2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
35 #include <sys/types.h>
37 #include <sys/resource.h>
39 #include <urcu/futex.h>
40 #include <urcu/uatomic.h>
45 #include <lttng/lttng.h>
46 #include <common/common.h>
47 #include <common/compat/poll.h>
48 #include <common/compat/socket.h>
49 #include <common/compat/endian.h>
50 #include <common/compat/getenv.h>
51 #include <common/defaults.h>
52 #include <common/daemonize.h>
53 #include <common/futex.h>
54 #include <common/sessiond-comm/sessiond-comm.h>
55 #include <common/sessiond-comm/inet.h>
56 #include <common/sessiond-comm/relayd.h>
57 #include <common/uri.h>
58 #include <common/utils.h>
59 #include <common/config/session-config.h>
60 #include <common/dynamic-buffer.h>
61 #include <common/buffer-view.h>
62 #include <common/fd-tracker/utils.h>
63 #include <urcu/rculist.h>
67 #include "ctf-trace.h"
70 #include "lttng-relayd.h"
72 #include "health-relayd.h"
73 #include "testpoint.h"
74 #include "viewer-stream.h"
77 #include "connection.h"
78 #include "tracefile-array.h"
79 #include "tcp_keep_alive.h"
81 enum relay_connection_status
{
82 RELAY_CONNECTION_STATUS_OK
,
83 /* An error occured while processing an event on the connection. */
84 RELAY_CONNECTION_STATUS_ERROR
,
85 /* Connection closed/shutdown cleanly. */
86 RELAY_CONNECTION_STATUS_CLOSED
,
89 /* command line options */
90 char *opt_output_path
, *opt_working_directory
;
91 static int opt_daemon
, opt_background
, opt_print_version
;
92 int opt_group_output_by_session
;
93 int opt_group_output_by_host
;
96 * We need to wait for listener and live listener threads, as well as
97 * health check thread, before being ready to signal readiness.
99 #define NR_LTTNG_RELAY_READY 3
100 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
102 /* Size of receive buffer. */
103 #define RECV_DATA_BUFFER_SIZE 65536
105 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
106 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
108 static struct lttng_uri
*control_uri
;
109 static struct lttng_uri
*data_uri
;
110 static struct lttng_uri
*live_uri
;
112 const char *progname
;
114 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
115 static int tracing_group_name_override
;
117 const char * const config_section_name
= "relayd";
120 * Quit pipe for all threads. This permits a single cancellation point
121 * for all threads when receiving an event on the pipe.
123 int thread_quit_pipe
[2] = { -1, -1 };
126 * This pipe is used to inform the worker thread that a command is queued and
127 * ready to be processed.
129 static int relay_conn_pipe
[2] = { -1, -1 };
131 /* Shared between threads */
132 static int dispatch_thread_exit
;
134 static pthread_t listener_thread
;
135 static pthread_t dispatcher_thread
;
136 static pthread_t worker_thread
;
137 static pthread_t health_thread
;
140 * last_relay_stream_id_lock protects last_relay_stream_id increment
141 * atomicity on 32-bit architectures.
143 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
144 static uint64_t last_relay_stream_id
;
147 * Relay command queue.
149 * The relay_thread_listener and relay_thread_dispatcher communicate with this
152 static struct relay_conn_queue relay_conn_queue
;
154 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
155 static unsigned int lttng_opt_fd_cap
;
157 /* Global relay stream hash table. */
158 struct lttng_ht
*relay_streams_ht
;
160 /* Global relay viewer stream hash table. */
161 struct lttng_ht
*viewer_streams_ht
;
163 /* Global relay sessions hash table. */
164 struct lttng_ht
*sessions_ht
;
166 /* Relayd health monitoring */
167 struct health_app
*health_relayd
;
169 /* Global fd tracker. */
170 struct fd_tracker
*the_fd_tracker
;
172 static struct option long_options
[] = {
173 { "control-port", 1, 0, 'C', },
174 { "data-port", 1, 0, 'D', },
175 { "live-port", 1, 0, 'L', },
176 { "daemonize", 0, 0, 'd', },
177 { "background", 0, 0, 'b', },
178 { "group", 1, 0, 'g', },
179 { "fd-cap", 1, 0, '\0', },
180 { "help", 0, 0, 'h', },
181 { "output", 1, 0, 'o', },
182 { "verbose", 0, 0, 'v', },
183 { "config", 1, 0, 'f' },
184 { "version", 0, 0, 'V' },
185 { "working-directory", 1, 0, 'w', },
186 { "group-output-by-session", 0, 0, 's', },
187 { "group-output-by-host", 0, 0, 'p', },
191 static const char *config_ignore_options
[] = { "help", "config", "version" };
193 static void print_version(void) {
194 fprintf(stdout
, "%s\n", VERSION
);
197 static void relayd_config_log(void)
199 DBG("LTTng-relayd " VERSION
" - " VERSION_NAME
"%s%s",
200 GIT_VERSION
[0] == '\0' ? "" : " - " GIT_VERSION
,
201 EXTRA_VERSION_NAME
[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME
);
202 if (EXTRA_VERSION_DESCRIPTION
[0] != '\0') {
203 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION
"\n");
205 if (EXTRA_VERSION_PATCHES
[0] != '\0') {
206 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES
"\n");
211 * Take an option from the getopt output and set it in the right variable to be
214 * Return 0 on success else a negative value.
216 static int set_option(int opt
, const char *arg
, const char *optname
)
222 if (!strcmp(optname
, "fd-cap")) {
226 v
= strtoul(arg
, NULL
, 0);
227 if (errno
!= 0 || !isdigit(arg
[0])) {
228 ERR("Wrong value in --fd-cap parameter: %s", arg
);
232 if (v
< DEFAULT_RELAYD_MINIMAL_FD_CAP
) {
233 ERR("File descriptor cap must be set to at least %d",
234 DEFAULT_RELAYD_MINIMAL_FD_CAP
);
237 ERR("File descriptor cap overflow in --fd-cap parameter: %s", arg
);
241 lttng_opt_fd_cap
= (unsigned int) v
;
242 DBG3("File descriptor cap set to %u", lttng_opt_fd_cap
);
245 fprintf(stderr
, "unknown option %s", optname
);
247 fprintf(stderr
, " with arg %s\n", arg
);
252 if (lttng_is_setuid_setgid()) {
253 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
254 "-C, --control-port");
256 ret
= uri_parse(arg
, &control_uri
);
258 ERR("Invalid control URI specified");
261 if (control_uri
->port
== 0) {
262 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
267 if (lttng_is_setuid_setgid()) {
268 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
271 ret
= uri_parse(arg
, &data_uri
);
273 ERR("Invalid data URI specified");
276 if (data_uri
->port
== 0) {
277 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
282 if (lttng_is_setuid_setgid()) {
283 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
286 ret
= uri_parse(arg
, &live_uri
);
288 ERR("Invalid live URI specified");
291 if (live_uri
->port
== 0) {
292 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
303 if (lttng_is_setuid_setgid()) {
304 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
307 tracing_group_name
= strdup(arg
);
308 if (tracing_group_name
== NULL
) {
313 tracing_group_name_override
= 1;
317 ret
= utils_show_man_page(8, "lttng-relayd");
319 ERR("Cannot view man page lttng-relayd(8)");
324 opt_print_version
= 1;
327 if (lttng_is_setuid_setgid()) {
328 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
331 ret
= asprintf(&opt_output_path
, "%s", arg
);
334 PERROR("asprintf opt_output_path");
340 if (lttng_is_setuid_setgid()) {
341 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
342 "-w, --working-directory");
344 ret
= asprintf(&opt_working_directory
, "%s", arg
);
347 PERROR("asprintf working_directory");
354 /* Verbose level can increase using multiple -v */
356 lttng_opt_verbose
= config_parse_value(arg
);
358 /* Only 3 level of verbosity (-vvv). */
359 if (lttng_opt_verbose
< 3) {
360 lttng_opt_verbose
+= 1;
365 if (opt_group_output_by_host
) {
366 ERR("Cannot set --group-output-by-session, --group-output-by-host already defined");
369 opt_group_output_by_session
= 1;
372 if (opt_group_output_by_session
) {
373 ERR("Cannot set --group-output-by-host, --group-output-by-session already defined");
376 opt_group_output_by_host
= 1;
379 /* Unknown option or other error.
380 * Error is printed by getopt, just return */
393 * config_entry_handler_cb used to handle options read from a config file.
394 * See config_entry_handler_cb comment in common/config/session-config.h for the
395 * return value conventions.
397 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
401 if (!entry
|| !entry
->name
|| !entry
->value
) {
406 /* Check if the option is to be ignored */
407 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
408 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
413 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
414 /* Ignore if entry name is not fully matched. */
415 if (strcmp(entry
->name
, long_options
[i
].name
)) {
420 * If the option takes no argument on the command line,
421 * we have to check if the value is "true". We support
422 * non-zero numeric values, true, on and yes.
424 if (!long_options
[i
].has_arg
) {
425 ret
= config_parse_value(entry
->value
);
428 WARN("Invalid configuration value \"%s\" for option %s",
429 entry
->value
, entry
->name
);
431 /* False, skip boolean config option. */
436 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
440 WARN("Unrecognized option \"%s\" in daemon configuration file.",
447 static void parse_env_options(void)
451 value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV
);
453 opt_working_directory
= value
;
457 static int set_options(int argc
, char **argv
)
459 int c
, ret
= 0, option_index
= 0, retval
= 0;
460 int orig_optopt
= optopt
, orig_optind
= optind
;
461 char *default_address
, *optstring
;
462 const char *config_path
= NULL
;
464 optstring
= utils_generate_optstring(long_options
,
465 sizeof(long_options
) / sizeof(struct option
));
471 /* Check for the --config option */
473 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
474 &option_index
)) != -1) {
478 } else if (c
!= 'f') {
482 if (lttng_is_setuid_setgid()) {
483 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
486 config_path
= utils_expand_path(optarg
);
488 ERR("Failed to resolve path: %s", optarg
);
493 ret
= config_get_section_entries(config_path
, config_section_name
,
494 config_entry_handler
, NULL
);
497 ERR("Invalid configuration option at line %i", ret
);
503 /* Reset getopt's global state */
504 optopt
= orig_optopt
;
505 optind
= orig_optind
;
507 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
512 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
519 /* assign default values */
520 if (control_uri
== NULL
) {
521 ret
= asprintf(&default_address
,
522 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
523 DEFAULT_NETWORK_CONTROL_PORT
);
525 PERROR("asprintf default data address");
530 ret
= uri_parse(default_address
, &control_uri
);
531 free(default_address
);
533 ERR("Invalid control URI specified");
538 if (data_uri
== NULL
) {
539 ret
= asprintf(&default_address
,
540 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
541 DEFAULT_NETWORK_DATA_PORT
);
543 PERROR("asprintf default data address");
548 ret
= uri_parse(default_address
, &data_uri
);
549 free(default_address
);
551 ERR("Invalid data URI specified");
556 if (live_uri
== NULL
) {
557 ret
= asprintf(&default_address
,
558 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
559 DEFAULT_NETWORK_VIEWER_PORT
);
561 PERROR("asprintf default viewer control address");
566 ret
= uri_parse(default_address
, &live_uri
);
567 free(default_address
);
569 ERR("Invalid viewer control URI specified");
574 if (lttng_opt_fd_cap
== 0) {
576 struct rlimit rlimit
;
578 ret
= getrlimit(RLIMIT_NOFILE
, &rlimit
);
580 PERROR("Failed to get file descriptor limit");
584 lttng_opt_fd_cap
= rlimit
.rlim_cur
;
587 if (!opt_group_output_by_session
&& !opt_group_output_by_host
) {
588 /* Group by host by default */
589 opt_group_output_by_host
= 1;
597 static void print_global_objects(void)
599 print_viewer_streams();
600 print_relay_streams();
607 static void relayd_cleanup(void)
609 print_global_objects();
613 if (viewer_streams_ht
)
614 lttng_ht_destroy(viewer_streams_ht
);
615 if (relay_streams_ht
)
616 lttng_ht_destroy(relay_streams_ht
);
618 lttng_ht_destroy(sessions_ht
);
620 /* free the dynamically allocated opt_output_path */
621 free(opt_output_path
);
623 /* Close thread quit pipes */
624 (void) fd_tracker_util_pipe_close(the_fd_tracker
, thread_quit_pipe
);
626 uri_free(control_uri
);
628 /* Live URI is freed in the live thread. */
630 if (tracing_group_name_override
) {
631 free((void *) tracing_group_name
);
636 * Write to writable pipe used to notify a thread.
638 static int notify_thread_pipe(int wpipe
)
642 ret
= lttng_write(wpipe
, "!", 1);
644 PERROR("write poll pipe");
652 static int notify_health_quit_pipe(int *pipe
)
656 ret
= lttng_write(pipe
[1], "4", 1);
658 PERROR("write relay health quit");
667 * Stop all relayd and relayd-live threads.
669 int lttng_relay_stop_threads(void)
673 /* Stopping all threads */
674 DBG("Terminating all threads");
675 if (notify_thread_pipe(thread_quit_pipe
[1])) {
676 ERR("write error on thread quit pipe");
680 if (notify_health_quit_pipe(health_quit_pipe
)) {
681 ERR("write error on health quit pipe");
684 /* Dispatch thread */
685 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
686 futex_nto1_wake(&relay_conn_queue
.futex
);
688 if (relayd_live_stop()) {
689 ERR("Error stopping live threads");
696 * Signal handler for the daemon
698 * Simply stop all worker threads, leaving main() return gracefully after
699 * joining all threads and calling cleanup().
701 static void sighandler(int sig
)
705 DBG("SIGINT caught");
706 if (lttng_relay_stop_threads()) {
707 ERR("Error stopping threads");
711 DBG("SIGTERM caught");
712 if (lttng_relay_stop_threads()) {
713 ERR("Error stopping threads");
717 CMM_STORE_SHARED(recv_child_signal
, 1);
725 * Setup signal handler for :
726 * SIGINT, SIGTERM, SIGPIPE
728 static int set_signal_handler(void)
734 if ((ret
= sigemptyset(&sigset
)) < 0) {
735 PERROR("sigemptyset");
742 sa
.sa_handler
= sighandler
;
743 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
748 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
753 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
758 sa
.sa_handler
= SIG_IGN
;
759 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
764 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
769 void lttng_relay_notify_ready(void)
771 /* Notify the parent of the fork() process that we are ready. */
772 if (opt_daemon
|| opt_background
) {
773 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
774 kill(child_ppid
, SIGUSR1
);
780 * Init thread quit pipe.
782 * Return -1 on error or 0 if all pipes are created.
784 static int init_thread_quit_pipe(void)
786 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
787 "Quit pipe", thread_quit_pipe
);
791 * Init health quit pipe.
793 * Return -1 on error or 0 if all pipes are created.
795 static int init_health_quit_pipe(void)
797 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
798 "Health quit pipe", health_quit_pipe
);
802 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
804 static int create_named_thread_poll_set(struct lttng_poll_event
*events
,
805 int size
, const char *name
)
809 if (events
== NULL
|| size
== 0) {
814 ret
= fd_tracker_util_poll_create(the_fd_tracker
,
815 name
, events
, 1, LTTNG_CLOEXEC
);
818 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
830 * Check if the thread quit pipe was triggered.
832 * Return 1 if it was triggered else 0;
834 static int check_thread_quit_pipe(int fd
, uint32_t events
)
836 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
843 static int create_sock(void *data
, int *out_fd
)
846 struct lttcomm_sock
*sock
= data
;
848 ret
= lttcomm_create_sock(sock
);
858 static int close_sock(void *data
, int *in_fd
)
860 struct lttcomm_sock
*sock
= data
;
862 return sock
->ops
->close(sock
);
865 static int accept_sock(void *data
, int *out_fd
)
868 /* Socks is an array of in_sock, out_sock. */
869 struct lttcomm_sock
**socks
= data
;
870 struct lttcomm_sock
*in_sock
= socks
[0];
872 socks
[1] = in_sock
->ops
->accept(in_sock
);
877 *out_fd
= socks
[1]->fd
;
883 * Create and init socket from uri.
885 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
,
889 struct lttcomm_sock
*sock
= NULL
;
890 char uri_str
[PATH_MAX
];
891 char *formated_name
= NULL
;
893 sock
= lttcomm_alloc_sock_from_uri(uri
);
895 ERR("Allocating socket");
900 * Don't fail to create the socket if the name can't be built as it is
901 * only used for debugging purposes.
903 ret
= uri_to_str_url(uri
, uri_str
, sizeof(uri_str
));
904 uri_str
[sizeof(uri_str
) - 1] = '\0';
906 ret
= asprintf(&formated_name
, "%s socket @ %s", name
,
909 formated_name
= NULL
;
913 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &sock_fd
,
914 (const char **) (formated_name
? &formated_name
: NULL
),
915 1, create_sock
, sock
);
917 DBG("Listening on %s socket %d", name
, sock
->fd
);
919 ret
= sock
->ops
->bind(sock
);
924 ret
= sock
->ops
->listen(sock
, -1);
934 lttcomm_destroy_sock(sock
);
940 struct lttcomm_sock
*accept_relayd_sock(struct lttcomm_sock
*listening_sock
,
944 struct lttcomm_sock
*socks
[2] = { listening_sock
, NULL
};
945 struct lttcomm_sock
*new_sock
= NULL
;
947 ret
= fd_tracker_open_unsuspendable_fd(
948 the_fd_tracker
, &out_fd
,
949 (const char **) &name
,
950 1, accept_sock
, &socks
);
955 DBG("%s accepted, socket %d", name
, new_sock
->fd
);
961 * This thread manages the listening for new connections on the network
963 static void *relay_thread_listener(void *data
)
965 int i
, ret
, pollfd
, err
= -1;
966 uint32_t revents
, nb_fd
;
967 struct lttng_poll_event events
;
968 struct lttcomm_sock
*control_sock
, *data_sock
;
970 DBG("[thread] Relay listener started");
972 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
974 health_code_update();
976 control_sock
= relay_socket_create(control_uri
, "Control listener");
978 goto error_sock_control
;
981 data_sock
= relay_socket_create(data_uri
, "Data listener");
983 goto error_sock_relay
;
987 * Pass 3 as size here for the thread quit pipe, control and
990 ret
= create_named_thread_poll_set(&events
, 3, "Listener thread epoll");
992 goto error_create_poll
;
995 /* Add the control socket */
996 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1001 /* Add the data socket */
1002 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1004 goto error_poll_add
;
1007 lttng_relay_notify_ready();
1009 if (testpoint(relayd_thread_listener
)) {
1010 goto error_testpoint
;
1014 health_code_update();
1016 DBG("Listener accepting connections");
1019 health_poll_entry();
1020 ret
= lttng_poll_wait(&events
, -1);
1024 * Restart interrupted system call.
1026 if (errno
== EINTR
) {
1034 DBG("Relay new connection received");
1035 for (i
= 0; i
< nb_fd
; i
++) {
1036 health_code_update();
1038 /* Fetch once the poll data */
1039 revents
= LTTNG_POLL_GETEV(&events
, i
);
1040 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1044 * No activity for this FD (poll
1050 /* Thread quit pipe has been closed. Killing thread. */
1051 ret
= check_thread_quit_pipe(pollfd
, revents
);
1057 if (revents
& LPOLLIN
) {
1059 * A new connection is requested, therefore a
1060 * sessiond/consumerd connection is allocated in
1061 * this thread, enqueued to a global queue and
1062 * dequeued (and freed) in the worker thread.
1065 struct relay_connection
*new_conn
;
1066 struct lttcomm_sock
*newsock
= NULL
;
1067 enum connection_type type
;
1069 if (pollfd
== data_sock
->fd
) {
1071 newsock
= accept_relayd_sock(data_sock
,
1072 "Data socket to relayd");
1074 assert(pollfd
== control_sock
->fd
);
1075 type
= RELAY_CONTROL
;
1076 newsock
= control_sock
->ops
->accept(control_sock
);
1077 DBG("Relay control connection accepted, socket %d",
1081 PERROR("accepting sock");
1085 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
1088 PERROR("setsockopt inet");
1089 lttcomm_destroy_sock(newsock
);
1093 ret
= socket_apply_keep_alive_config(newsock
->fd
);
1095 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1097 lttcomm_destroy_sock(newsock
);
1101 new_conn
= connection_create(newsock
, type
);
1103 lttcomm_destroy_sock(newsock
);
1107 /* Enqueue request for the dispatcher thread. */
1108 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
1112 * Wake the dispatch queue futex.
1113 * Implicit memory barrier with the
1114 * exchange in cds_wfcq_enqueue.
1116 futex_nto1_wake(&relay_conn_queue
.futex
);
1117 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1118 ERR("socket poll error");
1121 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1131 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
1133 if (data_sock
->fd
>= 0) {
1134 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1135 &data_sock
->fd
, 1, close_sock
,
1141 lttcomm_destroy_sock(data_sock
);
1143 if (control_sock
->fd
>= 0) {
1144 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1145 &control_sock
->fd
, 1, close_sock
,
1151 lttcomm_destroy_sock(control_sock
);
1155 ERR("Health error occurred in %s", __func__
);
1157 health_unregister(health_relayd
);
1158 DBG("Relay listener thread cleanup complete");
1159 lttng_relay_stop_threads();
1164 * This thread manages the dispatching of the requests to worker threads
1166 static void *relay_thread_dispatcher(void *data
)
1170 struct cds_wfcq_node
*node
;
1171 struct relay_connection
*new_conn
= NULL
;
1173 DBG("[thread] Relay dispatcher started");
1175 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
1177 if (testpoint(relayd_thread_dispatcher
)) {
1178 goto error_testpoint
;
1181 health_code_update();
1184 health_code_update();
1186 /* Atomically prepare the queue futex */
1187 futex_nto1_prepare(&relay_conn_queue
.futex
);
1189 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1194 health_code_update();
1196 /* Dequeue commands */
1197 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1198 &relay_conn_queue
.tail
);
1200 DBG("Woken up but nothing in the relay command queue");
1201 /* Continue thread execution */
1204 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1206 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1209 * Inform worker thread of the new request. This
1210 * call is blocking so we can be assured that
1211 * the data will be read at some point in time
1212 * or wait to the end of the world :)
1214 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1216 PERROR("write connection pipe");
1217 connection_put(new_conn
);
1220 } while (node
!= NULL
);
1222 /* Futex wait on queue. Blocking call on futex() */
1223 health_poll_entry();
1224 futex_nto1_wait(&relay_conn_queue
.futex
);
1228 /* Normal exit, no error */
1235 ERR("Health error occurred in %s", __func__
);
1237 health_unregister(health_relayd
);
1238 DBG("Dispatch thread dying");
1239 lttng_relay_stop_threads();
1244 * Set index data from the control port to a given index object.
1246 static int set_index_control_data(struct relay_index
*index
,
1247 struct lttcomm_relayd_index
*data
,
1248 struct relay_connection
*conn
)
1250 struct ctf_packet_index index_data
;
1253 * The index on disk is encoded in big endian.
1255 index_data
.packet_size
= htobe64(data
->packet_size
);
1256 index_data
.content_size
= htobe64(data
->content_size
);
1257 index_data
.timestamp_begin
= htobe64(data
->timestamp_begin
);
1258 index_data
.timestamp_end
= htobe64(data
->timestamp_end
);
1259 index_data
.events_discarded
= htobe64(data
->events_discarded
);
1260 index_data
.stream_id
= htobe64(data
->stream_id
);
1262 if (conn
->minor
>= 8) {
1263 index
->index_data
.stream_instance_id
= htobe64(data
->stream_instance_id
);
1264 index
->index_data
.packet_seq_num
= htobe64(data
->packet_seq_num
);
1267 return relay_index_set_data(index
, &index_data
);
1271 * Handle the RELAYD_CREATE_SESSION command.
1273 * On success, send back the session id or else return a negative value.
1275 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1276 struct relay_connection
*conn
,
1277 const struct lttng_buffer_view
*payload
)
1281 struct relay_session
*session
;
1282 struct lttcomm_relayd_status_session reply
;
1283 char session_name
[LTTNG_NAME_MAX
];
1284 char hostname
[LTTNG_HOST_NAME_MAX
];
1285 uint32_t live_timer
= 0;
1286 bool snapshot
= false;
1288 memset(session_name
, 0, LTTNG_NAME_MAX
);
1289 memset(hostname
, 0, LTTNG_HOST_NAME_MAX
);
1291 memset(&reply
, 0, sizeof(reply
));
1293 switch (conn
->minor
) {
1298 case 4: /* LTTng sessiond 2.4 */
1300 ret
= cmd_create_session_2_4(payload
, session_name
,
1301 hostname
, &live_timer
, &snapshot
);
1307 session
= session_create(session_name
, hostname
, live_timer
,
1308 snapshot
, conn
->major
, conn
->minor
);
1313 assert(!conn
->session
);
1314 conn
->session
= session
;
1315 DBG("Created session %" PRIu64
, session
->id
);
1317 reply
.session_id
= htobe64(session
->id
);
1321 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1323 reply
.ret_code
= htobe32(LTTNG_OK
);
1326 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1327 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1328 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1337 * When we have received all the streams and the metadata for a channel,
1338 * we make them visible to the viewer threads.
1340 static void publish_connection_local_streams(struct relay_connection
*conn
)
1342 struct relay_stream
*stream
;
1343 struct relay_session
*session
= conn
->session
;
1346 * We publish all streams belonging to a session atomically wrt
1349 pthread_mutex_lock(&session
->lock
);
1351 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1353 stream_publish(stream
);
1358 * Inform the viewer that there are new streams in the session.
1360 if (session
->viewer_attached
) {
1361 uatomic_set(&session
->new_streams
, 1);
1363 pthread_mutex_unlock(&session
->lock
);
1367 * relay_add_stream: allocate a new stream for a session
1369 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1370 struct relay_connection
*conn
,
1371 const struct lttng_buffer_view
*payload
)
1375 struct relay_session
*session
= conn
->session
;
1376 struct relay_stream
*stream
= NULL
;
1377 struct lttcomm_relayd_status_stream reply
;
1378 struct ctf_trace
*trace
= NULL
;
1379 uint64_t stream_handle
= -1ULL;
1380 char *path_name
= NULL
, *channel_name
= NULL
;
1381 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1383 if (!session
|| !conn
->version_check_done
) {
1384 ERR("Trying to add a stream before version check");
1386 goto end_no_session
;
1389 switch (session
->minor
) {
1390 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1391 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1392 &channel_name
, session
);
1394 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1396 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1397 &channel_name
, &tracefile_size
, &tracefile_count
,
1405 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1409 /* This stream here has one reference on the trace. */
1411 pthread_mutex_lock(&last_relay_stream_id_lock
);
1412 stream_handle
= ++last_relay_stream_id
;
1413 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1415 /* We pass ownership of path_name and channel_name. */
1416 stream
= stream_create(trace
, stream_handle
, path_name
,
1417 channel_name
, tracefile_size
, tracefile_count
);
1419 channel_name
= NULL
;
1422 * Streams are the owners of their trace. Reference to trace is
1423 * kept within stream_create().
1425 ctf_trace_put(trace
);
1428 memset(&reply
, 0, sizeof(reply
));
1429 reply
.handle
= htobe64(stream_handle
);
1431 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1433 reply
.ret_code
= htobe32(LTTNG_OK
);
1436 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1437 sizeof(struct lttcomm_relayd_status_stream
), 0);
1438 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1439 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1451 * relay_close_stream: close a specific stream
1453 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1454 struct relay_connection
*conn
,
1455 const struct lttng_buffer_view
*payload
)
1459 struct relay_session
*session
= conn
->session
;
1460 struct lttcomm_relayd_close_stream stream_info
;
1461 struct lttcomm_relayd_generic_reply reply
;
1462 struct relay_stream
*stream
;
1464 DBG("Close stream received");
1466 if (!session
|| !conn
->version_check_done
) {
1467 ERR("Trying to close a stream before version check");
1469 goto end_no_session
;
1472 if (payload
->size
< sizeof(stream_info
)) {
1473 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1474 sizeof(stream_info
), payload
->size
);
1476 goto end_no_session
;
1478 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1479 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1480 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1482 stream
= stream_get_by_id(stream_info
.stream_id
);
1489 * Set last_net_seq_num before the close flag. Required by data
1492 pthread_mutex_lock(&stream
->lock
);
1493 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1494 pthread_mutex_unlock(&stream
->lock
);
1497 * This is one of the conditions which may trigger a stream close
1498 * with the others being:
1499 * 1) A close command is received for a stream
1500 * 2) The control connection owning the stream is closed
1501 * 3) We have received all of the stream's data _after_ a close
1504 try_stream_close(stream
);
1505 if (stream
->is_metadata
) {
1506 struct relay_viewer_stream
*vstream
;
1508 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1510 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1512 * Since all the metadata has been sent to the
1513 * viewer and that we have a request to close
1514 * its stream, we can safely teardown the
1515 * corresponding metadata viewer stream.
1517 viewer_stream_put(vstream
);
1519 /* Put local reference. */
1520 viewer_stream_put(vstream
);
1527 memset(&reply
, 0, sizeof(reply
));
1529 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1531 reply
.ret_code
= htobe32(LTTNG_OK
);
1533 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1534 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1535 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1536 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1546 * relay_reset_metadata: reset a metadata stream
1549 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1550 struct relay_connection
*conn
,
1551 const struct lttng_buffer_view
*payload
)
1555 struct relay_session
*session
= conn
->session
;
1556 struct lttcomm_relayd_reset_metadata stream_info
;
1557 struct lttcomm_relayd_generic_reply reply
;
1558 struct relay_stream
*stream
;
1560 DBG("Reset metadata received");
1562 if (!session
|| !conn
->version_check_done
) {
1563 ERR("Trying to reset a metadata stream before version check");
1565 goto end_no_session
;
1568 if (payload
->size
< sizeof(stream_info
)) {
1569 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1570 sizeof(stream_info
), payload
->size
);
1572 goto end_no_session
;
1574 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1575 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1576 stream_info
.version
= be64toh(stream_info
.version
);
1578 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1580 /* Unsupported for live sessions for now. */
1581 if (session
->live_timer
!= 0) {
1586 stream
= stream_get_by_id(stream_info
.stream_id
);
1591 pthread_mutex_lock(&stream
->lock
);
1592 if (!stream
->is_metadata
) {
1597 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
1598 0, 0, -1, -1, stream
->stream_fd
->fd
, NULL
,
1599 &stream
->stream_fd
->fd
);
1601 ERR("Failed to rotate metadata file %s of channel %s",
1602 stream
->path_name
, stream
->channel_name
);
1607 pthread_mutex_unlock(&stream
->lock
);
1611 memset(&reply
, 0, sizeof(reply
));
1613 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1615 reply
.ret_code
= htobe32(LTTNG_OK
);
1617 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1618 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1619 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1620 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1630 * relay_unknown_command: send -1 if received unknown command
1632 static void relay_unknown_command(struct relay_connection
*conn
)
1634 struct lttcomm_relayd_generic_reply reply
;
1637 memset(&reply
, 0, sizeof(reply
));
1638 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1639 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1640 if (send_ret
< sizeof(reply
)) {
1641 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1646 * relay_start: send an acknowledgment to the client to tell if we are
1647 * ready to receive data. We are ready if a session is established.
1649 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1650 struct relay_connection
*conn
,
1651 const struct lttng_buffer_view
*payload
)
1655 struct lttcomm_relayd_generic_reply reply
;
1656 struct relay_session
*session
= conn
->session
;
1659 DBG("Trying to start the streaming without a session established");
1660 ret
= htobe32(LTTNG_ERR_UNK
);
1663 memset(&reply
, 0, sizeof(reply
));
1664 reply
.ret_code
= htobe32(LTTNG_OK
);
1665 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1667 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1668 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1677 * Append padding to the file pointed by the file descriptor fd.
1679 static int write_padding_to_file(int fd
, uint32_t size
)
1688 zeros
= zmalloc(size
);
1689 if (zeros
== NULL
) {
1690 PERROR("zmalloc zeros for padding");
1695 ret
= lttng_write(fd
, zeros
, size
);
1697 PERROR("write padding to file");
1707 * relay_recv_metadata: receive the metadata for the session.
1709 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1710 struct relay_connection
*conn
,
1711 const struct lttng_buffer_view
*payload
)
1715 struct relay_session
*session
= conn
->session
;
1716 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1717 struct relay_stream
*metadata_stream
;
1718 uint64_t metadata_payload_size
;
1721 ERR("Metadata sent before version check");
1726 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1727 ERR("Incorrect data size");
1731 metadata_payload_size
= recv_hdr
->data_size
-
1732 sizeof(struct lttcomm_relayd_metadata_payload
);
1734 memcpy(&metadata_payload_header
, payload
->data
,
1735 sizeof(metadata_payload_header
));
1736 metadata_payload_header
.stream_id
= be64toh(
1737 metadata_payload_header
.stream_id
);
1738 metadata_payload_header
.padding_size
= be32toh(
1739 metadata_payload_header
.padding_size
);
1741 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1742 if (!metadata_stream
) {
1747 pthread_mutex_lock(&metadata_stream
->lock
);
1749 size_ret
= lttng_write(metadata_stream
->stream_fd
->fd
,
1750 payload
->data
+ sizeof(metadata_payload_header
),
1751 metadata_payload_size
);
1752 if (size_ret
< metadata_payload_size
) {
1753 ERR("Relay error writing metadata on file");
1758 size_ret
= write_padding_to_file(metadata_stream
->stream_fd
->fd
,
1759 metadata_payload_header
.padding_size
);
1760 if (size_ret
< (int64_t) metadata_payload_header
.padding_size
) {
1765 metadata_stream
->metadata_received
+=
1766 metadata_payload_size
+ metadata_payload_header
.padding_size
;
1767 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1768 metadata_stream
->metadata_received
);
1771 pthread_mutex_unlock(&metadata_stream
->lock
);
1772 stream_put(metadata_stream
);
1778 * relay_send_version: send relayd version number
1780 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1781 struct relay_connection
*conn
,
1782 const struct lttng_buffer_view
*payload
)
1786 struct lttcomm_relayd_version reply
, msg
;
1787 bool compatible
= true;
1789 conn
->version_check_done
= true;
1791 /* Get version from the other side. */
1792 if (payload
->size
< sizeof(msg
)) {
1793 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1794 sizeof(msg
), payload
->size
);
1799 memcpy(&msg
, payload
->data
, sizeof(msg
));
1800 msg
.major
= be32toh(msg
.major
);
1801 msg
.minor
= be32toh(msg
.minor
);
1803 memset(&reply
, 0, sizeof(reply
));
1804 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1805 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1807 /* Major versions must be the same */
1808 if (reply
.major
!= msg
.major
) {
1809 DBG("Incompatible major versions (%u vs %u), deleting session",
1810 reply
.major
, msg
.major
);
1814 conn
->major
= reply
.major
;
1815 /* We adapt to the lowest compatible version */
1816 if (reply
.minor
<= msg
.minor
) {
1817 conn
->minor
= reply
.minor
;
1819 conn
->minor
= msg
.minor
;
1822 reply
.major
= htobe32(reply
.major
);
1823 reply
.minor
= htobe32(reply
.minor
);
1824 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1826 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1827 ERR("Failed to send \"send version\" command reply (ret = %zd)",
1840 DBG("Version check done using protocol %u.%u", conn
->major
,
1848 * Check for data pending for a given stream id from the session daemon.
1850 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1851 struct relay_connection
*conn
,
1852 const struct lttng_buffer_view
*payload
)
1854 struct relay_session
*session
= conn
->session
;
1855 struct lttcomm_relayd_data_pending msg
;
1856 struct lttcomm_relayd_generic_reply reply
;
1857 struct relay_stream
*stream
;
1861 DBG("Data pending command received");
1863 if (!session
|| !conn
->version_check_done
) {
1864 ERR("Trying to check for data before version check");
1866 goto end_no_session
;
1869 if (payload
->size
< sizeof(msg
)) {
1870 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
1871 sizeof(msg
), payload
->size
);
1873 goto end_no_session
;
1875 memcpy(&msg
, payload
->data
, sizeof(msg
));
1876 msg
.stream_id
= be64toh(msg
.stream_id
);
1877 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1879 stream
= stream_get_by_id(msg
.stream_id
);
1880 if (stream
== NULL
) {
1885 pthread_mutex_lock(&stream
->lock
);
1887 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1888 " and last_seq %" PRIu64
, msg
.stream_id
,
1889 stream
->prev_seq
, msg
.last_net_seq_num
);
1891 /* Avoid wrapping issue */
1892 if (((int64_t) (stream
->prev_seq
- msg
.last_net_seq_num
)) >= 0) {
1893 /* Data has in fact been written and is NOT pending */
1896 /* Data still being streamed thus pending */
1900 stream
->data_pending_check_done
= true;
1901 pthread_mutex_unlock(&stream
->lock
);
1906 memset(&reply
, 0, sizeof(reply
));
1907 reply
.ret_code
= htobe32(ret
);
1908 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1909 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1910 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
1920 * Wait for the control socket to reach a quiescent state.
1922 * Note that for now, when receiving this command from the session
1923 * daemon, this means that every subsequent commands or data received on
1924 * the control socket has been handled. So, this is why we simply return
1927 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
1928 struct relay_connection
*conn
,
1929 const struct lttng_buffer_view
*payload
)
1933 struct relay_stream
*stream
;
1934 struct lttcomm_relayd_quiescent_control msg
;
1935 struct lttcomm_relayd_generic_reply reply
;
1937 DBG("Checking quiescent state on control socket");
1939 if (!conn
->session
|| !conn
->version_check_done
) {
1940 ERR("Trying to check for data before version check");
1942 goto end_no_session
;
1945 if (payload
->size
< sizeof(msg
)) {
1946 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
1947 sizeof(msg
), payload
->size
);
1949 goto end_no_session
;
1951 memcpy(&msg
, payload
->data
, sizeof(msg
));
1952 msg
.stream_id
= be64toh(msg
.stream_id
);
1954 stream
= stream_get_by_id(msg
.stream_id
);
1958 pthread_mutex_lock(&stream
->lock
);
1959 stream
->data_pending_check_done
= true;
1960 pthread_mutex_unlock(&stream
->lock
);
1962 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
1965 memset(&reply
, 0, sizeof(reply
));
1966 reply
.ret_code
= htobe32(LTTNG_OK
);
1967 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1968 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1969 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
1981 * Initialize a data pending command. This means that a consumer is about
1982 * to ask for data pending for each stream it holds. Simply iterate over
1983 * all streams of a session and set the data_pending_check_done flag.
1985 * This command returns to the client a LTTNG_OK code.
1987 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1988 struct relay_connection
*conn
,
1989 const struct lttng_buffer_view
*payload
)
1993 struct lttng_ht_iter iter
;
1994 struct lttcomm_relayd_begin_data_pending msg
;
1995 struct lttcomm_relayd_generic_reply reply
;
1996 struct relay_stream
*stream
;
2001 DBG("Init streams for data pending");
2003 if (!conn
->session
|| !conn
->version_check_done
) {
2004 ERR("Trying to check for data before version check");
2006 goto end_no_session
;
2009 if (payload
->size
< sizeof(msg
)) {
2010 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2011 sizeof(msg
), payload
->size
);
2013 goto end_no_session
;
2015 memcpy(&msg
, payload
->data
, sizeof(msg
));
2016 msg
.session_id
= be64toh(msg
.session_id
);
2019 * Iterate over all streams to set the begin data pending flag.
2020 * For now, the streams are indexed by stream handle so we have
2021 * to iterate over all streams to find the one associated with
2022 * the right session_id.
2025 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2027 if (!stream_get(stream
)) {
2030 if (stream
->trace
->session
->id
== msg
.session_id
) {
2031 pthread_mutex_lock(&stream
->lock
);
2032 stream
->data_pending_check_done
= false;
2033 pthread_mutex_unlock(&stream
->lock
);
2034 DBG("Set begin data pending flag to stream %" PRIu64
,
2035 stream
->stream_handle
);
2041 memset(&reply
, 0, sizeof(reply
));
2042 /* All good, send back reply. */
2043 reply
.ret_code
= htobe32(LTTNG_OK
);
2045 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2046 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2047 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2059 * End data pending command. This will check, for a given session id, if
2060 * each stream associated with it has its data_pending_check_done flag
2061 * set. If not, this means that the client lost track of the stream but
2062 * the data is still being streamed on our side. In this case, we inform
2063 * the client that data is in flight.
2065 * Return to the client if there is data in flight or not with a ret_code.
2067 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2068 struct relay_connection
*conn
,
2069 const struct lttng_buffer_view
*payload
)
2073 struct lttng_ht_iter iter
;
2074 struct lttcomm_relayd_end_data_pending msg
;
2075 struct lttcomm_relayd_generic_reply reply
;
2076 struct relay_stream
*stream
;
2077 uint32_t is_data_inflight
= 0;
2079 DBG("End data pending command");
2081 if (!conn
->session
|| !conn
->version_check_done
) {
2082 ERR("Trying to check for data before version check");
2084 goto end_no_session
;
2087 if (payload
->size
< sizeof(msg
)) {
2088 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2089 sizeof(msg
), payload
->size
);
2091 goto end_no_session
;
2093 memcpy(&msg
, payload
->data
, sizeof(msg
));
2094 msg
.session_id
= be64toh(msg
.session_id
);
2097 * Iterate over all streams to see if the begin data pending
2101 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2103 if (!stream_get(stream
)) {
2106 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2110 pthread_mutex_lock(&stream
->lock
);
2111 if (!stream
->data_pending_check_done
) {
2112 if (!stream
->closed
|| !(((int64_t) (stream
->prev_seq
- stream
->last_net_seq_num
)) >= 0)) {
2113 is_data_inflight
= 1;
2114 DBG("Data is still in flight for stream %" PRIu64
,
2115 stream
->stream_handle
);
2116 pthread_mutex_unlock(&stream
->lock
);
2121 pthread_mutex_unlock(&stream
->lock
);
2126 memset(&reply
, 0, sizeof(reply
));
2127 /* All good, send back reply. */
2128 reply
.ret_code
= htobe32(is_data_inflight
);
2130 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2131 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2132 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2144 * Receive an index for a specific stream.
2146 * Return 0 on success else a negative value.
2148 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2149 struct relay_connection
*conn
,
2150 const struct lttng_buffer_view
*payload
)
2154 struct relay_session
*session
= conn
->session
;
2155 struct lttcomm_relayd_index index_info
;
2156 struct relay_index
*index
;
2157 struct lttcomm_relayd_generic_reply reply
;
2158 struct relay_stream
*stream
;
2163 DBG("Relay receiving index");
2165 if (!session
|| !conn
->version_check_done
) {
2166 ERR("Trying to close a stream before version check");
2168 goto end_no_session
;
2171 msg_len
= lttcomm_relayd_index_len(
2172 lttng_to_index_major(conn
->major
, conn
->minor
),
2173 lttng_to_index_minor(conn
->major
, conn
->minor
));
2174 if (payload
->size
< msg_len
) {
2175 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2176 msg_len
, payload
->size
);
2178 goto end_no_session
;
2180 memcpy(&index_info
, payload
->data
, msg_len
);
2181 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2182 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2183 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2184 index_info
.content_size
= be64toh(index_info
.content_size
);
2185 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2186 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2187 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2188 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2190 if (conn
->minor
>= 8) {
2191 index_info
.stream_instance_id
=
2192 be64toh(index_info
.stream_instance_id
);
2193 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2196 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2198 ERR("stream_get_by_id not found");
2202 pthread_mutex_lock(&stream
->lock
);
2204 /* Live beacon handling */
2205 if (index_info
.packet_size
== 0) {
2206 DBG("Received live beacon for stream %" PRIu64
,
2207 stream
->stream_handle
);
2210 * Only flag a stream inactive when it has already
2211 * received data and no indexes are in flight.
2213 if (stream
->index_received_seqcount
> 0
2214 && stream
->indexes_in_flight
== 0) {
2215 stream
->beacon_ts_end
= index_info
.timestamp_end
;
2218 goto end_stream_put
;
2220 stream
->beacon_ts_end
= -1ULL;
2223 if (stream
->ctf_stream_id
== -1ULL) {
2224 stream
->ctf_stream_id
= index_info
.stream_id
;
2226 index
= relay_index_get_by_id_or_create(stream
, index_info
.net_seq_num
);
2229 ERR("relay_index_get_by_id_or_create index NULL");
2230 goto end_stream_put
;
2232 if (set_index_control_data(index
, &index_info
, conn
)) {
2233 ERR("set_index_control_data error");
2234 relay_index_put(index
);
2236 goto end_stream_put
;
2238 ret
= relay_index_try_flush(index
);
2240 tracefile_array_commit_seq(stream
->tfa
);
2241 stream
->index_received_seqcount
++;
2242 } else if (ret
> 0) {
2246 ERR("relay_index_try_flush error %d", ret
);
2247 relay_index_put(index
);
2252 pthread_mutex_unlock(&stream
->lock
);
2257 memset(&reply
, 0, sizeof(reply
));
2259 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2261 reply
.ret_code
= htobe32(LTTNG_OK
);
2263 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2264 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2265 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2274 * Receive the streams_sent message.
2276 * Return 0 on success else a negative value.
2278 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2279 struct relay_connection
*conn
,
2280 const struct lttng_buffer_view
*payload
)
2284 struct lttcomm_relayd_generic_reply reply
;
2288 DBG("Relay receiving streams_sent");
2290 if (!conn
->session
|| !conn
->version_check_done
) {
2291 ERR("Trying to close a stream before version check");
2293 goto end_no_session
;
2297 * Publish every pending stream in the connection recv list which are
2298 * now ready to be used by the viewer.
2300 publish_connection_local_streams(conn
);
2302 memset(&reply
, 0, sizeof(reply
));
2303 reply
.ret_code
= htobe32(LTTNG_OK
);
2304 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2305 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2306 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2318 #define DBG_CMD(cmd_name, conn) \
2319 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2321 static int relay_process_control_command(struct relay_connection
*conn
,
2322 const struct lttcomm_relayd_hdr
*header
,
2323 const struct lttng_buffer_view
*payload
)
2327 switch (header
->cmd
) {
2328 case RELAYD_CREATE_SESSION
:
2329 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
2330 ret
= relay_create_session(header
, conn
, payload
);
2332 case RELAYD_ADD_STREAM
:
2333 DBG_CMD("RELAYD_ADD_STREAM", conn
);
2334 ret
= relay_add_stream(header
, conn
, payload
);
2336 case RELAYD_START_DATA
:
2337 DBG_CMD("RELAYD_START_DATA", conn
);
2338 ret
= relay_start(header
, conn
, payload
);
2340 case RELAYD_SEND_METADATA
:
2341 DBG_CMD("RELAYD_SEND_METADATA", conn
);
2342 ret
= relay_recv_metadata(header
, conn
, payload
);
2344 case RELAYD_VERSION
:
2345 DBG_CMD("RELAYD_VERSION", conn
);
2346 ret
= relay_send_version(header
, conn
, payload
);
2348 case RELAYD_CLOSE_STREAM
:
2349 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
2350 ret
= relay_close_stream(header
, conn
, payload
);
2352 case RELAYD_DATA_PENDING
:
2353 DBG_CMD("RELAYD_DATA_PENDING", conn
);
2354 ret
= relay_data_pending(header
, conn
, payload
);
2356 case RELAYD_QUIESCENT_CONTROL
:
2357 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
2358 ret
= relay_quiescent_control(header
, conn
, payload
);
2360 case RELAYD_BEGIN_DATA_PENDING
:
2361 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
2362 ret
= relay_begin_data_pending(header
, conn
, payload
);
2364 case RELAYD_END_DATA_PENDING
:
2365 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
2366 ret
= relay_end_data_pending(header
, conn
, payload
);
2368 case RELAYD_SEND_INDEX
:
2369 DBG_CMD("RELAYD_SEND_INDEX", conn
);
2370 ret
= relay_recv_index(header
, conn
, payload
);
2372 case RELAYD_STREAMS_SENT
:
2373 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
2374 ret
= relay_streams_sent(header
, conn
, payload
);
2376 case RELAYD_RESET_METADATA
:
2377 DBG_CMD("RELAYD_RESET_METADATA", conn
);
2378 ret
= relay_reset_metadata(header
, conn
, payload
);
2380 case RELAYD_UPDATE_SYNC_INFO
:
2382 ERR("Received unknown command (%u)", header
->cmd
);
2383 relay_unknown_command(conn
);
2392 static enum relay_connection_status
relay_process_control_receive_payload(
2393 struct relay_connection
*conn
)
2396 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2397 struct lttng_dynamic_buffer
*reception_buffer
=
2398 &conn
->protocol
.ctrl
.reception_buffer
;
2399 struct ctrl_connection_state_receive_payload
*state
=
2400 &conn
->protocol
.ctrl
.state
.receive_payload
;
2401 struct lttng_buffer_view payload_view
;
2403 if (state
->left_to_receive
== 0) {
2404 /* Short-circuit for payload-less commands. */
2405 goto reception_complete
;
2407 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2408 reception_buffer
->data
+ state
->received
,
2409 state
->left_to_receive
, MSG_DONTWAIT
);
2411 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2412 PERROR("Unable to receive command payload on sock %d",
2414 status
= RELAY_CONNECTION_STATUS_ERROR
;
2417 } else if (ret
== 0) {
2418 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2419 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2424 assert(ret
<= state
->left_to_receive
);
2426 state
->left_to_receive
-= ret
;
2427 state
->received
+= ret
;
2429 if (state
->left_to_receive
> 0) {
2431 * Can't transition to the protocol's next state, wait to
2432 * receive the rest of the header.
2434 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2435 state
->received
, state
->left_to_receive
,
2441 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
2442 conn
->sock
->fd
, state
->received
);
2444 * The payload required to process the command has been received.
2445 * A view to the reception buffer is forwarded to the various
2446 * commands and the state of the control is reset on success.
2448 * Commands are responsible for sending their reply to the peer.
2450 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
2452 ret
= relay_process_control_command(conn
,
2453 &state
->header
, &payload_view
);
2455 status
= RELAY_CONNECTION_STATUS_ERROR
;
2459 ret
= connection_reset_protocol_state(conn
);
2461 status
= RELAY_CONNECTION_STATUS_ERROR
;
2467 static enum relay_connection_status
relay_process_control_receive_header(
2468 struct relay_connection
*conn
)
2471 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2472 struct lttcomm_relayd_hdr header
;
2473 struct lttng_dynamic_buffer
*reception_buffer
=
2474 &conn
->protocol
.ctrl
.reception_buffer
;
2475 struct ctrl_connection_state_receive_header
*state
=
2476 &conn
->protocol
.ctrl
.state
.receive_header
;
2478 assert(state
->left_to_receive
!= 0);
2480 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2481 reception_buffer
->data
+ state
->received
,
2482 state
->left_to_receive
, MSG_DONTWAIT
);
2484 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2485 PERROR("Unable to receive control command header on sock %d",
2487 status
= RELAY_CONNECTION_STATUS_ERROR
;
2490 } else if (ret
== 0) {
2491 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2492 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2497 assert(ret
<= state
->left_to_receive
);
2499 state
->left_to_receive
-= ret
;
2500 state
->received
+= ret
;
2502 if (state
->left_to_receive
> 0) {
2504 * Can't transition to the protocol's next state, wait to
2505 * receive the rest of the header.
2507 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2508 state
->received
, state
->left_to_receive
,
2513 /* Transition to next state: receiving the command's payload. */
2514 conn
->protocol
.ctrl
.state_id
=
2515 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
2516 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
2517 header
.circuit_id
= be64toh(header
.circuit_id
);
2518 header
.data_size
= be64toh(header
.data_size
);
2519 header
.cmd
= be32toh(header
.cmd
);
2520 header
.cmd_version
= be32toh(header
.cmd_version
);
2521 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
2522 &header
, sizeof(header
));
2524 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
2525 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
2528 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
2529 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
2531 status
= RELAY_CONNECTION_STATUS_ERROR
;
2535 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
2537 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
2538 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
2541 status
= RELAY_CONNECTION_STATUS_ERROR
;
2545 if (header
.data_size
== 0) {
2547 * Manually invoke the next state as the poll loop
2548 * will not wake-up to allow us to proceed further.
2550 status
= relay_process_control_receive_payload(conn
);
2557 * Process the commands received on the control socket
2559 static enum relay_connection_status
relay_process_control(
2560 struct relay_connection
*conn
)
2562 enum relay_connection_status status
;
2564 switch (conn
->protocol
.ctrl
.state_id
) {
2565 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
2566 status
= relay_process_control_receive_header(conn
);
2568 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
2569 status
= relay_process_control_receive_payload(conn
);
2572 ERR("Unknown control connection protocol state encountered.");
2580 * Handle index for a data stream.
2582 * Called with the stream lock held.
2584 * Return 0 on success else a negative value.
2586 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2590 uint64_t data_offset
;
2591 struct relay_index
*index
;
2593 /* Get data offset because we are about to update the index. */
2594 data_offset
= htobe64(stream
->tracefile_size_current
);
2596 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
2597 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
2600 * Lookup for an existing index for that stream id/sequence
2601 * number. If it exists, the control thread has already received the
2602 * data for it, thus we need to write it to disk.
2604 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
2610 if (rotate_index
|| !stream
->index_file
) {
2611 uint32_t major
, minor
;
2613 /* Put ref on previous index_file. */
2614 if (stream
->index_file
) {
2615 lttng_index_file_put(stream
->index_file
);
2616 stream
->index_file
= NULL
;
2618 major
= stream
->trace
->session
->major
;
2619 minor
= stream
->trace
->session
->minor
;
2620 stream
->index_file
= lttng_index_file_create(stream
->path_name
,
2621 stream
->channel_name
,
2622 -1, -1, stream
->tracefile_size
,
2623 tracefile_array_get_file_index_head(stream
->tfa
),
2624 lttng_to_index_major(major
, minor
),
2625 lttng_to_index_minor(major
, minor
));
2626 if (!stream
->index_file
) {
2628 /* Put self-ref for this index due to error. */
2629 relay_index_put(index
);
2635 if (relay_index_set_file(index
, stream
->index_file
, data_offset
)) {
2637 /* Put self-ref for this index due to error. */
2638 relay_index_put(index
);
2643 ret
= relay_index_try_flush(index
);
2645 tracefile_array_commit_seq(stream
->tfa
);
2646 stream
->index_received_seqcount
++;
2647 } else if (ret
> 0) {
2651 /* Put self-ref for this index due to error. */
2652 relay_index_put(index
);
2660 static enum relay_connection_status
relay_process_data_receive_header(
2661 struct relay_connection
*conn
)
2664 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2665 struct data_connection_state_receive_header
*state
=
2666 &conn
->protocol
.data
.state
.receive_header
;
2667 struct lttcomm_relayd_data_hdr header
;
2668 struct relay_stream
*stream
;
2670 assert(state
->left_to_receive
!= 0);
2672 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2673 state
->header_reception_buffer
+ state
->received
,
2674 state
->left_to_receive
, MSG_DONTWAIT
);
2676 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2677 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2678 status
= RELAY_CONNECTION_STATUS_ERROR
;
2681 } else if (ret
== 0) {
2682 /* Orderly shutdown. Not necessary to print an error. */
2683 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2684 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2689 assert(ret
<= state
->left_to_receive
);
2691 state
->left_to_receive
-= ret
;
2692 state
->received
+= ret
;
2694 if (state
->left_to_receive
> 0) {
2696 * Can't transition to the protocol's next state, wait to
2697 * receive the rest of the header.
2699 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2700 state
->received
, state
->left_to_receive
,
2706 /* Transition to next state: receiving the payload. */
2707 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
2709 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
2710 header
.circuit_id
= be64toh(header
.circuit_id
);
2711 header
.stream_id
= be64toh(header
.stream_id
);
2712 header
.data_size
= be32toh(header
.data_size
);
2713 header
.net_seq_num
= be64toh(header
.net_seq_num
);
2714 header
.padding_size
= be32toh(header
.padding_size
);
2715 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
2717 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
2719 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
2720 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
2722 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
2723 conn
->sock
->fd
, header
.circuit_id
,
2724 header
.stream_id
, header
.data_size
,
2725 header
.net_seq_num
, header
.padding_size
);
2727 stream
= stream_get_by_id(header
.stream_id
);
2729 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
2731 /* Protocol error. */
2732 status
= RELAY_CONNECTION_STATUS_ERROR
;
2736 pthread_mutex_lock(&stream
->lock
);
2738 /* Check if a rotation is needed. */
2739 if (stream
->tracefile_size
> 0 &&
2740 (stream
->tracefile_size_current
+ header
.data_size
) >
2741 stream
->tracefile_size
) {
2742 uint64_t old_id
, new_id
;
2744 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
2745 tracefile_array_file_rotate(stream
->tfa
);
2747 /* new_id is updated by utils_rotate_stream_file. */
2750 ret
= utils_rotate_stream_file(stream
->path_name
,
2751 stream
->channel_name
, stream
->tracefile_size
,
2752 stream
->tracefile_count
, -1,
2753 -1, stream
->stream_fd
->fd
,
2754 &new_id
, &stream
->stream_fd
->fd
);
2756 ERR("Failed to rotate stream output file");
2757 status
= RELAY_CONNECTION_STATUS_ERROR
;
2758 goto end_stream_unlock
;
2762 * Reset current size because we just performed a stream
2765 stream
->tracefile_size_current
= 0;
2766 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= true;
2771 pthread_mutex_unlock(&stream
->lock
);
2777 static enum relay_connection_status
relay_process_data_receive_payload(
2778 struct relay_connection
*conn
)
2781 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2782 struct relay_stream
*stream
;
2783 struct data_connection_state_receive_payload
*state
=
2784 &conn
->protocol
.data
.state
.receive_payload
;
2785 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
2786 char data_buffer
[chunk_size
];
2787 bool partial_recv
= false;
2788 bool new_stream
= false, close_requested
= false;
2789 uint64_t left_to_receive
= state
->left_to_receive
;
2790 struct relay_session
*session
;
2792 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
2793 state
->header
.stream_id
, state
->header
.net_seq_num
,
2794 state
->received
, left_to_receive
);
2796 stream
= stream_get_by_id(state
->header
.stream_id
);
2798 /* Protocol error. */
2799 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
2800 state
->header
.stream_id
);
2801 status
= RELAY_CONNECTION_STATUS_ERROR
;
2805 pthread_mutex_lock(&stream
->lock
);
2806 session
= stream
->trace
->session
;
2807 if (!conn
->session
) {
2808 ret
= connection_set_session(conn
, session
);
2810 status
= RELAY_CONNECTION_STATUS_ERROR
;
2811 goto end_stream_unlock
;
2816 * The size of the "chunk" received on any iteration is bounded by:
2817 * - the data left to receive,
2818 * - the data immediately available on the socket,
2819 * - the on-stack data buffer
2821 while (left_to_receive
> 0 && !partial_recv
) {
2823 size_t recv_size
= min(left_to_receive
, chunk_size
);
2825 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
2826 recv_size
, MSG_DONTWAIT
);
2828 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2829 PERROR("Socket %d error", conn
->sock
->fd
);
2830 status
= RELAY_CONNECTION_STATUS_ERROR
;
2832 goto end_stream_unlock
;
2833 } else if (ret
== 0) {
2834 /* No more data ready to be consumed on socket. */
2835 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
2836 state
->header
.stream_id
);
2837 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2839 } else if (ret
< (int) recv_size
) {
2841 * All the data available on the socket has been
2844 partial_recv
= true;
2849 /* Write data to stream output fd. */
2850 write_ret
= lttng_write(stream
->stream_fd
->fd
, data_buffer
,
2852 if (write_ret
< (ssize_t
) recv_size
) {
2853 ERR("Relay error writing data to file");
2854 status
= RELAY_CONNECTION_STATUS_ERROR
;
2855 goto end_stream_unlock
;
2858 left_to_receive
-= recv_size
;
2859 state
->received
+= recv_size
;
2860 state
->left_to_receive
= left_to_receive
;
2862 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
2863 write_ret
, stream
->stream_handle
);
2866 if (state
->left_to_receive
> 0) {
2868 * Did not receive all the data expected, wait for more data to
2869 * become available on the socket.
2871 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
2872 state
->header
.stream_id
, state
->received
,
2873 state
->left_to_receive
);
2874 goto end_stream_unlock
;
2877 ret
= write_padding_to_file(stream
->stream_fd
->fd
,
2878 state
->header
.padding_size
);
2879 if ((int64_t) ret
< (int64_t) state
->header
.padding_size
) {
2880 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2881 stream
->stream_handle
,
2882 state
->header
.net_seq_num
, ret
);
2883 status
= RELAY_CONNECTION_STATUS_ERROR
;
2884 goto end_stream_unlock
;
2888 if (session
->minor
>= 4 && !session
->snapshot
) {
2889 ret
= handle_index_data(stream
, state
->header
.net_seq_num
,
2890 state
->rotate_index
);
2892 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2893 stream
->stream_handle
,
2894 state
->header
.net_seq_num
, ret
);
2895 status
= RELAY_CONNECTION_STATUS_ERROR
;
2896 goto end_stream_unlock
;
2900 stream
->tracefile_size_current
+= state
->header
.data_size
+
2901 state
->header
.padding_size
;
2903 if (stream
->prev_seq
== -1ULL) {
2907 stream
->prev_seq
= state
->header
.net_seq_num
;
2910 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
2911 * contents of *state which are aliased (union) to the same location as
2912 * the new state. Don't use it beyond this point.
2914 connection_reset_protocol_state(conn
);
2918 close_requested
= stream
->close_requested
;
2919 pthread_mutex_unlock(&stream
->lock
);
2920 if (close_requested
&& left_to_receive
== 0) {
2921 try_stream_close(stream
);
2925 pthread_mutex_lock(&session
->lock
);
2926 uatomic_set(&session
->new_streams
, 1);
2927 pthread_mutex_unlock(&session
->lock
);
2936 * relay_process_data: Process the data received on the data socket
2938 static enum relay_connection_status
relay_process_data(
2939 struct relay_connection
*conn
)
2941 enum relay_connection_status status
;
2943 switch (conn
->protocol
.data
.state_id
) {
2944 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
2945 status
= relay_process_data_receive_header(conn
);
2947 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
2948 status
= relay_process_data_receive_payload(conn
);
2951 ERR("Unexpected data connection communication state.");
2958 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2962 (void) lttng_poll_del(events
, pollfd
);
2964 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
, &pollfd
, 1,
2965 fd_tracker_util_close_fd
, NULL
);
2967 ERR("Closing pollfd %d", pollfd
);
2971 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
2972 int pollfd
, struct relay_connection
*conn
)
2974 const char *type_str
;
2976 switch (conn
->type
) {
2981 type_str
= "Control";
2983 case RELAY_VIEWER_COMMAND
:
2984 type_str
= "Viewer Command";
2986 case RELAY_VIEWER_NOTIFICATION
:
2987 type_str
= "Viewer Notification";
2990 type_str
= "Unknown";
2992 cleanup_connection_pollfd(events
, pollfd
);
2993 connection_put(conn
);
2994 DBG("%s connection closed with %d", type_str
, pollfd
);
2998 * This thread does the actual work
3000 static void *relay_thread_worker(void *data
)
3002 int ret
, err
= -1, last_seen_data_fd
= -1;
3004 struct lttng_poll_event events
;
3005 struct lttng_ht
*relay_connections_ht
;
3006 struct lttng_ht_iter iter
;
3007 struct relay_connection
*destroy_conn
= NULL
;
3009 DBG("[thread] Relay worker started");
3011 rcu_register_thread();
3013 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3015 if (testpoint(relayd_thread_worker
)) {
3016 goto error_testpoint
;
3019 health_code_update();
3021 /* table of connections indexed on socket */
3022 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3023 if (!relay_connections_ht
) {
3024 goto relay_connections_ht_error
;
3027 ret
= create_named_thread_poll_set(&events
, 2, "Worker thread epoll");
3029 goto error_poll_create
;
3032 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3039 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3041 health_code_update();
3043 /* Infinite blocking call, waiting for transmission */
3044 DBG3("Relayd worker thread polling...");
3045 health_poll_entry();
3046 ret
= lttng_poll_wait(&events
, -1);
3050 * Restart interrupted system call.
3052 if (errno
== EINTR
) {
3061 * Process control. The control connection is
3062 * prioritized so we don't starve it with high
3063 * throughput tracing data on the data connection.
3065 for (i
= 0; i
< nb_fd
; i
++) {
3066 /* Fetch once the poll data */
3067 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3068 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3070 health_code_update();
3074 * No activity for this FD (poll
3080 /* Thread quit pipe has been closed. Killing thread. */
3081 ret
= check_thread_quit_pipe(pollfd
, revents
);
3087 /* Inspect the relay conn pipe for new connection */
3088 if (pollfd
== relay_conn_pipe
[0]) {
3089 if (revents
& LPOLLIN
) {
3090 struct relay_connection
*conn
;
3092 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3096 lttng_poll_add(&events
, conn
->sock
->fd
,
3097 LPOLLIN
| LPOLLRDHUP
);
3098 connection_ht_add(relay_connections_ht
, conn
);
3099 DBG("Connection socket %d added", conn
->sock
->fd
);
3100 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3101 ERR("Relay connection pipe error");
3104 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3108 struct relay_connection
*ctrl_conn
;
3110 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3111 /* If not found, there is a synchronization issue. */
3114 if (ctrl_conn
->type
== RELAY_DATA
) {
3115 if (revents
& LPOLLIN
) {
3117 * Flag the last seen data fd not deleted. It will be
3118 * used as the last seen fd if any fd gets deleted in
3121 last_notdel_data_fd
= pollfd
;
3123 goto put_ctrl_connection
;
3125 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3127 if (revents
& LPOLLIN
) {
3128 enum relay_connection_status status
;
3130 status
= relay_process_control(ctrl_conn
);
3131 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3133 * On socket error flag the session as aborted to force
3134 * the cleanup of its stream otherwise it can leak
3135 * during the lifetime of the relayd.
3137 * This prevents situations in which streams can be
3138 * left opened because an index was received, the
3139 * control connection is closed, and the data
3140 * connection is closed (uncleanly) before the packet's
3143 * Since the control connection encountered an error,
3144 * it is okay to be conservative and close the
3145 * session right now as we can't rely on the protocol
3146 * being respected anymore.
3148 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3149 session_abort(ctrl_conn
->session
);
3152 /* Clear the connection on error or close. */
3153 relay_thread_close_connection(&events
,
3158 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3159 relay_thread_close_connection(&events
,
3161 if (last_seen_data_fd
== pollfd
) {
3162 last_seen_data_fd
= last_notdel_data_fd
;
3165 ERR("Unexpected poll events %u for control sock %d",
3167 connection_put(ctrl_conn
);
3170 put_ctrl_connection
:
3171 connection_put(ctrl_conn
);
3176 * The last loop handled a control request, go back to poll to make
3177 * sure we prioritise the control socket.
3183 if (last_seen_data_fd
>= 0) {
3184 for (i
= 0; i
< nb_fd
; i
++) {
3185 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3187 health_code_update();
3189 if (last_seen_data_fd
== pollfd
) {
3196 /* Process data connection. */
3197 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3198 /* Fetch the poll data. */
3199 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3200 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3201 struct relay_connection
*data_conn
;
3203 health_code_update();
3206 /* No activity for this FD (poll implementation). */
3210 /* Skip the command pipe. It's handled in the first loop. */
3211 if (pollfd
== relay_conn_pipe
[0]) {
3215 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3217 /* Skip it. Might be removed before. */
3220 if (data_conn
->type
== RELAY_CONTROL
) {
3221 goto put_data_connection
;
3223 assert(data_conn
->type
== RELAY_DATA
);
3225 if (revents
& LPOLLIN
) {
3226 enum relay_connection_status status
;
3228 status
= relay_process_data(data_conn
);
3229 /* Connection closed or error. */
3230 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3232 * On socket error flag the session as aborted to force
3233 * the cleanup of its stream otherwise it can leak
3234 * during the lifetime of the relayd.
3236 * This prevents situations in which streams can be
3237 * left opened because an index was received, the
3238 * control connection is closed, and the data
3239 * connection is closed (uncleanly) before the packet's
3242 * Since the data connection encountered an error,
3243 * it is okay to be conservative and close the
3244 * session right now as we can't rely on the protocol
3245 * being respected anymore.
3247 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3248 session_abort(data_conn
->session
);
3250 relay_thread_close_connection(&events
, pollfd
,
3253 * Every goto restart call sets the last seen fd where
3254 * here we don't really care since we gracefully
3255 * continue the loop after the connection is deleted.
3258 /* Keep last seen port. */
3259 last_seen_data_fd
= pollfd
;
3260 connection_put(data_conn
);
3263 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3264 relay_thread_close_connection(&events
, pollfd
,
3267 ERR("Unknown poll events %u for data sock %d",
3270 put_data_connection
:
3271 connection_put(data_conn
);
3273 last_seen_data_fd
= -1;
3276 /* Normal exit, no error */
3281 /* Cleanup reamaining connection object. */
3283 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3286 health_code_update();
3288 session_abort(destroy_conn
->session
);
3291 * No need to grab another ref, because we own
3294 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
3299 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
3301 lttng_ht_destroy(relay_connections_ht
);
3302 relay_connections_ht_error
:
3303 /* Close relay conn pipes */
3304 (void) fd_tracker_util_pipe_close(the_fd_tracker
,
3307 DBG("Thread exited with error");
3309 DBG("Worker thread cleanup complete");
3313 ERR("Health error occurred in %s", __func__
);
3315 health_unregister(health_relayd
);
3316 rcu_unregister_thread();
3317 lttng_relay_stop_threads();
3322 * Create the relay command pipe to wake thread_manage_apps.
3323 * Closed in cleanup().
3325 static int create_relay_conn_pipe(void)
3327 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
3328 "Relayd connection pipe", relay_conn_pipe
);
3332 int stdio_open(void *data
, int *fds
)
3334 fds
[0] = fileno(stdout
);
3335 fds
[1] = fileno(stderr
);
3340 int noop_close(void *data
, int *fds
)
3346 int track_stdio(void)
3349 const char *names
[] = { "stdout", "stderr" };
3351 return fd_tracker_open_unsuspendable_fd(the_fd_tracker
, fds
,
3352 names
, 2, stdio_open
, NULL
);
3356 void untrack_stdio(void)
3358 int fds
[] = { fileno(stdout
), fileno(stderr
) };
3361 * noop_close is used since we don't really want to close
3362 * the stdio output fds; we merely want to stop tracking them.
3364 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
3365 fds
, 2, noop_close
, NULL
);
3371 int main(int argc
, char **argv
)
3373 int ret
= 0, retval
= 0;
3376 /* Parse environment variables */
3377 parse_env_options();
3381 * Command line arguments overwrite environment.
3384 if (set_options(argc
, argv
)) {
3389 if (set_signal_handler()) {
3394 relayd_config_log();
3396 if (opt_print_version
) {
3402 ret
= fclose(stdin
);
3404 PERROR("Failed to close stdin");
3407 /* Try to create directory if -o, --output is specified. */
3408 if (opt_output_path
) {
3409 if (*opt_output_path
!= '/') {
3410 ERR("Please specify an absolute path for -o, --output PATH");
3415 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
3418 ERR("Unable to create %s", opt_output_path
);
3425 if (opt_daemon
|| opt_background
) {
3426 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
3434 if (opt_working_directory
) {
3435 ret
= utils_change_working_dir(opt_working_directory
);
3437 ERR("Changing working directory");
3442 * The RCU thread registration (and use, through the fd-tracker's
3443 * creation) is done after the daemonization to allow us to not
3444 * deal with liburcu's fork() management as the call RCU needs to
3447 rcu_register_thread();
3449 the_fd_tracker
= fd_tracker_create(lttng_opt_fd_cap
);
3450 if (!the_fd_tracker
) {
3455 ret
= track_stdio();
3461 /* Initialize thread health monitoring */
3462 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
3463 if (!health_relayd
) {
3464 PERROR("health_app_create error");
3466 goto exit_health_app_create
;
3469 /* Create thread quit pipe */
3470 if (init_thread_quit_pipe()) {
3472 goto exit_init_data
;
3475 /* Setup the thread apps communication pipe. */
3476 if (create_relay_conn_pipe()) {
3478 goto exit_init_data
;
3481 /* Init relay command queue. */
3482 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
3484 /* Initialize communication library */
3486 lttcomm_inet_init();
3488 /* tables of sessions indexed by session ID */
3489 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3492 goto exit_init_data
;
3495 /* tables of streams indexed by stream ID */
3496 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3497 if (!relay_streams_ht
) {
3499 goto exit_init_data
;
3502 /* tables of streams indexed by stream ID */
3503 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3504 if (!viewer_streams_ht
) {
3506 goto exit_init_data
;
3509 ret
= init_health_quit_pipe();
3512 goto exit_health_quit_pipe
;
3515 /* Create thread to manage the client socket */
3516 ret
= pthread_create(&health_thread
, default_pthread_attr(),
3517 thread_manage_health
, (void *) NULL
);
3520 PERROR("pthread_create health");
3522 goto exit_health_thread
;
3525 /* Setup the dispatcher thread */
3526 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
3527 relay_thread_dispatcher
, (void *) NULL
);
3530 PERROR("pthread_create dispatcher");
3532 goto exit_dispatcher_thread
;
3535 /* Setup the worker thread */
3536 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
3537 relay_thread_worker
, NULL
);
3540 PERROR("pthread_create worker");
3542 goto exit_worker_thread
;
3545 /* Setup the listener thread */
3546 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
3547 relay_thread_listener
, (void *) NULL
);
3550 PERROR("pthread_create listener");
3552 goto exit_listener_thread
;
3555 ret
= relayd_live_create(live_uri
);
3557 ERR("Starting live viewer threads");
3563 * This is where we start awaiting program completion (e.g. through
3564 * signal that asks threads to teardown).
3567 ret
= relayd_live_join();
3573 ret
= pthread_join(listener_thread
, &status
);
3576 PERROR("pthread_join listener_thread");
3580 exit_listener_thread
:
3581 ret
= pthread_join(worker_thread
, &status
);
3584 PERROR("pthread_join worker_thread");
3589 ret
= pthread_join(dispatcher_thread
, &status
);
3592 PERROR("pthread_join dispatcher_thread");
3595 exit_dispatcher_thread
:
3597 ret
= pthread_join(health_thread
, &status
);
3600 PERROR("pthread_join health_thread");
3605 (void) fd_tracker_util_pipe_close(the_fd_tracker
, health_quit_pipe
);
3606 exit_health_quit_pipe
:
3609 health_app_destroy(health_relayd
);
3610 exit_health_app_create
:
3613 * Wait for all pending call_rcu work to complete before tearing
3614 * down data structures. call_rcu worker may be trying to
3615 * perform lookups in those structures.
3620 /* Ensure all prior call_rcu are done. */
3625 * fd_tracker_destroy() will log the contents of the fd-tracker
3626 * if a leak is detected.
3628 fd_tracker_destroy(the_fd_tracker
);
3629 rcu_unregister_thread();