2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
35 #include <sys/types.h>
37 #include <sys/resource.h>
39 #include <urcu/futex.h>
40 #include <urcu/uatomic.h>
45 #include <lttng/lttng.h>
46 #include <common/common.h>
47 #include <common/compat/poll.h>
48 #include <common/compat/socket.h>
49 #include <common/compat/endian.h>
50 #include <common/compat/getenv.h>
51 #include <common/defaults.h>
52 #include <common/daemonize.h>
53 #include <common/futex.h>
54 #include <common/sessiond-comm/sessiond-comm.h>
55 #include <common/sessiond-comm/inet.h>
56 #include <common/sessiond-comm/relayd.h>
57 #include <common/uri.h>
58 #include <common/utils.h>
59 #include <common/config/session-config.h>
60 #include <common/dynamic-buffer.h>
61 #include <common/buffer-view.h>
62 #include <common/fd-tracker/utils.h>
63 #include <urcu/rculist.h>
67 #include "ctf-trace.h"
70 #include "lttng-relayd.h"
72 #include "health-relayd.h"
73 #include "testpoint.h"
74 #include "viewer-stream.h"
77 #include "connection.h"
78 #include "tracefile-array.h"
79 #include "tcp_keep_alive.h"
81 enum relay_connection_status
{
82 RELAY_CONNECTION_STATUS_OK
,
83 /* An error occured while processing an event on the connection. */
84 RELAY_CONNECTION_STATUS_ERROR
,
85 /* Connection closed/shutdown cleanly. */
86 RELAY_CONNECTION_STATUS_CLOSED
,
89 /* command line options */
90 char *opt_output_path
, *opt_working_directory
;
91 static int opt_daemon
, opt_background
, opt_print_version
;
92 int opt_group_output_by_session
;
93 int opt_group_output_by_host
;
96 * We need to wait for listener and live listener threads, as well as
97 * health check thread, before being ready to signal readiness.
99 #define NR_LTTNG_RELAY_READY 3
100 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
102 /* Size of receive buffer. */
103 #define RECV_DATA_BUFFER_SIZE 65536
105 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
106 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
108 static struct lttng_uri
*control_uri
;
109 static struct lttng_uri
*data_uri
;
110 static struct lttng_uri
*live_uri
;
112 const char *progname
;
114 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
115 static int tracing_group_name_override
;
117 const char * const config_section_name
= "relayd";
120 * Quit pipe for all threads. This permits a single cancellation point
121 * for all threads when receiving an event on the pipe.
123 int thread_quit_pipe
[2] = { -1, -1 };
126 * This pipe is used to inform the worker thread that a command is queued and
127 * ready to be processed.
129 static int relay_conn_pipe
[2] = { -1, -1 };
131 /* Shared between threads */
132 static int dispatch_thread_exit
;
134 static pthread_t listener_thread
;
135 static pthread_t dispatcher_thread
;
136 static pthread_t worker_thread
;
137 static pthread_t health_thread
;
140 * last_relay_stream_id_lock protects last_relay_stream_id increment
141 * atomicity on 32-bit architectures.
143 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
144 static uint64_t last_relay_stream_id
;
147 * Relay command queue.
149 * The relay_thread_listener and relay_thread_dispatcher communicate with this
152 static struct relay_conn_queue relay_conn_queue
;
154 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
155 static unsigned int lttng_opt_fd_cap
;
157 /* Global relay stream hash table. */
158 struct lttng_ht
*relay_streams_ht
;
160 /* Global relay viewer stream hash table. */
161 struct lttng_ht
*viewer_streams_ht
;
163 /* Global relay sessions hash table. */
164 struct lttng_ht
*sessions_ht
;
166 /* Relayd health monitoring */
167 struct health_app
*health_relayd
;
169 /* Global fd tracker. */
170 struct fd_tracker
*the_fd_tracker
;
172 static struct option long_options
[] = {
173 { "control-port", 1, 0, 'C', },
174 { "data-port", 1, 0, 'D', },
175 { "live-port", 1, 0, 'L', },
176 { "daemonize", 0, 0, 'd', },
177 { "background", 0, 0, 'b', },
178 { "group", 1, 0, 'g', },
179 { "fd-cap", 1, 0, '\0', },
180 { "help", 0, 0, 'h', },
181 { "output", 1, 0, 'o', },
182 { "verbose", 0, 0, 'v', },
183 { "config", 1, 0, 'f' },
184 { "version", 0, 0, 'V' },
185 { "working-directory", 1, 0, 'w', },
186 { "group-output-by-session", 0, 0, 's', },
187 { "group-output-by-host", 0, 0, 'p', },
191 static const char *config_ignore_options
[] = { "help", "config", "version" };
193 static void print_version(void) {
194 fprintf(stdout
, "%s\n", VERSION
);
197 static void relayd_config_log(void)
199 DBG("LTTng-relayd " VERSION
" - " VERSION_NAME
"%s%s",
200 GIT_VERSION
[0] == '\0' ? "" : " - " GIT_VERSION
,
201 EXTRA_VERSION_NAME
[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME
);
202 if (EXTRA_VERSION_DESCRIPTION
[0] != '\0') {
203 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION
"\n");
205 if (EXTRA_VERSION_PATCHES
[0] != '\0') {
206 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES
"\n");
211 * Take an option from the getopt output and set it in the right variable to be
214 * Return 0 on success else a negative value.
216 static int set_option(int opt
, const char *arg
, const char *optname
)
222 if (!strcmp(optname
, "fd-cap")) {
226 v
= strtoul(arg
, NULL
, 0);
227 if (errno
!= 0 || !isdigit(arg
[0])) {
228 ERR("Wrong value in --fd-cap parameter: %s", arg
);
232 if (v
< DEFAULT_RELAYD_MINIMAL_FD_CAP
) {
233 ERR("File descriptor cap must be set to at least %d",
234 DEFAULT_RELAYD_MINIMAL_FD_CAP
);
237 ERR("File descriptor cap overflow in --fd-cap parameter: %s", arg
);
241 lttng_opt_fd_cap
= (unsigned int) v
;
242 DBG3("File descriptor cap set to %u", lttng_opt_fd_cap
);
245 fprintf(stderr
, "unknown option %s", optname
);
247 fprintf(stderr
, " with arg %s\n", arg
);
252 if (lttng_is_setuid_setgid()) {
253 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
254 "-C, --control-port");
256 ret
= uri_parse(arg
, &control_uri
);
258 ERR("Invalid control URI specified");
261 if (control_uri
->port
== 0) {
262 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
267 if (lttng_is_setuid_setgid()) {
268 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
271 ret
= uri_parse(arg
, &data_uri
);
273 ERR("Invalid data URI specified");
276 if (data_uri
->port
== 0) {
277 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
282 if (lttng_is_setuid_setgid()) {
283 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
286 ret
= uri_parse(arg
, &live_uri
);
288 ERR("Invalid live URI specified");
291 if (live_uri
->port
== 0) {
292 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
303 if (lttng_is_setuid_setgid()) {
304 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
307 tracing_group_name
= strdup(arg
);
308 if (tracing_group_name
== NULL
) {
313 tracing_group_name_override
= 1;
317 ret
= utils_show_man_page(8, "lttng-relayd");
319 ERR("Cannot view man page lttng-relayd(8)");
324 opt_print_version
= 1;
327 if (lttng_is_setuid_setgid()) {
328 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
331 ret
= asprintf(&opt_output_path
, "%s", arg
);
334 PERROR("asprintf opt_output_path");
340 if (lttng_is_setuid_setgid()) {
341 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
342 "-w, --working-directory");
344 ret
= asprintf(&opt_working_directory
, "%s", arg
);
347 PERROR("asprintf working_directory");
354 /* Verbose level can increase using multiple -v */
356 lttng_opt_verbose
= config_parse_value(arg
);
358 /* Only 3 level of verbosity (-vvv). */
359 if (lttng_opt_verbose
< 3) {
360 lttng_opt_verbose
+= 1;
365 if (opt_group_output_by_host
) {
366 ERR("Cannot set --group-output-by-session, --group-output-by-host already defined");
369 opt_group_output_by_session
= 1;
372 if (opt_group_output_by_session
) {
373 ERR("Cannot set --group-output-by-host, --group-output-by-session already defined");
376 opt_group_output_by_host
= 1;
379 /* Unknown option or other error.
380 * Error is printed by getopt, just return */
393 * config_entry_handler_cb used to handle options read from a config file.
394 * See config_entry_handler_cb comment in common/config/session-config.h for the
395 * return value conventions.
397 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
401 if (!entry
|| !entry
->name
|| !entry
->value
) {
406 /* Check if the option is to be ignored */
407 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
408 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
413 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
414 /* Ignore if entry name is not fully matched. */
415 if (strcmp(entry
->name
, long_options
[i
].name
)) {
420 * If the option takes no argument on the command line,
421 * we have to check if the value is "true". We support
422 * non-zero numeric values, true, on and yes.
424 if (!long_options
[i
].has_arg
) {
425 ret
= config_parse_value(entry
->value
);
428 WARN("Invalid configuration value \"%s\" for option %s",
429 entry
->value
, entry
->name
);
431 /* False, skip boolean config option. */
436 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
440 WARN("Unrecognized option \"%s\" in daemon configuration file.",
447 static void parse_env_options(void)
451 value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV
);
453 opt_working_directory
= value
;
457 static int set_options(int argc
, char **argv
)
459 int c
, ret
= 0, option_index
= 0, retval
= 0;
460 int orig_optopt
= optopt
, orig_optind
= optind
;
461 char *default_address
, *optstring
;
462 const char *config_path
= NULL
;
464 optstring
= utils_generate_optstring(long_options
,
465 sizeof(long_options
) / sizeof(struct option
));
471 /* Check for the --config option */
473 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
474 &option_index
)) != -1) {
478 } else if (c
!= 'f') {
482 if (lttng_is_setuid_setgid()) {
483 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
486 config_path
= utils_expand_path(optarg
);
488 ERR("Failed to resolve path: %s", optarg
);
493 ret
= config_get_section_entries(config_path
, config_section_name
,
494 config_entry_handler
, NULL
);
497 ERR("Invalid configuration option at line %i", ret
);
503 /* Reset getopt's global state */
504 optopt
= orig_optopt
;
505 optind
= orig_optind
;
507 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
512 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
519 /* assign default values */
520 if (control_uri
== NULL
) {
521 ret
= asprintf(&default_address
,
522 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
523 DEFAULT_NETWORK_CONTROL_PORT
);
525 PERROR("asprintf default data address");
530 ret
= uri_parse(default_address
, &control_uri
);
531 free(default_address
);
533 ERR("Invalid control URI specified");
538 if (data_uri
== NULL
) {
539 ret
= asprintf(&default_address
,
540 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
541 DEFAULT_NETWORK_DATA_PORT
);
543 PERROR("asprintf default data address");
548 ret
= uri_parse(default_address
, &data_uri
);
549 free(default_address
);
551 ERR("Invalid data URI specified");
556 if (live_uri
== NULL
) {
557 ret
= asprintf(&default_address
,
558 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
559 DEFAULT_NETWORK_VIEWER_PORT
);
561 PERROR("asprintf default viewer control address");
566 ret
= uri_parse(default_address
, &live_uri
);
567 free(default_address
);
569 ERR("Invalid viewer control URI specified");
574 if (lttng_opt_fd_cap
== 0) {
576 struct rlimit rlimit
;
578 ret
= getrlimit(RLIMIT_NOFILE
, &rlimit
);
580 PERROR("Failed to get file descriptor limit");
584 lttng_opt_fd_cap
= rlimit
.rlim_cur
;
587 if (!opt_group_output_by_session
&& !opt_group_output_by_host
) {
588 /* Group by host by default */
589 opt_group_output_by_host
= 1;
597 static void print_global_objects(void)
599 print_viewer_streams();
600 print_relay_streams();
607 static void relayd_cleanup(void)
609 print_global_objects();
613 if (viewer_streams_ht
)
614 lttng_ht_destroy(viewer_streams_ht
);
615 if (relay_streams_ht
)
616 lttng_ht_destroy(relay_streams_ht
);
618 lttng_ht_destroy(sessions_ht
);
620 /* free the dynamically allocated opt_output_path */
621 free(opt_output_path
);
623 /* Close thread quit pipes */
624 (void) fd_tracker_util_pipe_close(the_fd_tracker
, thread_quit_pipe
);
626 uri_free(control_uri
);
628 /* Live URI is freed in the live thread. */
630 if (tracing_group_name_override
) {
631 free((void *) tracing_group_name
);
636 * Write to writable pipe used to notify a thread.
638 static int notify_thread_pipe(int wpipe
)
642 ret
= lttng_write(wpipe
, "!", 1);
644 PERROR("write poll pipe");
652 static int notify_health_quit_pipe(int *pipe
)
656 ret
= lttng_write(pipe
[1], "4", 1);
658 PERROR("write relay health quit");
667 * Stop all relayd and relayd-live threads.
669 int lttng_relay_stop_threads(void)
673 /* Stopping all threads */
674 DBG("Terminating all threads");
675 if (notify_thread_pipe(thread_quit_pipe
[1])) {
676 ERR("write error on thread quit pipe");
680 if (notify_health_quit_pipe(health_quit_pipe
)) {
681 ERR("write error on health quit pipe");
684 /* Dispatch thread */
685 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
686 futex_nto1_wake(&relay_conn_queue
.futex
);
688 if (relayd_live_stop()) {
689 ERR("Error stopping live threads");
696 * Signal handler for the daemon
698 * Simply stop all worker threads, leaving main() return gracefully after
699 * joining all threads and calling cleanup().
701 static void sighandler(int sig
)
705 DBG("SIGINT caught");
706 if (lttng_relay_stop_threads()) {
707 ERR("Error stopping threads");
711 DBG("SIGTERM caught");
712 if (lttng_relay_stop_threads()) {
713 ERR("Error stopping threads");
717 CMM_STORE_SHARED(recv_child_signal
, 1);
725 * Setup signal handler for :
726 * SIGINT, SIGTERM, SIGPIPE
728 static int set_signal_handler(void)
734 if ((ret
= sigemptyset(&sigset
)) < 0) {
735 PERROR("sigemptyset");
742 sa
.sa_handler
= sighandler
;
743 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
748 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
753 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
758 sa
.sa_handler
= SIG_IGN
;
759 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
764 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
769 void lttng_relay_notify_ready(void)
771 /* Notify the parent of the fork() process that we are ready. */
772 if (opt_daemon
|| opt_background
) {
773 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
774 kill(child_ppid
, SIGUSR1
);
780 * Init thread quit pipe.
782 * Return -1 on error or 0 if all pipes are created.
784 static int init_thread_quit_pipe(void)
786 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
787 "Quit pipe", thread_quit_pipe
);
791 * Init health quit pipe.
793 * Return -1 on error or 0 if all pipes are created.
795 static int init_health_quit_pipe(void)
797 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
798 "Health quit pipe", health_quit_pipe
);
802 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
804 static int create_named_thread_poll_set(struct lttng_poll_event
*events
,
805 int size
, const char *name
)
809 if (events
== NULL
|| size
== 0) {
814 ret
= fd_tracker_util_poll_create(the_fd_tracker
,
815 name
, events
, 1, LTTNG_CLOEXEC
);
818 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
830 * Check if the thread quit pipe was triggered.
832 * Return 1 if it was triggered else 0;
834 static int check_thread_quit_pipe(int fd
, uint32_t events
)
836 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
844 * Create and init socket from uri.
846 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
)
849 struct lttcomm_sock
*sock
= NULL
;
851 sock
= lttcomm_alloc_sock_from_uri(uri
);
853 ERR("Allocating socket");
857 ret
= lttcomm_create_sock(sock
);
861 DBG("Listening on sock %d", sock
->fd
);
863 ret
= sock
->ops
->bind(sock
);
868 ret
= sock
->ops
->listen(sock
, -1);
878 lttcomm_destroy_sock(sock
);
884 * This thread manages the listening for new connections on the network
886 static void *relay_thread_listener(void *data
)
888 int i
, ret
, pollfd
, err
= -1;
889 uint32_t revents
, nb_fd
;
890 struct lttng_poll_event events
;
891 struct lttcomm_sock
*control_sock
, *data_sock
;
893 DBG("[thread] Relay listener started");
895 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
897 health_code_update();
899 control_sock
= relay_socket_create(control_uri
);
901 goto error_sock_control
;
904 data_sock
= relay_socket_create(data_uri
);
906 goto error_sock_relay
;
910 * Pass 3 as size here for the thread quit pipe, control and
913 ret
= create_named_thread_poll_set(&events
, 3, "Listener thread epoll");
915 goto error_create_poll
;
918 /* Add the control socket */
919 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
924 /* Add the data socket */
925 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
930 lttng_relay_notify_ready();
932 if (testpoint(relayd_thread_listener
)) {
933 goto error_testpoint
;
937 health_code_update();
939 DBG("Listener accepting connections");
943 ret
= lttng_poll_wait(&events
, -1);
947 * Restart interrupted system call.
949 if (errno
== EINTR
) {
957 DBG("Relay new connection received");
958 for (i
= 0; i
< nb_fd
; i
++) {
959 health_code_update();
961 /* Fetch once the poll data */
962 revents
= LTTNG_POLL_GETEV(&events
, i
);
963 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
967 * No activity for this FD (poll
973 /* Thread quit pipe has been closed. Killing thread. */
974 ret
= check_thread_quit_pipe(pollfd
, revents
);
980 if (revents
& LPOLLIN
) {
982 * A new connection is requested, therefore a
983 * sessiond/consumerd connection is allocated in
984 * this thread, enqueued to a global queue and
985 * dequeued (and freed) in the worker thread.
988 struct relay_connection
*new_conn
;
989 struct lttcomm_sock
*newsock
;
990 enum connection_type type
;
992 if (pollfd
== data_sock
->fd
) {
994 newsock
= data_sock
->ops
->accept(data_sock
);
995 DBG("Relay data connection accepted, socket %d",
998 assert(pollfd
== control_sock
->fd
);
999 type
= RELAY_CONTROL
;
1000 newsock
= control_sock
->ops
->accept(control_sock
);
1001 DBG("Relay control connection accepted, socket %d",
1005 PERROR("accepting sock");
1009 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
1012 PERROR("setsockopt inet");
1013 lttcomm_destroy_sock(newsock
);
1017 ret
= socket_apply_keep_alive_config(newsock
->fd
);
1019 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1021 lttcomm_destroy_sock(newsock
);
1025 new_conn
= connection_create(newsock
, type
);
1027 lttcomm_destroy_sock(newsock
);
1031 /* Enqueue request for the dispatcher thread. */
1032 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
1036 * Wake the dispatch queue futex.
1037 * Implicit memory barrier with the
1038 * exchange in cds_wfcq_enqueue.
1040 futex_nto1_wake(&relay_conn_queue
.futex
);
1041 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1042 ERR("socket poll error");
1045 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1055 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
1057 if (data_sock
->fd
>= 0) {
1058 ret
= data_sock
->ops
->close(data_sock
);
1063 lttcomm_destroy_sock(data_sock
);
1065 if (control_sock
->fd
>= 0) {
1066 ret
= control_sock
->ops
->close(control_sock
);
1071 lttcomm_destroy_sock(control_sock
);
1075 ERR("Health error occurred in %s", __func__
);
1077 health_unregister(health_relayd
);
1078 DBG("Relay listener thread cleanup complete");
1079 lttng_relay_stop_threads();
1084 * This thread manages the dispatching of the requests to worker threads
1086 static void *relay_thread_dispatcher(void *data
)
1090 struct cds_wfcq_node
*node
;
1091 struct relay_connection
*new_conn
= NULL
;
1093 DBG("[thread] Relay dispatcher started");
1095 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
1097 if (testpoint(relayd_thread_dispatcher
)) {
1098 goto error_testpoint
;
1101 health_code_update();
1104 health_code_update();
1106 /* Atomically prepare the queue futex */
1107 futex_nto1_prepare(&relay_conn_queue
.futex
);
1109 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1114 health_code_update();
1116 /* Dequeue commands */
1117 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1118 &relay_conn_queue
.tail
);
1120 DBG("Woken up but nothing in the relay command queue");
1121 /* Continue thread execution */
1124 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1126 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1129 * Inform worker thread of the new request. This
1130 * call is blocking so we can be assured that
1131 * the data will be read at some point in time
1132 * or wait to the end of the world :)
1134 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1136 PERROR("write connection pipe");
1137 connection_put(new_conn
);
1140 } while (node
!= NULL
);
1142 /* Futex wait on queue. Blocking call on futex() */
1143 health_poll_entry();
1144 futex_nto1_wait(&relay_conn_queue
.futex
);
1148 /* Normal exit, no error */
1155 ERR("Health error occurred in %s", __func__
);
1157 health_unregister(health_relayd
);
1158 DBG("Dispatch thread dying");
1159 lttng_relay_stop_threads();
1164 * Set index data from the control port to a given index object.
1166 static int set_index_control_data(struct relay_index
*index
,
1167 struct lttcomm_relayd_index
*data
,
1168 struct relay_connection
*conn
)
1170 struct ctf_packet_index index_data
;
1173 * The index on disk is encoded in big endian.
1175 index_data
.packet_size
= htobe64(data
->packet_size
);
1176 index_data
.content_size
= htobe64(data
->content_size
);
1177 index_data
.timestamp_begin
= htobe64(data
->timestamp_begin
);
1178 index_data
.timestamp_end
= htobe64(data
->timestamp_end
);
1179 index_data
.events_discarded
= htobe64(data
->events_discarded
);
1180 index_data
.stream_id
= htobe64(data
->stream_id
);
1182 if (conn
->minor
>= 8) {
1183 index
->index_data
.stream_instance_id
= htobe64(data
->stream_instance_id
);
1184 index
->index_data
.packet_seq_num
= htobe64(data
->packet_seq_num
);
1187 return relay_index_set_data(index
, &index_data
);
1191 * Handle the RELAYD_CREATE_SESSION command.
1193 * On success, send back the session id or else return a negative value.
1195 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1196 struct relay_connection
*conn
,
1197 const struct lttng_buffer_view
*payload
)
1201 struct relay_session
*session
;
1202 struct lttcomm_relayd_status_session reply
;
1203 char session_name
[LTTNG_NAME_MAX
];
1204 char hostname
[LTTNG_HOST_NAME_MAX
];
1205 uint32_t live_timer
= 0;
1206 bool snapshot
= false;
1208 memset(session_name
, 0, LTTNG_NAME_MAX
);
1209 memset(hostname
, 0, LTTNG_HOST_NAME_MAX
);
1211 memset(&reply
, 0, sizeof(reply
));
1213 switch (conn
->minor
) {
1218 case 4: /* LTTng sessiond 2.4 */
1220 ret
= cmd_create_session_2_4(payload
, session_name
,
1221 hostname
, &live_timer
, &snapshot
);
1227 session
= session_create(session_name
, hostname
, live_timer
,
1228 snapshot
, conn
->major
, conn
->minor
);
1233 assert(!conn
->session
);
1234 conn
->session
= session
;
1235 DBG("Created session %" PRIu64
, session
->id
);
1237 reply
.session_id
= htobe64(session
->id
);
1241 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1243 reply
.ret_code
= htobe32(LTTNG_OK
);
1246 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1247 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1248 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1257 * When we have received all the streams and the metadata for a channel,
1258 * we make them visible to the viewer threads.
1260 static void publish_connection_local_streams(struct relay_connection
*conn
)
1262 struct relay_stream
*stream
;
1263 struct relay_session
*session
= conn
->session
;
1266 * We publish all streams belonging to a session atomically wrt
1269 pthread_mutex_lock(&session
->lock
);
1271 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1273 stream_publish(stream
);
1278 * Inform the viewer that there are new streams in the session.
1280 if (session
->viewer_attached
) {
1281 uatomic_set(&session
->new_streams
, 1);
1283 pthread_mutex_unlock(&session
->lock
);
1287 * relay_add_stream: allocate a new stream for a session
1289 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1290 struct relay_connection
*conn
,
1291 const struct lttng_buffer_view
*payload
)
1295 struct relay_session
*session
= conn
->session
;
1296 struct relay_stream
*stream
= NULL
;
1297 struct lttcomm_relayd_status_stream reply
;
1298 struct ctf_trace
*trace
= NULL
;
1299 uint64_t stream_handle
= -1ULL;
1300 char *path_name
= NULL
, *channel_name
= NULL
;
1301 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1303 if (!session
|| !conn
->version_check_done
) {
1304 ERR("Trying to add a stream before version check");
1306 goto end_no_session
;
1309 switch (session
->minor
) {
1310 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1311 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1312 &channel_name
, session
);
1314 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1316 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1317 &channel_name
, &tracefile_size
, &tracefile_count
,
1325 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1329 /* This stream here has one reference on the trace. */
1331 pthread_mutex_lock(&last_relay_stream_id_lock
);
1332 stream_handle
= ++last_relay_stream_id
;
1333 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1335 /* We pass ownership of path_name and channel_name. */
1336 stream
= stream_create(trace
, stream_handle
, path_name
,
1337 channel_name
, tracefile_size
, tracefile_count
);
1339 channel_name
= NULL
;
1342 * Streams are the owners of their trace. Reference to trace is
1343 * kept within stream_create().
1345 ctf_trace_put(trace
);
1348 memset(&reply
, 0, sizeof(reply
));
1349 reply
.handle
= htobe64(stream_handle
);
1351 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1353 reply
.ret_code
= htobe32(LTTNG_OK
);
1356 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1357 sizeof(struct lttcomm_relayd_status_stream
), 0);
1358 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1359 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1371 * relay_close_stream: close a specific stream
1373 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1374 struct relay_connection
*conn
,
1375 const struct lttng_buffer_view
*payload
)
1379 struct relay_session
*session
= conn
->session
;
1380 struct lttcomm_relayd_close_stream stream_info
;
1381 struct lttcomm_relayd_generic_reply reply
;
1382 struct relay_stream
*stream
;
1384 DBG("Close stream received");
1386 if (!session
|| !conn
->version_check_done
) {
1387 ERR("Trying to close a stream before version check");
1389 goto end_no_session
;
1392 if (payload
->size
< sizeof(stream_info
)) {
1393 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1394 sizeof(stream_info
), payload
->size
);
1396 goto end_no_session
;
1398 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1399 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1400 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1402 stream
= stream_get_by_id(stream_info
.stream_id
);
1409 * Set last_net_seq_num before the close flag. Required by data
1412 pthread_mutex_lock(&stream
->lock
);
1413 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1414 pthread_mutex_unlock(&stream
->lock
);
1417 * This is one of the conditions which may trigger a stream close
1418 * with the others being:
1419 * 1) A close command is received for a stream
1420 * 2) The control connection owning the stream is closed
1421 * 3) We have received all of the stream's data _after_ a close
1424 try_stream_close(stream
);
1425 if (stream
->is_metadata
) {
1426 struct relay_viewer_stream
*vstream
;
1428 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1430 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1432 * Since all the metadata has been sent to the
1433 * viewer and that we have a request to close
1434 * its stream, we can safely teardown the
1435 * corresponding metadata viewer stream.
1437 viewer_stream_put(vstream
);
1439 /* Put local reference. */
1440 viewer_stream_put(vstream
);
1447 memset(&reply
, 0, sizeof(reply
));
1449 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1451 reply
.ret_code
= htobe32(LTTNG_OK
);
1453 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1454 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1455 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1456 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1466 * relay_reset_metadata: reset a metadata stream
1469 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1470 struct relay_connection
*conn
,
1471 const struct lttng_buffer_view
*payload
)
1475 struct relay_session
*session
= conn
->session
;
1476 struct lttcomm_relayd_reset_metadata stream_info
;
1477 struct lttcomm_relayd_generic_reply reply
;
1478 struct relay_stream
*stream
;
1480 DBG("Reset metadata received");
1482 if (!session
|| !conn
->version_check_done
) {
1483 ERR("Trying to reset a metadata stream before version check");
1485 goto end_no_session
;
1488 if (payload
->size
< sizeof(stream_info
)) {
1489 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1490 sizeof(stream_info
), payload
->size
);
1492 goto end_no_session
;
1494 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1495 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1496 stream_info
.version
= be64toh(stream_info
.version
);
1498 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1500 /* Unsupported for live sessions for now. */
1501 if (session
->live_timer
!= 0) {
1506 stream
= stream_get_by_id(stream_info
.stream_id
);
1511 pthread_mutex_lock(&stream
->lock
);
1512 if (!stream
->is_metadata
) {
1517 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
1518 0, 0, -1, -1, stream
->stream_fd
->fd
, NULL
,
1519 &stream
->stream_fd
->fd
);
1521 ERR("Failed to rotate metadata file %s of channel %s",
1522 stream
->path_name
, stream
->channel_name
);
1527 pthread_mutex_unlock(&stream
->lock
);
1531 memset(&reply
, 0, sizeof(reply
));
1533 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1535 reply
.ret_code
= htobe32(LTTNG_OK
);
1537 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1538 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1539 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1540 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1550 * relay_unknown_command: send -1 if received unknown command
1552 static void relay_unknown_command(struct relay_connection
*conn
)
1554 struct lttcomm_relayd_generic_reply reply
;
1557 memset(&reply
, 0, sizeof(reply
));
1558 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1559 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1560 if (send_ret
< sizeof(reply
)) {
1561 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1566 * relay_start: send an acknowledgment to the client to tell if we are
1567 * ready to receive data. We are ready if a session is established.
1569 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1570 struct relay_connection
*conn
,
1571 const struct lttng_buffer_view
*payload
)
1575 struct lttcomm_relayd_generic_reply reply
;
1576 struct relay_session
*session
= conn
->session
;
1579 DBG("Trying to start the streaming without a session established");
1580 ret
= htobe32(LTTNG_ERR_UNK
);
1583 memset(&reply
, 0, sizeof(reply
));
1584 reply
.ret_code
= htobe32(LTTNG_OK
);
1585 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1587 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1588 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1597 * Append padding to the file pointed by the file descriptor fd.
1599 static int write_padding_to_file(int fd
, uint32_t size
)
1608 zeros
= zmalloc(size
);
1609 if (zeros
== NULL
) {
1610 PERROR("zmalloc zeros for padding");
1615 ret
= lttng_write(fd
, zeros
, size
);
1617 PERROR("write padding to file");
1627 * relay_recv_metadata: receive the metadata for the session.
1629 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1630 struct relay_connection
*conn
,
1631 const struct lttng_buffer_view
*payload
)
1635 struct relay_session
*session
= conn
->session
;
1636 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1637 struct relay_stream
*metadata_stream
;
1638 uint64_t metadata_payload_size
;
1641 ERR("Metadata sent before version check");
1646 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1647 ERR("Incorrect data size");
1651 metadata_payload_size
= recv_hdr
->data_size
-
1652 sizeof(struct lttcomm_relayd_metadata_payload
);
1654 memcpy(&metadata_payload_header
, payload
->data
,
1655 sizeof(metadata_payload_header
));
1656 metadata_payload_header
.stream_id
= be64toh(
1657 metadata_payload_header
.stream_id
);
1658 metadata_payload_header
.padding_size
= be32toh(
1659 metadata_payload_header
.padding_size
);
1661 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1662 if (!metadata_stream
) {
1667 pthread_mutex_lock(&metadata_stream
->lock
);
1669 size_ret
= lttng_write(metadata_stream
->stream_fd
->fd
,
1670 payload
->data
+ sizeof(metadata_payload_header
),
1671 metadata_payload_size
);
1672 if (size_ret
< metadata_payload_size
) {
1673 ERR("Relay error writing metadata on file");
1678 size_ret
= write_padding_to_file(metadata_stream
->stream_fd
->fd
,
1679 metadata_payload_header
.padding_size
);
1680 if (size_ret
< (int64_t) metadata_payload_header
.padding_size
) {
1685 metadata_stream
->metadata_received
+=
1686 metadata_payload_size
+ metadata_payload_header
.padding_size
;
1687 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1688 metadata_stream
->metadata_received
);
1691 pthread_mutex_unlock(&metadata_stream
->lock
);
1692 stream_put(metadata_stream
);
1698 * relay_send_version: send relayd version number
1700 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1701 struct relay_connection
*conn
,
1702 const struct lttng_buffer_view
*payload
)
1706 struct lttcomm_relayd_version reply
, msg
;
1707 bool compatible
= true;
1709 conn
->version_check_done
= true;
1711 /* Get version from the other side. */
1712 if (payload
->size
< sizeof(msg
)) {
1713 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1714 sizeof(msg
), payload
->size
);
1719 memcpy(&msg
, payload
->data
, sizeof(msg
));
1720 msg
.major
= be32toh(msg
.major
);
1721 msg
.minor
= be32toh(msg
.minor
);
1723 memset(&reply
, 0, sizeof(reply
));
1724 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1725 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1727 /* Major versions must be the same */
1728 if (reply
.major
!= msg
.major
) {
1729 DBG("Incompatible major versions (%u vs %u), deleting session",
1730 reply
.major
, msg
.major
);
1734 conn
->major
= reply
.major
;
1735 /* We adapt to the lowest compatible version */
1736 if (reply
.minor
<= msg
.minor
) {
1737 conn
->minor
= reply
.minor
;
1739 conn
->minor
= msg
.minor
;
1742 reply
.major
= htobe32(reply
.major
);
1743 reply
.minor
= htobe32(reply
.minor
);
1744 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1746 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1747 ERR("Failed to send \"send version\" command reply (ret = %zd)",
1760 DBG("Version check done using protocol %u.%u", conn
->major
,
1768 * Check for data pending for a given stream id from the session daemon.
1770 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1771 struct relay_connection
*conn
,
1772 const struct lttng_buffer_view
*payload
)
1774 struct relay_session
*session
= conn
->session
;
1775 struct lttcomm_relayd_data_pending msg
;
1776 struct lttcomm_relayd_generic_reply reply
;
1777 struct relay_stream
*stream
;
1781 DBG("Data pending command received");
1783 if (!session
|| !conn
->version_check_done
) {
1784 ERR("Trying to check for data before version check");
1786 goto end_no_session
;
1789 if (payload
->size
< sizeof(msg
)) {
1790 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
1791 sizeof(msg
), payload
->size
);
1793 goto end_no_session
;
1795 memcpy(&msg
, payload
->data
, sizeof(msg
));
1796 msg
.stream_id
= be64toh(msg
.stream_id
);
1797 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1799 stream
= stream_get_by_id(msg
.stream_id
);
1800 if (stream
== NULL
) {
1805 pthread_mutex_lock(&stream
->lock
);
1807 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1808 " and last_seq %" PRIu64
, msg
.stream_id
,
1809 stream
->prev_seq
, msg
.last_net_seq_num
);
1811 /* Avoid wrapping issue */
1812 if (((int64_t) (stream
->prev_seq
- msg
.last_net_seq_num
)) >= 0) {
1813 /* Data has in fact been written and is NOT pending */
1816 /* Data still being streamed thus pending */
1820 stream
->data_pending_check_done
= true;
1821 pthread_mutex_unlock(&stream
->lock
);
1826 memset(&reply
, 0, sizeof(reply
));
1827 reply
.ret_code
= htobe32(ret
);
1828 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1829 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1830 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
1840 * Wait for the control socket to reach a quiescent state.
1842 * Note that for now, when receiving this command from the session
1843 * daemon, this means that every subsequent commands or data received on
1844 * the control socket has been handled. So, this is why we simply return
1847 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
1848 struct relay_connection
*conn
,
1849 const struct lttng_buffer_view
*payload
)
1853 struct relay_stream
*stream
;
1854 struct lttcomm_relayd_quiescent_control msg
;
1855 struct lttcomm_relayd_generic_reply reply
;
1857 DBG("Checking quiescent state on control socket");
1859 if (!conn
->session
|| !conn
->version_check_done
) {
1860 ERR("Trying to check for data before version check");
1862 goto end_no_session
;
1865 if (payload
->size
< sizeof(msg
)) {
1866 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
1867 sizeof(msg
), payload
->size
);
1869 goto end_no_session
;
1871 memcpy(&msg
, payload
->data
, sizeof(msg
));
1872 msg
.stream_id
= be64toh(msg
.stream_id
);
1874 stream
= stream_get_by_id(msg
.stream_id
);
1878 pthread_mutex_lock(&stream
->lock
);
1879 stream
->data_pending_check_done
= true;
1880 pthread_mutex_unlock(&stream
->lock
);
1882 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
1885 memset(&reply
, 0, sizeof(reply
));
1886 reply
.ret_code
= htobe32(LTTNG_OK
);
1887 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1888 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1889 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
1901 * Initialize a data pending command. This means that a consumer is about
1902 * to ask for data pending for each stream it holds. Simply iterate over
1903 * all streams of a session and set the data_pending_check_done flag.
1905 * This command returns to the client a LTTNG_OK code.
1907 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1908 struct relay_connection
*conn
,
1909 const struct lttng_buffer_view
*payload
)
1913 struct lttng_ht_iter iter
;
1914 struct lttcomm_relayd_begin_data_pending msg
;
1915 struct lttcomm_relayd_generic_reply reply
;
1916 struct relay_stream
*stream
;
1921 DBG("Init streams for data pending");
1923 if (!conn
->session
|| !conn
->version_check_done
) {
1924 ERR("Trying to check for data before version check");
1926 goto end_no_session
;
1929 if (payload
->size
< sizeof(msg
)) {
1930 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
1931 sizeof(msg
), payload
->size
);
1933 goto end_no_session
;
1935 memcpy(&msg
, payload
->data
, sizeof(msg
));
1936 msg
.session_id
= be64toh(msg
.session_id
);
1939 * Iterate over all streams to set the begin data pending flag.
1940 * For now, the streams are indexed by stream handle so we have
1941 * to iterate over all streams to find the one associated with
1942 * the right session_id.
1945 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1947 if (!stream_get(stream
)) {
1950 if (stream
->trace
->session
->id
== msg
.session_id
) {
1951 pthread_mutex_lock(&stream
->lock
);
1952 stream
->data_pending_check_done
= false;
1953 pthread_mutex_unlock(&stream
->lock
);
1954 DBG("Set begin data pending flag to stream %" PRIu64
,
1955 stream
->stream_handle
);
1961 memset(&reply
, 0, sizeof(reply
));
1962 /* All good, send back reply. */
1963 reply
.ret_code
= htobe32(LTTNG_OK
);
1965 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1966 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1967 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
1979 * End data pending command. This will check, for a given session id, if
1980 * each stream associated with it has its data_pending_check_done flag
1981 * set. If not, this means that the client lost track of the stream but
1982 * the data is still being streamed on our side. In this case, we inform
1983 * the client that data is in flight.
1985 * Return to the client if there is data in flight or not with a ret_code.
1987 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
1988 struct relay_connection
*conn
,
1989 const struct lttng_buffer_view
*payload
)
1993 struct lttng_ht_iter iter
;
1994 struct lttcomm_relayd_end_data_pending msg
;
1995 struct lttcomm_relayd_generic_reply reply
;
1996 struct relay_stream
*stream
;
1997 uint32_t is_data_inflight
= 0;
1999 DBG("End data pending command");
2001 if (!conn
->session
|| !conn
->version_check_done
) {
2002 ERR("Trying to check for data before version check");
2004 goto end_no_session
;
2007 if (payload
->size
< sizeof(msg
)) {
2008 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2009 sizeof(msg
), payload
->size
);
2011 goto end_no_session
;
2013 memcpy(&msg
, payload
->data
, sizeof(msg
));
2014 msg
.session_id
= be64toh(msg
.session_id
);
2017 * Iterate over all streams to see if the begin data pending
2021 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2023 if (!stream_get(stream
)) {
2026 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2030 pthread_mutex_lock(&stream
->lock
);
2031 if (!stream
->data_pending_check_done
) {
2032 if (!stream
->closed
|| !(((int64_t) (stream
->prev_seq
- stream
->last_net_seq_num
)) >= 0)) {
2033 is_data_inflight
= 1;
2034 DBG("Data is still in flight for stream %" PRIu64
,
2035 stream
->stream_handle
);
2036 pthread_mutex_unlock(&stream
->lock
);
2041 pthread_mutex_unlock(&stream
->lock
);
2046 memset(&reply
, 0, sizeof(reply
));
2047 /* All good, send back reply. */
2048 reply
.ret_code
= htobe32(is_data_inflight
);
2050 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2051 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2052 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2064 * Receive an index for a specific stream.
2066 * Return 0 on success else a negative value.
2068 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2069 struct relay_connection
*conn
,
2070 const struct lttng_buffer_view
*payload
)
2074 struct relay_session
*session
= conn
->session
;
2075 struct lttcomm_relayd_index index_info
;
2076 struct relay_index
*index
;
2077 struct lttcomm_relayd_generic_reply reply
;
2078 struct relay_stream
*stream
;
2083 DBG("Relay receiving index");
2085 if (!session
|| !conn
->version_check_done
) {
2086 ERR("Trying to close a stream before version check");
2088 goto end_no_session
;
2091 msg_len
= lttcomm_relayd_index_len(
2092 lttng_to_index_major(conn
->major
, conn
->minor
),
2093 lttng_to_index_minor(conn
->major
, conn
->minor
));
2094 if (payload
->size
< msg_len
) {
2095 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2096 msg_len
, payload
->size
);
2098 goto end_no_session
;
2100 memcpy(&index_info
, payload
->data
, msg_len
);
2101 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2102 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2103 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2104 index_info
.content_size
= be64toh(index_info
.content_size
);
2105 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2106 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2107 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2108 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2110 if (conn
->minor
>= 8) {
2111 index_info
.stream_instance_id
=
2112 be64toh(index_info
.stream_instance_id
);
2113 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2116 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2118 ERR("stream_get_by_id not found");
2122 pthread_mutex_lock(&stream
->lock
);
2124 /* Live beacon handling */
2125 if (index_info
.packet_size
== 0) {
2126 DBG("Received live beacon for stream %" PRIu64
,
2127 stream
->stream_handle
);
2130 * Only flag a stream inactive when it has already
2131 * received data and no indexes are in flight.
2133 if (stream
->index_received_seqcount
> 0
2134 && stream
->indexes_in_flight
== 0) {
2135 stream
->beacon_ts_end
= index_info
.timestamp_end
;
2138 goto end_stream_put
;
2140 stream
->beacon_ts_end
= -1ULL;
2143 if (stream
->ctf_stream_id
== -1ULL) {
2144 stream
->ctf_stream_id
= index_info
.stream_id
;
2146 index
= relay_index_get_by_id_or_create(stream
, index_info
.net_seq_num
);
2149 ERR("relay_index_get_by_id_or_create index NULL");
2150 goto end_stream_put
;
2152 if (set_index_control_data(index
, &index_info
, conn
)) {
2153 ERR("set_index_control_data error");
2154 relay_index_put(index
);
2156 goto end_stream_put
;
2158 ret
= relay_index_try_flush(index
);
2160 tracefile_array_commit_seq(stream
->tfa
);
2161 stream
->index_received_seqcount
++;
2162 } else if (ret
> 0) {
2166 ERR("relay_index_try_flush error %d", ret
);
2167 relay_index_put(index
);
2172 pthread_mutex_unlock(&stream
->lock
);
2177 memset(&reply
, 0, sizeof(reply
));
2179 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2181 reply
.ret_code
= htobe32(LTTNG_OK
);
2183 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2184 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2185 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2194 * Receive the streams_sent message.
2196 * Return 0 on success else a negative value.
2198 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2199 struct relay_connection
*conn
,
2200 const struct lttng_buffer_view
*payload
)
2204 struct lttcomm_relayd_generic_reply reply
;
2208 DBG("Relay receiving streams_sent");
2210 if (!conn
->session
|| !conn
->version_check_done
) {
2211 ERR("Trying to close a stream before version check");
2213 goto end_no_session
;
2217 * Publish every pending stream in the connection recv list which are
2218 * now ready to be used by the viewer.
2220 publish_connection_local_streams(conn
);
2222 memset(&reply
, 0, sizeof(reply
));
2223 reply
.ret_code
= htobe32(LTTNG_OK
);
2224 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2225 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2226 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2238 #define DBG_CMD(cmd_name, conn) \
2239 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2241 static int relay_process_control_command(struct relay_connection
*conn
,
2242 const struct lttcomm_relayd_hdr
*header
,
2243 const struct lttng_buffer_view
*payload
)
2247 switch (header
->cmd
) {
2248 case RELAYD_CREATE_SESSION
:
2249 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
2250 ret
= relay_create_session(header
, conn
, payload
);
2252 case RELAYD_ADD_STREAM
:
2253 DBG_CMD("RELAYD_ADD_STREAM", conn
);
2254 ret
= relay_add_stream(header
, conn
, payload
);
2256 case RELAYD_START_DATA
:
2257 DBG_CMD("RELAYD_START_DATA", conn
);
2258 ret
= relay_start(header
, conn
, payload
);
2260 case RELAYD_SEND_METADATA
:
2261 DBG_CMD("RELAYD_SEND_METADATA", conn
);
2262 ret
= relay_recv_metadata(header
, conn
, payload
);
2264 case RELAYD_VERSION
:
2265 DBG_CMD("RELAYD_VERSION", conn
);
2266 ret
= relay_send_version(header
, conn
, payload
);
2268 case RELAYD_CLOSE_STREAM
:
2269 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
2270 ret
= relay_close_stream(header
, conn
, payload
);
2272 case RELAYD_DATA_PENDING
:
2273 DBG_CMD("RELAYD_DATA_PENDING", conn
);
2274 ret
= relay_data_pending(header
, conn
, payload
);
2276 case RELAYD_QUIESCENT_CONTROL
:
2277 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
2278 ret
= relay_quiescent_control(header
, conn
, payload
);
2280 case RELAYD_BEGIN_DATA_PENDING
:
2281 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
2282 ret
= relay_begin_data_pending(header
, conn
, payload
);
2284 case RELAYD_END_DATA_PENDING
:
2285 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
2286 ret
= relay_end_data_pending(header
, conn
, payload
);
2288 case RELAYD_SEND_INDEX
:
2289 DBG_CMD("RELAYD_SEND_INDEX", conn
);
2290 ret
= relay_recv_index(header
, conn
, payload
);
2292 case RELAYD_STREAMS_SENT
:
2293 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
2294 ret
= relay_streams_sent(header
, conn
, payload
);
2296 case RELAYD_RESET_METADATA
:
2297 DBG_CMD("RELAYD_RESET_METADATA", conn
);
2298 ret
= relay_reset_metadata(header
, conn
, payload
);
2300 case RELAYD_UPDATE_SYNC_INFO
:
2302 ERR("Received unknown command (%u)", header
->cmd
);
2303 relay_unknown_command(conn
);
2312 static enum relay_connection_status
relay_process_control_receive_payload(
2313 struct relay_connection
*conn
)
2316 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2317 struct lttng_dynamic_buffer
*reception_buffer
=
2318 &conn
->protocol
.ctrl
.reception_buffer
;
2319 struct ctrl_connection_state_receive_payload
*state
=
2320 &conn
->protocol
.ctrl
.state
.receive_payload
;
2321 struct lttng_buffer_view payload_view
;
2323 if (state
->left_to_receive
== 0) {
2324 /* Short-circuit for payload-less commands. */
2325 goto reception_complete
;
2327 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2328 reception_buffer
->data
+ state
->received
,
2329 state
->left_to_receive
, MSG_DONTWAIT
);
2331 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2332 PERROR("Unable to receive command payload on sock %d",
2334 status
= RELAY_CONNECTION_STATUS_ERROR
;
2337 } else if (ret
== 0) {
2338 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2339 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2344 assert(ret
<= state
->left_to_receive
);
2346 state
->left_to_receive
-= ret
;
2347 state
->received
+= ret
;
2349 if (state
->left_to_receive
> 0) {
2351 * Can't transition to the protocol's next state, wait to
2352 * receive the rest of the header.
2354 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2355 state
->received
, state
->left_to_receive
,
2361 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
2362 conn
->sock
->fd
, state
->received
);
2364 * The payload required to process the command has been received.
2365 * A view to the reception buffer is forwarded to the various
2366 * commands and the state of the control is reset on success.
2368 * Commands are responsible for sending their reply to the peer.
2370 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
2372 ret
= relay_process_control_command(conn
,
2373 &state
->header
, &payload_view
);
2375 status
= RELAY_CONNECTION_STATUS_ERROR
;
2379 ret
= connection_reset_protocol_state(conn
);
2381 status
= RELAY_CONNECTION_STATUS_ERROR
;
2387 static enum relay_connection_status
relay_process_control_receive_header(
2388 struct relay_connection
*conn
)
2391 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2392 struct lttcomm_relayd_hdr header
;
2393 struct lttng_dynamic_buffer
*reception_buffer
=
2394 &conn
->protocol
.ctrl
.reception_buffer
;
2395 struct ctrl_connection_state_receive_header
*state
=
2396 &conn
->protocol
.ctrl
.state
.receive_header
;
2398 assert(state
->left_to_receive
!= 0);
2400 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2401 reception_buffer
->data
+ state
->received
,
2402 state
->left_to_receive
, MSG_DONTWAIT
);
2404 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2405 PERROR("Unable to receive control command header on sock %d",
2407 status
= RELAY_CONNECTION_STATUS_ERROR
;
2410 } else if (ret
== 0) {
2411 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2412 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2417 assert(ret
<= state
->left_to_receive
);
2419 state
->left_to_receive
-= ret
;
2420 state
->received
+= ret
;
2422 if (state
->left_to_receive
> 0) {
2424 * Can't transition to the protocol's next state, wait to
2425 * receive the rest of the header.
2427 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2428 state
->received
, state
->left_to_receive
,
2433 /* Transition to next state: receiving the command's payload. */
2434 conn
->protocol
.ctrl
.state_id
=
2435 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
2436 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
2437 header
.circuit_id
= be64toh(header
.circuit_id
);
2438 header
.data_size
= be64toh(header
.data_size
);
2439 header
.cmd
= be32toh(header
.cmd
);
2440 header
.cmd_version
= be32toh(header
.cmd_version
);
2441 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
2442 &header
, sizeof(header
));
2444 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
2445 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
2448 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
2449 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
2451 status
= RELAY_CONNECTION_STATUS_ERROR
;
2455 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
2457 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
2458 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
2461 status
= RELAY_CONNECTION_STATUS_ERROR
;
2465 if (header
.data_size
== 0) {
2467 * Manually invoke the next state as the poll loop
2468 * will not wake-up to allow us to proceed further.
2470 status
= relay_process_control_receive_payload(conn
);
2477 * Process the commands received on the control socket
2479 static enum relay_connection_status
relay_process_control(
2480 struct relay_connection
*conn
)
2482 enum relay_connection_status status
;
2484 switch (conn
->protocol
.ctrl
.state_id
) {
2485 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
2486 status
= relay_process_control_receive_header(conn
);
2488 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
2489 status
= relay_process_control_receive_payload(conn
);
2492 ERR("Unknown control connection protocol state encountered.");
2500 * Handle index for a data stream.
2502 * Called with the stream lock held.
2504 * Return 0 on success else a negative value.
2506 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2510 uint64_t data_offset
;
2511 struct relay_index
*index
;
2513 /* Get data offset because we are about to update the index. */
2514 data_offset
= htobe64(stream
->tracefile_size_current
);
2516 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
2517 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
2520 * Lookup for an existing index for that stream id/sequence
2521 * number. If it exists, the control thread has already received the
2522 * data for it, thus we need to write it to disk.
2524 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
2530 if (rotate_index
|| !stream
->index_file
) {
2531 uint32_t major
, minor
;
2533 /* Put ref on previous index_file. */
2534 if (stream
->index_file
) {
2535 lttng_index_file_put(stream
->index_file
);
2536 stream
->index_file
= NULL
;
2538 major
= stream
->trace
->session
->major
;
2539 minor
= stream
->trace
->session
->minor
;
2540 stream
->index_file
= lttng_index_file_create(stream
->path_name
,
2541 stream
->channel_name
,
2542 -1, -1, stream
->tracefile_size
,
2543 tracefile_array_get_file_index_head(stream
->tfa
),
2544 lttng_to_index_major(major
, minor
),
2545 lttng_to_index_minor(major
, minor
));
2546 if (!stream
->index_file
) {
2548 /* Put self-ref for this index due to error. */
2549 relay_index_put(index
);
2555 if (relay_index_set_file(index
, stream
->index_file
, data_offset
)) {
2557 /* Put self-ref for this index due to error. */
2558 relay_index_put(index
);
2563 ret
= relay_index_try_flush(index
);
2565 tracefile_array_commit_seq(stream
->tfa
);
2566 stream
->index_received_seqcount
++;
2567 } else if (ret
> 0) {
2571 /* Put self-ref for this index due to error. */
2572 relay_index_put(index
);
2580 static enum relay_connection_status
relay_process_data_receive_header(
2581 struct relay_connection
*conn
)
2584 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2585 struct data_connection_state_receive_header
*state
=
2586 &conn
->protocol
.data
.state
.receive_header
;
2587 struct lttcomm_relayd_data_hdr header
;
2588 struct relay_stream
*stream
;
2590 assert(state
->left_to_receive
!= 0);
2592 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
2593 state
->header_reception_buffer
+ state
->received
,
2594 state
->left_to_receive
, MSG_DONTWAIT
);
2596 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2597 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2598 status
= RELAY_CONNECTION_STATUS_ERROR
;
2601 } else if (ret
== 0) {
2602 /* Orderly shutdown. Not necessary to print an error. */
2603 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
2604 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2609 assert(ret
<= state
->left_to_receive
);
2611 state
->left_to_receive
-= ret
;
2612 state
->received
+= ret
;
2614 if (state
->left_to_receive
> 0) {
2616 * Can't transition to the protocol's next state, wait to
2617 * receive the rest of the header.
2619 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
2620 state
->received
, state
->left_to_receive
,
2626 /* Transition to next state: receiving the payload. */
2627 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
2629 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
2630 header
.circuit_id
= be64toh(header
.circuit_id
);
2631 header
.stream_id
= be64toh(header
.stream_id
);
2632 header
.data_size
= be32toh(header
.data_size
);
2633 header
.net_seq_num
= be64toh(header
.net_seq_num
);
2634 header
.padding_size
= be32toh(header
.padding_size
);
2635 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
2637 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
2639 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
2640 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
2642 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
2643 conn
->sock
->fd
, header
.circuit_id
,
2644 header
.stream_id
, header
.data_size
,
2645 header
.net_seq_num
, header
.padding_size
);
2647 stream
= stream_get_by_id(header
.stream_id
);
2649 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
2651 /* Protocol error. */
2652 status
= RELAY_CONNECTION_STATUS_ERROR
;
2656 pthread_mutex_lock(&stream
->lock
);
2658 /* Check if a rotation is needed. */
2659 if (stream
->tracefile_size
> 0 &&
2660 (stream
->tracefile_size_current
+ header
.data_size
) >
2661 stream
->tracefile_size
) {
2662 uint64_t old_id
, new_id
;
2664 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
2665 tracefile_array_file_rotate(stream
->tfa
);
2667 /* new_id is updated by utils_rotate_stream_file. */
2670 ret
= utils_rotate_stream_file(stream
->path_name
,
2671 stream
->channel_name
, stream
->tracefile_size
,
2672 stream
->tracefile_count
, -1,
2673 -1, stream
->stream_fd
->fd
,
2674 &new_id
, &stream
->stream_fd
->fd
);
2676 ERR("Failed to rotate stream output file");
2677 status
= RELAY_CONNECTION_STATUS_ERROR
;
2678 goto end_stream_unlock
;
2682 * Reset current size because we just performed a stream
2685 stream
->tracefile_size_current
= 0;
2686 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= true;
2691 pthread_mutex_unlock(&stream
->lock
);
2697 static enum relay_connection_status
relay_process_data_receive_payload(
2698 struct relay_connection
*conn
)
2701 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
2702 struct relay_stream
*stream
;
2703 struct data_connection_state_receive_payload
*state
=
2704 &conn
->protocol
.data
.state
.receive_payload
;
2705 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
2706 char data_buffer
[chunk_size
];
2707 bool partial_recv
= false;
2708 bool new_stream
= false, close_requested
= false;
2709 uint64_t left_to_receive
= state
->left_to_receive
;
2710 struct relay_session
*session
;
2712 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
2713 state
->header
.stream_id
, state
->header
.net_seq_num
,
2714 state
->received
, left_to_receive
);
2716 stream
= stream_get_by_id(state
->header
.stream_id
);
2718 /* Protocol error. */
2719 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
2720 state
->header
.stream_id
);
2721 status
= RELAY_CONNECTION_STATUS_ERROR
;
2725 pthread_mutex_lock(&stream
->lock
);
2726 session
= stream
->trace
->session
;
2727 if (!conn
->session
) {
2728 ret
= connection_set_session(conn
, session
);
2730 status
= RELAY_CONNECTION_STATUS_ERROR
;
2731 goto end_stream_unlock
;
2736 * The size of the "chunk" received on any iteration is bounded by:
2737 * - the data left to receive,
2738 * - the data immediately available on the socket,
2739 * - the on-stack data buffer
2741 while (left_to_receive
> 0 && !partial_recv
) {
2743 size_t recv_size
= min(left_to_receive
, chunk_size
);
2745 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
2746 recv_size
, MSG_DONTWAIT
);
2748 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
2749 PERROR("Socket %d error", conn
->sock
->fd
);
2750 status
= RELAY_CONNECTION_STATUS_ERROR
;
2752 goto end_stream_unlock
;
2753 } else if (ret
== 0) {
2754 /* No more data ready to be consumed on socket. */
2755 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
2756 state
->header
.stream_id
);
2757 status
= RELAY_CONNECTION_STATUS_CLOSED
;
2759 } else if (ret
< (int) recv_size
) {
2761 * All the data available on the socket has been
2764 partial_recv
= true;
2769 /* Write data to stream output fd. */
2770 write_ret
= lttng_write(stream
->stream_fd
->fd
, data_buffer
,
2772 if (write_ret
< (ssize_t
) recv_size
) {
2773 ERR("Relay error writing data to file");
2774 status
= RELAY_CONNECTION_STATUS_ERROR
;
2775 goto end_stream_unlock
;
2778 left_to_receive
-= recv_size
;
2779 state
->received
+= recv_size
;
2780 state
->left_to_receive
= left_to_receive
;
2782 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
2783 write_ret
, stream
->stream_handle
);
2786 if (state
->left_to_receive
> 0) {
2788 * Did not receive all the data expected, wait for more data to
2789 * become available on the socket.
2791 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
2792 state
->header
.stream_id
, state
->received
,
2793 state
->left_to_receive
);
2794 goto end_stream_unlock
;
2797 ret
= write_padding_to_file(stream
->stream_fd
->fd
,
2798 state
->header
.padding_size
);
2799 if ((int64_t) ret
< (int64_t) state
->header
.padding_size
) {
2800 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2801 stream
->stream_handle
,
2802 state
->header
.net_seq_num
, ret
);
2803 status
= RELAY_CONNECTION_STATUS_ERROR
;
2804 goto end_stream_unlock
;
2808 if (session
->minor
>= 4 && !session
->snapshot
) {
2809 ret
= handle_index_data(stream
, state
->header
.net_seq_num
,
2810 state
->rotate_index
);
2812 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2813 stream
->stream_handle
,
2814 state
->header
.net_seq_num
, ret
);
2815 status
= RELAY_CONNECTION_STATUS_ERROR
;
2816 goto end_stream_unlock
;
2820 stream
->tracefile_size_current
+= state
->header
.data_size
+
2821 state
->header
.padding_size
;
2823 if (stream
->prev_seq
== -1ULL) {
2827 stream
->prev_seq
= state
->header
.net_seq_num
;
2830 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
2831 * contents of *state which are aliased (union) to the same location as
2832 * the new state. Don't use it beyond this point.
2834 connection_reset_protocol_state(conn
);
2838 close_requested
= stream
->close_requested
;
2839 pthread_mutex_unlock(&stream
->lock
);
2840 if (close_requested
&& left_to_receive
== 0) {
2841 try_stream_close(stream
);
2845 pthread_mutex_lock(&session
->lock
);
2846 uatomic_set(&session
->new_streams
, 1);
2847 pthread_mutex_unlock(&session
->lock
);
2856 * relay_process_data: Process the data received on the data socket
2858 static enum relay_connection_status
relay_process_data(
2859 struct relay_connection
*conn
)
2861 enum relay_connection_status status
;
2863 switch (conn
->protocol
.data
.state_id
) {
2864 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
2865 status
= relay_process_data_receive_header(conn
);
2867 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
2868 status
= relay_process_data_receive_payload(conn
);
2871 ERR("Unexpected data connection communication state.");
2878 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2882 (void) lttng_poll_del(events
, pollfd
);
2884 ret
= close(pollfd
);
2886 ERR("Closing pollfd %d", pollfd
);
2890 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
2891 int pollfd
, struct relay_connection
*conn
)
2893 const char *type_str
;
2895 switch (conn
->type
) {
2900 type_str
= "Control";
2902 case RELAY_VIEWER_COMMAND
:
2903 type_str
= "Viewer Command";
2905 case RELAY_VIEWER_NOTIFICATION
:
2906 type_str
= "Viewer Notification";
2909 type_str
= "Unknown";
2911 cleanup_connection_pollfd(events
, pollfd
);
2912 connection_put(conn
);
2913 DBG("%s connection closed with %d", type_str
, pollfd
);
2917 * This thread does the actual work
2919 static void *relay_thread_worker(void *data
)
2921 int ret
, err
= -1, last_seen_data_fd
= -1;
2923 struct lttng_poll_event events
;
2924 struct lttng_ht
*relay_connections_ht
;
2925 struct lttng_ht_iter iter
;
2926 struct relay_connection
*destroy_conn
= NULL
;
2928 DBG("[thread] Relay worker started");
2930 rcu_register_thread();
2932 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2934 if (testpoint(relayd_thread_worker
)) {
2935 goto error_testpoint
;
2938 health_code_update();
2940 /* table of connections indexed on socket */
2941 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2942 if (!relay_connections_ht
) {
2943 goto relay_connections_ht_error
;
2946 ret
= create_named_thread_poll_set(&events
, 2, "Worker thread epoll");
2948 goto error_poll_create
;
2951 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2958 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2960 health_code_update();
2962 /* Infinite blocking call, waiting for transmission */
2963 DBG3("Relayd worker thread polling...");
2964 health_poll_entry();
2965 ret
= lttng_poll_wait(&events
, -1);
2969 * Restart interrupted system call.
2971 if (errno
== EINTR
) {
2980 * Process control. The control connection is
2981 * prioritized so we don't starve it with high
2982 * throughput tracing data on the data connection.
2984 for (i
= 0; i
< nb_fd
; i
++) {
2985 /* Fetch once the poll data */
2986 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2987 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2989 health_code_update();
2993 * No activity for this FD (poll
2999 /* Thread quit pipe has been closed. Killing thread. */
3000 ret
= check_thread_quit_pipe(pollfd
, revents
);
3006 /* Inspect the relay conn pipe for new connection */
3007 if (pollfd
== relay_conn_pipe
[0]) {
3008 if (revents
& LPOLLIN
) {
3009 struct relay_connection
*conn
;
3011 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3015 lttng_poll_add(&events
, conn
->sock
->fd
,
3016 LPOLLIN
| LPOLLRDHUP
);
3017 connection_ht_add(relay_connections_ht
, conn
);
3018 DBG("Connection socket %d added", conn
->sock
->fd
);
3019 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3020 ERR("Relay connection pipe error");
3023 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3027 struct relay_connection
*ctrl_conn
;
3029 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3030 /* If not found, there is a synchronization issue. */
3033 if (ctrl_conn
->type
== RELAY_DATA
) {
3034 if (revents
& LPOLLIN
) {
3036 * Flag the last seen data fd not deleted. It will be
3037 * used as the last seen fd if any fd gets deleted in
3040 last_notdel_data_fd
= pollfd
;
3042 goto put_ctrl_connection
;
3044 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3046 if (revents
& LPOLLIN
) {
3047 enum relay_connection_status status
;
3049 status
= relay_process_control(ctrl_conn
);
3050 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3052 * On socket error flag the session as aborted to force
3053 * the cleanup of its stream otherwise it can leak
3054 * during the lifetime of the relayd.
3056 * This prevents situations in which streams can be
3057 * left opened because an index was received, the
3058 * control connection is closed, and the data
3059 * connection is closed (uncleanly) before the packet's
3062 * Since the control connection encountered an error,
3063 * it is okay to be conservative and close the
3064 * session right now as we can't rely on the protocol
3065 * being respected anymore.
3067 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3068 session_abort(ctrl_conn
->session
);
3071 /* Clear the connection on error or close. */
3072 relay_thread_close_connection(&events
,
3077 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3078 relay_thread_close_connection(&events
,
3080 if (last_seen_data_fd
== pollfd
) {
3081 last_seen_data_fd
= last_notdel_data_fd
;
3084 ERR("Unexpected poll events %u for control sock %d",
3086 connection_put(ctrl_conn
);
3089 put_ctrl_connection
:
3090 connection_put(ctrl_conn
);
3095 * The last loop handled a control request, go back to poll to make
3096 * sure we prioritise the control socket.
3102 if (last_seen_data_fd
>= 0) {
3103 for (i
= 0; i
< nb_fd
; i
++) {
3104 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3106 health_code_update();
3108 if (last_seen_data_fd
== pollfd
) {
3115 /* Process data connection. */
3116 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3117 /* Fetch the poll data. */
3118 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3119 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3120 struct relay_connection
*data_conn
;
3122 health_code_update();
3125 /* No activity for this FD (poll implementation). */
3129 /* Skip the command pipe. It's handled in the first loop. */
3130 if (pollfd
== relay_conn_pipe
[0]) {
3134 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3136 /* Skip it. Might be removed before. */
3139 if (data_conn
->type
== RELAY_CONTROL
) {
3140 goto put_data_connection
;
3142 assert(data_conn
->type
== RELAY_DATA
);
3144 if (revents
& LPOLLIN
) {
3145 enum relay_connection_status status
;
3147 status
= relay_process_data(data_conn
);
3148 /* Connection closed or error. */
3149 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3151 * On socket error flag the session as aborted to force
3152 * the cleanup of its stream otherwise it can leak
3153 * during the lifetime of the relayd.
3155 * This prevents situations in which streams can be
3156 * left opened because an index was received, the
3157 * control connection is closed, and the data
3158 * connection is closed (uncleanly) before the packet's
3161 * Since the data connection encountered an error,
3162 * it is okay to be conservative and close the
3163 * session right now as we can't rely on the protocol
3164 * being respected anymore.
3166 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3167 session_abort(data_conn
->session
);
3169 relay_thread_close_connection(&events
, pollfd
,
3172 * Every goto restart call sets the last seen fd where
3173 * here we don't really care since we gracefully
3174 * continue the loop after the connection is deleted.
3177 /* Keep last seen port. */
3178 last_seen_data_fd
= pollfd
;
3179 connection_put(data_conn
);
3182 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3183 relay_thread_close_connection(&events
, pollfd
,
3186 ERR("Unknown poll events %u for data sock %d",
3189 put_data_connection
:
3190 connection_put(data_conn
);
3192 last_seen_data_fd
= -1;
3195 /* Normal exit, no error */
3200 /* Cleanup reamaining connection object. */
3202 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3205 health_code_update();
3207 session_abort(destroy_conn
->session
);
3210 * No need to grab another ref, because we own
3213 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
3218 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
3220 lttng_ht_destroy(relay_connections_ht
);
3221 relay_connections_ht_error
:
3222 /* Close relay conn pipes */
3223 (void) fd_tracker_util_pipe_close(the_fd_tracker
,
3226 DBG("Thread exited with error");
3228 DBG("Worker thread cleanup complete");
3232 ERR("Health error occurred in %s", __func__
);
3234 health_unregister(health_relayd
);
3235 rcu_unregister_thread();
3236 lttng_relay_stop_threads();
3241 * Create the relay command pipe to wake thread_manage_apps.
3242 * Closed in cleanup().
3244 static int create_relay_conn_pipe(void)
3246 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
3247 "Relayd connection pipe", relay_conn_pipe
);
3251 int stdio_open(void *data
, int *fds
)
3253 fds
[0] = fileno(stdout
);
3254 fds
[1] = fileno(stderr
);
3259 int noop_close(void *data
, int *fds
)
3265 int track_stdio(void)
3268 const char *names
[] = { "stdout", "stderr" };
3270 return fd_tracker_open_unsuspendable_fd(the_fd_tracker
, fds
,
3271 names
, 2, stdio_open
, NULL
);
3275 void untrack_stdio(void)
3277 int fds
[] = { fileno(stdout
), fileno(stderr
) };
3280 * noop_close is used since we don't really want to close
3281 * the stdio output fds; we merely want to stop tracking them.
3283 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
3284 fds
, 2, noop_close
, NULL
);
3290 int main(int argc
, char **argv
)
3292 int ret
= 0, retval
= 0;
3295 /* Parse environment variables */
3296 parse_env_options();
3300 * Command line arguments overwrite environment.
3303 if (set_options(argc
, argv
)) {
3308 if (set_signal_handler()) {
3313 relayd_config_log();
3315 if (opt_print_version
) {
3321 ret
= fclose(stdin
);
3323 PERROR("Failed to close stdin");
3326 /* Try to create directory if -o, --output is specified. */
3327 if (opt_output_path
) {
3328 if (*opt_output_path
!= '/') {
3329 ERR("Please specify an absolute path for -o, --output PATH");
3334 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
3337 ERR("Unable to create %s", opt_output_path
);
3344 if (opt_daemon
|| opt_background
) {
3345 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
3353 if (opt_working_directory
) {
3354 ret
= utils_change_working_dir(opt_working_directory
);
3356 ERR("Changing working directory");
3361 * The RCU thread registration (and use, through the fd-tracker's
3362 * creation) is done after the daemonization to allow us to not
3363 * deal with liburcu's fork() management as the call RCU needs to
3366 rcu_register_thread();
3368 the_fd_tracker
= fd_tracker_create(lttng_opt_fd_cap
);
3369 if (!the_fd_tracker
) {
3374 ret
= track_stdio();
3380 /* Initialize thread health monitoring */
3381 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
3382 if (!health_relayd
) {
3383 PERROR("health_app_create error");
3385 goto exit_health_app_create
;
3388 /* Create thread quit pipe */
3389 if (init_thread_quit_pipe()) {
3391 goto exit_init_data
;
3394 /* Setup the thread apps communication pipe. */
3395 if (create_relay_conn_pipe()) {
3397 goto exit_init_data
;
3400 /* Init relay command queue. */
3401 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
3403 /* Initialize communication library */
3405 lttcomm_inet_init();
3407 /* tables of sessions indexed by session ID */
3408 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3411 goto exit_init_data
;
3414 /* tables of streams indexed by stream ID */
3415 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3416 if (!relay_streams_ht
) {
3418 goto exit_init_data
;
3421 /* tables of streams indexed by stream ID */
3422 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3423 if (!viewer_streams_ht
) {
3425 goto exit_init_data
;
3428 ret
= init_health_quit_pipe();
3431 goto exit_health_quit_pipe
;
3434 /* Create thread to manage the client socket */
3435 ret
= pthread_create(&health_thread
, default_pthread_attr(),
3436 thread_manage_health
, (void *) NULL
);
3439 PERROR("pthread_create health");
3441 goto exit_health_thread
;
3444 /* Setup the dispatcher thread */
3445 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
3446 relay_thread_dispatcher
, (void *) NULL
);
3449 PERROR("pthread_create dispatcher");
3451 goto exit_dispatcher_thread
;
3454 /* Setup the worker thread */
3455 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
3456 relay_thread_worker
, NULL
);
3459 PERROR("pthread_create worker");
3461 goto exit_worker_thread
;
3464 /* Setup the listener thread */
3465 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
3466 relay_thread_listener
, (void *) NULL
);
3469 PERROR("pthread_create listener");
3471 goto exit_listener_thread
;
3474 ret
= relayd_live_create(live_uri
);
3476 ERR("Starting live viewer threads");
3482 * This is where we start awaiting program completion (e.g. through
3483 * signal that asks threads to teardown).
3486 ret
= relayd_live_join();
3492 ret
= pthread_join(listener_thread
, &status
);
3495 PERROR("pthread_join listener_thread");
3499 exit_listener_thread
:
3500 ret
= pthread_join(worker_thread
, &status
);
3503 PERROR("pthread_join worker_thread");
3508 ret
= pthread_join(dispatcher_thread
, &status
);
3511 PERROR("pthread_join dispatcher_thread");
3514 exit_dispatcher_thread
:
3516 ret
= pthread_join(health_thread
, &status
);
3519 PERROR("pthread_join health_thread");
3524 (void) fd_tracker_util_pipe_close(the_fd_tracker
, health_quit_pipe
);
3525 exit_health_quit_pipe
:
3528 health_app_destroy(health_relayd
);
3529 exit_health_app_create
:
3532 * Wait for all pending call_rcu work to complete before tearing
3533 * down data structures. call_rcu worker may be trying to
3534 * perform lookups in those structures.
3539 /* Ensure all prior call_rcu are done. */
3544 * fd_tracker_destroy() will log the contents of the fd-tracker
3545 * if a leak is detected.
3547 fd_tracker_destroy(the_fd_tracker
);
3548 rcu_unregister_thread();