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
= stream_fd_rotate(stream
->stream_fd
,
1551 stream
->path_name
, stream
->channel_name
, 0, 0, NULL
);
1553 ERR("Failed to rotate metadata file %s of channel %s",
1554 stream
->path_name
, stream
->channel_name
);
1559 pthread_mutex_unlock(&stream
->lock
);
1563 memset(&reply
, 0, sizeof(reply
));
1565 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1567 reply
.ret_code
= htobe32(LTTNG_OK
);
1569 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1570 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1571 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1572 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1582 * relay_unknown_command: send -1 if received unknown command
1584 static void relay_unknown_command(struct relay_connection
*conn
)
1586 struct lttcomm_relayd_generic_reply reply
;
1589 memset(&reply
, 0, sizeof(reply
));
1590 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1591 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1592 if (send_ret
< sizeof(reply
)) {
1593 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1598 * relay_start: send an acknowledgment to the client to tell if we are
1599 * ready to receive data. We are ready if a session is established.
1601 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1602 struct relay_connection
*conn
,
1603 const struct lttng_buffer_view
*payload
)
1607 struct lttcomm_relayd_generic_reply reply
;
1608 struct relay_session
*session
= conn
->session
;
1611 DBG("Trying to start the streaming without a session established");
1612 ret
= htobe32(LTTNG_ERR_UNK
);
1615 memset(&reply
, 0, sizeof(reply
));
1616 reply
.ret_code
= htobe32(LTTNG_OK
);
1617 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1619 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1620 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1629 * Append padding to the file pointed by the file descriptor fd.
1631 static int write_padding_to_file(int fd
, uint32_t size
)
1640 zeros
= zmalloc(size
);
1641 if (zeros
== NULL
) {
1642 PERROR("zmalloc zeros for padding");
1647 ret
= lttng_write(fd
, zeros
, size
);
1649 PERROR("write padding to file");
1659 * Close the current index file if it is open, and create a new one.
1661 * Return 0 on success, -1 on error.
1664 int create_rotate_index_file(struct relay_stream
*stream
)
1667 uint32_t major
, minor
;
1669 /* Put ref on previous index_file. */
1670 if (stream
->index_file
) {
1671 relay_index_file_put(stream
->index_file
);
1672 stream
->index_file
= NULL
;
1674 major
= stream
->trace
->session
->major
;
1675 minor
= stream
->trace
->session
->minor
;
1676 stream
->index_file
= relay_index_file_create(stream
->path_name
,
1677 stream
->channel_name
,
1678 stream
->tracefile_size
,
1679 tracefile_array_get_file_index_head(stream
->tfa
),
1680 lttng_to_index_major(major
, minor
),
1681 lttng_to_index_minor(major
, minor
));
1682 if (!stream
->index_file
) {
1694 int do_rotate_stream(struct relay_stream
*stream
)
1698 /* Perform the stream rotation. */
1699 ret
= stream_fd_rotate(stream
->stream_fd
, stream
->path_name
,
1700 stream
->channel_name
, stream
->tracefile_size
,
1701 stream
->tracefile_count
, NULL
);
1703 ERR("Rotating stream output file");
1706 stream
->tracefile_size_current
= 0;
1708 /* Rotate also the index if the stream is not a metadata stream. */
1709 if (!stream
->is_metadata
) {
1710 ret
= create_rotate_index_file(stream
);
1712 ERR("Failed to rotate index file");
1717 stream
->rotate_at_seq_num
= -1ULL;
1718 stream
->pos_after_last_complete_data_index
= 0;
1725 * If too much data has been written in a tracefile before we received the
1726 * rotation command, we have to move the excess data to the new tracefile and
1727 * perform the rotation. This can happen because the control and data
1728 * connections are separate, the indexes as well as the commands arrive from
1729 * the control connection and we have no control over the order so we could be
1730 * in a situation where too much data has been received on the data connection
1731 * before the rotation command on the control connection arrives. We don't need
1732 * to update the index because its order is guaranteed with the rotation
1736 int rotate_truncate_stream(struct relay_stream
*stream
)
1738 int ret
, old_fd
= -1, new_fd
= -1;
1740 uint64_t diff
, pos
= 0;
1741 char buf
[FILE_COPY_BUFFER_SIZE
];
1742 struct stream_fd
*new_stream_fd
= NULL
;
1744 assert(!stream
->is_metadata
);
1746 assert(stream
->tracefile_size_current
>
1747 stream
->pos_after_last_complete_data_index
);
1748 diff
= stream
->tracefile_size_current
-
1749 stream
->pos_after_last_complete_data_index
;
1751 /* Create the new tracefile. */
1752 new_stream_fd
= stream_fd_create(stream
->path_name
, stream
->channel_name
,
1753 stream
->tracefile_size
, stream
->tracefile_count
, NULL
);
1754 if (new_stream_fd
< 0) {
1755 ERR("Failed to create new stream file at path %s for channel %s",
1756 stream
->path_name
, stream
->channel_name
);
1761 old_fd
= stream_fd_get_fd(stream
->stream_fd
);
1767 new_fd
= stream_fd_get_fd(new_stream_fd
);
1774 * Rewind the current tracefile to the position at which the rotation
1775 * should have occured.
1777 lseek_ret
= lseek(old_fd
, stream
->pos_after_last_complete_data_index
,
1779 if (lseek_ret
< 0) {
1780 PERROR("seek truncate stream");
1785 /* Move data from the old file to the new file. */
1786 while (pos
< diff
) {
1787 uint64_t count
, bytes_left
;
1790 bytes_left
= diff
- pos
;
1791 count
= bytes_left
> sizeof(buf
) ? sizeof(buf
) : bytes_left
;
1792 assert(count
<= SIZE_MAX
);
1794 io_ret
= lttng_read(old_fd
, buf
, count
);
1795 if (io_ret
< (ssize_t
) count
) {
1796 char error_string
[256];
1798 snprintf(error_string
, sizeof(error_string
),
1799 "Failed to read %" PRIu64
" bytes from in rotate_truncate_stream(), returned %zi",
1802 PERROR("%s", error_string
);
1804 ERR("%s", error_string
);
1810 io_ret
= lttng_write(new_fd
, buf
, count
);
1811 if (io_ret
< (ssize_t
) count
) {
1812 char error_string
[256];
1814 snprintf(error_string
, sizeof(error_string
),
1815 "Failed to write %" PRIu64
" bytes from fd %i in rotate_truncate_stream(), returned %zi",
1816 count
, new_fd
, io_ret
);
1818 PERROR("%s", error_string
);
1820 ERR("%s", error_string
);
1829 stream_fd_put_fd(new_stream_fd
);
1832 /* Truncate the file to get rid of the excess data. */
1833 ret
= ftruncate(old_fd
, stream
->pos_after_last_complete_data_index
);
1835 PERROR("ftruncate");
1839 stream_fd_put_fd(stream
->stream_fd
);
1843 * Swap the old stream_fd with the new, releasing the stream's
1846 stream_fd_put(stream
->stream_fd
);
1847 stream
->stream_fd
= new_stream_fd
;
1849 ret
= create_rotate_index_file(stream
);
1851 ERR("Rotate stream index file");
1856 * Update the offset and FD of all the eventual indexes created by the
1857 * data connection before the rotation command arrived.
1859 ret
= relay_index_switch_all_files(stream
);
1861 ERR("Failed to rotate index file");
1865 stream
->tracefile_size_current
= diff
;
1866 stream
->pos_after_last_complete_data_index
= 0;
1867 stream
->rotate_at_seq_num
= -1ULL;
1873 stream_fd_put_fd(stream
->stream_fd
);
1876 stream_fd_put_fd(new_stream_fd
);
1883 * Check if a stream should perform a rotation (for session rotation).
1884 * Must be called with the stream lock held.
1886 * Return 0 on success, a negative value on error.
1889 int try_rotate_stream(struct relay_stream
*stream
)
1893 /* No rotation expected. */
1894 if (stream
->rotate_at_seq_num
== -1ULL) {
1898 if (stream
->prev_seq
< stream
->rotate_at_seq_num
||
1899 stream
->prev_seq
== -1ULL) {
1900 DBG("Stream %" PRIu64
" no yet ready for rotation",
1901 stream
->stream_handle
);
1903 } else if (stream
->prev_seq
> stream
->rotate_at_seq_num
) {
1904 DBG("Rotation after too much data has been written in tracefile "
1905 "for stream %" PRIu64
", need to truncate before "
1906 "rotating", stream
->stream_handle
);
1907 ret
= rotate_truncate_stream(stream
);
1909 ERR("Failed to truncate stream");
1913 /* stream->prev_seq == stream->rotate_at_seq_num */
1914 DBG("Stream %" PRIu64
" ready for rotation",
1915 stream
->stream_handle
);
1916 ret
= do_rotate_stream(stream
);
1924 * relay_recv_metadata: receive the metadata for the session.
1926 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1927 struct relay_connection
*conn
,
1928 const struct lttng_buffer_view
*payload
)
1932 struct relay_session
*session
= conn
->session
;
1933 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1934 struct relay_stream
*metadata_stream
;
1935 uint64_t metadata_payload_size
;
1936 int metadata_fd
= -1;
1939 ERR("Metadata sent before version check");
1944 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1945 ERR("Incorrect data size");
1949 metadata_payload_size
= recv_hdr
->data_size
-
1950 sizeof(struct lttcomm_relayd_metadata_payload
);
1952 memcpy(&metadata_payload_header
, payload
->data
,
1953 sizeof(metadata_payload_header
));
1954 metadata_payload_header
.stream_id
= be64toh(
1955 metadata_payload_header
.stream_id
);
1956 metadata_payload_header
.padding_size
= be32toh(
1957 metadata_payload_header
.padding_size
);
1959 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1960 if (!metadata_stream
) {
1965 pthread_mutex_lock(&metadata_stream
->lock
);
1967 metadata_fd
= stream_fd_get_fd(metadata_stream
->stream_fd
);
1968 if (metadata_fd
< 0) {
1971 size_ret
= lttng_write(metadata_fd
,
1972 payload
->data
+ sizeof(metadata_payload_header
),
1973 metadata_payload_size
);
1974 if (size_ret
< metadata_payload_size
) {
1975 ERR("Relay error writing metadata on file");
1980 size_ret
= write_padding_to_file(metadata_fd
,
1981 metadata_payload_header
.padding_size
);
1982 if (size_ret
< (int64_t) metadata_payload_header
.padding_size
) {
1987 metadata_stream
->metadata_received
+=
1988 metadata_payload_size
+ metadata_payload_header
.padding_size
;
1989 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1990 metadata_stream
->metadata_received
);
1992 ret
= try_rotate_stream(metadata_stream
);
1998 stream_fd_put_fd(metadata_stream
->stream_fd
);
2000 pthread_mutex_unlock(&metadata_stream
->lock
);
2001 stream_put(metadata_stream
);
2007 * relay_send_version: send relayd version number
2009 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
2010 struct relay_connection
*conn
,
2011 const struct lttng_buffer_view
*payload
)
2015 struct lttcomm_relayd_version reply
, msg
;
2016 bool compatible
= true;
2018 conn
->version_check_done
= true;
2020 /* Get version from the other side. */
2021 if (payload
->size
< sizeof(msg
)) {
2022 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
2023 sizeof(msg
), payload
->size
);
2028 memcpy(&msg
, payload
->data
, sizeof(msg
));
2029 msg
.major
= be32toh(msg
.major
);
2030 msg
.minor
= be32toh(msg
.minor
);
2032 memset(&reply
, 0, sizeof(reply
));
2033 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
2034 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
2036 /* Major versions must be the same */
2037 if (reply
.major
!= msg
.major
) {
2038 DBG("Incompatible major versions (%u vs %u), deleting session",
2039 reply
.major
, msg
.major
);
2043 conn
->major
= reply
.major
;
2044 /* We adapt to the lowest compatible version */
2045 if (reply
.minor
<= msg
.minor
) {
2046 conn
->minor
= reply
.minor
;
2048 conn
->minor
= msg
.minor
;
2051 reply
.major
= htobe32(reply
.major
);
2052 reply
.minor
= htobe32(reply
.minor
);
2053 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2055 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2056 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2069 DBG("Version check done using protocol %u.%u", conn
->major
,
2077 * Check for data pending for a given stream id from the session daemon.
2079 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2080 struct relay_connection
*conn
,
2081 const struct lttng_buffer_view
*payload
)
2083 struct relay_session
*session
= conn
->session
;
2084 struct lttcomm_relayd_data_pending msg
;
2085 struct lttcomm_relayd_generic_reply reply
;
2086 struct relay_stream
*stream
;
2090 DBG("Data pending command received");
2092 if (!session
|| !conn
->version_check_done
) {
2093 ERR("Trying to check for data before version check");
2095 goto end_no_session
;
2098 if (payload
->size
< sizeof(msg
)) {
2099 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2100 sizeof(msg
), payload
->size
);
2102 goto end_no_session
;
2104 memcpy(&msg
, payload
->data
, sizeof(msg
));
2105 msg
.stream_id
= be64toh(msg
.stream_id
);
2106 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
2108 stream
= stream_get_by_id(msg
.stream_id
);
2109 if (stream
== NULL
) {
2114 pthread_mutex_lock(&stream
->lock
);
2116 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
2117 " and last_seq %" PRIu64
, msg
.stream_id
,
2118 stream
->prev_seq
, msg
.last_net_seq_num
);
2120 /* Avoid wrapping issue */
2121 if (((int64_t) (stream
->prev_seq
- msg
.last_net_seq_num
)) >= 0) {
2122 /* Data has in fact been written and is NOT pending */
2125 /* Data still being streamed thus pending */
2129 stream
->data_pending_check_done
= true;
2130 pthread_mutex_unlock(&stream
->lock
);
2135 memset(&reply
, 0, sizeof(reply
));
2136 reply
.ret_code
= htobe32(ret
);
2137 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2138 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2139 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2149 * Wait for the control socket to reach a quiescent state.
2151 * Note that for now, when receiving this command from the session
2152 * daemon, this means that every subsequent commands or data received on
2153 * the control socket has been handled. So, this is why we simply return
2156 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
2157 struct relay_connection
*conn
,
2158 const struct lttng_buffer_view
*payload
)
2162 struct relay_stream
*stream
;
2163 struct lttcomm_relayd_quiescent_control msg
;
2164 struct lttcomm_relayd_generic_reply reply
;
2166 DBG("Checking quiescent state on control socket");
2168 if (!conn
->session
|| !conn
->version_check_done
) {
2169 ERR("Trying to check for data before version check");
2171 goto end_no_session
;
2174 if (payload
->size
< sizeof(msg
)) {
2175 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2176 sizeof(msg
), payload
->size
);
2178 goto end_no_session
;
2180 memcpy(&msg
, payload
->data
, sizeof(msg
));
2181 msg
.stream_id
= be64toh(msg
.stream_id
);
2183 stream
= stream_get_by_id(msg
.stream_id
);
2187 pthread_mutex_lock(&stream
->lock
);
2188 stream
->data_pending_check_done
= true;
2189 pthread_mutex_unlock(&stream
->lock
);
2191 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
2194 memset(&reply
, 0, sizeof(reply
));
2195 reply
.ret_code
= htobe32(LTTNG_OK
);
2196 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2197 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2198 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2210 * Initialize a data pending command. This means that a consumer is about
2211 * to ask for data pending for each stream it holds. Simply iterate over
2212 * all streams of a session and set the data_pending_check_done flag.
2214 * This command returns to the client a LTTNG_OK code.
2216 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2217 struct relay_connection
*conn
,
2218 const struct lttng_buffer_view
*payload
)
2222 struct lttng_ht_iter iter
;
2223 struct lttcomm_relayd_begin_data_pending msg
;
2224 struct lttcomm_relayd_generic_reply reply
;
2225 struct relay_stream
*stream
;
2230 DBG("Init streams for data pending");
2232 if (!conn
->session
|| !conn
->version_check_done
) {
2233 ERR("Trying to check for data before version check");
2235 goto end_no_session
;
2238 if (payload
->size
< sizeof(msg
)) {
2239 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2240 sizeof(msg
), payload
->size
);
2242 goto end_no_session
;
2244 memcpy(&msg
, payload
->data
, sizeof(msg
));
2245 msg
.session_id
= be64toh(msg
.session_id
);
2248 * Iterate over all streams to set the begin data pending flag.
2249 * For now, the streams are indexed by stream handle so we have
2250 * to iterate over all streams to find the one associated with
2251 * the right session_id.
2254 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2256 if (!stream_get(stream
)) {
2259 if (stream
->trace
->session
->id
== msg
.session_id
) {
2260 pthread_mutex_lock(&stream
->lock
);
2261 stream
->data_pending_check_done
= false;
2262 pthread_mutex_unlock(&stream
->lock
);
2263 DBG("Set begin data pending flag to stream %" PRIu64
,
2264 stream
->stream_handle
);
2270 memset(&reply
, 0, sizeof(reply
));
2271 /* All good, send back reply. */
2272 reply
.ret_code
= htobe32(LTTNG_OK
);
2274 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2275 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2276 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2288 * End data pending command. This will check, for a given session id, if
2289 * each stream associated with it has its data_pending_check_done flag
2290 * set. If not, this means that the client lost track of the stream but
2291 * the data is still being streamed on our side. In this case, we inform
2292 * the client that data is in flight.
2294 * Return to the client if there is data in flight or not with a ret_code.
2296 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2297 struct relay_connection
*conn
,
2298 const struct lttng_buffer_view
*payload
)
2302 struct lttng_ht_iter iter
;
2303 struct lttcomm_relayd_end_data_pending msg
;
2304 struct lttcomm_relayd_generic_reply reply
;
2305 struct relay_stream
*stream
;
2306 uint32_t is_data_inflight
= 0;
2308 DBG("End data pending command");
2310 if (!conn
->session
|| !conn
->version_check_done
) {
2311 ERR("Trying to check for data before version check");
2313 goto end_no_session
;
2316 if (payload
->size
< sizeof(msg
)) {
2317 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2318 sizeof(msg
), payload
->size
);
2320 goto end_no_session
;
2322 memcpy(&msg
, payload
->data
, sizeof(msg
));
2323 msg
.session_id
= be64toh(msg
.session_id
);
2326 * Iterate over all streams to see if the begin data pending
2330 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2332 if (!stream_get(stream
)) {
2335 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2339 pthread_mutex_lock(&stream
->lock
);
2340 if (!stream
->data_pending_check_done
) {
2341 if (!stream
->closed
|| !(((int64_t) (stream
->prev_seq
- stream
->last_net_seq_num
)) >= 0)) {
2342 is_data_inflight
= 1;
2343 DBG("Data is still in flight for stream %" PRIu64
,
2344 stream
->stream_handle
);
2345 pthread_mutex_unlock(&stream
->lock
);
2350 pthread_mutex_unlock(&stream
->lock
);
2355 memset(&reply
, 0, sizeof(reply
));
2356 /* All good, send back reply. */
2357 reply
.ret_code
= htobe32(is_data_inflight
);
2359 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2360 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2361 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2373 * Receive an index for a specific stream.
2375 * Return 0 on success else a negative value.
2377 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2378 struct relay_connection
*conn
,
2379 const struct lttng_buffer_view
*payload
)
2383 struct relay_session
*session
= conn
->session
;
2384 struct lttcomm_relayd_index index_info
;
2385 struct relay_index
*index
;
2386 struct lttcomm_relayd_generic_reply reply
;
2387 struct relay_stream
*stream
;
2392 DBG("Relay receiving index");
2394 if (!session
|| !conn
->version_check_done
) {
2395 ERR("Trying to close a stream before version check");
2397 goto end_no_session
;
2400 msg_len
= lttcomm_relayd_index_len(
2401 lttng_to_index_major(conn
->major
, conn
->minor
),
2402 lttng_to_index_minor(conn
->major
, conn
->minor
));
2403 if (payload
->size
< msg_len
) {
2404 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2405 msg_len
, payload
->size
);
2407 goto end_no_session
;
2409 memcpy(&index_info
, payload
->data
, msg_len
);
2410 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2411 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2412 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2413 index_info
.content_size
= be64toh(index_info
.content_size
);
2414 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2415 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2416 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2417 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2419 if (conn
->minor
>= 8) {
2420 index_info
.stream_instance_id
=
2421 be64toh(index_info
.stream_instance_id
);
2422 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2425 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2427 ERR("stream_get_by_id not found");
2431 pthread_mutex_lock(&stream
->lock
);
2433 /* Live beacon handling */
2434 if (index_info
.packet_size
== 0) {
2435 DBG("Received live beacon for stream %" PRIu64
,
2436 stream
->stream_handle
);
2439 * Only flag a stream inactive when it has already
2440 * received data and no indexes are in flight.
2442 if (stream
->index_received_seqcount
> 0
2443 && stream
->indexes_in_flight
== 0) {
2444 stream
->beacon_ts_end
= index_info
.timestamp_end
;
2447 goto end_stream_put
;
2449 stream
->beacon_ts_end
= -1ULL;
2452 if (stream
->ctf_stream_id
== -1ULL) {
2453 stream
->ctf_stream_id
= index_info
.stream_id
;
2455 index
= relay_index_get_by_id_or_create(stream
, index_info
.net_seq_num
);
2458 ERR("relay_index_get_by_id_or_create index NULL");
2459 goto end_stream_put
;
2461 if (set_index_control_data(index
, &index_info
, conn
)) {
2462 ERR("set_index_control_data error");
2463 relay_index_put(index
);
2465 goto end_stream_put
;
2467 ret
= relay_index_try_flush(index
);
2469 tracefile_array_commit_seq(stream
->tfa
);
2470 stream
->index_received_seqcount
++;
2471 stream
->pos_after_last_complete_data_index
+= index
->total_size
;
2472 } else if (ret
> 0) {
2476 ERR("relay_index_try_flush error %d", ret
);
2477 relay_index_put(index
);
2482 pthread_mutex_unlock(&stream
->lock
);
2487 memset(&reply
, 0, sizeof(reply
));
2489 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2491 reply
.ret_code
= htobe32(LTTNG_OK
);
2493 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2494 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2495 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2504 * Receive the streams_sent message.
2506 * Return 0 on success else a negative value.
2508 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2509 struct relay_connection
*conn
,
2510 const struct lttng_buffer_view
*payload
)
2514 struct lttcomm_relayd_generic_reply reply
;
2518 DBG("Relay receiving streams_sent");
2520 if (!conn
->session
|| !conn
->version_check_done
) {
2521 ERR("Trying to close a stream before version check");
2523 goto end_no_session
;
2527 * Publish every pending stream in the connection recv list which are
2528 * now ready to be used by the viewer.
2530 publish_connection_local_streams(conn
);
2532 memset(&reply
, 0, sizeof(reply
));
2533 reply
.ret_code
= htobe32(LTTNG_OK
);
2534 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2535 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2536 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2549 * relay_rotate_session_stream: rotate a stream to a new tracefile for the session
2550 * rotation feature (not the tracefile rotation feature).
2552 static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
2553 struct relay_connection
*conn
,
2554 const struct lttng_buffer_view
*payload
)
2558 struct relay_session
*session
= conn
->session
;
2559 struct lttcomm_relayd_rotate_stream stream_info
;
2560 struct lttcomm_relayd_generic_reply reply
;
2561 struct relay_stream
*stream
;
2564 struct lttng_buffer_view new_path_view
;
2566 DBG("Rotate stream received");
2568 if (!session
|| !conn
->version_check_done
) {
2569 ERR("Trying to rotate a stream before version check");
2574 if (session
->major
== 2 && session
->minor
< 11) {
2575 ERR("Unsupported feature before 2.11");
2580 header_len
= sizeof(struct lttcomm_relayd_rotate_stream
);
2582 if (payload
->size
< header_len
) {
2583 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2584 header_len
, payload
->size
);
2589 memcpy(&stream_info
, payload
->data
, header_len
);
2591 /* Convert to host */
2592 stream_info
.pathname_length
= be32toh(stream_info
.pathname_length
);
2593 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
2594 stream_info
.new_chunk_id
= be64toh(stream_info
.new_chunk_id
);
2595 stream_info
.rotate_at_seq_num
= be64toh(stream_info
.rotate_at_seq_num
);
2597 path_len
= stream_info
.pathname_length
;
2598 if (payload
->size
< header_len
+ path_len
) {
2599 ERR("Unexpected payload size in \"relay_rotate_session_stream\" including path: expected >= %zu bytes, got %zu bytes",
2600 header_len
+ path_len
, payload
->size
);
2605 /* Ensure it fits in local filename length. */
2606 if (path_len
>= LTTNG_PATH_MAX
) {
2607 ret
= -ENAMETOOLONG
;
2608 ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes",
2609 path_len
, LTTNG_PATH_MAX
);
2613 new_path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2614 stream_info
.pathname_length
);
2616 stream
= stream_get_by_id(stream_info
.stream_id
);
2622 pthread_mutex_lock(&stream
->lock
);
2625 * Update the trace path (just the folder, the stream name does not
2628 free(stream
->path_name
);
2629 stream
->path_name
= create_output_path(new_path_view
.data
);
2630 if (!stream
->path_name
) {
2631 ERR("Failed to create a new output path");
2633 goto end_stream_unlock
;
2635 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
,
2638 ERR("relay creating output directory");
2640 goto end_stream_unlock
;
2643 assert(stream
->current_chunk_id
.is_set
);
2644 stream
->current_chunk_id
.value
= stream_info
.new_chunk_id
;
2646 if (stream
->is_metadata
) {
2648 * The metadata stream is sent only over the control connection
2649 * so we know we have all the data to perform the stream
2652 ret
= do_rotate_stream(stream
);
2654 stream
->rotate_at_seq_num
= stream_info
.rotate_at_seq_num
;
2655 ret
= try_rotate_stream(stream
);
2658 goto end_stream_unlock
;
2662 pthread_mutex_unlock(&stream
->lock
);
2665 memset(&reply
, 0, sizeof(reply
));
2667 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2669 reply
.ret_code
= htobe32(LTTNG_OK
);
2671 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2672 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2673 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2674 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2684 * relay_mkdir: Create a folder on the disk.
2686 static int relay_mkdir(const struct lttcomm_relayd_hdr
*recv_hdr
,
2687 struct relay_connection
*conn
,
2688 const struct lttng_buffer_view
*payload
)
2691 struct relay_session
*session
= conn
->session
;
2692 struct lttcomm_relayd_mkdir path_info_header
;
2693 struct lttcomm_relayd_generic_reply reply
;
2697 struct lttng_buffer_view path_view
;
2699 if (!session
|| !conn
->version_check_done
) {
2700 ERR("Trying to create a directory before version check");
2702 goto end_no_session
;
2705 if (session
->major
== 2 && session
->minor
< 11) {
2707 * This client is not supposed to use this command since
2708 * it predates its introduction.
2710 ERR("relay_mkdir command is unsupported before LTTng 2.11");
2712 goto end_no_session
;
2715 header_len
= sizeof(path_info_header
);
2716 if (payload
->size
< header_len
) {
2717 ERR("Unexpected payload size in \"relay_mkdir\": expected >= %zu bytes, got %zu bytes",
2718 header_len
, payload
->size
);
2720 goto end_no_session
;
2723 memcpy(&path_info_header
, payload
->data
, header_len
);
2725 path_info_header
.length
= be32toh(path_info_header
.length
);
2727 if (payload
->size
< header_len
+ path_info_header
.length
) {
2728 ERR("Unexpected payload size in \"relay_mkdir\" including path: expected >= %zu bytes, got %zu bytes",
2729 header_len
+ path_info_header
.length
, payload
->size
);
2731 goto end_no_session
;
2734 /* Ensure that it fits in local path length. */
2735 if (path_info_header
.length
>= LTTNG_PATH_MAX
) {
2736 ret
= -ENAMETOOLONG
;
2737 ERR("Path name argument of mkdir command (%" PRIu32
" bytes) exceeds the maximal length allowed (%d bytes)",
2738 path_info_header
.length
, LTTNG_PATH_MAX
);
2742 path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2743 path_info_header
.length
);
2745 path
= create_output_path(path_view
.data
);
2747 ERR("Failed to create output path");
2752 ret
= utils_mkdir_recursive(path
, S_IRWXU
| S_IRWXG
, -1, -1);
2754 ERR("relay creating output directory");
2761 memset(&reply
, 0, sizeof(reply
));
2763 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2765 reply
.ret_code
= htobe32(LTTNG_OK
);
2767 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2768 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2769 ERR("Failed to send \"mkdir\" command reply (ret = %zd)", send_ret
);
2778 static int validate_rotate_rename_path_length(const char *path_type
,
2779 uint32_t path_length
)
2783 if (path_length
> LTTNG_PATH_MAX
) {
2784 ret
= -ENAMETOOLONG
;
2785 ERR("rotate rename \"%s\" path name length (%" PRIu32
" bytes) exceeds the allowed size of %i bytes",
2786 path_type
, path_length
, LTTNG_PATH_MAX
);
2787 } else if (path_length
== 0) {
2789 ERR("rotate rename \"%s\" path name has an illegal length of 0", path_type
);
2795 * relay_rotate_rename: rename the trace folder after a rotation is
2796 * completed. We are not closing any fd here, just moving the folder, so it
2797 * works even if data is still in-flight.
2799 static int relay_rotate_rename(const struct lttcomm_relayd_hdr
*recv_hdr
,
2800 struct relay_connection
*conn
,
2801 const struct lttng_buffer_view
*payload
)
2805 struct relay_session
*session
= conn
->session
;
2806 struct lttcomm_relayd_generic_reply reply
;
2807 struct lttcomm_relayd_rotate_rename header
;
2809 size_t received_paths_size
;
2810 char *complete_old_path
= NULL
, *complete_new_path
= NULL
;
2811 struct lttng_buffer_view old_path_view
;
2812 struct lttng_buffer_view new_path_view
;
2814 if (!session
|| !conn
->version_check_done
) {
2815 ERR("Trying to rename a trace folder before version check");
2820 if (session
->major
== 2 && session
->minor
< 11) {
2821 ERR("relay_rotate_rename command is unsupported before LTTng 2.11");
2826 header_len
= sizeof(header
);
2827 if (payload
->size
< header_len
) {
2828 ERR("Unexpected payload size in \"relay_rotate_rename\": expected >= %zu bytes, got %zu bytes",
2829 header_len
, payload
->size
);
2834 memcpy(&header
, payload
->data
, header_len
);
2836 header
.old_path_length
= be32toh(header
.old_path_length
);
2837 header
.new_path_length
= be32toh(header
.new_path_length
);
2838 received_paths_size
= header
.old_path_length
+ header
.new_path_length
;
2840 if (payload
->size
< header_len
+ received_paths_size
) {
2841 ERR("Unexpected payload size in \"relay_rotate_rename\" including paths: expected >= %zu bytes, got %zu bytes",
2842 header_len
, payload
->size
);
2847 /* Ensure the paths don't exceed their allowed size. */
2848 ret
= validate_rotate_rename_path_length("old", header
.old_path_length
);
2852 ret
= validate_rotate_rename_path_length("new", header
.new_path_length
);
2857 old_path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2858 header
.old_path_length
);
2859 new_path_view
= lttng_buffer_view_from_view(payload
,
2860 header_len
+ header
.old_path_length
,
2861 header
.new_path_length
);
2863 /* Validate that both paths received are NULL terminated. */
2864 if (old_path_view
.data
[old_path_view
.size
- 1] != '\0') {
2865 ERR("relay_rotate_rename command's \"old\" path is invalid (not NULL terminated)");
2869 if (new_path_view
.data
[new_path_view
.size
- 1] != '\0') {
2870 ERR("relay_rotate_rename command's \"new\" path is invalid (not NULL terminated)");
2875 complete_old_path
= create_output_path(old_path_view
.data
);
2876 if (!complete_old_path
) {
2877 ERR("Failed to build old output path in rotate_rename command");
2882 complete_new_path
= create_output_path(new_path_view
.data
);
2883 if (!complete_new_path
) {
2884 ERR("Failed to build new output path in rotate_rename command");
2889 ret
= utils_mkdir_recursive(complete_new_path
, S_IRWXU
| S_IRWXG
,
2892 ERR("Failed to mkdir() rotate_rename's \"new\" output directory at \"%s\"",
2898 * If a domain has not yet created its channel, the domain-specific
2899 * folder might not exist, but this is not an error.
2901 ret
= rename(complete_old_path
, complete_new_path
);
2902 if (ret
< 0 && errno
!= ENOENT
) {
2903 PERROR("Renaming chunk in rotate_rename command from \"%s\" to \"%s\"",
2904 complete_old_path
, complete_new_path
);
2910 memset(&reply
, 0, sizeof(reply
));
2912 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2914 reply
.ret_code
= htobe32(LTTNG_OK
);
2916 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2918 if (send_ret
< sizeof(reply
)) {
2919 ERR("Failed to send \"rotate rename\" command reply (ret = %zd)",
2925 free(complete_old_path
);
2926 free(complete_new_path
);
2931 * Check if all the streams in the session have completed the last rotation.
2932 * The chunk_id value is used to distinguish the cases where a stream was
2933 * closed on the consumerd before the rotation started but it still active on
2934 * the relayd, and the case where a stream appeared on the consumerd/relayd
2935 * after the last rotation started (in that case, it is already writing in the
2936 * new chunk folder).
2939 int relay_rotate_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2940 struct relay_connection
*conn
,
2941 const struct lttng_buffer_view
*payload
)
2943 struct relay_session
*session
= conn
->session
;
2944 struct lttcomm_relayd_rotate_pending msg
;
2945 struct lttcomm_relayd_rotate_pending_reply reply
;
2946 struct lttng_ht_iter iter
;
2947 struct relay_stream
*stream
;
2951 bool rotate_pending
= false;
2953 DBG("Rotate pending command received");
2955 if (!session
|| !conn
->version_check_done
) {
2956 ERR("Trying to check for data before version check");
2961 if (session
->major
== 2 && session
->minor
< 11) {
2962 ERR("Unsupported feature before 2.11");
2967 if (payload
->size
< sizeof(msg
)) {
2968 ERR("Unexpected payload size in \"relay_rotate_pending\": expected >= %zu bytes, got %zu bytes",
2969 sizeof(msg
), payload
->size
);
2974 memcpy(&msg
, payload
->data
, sizeof(msg
));
2976 chunk_id
= be64toh(msg
.chunk_id
);
2978 DBG("Evaluating rotate pending for session \"%s\" and chunk id %" PRIu64
,
2979 session
->session_name
, chunk_id
);
2982 * Iterate over all the streams in the session and check if they are
2983 * still waiting for data to perform their rotation.
2986 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2988 if (!stream_get(stream
)) {
2991 if (stream
->trace
->session
!= session
) {
2995 pthread_mutex_lock(&stream
->lock
);
2996 if (stream
->rotate_at_seq_num
!= -1ULL) {
2997 /* We have not yet performed the rotation. */
2998 rotate_pending
= true;
2999 DBG("Stream %" PRIu64
" is still rotating",
3000 stream
->stream_handle
);
3001 } else if (stream
->current_chunk_id
.value
< chunk_id
) {
3003 * Stream closed on the consumer but still active on the
3006 rotate_pending
= true;
3007 DBG("Stream %" PRIu64
" did not exist on the consumer "
3008 "when the last rotation started, but is"
3009 "still waiting for data before getting"
3011 stream
->stream_handle
);
3013 pthread_mutex_unlock(&stream
->lock
);
3015 if (rotate_pending
) {
3022 memset(&reply
, 0, sizeof(reply
));
3023 reply
.generic
.ret_code
= htobe32((uint32_t) LTTNG_OK
);
3024 reply
.is_pending
= (uint8_t) !!rotate_pending
;
3025 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
3027 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3028 ERR("Failed to send \"rotate pending\" command reply (ret = %zd)",
3037 #define DBG_CMD(cmd_name, conn) \
3038 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
3040 static int relay_process_control_command(struct relay_connection
*conn
,
3041 const struct lttcomm_relayd_hdr
*header
,
3042 const struct lttng_buffer_view
*payload
)
3046 switch (header
->cmd
) {
3047 case RELAYD_CREATE_SESSION
:
3048 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
3049 ret
= relay_create_session(header
, conn
, payload
);
3051 case RELAYD_ADD_STREAM
:
3052 DBG_CMD("RELAYD_ADD_STREAM", conn
);
3053 ret
= relay_add_stream(header
, conn
, payload
);
3055 case RELAYD_START_DATA
:
3056 DBG_CMD("RELAYD_START_DATA", conn
);
3057 ret
= relay_start(header
, conn
, payload
);
3059 case RELAYD_SEND_METADATA
:
3060 DBG_CMD("RELAYD_SEND_METADATA", conn
);
3061 ret
= relay_recv_metadata(header
, conn
, payload
);
3063 case RELAYD_VERSION
:
3064 DBG_CMD("RELAYD_VERSION", conn
);
3065 ret
= relay_send_version(header
, conn
, payload
);
3067 case RELAYD_CLOSE_STREAM
:
3068 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
3069 ret
= relay_close_stream(header
, conn
, payload
);
3071 case RELAYD_DATA_PENDING
:
3072 DBG_CMD("RELAYD_DATA_PENDING", conn
);
3073 ret
= relay_data_pending(header
, conn
, payload
);
3075 case RELAYD_QUIESCENT_CONTROL
:
3076 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
3077 ret
= relay_quiescent_control(header
, conn
, payload
);
3079 case RELAYD_BEGIN_DATA_PENDING
:
3080 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
3081 ret
= relay_begin_data_pending(header
, conn
, payload
);
3083 case RELAYD_END_DATA_PENDING
:
3084 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
3085 ret
= relay_end_data_pending(header
, conn
, payload
);
3087 case RELAYD_SEND_INDEX
:
3088 DBG_CMD("RELAYD_SEND_INDEX", conn
);
3089 ret
= relay_recv_index(header
, conn
, payload
);
3091 case RELAYD_STREAMS_SENT
:
3092 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
3093 ret
= relay_streams_sent(header
, conn
, payload
);
3095 case RELAYD_RESET_METADATA
:
3096 DBG_CMD("RELAYD_RESET_METADATA", conn
);
3097 ret
= relay_reset_metadata(header
, conn
, payload
);
3099 case RELAYD_ROTATE_STREAM
:
3100 DBG_CMD("RELAYD_ROTATE_STREAM", conn
);
3101 ret
= relay_rotate_session_stream(header
, conn
, payload
);
3103 case RELAYD_ROTATE_RENAME
:
3104 DBG_CMD("RELAYD_ROTATE_RENAME", conn
);
3105 ret
= relay_rotate_rename(header
, conn
, payload
);
3107 case RELAYD_ROTATE_PENDING
:
3108 DBG_CMD("RELAYD_ROTATE_PENDING", conn
);
3109 ret
= relay_rotate_pending(header
, conn
, payload
);
3112 DBG_CMD("RELAYD_MKDIR", conn
);
3113 ret
= relay_mkdir(header
, conn
, payload
);
3115 case RELAYD_UPDATE_SYNC_INFO
:
3117 ERR("Received unknown command (%u)", header
->cmd
);
3118 relay_unknown_command(conn
);
3127 static enum relay_connection_status
relay_process_control_receive_payload(
3128 struct relay_connection
*conn
)
3131 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3132 struct lttng_dynamic_buffer
*reception_buffer
=
3133 &conn
->protocol
.ctrl
.reception_buffer
;
3134 struct ctrl_connection_state_receive_payload
*state
=
3135 &conn
->protocol
.ctrl
.state
.receive_payload
;
3136 struct lttng_buffer_view payload_view
;
3138 if (state
->left_to_receive
== 0) {
3139 /* Short-circuit for payload-less commands. */
3140 goto reception_complete
;
3143 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3144 reception_buffer
->data
+ state
->received
,
3145 state
->left_to_receive
, MSG_DONTWAIT
);
3147 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3148 PERROR("Unable to receive command payload on sock %d",
3150 status
= RELAY_CONNECTION_STATUS_ERROR
;
3153 } else if (ret
== 0) {
3154 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3155 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3160 assert(ret
<= state
->left_to_receive
);
3162 state
->left_to_receive
-= ret
;
3163 state
->received
+= ret
;
3165 if (state
->left_to_receive
> 0) {
3167 * Can't transition to the protocol's next state, wait to
3168 * receive the rest of the header.
3170 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3171 state
->received
, state
->left_to_receive
,
3177 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
3178 conn
->sock
->fd
, state
->received
);
3180 * The payload required to process the command has been received.
3181 * A view to the reception buffer is forwarded to the various
3182 * commands and the state of the control is reset on success.
3184 * Commands are responsible for sending their reply to the peer.
3186 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
3188 ret
= relay_process_control_command(conn
,
3189 &state
->header
, &payload_view
);
3191 status
= RELAY_CONNECTION_STATUS_ERROR
;
3195 ret
= connection_reset_protocol_state(conn
);
3197 status
= RELAY_CONNECTION_STATUS_ERROR
;
3203 static enum relay_connection_status
relay_process_control_receive_header(
3204 struct relay_connection
*conn
)
3207 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3208 struct lttcomm_relayd_hdr header
;
3209 struct lttng_dynamic_buffer
*reception_buffer
=
3210 &conn
->protocol
.ctrl
.reception_buffer
;
3211 struct ctrl_connection_state_receive_header
*state
=
3212 &conn
->protocol
.ctrl
.state
.receive_header
;
3214 assert(state
->left_to_receive
!= 0);
3216 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3217 reception_buffer
->data
+ state
->received
,
3218 state
->left_to_receive
, MSG_DONTWAIT
);
3220 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3221 PERROR("Unable to receive control command header on sock %d",
3223 status
= RELAY_CONNECTION_STATUS_ERROR
;
3226 } else if (ret
== 0) {
3227 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3228 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3233 assert(ret
<= state
->left_to_receive
);
3235 state
->left_to_receive
-= ret
;
3236 state
->received
+= ret
;
3238 if (state
->left_to_receive
> 0) {
3240 * Can't transition to the protocol's next state, wait to
3241 * receive the rest of the header.
3243 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3244 state
->received
, state
->left_to_receive
,
3249 /* Transition to next state: receiving the command's payload. */
3250 conn
->protocol
.ctrl
.state_id
=
3251 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3252 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
3253 header
.circuit_id
= be64toh(header
.circuit_id
);
3254 header
.data_size
= be64toh(header
.data_size
);
3255 header
.cmd
= be32toh(header
.cmd
);
3256 header
.cmd_version
= be32toh(header
.cmd_version
);
3257 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
3258 &header
, sizeof(header
));
3260 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
3261 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
3264 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
3265 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
3267 status
= RELAY_CONNECTION_STATUS_ERROR
;
3271 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
3273 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
3274 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
3277 status
= RELAY_CONNECTION_STATUS_ERROR
;
3281 if (header
.data_size
== 0) {
3283 * Manually invoke the next state as the poll loop
3284 * will not wake-up to allow us to proceed further.
3286 status
= relay_process_control_receive_payload(conn
);
3293 * Process the commands received on the control socket
3295 static enum relay_connection_status
relay_process_control(
3296 struct relay_connection
*conn
)
3298 enum relay_connection_status status
;
3300 switch (conn
->protocol
.ctrl
.state_id
) {
3301 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
3302 status
= relay_process_control_receive_header(conn
);
3304 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3305 status
= relay_process_control_receive_payload(conn
);
3308 ERR("Unknown control connection protocol state encountered.");
3316 * Handle index for a data stream.
3318 * Called with the stream lock held.
3320 * Return 0 on success else a negative value.
3322 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
3323 bool rotate_index
, bool *flushed
, uint64_t total_size
)
3326 uint64_t data_offset
;
3327 struct relay_index
*index
;
3329 /* Get data offset because we are about to update the index. */
3330 data_offset
= htobe64(stream
->tracefile_size_current
);
3332 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
3333 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
3336 * Lookup for an existing index for that stream id/sequence
3337 * number. If it exists, the control thread has already received the
3338 * data for it, thus we need to write it to disk.
3340 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
3346 if (rotate_index
|| !stream
->index_file
) {
3347 ret
= create_rotate_index_file(stream
);
3349 ERR("Failed to rotate index");
3350 /* Put self-ref for this index due to error. */
3351 relay_index_put(index
);
3357 if (relay_index_set_file(index
, stream
->index_file
, data_offset
)) {
3359 /* Put self-ref for this index due to error. */
3360 relay_index_put(index
);
3365 ret
= relay_index_try_flush(index
);
3367 tracefile_array_commit_seq(stream
->tfa
);
3368 stream
->index_received_seqcount
++;
3370 } else if (ret
> 0) {
3371 index
->total_size
= total_size
;
3375 /* Put self-ref for this index due to error. */
3376 relay_index_put(index
);
3384 static enum relay_connection_status
relay_process_data_receive_header(
3385 struct relay_connection
*conn
)
3388 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3389 struct data_connection_state_receive_header
*state
=
3390 &conn
->protocol
.data
.state
.receive_header
;
3391 struct lttcomm_relayd_data_hdr header
;
3392 struct relay_stream
*stream
;
3394 assert(state
->left_to_receive
!= 0);
3396 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3397 state
->header_reception_buffer
+ state
->received
,
3398 state
->left_to_receive
, MSG_DONTWAIT
);
3400 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3401 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
3402 status
= RELAY_CONNECTION_STATUS_ERROR
;
3405 } else if (ret
== 0) {
3406 /* Orderly shutdown. Not necessary to print an error. */
3407 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3408 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3413 assert(ret
<= state
->left_to_receive
);
3415 state
->left_to_receive
-= ret
;
3416 state
->received
+= ret
;
3418 if (state
->left_to_receive
> 0) {
3420 * Can't transition to the protocol's next state, wait to
3421 * receive the rest of the header.
3423 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3424 state
->received
, state
->left_to_receive
,
3429 /* Transition to next state: receiving the payload. */
3430 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3432 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3433 header
.circuit_id
= be64toh(header
.circuit_id
);
3434 header
.stream_id
= be64toh(header
.stream_id
);
3435 header
.data_size
= be32toh(header
.data_size
);
3436 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3437 header
.padding_size
= be32toh(header
.padding_size
);
3438 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3440 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3442 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3443 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3445 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3446 conn
->sock
->fd
, header
.circuit_id
,
3447 header
.stream_id
, header
.data_size
,
3448 header
.net_seq_num
, header
.padding_size
);
3450 stream
= stream_get_by_id(header
.stream_id
);
3452 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3454 /* Protocol error. */
3455 status
= RELAY_CONNECTION_STATUS_ERROR
;
3459 pthread_mutex_lock(&stream
->lock
);
3461 /* Check if a rotation is needed. */
3462 if (stream
->tracefile_size
> 0 &&
3463 (stream
->tracefile_size_current
+ header
.data_size
) >
3464 stream
->tracefile_size
) {
3465 uint64_t old_id
, new_id
;
3467 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
3468 tracefile_array_file_rotate(stream
->tfa
);
3470 /* new_id is updated by utils_rotate_stream_file. */
3473 ret
= stream_fd_rotate(stream
->stream_fd
, stream
->path_name
,
3474 stream
->channel_name
, stream
->tracefile_size
,
3475 stream
->tracefile_count
, &new_id
);
3477 ERR("Failed to rotate stream output file");
3478 status
= RELAY_CONNECTION_STATUS_ERROR
;
3479 goto end_stream_unlock
;
3483 * Reset current size because we just performed a stream
3486 stream
->tracefile_size_current
= 0;
3487 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= true;
3491 pthread_mutex_unlock(&stream
->lock
);
3497 static enum relay_connection_status
relay_process_data_receive_payload(
3498 struct relay_connection
*conn
)
3501 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3502 struct relay_stream
*stream
;
3503 struct data_connection_state_receive_payload
*state
=
3504 &conn
->protocol
.data
.state
.receive_payload
;
3505 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3506 char data_buffer
[chunk_size
];
3507 bool partial_recv
= false;
3508 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3509 uint64_t left_to_receive
= state
->left_to_receive
;
3510 struct relay_session
*session
;
3513 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3514 state
->header
.stream_id
, state
->header
.net_seq_num
,
3515 state
->received
, left_to_receive
);
3517 stream
= stream_get_by_id(state
->header
.stream_id
);
3519 /* Protocol error. */
3520 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3521 state
->header
.stream_id
);
3522 status
= RELAY_CONNECTION_STATUS_ERROR
;
3526 pthread_mutex_lock(&stream
->lock
);
3527 session
= stream
->trace
->session
;
3528 if (!conn
->session
) {
3529 ret
= connection_set_session(conn
, session
);
3531 status
= RELAY_CONNECTION_STATUS_ERROR
;
3532 goto end_stream_unlock
;
3536 stream_fd
= stream_fd_get_fd(stream
->stream_fd
);
3537 if (stream_fd
< 0) {
3538 status
= RELAY_CONNECTION_STATUS_ERROR
;
3539 goto end_stream_unlock
;
3543 * The size of the "chunk" received on any iteration is bounded by:
3544 * - the data left to receive,
3545 * - the data immediately available on the socket,
3546 * - the on-stack data buffer
3548 while (left_to_receive
> 0 && !partial_recv
) {
3550 size_t recv_size
= min(left_to_receive
, chunk_size
);
3552 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3553 recv_size
, MSG_DONTWAIT
);
3555 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3556 PERROR("Socket %d error", conn
->sock
->fd
);
3557 status
= RELAY_CONNECTION_STATUS_ERROR
;
3560 } else if (ret
== 0) {
3561 /* No more data ready to be consumed on socket. */
3562 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3563 state
->header
.stream_id
);
3564 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3566 } else if (ret
< (int) recv_size
) {
3568 * All the data available on the socket has been
3571 partial_recv
= true;
3576 /* Write data to stream output fd. */
3577 write_ret
= lttng_write(stream_fd
, data_buffer
,
3579 if (write_ret
< (ssize_t
) recv_size
) {
3580 ERR("Relay error writing data to file");
3581 status
= RELAY_CONNECTION_STATUS_ERROR
;
3585 left_to_receive
-= recv_size
;
3586 state
->received
+= recv_size
;
3587 state
->left_to_receive
= left_to_receive
;
3589 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
3590 write_ret
, stream
->stream_handle
);
3593 if (state
->left_to_receive
> 0) {
3595 * Did not receive all the data expected, wait for more data to
3596 * become available on the socket.
3598 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3599 state
->header
.stream_id
, state
->received
,
3600 state
->left_to_receive
);
3604 ret
= write_padding_to_file(stream_fd
,
3605 state
->header
.padding_size
);
3606 if ((int64_t) ret
< (int64_t) state
->header
.padding_size
) {
3607 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3608 stream
->stream_handle
,
3609 state
->header
.net_seq_num
, ret
);
3610 status
= RELAY_CONNECTION_STATUS_ERROR
;
3615 if (session
->minor
>= 4 && !session
->snapshot
) {
3616 ret
= handle_index_data(stream
, state
->header
.net_seq_num
,
3617 state
->rotate_index
, &index_flushed
, state
->header
.data_size
+ state
->header
.padding_size
);
3619 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3620 stream
->stream_handle
,
3621 state
->header
.net_seq_num
, ret
);
3622 status
= RELAY_CONNECTION_STATUS_ERROR
;
3627 stream
->tracefile_size_current
+= state
->header
.data_size
+
3628 state
->header
.padding_size
;
3630 if (stream
->prev_seq
== -1ULL) {
3633 if (index_flushed
) {
3634 stream
->pos_after_last_complete_data_index
=
3635 stream
->tracefile_size_current
;
3638 stream
->prev_seq
= state
->header
.net_seq_num
;
3641 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3642 * contents of *state which are aliased (union) to the same location as
3643 * the new state. Don't use it beyond this point.
3645 connection_reset_protocol_state(conn
);
3648 ret
= try_rotate_stream(stream
);
3650 status
= RELAY_CONNECTION_STATUS_ERROR
;
3654 (void) stream_fd_put_fd(stream
->stream_fd
);
3656 close_requested
= stream
->close_requested
;
3657 pthread_mutex_unlock(&stream
->lock
);
3658 if (close_requested
&& left_to_receive
== 0) {
3659 try_stream_close(stream
);
3663 pthread_mutex_lock(&session
->lock
);
3664 uatomic_set(&session
->new_streams
, 1);
3665 pthread_mutex_unlock(&session
->lock
);
3674 * relay_process_data: Process the data received on the data socket
3676 static enum relay_connection_status
relay_process_data(
3677 struct relay_connection
*conn
)
3679 enum relay_connection_status status
;
3681 switch (conn
->protocol
.data
.state_id
) {
3682 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3683 status
= relay_process_data_receive_header(conn
);
3685 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3686 status
= relay_process_data_receive_payload(conn
);
3689 ERR("Unexpected data connection communication state.");
3696 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3700 (void) lttng_poll_del(events
, pollfd
);
3702 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
, &pollfd
, 1,
3703 fd_tracker_util_close_fd
, NULL
);
3705 ERR("Closing pollfd %d", pollfd
);
3709 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3710 int pollfd
, struct relay_connection
*conn
)
3712 const char *type_str
;
3714 switch (conn
->type
) {
3719 type_str
= "Control";
3721 case RELAY_VIEWER_COMMAND
:
3722 type_str
= "Viewer Command";
3724 case RELAY_VIEWER_NOTIFICATION
:
3725 type_str
= "Viewer Notification";
3728 type_str
= "Unknown";
3730 cleanup_connection_pollfd(events
, pollfd
);
3731 connection_put(conn
);
3732 DBG("%s connection closed with %d", type_str
, pollfd
);
3736 * This thread does the actual work
3738 static void *relay_thread_worker(void *data
)
3740 int ret
, err
= -1, last_seen_data_fd
= -1;
3742 struct lttng_poll_event events
;
3743 struct lttng_ht
*relay_connections_ht
;
3744 struct lttng_ht_iter iter
;
3745 struct relay_connection
*destroy_conn
= NULL
;
3747 DBG("[thread] Relay worker started");
3749 rcu_register_thread();
3751 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3753 if (testpoint(relayd_thread_worker
)) {
3754 goto error_testpoint
;
3757 health_code_update();
3759 /* table of connections indexed on socket */
3760 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3761 if (!relay_connections_ht
) {
3762 goto relay_connections_ht_error
;
3765 ret
= create_named_thread_poll_set(&events
, 2, "Worker thread epoll");
3767 goto error_poll_create
;
3770 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3777 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3779 health_code_update();
3781 /* Infinite blocking call, waiting for transmission */
3782 DBG3("Relayd worker thread polling...");
3783 health_poll_entry();
3784 ret
= lttng_poll_wait(&events
, -1);
3788 * Restart interrupted system call.
3790 if (errno
== EINTR
) {
3799 * Process control. The control connection is
3800 * prioritized so we don't starve it with high
3801 * throughput tracing data on the data connection.
3803 for (i
= 0; i
< nb_fd
; i
++) {
3804 /* Fetch once the poll data */
3805 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3806 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3808 health_code_update();
3812 * No activity for this FD (poll
3818 /* Thread quit pipe has been closed. Killing thread. */
3819 ret
= check_thread_quit_pipe(pollfd
, revents
);
3825 /* Inspect the relay conn pipe for new connection */
3826 if (pollfd
== relay_conn_pipe
[0]) {
3827 if (revents
& LPOLLIN
) {
3828 struct relay_connection
*conn
;
3830 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3834 lttng_poll_add(&events
, conn
->sock
->fd
,
3835 LPOLLIN
| LPOLLRDHUP
);
3836 connection_ht_add(relay_connections_ht
, conn
);
3837 DBG("Connection socket %d added", conn
->sock
->fd
);
3838 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3839 ERR("Relay connection pipe error");
3842 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3846 struct relay_connection
*ctrl_conn
;
3848 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3849 /* If not found, there is a synchronization issue. */
3852 if (ctrl_conn
->type
== RELAY_DATA
) {
3853 if (revents
& LPOLLIN
) {
3855 * Flag the last seen data fd not deleted. It will be
3856 * used as the last seen fd if any fd gets deleted in
3859 last_notdel_data_fd
= pollfd
;
3861 goto put_ctrl_connection
;
3863 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3865 if (revents
& LPOLLIN
) {
3866 enum relay_connection_status status
;
3868 status
= relay_process_control(ctrl_conn
);
3869 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3871 * On socket error flag the session as aborted to force
3872 * the cleanup of its stream otherwise it can leak
3873 * during the lifetime of the relayd.
3875 * This prevents situations in which streams can be
3876 * left opened because an index was received, the
3877 * control connection is closed, and the data
3878 * connection is closed (uncleanly) before the packet's
3881 * Since the control connection encountered an error,
3882 * it is okay to be conservative and close the
3883 * session right now as we can't rely on the protocol
3884 * being respected anymore.
3886 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3887 session_abort(ctrl_conn
->session
);
3890 /* Clear the connection on error or close. */
3891 relay_thread_close_connection(&events
,
3896 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3897 relay_thread_close_connection(&events
,
3899 if (last_seen_data_fd
== pollfd
) {
3900 last_seen_data_fd
= last_notdel_data_fd
;
3903 ERR("Unexpected poll events %u for control sock %d",
3905 connection_put(ctrl_conn
);
3908 put_ctrl_connection
:
3909 connection_put(ctrl_conn
);
3914 * The last loop handled a control request, go back to poll to make
3915 * sure we prioritise the control socket.
3921 if (last_seen_data_fd
>= 0) {
3922 for (i
= 0; i
< nb_fd
; i
++) {
3923 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3925 health_code_update();
3927 if (last_seen_data_fd
== pollfd
) {
3934 /* Process data connection. */
3935 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3936 /* Fetch the poll data. */
3937 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3938 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3939 struct relay_connection
*data_conn
;
3941 health_code_update();
3944 /* No activity for this FD (poll implementation). */
3948 /* Skip the command pipe. It's handled in the first loop. */
3949 if (pollfd
== relay_conn_pipe
[0]) {
3953 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3955 /* Skip it. Might be removed before. */
3958 if (data_conn
->type
== RELAY_CONTROL
) {
3959 goto put_data_connection
;
3961 assert(data_conn
->type
== RELAY_DATA
);
3963 if (revents
& LPOLLIN
) {
3964 enum relay_connection_status status
;
3966 status
= relay_process_data(data_conn
);
3967 /* Connection closed or error. */
3968 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3970 * On socket error flag the session as aborted to force
3971 * the cleanup of its stream otherwise it can leak
3972 * during the lifetime of the relayd.
3974 * This prevents situations in which streams can be
3975 * left opened because an index was received, the
3976 * control connection is closed, and the data
3977 * connection is closed (uncleanly) before the packet's
3980 * Since the data connection encountered an error,
3981 * it is okay to be conservative and close the
3982 * session right now as we can't rely on the protocol
3983 * being respected anymore.
3985 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3986 session_abort(data_conn
->session
);
3988 relay_thread_close_connection(&events
, pollfd
,
3991 * Every goto restart call sets the last seen fd where
3992 * here we don't really care since we gracefully
3993 * continue the loop after the connection is deleted.
3996 /* Keep last seen port. */
3997 last_seen_data_fd
= pollfd
;
3998 connection_put(data_conn
);
4001 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4002 relay_thread_close_connection(&events
, pollfd
,
4005 ERR("Unknown poll events %u for data sock %d",
4008 put_data_connection
:
4009 connection_put(data_conn
);
4011 last_seen_data_fd
= -1;
4014 /* Normal exit, no error */
4019 /* Cleanup remaining connection object. */
4021 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
4024 health_code_update();
4026 session_abort(destroy_conn
->session
);
4029 * No need to grab another ref, because we own
4032 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
4037 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
4039 lttng_ht_destroy(relay_connections_ht
);
4040 relay_connections_ht_error
:
4041 /* Close relay conn pipes */
4042 (void) fd_tracker_util_pipe_close(the_fd_tracker
,
4045 DBG("Thread exited with error");
4047 DBG("Worker thread cleanup complete");
4051 ERR("Health error occurred in %s", __func__
);
4053 health_unregister(health_relayd
);
4054 rcu_unregister_thread();
4055 lttng_relay_stop_threads();
4060 * Create the relay command pipe to wake thread_manage_apps.
4061 * Closed in cleanup().
4063 static int create_relay_conn_pipe(void)
4065 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
4066 "Relayd connection pipe", relay_conn_pipe
);
4070 int stdio_open(void *data
, int *fds
)
4072 fds
[0] = fileno(stdout
);
4073 fds
[1] = fileno(stderr
);
4078 int noop_close(void *data
, int *fds
)
4084 int track_stdio(void)
4087 const char *names
[] = { "stdout", "stderr" };
4089 return fd_tracker_open_unsuspendable_fd(the_fd_tracker
, fds
,
4090 names
, 2, stdio_open
, NULL
);
4094 void untrack_stdio(void)
4096 int fds
[] = { fileno(stdout
), fileno(stderr
) };
4099 * noop_close is used since we don't really want to close
4100 * the stdio output fds; we merely want to stop tracking them.
4102 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
4103 fds
, 2, noop_close
, NULL
);
4109 int main(int argc
, char **argv
)
4111 int ret
= 0, retval
= 0;
4114 /* Parse arguments */
4116 if (set_options(argc
, argv
)) {
4121 if (set_signal_handler()) {
4126 ret
= fclose(stdin
);
4128 PERROR("Failed to close stdin");
4131 /* Try to create directory if -o, --output is specified. */
4132 if (opt_output_path
) {
4133 if (*opt_output_path
!= '/') {
4134 ERR("Please specify an absolute path for -o, --output PATH");
4139 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
4142 ERR("Unable to create %s", opt_output_path
);
4149 if (opt_daemon
|| opt_background
) {
4150 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4159 * The RCU thread registration (and use, through the fd-tracker's
4160 * creation) is done after the daemonization to allow us to not
4161 * deal with liburcu's fork() management as the call RCU needs to
4164 rcu_register_thread();
4166 the_fd_tracker
= fd_tracker_create(lttng_opt_fd_cap
);
4167 if (!the_fd_tracker
) {
4172 ret
= track_stdio();
4178 /* Initialize thread health monitoring */
4179 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
4180 if (!health_relayd
) {
4181 PERROR("health_app_create error");
4183 goto exit_health_app_create
;
4186 /* Create thread quit pipe */
4187 if (init_thread_quit_pipe()) {
4189 goto exit_init_data
;
4192 /* Setup the thread apps communication pipe. */
4193 if (create_relay_conn_pipe()) {
4195 goto exit_init_data
;
4198 /* Init relay command queue. */
4199 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
4201 /* Initialize communication library */
4203 lttcomm_inet_init();
4205 /* tables of sessions indexed by session ID */
4206 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4209 goto exit_init_data
;
4212 /* tables of streams indexed by stream ID */
4213 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4214 if (!relay_streams_ht
) {
4216 goto exit_init_data
;
4219 /* tables of streams indexed by stream ID */
4220 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4221 if (!viewer_streams_ht
) {
4223 goto exit_init_data
;
4226 ret
= init_health_quit_pipe();
4229 goto exit_health_quit_pipe
;
4232 /* Create thread to manage the client socket */
4233 ret
= pthread_create(&health_thread
, default_pthread_attr(),
4234 thread_manage_health
, (void *) NULL
);
4237 PERROR("pthread_create health");
4239 goto exit_health_thread
;
4242 /* Setup the dispatcher thread */
4243 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
4244 relay_thread_dispatcher
, (void *) NULL
);
4247 PERROR("pthread_create dispatcher");
4249 goto exit_dispatcher_thread
;
4252 /* Setup the worker thread */
4253 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
4254 relay_thread_worker
, NULL
);
4257 PERROR("pthread_create worker");
4259 goto exit_worker_thread
;
4262 /* Setup the listener thread */
4263 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
4264 relay_thread_listener
, (void *) NULL
);
4267 PERROR("pthread_create listener");
4269 goto exit_listener_thread
;
4272 ret
= relayd_live_create(live_uri
);
4274 ERR("Starting live viewer threads");
4280 * This is where we start awaiting program completion (e.g. through
4281 * signal that asks threads to teardown).
4284 ret
= relayd_live_join();
4290 ret
= pthread_join(listener_thread
, &status
);
4293 PERROR("pthread_join listener_thread");
4297 exit_listener_thread
:
4298 ret
= pthread_join(worker_thread
, &status
);
4301 PERROR("pthread_join worker_thread");
4306 ret
= pthread_join(dispatcher_thread
, &status
);
4309 PERROR("pthread_join dispatcher_thread");
4312 exit_dispatcher_thread
:
4314 ret
= pthread_join(health_thread
, &status
);
4317 PERROR("pthread_join health_thread");
4322 (void) fd_tracker_util_pipe_close(the_fd_tracker
, health_quit_pipe
);
4323 exit_health_quit_pipe
:
4326 health_app_destroy(health_relayd
);
4327 exit_health_app_create
:
4330 * Wait for all pending call_rcu work to complete before tearing
4331 * down data structures. call_rcu worker may be trying to
4332 * perform lookups in those structures.
4337 /* Ensure all prior call_rcu are done. */
4342 * fd_tracker_destroy() will log the contents of the fd-tracker
4343 * if a leak is detected.
4345 fd_tracker_destroy(the_fd_tracker
);
4346 rcu_unregister_thread();