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/align.h>
60 #include <common/config/session-config.h>
61 #include <common/dynamic-buffer.h>
62 #include <common/buffer-view.h>
63 #include <common/fd-tracker/utils.h>
64 #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 static const char *help_msg
=
82 #ifdef LTTNG_EMBED_HELP
83 #include <lttng-relayd.8.h>
89 enum relay_connection_status
{
90 RELAY_CONNECTION_STATUS_OK
,
91 /* An error occured while processing an event on the connection. */
92 RELAY_CONNECTION_STATUS_ERROR
,
93 /* Connection closed/shutdown cleanly. */
94 RELAY_CONNECTION_STATUS_CLOSED
,
97 /* command line options */
98 char *opt_output_path
;
99 static int opt_daemon
, opt_background
;
102 * We need to wait for listener and live listener threads, as well as
103 * health check thread, before being ready to signal readiness.
105 #define NR_LTTNG_RELAY_READY 3
106 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
108 /* Size of receive buffer. */
109 #define RECV_DATA_BUFFER_SIZE 65536
110 #define FILE_COPY_BUFFER_SIZE 65536
112 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
113 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
115 static struct lttng_uri
*control_uri
;
116 static struct lttng_uri
*data_uri
;
117 static struct lttng_uri
*live_uri
;
119 const char *progname
;
121 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
122 static int tracing_group_name_override
;
124 const char * const config_section_name
= "relayd";
127 * Quit pipe for all threads. This permits a single cancellation point
128 * for all threads when receiving an event on the pipe.
130 int thread_quit_pipe
[2] = { -1, -1 };
133 * This pipe is used to inform the worker thread that a command is queued and
134 * ready to be processed.
136 static int relay_conn_pipe
[2] = { -1, -1 };
138 /* Shared between threads */
139 static int dispatch_thread_exit
;
141 static pthread_t listener_thread
;
142 static pthread_t dispatcher_thread
;
143 static pthread_t worker_thread
;
144 static pthread_t health_thread
;
147 * last_relay_stream_id_lock protects last_relay_stream_id increment
148 * atomicity on 32-bit architectures.
150 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
151 static uint64_t last_relay_stream_id
;
154 * Relay command queue.
156 * The relay_thread_listener and relay_thread_dispatcher communicate with this
159 static struct relay_conn_queue relay_conn_queue
;
161 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
162 static unsigned int lttng_opt_fd_cap
;
164 /* Global relay stream hash table. */
165 struct lttng_ht
*relay_streams_ht
;
167 /* Global relay viewer stream hash table. */
168 struct lttng_ht
*viewer_streams_ht
;
170 /* Global relay sessions hash table. */
171 struct lttng_ht
*sessions_ht
;
173 /* Relayd health monitoring */
174 struct health_app
*health_relayd
;
176 /* Global fd tracker. */
177 struct fd_tracker
*the_fd_tracker
;
179 static struct option long_options
[] = {
180 { "control-port", 1, 0, 'C', },
181 { "data-port", 1, 0, 'D', },
182 { "live-port", 1, 0, 'L', },
183 { "daemonize", 0, 0, 'd', },
184 { "background", 0, 0, 'b', },
185 { "group", 1, 0, 'g', },
186 { "fd-cap", 1, 0, '\0', },
187 { "help", 0, 0, 'h', },
188 { "output", 1, 0, 'o', },
189 { "verbose", 0, 0, 'v', },
190 { "config", 1, 0, 'f' },
191 { "version", 0, 0, 'V' },
195 static const char *config_ignore_options
[] = { "help", "config", "version" };
198 * Take an option from the getopt output and set it in the right variable to be
201 * Return 0 on success else a negative value.
203 static int set_option(int opt
, const char *arg
, const char *optname
)
209 if (!strcmp(optname
, "fd-cap")) {
213 v
= strtoul(arg
, NULL
, 0);
214 if (errno
!= 0 || !isdigit(arg
[0])) {
215 ERR("Wrong value in --fd-cap parameter: %s", arg
);
219 if (v
< DEFAULT_RELAYD_MINIMAL_FD_CAP
) {
220 ERR("File descriptor cap must be set to at least %d",
221 DEFAULT_RELAYD_MINIMAL_FD_CAP
);
224 ERR("File descriptor cap overflow in --fd-cap parameter: %s", arg
);
228 lttng_opt_fd_cap
= (unsigned int) v
;
229 DBG3("File descriptor cap set to %u", lttng_opt_fd_cap
);
232 fprintf(stderr
, "unknown option %s", optname
);
234 fprintf(stderr
, " with arg %s\n", arg
);
239 if (lttng_is_setuid_setgid()) {
240 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
241 "-C, --control-port");
243 ret
= uri_parse(arg
, &control_uri
);
245 ERR("Invalid control URI specified");
248 if (control_uri
->port
== 0) {
249 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
254 if (lttng_is_setuid_setgid()) {
255 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
258 ret
= uri_parse(arg
, &data_uri
);
260 ERR("Invalid data URI specified");
263 if (data_uri
->port
== 0) {
264 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
269 if (lttng_is_setuid_setgid()) {
270 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
273 ret
= uri_parse(arg
, &live_uri
);
275 ERR("Invalid live URI specified");
278 if (live_uri
->port
== 0) {
279 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
290 if (lttng_is_setuid_setgid()) {
291 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
294 tracing_group_name
= strdup(arg
);
295 if (tracing_group_name
== NULL
) {
300 tracing_group_name_override
= 1;
304 ret
= utils_show_help(8, "lttng-relayd", help_msg
);
306 ERR("Cannot show --help for `lttng-relayd`");
311 fprintf(stdout
, "%s\n", VERSION
);
314 if (lttng_is_setuid_setgid()) {
315 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
318 ret
= asprintf(&opt_output_path
, "%s", arg
);
321 PERROR("asprintf opt_output_path");
327 /* Verbose level can increase using multiple -v */
329 lttng_opt_verbose
= config_parse_value(arg
);
331 /* Only 3 level of verbosity (-vvv). */
332 if (lttng_opt_verbose
< 3) {
333 lttng_opt_verbose
+= 1;
338 /* Unknown option or other error.
339 * Error is printed by getopt, just return */
352 * config_entry_handler_cb used to handle options read from a config file.
353 * See config_entry_handler_cb comment in common/config/session-config.h for the
354 * return value conventions.
356 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
360 if (!entry
|| !entry
->name
|| !entry
->value
) {
365 /* Check if the option is to be ignored */
366 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
367 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
372 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
373 /* Ignore if entry name is not fully matched. */
374 if (strcmp(entry
->name
, long_options
[i
].name
)) {
379 * If the option takes no argument on the command line,
380 * we have to check if the value is "true". We support
381 * non-zero numeric values, true, on and yes.
383 if (!long_options
[i
].has_arg
) {
384 ret
= config_parse_value(entry
->value
);
387 WARN("Invalid configuration value \"%s\" for option %s",
388 entry
->value
, entry
->name
);
390 /* False, skip boolean config option. */
395 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
399 WARN("Unrecognized option \"%s\" in daemon configuration file.",
406 static int set_options(int argc
, char **argv
)
408 int c
, ret
= 0, option_index
= 0, retval
= 0;
409 int orig_optopt
= optopt
, orig_optind
= optind
;
410 char *default_address
, *optstring
;
411 const char *config_path
= NULL
;
413 optstring
= utils_generate_optstring(long_options
,
414 sizeof(long_options
) / sizeof(struct option
));
420 /* Check for the --config option */
422 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
423 &option_index
)) != -1) {
427 } else if (c
!= 'f') {
431 if (lttng_is_setuid_setgid()) {
432 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
435 config_path
= utils_expand_path(optarg
);
437 ERR("Failed to resolve path: %s", optarg
);
442 ret
= config_get_section_entries(config_path
, config_section_name
,
443 config_entry_handler
, NULL
);
446 ERR("Invalid configuration option at line %i", ret
);
452 /* Reset getopt's global state */
453 optopt
= orig_optopt
;
454 optind
= orig_optind
;
456 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
461 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
468 /* assign default values */
469 if (control_uri
== NULL
) {
470 ret
= asprintf(&default_address
,
471 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
472 DEFAULT_NETWORK_CONTROL_PORT
);
474 PERROR("asprintf default data address");
479 ret
= uri_parse(default_address
, &control_uri
);
480 free(default_address
);
482 ERR("Invalid control URI specified");
487 if (data_uri
== NULL
) {
488 ret
= asprintf(&default_address
,
489 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
490 DEFAULT_NETWORK_DATA_PORT
);
492 PERROR("asprintf default data address");
497 ret
= uri_parse(default_address
, &data_uri
);
498 free(default_address
);
500 ERR("Invalid data URI specified");
505 if (live_uri
== NULL
) {
506 ret
= asprintf(&default_address
,
507 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
508 DEFAULT_NETWORK_VIEWER_PORT
);
510 PERROR("asprintf default viewer control address");
515 ret
= uri_parse(default_address
, &live_uri
);
516 free(default_address
);
518 ERR("Invalid viewer control URI specified");
523 if (lttng_opt_fd_cap
== 0) {
525 struct rlimit rlimit
;
527 ret
= getrlimit(RLIMIT_NOFILE
, &rlimit
);
529 PERROR("Failed to get file descriptor limit");
533 lttng_opt_fd_cap
= rlimit
.rlim_cur
;
541 static void print_global_objects(void)
543 print_viewer_streams();
544 print_relay_streams();
551 static void relayd_cleanup(void)
553 print_global_objects();
557 if (viewer_streams_ht
)
558 lttng_ht_destroy(viewer_streams_ht
);
559 if (relay_streams_ht
)
560 lttng_ht_destroy(relay_streams_ht
);
562 lttng_ht_destroy(sessions_ht
);
564 /* free the dynamically allocated opt_output_path */
565 free(opt_output_path
);
567 /* Close thread quit pipes */
568 (void) fd_tracker_util_pipe_close(the_fd_tracker
, thread_quit_pipe
);
570 uri_free(control_uri
);
572 /* Live URI is freed in the live thread. */
574 if (tracing_group_name_override
) {
575 free((void *) tracing_group_name
);
580 * Write to writable pipe used to notify a thread.
582 static int notify_thread_pipe(int wpipe
)
586 ret
= lttng_write(wpipe
, "!", 1);
588 PERROR("write poll pipe");
596 static int notify_health_quit_pipe(int *pipe
)
600 ret
= lttng_write(pipe
[1], "4", 1);
602 PERROR("write relay health quit");
611 * Stop all relayd and relayd-live threads.
613 int lttng_relay_stop_threads(void)
617 /* Stopping all threads */
618 DBG("Terminating all threads");
619 if (notify_thread_pipe(thread_quit_pipe
[1])) {
620 ERR("write error on thread quit pipe");
624 if (notify_health_quit_pipe(health_quit_pipe
)) {
625 ERR("write error on health quit pipe");
628 /* Dispatch thread */
629 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
630 futex_nto1_wake(&relay_conn_queue
.futex
);
632 if (relayd_live_stop()) {
633 ERR("Error stopping live threads");
640 * Signal handler for the daemon
642 * Simply stop all worker threads, leaving main() return gracefully after
643 * joining all threads and calling cleanup().
645 static void sighandler(int sig
)
649 DBG("SIGINT caught");
650 if (lttng_relay_stop_threads()) {
651 ERR("Error stopping threads");
655 DBG("SIGTERM caught");
656 if (lttng_relay_stop_threads()) {
657 ERR("Error stopping threads");
661 CMM_STORE_SHARED(recv_child_signal
, 1);
669 * Setup signal handler for :
670 * SIGINT, SIGTERM, SIGPIPE
672 static int set_signal_handler(void)
678 if ((ret
= sigemptyset(&sigset
)) < 0) {
679 PERROR("sigemptyset");
686 sa
.sa_handler
= sighandler
;
687 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
692 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
697 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
702 sa
.sa_handler
= SIG_IGN
;
703 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
708 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
713 void lttng_relay_notify_ready(void)
715 /* Notify the parent of the fork() process that we are ready. */
716 if (opt_daemon
|| opt_background
) {
717 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
718 kill(child_ppid
, SIGUSR1
);
724 * Init thread quit pipe.
726 * Return -1 on error or 0 if all pipes are created.
728 static int init_thread_quit_pipe(void)
730 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
731 "Quit pipe", thread_quit_pipe
);
735 * Init health quit pipe.
737 * Return -1 on error or 0 if all pipes are created.
739 static int init_health_quit_pipe(void)
741 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
742 "Health quit pipe", health_quit_pipe
);
746 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
748 static int create_named_thread_poll_set(struct lttng_poll_event
*events
,
749 int size
, const char *name
)
753 if (events
== NULL
|| size
== 0) {
758 ret
= fd_tracker_util_poll_create(the_fd_tracker
,
759 name
, events
, 1, LTTNG_CLOEXEC
);
762 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
774 * Check if the thread quit pipe was triggered.
776 * Return 1 if it was triggered else 0;
778 static int check_thread_quit_pipe(int fd
, uint32_t events
)
780 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
787 static int create_sock(void *data
, int *out_fd
)
790 struct lttcomm_sock
*sock
= data
;
792 ret
= lttcomm_create_sock(sock
);
802 static int close_sock(void *data
, int *in_fd
)
804 struct lttcomm_sock
*sock
= data
;
806 return sock
->ops
->close(sock
);
809 static int accept_sock(void *data
, int *out_fd
)
812 /* Socks is an array of in_sock, out_sock. */
813 struct lttcomm_sock
**socks
= data
;
814 struct lttcomm_sock
*in_sock
= socks
[0];
816 socks
[1] = in_sock
->ops
->accept(in_sock
);
821 *out_fd
= socks
[1]->fd
;
827 * Create and init socket from uri.
829 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
,
833 struct lttcomm_sock
*sock
= NULL
;
834 char uri_str
[PATH_MAX
];
835 char *formated_name
= NULL
;
837 sock
= lttcomm_alloc_sock_from_uri(uri
);
839 ERR("Allocating socket");
844 * Don't fail to create the socket if the name can't be built as it is
845 * only used for debugging purposes.
847 ret
= uri_to_str_url(uri
, uri_str
, sizeof(uri_str
));
848 uri_str
[sizeof(uri_str
) - 1] = '\0';
850 ret
= asprintf(&formated_name
, "%s socket @ %s", name
,
853 formated_name
= NULL
;
857 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &sock_fd
,
858 (const char **) (formated_name
? &formated_name
: NULL
),
859 1, create_sock
, sock
);
861 DBG("Listening on %s socket %d", name
, sock
->fd
);
863 ret
= sock
->ops
->bind(sock
);
865 PERROR("Failed to bind socket");
869 ret
= sock
->ops
->listen(sock
, -1);
879 lttcomm_destroy_sock(sock
);
885 struct lttcomm_sock
*accept_relayd_sock(struct lttcomm_sock
*listening_sock
,
889 struct lttcomm_sock
*socks
[2] = { listening_sock
, NULL
};
890 struct lttcomm_sock
*new_sock
= NULL
;
892 ret
= fd_tracker_open_unsuspendable_fd(
893 the_fd_tracker
, &out_fd
,
894 (const char **) &name
,
895 1, accept_sock
, &socks
);
900 DBG("%s accepted, socket %d", name
, new_sock
->fd
);
906 * This thread manages the listening for new connections on the network
908 static void *relay_thread_listener(void *data
)
910 int i
, ret
, pollfd
, err
= -1;
911 uint32_t revents
, nb_fd
;
912 struct lttng_poll_event events
;
913 struct lttcomm_sock
*control_sock
, *data_sock
;
915 DBG("[thread] Relay listener started");
917 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
919 health_code_update();
921 control_sock
= relay_socket_create(control_uri
, "Control listener");
923 goto error_sock_control
;
926 data_sock
= relay_socket_create(data_uri
, "Data listener");
928 goto error_sock_relay
;
932 * Pass 3 as size here for the thread quit pipe, control and
935 ret
= create_named_thread_poll_set(&events
, 3, "Listener thread epoll");
937 goto error_create_poll
;
940 /* Add the control socket */
941 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
946 /* Add the data socket */
947 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
952 lttng_relay_notify_ready();
954 if (testpoint(relayd_thread_listener
)) {
955 goto error_testpoint
;
959 health_code_update();
961 DBG("Listener accepting connections");
965 ret
= lttng_poll_wait(&events
, -1);
969 * Restart interrupted system call.
971 if (errno
== EINTR
) {
979 DBG("Relay new connection received");
980 for (i
= 0; i
< nb_fd
; i
++) {
981 health_code_update();
983 /* Fetch once the poll data */
984 revents
= LTTNG_POLL_GETEV(&events
, i
);
985 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
989 * No activity for this FD (poll
995 /* Thread quit pipe has been closed. Killing thread. */
996 ret
= check_thread_quit_pipe(pollfd
, revents
);
1002 if (revents
& LPOLLIN
) {
1004 * A new connection is requested, therefore a
1005 * sessiond/consumerd connection is allocated in
1006 * this thread, enqueued to a global queue and
1007 * dequeued (and freed) in the worker thread.
1010 struct relay_connection
*new_conn
;
1011 struct lttcomm_sock
*newsock
= NULL
;
1012 enum connection_type type
;
1014 if (pollfd
== data_sock
->fd
) {
1016 newsock
= accept_relayd_sock(data_sock
,
1017 "Data socket to relayd");
1019 assert(pollfd
== control_sock
->fd
);
1020 type
= RELAY_CONTROL
;
1021 newsock
= accept_relayd_sock(control_sock
,
1022 "Control socket to relayd");
1025 PERROR("accepting sock");
1029 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
1032 PERROR("setsockopt inet");
1033 lttcomm_destroy_sock(newsock
);
1037 ret
= socket_apply_keep_alive_config(newsock
->fd
);
1039 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1041 lttcomm_destroy_sock(newsock
);
1045 new_conn
= connection_create(newsock
, type
);
1047 lttcomm_destroy_sock(newsock
);
1051 /* Enqueue request for the dispatcher thread. */
1052 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
1056 * Wake the dispatch queue futex.
1057 * Implicit memory barrier with the
1058 * exchange in cds_wfcq_enqueue.
1060 futex_nto1_wake(&relay_conn_queue
.futex
);
1061 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1062 ERR("socket poll error");
1065 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1075 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
1077 if (data_sock
->fd
>= 0) {
1078 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1079 &data_sock
->fd
, 1, close_sock
,
1085 lttcomm_destroy_sock(data_sock
);
1087 if (control_sock
->fd
>= 0) {
1088 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1089 &control_sock
->fd
, 1, close_sock
,
1095 lttcomm_destroy_sock(control_sock
);
1099 ERR("Health error occurred in %s", __func__
);
1101 health_unregister(health_relayd
);
1102 DBG("Relay listener thread cleanup complete");
1103 lttng_relay_stop_threads();
1108 * This thread manages the dispatching of the requests to worker threads
1110 static void *relay_thread_dispatcher(void *data
)
1114 struct cds_wfcq_node
*node
;
1115 struct relay_connection
*new_conn
= NULL
;
1117 DBG("[thread] Relay dispatcher started");
1119 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
1121 if (testpoint(relayd_thread_dispatcher
)) {
1122 goto error_testpoint
;
1125 health_code_update();
1128 health_code_update();
1130 /* Atomically prepare the queue futex */
1131 futex_nto1_prepare(&relay_conn_queue
.futex
);
1133 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1138 health_code_update();
1140 /* Dequeue commands */
1141 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1142 &relay_conn_queue
.tail
);
1144 DBG("Woken up but nothing in the relay command queue");
1145 /* Continue thread execution */
1148 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1150 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1153 * Inform worker thread of the new request. This
1154 * call is blocking so we can be assured that
1155 * the data will be read at some point in time
1156 * or wait to the end of the world :)
1158 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1160 PERROR("write connection pipe");
1161 connection_put(new_conn
);
1164 } while (node
!= NULL
);
1166 /* Futex wait on queue. Blocking call on futex() */
1167 health_poll_entry();
1168 futex_nto1_wait(&relay_conn_queue
.futex
);
1172 /* Normal exit, no error */
1179 ERR("Health error occurred in %s", __func__
);
1181 health_unregister(health_relayd
);
1182 DBG("Dispatch thread dying");
1183 lttng_relay_stop_threads();
1188 * Set index data from the control port to a given index object.
1190 static int set_index_control_data(struct relay_index
*index
,
1191 struct lttcomm_relayd_index
*data
,
1192 struct relay_connection
*conn
)
1194 struct ctf_packet_index index_data
;
1197 * The index on disk is encoded in big endian.
1199 index_data
.packet_size
= htobe64(data
->packet_size
);
1200 index_data
.content_size
= htobe64(data
->content_size
);
1201 index_data
.timestamp_begin
= htobe64(data
->timestamp_begin
);
1202 index_data
.timestamp_end
= htobe64(data
->timestamp_end
);
1203 index_data
.events_discarded
= htobe64(data
->events_discarded
);
1204 index_data
.stream_id
= htobe64(data
->stream_id
);
1206 if (conn
->minor
>= 8) {
1207 index
->index_data
.stream_instance_id
= htobe64(data
->stream_instance_id
);
1208 index
->index_data
.packet_seq_num
= htobe64(data
->packet_seq_num
);
1211 return relay_index_set_data(index
, &index_data
);
1215 * Handle the RELAYD_CREATE_SESSION command.
1217 * On success, send back the session id or else return a negative value.
1219 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1220 struct relay_connection
*conn
,
1221 const struct lttng_buffer_view
*payload
)
1225 struct relay_session
*session
;
1226 struct lttcomm_relayd_status_session reply
;
1227 char session_name
[LTTNG_NAME_MAX
];
1228 char hostname
[LTTNG_HOST_NAME_MAX
];
1229 uint32_t live_timer
= 0;
1230 bool snapshot
= false;
1232 memset(session_name
, 0, LTTNG_NAME_MAX
);
1233 memset(hostname
, 0, LTTNG_HOST_NAME_MAX
);
1235 memset(&reply
, 0, sizeof(reply
));
1237 if (conn
->minor
< 4) {
1238 /* From 2.1 to 2.3 */
1240 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1241 /* From 2.4 to 2.10 */
1242 ret
= cmd_create_session_2_4(payload
, session_name
,
1243 hostname
, &live_timer
, &snapshot
);
1245 /* From 2.11 to ... */
1246 ret
= cmd_create_session_2_11(payload
, session_name
,
1247 hostname
, &live_timer
, &snapshot
);
1254 session
= session_create(session_name
, hostname
, live_timer
,
1255 snapshot
, conn
->major
, conn
->minor
);
1260 assert(!conn
->session
);
1261 conn
->session
= session
;
1262 DBG("Created session %" PRIu64
, session
->id
);
1264 reply
.session_id
= htobe64(session
->id
);
1268 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1270 reply
.ret_code
= htobe32(LTTNG_OK
);
1273 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1274 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1275 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1284 * When we have received all the streams and the metadata for a channel,
1285 * we make them visible to the viewer threads.
1287 static void publish_connection_local_streams(struct relay_connection
*conn
)
1289 struct relay_stream
*stream
;
1290 struct relay_session
*session
= conn
->session
;
1293 * We publish all streams belonging to a session atomically wrt
1296 pthread_mutex_lock(&session
->lock
);
1298 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1300 stream_publish(stream
);
1305 * Inform the viewer that there are new streams in the session.
1307 if (session
->viewer_attached
) {
1308 uatomic_set(&session
->new_streams
, 1);
1310 pthread_mutex_unlock(&session
->lock
);
1314 * relay_add_stream: allocate a new stream for a session
1316 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1317 struct relay_connection
*conn
,
1318 const struct lttng_buffer_view
*payload
)
1322 struct relay_session
*session
= conn
->session
;
1323 struct relay_stream
*stream
= NULL
;
1324 struct lttcomm_relayd_status_stream reply
;
1325 struct ctf_trace
*trace
= NULL
;
1326 uint64_t stream_handle
= -1ULL;
1327 char *path_name
= NULL
, *channel_name
= NULL
;
1328 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1329 struct relay_stream_chunk_id stream_chunk_id
= { 0 };
1331 if (!session
|| !conn
->version_check_done
) {
1332 ERR("Trying to add a stream before version check");
1334 goto end_no_session
;
1337 if (session
->minor
== 1) {
1339 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1341 } else if (session
->minor
> 1 && session
->minor
< 11) {
1342 /* From 2.2 to 2.10 */
1343 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1344 &channel_name
, &tracefile_size
, &tracefile_count
);
1346 /* From 2.11 to ... */
1347 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1348 &channel_name
, &tracefile_size
, &tracefile_count
,
1349 &stream_chunk_id
.value
);
1350 stream_chunk_id
.is_set
= true;
1357 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1361 /* This stream here has one reference on the trace. */
1363 pthread_mutex_lock(&last_relay_stream_id_lock
);
1364 stream_handle
= ++last_relay_stream_id
;
1365 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1367 /* We pass ownership of path_name and channel_name. */
1368 stream
= stream_create(trace
, stream_handle
, path_name
,
1369 channel_name
, tracefile_size
, tracefile_count
,
1372 channel_name
= NULL
;
1375 * Streams are the owners of their trace. Reference to trace is
1376 * kept within stream_create().
1378 ctf_trace_put(trace
);
1381 memset(&reply
, 0, sizeof(reply
));
1382 reply
.handle
= htobe64(stream_handle
);
1384 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1386 reply
.ret_code
= htobe32(LTTNG_OK
);
1389 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1390 sizeof(struct lttcomm_relayd_status_stream
), 0);
1391 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1392 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1404 * relay_close_stream: close a specific stream
1406 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1407 struct relay_connection
*conn
,
1408 const struct lttng_buffer_view
*payload
)
1412 struct relay_session
*session
= conn
->session
;
1413 struct lttcomm_relayd_close_stream stream_info
;
1414 struct lttcomm_relayd_generic_reply reply
;
1415 struct relay_stream
*stream
;
1417 DBG("Close stream received");
1419 if (!session
|| !conn
->version_check_done
) {
1420 ERR("Trying to close a stream before version check");
1422 goto end_no_session
;
1425 if (payload
->size
< sizeof(stream_info
)) {
1426 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1427 sizeof(stream_info
), payload
->size
);
1429 goto end_no_session
;
1431 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1432 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1433 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1435 stream
= stream_get_by_id(stream_info
.stream_id
);
1442 * Set last_net_seq_num before the close flag. Required by data
1445 pthread_mutex_lock(&stream
->lock
);
1446 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1447 pthread_mutex_unlock(&stream
->lock
);
1450 * This is one of the conditions which may trigger a stream close
1451 * with the others being:
1452 * 1) A close command is received for a stream
1453 * 2) The control connection owning the stream is closed
1454 * 3) We have received all of the stream's data _after_ a close
1457 try_stream_close(stream
);
1458 if (stream
->is_metadata
) {
1459 struct relay_viewer_stream
*vstream
;
1461 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1463 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1465 * Since all the metadata has been sent to the
1466 * viewer and that we have a request to close
1467 * its stream, we can safely teardown the
1468 * corresponding metadata viewer stream.
1470 viewer_stream_put(vstream
);
1472 /* Put local reference. */
1473 viewer_stream_put(vstream
);
1480 memset(&reply
, 0, sizeof(reply
));
1482 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1484 reply
.ret_code
= htobe32(LTTNG_OK
);
1486 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1487 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1488 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1489 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1499 * relay_reset_metadata: reset a metadata stream
1502 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1503 struct relay_connection
*conn
,
1504 const struct lttng_buffer_view
*payload
)
1508 struct relay_session
*session
= conn
->session
;
1509 struct lttcomm_relayd_reset_metadata stream_info
;
1510 struct lttcomm_relayd_generic_reply reply
;
1511 struct relay_stream
*stream
;
1513 DBG("Reset metadata received");
1515 if (!session
|| !conn
->version_check_done
) {
1516 ERR("Trying to reset a metadata stream before version check");
1518 goto end_no_session
;
1521 if (payload
->size
< sizeof(stream_info
)) {
1522 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1523 sizeof(stream_info
), payload
->size
);
1525 goto end_no_session
;
1527 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1528 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1529 stream_info
.version
= be64toh(stream_info
.version
);
1531 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1533 /* Unsupported for live sessions for now. */
1534 if (session
->live_timer
!= 0) {
1539 stream
= stream_get_by_id(stream_info
.stream_id
);
1544 pthread_mutex_lock(&stream
->lock
);
1545 if (!stream
->is_metadata
) {
1550 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
1551 0, 0, -1, -1, stream
->stream_fd
->fd
, NULL
,
1552 &stream
->stream_fd
->fd
);
1554 ERR("Failed to rotate metadata file %s of channel %s",
1555 stream
->path_name
, stream
->channel_name
);
1560 pthread_mutex_unlock(&stream
->lock
);
1564 memset(&reply
, 0, sizeof(reply
));
1566 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1568 reply
.ret_code
= htobe32(LTTNG_OK
);
1570 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1571 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1572 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1573 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1583 * relay_unknown_command: send -1 if received unknown command
1585 static void relay_unknown_command(struct relay_connection
*conn
)
1587 struct lttcomm_relayd_generic_reply reply
;
1590 memset(&reply
, 0, sizeof(reply
));
1591 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1592 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1593 if (send_ret
< sizeof(reply
)) {
1594 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1599 * relay_start: send an acknowledgment to the client to tell if we are
1600 * ready to receive data. We are ready if a session is established.
1602 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1603 struct relay_connection
*conn
,
1604 const struct lttng_buffer_view
*payload
)
1608 struct lttcomm_relayd_generic_reply reply
;
1609 struct relay_session
*session
= conn
->session
;
1612 DBG("Trying to start the streaming without a session established");
1613 ret
= htobe32(LTTNG_ERR_UNK
);
1616 memset(&reply
, 0, sizeof(reply
));
1617 reply
.ret_code
= htobe32(LTTNG_OK
);
1618 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1620 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1621 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1630 * Append padding to the file pointed by the file descriptor fd.
1632 static int write_padding_to_file(int fd
, uint32_t size
)
1641 zeros
= zmalloc(size
);
1642 if (zeros
== NULL
) {
1643 PERROR("zmalloc zeros for padding");
1648 ret
= lttng_write(fd
, zeros
, size
);
1650 PERROR("write padding to file");
1660 * Close the current index file if it is open, and create a new one.
1662 * Return 0 on success, -1 on error.
1665 int create_rotate_index_file(struct relay_stream
*stream
)
1668 uint32_t major
, minor
;
1670 /* Put ref on previous index_file. */
1671 if (stream
->index_file
) {
1672 lttng_index_file_put(stream
->index_file
);
1673 stream
->index_file
= NULL
;
1675 major
= stream
->trace
->session
->major
;
1676 minor
= stream
->trace
->session
->minor
;
1677 stream
->index_file
= lttng_index_file_create(stream
->path_name
,
1678 stream
->channel_name
,
1679 -1, -1, stream
->tracefile_size
,
1680 tracefile_array_get_file_index_head(stream
->tfa
),
1681 lttng_to_index_major(major
, minor
),
1682 lttng_to_index_minor(major
, minor
));
1683 if (!stream
->index_file
) {
1695 int do_rotate_stream(struct relay_stream
*stream
)
1699 /* Perform the stream rotation. */
1700 ret
= utils_rotate_stream_file(stream
->path_name
,
1701 stream
->channel_name
, stream
->tracefile_size
,
1702 stream
->tracefile_count
, -1,
1703 -1, stream
->stream_fd
->fd
,
1704 NULL
, &stream
->stream_fd
->fd
);
1706 ERR("Rotating stream output file");
1709 stream
->tracefile_size_current
= 0;
1711 /* Rotate also the index if the stream is not a metadata stream. */
1712 if (!stream
->is_metadata
) {
1713 ret
= create_rotate_index_file(stream
);
1715 ERR("Failed to rotate index file");
1720 stream
->rotate_at_seq_num
= -1ULL;
1721 stream
->pos_after_last_complete_data_index
= 0;
1728 * If too much data has been written in a tracefile before we received the
1729 * rotation command, we have to move the excess data to the new tracefile and
1730 * perform the rotation. This can happen because the control and data
1731 * connections are separate, the indexes as well as the commands arrive from
1732 * the control connection and we have no control over the order so we could be
1733 * in a situation where too much data has been received on the data connection
1734 * before the rotation command on the control connection arrives. We don't need
1735 * to update the index because its order is guaranteed with the rotation
1739 int rotate_truncate_stream(struct relay_stream
*stream
)
1743 uint64_t diff
, pos
= 0;
1744 char buf
[FILE_COPY_BUFFER_SIZE
];
1746 assert(!stream
->is_metadata
);
1748 assert(stream
->tracefile_size_current
>
1749 stream
->pos_after_last_complete_data_index
);
1750 diff
= stream
->tracefile_size_current
-
1751 stream
->pos_after_last_complete_data_index
;
1753 /* Create the new tracefile. */
1754 new_fd
= utils_create_stream_file(stream
->path_name
,
1755 stream
->channel_name
,
1756 stream
->tracefile_size
, stream
->tracefile_count
,
1757 /* uid */ -1, /* gid */ -1, /* suffix */ NULL
);
1759 ERR("Failed to create new stream file at path %s for channel %s",
1760 stream
->path_name
, stream
->channel_name
);
1766 * Rewind the current tracefile to the position at which the rotation
1767 * should have occured.
1769 lseek_ret
= lseek(stream
->stream_fd
->fd
,
1770 stream
->pos_after_last_complete_data_index
, SEEK_SET
);
1771 if (lseek_ret
< 0) {
1772 PERROR("seek truncate stream");
1777 /* Move data from the old file to the new file. */
1778 while (pos
< diff
) {
1779 uint64_t count
, bytes_left
;
1782 bytes_left
= diff
- pos
;
1783 count
= bytes_left
> sizeof(buf
) ? sizeof(buf
) : bytes_left
;
1784 assert(count
<= SIZE_MAX
);
1786 io_ret
= lttng_read(stream
->stream_fd
->fd
, buf
, count
);
1787 if (io_ret
< (ssize_t
) count
) {
1788 char error_string
[256];
1790 snprintf(error_string
, sizeof(error_string
),
1791 "Failed to read %" PRIu64
" bytes from fd %i in rotate_truncate_stream(), returned %zi",
1792 count
, stream
->stream_fd
->fd
, io_ret
);
1794 PERROR("%s", error_string
);
1796 ERR("%s", error_string
);
1802 io_ret
= lttng_write(new_fd
, buf
, count
);
1803 if (io_ret
< (ssize_t
) count
) {
1804 char error_string
[256];
1806 snprintf(error_string
, sizeof(error_string
),
1807 "Failed to write %" PRIu64
" bytes from fd %i in rotate_truncate_stream(), returned %zi",
1808 count
, new_fd
, io_ret
);
1810 PERROR("%s", error_string
);
1812 ERR("%s", error_string
);
1821 /* Truncate the file to get rid of the excess data. */
1822 ret
= ftruncate(stream
->stream_fd
->fd
,
1823 stream
->pos_after_last_complete_data_index
);
1825 PERROR("ftruncate");
1829 ret
= close(stream
->stream_fd
->fd
);
1831 PERROR("Closing tracefile");
1835 ret
= create_rotate_index_file(stream
);
1837 ERR("Rotate stream index file");
1842 * Update the offset and FD of all the eventual indexes created by the
1843 * data connection before the rotation command arrived.
1845 ret
= relay_index_switch_all_files(stream
);
1847 ERR("Failed to rotate index file");
1851 stream
->stream_fd
->fd
= new_fd
;
1852 stream
->tracefile_size_current
= diff
;
1853 stream
->pos_after_last_complete_data_index
= 0;
1854 stream
->rotate_at_seq_num
= -1ULL;
1863 * Check if a stream should perform a rotation (for session rotation).
1864 * Must be called with the stream lock held.
1866 * Return 0 on success, a negative value on error.
1869 int try_rotate_stream(struct relay_stream
*stream
)
1873 /* No rotation expected. */
1874 if (stream
->rotate_at_seq_num
== -1ULL) {
1878 if (stream
->prev_seq
< stream
->rotate_at_seq_num
||
1879 stream
->prev_seq
== -1ULL) {
1880 DBG("Stream %" PRIu64
" no yet ready for rotation",
1881 stream
->stream_handle
);
1883 } else if (stream
->prev_seq
> stream
->rotate_at_seq_num
) {
1884 DBG("Rotation after too much data has been written in tracefile "
1885 "for stream %" PRIu64
", need to truncate before "
1886 "rotating", stream
->stream_handle
);
1887 ret
= rotate_truncate_stream(stream
);
1889 ERR("Failed to truncate stream");
1893 /* stream->prev_seq == stream->rotate_at_seq_num */
1894 DBG("Stream %" PRIu64
" ready for rotation",
1895 stream
->stream_handle
);
1896 ret
= do_rotate_stream(stream
);
1904 * relay_recv_metadata: receive the metadata for the session.
1906 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1907 struct relay_connection
*conn
,
1908 const struct lttng_buffer_view
*payload
)
1912 struct relay_session
*session
= conn
->session
;
1913 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1914 struct relay_stream
*metadata_stream
;
1915 uint64_t metadata_payload_size
;
1918 ERR("Metadata sent before version check");
1923 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1924 ERR("Incorrect data size");
1928 metadata_payload_size
= recv_hdr
->data_size
-
1929 sizeof(struct lttcomm_relayd_metadata_payload
);
1931 memcpy(&metadata_payload_header
, payload
->data
,
1932 sizeof(metadata_payload_header
));
1933 metadata_payload_header
.stream_id
= be64toh(
1934 metadata_payload_header
.stream_id
);
1935 metadata_payload_header
.padding_size
= be32toh(
1936 metadata_payload_header
.padding_size
);
1938 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1939 if (!metadata_stream
) {
1944 pthread_mutex_lock(&metadata_stream
->lock
);
1946 size_ret
= lttng_write(metadata_stream
->stream_fd
->fd
,
1947 payload
->data
+ sizeof(metadata_payload_header
),
1948 metadata_payload_size
);
1949 if (size_ret
< metadata_payload_size
) {
1950 ERR("Relay error writing metadata on file");
1955 size_ret
= write_padding_to_file(metadata_stream
->stream_fd
->fd
,
1956 metadata_payload_header
.padding_size
);
1957 if (size_ret
< (int64_t) metadata_payload_header
.padding_size
) {
1962 metadata_stream
->metadata_received
+=
1963 metadata_payload_size
+ metadata_payload_header
.padding_size
;
1964 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1965 metadata_stream
->metadata_received
);
1967 ret
= try_rotate_stream(metadata_stream
);
1973 pthread_mutex_unlock(&metadata_stream
->lock
);
1974 stream_put(metadata_stream
);
1980 * relay_send_version: send relayd version number
1982 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1983 struct relay_connection
*conn
,
1984 const struct lttng_buffer_view
*payload
)
1988 struct lttcomm_relayd_version reply
, msg
;
1989 bool compatible
= true;
1991 conn
->version_check_done
= true;
1993 /* Get version from the other side. */
1994 if (payload
->size
< sizeof(msg
)) {
1995 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1996 sizeof(msg
), payload
->size
);
2001 memcpy(&msg
, payload
->data
, sizeof(msg
));
2002 msg
.major
= be32toh(msg
.major
);
2003 msg
.minor
= be32toh(msg
.minor
);
2005 memset(&reply
, 0, sizeof(reply
));
2006 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
2007 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
2009 /* Major versions must be the same */
2010 if (reply
.major
!= msg
.major
) {
2011 DBG("Incompatible major versions (%u vs %u), deleting session",
2012 reply
.major
, msg
.major
);
2016 conn
->major
= reply
.major
;
2017 /* We adapt to the lowest compatible version */
2018 if (reply
.minor
<= msg
.minor
) {
2019 conn
->minor
= reply
.minor
;
2021 conn
->minor
= msg
.minor
;
2024 reply
.major
= htobe32(reply
.major
);
2025 reply
.minor
= htobe32(reply
.minor
);
2026 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2028 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2029 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2042 DBG("Version check done using protocol %u.%u", conn
->major
,
2050 * Check for data pending for a given stream id from the session daemon.
2052 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2053 struct relay_connection
*conn
,
2054 const struct lttng_buffer_view
*payload
)
2056 struct relay_session
*session
= conn
->session
;
2057 struct lttcomm_relayd_data_pending msg
;
2058 struct lttcomm_relayd_generic_reply reply
;
2059 struct relay_stream
*stream
;
2063 DBG("Data pending command received");
2065 if (!session
|| !conn
->version_check_done
) {
2066 ERR("Trying to check for data before version check");
2068 goto end_no_session
;
2071 if (payload
->size
< sizeof(msg
)) {
2072 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2073 sizeof(msg
), payload
->size
);
2075 goto end_no_session
;
2077 memcpy(&msg
, payload
->data
, sizeof(msg
));
2078 msg
.stream_id
= be64toh(msg
.stream_id
);
2079 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
2081 stream
= stream_get_by_id(msg
.stream_id
);
2082 if (stream
== NULL
) {
2087 pthread_mutex_lock(&stream
->lock
);
2089 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
2090 " and last_seq %" PRIu64
, msg
.stream_id
,
2091 stream
->prev_seq
, msg
.last_net_seq_num
);
2093 /* Avoid wrapping issue */
2094 if (((int64_t) (stream
->prev_seq
- msg
.last_net_seq_num
)) >= 0) {
2095 /* Data has in fact been written and is NOT pending */
2098 /* Data still being streamed thus pending */
2102 stream
->data_pending_check_done
= true;
2103 pthread_mutex_unlock(&stream
->lock
);
2108 memset(&reply
, 0, sizeof(reply
));
2109 reply
.ret_code
= htobe32(ret
);
2110 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2111 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2112 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2122 * Wait for the control socket to reach a quiescent state.
2124 * Note that for now, when receiving this command from the session
2125 * daemon, this means that every subsequent commands or data received on
2126 * the control socket has been handled. So, this is why we simply return
2129 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
2130 struct relay_connection
*conn
,
2131 const struct lttng_buffer_view
*payload
)
2135 struct relay_stream
*stream
;
2136 struct lttcomm_relayd_quiescent_control msg
;
2137 struct lttcomm_relayd_generic_reply reply
;
2139 DBG("Checking quiescent state on control socket");
2141 if (!conn
->session
|| !conn
->version_check_done
) {
2142 ERR("Trying to check for data before version check");
2144 goto end_no_session
;
2147 if (payload
->size
< sizeof(msg
)) {
2148 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2149 sizeof(msg
), payload
->size
);
2151 goto end_no_session
;
2153 memcpy(&msg
, payload
->data
, sizeof(msg
));
2154 msg
.stream_id
= be64toh(msg
.stream_id
);
2156 stream
= stream_get_by_id(msg
.stream_id
);
2160 pthread_mutex_lock(&stream
->lock
);
2161 stream
->data_pending_check_done
= true;
2162 pthread_mutex_unlock(&stream
->lock
);
2164 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
2167 memset(&reply
, 0, sizeof(reply
));
2168 reply
.ret_code
= htobe32(LTTNG_OK
);
2169 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2170 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2171 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2183 * Initialize a data pending command. This means that a consumer is about
2184 * to ask for data pending for each stream it holds. Simply iterate over
2185 * all streams of a session and set the data_pending_check_done flag.
2187 * This command returns to the client a LTTNG_OK code.
2189 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2190 struct relay_connection
*conn
,
2191 const struct lttng_buffer_view
*payload
)
2195 struct lttng_ht_iter iter
;
2196 struct lttcomm_relayd_begin_data_pending msg
;
2197 struct lttcomm_relayd_generic_reply reply
;
2198 struct relay_stream
*stream
;
2203 DBG("Init streams for data pending");
2205 if (!conn
->session
|| !conn
->version_check_done
) {
2206 ERR("Trying to check for data before version check");
2208 goto end_no_session
;
2211 if (payload
->size
< sizeof(msg
)) {
2212 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2213 sizeof(msg
), payload
->size
);
2215 goto end_no_session
;
2217 memcpy(&msg
, payload
->data
, sizeof(msg
));
2218 msg
.session_id
= be64toh(msg
.session_id
);
2221 * Iterate over all streams to set the begin data pending flag.
2222 * For now, the streams are indexed by stream handle so we have
2223 * to iterate over all streams to find the one associated with
2224 * the right session_id.
2227 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2229 if (!stream_get(stream
)) {
2232 if (stream
->trace
->session
->id
== msg
.session_id
) {
2233 pthread_mutex_lock(&stream
->lock
);
2234 stream
->data_pending_check_done
= false;
2235 pthread_mutex_unlock(&stream
->lock
);
2236 DBG("Set begin data pending flag to stream %" PRIu64
,
2237 stream
->stream_handle
);
2243 memset(&reply
, 0, sizeof(reply
));
2244 /* All good, send back reply. */
2245 reply
.ret_code
= htobe32(LTTNG_OK
);
2247 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2248 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2249 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2261 * End data pending command. This will check, for a given session id, if
2262 * each stream associated with it has its data_pending_check_done flag
2263 * set. If not, this means that the client lost track of the stream but
2264 * the data is still being streamed on our side. In this case, we inform
2265 * the client that data is in flight.
2267 * Return to the client if there is data in flight or not with a ret_code.
2269 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2270 struct relay_connection
*conn
,
2271 const struct lttng_buffer_view
*payload
)
2275 struct lttng_ht_iter iter
;
2276 struct lttcomm_relayd_end_data_pending msg
;
2277 struct lttcomm_relayd_generic_reply reply
;
2278 struct relay_stream
*stream
;
2279 uint32_t is_data_inflight
= 0;
2281 DBG("End data pending command");
2283 if (!conn
->session
|| !conn
->version_check_done
) {
2284 ERR("Trying to check for data before version check");
2286 goto end_no_session
;
2289 if (payload
->size
< sizeof(msg
)) {
2290 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2291 sizeof(msg
), payload
->size
);
2293 goto end_no_session
;
2295 memcpy(&msg
, payload
->data
, sizeof(msg
));
2296 msg
.session_id
= be64toh(msg
.session_id
);
2299 * Iterate over all streams to see if the begin data pending
2303 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2305 if (!stream_get(stream
)) {
2308 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2312 pthread_mutex_lock(&stream
->lock
);
2313 if (!stream
->data_pending_check_done
) {
2314 if (!stream
->closed
|| !(((int64_t) (stream
->prev_seq
- stream
->last_net_seq_num
)) >= 0)) {
2315 is_data_inflight
= 1;
2316 DBG("Data is still in flight for stream %" PRIu64
,
2317 stream
->stream_handle
);
2318 pthread_mutex_unlock(&stream
->lock
);
2323 pthread_mutex_unlock(&stream
->lock
);
2328 memset(&reply
, 0, sizeof(reply
));
2329 /* All good, send back reply. */
2330 reply
.ret_code
= htobe32(is_data_inflight
);
2332 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2333 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2334 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2346 * Receive an index for a specific stream.
2348 * Return 0 on success else a negative value.
2350 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2351 struct relay_connection
*conn
,
2352 const struct lttng_buffer_view
*payload
)
2356 struct relay_session
*session
= conn
->session
;
2357 struct lttcomm_relayd_index index_info
;
2358 struct relay_index
*index
;
2359 struct lttcomm_relayd_generic_reply reply
;
2360 struct relay_stream
*stream
;
2365 DBG("Relay receiving index");
2367 if (!session
|| !conn
->version_check_done
) {
2368 ERR("Trying to close a stream before version check");
2370 goto end_no_session
;
2373 msg_len
= lttcomm_relayd_index_len(
2374 lttng_to_index_major(conn
->major
, conn
->minor
),
2375 lttng_to_index_minor(conn
->major
, conn
->minor
));
2376 if (payload
->size
< msg_len
) {
2377 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2378 msg_len
, payload
->size
);
2380 goto end_no_session
;
2382 memcpy(&index_info
, payload
->data
, msg_len
);
2383 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2384 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2385 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2386 index_info
.content_size
= be64toh(index_info
.content_size
);
2387 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2388 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2389 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2390 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2392 if (conn
->minor
>= 8) {
2393 index_info
.stream_instance_id
=
2394 be64toh(index_info
.stream_instance_id
);
2395 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2398 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2400 ERR("stream_get_by_id not found");
2404 pthread_mutex_lock(&stream
->lock
);
2406 /* Live beacon handling */
2407 if (index_info
.packet_size
== 0) {
2408 DBG("Received live beacon for stream %" PRIu64
,
2409 stream
->stream_handle
);
2412 * Only flag a stream inactive when it has already
2413 * received data and no indexes are in flight.
2415 if (stream
->index_received_seqcount
> 0
2416 && stream
->indexes_in_flight
== 0) {
2417 stream
->beacon_ts_end
= index_info
.timestamp_end
;
2420 goto end_stream_put
;
2422 stream
->beacon_ts_end
= -1ULL;
2425 if (stream
->ctf_stream_id
== -1ULL) {
2426 stream
->ctf_stream_id
= index_info
.stream_id
;
2428 index
= relay_index_get_by_id_or_create(stream
, index_info
.net_seq_num
);
2431 ERR("relay_index_get_by_id_or_create index NULL");
2432 goto end_stream_put
;
2434 if (set_index_control_data(index
, &index_info
, conn
)) {
2435 ERR("set_index_control_data error");
2436 relay_index_put(index
);
2438 goto end_stream_put
;
2440 ret
= relay_index_try_flush(index
);
2442 tracefile_array_commit_seq(stream
->tfa
);
2443 stream
->index_received_seqcount
++;
2444 stream
->pos_after_last_complete_data_index
+= index
->total_size
;
2445 } else if (ret
> 0) {
2449 ERR("relay_index_try_flush error %d", ret
);
2450 relay_index_put(index
);
2455 pthread_mutex_unlock(&stream
->lock
);
2460 memset(&reply
, 0, sizeof(reply
));
2462 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2464 reply
.ret_code
= htobe32(LTTNG_OK
);
2466 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2467 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2468 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2477 * Receive the streams_sent message.
2479 * Return 0 on success else a negative value.
2481 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2482 struct relay_connection
*conn
,
2483 const struct lttng_buffer_view
*payload
)
2487 struct lttcomm_relayd_generic_reply reply
;
2491 DBG("Relay receiving streams_sent");
2493 if (!conn
->session
|| !conn
->version_check_done
) {
2494 ERR("Trying to close a stream before version check");
2496 goto end_no_session
;
2500 * Publish every pending stream in the connection recv list which are
2501 * now ready to be used by the viewer.
2503 publish_connection_local_streams(conn
);
2505 memset(&reply
, 0, sizeof(reply
));
2506 reply
.ret_code
= htobe32(LTTNG_OK
);
2507 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2508 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2509 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2522 * relay_rotate_session_stream: rotate a stream to a new tracefile for the session
2523 * rotation feature (not the tracefile rotation feature).
2525 static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
2526 struct relay_connection
*conn
,
2527 const struct lttng_buffer_view
*payload
)
2531 struct relay_session
*session
= conn
->session
;
2532 struct lttcomm_relayd_rotate_stream stream_info
;
2533 struct lttcomm_relayd_generic_reply reply
;
2534 struct relay_stream
*stream
;
2537 struct lttng_buffer_view new_path_view
;
2539 DBG("Rotate stream received");
2541 if (!session
|| !conn
->version_check_done
) {
2542 ERR("Trying to rotate a stream before version check");
2547 if (session
->major
== 2 && session
->minor
< 11) {
2548 ERR("Unsupported feature before 2.11");
2553 header_len
= sizeof(struct lttcomm_relayd_rotate_stream
);
2555 if (payload
->size
< header_len
) {
2556 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2557 header_len
, payload
->size
);
2562 memcpy(&stream_info
, payload
->data
, header_len
);
2564 /* Convert to host */
2565 stream_info
.pathname_length
= be32toh(stream_info
.pathname_length
);
2566 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
2567 stream_info
.new_chunk_id
= be64toh(stream_info
.new_chunk_id
);
2568 stream_info
.rotate_at_seq_num
= be64toh(stream_info
.rotate_at_seq_num
);
2570 path_len
= stream_info
.pathname_length
;
2571 if (payload
->size
< header_len
+ path_len
) {
2572 ERR("Unexpected payload size in \"relay_rotate_session_stream\" including path: expected >= %zu bytes, got %zu bytes",
2573 header_len
+ path_len
, payload
->size
);
2578 /* Ensure it fits in local filename length. */
2579 if (path_len
>= LTTNG_PATH_MAX
) {
2580 ret
= -ENAMETOOLONG
;
2581 ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes",
2582 path_len
, LTTNG_PATH_MAX
);
2586 new_path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2587 stream_info
.pathname_length
);
2589 stream
= stream_get_by_id(stream_info
.stream_id
);
2595 pthread_mutex_lock(&stream
->lock
);
2598 * Update the trace path (just the folder, the stream name does not
2601 free(stream
->path_name
);
2602 stream
->path_name
= create_output_path(new_path_view
.data
);
2603 if (!stream
->path_name
) {
2604 ERR("Failed to create a new output path");
2606 goto end_stream_unlock
;
2608 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
,
2611 ERR("relay creating output directory");
2613 goto end_stream_unlock
;
2616 assert(stream
->current_chunk_id
.is_set
);
2617 stream
->current_chunk_id
.value
= stream_info
.new_chunk_id
;
2619 if (stream
->is_metadata
) {
2621 * The metadata stream is sent only over the control connection
2622 * so we know we have all the data to perform the stream
2625 ret
= do_rotate_stream(stream
);
2627 stream
->rotate_at_seq_num
= stream_info
.rotate_at_seq_num
;
2628 ret
= try_rotate_stream(stream
);
2631 goto end_stream_unlock
;
2635 pthread_mutex_unlock(&stream
->lock
);
2638 memset(&reply
, 0, sizeof(reply
));
2640 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2642 reply
.ret_code
= htobe32(LTTNG_OK
);
2644 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2645 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2646 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2647 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2657 * relay_mkdir: Create a folder on the disk.
2659 static int relay_mkdir(const struct lttcomm_relayd_hdr
*recv_hdr
,
2660 struct relay_connection
*conn
,
2661 const struct lttng_buffer_view
*payload
)
2664 struct relay_session
*session
= conn
->session
;
2665 struct lttcomm_relayd_mkdir path_info_header
;
2666 struct lttcomm_relayd_generic_reply reply
;
2670 struct lttng_buffer_view path_view
;
2672 if (!session
|| !conn
->version_check_done
) {
2673 ERR("Trying to create a directory before version check");
2675 goto end_no_session
;
2678 if (session
->major
== 2 && session
->minor
< 11) {
2680 * This client is not supposed to use this command since
2681 * it predates its introduction.
2683 ERR("relay_mkdir command is unsupported before LTTng 2.11");
2685 goto end_no_session
;
2688 header_len
= sizeof(path_info_header
);
2689 if (payload
->size
< header_len
) {
2690 ERR("Unexpected payload size in \"relay_mkdir\": expected >= %zu bytes, got %zu bytes",
2691 header_len
, payload
->size
);
2693 goto end_no_session
;
2696 memcpy(&path_info_header
, payload
->data
, header_len
);
2698 path_info_header
.length
= be32toh(path_info_header
.length
);
2700 if (payload
->size
< header_len
+ path_info_header
.length
) {
2701 ERR("Unexpected payload size in \"relay_mkdir\" including path: expected >= %zu bytes, got %zu bytes",
2702 header_len
+ path_info_header
.length
, payload
->size
);
2704 goto end_no_session
;
2707 /* Ensure that it fits in local path length. */
2708 if (path_info_header
.length
>= LTTNG_PATH_MAX
) {
2709 ret
= -ENAMETOOLONG
;
2710 ERR("Path name argument of mkdir command (%" PRIu32
" bytes) exceeds the maximal length allowed (%d bytes)",
2711 path_info_header
.length
, LTTNG_PATH_MAX
);
2715 path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2716 path_info_header
.length
);
2718 path
= create_output_path(path_view
.data
);
2720 ERR("Failed to create output path");
2725 ret
= utils_mkdir_recursive(path
, S_IRWXU
| S_IRWXG
, -1, -1);
2727 ERR("relay creating output directory");
2734 memset(&reply
, 0, sizeof(reply
));
2736 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2738 reply
.ret_code
= htobe32(LTTNG_OK
);
2740 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2741 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2742 ERR("Failed to send \"mkdir\" command reply (ret = %zd)", send_ret
);
2751 static int validate_rotate_rename_path_length(const char *path_type
,
2752 uint32_t path_length
)
2756 if (path_length
> LTTNG_PATH_MAX
) {
2757 ret
= -ENAMETOOLONG
;
2758 ERR("rotate rename \"%s\" path name length (%" PRIu32
" bytes) exceeds the allowed size of %i bytes",
2759 path_type
, path_length
, LTTNG_PATH_MAX
);
2760 } else if (path_length
== 0) {
2762 ERR("rotate rename \"%s\" path name has an illegal length of 0", path_type
);
2768 * relay_rotate_rename: rename the trace folder after a rotation is
2769 * completed. We are not closing any fd here, just moving the folder, so it
2770 * works even if data is still in-flight.
2772 static int relay_rotate_rename(const struct lttcomm_relayd_hdr
*recv_hdr
,
2773 struct relay_connection
*conn
,
2774 const struct lttng_buffer_view
*payload
)
2778 struct relay_session
*session
= conn
->session
;
2779 struct lttcomm_relayd_generic_reply reply
;
2780 struct lttcomm_relayd_rotate_rename header
;
2782 size_t received_paths_size
;
2783 char *complete_old_path
= NULL
, *complete_new_path
= NULL
;
2784 struct lttng_buffer_view old_path_view
;
2785 struct lttng_buffer_view new_path_view
;
2787 if (!session
|| !conn
->version_check_done
) {
2788 ERR("Trying to rename a trace folder before version check");
2793 if (session
->major
== 2 && session
->minor
< 11) {
2794 ERR("relay_rotate_rename command is unsupported before LTTng 2.11");
2799 header_len
= sizeof(header
);
2800 if (payload
->size
< header_len
) {
2801 ERR("Unexpected payload size in \"relay_rotate_rename\": expected >= %zu bytes, got %zu bytes",
2802 header_len
, payload
->size
);
2807 memcpy(&header
, payload
->data
, header_len
);
2809 header
.old_path_length
= be32toh(header
.old_path_length
);
2810 header
.new_path_length
= be32toh(header
.new_path_length
);
2811 received_paths_size
= header
.old_path_length
+ header
.new_path_length
;
2813 if (payload
->size
< header_len
+ received_paths_size
) {
2814 ERR("Unexpected payload size in \"relay_rotate_rename\" including paths: expected >= %zu bytes, got %zu bytes",
2815 header_len
, payload
->size
);
2820 /* Ensure the paths don't exceed their allowed size. */
2821 ret
= validate_rotate_rename_path_length("old", header
.old_path_length
);
2825 ret
= validate_rotate_rename_path_length("new", header
.new_path_length
);
2830 old_path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2831 header
.old_path_length
);
2832 new_path_view
= lttng_buffer_view_from_view(payload
,
2833 header_len
+ header
.old_path_length
,
2834 header
.new_path_length
);
2836 /* Validate that both paths received are NULL terminated. */
2837 if (old_path_view
.data
[old_path_view
.size
- 1] != '\0') {
2838 ERR("relay_rotate_rename command's \"old\" path is invalid (not NULL terminated)");
2842 if (new_path_view
.data
[new_path_view
.size
- 1] != '\0') {
2843 ERR("relay_rotate_rename command's \"new\" path is invalid (not NULL terminated)");
2848 complete_old_path
= create_output_path(old_path_view
.data
);
2849 if (!complete_old_path
) {
2850 ERR("Failed to build old output path in rotate_rename command");
2855 complete_new_path
= create_output_path(new_path_view
.data
);
2856 if (!complete_new_path
) {
2857 ERR("Failed to build new output path in rotate_rename command");
2862 ret
= utils_mkdir_recursive(complete_new_path
, S_IRWXU
| S_IRWXG
,
2865 ERR("Failed to mkdir() rotate_rename's \"new\" output directory at \"%s\"",
2871 * If a domain has not yet created its channel, the domain-specific
2872 * folder might not exist, but this is not an error.
2874 ret
= rename(complete_old_path
, complete_new_path
);
2875 if (ret
< 0 && errno
!= ENOENT
) {
2876 PERROR("Renaming chunk in rotate_rename command from \"%s\" to \"%s\"",
2877 complete_old_path
, complete_new_path
);
2883 memset(&reply
, 0, sizeof(reply
));
2885 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2887 reply
.ret_code
= htobe32(LTTNG_OK
);
2889 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2891 if (send_ret
< sizeof(reply
)) {
2892 ERR("Failed to send \"rotate rename\" command reply (ret = %zd)",
2898 free(complete_old_path
);
2899 free(complete_new_path
);
2904 * Check if all the streams in the session have completed the last rotation.
2905 * The chunk_id value is used to distinguish the cases where a stream was
2906 * closed on the consumerd before the rotation started but it still active on
2907 * the relayd, and the case where a stream appeared on the consumerd/relayd
2908 * after the last rotation started (in that case, it is already writing in the
2909 * new chunk folder).
2912 int relay_rotate_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2913 struct relay_connection
*conn
,
2914 const struct lttng_buffer_view
*payload
)
2916 struct relay_session
*session
= conn
->session
;
2917 struct lttcomm_relayd_rotate_pending msg
;
2918 struct lttcomm_relayd_rotate_pending_reply reply
;
2919 struct lttng_ht_iter iter
;
2920 struct relay_stream
*stream
;
2924 bool rotate_pending
= false;
2926 DBG("Rotate pending command received");
2928 if (!session
|| !conn
->version_check_done
) {
2929 ERR("Trying to check for data before version check");
2934 if (session
->major
== 2 && session
->minor
< 11) {
2935 ERR("Unsupported feature before 2.11");
2940 if (payload
->size
< sizeof(msg
)) {
2941 ERR("Unexpected payload size in \"relay_rotate_pending\": expected >= %zu bytes, got %zu bytes",
2942 sizeof(msg
), payload
->size
);
2947 memcpy(&msg
, payload
->data
, sizeof(msg
));
2949 chunk_id
= be64toh(msg
.chunk_id
);
2951 DBG("Evaluating rotate pending for session \"%s\" and chunk id %" PRIu64
,
2952 session
->session_name
, chunk_id
);
2955 * Iterate over all the streams in the session and check if they are
2956 * still waiting for data to perform their rotation.
2959 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2961 if (!stream_get(stream
)) {
2964 if (stream
->trace
->session
!= session
) {
2968 pthread_mutex_lock(&stream
->lock
);
2969 if (stream
->rotate_at_seq_num
!= -1ULL) {
2970 /* We have not yet performed the rotation. */
2971 rotate_pending
= true;
2972 DBG("Stream %" PRIu64
" is still rotating",
2973 stream
->stream_handle
);
2974 } else if (stream
->current_chunk_id
.value
< chunk_id
) {
2976 * Stream closed on the consumer but still active on the
2979 rotate_pending
= true;
2980 DBG("Stream %" PRIu64
" did not exist on the consumer "
2981 "when the last rotation started, but is"
2982 "still waiting for data before getting"
2984 stream
->stream_handle
);
2986 pthread_mutex_unlock(&stream
->lock
);
2988 if (rotate_pending
) {
2995 memset(&reply
, 0, sizeof(reply
));
2996 reply
.generic
.ret_code
= htobe32((uint32_t) LTTNG_OK
);
2997 reply
.is_pending
= (uint8_t) !!rotate_pending
;
2998 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
3000 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3001 ERR("Failed to send \"rotate pending\" command reply (ret = %zd)",
3010 #define DBG_CMD(cmd_name, conn) \
3011 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
3013 static int relay_process_control_command(struct relay_connection
*conn
,
3014 const struct lttcomm_relayd_hdr
*header
,
3015 const struct lttng_buffer_view
*payload
)
3019 switch (header
->cmd
) {
3020 case RELAYD_CREATE_SESSION
:
3021 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
3022 ret
= relay_create_session(header
, conn
, payload
);
3024 case RELAYD_ADD_STREAM
:
3025 DBG_CMD("RELAYD_ADD_STREAM", conn
);
3026 ret
= relay_add_stream(header
, conn
, payload
);
3028 case RELAYD_START_DATA
:
3029 DBG_CMD("RELAYD_START_DATA", conn
);
3030 ret
= relay_start(header
, conn
, payload
);
3032 case RELAYD_SEND_METADATA
:
3033 DBG_CMD("RELAYD_SEND_METADATA", conn
);
3034 ret
= relay_recv_metadata(header
, conn
, payload
);
3036 case RELAYD_VERSION
:
3037 DBG_CMD("RELAYD_VERSION", conn
);
3038 ret
= relay_send_version(header
, conn
, payload
);
3040 case RELAYD_CLOSE_STREAM
:
3041 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
3042 ret
= relay_close_stream(header
, conn
, payload
);
3044 case RELAYD_DATA_PENDING
:
3045 DBG_CMD("RELAYD_DATA_PENDING", conn
);
3046 ret
= relay_data_pending(header
, conn
, payload
);
3048 case RELAYD_QUIESCENT_CONTROL
:
3049 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
3050 ret
= relay_quiescent_control(header
, conn
, payload
);
3052 case RELAYD_BEGIN_DATA_PENDING
:
3053 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
3054 ret
= relay_begin_data_pending(header
, conn
, payload
);
3056 case RELAYD_END_DATA_PENDING
:
3057 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
3058 ret
= relay_end_data_pending(header
, conn
, payload
);
3060 case RELAYD_SEND_INDEX
:
3061 DBG_CMD("RELAYD_SEND_INDEX", conn
);
3062 ret
= relay_recv_index(header
, conn
, payload
);
3064 case RELAYD_STREAMS_SENT
:
3065 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
3066 ret
= relay_streams_sent(header
, conn
, payload
);
3068 case RELAYD_RESET_METADATA
:
3069 DBG_CMD("RELAYD_RESET_METADATA", conn
);
3070 ret
= relay_reset_metadata(header
, conn
, payload
);
3072 case RELAYD_ROTATE_STREAM
:
3073 DBG_CMD("RELAYD_ROTATE_STREAM", conn
);
3074 ret
= relay_rotate_session_stream(header
, conn
, payload
);
3076 case RELAYD_ROTATE_RENAME
:
3077 DBG_CMD("RELAYD_ROTATE_RENAME", conn
);
3078 ret
= relay_rotate_rename(header
, conn
, payload
);
3080 case RELAYD_ROTATE_PENDING
:
3081 DBG_CMD("RELAYD_ROTATE_PENDING", conn
);
3082 ret
= relay_rotate_pending(header
, conn
, payload
);
3085 DBG_CMD("RELAYD_MKDIR", conn
);
3086 ret
= relay_mkdir(header
, conn
, payload
);
3088 case RELAYD_UPDATE_SYNC_INFO
:
3090 ERR("Received unknown command (%u)", header
->cmd
);
3091 relay_unknown_command(conn
);
3100 static enum relay_connection_status
relay_process_control_receive_payload(
3101 struct relay_connection
*conn
)
3104 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3105 struct lttng_dynamic_buffer
*reception_buffer
=
3106 &conn
->protocol
.ctrl
.reception_buffer
;
3107 struct ctrl_connection_state_receive_payload
*state
=
3108 &conn
->protocol
.ctrl
.state
.receive_payload
;
3109 struct lttng_buffer_view payload_view
;
3111 if (state
->left_to_receive
== 0) {
3112 /* Short-circuit for payload-less commands. */
3113 goto reception_complete
;
3116 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3117 reception_buffer
->data
+ state
->received
,
3118 state
->left_to_receive
, MSG_DONTWAIT
);
3120 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3121 PERROR("Unable to receive command payload on sock %d",
3123 status
= RELAY_CONNECTION_STATUS_ERROR
;
3126 } else if (ret
== 0) {
3127 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3128 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3133 assert(ret
<= state
->left_to_receive
);
3135 state
->left_to_receive
-= ret
;
3136 state
->received
+= ret
;
3138 if (state
->left_to_receive
> 0) {
3140 * Can't transition to the protocol's next state, wait to
3141 * receive the rest of the header.
3143 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3144 state
->received
, state
->left_to_receive
,
3150 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
3151 conn
->sock
->fd
, state
->received
);
3153 * The payload required to process the command has been received.
3154 * A view to the reception buffer is forwarded to the various
3155 * commands and the state of the control is reset on success.
3157 * Commands are responsible for sending their reply to the peer.
3159 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
3161 ret
= relay_process_control_command(conn
,
3162 &state
->header
, &payload_view
);
3164 status
= RELAY_CONNECTION_STATUS_ERROR
;
3168 ret
= connection_reset_protocol_state(conn
);
3170 status
= RELAY_CONNECTION_STATUS_ERROR
;
3176 static enum relay_connection_status
relay_process_control_receive_header(
3177 struct relay_connection
*conn
)
3180 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3181 struct lttcomm_relayd_hdr header
;
3182 struct lttng_dynamic_buffer
*reception_buffer
=
3183 &conn
->protocol
.ctrl
.reception_buffer
;
3184 struct ctrl_connection_state_receive_header
*state
=
3185 &conn
->protocol
.ctrl
.state
.receive_header
;
3187 assert(state
->left_to_receive
!= 0);
3189 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3190 reception_buffer
->data
+ state
->received
,
3191 state
->left_to_receive
, MSG_DONTWAIT
);
3193 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3194 PERROR("Unable to receive control command header on sock %d",
3196 status
= RELAY_CONNECTION_STATUS_ERROR
;
3199 } else if (ret
== 0) {
3200 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3201 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3206 assert(ret
<= state
->left_to_receive
);
3208 state
->left_to_receive
-= ret
;
3209 state
->received
+= ret
;
3211 if (state
->left_to_receive
> 0) {
3213 * Can't transition to the protocol's next state, wait to
3214 * receive the rest of the header.
3216 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3217 state
->received
, state
->left_to_receive
,
3222 /* Transition to next state: receiving the command's payload. */
3223 conn
->protocol
.ctrl
.state_id
=
3224 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3225 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
3226 header
.circuit_id
= be64toh(header
.circuit_id
);
3227 header
.data_size
= be64toh(header
.data_size
);
3228 header
.cmd
= be32toh(header
.cmd
);
3229 header
.cmd_version
= be32toh(header
.cmd_version
);
3230 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
3231 &header
, sizeof(header
));
3233 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
3234 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
3237 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
3238 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
3240 status
= RELAY_CONNECTION_STATUS_ERROR
;
3244 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
3246 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
3247 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
3250 status
= RELAY_CONNECTION_STATUS_ERROR
;
3254 if (header
.data_size
== 0) {
3256 * Manually invoke the next state as the poll loop
3257 * will not wake-up to allow us to proceed further.
3259 status
= relay_process_control_receive_payload(conn
);
3266 * Process the commands received on the control socket
3268 static enum relay_connection_status
relay_process_control(
3269 struct relay_connection
*conn
)
3271 enum relay_connection_status status
;
3273 switch (conn
->protocol
.ctrl
.state_id
) {
3274 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
3275 status
= relay_process_control_receive_header(conn
);
3277 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3278 status
= relay_process_control_receive_payload(conn
);
3281 ERR("Unknown control connection protocol state encountered.");
3289 * Handle index for a data stream.
3291 * Called with the stream lock held.
3293 * Return 0 on success else a negative value.
3295 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
3296 bool rotate_index
, bool *flushed
, uint64_t total_size
)
3299 uint64_t data_offset
;
3300 struct relay_index
*index
;
3302 /* Get data offset because we are about to update the index. */
3303 data_offset
= htobe64(stream
->tracefile_size_current
);
3305 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
3306 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
3309 * Lookup for an existing index for that stream id/sequence
3310 * number. If it exists, the control thread has already received the
3311 * data for it, thus we need to write it to disk.
3313 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
3319 if (rotate_index
|| !stream
->index_file
) {
3320 ret
= create_rotate_index_file(stream
);
3322 ERR("Failed to rotate index");
3323 /* Put self-ref for this index due to error. */
3324 relay_index_put(index
);
3330 if (relay_index_set_file(index
, stream
->index_file
, data_offset
)) {
3332 /* Put self-ref for this index due to error. */
3333 relay_index_put(index
);
3338 ret
= relay_index_try_flush(index
);
3340 tracefile_array_commit_seq(stream
->tfa
);
3341 stream
->index_received_seqcount
++;
3343 } else if (ret
> 0) {
3344 index
->total_size
= total_size
;
3348 /* Put self-ref for this index due to error. */
3349 relay_index_put(index
);
3357 static enum relay_connection_status
relay_process_data_receive_header(
3358 struct relay_connection
*conn
)
3361 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3362 struct data_connection_state_receive_header
*state
=
3363 &conn
->protocol
.data
.state
.receive_header
;
3364 struct lttcomm_relayd_data_hdr header
;
3365 struct relay_stream
*stream
;
3367 assert(state
->left_to_receive
!= 0);
3369 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3370 state
->header_reception_buffer
+ state
->received
,
3371 state
->left_to_receive
, MSG_DONTWAIT
);
3373 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3374 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
3375 status
= RELAY_CONNECTION_STATUS_ERROR
;
3378 } else if (ret
== 0) {
3379 /* Orderly shutdown. Not necessary to print an error. */
3380 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3381 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3386 assert(ret
<= state
->left_to_receive
);
3388 state
->left_to_receive
-= ret
;
3389 state
->received
+= ret
;
3391 if (state
->left_to_receive
> 0) {
3393 * Can't transition to the protocol's next state, wait to
3394 * receive the rest of the header.
3396 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3397 state
->received
, state
->left_to_receive
,
3402 /* Transition to next state: receiving the payload. */
3403 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3405 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3406 header
.circuit_id
= be64toh(header
.circuit_id
);
3407 header
.stream_id
= be64toh(header
.stream_id
);
3408 header
.data_size
= be32toh(header
.data_size
);
3409 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3410 header
.padding_size
= be32toh(header
.padding_size
);
3411 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3413 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3415 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3416 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3418 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3419 conn
->sock
->fd
, header
.circuit_id
,
3420 header
.stream_id
, header
.data_size
,
3421 header
.net_seq_num
, header
.padding_size
);
3423 stream
= stream_get_by_id(header
.stream_id
);
3425 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3427 /* Protocol error. */
3428 status
= RELAY_CONNECTION_STATUS_ERROR
;
3432 pthread_mutex_lock(&stream
->lock
);
3434 /* Check if a rotation is needed. */
3435 if (stream
->tracefile_size
> 0 &&
3436 (stream
->tracefile_size_current
+ header
.data_size
) >
3437 stream
->tracefile_size
) {
3438 uint64_t old_id
, new_id
;
3440 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
3441 tracefile_array_file_rotate(stream
->tfa
);
3443 /* new_id is updated by utils_rotate_stream_file. */
3446 ret
= utils_rotate_stream_file(stream
->path_name
,
3447 stream
->channel_name
, stream
->tracefile_size
,
3448 stream
->tracefile_count
, -1,
3449 -1, stream
->stream_fd
->fd
,
3450 &new_id
, &stream
->stream_fd
->fd
);
3452 ERR("Failed to rotate stream output file");
3453 status
= RELAY_CONNECTION_STATUS_ERROR
;
3454 goto end_stream_unlock
;
3458 * Reset current size because we just performed a stream
3461 stream
->tracefile_size_current
= 0;
3462 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= true;
3466 pthread_mutex_unlock(&stream
->lock
);
3472 static enum relay_connection_status
relay_process_data_receive_payload(
3473 struct relay_connection
*conn
)
3476 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3477 struct relay_stream
*stream
;
3478 struct data_connection_state_receive_payload
*state
=
3479 &conn
->protocol
.data
.state
.receive_payload
;
3480 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3481 char data_buffer
[chunk_size
];
3482 bool partial_recv
= false;
3483 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3484 uint64_t left_to_receive
= state
->left_to_receive
;
3485 struct relay_session
*session
;
3487 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3488 state
->header
.stream_id
, state
->header
.net_seq_num
,
3489 state
->received
, left_to_receive
);
3491 stream
= stream_get_by_id(state
->header
.stream_id
);
3493 /* Protocol error. */
3494 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3495 state
->header
.stream_id
);
3496 status
= RELAY_CONNECTION_STATUS_ERROR
;
3500 pthread_mutex_lock(&stream
->lock
);
3501 session
= stream
->trace
->session
;
3502 if (!conn
->session
) {
3503 ret
= connection_set_session(conn
, session
);
3505 status
= RELAY_CONNECTION_STATUS_ERROR
;
3506 goto end_stream_unlock
;
3511 * The size of the "chunk" received on any iteration is bounded by:
3512 * - the data left to receive,
3513 * - the data immediately available on the socket,
3514 * - the on-stack data buffer
3516 while (left_to_receive
> 0 && !partial_recv
) {
3518 size_t recv_size
= min(left_to_receive
, chunk_size
);
3520 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3521 recv_size
, MSG_DONTWAIT
);
3523 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3524 PERROR("Socket %d error", conn
->sock
->fd
);
3525 status
= RELAY_CONNECTION_STATUS_ERROR
;
3527 goto end_stream_unlock
;
3528 } else if (ret
== 0) {
3529 /* No more data ready to be consumed on socket. */
3530 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3531 state
->header
.stream_id
);
3532 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3534 } else if (ret
< (int) recv_size
) {
3536 * All the data available on the socket has been
3539 partial_recv
= true;
3544 /* Write data to stream output fd. */
3545 write_ret
= lttng_write(stream
->stream_fd
->fd
, data_buffer
,
3547 if (write_ret
< (ssize_t
) recv_size
) {
3548 ERR("Relay error writing data to file");
3549 status
= RELAY_CONNECTION_STATUS_ERROR
;
3550 goto end_stream_unlock
;
3553 left_to_receive
-= recv_size
;
3554 state
->received
+= recv_size
;
3555 state
->left_to_receive
= left_to_receive
;
3557 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
3558 write_ret
, stream
->stream_handle
);
3561 if (state
->left_to_receive
> 0) {
3563 * Did not receive all the data expected, wait for more data to
3564 * become available on the socket.
3566 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3567 state
->header
.stream_id
, state
->received
,
3568 state
->left_to_receive
);
3569 goto end_stream_unlock
;
3572 ret
= write_padding_to_file(stream
->stream_fd
->fd
,
3573 state
->header
.padding_size
);
3574 if ((int64_t) ret
< (int64_t) state
->header
.padding_size
) {
3575 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3576 stream
->stream_handle
,
3577 state
->header
.net_seq_num
, ret
);
3578 status
= RELAY_CONNECTION_STATUS_ERROR
;
3579 goto end_stream_unlock
;
3583 if (session
->minor
>= 4 && !session
->snapshot
) {
3584 ret
= handle_index_data(stream
, state
->header
.net_seq_num
,
3585 state
->rotate_index
, &index_flushed
, state
->header
.data_size
+ state
->header
.padding_size
);
3587 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3588 stream
->stream_handle
,
3589 state
->header
.net_seq_num
, ret
);
3590 status
= RELAY_CONNECTION_STATUS_ERROR
;
3591 goto end_stream_unlock
;
3595 stream
->tracefile_size_current
+= state
->header
.data_size
+
3596 state
->header
.padding_size
;
3598 if (stream
->prev_seq
== -1ULL) {
3601 if (index_flushed
) {
3602 stream
->pos_after_last_complete_data_index
=
3603 stream
->tracefile_size_current
;
3606 stream
->prev_seq
= state
->header
.net_seq_num
;
3609 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3610 * contents of *state which are aliased (union) to the same location as
3611 * the new state. Don't use it beyond this point.
3613 connection_reset_protocol_state(conn
);
3616 ret
= try_rotate_stream(stream
);
3618 status
= RELAY_CONNECTION_STATUS_ERROR
;
3619 goto end_stream_unlock
;
3623 close_requested
= stream
->close_requested
;
3624 pthread_mutex_unlock(&stream
->lock
);
3625 if (close_requested
&& left_to_receive
== 0) {
3626 try_stream_close(stream
);
3630 pthread_mutex_lock(&session
->lock
);
3631 uatomic_set(&session
->new_streams
, 1);
3632 pthread_mutex_unlock(&session
->lock
);
3641 * relay_process_data: Process the data received on the data socket
3643 static enum relay_connection_status
relay_process_data(
3644 struct relay_connection
*conn
)
3646 enum relay_connection_status status
;
3648 switch (conn
->protocol
.data
.state_id
) {
3649 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3650 status
= relay_process_data_receive_header(conn
);
3652 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3653 status
= relay_process_data_receive_payload(conn
);
3656 ERR("Unexpected data connection communication state.");
3663 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3667 (void) lttng_poll_del(events
, pollfd
);
3669 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
, &pollfd
, 1,
3670 fd_tracker_util_close_fd
, NULL
);
3672 ERR("Closing pollfd %d", pollfd
);
3676 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3677 int pollfd
, struct relay_connection
*conn
)
3679 const char *type_str
;
3681 switch (conn
->type
) {
3686 type_str
= "Control";
3688 case RELAY_VIEWER_COMMAND
:
3689 type_str
= "Viewer Command";
3691 case RELAY_VIEWER_NOTIFICATION
:
3692 type_str
= "Viewer Notification";
3695 type_str
= "Unknown";
3697 cleanup_connection_pollfd(events
, pollfd
);
3698 connection_put(conn
);
3699 DBG("%s connection closed with %d", type_str
, pollfd
);
3703 * This thread does the actual work
3705 static void *relay_thread_worker(void *data
)
3707 int ret
, err
= -1, last_seen_data_fd
= -1;
3709 struct lttng_poll_event events
;
3710 struct lttng_ht
*relay_connections_ht
;
3711 struct lttng_ht_iter iter
;
3712 struct relay_connection
*destroy_conn
= NULL
;
3714 DBG("[thread] Relay worker started");
3716 rcu_register_thread();
3718 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3720 if (testpoint(relayd_thread_worker
)) {
3721 goto error_testpoint
;
3724 health_code_update();
3726 /* table of connections indexed on socket */
3727 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3728 if (!relay_connections_ht
) {
3729 goto relay_connections_ht_error
;
3732 ret
= create_named_thread_poll_set(&events
, 2, "Worker thread epoll");
3734 goto error_poll_create
;
3737 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3744 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3746 health_code_update();
3748 /* Infinite blocking call, waiting for transmission */
3749 DBG3("Relayd worker thread polling...");
3750 health_poll_entry();
3751 ret
= lttng_poll_wait(&events
, -1);
3755 * Restart interrupted system call.
3757 if (errno
== EINTR
) {
3766 * Process control. The control connection is
3767 * prioritized so we don't starve it with high
3768 * throughput tracing data on the data connection.
3770 for (i
= 0; i
< nb_fd
; i
++) {
3771 /* Fetch once the poll data */
3772 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3773 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3775 health_code_update();
3779 * No activity for this FD (poll
3785 /* Thread quit pipe has been closed. Killing thread. */
3786 ret
= check_thread_quit_pipe(pollfd
, revents
);
3792 /* Inspect the relay conn pipe for new connection */
3793 if (pollfd
== relay_conn_pipe
[0]) {
3794 if (revents
& LPOLLIN
) {
3795 struct relay_connection
*conn
;
3797 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3801 lttng_poll_add(&events
, conn
->sock
->fd
,
3802 LPOLLIN
| LPOLLRDHUP
);
3803 connection_ht_add(relay_connections_ht
, conn
);
3804 DBG("Connection socket %d added", conn
->sock
->fd
);
3805 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3806 ERR("Relay connection pipe error");
3809 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3813 struct relay_connection
*ctrl_conn
;
3815 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3816 /* If not found, there is a synchronization issue. */
3819 if (ctrl_conn
->type
== RELAY_DATA
) {
3820 if (revents
& LPOLLIN
) {
3822 * Flag the last seen data fd not deleted. It will be
3823 * used as the last seen fd if any fd gets deleted in
3826 last_notdel_data_fd
= pollfd
;
3828 goto put_ctrl_connection
;
3830 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3832 if (revents
& LPOLLIN
) {
3833 enum relay_connection_status status
;
3835 status
= relay_process_control(ctrl_conn
);
3836 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3838 * On socket error flag the session as aborted to force
3839 * the cleanup of its stream otherwise it can leak
3840 * during the lifetime of the relayd.
3842 * This prevents situations in which streams can be
3843 * left opened because an index was received, the
3844 * control connection is closed, and the data
3845 * connection is closed (uncleanly) before the packet's
3848 * Since the control connection encountered an error,
3849 * it is okay to be conservative and close the
3850 * session right now as we can't rely on the protocol
3851 * being respected anymore.
3853 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3854 session_abort(ctrl_conn
->session
);
3857 /* Clear the connection on error or close. */
3858 relay_thread_close_connection(&events
,
3863 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3864 relay_thread_close_connection(&events
,
3866 if (last_seen_data_fd
== pollfd
) {
3867 last_seen_data_fd
= last_notdel_data_fd
;
3870 ERR("Unexpected poll events %u for control sock %d",
3872 connection_put(ctrl_conn
);
3875 put_ctrl_connection
:
3876 connection_put(ctrl_conn
);
3881 * The last loop handled a control request, go back to poll to make
3882 * sure we prioritise the control socket.
3888 if (last_seen_data_fd
>= 0) {
3889 for (i
= 0; i
< nb_fd
; i
++) {
3890 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3892 health_code_update();
3894 if (last_seen_data_fd
== pollfd
) {
3901 /* Process data connection. */
3902 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3903 /* Fetch the poll data. */
3904 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3905 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3906 struct relay_connection
*data_conn
;
3908 health_code_update();
3911 /* No activity for this FD (poll implementation). */
3915 /* Skip the command pipe. It's handled in the first loop. */
3916 if (pollfd
== relay_conn_pipe
[0]) {
3920 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3922 /* Skip it. Might be removed before. */
3925 if (data_conn
->type
== RELAY_CONTROL
) {
3926 goto put_data_connection
;
3928 assert(data_conn
->type
== RELAY_DATA
);
3930 if (revents
& LPOLLIN
) {
3931 enum relay_connection_status status
;
3933 status
= relay_process_data(data_conn
);
3934 /* Connection closed or error. */
3935 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3937 * On socket error flag the session as aborted to force
3938 * the cleanup of its stream otherwise it can leak
3939 * during the lifetime of the relayd.
3941 * This prevents situations in which streams can be
3942 * left opened because an index was received, the
3943 * control connection is closed, and the data
3944 * connection is closed (uncleanly) before the packet's
3947 * Since the data connection encountered an error,
3948 * it is okay to be conservative and close the
3949 * session right now as we can't rely on the protocol
3950 * being respected anymore.
3952 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3953 session_abort(data_conn
->session
);
3955 relay_thread_close_connection(&events
, pollfd
,
3958 * Every goto restart call sets the last seen fd where
3959 * here we don't really care since we gracefully
3960 * continue the loop after the connection is deleted.
3963 /* Keep last seen port. */
3964 last_seen_data_fd
= pollfd
;
3965 connection_put(data_conn
);
3968 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3969 relay_thread_close_connection(&events
, pollfd
,
3972 ERR("Unknown poll events %u for data sock %d",
3975 put_data_connection
:
3976 connection_put(data_conn
);
3978 last_seen_data_fd
= -1;
3981 /* Normal exit, no error */
3986 /* Cleanup remaining connection object. */
3988 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3991 health_code_update();
3993 session_abort(destroy_conn
->session
);
3996 * No need to grab another ref, because we own
3999 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
4004 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
4006 lttng_ht_destroy(relay_connections_ht
);
4007 relay_connections_ht_error
:
4008 /* Close relay conn pipes */
4009 (void) fd_tracker_util_pipe_close(the_fd_tracker
,
4012 DBG("Thread exited with error");
4014 DBG("Worker thread cleanup complete");
4018 ERR("Health error occurred in %s", __func__
);
4020 health_unregister(health_relayd
);
4021 rcu_unregister_thread();
4022 lttng_relay_stop_threads();
4027 * Create the relay command pipe to wake thread_manage_apps.
4028 * Closed in cleanup().
4030 static int create_relay_conn_pipe(void)
4032 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
4033 "Relayd connection pipe", relay_conn_pipe
);
4037 int stdio_open(void *data
, int *fds
)
4039 fds
[0] = fileno(stdout
);
4040 fds
[1] = fileno(stderr
);
4045 int noop_close(void *data
, int *fds
)
4051 int track_stdio(void)
4054 const char *names
[] = { "stdout", "stderr" };
4056 return fd_tracker_open_unsuspendable_fd(the_fd_tracker
, fds
,
4057 names
, 2, stdio_open
, NULL
);
4061 void untrack_stdio(void)
4063 int fds
[] = { fileno(stdout
), fileno(stderr
) };
4066 * noop_close is used since we don't really want to close
4067 * the stdio output fds; we merely want to stop tracking them.
4069 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
4070 fds
, 2, noop_close
, NULL
);
4076 int main(int argc
, char **argv
)
4078 int ret
= 0, retval
= 0;
4081 /* Parse arguments */
4083 if (set_options(argc
, argv
)) {
4088 if (set_signal_handler()) {
4093 ret
= fclose(stdin
);
4095 PERROR("Failed to close stdin");
4098 /* Try to create directory if -o, --output is specified. */
4099 if (opt_output_path
) {
4100 if (*opt_output_path
!= '/') {
4101 ERR("Please specify an absolute path for -o, --output PATH");
4106 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
4109 ERR("Unable to create %s", opt_output_path
);
4116 if (opt_daemon
|| opt_background
) {
4117 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4126 * The RCU thread registration (and use, through the fd-tracker's
4127 * creation) is done after the daemonization to allow us to not
4128 * deal with liburcu's fork() management as the call RCU needs to
4131 rcu_register_thread();
4133 the_fd_tracker
= fd_tracker_create(lttng_opt_fd_cap
);
4134 if (!the_fd_tracker
) {
4139 ret
= track_stdio();
4145 /* Initialize thread health monitoring */
4146 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
4147 if (!health_relayd
) {
4148 PERROR("health_app_create error");
4150 goto exit_health_app_create
;
4153 /* Create thread quit pipe */
4154 if (init_thread_quit_pipe()) {
4156 goto exit_init_data
;
4159 /* Setup the thread apps communication pipe. */
4160 if (create_relay_conn_pipe()) {
4162 goto exit_init_data
;
4165 /* Init relay command queue. */
4166 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
4168 /* Initialize communication library */
4170 lttcomm_inet_init();
4172 /* tables of sessions indexed by session ID */
4173 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4176 goto exit_init_data
;
4179 /* tables of streams indexed by stream ID */
4180 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4181 if (!relay_streams_ht
) {
4183 goto exit_init_data
;
4186 /* tables of streams indexed by stream ID */
4187 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4188 if (!viewer_streams_ht
) {
4190 goto exit_init_data
;
4193 ret
= init_health_quit_pipe();
4196 goto exit_health_quit_pipe
;
4199 /* Create thread to manage the client socket */
4200 ret
= pthread_create(&health_thread
, default_pthread_attr(),
4201 thread_manage_health
, (void *) NULL
);
4204 PERROR("pthread_create health");
4206 goto exit_health_thread
;
4209 /* Setup the dispatcher thread */
4210 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
4211 relay_thread_dispatcher
, (void *) NULL
);
4214 PERROR("pthread_create dispatcher");
4216 goto exit_dispatcher_thread
;
4219 /* Setup the worker thread */
4220 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
4221 relay_thread_worker
, NULL
);
4224 PERROR("pthread_create worker");
4226 goto exit_worker_thread
;
4229 /* Setup the listener thread */
4230 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
4231 relay_thread_listener
, (void *) NULL
);
4234 PERROR("pthread_create listener");
4236 goto exit_listener_thread
;
4239 ret
= relayd_live_create(live_uri
);
4241 ERR("Starting live viewer threads");
4247 * This is where we start awaiting program completion (e.g. through
4248 * signal that asks threads to teardown).
4251 ret
= relayd_live_join();
4257 ret
= pthread_join(listener_thread
, &status
);
4260 PERROR("pthread_join listener_thread");
4264 exit_listener_thread
:
4265 ret
= pthread_join(worker_thread
, &status
);
4268 PERROR("pthread_join worker_thread");
4273 ret
= pthread_join(dispatcher_thread
, &status
);
4276 PERROR("pthread_join dispatcher_thread");
4279 exit_dispatcher_thread
:
4281 ret
= pthread_join(health_thread
, &status
);
4284 PERROR("pthread_join health_thread");
4289 (void) fd_tracker_util_pipe_close(the_fd_tracker
, health_quit_pipe
);
4290 exit_health_quit_pipe
:
4293 health_app_destroy(health_relayd
);
4294 exit_health_app_create
:
4297 * Wait for all pending call_rcu work to complete before tearing
4298 * down data structures. call_rcu worker may be trying to
4299 * perform lookups in those structures.
4304 /* Ensure all prior call_rcu are done. */
4309 * fd_tracker_destroy() will log the contents of the fd-tracker
4310 * if a leak is detected.
4312 fd_tracker_destroy(the_fd_tracker
);
4313 rcu_unregister_thread();