293 uint32_t is_data_inflight
= 0;
2295 DBG("End data pending command");
2297 if (!conn
->session
|| !conn
->version_check_done
) {
2298 ERR("Trying to check for data before version check");
2300 goto end_no_session
;
2303 if (payload
->size
< sizeof(msg
)) {
2304 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2305 sizeof(msg
), payload
->size
);
2307 goto end_no_session
;
2309 memcpy(&msg
, payload
->data
, sizeof(msg
));
2310 msg
.session_id
= be64toh(msg
.session_id
);
2313 * Iterate over all streams to see if the begin data pending
2317 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2319 if (!stream_get(stream
)) {
2322 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2326 pthread_mutex_lock(&stream
->lock
);
2327 if (!stream
->data_pending_check_done
) {
2328 uint64_t stream_seq
;
2330 if (session_streams_have_index(conn
->session
)) {
2332 * Ensure that both the index and stream data have been
2333 * flushed up to the requested point.
2335 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2337 stream_seq
= stream
->prev_data_seq
;
2339 if (!stream
->closed
|| !(((int64_t) (stream_seq
- stream
->last_net_seq_num
)) >= 0)) {
2340 is_data_inflight
= 1;
2341 DBG("Data is still in flight for stream %" PRIu64
,
2342 stream
->stream_handle
);
2343 pthread_mutex_unlock(&stream
->lock
);
2348 pthread_mutex_unlock(&stream
->lock
);
2353 memset(&reply
, 0, sizeof(reply
));
2354 /* All good, send back reply. */
2355 reply
.ret_code
= htobe32(is_data_inflight
);
2357 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2358 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2359 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2371 * Receive an index for a specific stream.
2373 * Return 0 on success else a negative value.
2375 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2376 struct relay_connection
*conn
,
2377 const struct lttng_buffer_view
*payload
)
2381 struct relay_session
*session
= conn
->session
;
2382 struct lttcomm_relayd_index index_info
;
2383 struct lttcomm_relayd_generic_reply reply
;
2384 struct relay_stream
*stream
;
2389 DBG("Relay receiving index");
2391 if (!session
|| !conn
->version_check_done
) {
2392 ERR("Trying to close a stream before version check");
2394 goto end_no_session
;
2397 msg_len
= lttcomm_relayd_index_len(
2398 lttng_to_index_major(conn
->major
, conn
->minor
),
2399 lttng_to_index_minor(conn
->major
, conn
->minor
));
2400 if (payload
->size
< msg_len
) {
2401 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2402 msg_len
, payload
->size
);
2404 goto end_no_session
;
2406 memcpy(&index_info
, payload
->data
, msg_len
);
2407 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2408 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2409 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2410 index_info
.content_size
= be64toh(index_info
.content_size
);
2411 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2412 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2413 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2414 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2416 if (conn
->minor
>= 8) {
2417 index_info
.stream_instance_id
=
2418 be64toh(index_info
.stream_instance_id
);
2419 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2421 index_info
.stream_instance_id
= -1ULL;
2422 index_info
.packet_seq_num
= -1ULL;
2425 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2427 ERR("stream_get_by_id not found");
2432 pthread_mutex_lock(&stream
->lock
);
2433 ret
= stream_add_index(stream
, &index_info
);
2434 pthread_mutex_unlock(&stream
->lock
);
2436 goto end_stream_put
;
2442 memset(&reply
, 0, sizeof(reply
));
2444 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2446 reply
.ret_code
= htobe32(LTTNG_OK
);
2448 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2449 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2450 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2459 * Receive the streams_sent message.
2461 * Return 0 on success else a negative value.
2463 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2464 struct relay_connection
*conn
,
2465 const struct lttng_buffer_view
*payload
)
2469 struct lttcomm_relayd_generic_reply reply
;
2473 DBG("Relay receiving streams_sent");
2475 if (!conn
->session
|| !conn
->version_check_done
) {
2476 ERR("Trying to close a stream before version check");
2478 goto end_no_session
;
2482 * Publish every pending stream in the connection recv list which are
2483 * now ready to be used by the viewer.
2485 publish_connection_local_streams(conn
);
2487 memset(&reply
, 0, sizeof(reply
));
2488 reply
.ret_code
= htobe32(LTTNG_OK
);
2489 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2490 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2491 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2504 * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2505 * session rotation feature (not the tracefile rotation feature).
2507 static int relay_rotate_session_streams(
2508 const struct lttcomm_relayd_hdr
*recv_hdr
,
2509 struct relay_connection
*conn
,
2510 const struct lttng_buffer_view
*payload
)
2515 enum lttng_error_code reply_code
= LTTNG_ERR_UNK
;
2516 struct relay_session
*session
= conn
->session
;
2517 struct lttcomm_relayd_rotate_streams rotate_streams
;
2518 struct lttcomm_relayd_generic_reply reply
= {};
2519 struct relay_stream
*stream
= NULL
;
2520 const size_t header_len
= sizeof(struct lttcomm_relayd_rotate_streams
);
2521 struct lttng_trace_chunk
*next_trace_chunk
= NULL
;
2522 struct lttng_buffer_view stream_positions
;
2523 char chunk_id_buf
[MAX_INT_DEC_LEN(uint64_t)];
2524 const char *chunk_id_str
= "none";
2526 if (!session
|| !conn
->version_check_done
) {
2527 ERR("Trying to rotate a stream before version check");
2532 if (session
->major
== 2 && session
->minor
< 11) {
2533 ERR("Unsupported feature before 2.11");
2538 if (payload
->size
< header_len
) {
2539 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2540 header_len
, payload
->size
);
2545 memcpy(&rotate_streams
, payload
->data
, header_len
);
2547 /* Convert header to host endianness. */
2548 rotate_streams
= (typeof(rotate_streams
)) {
2549 .stream_count
= be32toh(rotate_streams
.stream_count
),
2550 .new_chunk_id
= (typeof(rotate_streams
.new_chunk_id
)) {
2551 .is_set
= !!rotate_streams
.new_chunk_id
.is_set
,
2552 .value
= be64toh(rotate_streams
.new_chunk_id
.value
),
2556 if (rotate_streams
.new_chunk_id
.is_set
) {
2558 * Retrieve the trace chunk the stream must transition to. As
2559 * per the protocol, this chunk should have been created
2560 * before this command is received.
2562 next_trace_chunk
= sessiond_trace_chunk_registry_get_chunk(
2563 sessiond_trace_chunk_registry
,
2564 session
->sessiond_uuid
, session
->id
,
2565 rotate_streams
.new_chunk_id
.value
);
2566 if (!next_trace_chunk
) {
2567 char uuid_str
[LTTNG_UUID_STR_LEN
];
2569 lttng_uuid_to_str(session
->sessiond_uuid
, uuid_str
);
2570 ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2571 ", trace_chunk_id = %" PRIu64
,
2572 uuid_str
, session
->id
,
2573 rotate_streams
.new_chunk_id
.value
);
2574 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2579 ret
= snprintf(chunk_id_buf
, sizeof(chunk_id_buf
), "%" PRIu64
,
2580 rotate_streams
.new_chunk_id
.value
);
2581 if (ret
< 0 || ret
>= sizeof(chunk_id_buf
)) {
2582 chunk_id_str
= "formatting error";
2584 chunk_id_str
= chunk_id_buf
;
2588 DBG("Rotate %" PRIu32
" streams of session \"%s\" to chunk \"%s\"",
2589 rotate_streams
.stream_count
, session
->session_name
,
2592 stream_positions
= lttng_buffer_view_from_view(payload
,
2593 sizeof(rotate_streams
), -1);
2594 if (!stream_positions
.data
||
2595 stream_positions
.size
<
2596 (rotate_streams
.stream_count
*
2597 sizeof(struct lttcomm_relayd_stream_rotation_position
))) {
2598 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2603 for (i
= 0; i
< rotate_streams
.stream_count
; i
++) {
2604 struct lttcomm_relayd_stream_rotation_position
*position_comm
=
2605 &((typeof(position_comm
)) stream_positions
.data
)[i
];
2606 const struct lttcomm_relayd_stream_rotation_position pos
= {
2607 .stream_id
= be64toh(position_comm
->stream_id
),
2608 .rotate_at_seq_num
= be64toh(
2609 position_comm
->rotate_at_seq_num
),
2612 stream
= stream_get_by_id(pos
.stream_id
);
2614 reply_code
= LTTNG_ERR_INVALID
;
2619 pthread_mutex_lock(&stream
->lock
);
2620 ret
= stream_set_pending_rotation(stream
, next_trace_chunk
,
2621 pos
.rotate_at_seq_num
);
2622 pthread_mutex_unlock(&stream
->lock
);
2624 reply_code
= LTTNG_ERR_FILE_CREATION_ERROR
;
2632 reply_code
= LTTNG_OK
;
2639 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2640 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2641 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2642 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2643 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2648 lttng_trace_chunk_put(next_trace_chunk
);
2655 * relay_create_trace_chunk: create a new trace chunk
2657 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2658 struct relay_connection
*conn
,
2659 const struct lttng_buffer_view
*payload
)
2663 struct relay_session
*session
= conn
->session
;
2664 struct lttcomm_relayd_create_trace_chunk
*msg
;
2665 struct lttcomm_relayd_generic_reply reply
= {};
2666 struct lttng_buffer_view header_view
;
2667 struct lttng_buffer_view chunk_name_view
;
2668 struct lttng_trace_chunk
*chunk
= NULL
, *published_chunk
= NULL
;
2669 enum lttng_error_code reply_code
= LTTNG_OK
;
2670 enum lttng_trace_chunk_status chunk_status
;
2671 const char *new_path
;
2673 if (!session
|| !conn
->version_check_done
) {
2674 ERR("Trying to create a trace chunk before version check");
2679 if (session
->major
== 2 && session
->minor
< 11) {
2680 ERR("Chunk creation command is unsupported before 2.11");
2685 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2686 if (!header_view
.data
) {
2687 ERR("Failed to receive payload of chunk creation command");
2692 /* Convert to host endianness. */
2693 msg
= (typeof(msg
)) header_view
.data
;
2694 msg
->chunk_id
= be64toh(msg
->chunk_id
);
2695 msg
->creation_timestamp
= be64toh(msg
->creation_timestamp
);
2696 msg
->override_name_length
= be32toh(msg
->override_name_length
);
2698 if (session
->current_trace_chunk
&&
2699 !lttng_trace_chunk_get_name_overridden(session
->current_trace_chunk
)) {
2700 chunk_status
= lttng_trace_chunk_rename_path(session
->current_trace_chunk
,
2701 DEFAULT_CHUNK_TMP_OLD_DIRECTORY
);
2702 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2703 ERR("Failed to rename old chunk");
2705 reply_code
= LTTNG_ERR_UNK
;
2709 session
->ongoing_rotation
= true;
2710 if (!session
->current_trace_chunk
) {
2711 if (!session
->has_rotated
) {
2717 new_path
= DEFAULT_CHUNK_TMP_NEW_DIRECTORY
;
2719 chunk
= lttng_trace_chunk_create(
2720 msg
->chunk_id
, msg
->creation_timestamp
, new_path
);
2722 ERR("Failed to create trace chunk in trace chunk creation command");
2724 reply_code
= LTTNG_ERR_NOMEM
;
2727 lttng_trace_chunk_set_fd_tracker(chunk
, the_fd_tracker
);
2729 if (msg
->override_name_length
) {
2732 chunk_name_view
= lttng_buffer_view_from_view(payload
,
2734 msg
->override_name_length
);
2735 name
= chunk_name_view
.data
;
2736 if (!name
|| name
[msg
->override_name_length
- 1]) {
2737 ERR("Failed to receive payload of chunk creation command");
2739 reply_code
= LTTNG_ERR_INVALID
;
2743 chunk_status
= lttng_trace_chunk_override_name(
2744 chunk
, chunk_name_view
.data
);
2745 switch (chunk_status
) {
2746 case LTTNG_TRACE_CHUNK_STATUS_OK
:
2748 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT
:
2749 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2750 reply_code
= LTTNG_ERR_INVALID
;
2754 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2755 reply_code
= LTTNG_ERR_UNK
;
2761 chunk_status
= lttng_trace_chunk_set_credentials_current_user(chunk
);
2762 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2763 reply_code
= LTTNG_ERR_UNK
;
2768 assert(conn
->session
->output_directory
);
2769 chunk_status
= lttng_trace_chunk_set_as_owner(chunk
,
2770 conn
->session
->output_directory
);
2771 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2772 reply_code
= LTTNG_ERR_UNK
;
2777 published_chunk
= sessiond_trace_chunk_registry_publish_chunk(
2778 sessiond_trace_chunk_registry
,
2779 conn
->session
->sessiond_uuid
,
2782 if (!published_chunk
) {
2783 char uuid_str
[LTTNG_UUID_STR_LEN
];
2785 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2786 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2791 reply_code
= LTTNG_ERR_NOMEM
;
2795 pthread_mutex_lock(&conn
->session
->lock
);
2796 if (conn
->session
->pending_closure_trace_chunk
) {
2798 * Invalid; this means a second create_trace_chunk command was
2799 * received before a close_trace_chunk.
2801 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2802 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2804 goto end_unlock_session
;
2806 conn
->session
->pending_closure_trace_chunk
=
2807 conn
->session
->current_trace_chunk
;
2808 conn
->session
->current_trace_chunk
= published_chunk
;
2809 published_chunk
= NULL
;
2810 if (!conn
->session
->pending_closure_trace_chunk
) {
2811 session
->ongoing_rotation
= false;
2814 pthread_mutex_unlock(&conn
->session
->lock
);
2816 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2817 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2819 sizeof(struct lttcomm_relayd_generic_reply
),
2821 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2822 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2827 lttng_trace_chunk_put(chunk
);
2828 lttng_trace_chunk_put(published_chunk
);
2833 * relay_close_trace_chunk: close a trace chunk
2835 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2836 struct relay_connection
*conn
,
2837 const struct lttng_buffer_view
*payload
)
2839 int ret
= 0, buf_ret
;
2841 struct relay_session
*session
= conn
->session
;
2842 struct lttcomm_relayd_close_trace_chunk
*msg
;
2843 struct lttcomm_relayd_close_trace_chunk_reply reply
= {};
2844 struct lttng_buffer_view header_view
;
2845 struct lttng_trace_chunk
*chunk
= NULL
;
2846 enum lttng_error_code reply_code
= LTTNG_OK
;
2847 enum lttng_trace_chunk_status chunk_status
;
2849 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type
) close_command
= {};
2850 time_t close_timestamp
;
2851 char closed_trace_chunk_path
[LTTNG_PATH_MAX
];
2852 size_t path_length
= 0;
2853 const char *chunk_name
= NULL
;
2854 struct lttng_dynamic_buffer reply_payload
;
2855 const char *new_path
;
2857 lttng_dynamic_buffer_init(&reply_payload
);
2859 if (!session
|| !conn
->version_check_done
) {
2860 ERR("Trying to close a trace chunk before version check");
2865 if (session
->major
== 2 && session
->minor
< 11) {
2866 ERR("Chunk close command is unsupported before 2.11");
2871 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2872 if (!header_view
.data
) {
2873 ERR("Failed to receive payload of chunk close command");
2878 /* Convert to host endianness. */
2879 msg
= (typeof(msg
)) header_view
.data
;
2880 chunk_id
= be64toh(msg
->chunk_id
);
2881 close_timestamp
= (time_t) be64toh(msg
->close_timestamp
);
2882 close_command
= (typeof(close_command
)){
2883 .value
= be32toh(msg
->close_command
.value
),
2884 .is_set
= msg
->close_command
.is_set
,
2887 chunk
= sessiond_trace_chunk_registry_get_chunk(
2888 sessiond_trace_chunk_registry
,
2889 conn
->session
->sessiond_uuid
,
2893 char uuid_str
[LTTNG_UUID_STR_LEN
];
2895 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2896 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2901 reply_code
= LTTNG_ERR_NOMEM
;
2905 pthread_mutex_lock(&session
->lock
);
2906 if (close_command
.is_set
&&
2907 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE
) {
2909 * Clear command. It is a protocol error to ask for a
2910 * clear on a relay which does not allow it. Querying
2911 * the configuration allows figuring out whether
2912 * clearing is allowed before doing the clear.
2914 if (!opt_allow_clear
) {
2916 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2917 goto end_unlock_session
;
2920 if (session
->pending_closure_trace_chunk
&&
2921 session
->pending_closure_trace_chunk
!= chunk
) {
2922 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
2923 session
->session_name
);
2924 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2926 goto end_unlock_session
;
2929 if (session
->current_trace_chunk
&& session
->current_trace_chunk
!= chunk
&&
2930 !lttng_trace_chunk_get_name_overridden(session
->current_trace_chunk
)) {
2931 if (close_command
.is_set
&&
2932 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE
&&
2933 !session
->has_rotated
) {
2934 /* New chunk stays in session output directory. */
2937 /* Use chunk name for new chunk. */
2940 /* Rename new chunk path. */
2941 chunk_status
= lttng_trace_chunk_rename_path(session
->current_trace_chunk
,
2943 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2947 session
->ongoing_rotation
= false;
2949 if ((!close_command
.is_set
||
2950 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION
) &&
2951 !lttng_trace_chunk_get_name_overridden(chunk
)) {
2952 const char *old_path
;
2954 if (!session
->has_rotated
) {
2959 /* We need to move back the .tmp_old_chunk to its rightful place. */
2960 chunk_status
= lttng_trace_chunk_rename_path(chunk
, old_path
);
2961 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2966 chunk_status
= lttng_trace_chunk_set_close_timestamp(
2967 chunk
, close_timestamp
);
2968 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2969 ERR("Failed to set trace chunk close timestamp");
2971 reply_code
= LTTNG_ERR_UNK
;
2972 goto end_unlock_session
;
2975 if (close_command
.is_set
) {
2976 chunk_status
= lttng_trace_chunk_set_close_command(
2977 chunk
, close_command
.value
);
2978 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2980 reply_code
= LTTNG_ERR_INVALID
;
2981 goto end_unlock_session
;
2984 chunk_status
= lttng_trace_chunk_get_name(chunk
, &chunk_name
, NULL
);
2985 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2986 ERR("Failed to get chunk name");
2988 reply_code
= LTTNG_ERR_UNK
;
2989 goto end_unlock_session
;
2991 if (!session
->has_rotated
&& !session
->snapshot
) {
2992 ret
= lttng_strncpy(closed_trace_chunk_path
,
2993 session
->output_path
,
2994 sizeof(closed_trace_chunk_path
));
2996 ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
2997 strlen(session
->output_path
),
2998 sizeof(closed_trace_chunk_path
));
2999 reply_code
= LTTNG_ERR_NOMEM
;
3001 goto end_unlock_session
;
3004 if (session
->snapshot
) {
3005 ret
= snprintf(closed_trace_chunk_path
,
3006 sizeof(closed_trace_chunk_path
),
3007 "%s/%s", session
->output_path
,
3010 ret
= snprintf(closed_trace_chunk_path
,
3011 sizeof(closed_trace_chunk_path
),
3012 "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY
3014 session
->output_path
, chunk_name
);
3016 if (ret
< 0 || ret
== sizeof(closed_trace_chunk_path
)) {
3017 ERR("Failed to format closed trace chunk resulting path");
3018 reply_code
= ret
< 0 ? LTTNG_ERR_UNK
: LTTNG_ERR_NOMEM
;
3020 goto end_unlock_session
;
3023 if (close_command
.is_set
&&
3024 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED
) {
3025 session
->has_rotated
= true;
3027 DBG("Reply chunk path on close: %s", closed_trace_chunk_path
);
3028 path_length
= strlen(closed_trace_chunk_path
) + 1;
3029 if (path_length
> UINT32_MAX
) {
3030 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3032 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
3033 goto end_unlock_session
;
3036 if (session
->current_trace_chunk
== chunk
) {
3038 * After a trace chunk close command, no new streams
3039 * referencing the chunk may be created. Hence, on the
3040 * event that no new trace chunk have been created for
3041 * the session, the reference to the current trace chunk
3042 * is released in order to allow it to be reclaimed when
3043 * the last stream releases its reference to it.
3045 lttng_trace_chunk_put(session
->current_trace_chunk
);
3046 session
->current_trace_chunk
= NULL
;
3048 lttng_trace_chunk_put(session
->pending_closure_trace_chunk
);
3049 session
->pending_closure_trace_chunk
= NULL
;
3051 pthread_mutex_unlock(&session
->lock
);
3054 reply
.generic
.ret_code
= htobe32((uint32_t) reply_code
);
3055 reply
.path_length
= htobe32((uint32_t) path_length
);
3056 buf_ret
= lttng_dynamic_buffer_append(
3057 &reply_payload
, &reply
, sizeof(reply
));
3059 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3063 if (reply_code
== LTTNG_OK
) {
3064 buf_ret
= lttng_dynamic_buffer_append(&reply_payload
,
3065 closed_trace_chunk_path
, path_length
);
3067 ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3072 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
3076 if (send_ret
< reply_payload
.size
) {
3077 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3078 reply_payload
.size
, send_ret
);
3083 lttng_trace_chunk_put(chunk
);
3084 lttng_dynamic_buffer_reset(&reply_payload
);
3089 * relay_trace_chunk_exists: check if a trace chunk exists
3091 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr
*recv_hdr
,
3092 struct relay_connection
*conn
,
3093 const struct lttng_buffer_view
*payload
)
3097 struct relay_session
*session
= conn
->session
;
3098 struct lttcomm_relayd_trace_chunk_exists
*msg
;
3099 struct lttcomm_relayd_trace_chunk_exists_reply reply
= {};
3100 struct lttng_buffer_view header_view
;
3104 if (!session
|| !conn
->version_check_done
) {
3105 ERR("Trying to close a trace chunk before version check");
3110 if (session
->major
== 2 && session
->minor
< 11) {
3111 ERR("Chunk close command is unsupported before 2.11");
3116 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
3117 if (!header_view
.data
) {
3118 ERR("Failed to receive payload of chunk close command");
3123 /* Convert to host endianness. */
3124 msg
= (typeof(msg
)) header_view
.data
;
3125 chunk_id
= be64toh(msg
->chunk_id
);
3127 ret
= sessiond_trace_chunk_registry_chunk_exists(
3128 sessiond_trace_chunk_registry
,
3129 conn
->session
->sessiond_uuid
,
3131 chunk_id
, &chunk_exists
);
3133 * If ret is not 0, send the reply and report the error to the caller.
3134 * It is a protocol (or internal) error and the session/connection
3135 * should be torn down.
3137 reply
= (typeof(reply
)){
3138 .generic
.ret_code
= htobe32((uint32_t)
3139 (ret
== 0 ? LTTNG_OK
: LTTNG_ERR_INVALID_PROTOCOL
)),
3140 .trace_chunk_exists
= ret
== 0 ? chunk_exists
: 0,
3142 send_ret
= conn
->sock
->ops
->sendmsg(
3143 conn
->sock
, &reply
, sizeof(reply
), 0);
3144 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3145 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
3154 * relay_get_configuration: query whether feature is available
3156 static int relay_get_configuration(const struct lttcomm_relayd_hdr
*recv_hdr
,
3157 struct relay_connection
*conn
,
3158 const struct lttng_buffer_view
*payload
)
3162 struct lttcomm_relayd_get_configuration
*msg
;
3163 struct lttcomm_relayd_get_configuration_reply reply
= {};
3164 struct lttng_buffer_view header_view
;
3165 uint64_t query_flags
= 0;
3166 uint64_t result_flags
= 0;
3168 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
3169 if (!header_view
.data
) {
3170 ERR("Failed to receive payload of chunk close command");
3175 /* Convert to host endianness. */
3176 msg
= (typeof(msg
)) header_view
.data
;
3177 query_flags
= be64toh(msg
->query_flags
);
3180 ret
= LTTNG_ERR_INVALID_PROTOCOL
;
3183 if (opt_allow_clear
) {
3184 result_flags
|= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED
;
3188 reply
= (typeof(reply
)){
3189 .generic
.ret_code
= htobe32((uint32_t)
3190 (ret
== 0 ? LTTNG_OK
: LTTNG_ERR_INVALID_PROTOCOL
)),
3191 .relayd_configuration_flags
= htobe64(result_flags
),
3193 send_ret
= conn
->sock
->ops
->sendmsg(
3194 conn
->sock
, &reply
, sizeof(reply
), 0);
3195 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3196 ERR("Failed to send \"get configuration\" command reply (ret = %zd)",
3204 #define DBG_CMD(cmd_name, conn) \
3205 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
3207 static int relay_process_control_command(struct relay_connection
*conn
,
3208 const struct lttcomm_relayd_hdr
*header
,
3209 const struct lttng_buffer_view
*payload
)
3213 switch (header
->cmd
) {
3214 case RELAYD_CREATE_SESSION
:
3215 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
3216 ret
= relay_create_session(header
, conn
, payload
);
3218 case RELAYD_ADD_STREAM
:
3219 DBG_CMD("RELAYD_ADD_STREAM", conn
);
3220 ret
= relay_add_stream(header
, conn
, payload
);
3222 case RELAYD_START_DATA
:
3223 DBG_CMD("RELAYD_START_DATA", conn
);
3224 ret
= relay_start(header
, conn
, payload
);
3226 case RELAYD_SEND_METADATA
:
3227 DBG_CMD("RELAYD_SEND_METADATA", conn
);
3228 ret
= relay_recv_metadata(header
, conn
, payload
);
3230 case RELAYD_VERSION
:
3231 DBG_CMD("RELAYD_VERSION", conn
);
3232 ret
= relay_send_version(header
, conn
, payload
);
3234 case RELAYD_CLOSE_STREAM
:
3235 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
3236 ret
= relay_close_stream(header
, conn
, payload
);
3238 case RELAYD_DATA_PENDING
:
3239 DBG_CMD("RELAYD_DATA_PENDING", conn
);
3240 ret
= relay_data_pending(header
, conn
, payload
);
3242 case RELAYD_QUIESCENT_CONTROL
:
3243 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
3244 ret
= relay_quiescent_control(header
, conn
, payload
);
3246 case RELAYD_BEGIN_DATA_PENDING
:
3247 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
3248 ret
= relay_begin_data_pending(header
, conn
, payload
);
3250 case RELAYD_END_DATA_PENDING
:
3251 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
3252 ret
= relay_end_data_pending(header
, conn
, payload
);
3254 case RELAYD_SEND_INDEX
:
3255 DBG_CMD("RELAYD_SEND_INDEX", conn
);
3256 ret
= relay_recv_index(header
, conn
, payload
);
3258 case RELAYD_STREAMS_SENT
:
3259 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
3260 ret
= relay_streams_sent(header
, conn
, payload
);
3262 case RELAYD_RESET_METADATA
:
3263 DBG_CMD("RELAYD_RESET_METADATA", conn
);
3264 ret
= relay_reset_metadata(header
, conn
, payload
);
3266 case RELAYD_ROTATE_STREAMS
:
3267 DBG_CMD("RELAYD_ROTATE_STREAMS", conn
);
3268 ret
= relay_rotate_session_streams(header
, conn
, payload
);
3270 case RELAYD_CREATE_TRACE_CHUNK
:
3271 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn
);
3272 ret
= relay_create_trace_chunk(header
, conn
, payload
);
3274 case RELAYD_CLOSE_TRACE_CHUNK
:
3275 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn
);
3276 ret
= relay_close_trace_chunk(header
, conn
, payload
);
3278 case RELAYD_TRACE_CHUNK_EXISTS
:
3279 DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn
);
3280 ret
= relay_trace_chunk_exists(header
, conn
, payload
);
3282 case RELAYD_GET_CONFIGURATION
:
3283 DBG_CMD("RELAYD_GET_CONFIGURATION", conn
);
3284 ret
= relay_get_configuration(header
, conn
, payload
);
3286 case RELAYD_UPDATE_SYNC_INFO
:
3288 ERR("Received unknown command (%u)", header
->cmd
);
3289 relay_unknown_command(conn
);
3298 static enum relay_connection_status
relay_process_control_receive_payload(
3299 struct relay_connection
*conn
)
3302 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3303 struct lttng_dynamic_buffer
*reception_buffer
=
3304 &conn
->protocol
.ctrl
.reception_buffer
;
3305 struct ctrl_connection_state_receive_payload
*state
=
3306 &conn
->protocol
.ctrl
.state
.receive_payload
;
3307 struct lttng_buffer_view payload_view
;
3309 if (state
->left_to_receive
== 0) {
3310 /* Short-circuit for payload-less commands. */
3311 goto reception_complete
;
3314 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3315 reception_buffer
->data
+ state
->received
,
3316 state
->left_to_receive
, MSG_DONTWAIT
);
3318 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3319 PERROR("Unable to receive command payload on sock %d",
3321 status
= RELAY_CONNECTION_STATUS_ERROR
;
3324 } else if (ret
== 0) {
3325 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3326 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3331 assert(ret
<= state
->left_to_receive
);
3333 state
->left_to_receive
-= ret
;
3334 state
->received
+= ret
;
3336 if (state
->left_to_receive
> 0) {
3338 * Can't transition to the protocol's next state, wait to
3339 * receive the rest of the header.
3341 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3342 state
->received
, state
->left_to_receive
,
3348 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
3349 conn
->sock
->fd
, state
->received
);
3351 * The payload required to process the command has been received.
3352 * A view to the reception buffer is forwarded to the various
3353 * commands and the state of the control is reset on success.
3355 * Commands are responsible for sending their reply to the peer.
3357 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
3359 ret
= relay_process_control_command(conn
,
3360 &state
->header
, &payload_view
);
3362 status
= RELAY_CONNECTION_STATUS_ERROR
;
3366 ret
= connection_reset_protocol_state(conn
);
3368 status
= RELAY_CONNECTION_STATUS_ERROR
;
3374 static enum relay_connection_status
relay_process_control_receive_header(
3375 struct relay_connection
*conn
)
3378 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3379 struct lttcomm_relayd_hdr header
;
3380 struct lttng_dynamic_buffer
*reception_buffer
=
3381 &conn
->protocol
.ctrl
.reception_buffer
;
3382 struct ctrl_connection_state_receive_header
*state
=
3383 &conn
->protocol
.ctrl
.state
.receive_header
;
3385 assert(state
->left_to_receive
!= 0);
3387 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3388 reception_buffer
->data
+ state
->received
,
3389 state
->left_to_receive
, MSG_DONTWAIT
);
3391 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3392 PERROR("Unable to receive control command header on sock %d",
3394 status
= RELAY_CONNECTION_STATUS_ERROR
;
3397 } else if (ret
== 0) {
3398 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3399 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3404 assert(ret
<= state
->left_to_receive
);
3406 state
->left_to_receive
-= ret
;
3407 state
->received
+= ret
;
3409 if (state
->left_to_receive
> 0) {
3411 * Can't transition to the protocol's next state, wait to
3412 * receive the rest of the header.
3414 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3415 state
->received
, state
->left_to_receive
,
3420 /* Transition to next state: receiving the command's payload. */
3421 conn
->protocol
.ctrl
.state_id
=
3422 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3423 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
3424 header
.circuit_id
= be64toh(header
.circuit_id
);
3425 header
.data_size
= be64toh(header
.data_size
);
3426 header
.cmd
= be32toh(header
.cmd
);
3427 header
.cmd_version
= be32toh(header
.cmd_version
);
3428 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
3429 &header
, sizeof(header
));
3431 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
3432 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
3435 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
3436 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
3438 status
= RELAY_CONNECTION_STATUS_ERROR
;
3442 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
3444 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
3445 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
3448 status
= RELAY_CONNECTION_STATUS_ERROR
;
3452 if (header
.data_size
== 0) {
3454 * Manually invoke the next state as the poll loop
3455 * will not wake-up to allow us to proceed further.
3457 status
= relay_process_control_receive_payload(conn
);
3464 * Process the commands received on the control socket
3466 static enum relay_connection_status
relay_process_control(
3467 struct relay_connection
*conn
)
3469 enum relay_connection_status status
;
3471 switch (conn
->protocol
.ctrl
.state_id
) {
3472 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
3473 status
= relay_process_control_receive_header(conn
);
3475 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3476 status
= relay_process_control_receive_payload(conn
);
3479 ERR("Unknown control connection protocol state encountered.");
3486 static enum relay_connection_status
relay_process_data_receive_header(
3487 struct relay_connection
*conn
)
3490 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3491 struct data_connection_state_receive_header
*state
=
3492 &conn
->protocol
.data
.state
.receive_header
;
3493 struct lttcomm_relayd_data_hdr header
;
3494 struct relay_stream
*stream
;
3496 assert(state
->left_to_receive
!= 0);
3498 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3499 state
->header_reception_buffer
+ state
->received
,
3500 state
->left_to_receive
, MSG_DONTWAIT
);
3502 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3503 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
3504 status
= RELAY_CONNECTION_STATUS_ERROR
;
3507 } else if (ret
== 0) {
3508 /* Orderly shutdown. Not necessary to print an error. */
3509 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3510 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3515 assert(ret
<= state
->left_to_receive
);
3517 state
->left_to_receive
-= ret
;
3518 state
->received
+= ret
;
3520 if (state
->left_to_receive
> 0) {
3522 * Can't transition to the protocol's next state, wait to
3523 * receive the rest of the header.
3525 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3526 state
->received
, state
->left_to_receive
,
3531 /* Transition to next state: receiving the payload. */
3532 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3534 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3535 header
.circuit_id
= be64toh(header
.circuit_id
);
3536 header
.stream_id
= be64toh(header
.stream_id
);
3537 header
.data_size
= be32toh(header
.data_size
);
3538 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3539 header
.padding_size
= be32toh(header
.padding_size
);
3540 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3542 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3544 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3545 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3547 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3548 conn
->sock
->fd
, header
.circuit_id
,
3549 header
.stream_id
, header
.data_size
,
3550 header
.net_seq_num
, header
.padding_size
);
3552 stream
= stream_get_by_id(header
.stream_id
);
3554 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3556 /* Protocol error. */
3557 status
= RELAY_CONNECTION_STATUS_ERROR
;
3561 pthread_mutex_lock(&stream
->lock
);
3562 /* Prepare stream for the reception of a new packet. */
3563 ret
= stream_init_packet(stream
, header
.data_size
,
3564 &conn
->protocol
.data
.state
.receive_payload
.rotate_index
);
3565 pthread_mutex_unlock(&stream
->lock
);
3567 ERR("Failed to rotate stream output file");
3568 status
= RELAY_CONNECTION_STATUS_ERROR
;
3569 goto end_stream_unlock
;
3578 static enum relay_connection_status
relay_process_data_receive_payload(
3579 struct relay_connection
*conn
)
3582 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3583 struct relay_stream
*stream
;
3584 struct data_connection_state_receive_payload
*state
=
3585 &conn
->protocol
.data
.state
.receive_payload
;
3586 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3587 char data_buffer
[chunk_size
];
3588 bool partial_recv
= false;
3589 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3590 uint64_t left_to_receive
= state
->left_to_receive
;
3591 struct relay_session
*session
;
3593 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3594 state
->header
.stream_id
, state
->header
.net_seq_num
,
3595 state
->received
, left_to_receive
);
3597 stream
= stream_get_by_id(state
->header
.stream_id
);
3599 /* Protocol error. */
3600 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3601 state
->header
.stream_id
);
3602 status
= RELAY_CONNECTION_STATUS_ERROR
;
3606 pthread_mutex_lock(&stream
->lock
);
3607 session
= stream
->trace
->session
;
3608 if (!conn
->session
) {
3609 ret
= connection_set_session(conn
, session
);
3611 status
= RELAY_CONNECTION_STATUS_ERROR
;
3612 goto end_stream_unlock
;
3617 * The size of the "chunk" received on any iteration is bounded by:
3618 * - the data left to receive,
3619 * - the data immediately available on the socket,
3620 * - the on-stack data buffer
3622 while (left_to_receive
> 0 && !partial_recv
) {
3623 size_t recv_size
= min(left_to_receive
, chunk_size
);
3624 struct lttng_buffer_view packet_chunk
;
3626 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3627 recv_size
, MSG_DONTWAIT
);
3629 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3630 PERROR("Socket %d error", conn
->sock
->fd
);
3631 status
= RELAY_CONNECTION_STATUS_ERROR
;
3633 goto end_stream_unlock
;
3634 } else if (ret
== 0) {
3635 /* No more data ready to be consumed on socket. */
3636 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3637 state
->header
.stream_id
);
3638 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3640 } else if (ret
< (int) recv_size
) {
3642 * All the data available on the socket has been
3645 partial_recv
= true;
3649 packet_chunk
= lttng_buffer_view_init(data_buffer
,
3651 assert(packet_chunk
.data
);
3653 ret
= stream_write(stream
, &packet_chunk
, 0);
3655 ERR("Relay error writing data to file");
3656 status
= RELAY_CONNECTION_STATUS_ERROR
;
3657 goto end_stream_unlock
;
3660 left_to_receive
-= recv_size
;
3661 state
->received
+= recv_size
;
3662 state
->left_to_receive
= left_to_receive
;
3665 if (state
->left_to_receive
> 0) {
3667 * Did not receive all the data expected, wait for more data to
3668 * become available on the socket.
3670 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3671 state
->header
.stream_id
, state
->received
,
3672 state
->left_to_receive
);
3673 goto end_stream_unlock
;
3676 ret
= stream_write(stream
, NULL
, state
->header
.padding_size
);
3678 status
= RELAY_CONNECTION_STATUS_ERROR
;
3679 goto end_stream_unlock
;
3682 if (session_streams_have_index(session
)) {
3683 ret
= stream_update_index(stream
, state
->header
.net_seq_num
,
3684 state
->rotate_index
, &index_flushed
,
3685 state
->header
.data_size
+ state
->header
.padding_size
);
3687 ERR("Failed to update index: stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3688 stream
->stream_handle
,
3689 state
->header
.net_seq_num
, ret
);
3690 status
= RELAY_CONNECTION_STATUS_ERROR
;
3691 goto end_stream_unlock
;
3695 if (stream
->prev_data_seq
== -1ULL) {
3699 ret
= stream_complete_packet(stream
, state
->header
.data_size
+
3700 state
->header
.padding_size
, state
->header
.net_seq_num
,
3703 status
= RELAY_CONNECTION_STATUS_ERROR
;
3704 goto end_stream_unlock
;
3708 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3709 * contents of *state which are aliased (union) to the same location as
3710 * the new state. Don't use it beyond this point.
3712 connection_reset_protocol_state(conn
);
3716 close_requested
= stream
->close_requested
;
3717 pthread_mutex_unlock(&stream
->lock
);
3718 if (close_requested
&& left_to_receive
== 0) {
3719 try_stream_close(stream
);
3723 pthread_mutex_lock(&session
->lock
);
3724 uatomic_set(&session
->new_streams
, 1);
3725 pthread_mutex_unlock(&session
->lock
);
3734 * relay_process_data: Process the data received on the data socket
3736 static enum relay_connection_status
relay_process_data(
3737 struct relay_connection
*conn
)
3739 enum relay_connection_status status
;
3741 switch (conn
->protocol
.data
.state_id
) {
3742 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3743 status
= relay_process_data_receive_header(conn
);
3745 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3746 status
= relay_process_data_receive_payload(conn
);
3749 ERR("Unexpected data connection communication state.");
3756 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3760 (void) lttng_poll_del(events
, pollfd
);
3762 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
, &pollfd
, 1,
3763 fd_tracker_util_close_fd
, NULL
);
3765 ERR("Closing pollfd %d", pollfd
);
3769 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3770 int pollfd
, struct relay_connection
*conn
)
3772 const char *type_str
;
3774 switch (conn
->type
) {
3779 type_str
= "Control";
3781 case RELAY_VIEWER_COMMAND
:
3782 type_str
= "Viewer Command";
3784 case RELAY_VIEWER_NOTIFICATION
:
3785 type_str
= "Viewer Notification";
3788 type_str
= "Unknown";
3790 cleanup_connection_pollfd(events
, pollfd
);
3791 connection_put(conn
);
3792 DBG("%s connection closed with %d", type_str
, pollfd
);
3796 * This thread does the actual work
3798 static void *relay_thread_worker(void *data
)
3800 int ret
, err
= -1, last_seen_data_fd
= -1;
3802 struct lttng_poll_event events
;
3803 struct lttng_ht
*relay_connections_ht
;
3804 struct lttng_ht_iter iter
;
3805 struct relay_connection
*destroy_conn
= NULL
;
3807 DBG("[thread] Relay worker started");
3809 rcu_register_thread();
3811 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3813 if (testpoint(relayd_thread_worker
)) {
3814 goto error_testpoint
;
3817 health_code_update();
3819 /* table of connections indexed on socket */
3820 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3821 if (!relay_connections_ht
) {
3822 goto relay_connections_ht_error
;
3825 ret
= create_named_thread_poll_set(&events
, 2, "Worker thread epoll");
3827 goto error_poll_create
;
3830 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3837 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3839 health_code_update();
3841 /* Infinite blocking call, waiting for transmission */
3842 DBG3("Relayd worker thread polling...");
3843 health_poll_entry();
3844 ret
= lttng_poll_wait(&events
, -1);
3848 * Restart interrupted system call.
3850 if (errno
== EINTR
) {
3859 * Process control. The control connection is
3860 * prioritized so we don't starve it with high
3861 * throughput tracing data on the data connection.
3863 for (i
= 0; i
< nb_fd
; i
++) {
3864 /* Fetch once the poll data */
3865 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3866 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3868 health_code_update();
3870 /* Thread quit pipe has been closed. Killing thread. */
3871 ret
= check_thread_quit_pipe(pollfd
, revents
);
3877 /* Inspect the relay conn pipe for new connection */
3878 if (pollfd
== relay_conn_pipe
[0]) {
3879 if (revents
& LPOLLIN
) {
3880 struct relay_connection
*conn
;
3882 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3886 ret
= lttng_poll_add(&events
,
3888 LPOLLIN
| LPOLLRDHUP
);
3890 ERR("Failed to add new connection file descriptor to poll set");
3893 connection_ht_add(relay_connections_ht
, conn
);
3894 DBG("Connection socket %d added", conn
->sock
->fd
);
3895 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3896 ERR("Relay connection pipe error");
3899 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3903 struct relay_connection
*ctrl_conn
;
3905 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3906 /* If not found, there is a synchronization issue. */
3909 if (ctrl_conn
->type
== RELAY_DATA
) {
3910 if (revents
& LPOLLIN
) {
3912 * Flag the last seen data fd not deleted. It will be
3913 * used as the last seen fd if any fd gets deleted in
3916 last_notdel_data_fd
= pollfd
;
3918 goto put_ctrl_connection
;
3920 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3922 if (revents
& LPOLLIN
) {
3923 enum relay_connection_status status
;
3925 status
= relay_process_control(ctrl_conn
);
3926 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3928 * On socket error flag the session as aborted to force
3929 * the cleanup of its stream otherwise it can leak
3930 * during the lifetime of the relayd.
3932 * This prevents situations in which streams can be
3933 * left opened because an index was received, the
3934 * control connection is closed, and the data
3935 * connection is closed (uncleanly) before the packet's
3938 * Since the control connection encountered an error,
3939 * it is okay to be conservative and close the
3940 * session right now as we can't rely on the protocol
3941 * being respected anymore.
3943 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3944 session_abort(ctrl_conn
->session
);
3947 /* Clear the connection on error or close. */
3948 relay_thread_close_connection(&events
,
3953 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3954 relay_thread_close_connection(&events
,
3956 if (last_seen_data_fd
== pollfd
) {
3957 last_seen_data_fd
= last_notdel_data_fd
;
3960 ERR("Unexpected poll events %u for control sock %d",
3962 connection_put(ctrl_conn
);
3965 put_ctrl_connection
:
3966 connection_put(ctrl_conn
);
3971 * The last loop handled a control request, go back to poll to make
3972 * sure we prioritise the control socket.
3978 if (last_seen_data_fd
>= 0) {
3979 for (i
= 0; i
< nb_fd
; i
++) {
3980 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3982 health_code_update();
3984 if (last_seen_data_fd
== pollfd
) {
3991 /* Process data connection. */
3992 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3993 /* Fetch the poll data. */
3994 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3995 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3996 struct relay_connection
*data_conn
;
3998 health_code_update();
4001 /* No activity for this FD (poll implementation). */
4005 /* Skip the command pipe. It's handled in the first loop. */
4006 if (pollfd
== relay_conn_pipe
[0]) {
4010 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
4012 /* Skip it. Might be removed before. */
4015 if (data_conn
->type
== RELAY_CONTROL
) {
4016 goto put_data_connection
;
4018 assert(data_conn
->type
== RELAY_DATA
);
4020 if (revents
& LPOLLIN
) {
4021 enum relay_connection_status status
;
4023 status
= relay_process_data(data_conn
);
4024 /* Connection closed or error. */
4025 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
4027 * On socket error flag the session as aborted to force
4028 * the cleanup of its stream otherwise it can leak
4029 * during the lifetime of the relayd.
4031 * This prevents situations in which streams can be
4032 * left opened because an index was received, the
4033 * control connection is closed, and the data
4034 * connection is closed (uncleanly) before the packet's
4037 * Since the data connection encountered an error,
4038 * it is okay to be conservative and close the
4039 * session right now as we can't rely on the protocol
4040 * being respected anymore.
4042 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
4043 session_abort(data_conn
->session
);
4045 relay_thread_close_connection(&events
, pollfd
,
4048 * Every goto restart call sets the last seen fd where
4049 * here we don't really care since we gracefully
4050 * continue the loop after the connection is deleted.
4053 /* Keep last seen port. */
4054 last_seen_data_fd
= pollfd
;
4055 connection_put(data_conn
);
4058 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4059 relay_thread_close_connection(&events
, pollfd
,
4062 ERR("Unknown poll events %u for data sock %d",
4065 put_data_connection
:
4066 connection_put(data_conn
);
4068 last_seen_data_fd
= -1;
4071 /* Normal exit, no error */
4076 /* Cleanup remaining connection object. */
4078 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
4081 health_code_update();
4083 session_abort(destroy_conn
->session
);
4086 * No need to grab another ref, because we own
4089 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
4094 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
4096 lttng_ht_destroy(relay_connections_ht
);
4097 relay_connections_ht_error
:
4098 /* Close relay conn pipes */
4099 (void) fd_tracker_util_pipe_close(the_fd_tracker
,
4102 DBG("Thread exited with error");
4104 DBG("Worker thread cleanup complete");
4108 ERR("Health error occurred in %s", __func__
);
4110 health_unregister(health_relayd
);
4111 rcu_unregister_thread();
4112 lttng_relay_stop_threads();
4117 * Create the relay command pipe to wake thread_manage_apps.
4118 * Closed in cleanup().
4120 static int create_relay_conn_pipe(void)
4122 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
4123 "Relayd connection pipe", relay_conn_pipe
);
4126 static int stdio_open(void *data
, int *fds
)
4128 fds
[0] = fileno(stdout
);
4129 fds
[1] = fileno(stderr
);
4133 static int track_stdio(void)
4136 const char *names
[] = { "stdout", "stderr" };
4138 return fd_tracker_open_unsuspendable_fd(the_fd_tracker
, fds
,
4139 names
, 2, stdio_open
, NULL
);
4145 int main(int argc
, char **argv
)
4147 bool thread_is_rcu_registered
= false;
4148 int ret
= 0, retval
= 0;
4150 char *unlinked_file_directory_path
= NULL
, *output_path
= NULL
;
4152 /* Parse environment variables */
4153 ret
= parse_env_options();
4161 * Command line arguments overwrite environment.
4164 if (set_options(argc
, argv
)) {
4169 if (set_signal_handler()) {
4174 relayd_config_log();
4176 if (opt_print_version
) {
4182 ret
= fclose(stdin
);
4184 PERROR("Failed to close stdin");
4188 DBG("Clear command %s", opt_allow_clear
? "allowed" : "disallowed");
4190 /* Try to create directory if -o, --output is specified. */
4191 if (opt_output_path
) {
4192 if (*opt_output_path
!= '/') {
4193 ERR("Please specify an absolute path for -o, --output PATH");
4198 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
4201 ERR("Unable to create %s", opt_output_path
);
4208 if (opt_daemon
|| opt_background
) {
4209 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4217 if (opt_working_directory
) {
4218 ret
= utils_change_working_directory(opt_working_directory
);
4220 /* All errors are already logged. */
4225 sessiond_trace_chunk_registry
= sessiond_trace_chunk_registry_create();
4226 if (!sessiond_trace_chunk_registry
) {
4227 ERR("Failed to initialize session daemon trace chunk registry");
4233 * The RCU thread registration (and use, through the fd-tracker's
4234 * creation) is done after the daemonization to allow us to not
4235 * deal with liburcu's fork() management as the call RCU needs to
4238 rcu_register_thread();
4239 thread_is_rcu_registered
= true;
4241 output_path
= create_output_path("");
4243 ERR("Failed to get output path");
4247 ret
= asprintf(&unlinked_file_directory_path
, "%s/%s", output_path
,
4248 DEFAULT_UNLINKED_FILES_DIRECTORY
);
4251 ERR("Failed to format unlinked file directory path");
4255 the_fd_tracker
= fd_tracker_create(
4256 unlinked_file_directory_path
, lttng_opt_fd_pool_size
);
4257 free(unlinked_file_directory_path
);
4258 if (!the_fd_tracker
) {
4263 ret
= track_stdio();
4269 /* Initialize thread health monitoring */
4270 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
4271 if (!health_relayd
) {
4272 PERROR("health_app_create error");
4277 /* Create thread quit pipe */
4278 if (init_thread_quit_pipe()) {
4283 /* Setup the thread apps communication pipe. */
4284 if (create_relay_conn_pipe()) {
4289 /* Init relay command queue. */
4290 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
4292 /* Initialize communication library */
4294 lttcomm_inet_init();
4296 /* tables of sessions indexed by session ID */
4297 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4303 /* tables of streams indexed by stream ID */
4304 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4305 if (!relay_streams_ht
) {
4310 /* tables of streams indexed by stream ID */
4311 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4312 if (!viewer_streams_ht
) {
4317 ret
= init_health_quit_pipe();
4323 /* Create thread to manage the client socket */
4324 ret
= pthread_create(&health_thread
, default_pthread_attr(),
4325 thread_manage_health
, (void *) NULL
);
4328 PERROR("pthread_create health");
4333 /* Setup the dispatcher thread */
4334 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
4335 relay_thread_dispatcher
, (void *) NULL
);
4338 PERROR("pthread_create dispatcher");
4340 goto exit_dispatcher_thread
;
4343 /* Setup the worker thread */
4344 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
4345 relay_thread_worker
, NULL
);
4348 PERROR("pthread_create worker");
4350 goto exit_worker_thread
;
4353 /* Setup the listener thread */
4354 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
4355 relay_thread_listener
, (void *) NULL
);
4358 PERROR("pthread_create listener");
4360 goto exit_listener_thread
;
4363 ret
= relayd_live_create(live_uri
);
4365 ERR("Starting live viewer threads");
4371 * This is where we start awaiting program completion (e.g. through
4372 * signal that asks threads to teardown).
4375 ret
= relayd_live_join();
4381 ret
= pthread_join(listener_thread
, &status
);
4384 PERROR("pthread_join listener_thread");
4388 exit_listener_thread
:
4389 ret
= pthread_join(worker_thread
, &status
);
4392 PERROR("pthread_join worker_thread");
4397 ret
= pthread_join(dispatcher_thread
, &status
);
4400 PERROR("pthread_join dispatcher_thread");
4403 exit_dispatcher_thread
:
4405 ret
= pthread_join(health_thread
, &status
);
4408 PERROR("pthread_join health_thread");
4413 * Wait for all pending call_rcu work to complete before tearing
4414 * down data structures. call_rcu worker may be trying to
4415 * perform lookups in those structures.
4420 /* Ensure all prior call_rcu are done. */
4423 if (thread_is_rcu_registered
) {
4424 rcu_unregister_thread();