2 * Copyright (C) 2012 Julien Desfossez <jdesfossez@efficios.com>
3 * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
4 * Copyright (C) 2013 Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * Copyright (C) 2015 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * SPDX-License-Identifier: GPL-2.0-only
21 #include <sys/mount.h>
22 #include <sys/resource.h>
23 #include <sys/socket.h>
25 #include <sys/types.h>
27 #include <sys/resource.h>
29 #include <urcu/futex.h>
30 #include <urcu/uatomic.h>
31 #include <urcu/rculist.h>
38 #include <lttng/lttng.h>
39 #include <common/common.h>
40 #include <common/compat/poll.h>
41 #include <common/compat/socket.h>
42 #include <common/compat/endian.h>
43 #include <common/compat/getenv.h>
44 #include <common/defaults.h>
45 #include <common/daemonize.h>
46 #include <common/futex.h>
47 #include <common/sessiond-comm/sessiond-comm.h>
48 #include <common/sessiond-comm/inet.h>
49 #include <common/sessiond-comm/relayd.h>
50 #include <common/uri.h>
51 #include <common/utils.h>
52 #include <common/path.h>
53 #include <common/align.h>
54 #include <common/ini-config/ini-config.h>
55 #include <common/dynamic-buffer.h>
56 #include <common/buffer-view.h>
57 #include <common/string-utils/format.h>
58 #include <common/fd-tracker/fd-tracker.h>
59 #include <common/fd-tracker/utils.h>
61 #include "backward-compatibility-group-by.h"
63 #include "connection.h"
64 #include "ctf-trace.h"
65 #include "health-relayd.h"
68 #include "lttng-relayd.h"
70 #include "sessiond-trace-chunks.h"
72 #include "tcp_keep_alive.h"
73 #include "testpoint.h"
74 #include "tracefile-array.h"
77 #include "viewer-stream.h"
79 static const char *help_msg
=
80 #ifdef LTTNG_EMBED_HELP
81 #include <lttng-relayd.8.h>
87 enum relay_connection_status
{
88 RELAY_CONNECTION_STATUS_OK
,
89 /* An error occurred while processing an event on the connection. */
90 RELAY_CONNECTION_STATUS_ERROR
,
91 /* Connection closed/shutdown cleanly. */
92 RELAY_CONNECTION_STATUS_CLOSED
,
95 /* command line options */
96 char *opt_output_path
, *opt_working_directory
;
97 static int opt_daemon
, opt_background
, opt_print_version
, opt_allow_clear
= 1;
98 enum relay_group_output_by opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
;
100 /* Argument variables */
101 int lttng_opt_quiet
; /* not static in error.h */
102 int lttng_opt_verbose
; /* not static in error.h */
103 int lttng_opt_mi
; /* not static in error.h */
106 * We need to wait for listener and live listener threads, as well as
107 * health check thread, before being ready to signal readiness.
109 #define NR_LTTNG_RELAY_READY 3
110 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
112 /* Size of receive buffer. */
113 #define RECV_DATA_BUFFER_SIZE 65536
115 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
116 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
118 static struct lttng_uri
*control_uri
;
119 static struct lttng_uri
*data_uri
;
120 static struct lttng_uri
*live_uri
;
122 const char *progname
;
124 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
125 static int tracing_group_name_override
;
127 const char * const config_section_name
= "relayd";
130 * Quit pipe for all threads. This permits a single cancellation point
131 * for all threads when receiving an event on the pipe.
133 int thread_quit_pipe
[2] = { -1, -1 };
136 * This pipe is used to inform the worker thread that a command is queued and
137 * ready to be processed.
139 static int relay_conn_pipe
[2] = { -1, -1 };
141 /* Shared between threads */
142 static int dispatch_thread_exit
;
144 static pthread_t listener_thread
;
145 static pthread_t dispatcher_thread
;
146 static pthread_t worker_thread
;
147 static pthread_t health_thread
;
150 * last_relay_stream_id_lock protects last_relay_stream_id increment
151 * atomicity on 32-bit architectures.
153 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
154 static uint64_t last_relay_stream_id
;
157 * Relay command queue.
159 * The relay_thread_listener and relay_thread_dispatcher communicate with this
162 static struct relay_conn_queue relay_conn_queue
;
164 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
165 static unsigned int lttng_opt_fd_pool_size
= -1;
167 /* Global relay stream hash table. */
168 struct lttng_ht
*relay_streams_ht
;
170 /* Global relay viewer stream hash table. */
171 struct lttng_ht
*viewer_streams_ht
;
173 /* Global relay sessions hash table. */
174 struct lttng_ht
*sessions_ht
;
176 /* Relayd health monitoring */
177 struct health_app
*health_relayd
;
179 struct sessiond_trace_chunk_registry
*sessiond_trace_chunk_registry
;
181 /* Global fd tracker. */
182 struct fd_tracker
*the_fd_tracker
;
184 static struct option long_options
[] = {
185 { "control-port", 1, 0, 'C', },
186 { "data-port", 1, 0, 'D', },
187 { "live-port", 1, 0, 'L', },
188 { "daemonize", 0, 0, 'd', },
189 { "background", 0, 0, 'b', },
190 { "group", 1, 0, 'g', },
191 { "fd-pool-size", 1, 0, '\0', },
192 { "help", 0, 0, 'h', },
193 { "output", 1, 0, 'o', },
194 { "verbose", 0, 0, 'v', },
195 { "config", 1, 0, 'f' },
196 { "version", 0, 0, 'V' },
197 { "working-directory", 1, 0, 'w', },
198 { "group-output-by-session", 0, 0, 's', },
199 { "group-output-by-host", 0, 0, 'p', },
200 { "disallow-clear", 0, 0, 'x' },
204 static const char *config_ignore_options
[] = { "help", "config", "version" };
206 static void print_version(void) {
207 fprintf(stdout
, "%s\n", VERSION
);
210 static void relayd_config_log(void)
212 DBG("LTTng-relayd " VERSION
" - " VERSION_NAME
"%s%s",
213 GIT_VERSION
[0] == '\0' ? "" : " - " GIT_VERSION
,
214 EXTRA_VERSION_NAME
[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME
);
215 if (EXTRA_VERSION_DESCRIPTION
[0] != '\0') {
216 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION
"\n");
218 if (EXTRA_VERSION_PATCHES
[0] != '\0') {
219 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES
"\n");
224 * Take an option from the getopt output and set it in the right variable to be
227 * Return 0 on success else a negative value.
229 static int set_option(int opt
, const char *arg
, const char *optname
)
235 if (!strcmp(optname
, "fd-pool-size")) {
239 v
= strtoul(arg
, NULL
, 0);
240 if (errno
!= 0 || !isdigit((unsigned char) arg
[0])) {
241 ERR("Wrong value in --fd-pool-size parameter: %s", arg
);
246 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s", arg
);
250 lttng_opt_fd_pool_size
= (unsigned int) v
;
252 fprintf(stderr
, "unknown option %s", optname
);
254 fprintf(stderr
, " with arg %s\n", arg
);
259 if (lttng_is_setuid_setgid()) {
260 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
261 "-C, --control-port");
263 ret
= uri_parse(arg
, &control_uri
);
265 ERR("Invalid control URI specified");
268 if (control_uri
->port
== 0) {
269 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
274 if (lttng_is_setuid_setgid()) {
275 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
278 ret
= uri_parse(arg
, &data_uri
);
280 ERR("Invalid data URI specified");
283 if (data_uri
->port
== 0) {
284 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
289 if (lttng_is_setuid_setgid()) {
290 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
293 ret
= uri_parse(arg
, &live_uri
);
295 ERR("Invalid live URI specified");
298 if (live_uri
->port
== 0) {
299 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
310 if (lttng_is_setuid_setgid()) {
311 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
314 tracing_group_name
= strdup(arg
);
315 if (tracing_group_name
== NULL
) {
320 tracing_group_name_override
= 1;
324 ret
= utils_show_help(8, "lttng-relayd", help_msg
);
326 ERR("Cannot show --help for `lttng-relayd`");
331 opt_print_version
= 1;
334 if (lttng_is_setuid_setgid()) {
335 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
338 ret
= asprintf(&opt_output_path
, "%s", arg
);
341 PERROR("asprintf opt_output_path");
347 if (lttng_is_setuid_setgid()) {
348 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
349 "-w, --working-directory");
351 ret
= asprintf(&opt_working_directory
, "%s", arg
);
354 PERROR("asprintf opt_working_directory");
361 /* Verbose level can increase using multiple -v */
363 lttng_opt_verbose
= config_parse_value(arg
);
365 /* Only 3 level of verbosity (-vvv). */
366 if (lttng_opt_verbose
< 3) {
367 lttng_opt_verbose
+= 1;
372 if (opt_group_output_by
!= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
373 ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
376 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_SESSION
;
379 if (opt_group_output_by
!= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
380 ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
383 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_HOST
;
390 /* Unknown option or other error.
391 * Error is printed by getopt, just return */
404 * config_entry_handler_cb used to handle options read from a config file.
405 * See config_entry_handler_cb comment in common/config/session-config.h for the
406 * return value conventions.
408 static int config_entry_handler(const struct config_entry
*entry
,
409 void *unused
__attribute__((unused
)))
413 if (!entry
|| !entry
->name
|| !entry
->value
) {
418 /* Check if the option is to be ignored */
419 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
420 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
425 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
426 /* Ignore if entry name is not fully matched. */
427 if (strcmp(entry
->name
, long_options
[i
].name
)) {
432 * If the option takes no argument on the command line,
433 * we have to check if the value is "true". We support
434 * non-zero numeric values, true, on and yes.
436 if (!long_options
[i
].has_arg
) {
437 ret
= config_parse_value(entry
->value
);
440 WARN("Invalid configuration value \"%s\" for option %s",
441 entry
->value
, entry
->name
);
443 /* False, skip boolean config option. */
448 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
452 WARN("Unrecognized option \"%s\" in daemon configuration file.",
459 static int parse_env_options(void)
464 value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV
);
466 opt_working_directory
= strdup(value
);
467 if (!opt_working_directory
) {
468 ERR("Failed to allocate working directory string (\"%s\")",
476 static int set_fd_pool_size(void)
479 struct rlimit rlimit
;
481 ret
= getrlimit(RLIMIT_NOFILE
, &rlimit
);
483 PERROR("Failed to get file descriptor limit");
488 DBG("File descriptor count limits are %" PRIu64
" (soft) and %" PRIu64
" (hard)",
489 (uint64_t) rlimit
.rlim_cur
,
490 (uint64_t) rlimit
.rlim_max
);
491 if (lttng_opt_fd_pool_size
== -1) {
492 /* Use default value (soft limit - reserve). */
493 if (rlimit
.rlim_cur
< DEFAULT_RELAYD_MIN_FD_POOL_SIZE
) {
494 ERR("The process' file number limit is too low (%" PRIu64
"). The process' file number limit must be set to at least %i.",
495 (uint64_t) rlimit
.rlim_cur
, DEFAULT_RELAYD_MIN_FD_POOL_SIZE
);
499 lttng_opt_fd_pool_size
= rlimit
.rlim_cur
-
500 DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
;
504 if (lttng_opt_fd_pool_size
< DEFAULT_RELAYD_MIN_FD_POOL_SIZE
) {
505 ERR("File descriptor pool size must be set to at least %d",
506 DEFAULT_RELAYD_MIN_FD_POOL_SIZE
);
511 if (lttng_opt_fd_pool_size
> rlimit
.rlim_cur
) {
512 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64
").",
513 lttng_opt_fd_pool_size
, (uint64_t) rlimit
.rlim_cur
);
518 DBG("File descriptor pool size argument (%u) adjusted to %u to accommodates transient fd uses",
519 lttng_opt_fd_pool_size
,
520 lttng_opt_fd_pool_size
- DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
);
521 lttng_opt_fd_pool_size
-= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
;
526 static int set_options(int argc
, char **argv
)
528 int c
, ret
= 0, option_index
= 0, retval
= 0;
529 int orig_optopt
= optopt
, orig_optind
= optind
;
530 char *default_address
, *optstring
;
531 char *config_path
= NULL
;
533 optstring
= utils_generate_optstring(long_options
,
534 sizeof(long_options
) / sizeof(struct option
));
540 /* Check for the --config option */
542 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
543 &option_index
)) != -1) {
547 } else if (c
!= 'f') {
551 if (lttng_is_setuid_setgid()) {
552 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
556 config_path
= utils_expand_path(optarg
);
558 ERR("Failed to resolve path: %s", optarg
);
563 ret
= config_get_section_entries(config_path
, config_section_name
,
564 config_entry_handler
, NULL
);
567 ERR("Invalid configuration option at line %i", ret
);
573 /* Reset getopt's global state */
574 optopt
= orig_optopt
;
575 optind
= orig_optind
;
577 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
582 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
589 /* assign default values */
590 if (control_uri
== NULL
) {
591 ret
= asprintf(&default_address
,
592 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
593 DEFAULT_NETWORK_CONTROL_PORT
);
595 PERROR("asprintf default data address");
600 ret
= uri_parse(default_address
, &control_uri
);
601 free(default_address
);
603 ERR("Invalid control URI specified");
608 if (data_uri
== NULL
) {
609 ret
= asprintf(&default_address
,
610 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
611 DEFAULT_NETWORK_DATA_PORT
);
613 PERROR("asprintf default data address");
618 ret
= uri_parse(default_address
, &data_uri
);
619 free(default_address
);
621 ERR("Invalid data URI specified");
626 if (live_uri
== NULL
) {
627 ret
= asprintf(&default_address
,
628 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
629 DEFAULT_NETWORK_VIEWER_PORT
);
631 PERROR("asprintf default viewer control address");
636 ret
= uri_parse(default_address
, &live_uri
);
637 free(default_address
);
639 ERR("Invalid viewer control URI specified");
644 ret
= set_fd_pool_size();
650 if (opt_group_output_by
== RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
651 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_HOST
;
653 if (opt_allow_clear
) {
654 /* Check if env variable exists. */
655 const char *value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV
);
657 ret
= config_parse_value(value
);
659 ERR("Invalid value for %s specified", DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV
);
663 opt_allow_clear
= !ret
;
673 static void print_global_objects(void)
675 print_viewer_streams();
676 print_relay_streams();
680 static int noop_close(void *data
__attribute__((unused
)),
681 int *fds
__attribute__((unused
)))
686 static void untrack_stdio(void)
688 int fds
[] = { fileno(stdout
), fileno(stderr
) };
691 * noop_close is used since we don't really want to close
692 * the stdio output fds; we merely want to stop tracking them.
694 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
695 fds
, 2, noop_close
, NULL
);
701 static void relayd_cleanup(void)
703 print_global_objects();
707 if (viewer_streams_ht
)
708 lttng_ht_destroy(viewer_streams_ht
);
709 if (relay_streams_ht
)
710 lttng_ht_destroy(relay_streams_ht
);
712 lttng_ht_destroy(sessions_ht
);
714 free(opt_output_path
);
715 free(opt_working_directory
);
718 health_app_destroy(health_relayd
);
720 /* Close thread quit pipes */
721 if (health_quit_pipe
[0] != -1) {
722 (void) fd_tracker_util_pipe_close(
723 the_fd_tracker
, health_quit_pipe
);
725 if (thread_quit_pipe
[0] != -1) {
726 (void) fd_tracker_util_pipe_close(
727 the_fd_tracker
, thread_quit_pipe
);
729 if (sessiond_trace_chunk_registry
) {
730 sessiond_trace_chunk_registry_destroy(
731 sessiond_trace_chunk_registry
);
733 if (the_fd_tracker
) {
736 * fd_tracker_destroy() will log the contents of the fd-tracker
737 * if a leak is detected.
739 fd_tracker_destroy(the_fd_tracker
);
742 uri_free(control_uri
);
744 /* Live URI is freed in the live thread. */
746 if (tracing_group_name_override
) {
747 free((void *) tracing_group_name
);
752 * Write to writable pipe used to notify a thread.
754 static int notify_thread_pipe(int wpipe
)
758 ret
= lttng_write(wpipe
, "!", 1);
760 PERROR("write poll pipe");
768 static int notify_health_quit_pipe(int *pipe
)
772 ret
= lttng_write(pipe
[1], "4", 1);
774 PERROR("write relay health quit");
783 * Stop all relayd and relayd-live threads.
785 int lttng_relay_stop_threads(void)
789 /* Stopping all threads */
790 DBG("Terminating all threads");
791 if (notify_thread_pipe(thread_quit_pipe
[1])) {
792 ERR("write error on thread quit pipe");
796 if (notify_health_quit_pipe(health_quit_pipe
)) {
797 ERR("write error on health quit pipe");
800 /* Dispatch thread */
801 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
802 futex_nto1_wake(&relay_conn_queue
.futex
);
804 if (relayd_live_stop()) {
805 ERR("Error stopping live threads");
812 * Signal handler for the daemon
814 * Simply stop all worker threads, leaving main() return gracefully after
815 * joining all threads and calling cleanup().
817 static void sighandler(int sig
)
821 DBG("SIGINT caught");
822 if (lttng_relay_stop_threads()) {
823 ERR("Error stopping threads");
827 DBG("SIGTERM caught");
828 if (lttng_relay_stop_threads()) {
829 ERR("Error stopping threads");
833 CMM_STORE_SHARED(recv_child_signal
, 1);
841 * Setup signal handler for :
842 * SIGINT, SIGTERM, SIGPIPE
844 static int set_signal_handler(void)
850 if ((ret
= sigemptyset(&sigset
)) < 0) {
851 PERROR("sigemptyset");
858 sa
.sa_handler
= sighandler
;
859 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
864 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
869 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
874 sa
.sa_handler
= SIG_IGN
;
875 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
880 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
885 void lttng_relay_notify_ready(void)
887 /* Notify the parent of the fork() process that we are ready. */
888 if (opt_daemon
|| opt_background
) {
889 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
890 kill(child_ppid
, SIGUSR1
);
896 * Init thread quit pipe.
898 * Return -1 on error or 0 if all pipes are created.
900 static int init_thread_quit_pipe(void)
902 return fd_tracker_util_pipe_open_cloexec(
903 the_fd_tracker
, "Quit pipe", thread_quit_pipe
);
907 * Init health quit pipe.
909 * Return -1 on error or 0 if all pipes are created.
911 static int init_health_quit_pipe(void)
913 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
914 "Health quit pipe", health_quit_pipe
);
918 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
920 static int create_named_thread_poll_set(struct lttng_poll_event
*events
,
921 int size
, const char *name
)
925 if (events
== NULL
|| size
== 0) {
930 ret
= fd_tracker_util_poll_create(the_fd_tracker
,
931 name
, events
, 1, LTTNG_CLOEXEC
);
933 PERROR("Failed to create \"%s\" poll file descriptor", name
);
938 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
950 * Check if the thread quit pipe was triggered.
952 * Return 1 if it was triggered else 0;
954 static int check_thread_quit_pipe(int fd
, uint32_t events
)
956 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
963 static int create_sock(void *data
, int *out_fd
)
966 struct lttcomm_sock
*sock
= (lttcomm_sock
*) data
;
968 ret
= lttcomm_create_sock(sock
);
978 static int close_sock(void *data
, int *in_fd
__attribute__((unused
)))
980 struct lttcomm_sock
*sock
= (lttcomm_sock
*) data
;
982 return sock
->ops
->close(sock
);
985 static int accept_sock(void *data
, int *out_fd
)
988 /* Socks is an array of in_sock, out_sock. */
989 struct lttcomm_sock
**socks
= (lttcomm_sock
**) data
;
990 struct lttcomm_sock
*in_sock
= socks
[0];
992 socks
[1] = in_sock
->ops
->accept(in_sock
);
997 *out_fd
= socks
[1]->fd
;
1003 * Create and init socket from uri.
1005 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
,
1009 struct lttcomm_sock
*sock
= NULL
;
1010 char uri_str
[PATH_MAX
];
1011 char *formated_name
= NULL
;
1013 sock
= lttcomm_alloc_sock_from_uri(uri
);
1015 ERR("Allocating socket");
1020 * Don't fail to create the socket if the name can't be built as it is
1021 * only used for debugging purposes.
1023 ret
= uri_to_str_url(uri
, uri_str
, sizeof(uri_str
));
1024 uri_str
[sizeof(uri_str
) - 1] = '\0';
1026 ret
= asprintf(&formated_name
, "%s socket @ %s", name
,
1029 formated_name
= NULL
;
1033 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &sock_fd
,
1034 (const char **) (formated_name
? &formated_name
: NULL
),
1035 1, create_sock
, sock
);
1037 PERROR("Failed to open \"%s\" relay socket",
1038 formated_name
?: "Unknown");
1041 DBG("Listening on %s socket %d", name
, sock
->fd
);
1043 ret
= sock
->ops
->bind(sock
);
1045 PERROR("Failed to bind socket");
1049 ret
= sock
->ops
->listen(sock
, -1);
1055 free(formated_name
);
1060 lttcomm_destroy_sock(sock
);
1062 free(formated_name
);
1067 struct lttcomm_sock
*accept_relayd_sock(struct lttcomm_sock
*listening_sock
,
1071 struct lttcomm_sock
*socks
[2] = { listening_sock
, NULL
};
1072 struct lttcomm_sock
*new_sock
= NULL
;
1074 ret
= fd_tracker_open_unsuspendable_fd(
1075 the_fd_tracker
, &out_fd
,
1076 (const char **) &name
,
1077 1, accept_sock
, &socks
);
1081 new_sock
= socks
[1];
1082 DBG("%s accepted, socket %d", name
, new_sock
->fd
);
1088 * This thread manages the listening for new connections on the network
1090 static void *relay_thread_listener(void *data
__attribute__((unused
)))
1092 int i
, ret
, pollfd
, err
= -1;
1093 uint32_t revents
, nb_fd
;
1094 struct lttng_poll_event events
;
1095 struct lttcomm_sock
*control_sock
, *data_sock
;
1097 DBG("[thread] Relay listener started");
1099 rcu_register_thread();
1100 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
1102 health_code_update();
1104 control_sock
= relay_socket_create(control_uri
, "Control listener");
1105 if (!control_sock
) {
1106 goto error_sock_control
;
1109 data_sock
= relay_socket_create(data_uri
, "Data listener");
1111 goto error_sock_relay
;
1115 * Pass 3 as size here for the thread quit pipe, control and
1118 ret
= create_named_thread_poll_set(&events
, 3, "Listener thread epoll");
1120 goto error_create_poll
;
1123 /* Add the control socket */
1124 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1126 goto error_poll_add
;
1129 /* Add the data socket */
1130 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1132 goto error_poll_add
;
1135 lttng_relay_notify_ready();
1137 if (testpoint(relayd_thread_listener
)) {
1138 goto error_testpoint
;
1142 health_code_update();
1144 DBG("Listener accepting connections");
1147 health_poll_entry();
1148 ret
= lttng_poll_wait(&events
, -1);
1152 * Restart interrupted system call.
1154 if (errno
== EINTR
) {
1162 DBG("Relay new connection received");
1163 for (i
= 0; i
< nb_fd
; i
++) {
1164 health_code_update();
1166 /* Fetch once the poll data */
1167 revents
= LTTNG_POLL_GETEV(&events
, i
);
1168 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1170 /* Thread quit pipe has been closed. Killing thread. */
1171 ret
= check_thread_quit_pipe(pollfd
, revents
);
1177 if (revents
& LPOLLIN
) {
1179 * A new connection is requested, therefore a
1180 * sessiond/consumerd connection is allocated in
1181 * this thread, enqueued to a global queue and
1182 * dequeued (and freed) in the worker thread.
1185 struct relay_connection
*new_conn
;
1186 struct lttcomm_sock
*newsock
= NULL
;
1187 enum connection_type type
;
1189 if (pollfd
== data_sock
->fd
) {
1191 newsock
= accept_relayd_sock(data_sock
,
1192 "Data socket to relayd");
1194 LTTNG_ASSERT(pollfd
== control_sock
->fd
);
1195 type
= RELAY_CONTROL
;
1196 newsock
= accept_relayd_sock(control_sock
,
1197 "Control socket to relayd");
1200 PERROR("accepting sock");
1204 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
1207 PERROR("setsockopt inet");
1208 lttcomm_destroy_sock(newsock
);
1212 ret
= socket_apply_keep_alive_config(newsock
->fd
);
1214 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1216 lttcomm_destroy_sock(newsock
);
1220 new_conn
= connection_create(newsock
, type
);
1222 lttcomm_destroy_sock(newsock
);
1226 /* Enqueue request for the dispatcher thread. */
1227 cds_wfcq_head_ptr_t head
;
1228 head
.h
= &relay_conn_queue
.head
;
1229 cds_wfcq_enqueue(head
, &relay_conn_queue
.tail
,
1233 * Wake the dispatch queue futex.
1234 * Implicit memory barrier with the
1235 * exchange in cds_wfcq_enqueue.
1237 futex_nto1_wake(&relay_conn_queue
.futex
);
1238 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1239 ERR("socket poll error");
1242 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1252 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
1254 if (data_sock
->fd
>= 0) {
1255 int data_sock_fd
= data_sock
->fd
;
1257 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1258 &data_sock_fd
, 1, close_sock
,
1261 PERROR("Failed to close the data listener socket file descriptor");
1265 lttcomm_destroy_sock(data_sock
);
1267 if (control_sock
->fd
>= 0) {
1268 int control_sock_fd
= control_sock
->fd
;
1270 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1271 &control_sock_fd
, 1, close_sock
,
1274 PERROR("Failed to close the control listener socket file descriptor");
1276 control_sock
->fd
= -1;
1278 lttcomm_destroy_sock(control_sock
);
1282 ERR("Health error occurred in %s", __func__
);
1284 health_unregister(health_relayd
);
1285 rcu_unregister_thread();
1286 DBG("Relay listener thread cleanup complete");
1287 lttng_relay_stop_threads();
1292 * This thread manages the dispatching of the requests to worker threads
1294 static void *relay_thread_dispatcher(void *data
__attribute__((unused
)))
1298 struct cds_wfcq_node
*node
;
1299 struct relay_connection
*new_conn
= NULL
;
1301 DBG("[thread] Relay dispatcher started");
1303 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
1305 if (testpoint(relayd_thread_dispatcher
)) {
1306 goto error_testpoint
;
1309 health_code_update();
1312 health_code_update();
1314 /* Atomically prepare the queue futex */
1315 futex_nto1_prepare(&relay_conn_queue
.futex
);
1317 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1322 health_code_update();
1324 /* Dequeue commands */
1325 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1326 &relay_conn_queue
.tail
);
1328 DBG("Woken up but nothing in the relay command queue");
1329 /* Continue thread execution */
1332 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1334 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1337 * Inform worker thread of the new request. This
1338 * call is blocking so we can be assured that
1339 * the data will be read at some point in time
1340 * or wait to the end of the world :)
1342 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1344 PERROR("write connection pipe");
1345 connection_put(new_conn
);
1348 } while (node
!= NULL
);
1350 /* Futex wait on queue. Blocking call on futex() */
1351 health_poll_entry();
1352 futex_nto1_wait(&relay_conn_queue
.futex
);
1356 /* Normal exit, no error */
1363 ERR("Health error occurred in %s", __func__
);
1365 health_unregister(health_relayd
);
1366 DBG("Dispatch thread dying");
1367 lttng_relay_stop_threads();
1371 static bool session_streams_have_index(const struct relay_session
*session
)
1373 return session
->minor
>= 4 && !session
->snapshot
;
1377 * Handle the RELAYD_CREATE_SESSION command.
1379 * On success, send back the session id or else return a negative value.
1381 static int relay_create_session(
1382 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
1383 struct relay_connection
*conn
,
1384 const struct lttng_buffer_view
*payload
)
1388 struct relay_session
*session
= NULL
;
1389 struct lttcomm_relayd_create_session_reply_2_11 reply
= {};
1390 char session_name
[LTTNG_NAME_MAX
] = {};
1391 char hostname
[LTTNG_HOST_NAME_MAX
] = {};
1392 uint32_t live_timer
= 0;
1393 bool snapshot
= false;
1394 bool session_name_contains_creation_timestamp
= false;
1395 /* Left nil for peers < 2.11. */
1396 char base_path
[LTTNG_PATH_MAX
] = {};
1397 lttng_uuid sessiond_uuid
= {};
1398 LTTNG_OPTIONAL(uint64_t) id_sessiond
= {};
1399 LTTNG_OPTIONAL(uint64_t) current_chunk_id
= {};
1400 LTTNG_OPTIONAL(time_t) creation_time
= {};
1401 struct lttng_dynamic_buffer reply_payload
;
1403 lttng_dynamic_buffer_init(&reply_payload
);
1405 if (conn
->minor
< 4) {
1406 /* From 2.1 to 2.3 */
1408 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1409 /* From 2.4 to 2.10 */
1410 ret
= cmd_create_session_2_4(payload
, session_name
,
1411 hostname
, &live_timer
, &snapshot
);
1413 bool has_current_chunk
;
1414 uint64_t current_chunk_id_value
;
1415 time_t creation_time_value
;
1416 uint64_t id_sessiond_value
;
1418 /* From 2.11 to ... */
1419 ret
= cmd_create_session_2_11(payload
, session_name
, hostname
,
1420 base_path
, &live_timer
, &snapshot
, &id_sessiond_value
,
1421 sessiond_uuid
, &has_current_chunk
,
1422 ¤t_chunk_id_value
, &creation_time_value
,
1423 &session_name_contains_creation_timestamp
);
1424 if (lttng_uuid_is_nil(sessiond_uuid
)) {
1425 /* The nil UUID is reserved for pre-2.11 clients. */
1426 ERR("Illegal nil UUID announced by peer in create session command");
1430 LTTNG_OPTIONAL_SET(&id_sessiond
, id_sessiond_value
);
1431 LTTNG_OPTIONAL_SET(&creation_time
, creation_time_value
);
1432 if (has_current_chunk
) {
1433 LTTNG_OPTIONAL_SET(¤t_chunk_id
,
1434 current_chunk_id_value
);
1442 session
= session_create(session_name
, hostname
, base_path
, live_timer
,
1443 snapshot
, sessiond_uuid
,
1444 id_sessiond
.is_set
? &id_sessiond
.value
: NULL
,
1445 current_chunk_id
.is_set
? ¤t_chunk_id
.value
: NULL
,
1446 creation_time
.is_set
? &creation_time
.value
: NULL
,
1447 conn
->major
, conn
->minor
,
1448 session_name_contains_creation_timestamp
);
1453 LTTNG_ASSERT(!conn
->session
);
1454 conn
->session
= session
;
1455 DBG("Created session %" PRIu64
, session
->id
);
1457 reply
.generic
.session_id
= htobe64(session
->id
);
1461 reply
.generic
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1463 reply
.generic
.ret_code
= htobe32(LTTNG_OK
);
1466 if (conn
->minor
< 11) {
1467 /* From 2.1 to 2.10 */
1468 ret
= lttng_dynamic_buffer_append(&reply_payload
,
1469 &reply
.generic
, sizeof(reply
.generic
));
1471 ERR("Failed to append \"create session\" command reply header to payload buffer");
1476 const uint32_t output_path_length
=
1477 session
? strlen(session
->output_path
) + 1 : 0;
1479 reply
.output_path_length
= htobe32(output_path_length
);
1480 ret
= lttng_dynamic_buffer_append(
1481 &reply_payload
, &reply
, sizeof(reply
));
1483 ERR("Failed to append \"create session\" command reply header to payload buffer");
1487 if (output_path_length
) {
1488 ret
= lttng_dynamic_buffer_append(&reply_payload
,
1489 session
->output_path
,
1490 output_path_length
);
1492 ERR("Failed to append \"create session\" command reply path to payload buffer");
1498 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, reply_payload
.data
,
1499 reply_payload
.size
, 0);
1500 if (send_ret
< (ssize_t
) reply_payload
.size
) {
1501 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1502 reply_payload
.size
, send_ret
);
1506 if (ret
< 0 && session
) {
1507 session_put(session
);
1509 lttng_dynamic_buffer_reset(&reply_payload
);
1514 * When we have received all the streams and the metadata for a channel,
1515 * we make them visible to the viewer threads.
1517 static void publish_connection_local_streams(struct relay_connection
*conn
)
1519 struct relay_stream
*stream
;
1520 struct relay_session
*session
= conn
->session
;
1523 * We publish all streams belonging to a session atomically wrt
1526 pthread_mutex_lock(&session
->lock
);
1528 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1530 stream_publish(stream
);
1535 * Inform the viewer that there are new streams in the session.
1537 if (session
->viewer_attached
) {
1538 uatomic_set(&session
->new_streams
, 1);
1540 pthread_mutex_unlock(&session
->lock
);
1543 static int conform_channel_path(char *channel_path
)
1547 if (strstr("../", channel_path
)) {
1548 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1554 if (*channel_path
== '/') {
1555 const size_t len
= strlen(channel_path
);
1558 * Channel paths from peers prior to 2.11 are expressed as an
1559 * absolute path that is, in reality, relative to the relay
1560 * daemon's output directory. Remove the leading slash so it
1561 * is correctly interpreted as a relative path later on.
1563 * len (and not len - 1) is used to copy the trailing NULL.
1565 bcopy(channel_path
+ 1, channel_path
, len
);
1572 * relay_add_stream: allocate a new stream for a session
1574 static int relay_add_stream(
1575 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
1576 struct relay_connection
*conn
,
1577 const struct lttng_buffer_view
*payload
)
1581 struct relay_session
*session
= conn
->session
;
1582 struct relay_stream
*stream
= NULL
;
1583 struct lttcomm_relayd_status_stream reply
;
1584 struct ctf_trace
*trace
= NULL
;
1585 uint64_t stream_handle
= -1ULL;
1586 char *path_name
= NULL
, *channel_name
= NULL
;
1587 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1588 LTTNG_OPTIONAL(uint64_t) stream_chunk_id
= {};
1590 if (!session
|| !conn
->version_check_done
) {
1591 ERR("Trying to add a stream before version check");
1593 goto end_no_session
;
1596 if (session
->minor
== 1) {
1598 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1600 } else if (session
->minor
> 1 && session
->minor
< 11) {
1601 /* From 2.2 to 2.10 */
1602 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1603 &channel_name
, &tracefile_size
, &tracefile_count
);
1605 /* From 2.11 to ... */
1606 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1607 &channel_name
, &tracefile_size
, &tracefile_count
,
1608 &stream_chunk_id
.value
);
1609 stream_chunk_id
.is_set
= true;
1616 if (conform_channel_path(path_name
)) {
1621 * Backward compatibility for --group-output-by-session.
1622 * Prior to lttng 2.11, the complete path is passed by the stream.
1623 * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1624 * >=2.11 the chunk is responsible for the output path. When dealing
1625 * with producer < 2.11 the chunk output_path is the root output path
1626 * and the stream carries the complete path (path_name).
1627 * To support --group-output-by-session with older producer (<2.11), we
1628 * need to craft the path based on the stream path.
1630 if (opt_group_output_by
== RELAYD_GROUP_OUTPUT_BY_SESSION
) {
1631 if (conn
->minor
< 4) {
1633 * From 2.1 to 2.3, the session_name is not passed on
1634 * the RELAYD_CREATE_SESSION command. The session name
1635 * is necessary to detect the presence of a base_path
1636 * inside the stream path. Without it we cannot perform
1637 * a valid group-output-by-session transformation.
1639 WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1640 " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1641 session
->id
, path_name
);
1642 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1643 char *group_by_session_path_name
;
1645 LTTNG_ASSERT(session
->session_name
[0] != '\0');
1647 group_by_session_path_name
=
1648 backward_compat_group_by_session(
1650 session
->session_name
,
1651 session
->creation_time
.value
);
1652 if (!group_by_session_path_name
) {
1653 ERR("Failed to apply group by session to stream of session %" PRIu64
,
1658 DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1659 path_name
, group_by_session_path_name
);
1662 path_name
= group_by_session_path_name
;
1666 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1671 /* This stream here has one reference on the trace. */
1672 pthread_mutex_lock(&last_relay_stream_id_lock
);
1673 stream_handle
= ++last_relay_stream_id
;
1674 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1676 /* We pass ownership of path_name and channel_name. */
1677 stream
= stream_create(trace
, stream_handle
, path_name
,
1678 channel_name
, tracefile_size
, tracefile_count
);
1680 channel_name
= NULL
;
1683 * Streams are the owners of their trace. Reference to trace is
1684 * kept within stream_create().
1686 ctf_trace_put(trace
);
1689 memset(&reply
, 0, sizeof(reply
));
1690 reply
.handle
= htobe64(stream_handle
);
1692 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1694 reply
.ret_code
= htobe32(LTTNG_OK
);
1697 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1698 sizeof(struct lttcomm_relayd_status_stream
), 0);
1699 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1700 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1712 * relay_close_stream: close a specific stream
1714 static int relay_close_stream(
1715 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
1716 struct relay_connection
*conn
,
1717 const struct lttng_buffer_view
*payload
)
1721 struct relay_session
*session
= conn
->session
;
1722 struct lttcomm_relayd_close_stream stream_info
;
1723 struct lttcomm_relayd_generic_reply reply
;
1724 struct relay_stream
*stream
;
1726 DBG("Close stream received");
1728 if (!session
|| !conn
->version_check_done
) {
1729 ERR("Trying to close a stream before version check");
1731 goto end_no_session
;
1734 if (payload
->size
< sizeof(stream_info
)) {
1735 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1736 sizeof(stream_info
), payload
->size
);
1738 goto end_no_session
;
1740 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1741 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1742 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1744 stream
= stream_get_by_id(stream_info
.stream_id
);
1751 * Set last_net_seq_num before the close flag. Required by data
1754 pthread_mutex_lock(&stream
->lock
);
1755 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1756 pthread_mutex_unlock(&stream
->lock
);
1759 * This is one of the conditions which may trigger a stream close
1760 * with the others being:
1761 * 1) A close command is received for a stream
1762 * 2) The control connection owning the stream is closed
1763 * 3) We have received all of the stream's data _after_ a close
1766 try_stream_close(stream
);
1771 memset(&reply
, 0, sizeof(reply
));
1773 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1775 reply
.ret_code
= htobe32(LTTNG_OK
);
1777 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1778 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1779 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1780 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1790 * relay_reset_metadata: reset a metadata stream
1793 int relay_reset_metadata(
1794 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
1795 struct relay_connection
*conn
,
1796 const struct lttng_buffer_view
*payload
)
1800 struct relay_session
*session
= conn
->session
;
1801 struct lttcomm_relayd_reset_metadata stream_info
;
1802 struct lttcomm_relayd_generic_reply reply
;
1803 struct relay_stream
*stream
;
1805 DBG("Reset metadata received");
1807 if (!session
|| !conn
->version_check_done
) {
1808 ERR("Trying to reset a metadata stream before version check");
1810 goto end_no_session
;
1813 if (payload
->size
< sizeof(stream_info
)) {
1814 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1815 sizeof(stream_info
), payload
->size
);
1817 goto end_no_session
;
1819 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1820 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1821 stream_info
.version
= be64toh(stream_info
.version
);
1823 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1825 /* Unsupported for live sessions for now. */
1826 if (session
->live_timer
!= 0) {
1831 stream
= stream_get_by_id(stream_info
.stream_id
);
1836 pthread_mutex_lock(&stream
->lock
);
1837 if (!stream
->is_metadata
) {
1842 ret
= stream_reset_file(stream
);
1844 ERR("Failed to reset metadata stream %" PRIu64
1845 ": stream_path = %s, channel = %s",
1846 stream
->stream_handle
, stream
->path_name
,
1847 stream
->channel_name
);
1851 pthread_mutex_unlock(&stream
->lock
);
1855 memset(&reply
, 0, sizeof(reply
));
1857 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1859 reply
.ret_code
= htobe32(LTTNG_OK
);
1861 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1862 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1863 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1864 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1874 * relay_unknown_command: send -1 if received unknown command
1876 static void relay_unknown_command(struct relay_connection
*conn
)
1878 struct lttcomm_relayd_generic_reply reply
;
1881 memset(&reply
, 0, sizeof(reply
));
1882 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1883 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1884 if (send_ret
< sizeof(reply
)) {
1885 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1890 * relay_start: send an acknowledgment to the client to tell if we are
1891 * ready to receive data. We are ready if a session is established.
1893 static int relay_start(
1894 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
1895 struct relay_connection
*conn
,
1896 const struct lttng_buffer_view
*payload
__attribute__((unused
)))
1900 struct lttcomm_relayd_generic_reply reply
;
1901 struct relay_session
*session
= conn
->session
;
1904 DBG("Trying to start the streaming without a session established");
1905 ret
= htobe32(LTTNG_ERR_UNK
);
1908 memset(&reply
, 0, sizeof(reply
));
1909 reply
.ret_code
= htobe32(LTTNG_OK
);
1910 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1912 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1913 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1922 * relay_recv_metadata: receive the metadata for the session.
1924 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1925 struct relay_connection
*conn
,
1926 const struct lttng_buffer_view
*payload
)
1929 struct relay_session
*session
= conn
->session
;
1930 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1931 struct relay_stream
*metadata_stream
;
1932 uint64_t metadata_payload_size
;
1933 struct lttng_buffer_view packet_view
;
1936 ERR("Metadata sent before version check");
1941 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1942 ERR("Incorrect data size");
1946 metadata_payload_size
= recv_hdr
->data_size
-
1947 sizeof(struct lttcomm_relayd_metadata_payload
);
1949 memcpy(&metadata_payload_header
, payload
->data
,
1950 sizeof(metadata_payload_header
));
1951 metadata_payload_header
.stream_id
= be64toh(
1952 metadata_payload_header
.stream_id
);
1953 metadata_payload_header
.padding_size
= be32toh(
1954 metadata_payload_header
.padding_size
);
1956 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1957 if (!metadata_stream
) {
1962 packet_view
= lttng_buffer_view_from_view(payload
,
1963 sizeof(metadata_payload_header
), metadata_payload_size
);
1964 if (!lttng_buffer_view_is_valid(&packet_view
)) {
1965 ERR("Invalid metadata packet length announced by header");
1970 pthread_mutex_lock(&metadata_stream
->lock
);
1971 ret
= stream_write(metadata_stream
, &packet_view
,
1972 metadata_payload_header
.padding_size
);
1973 pthread_mutex_unlock(&metadata_stream
->lock
);
1979 stream_put(metadata_stream
);
1985 * relay_send_version: send relayd version number
1987 static int relay_send_version(
1988 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
1989 struct relay_connection
*conn
,
1990 const struct lttng_buffer_view
*payload
)
1994 struct lttcomm_relayd_version reply
, msg
;
1995 bool compatible
= true;
1997 conn
->version_check_done
= true;
1999 /* Get version from the other side. */
2000 if (payload
->size
< sizeof(msg
)) {
2001 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
2002 sizeof(msg
), payload
->size
);
2007 memcpy(&msg
, payload
->data
, sizeof(msg
));
2008 msg
.major
= be32toh(msg
.major
);
2009 msg
.minor
= be32toh(msg
.minor
);
2011 memset(&reply
, 0, sizeof(reply
));
2012 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
2013 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
2015 /* Major versions must be the same */
2016 if (reply
.major
!= msg
.major
) {
2017 DBG("Incompatible major versions (%u vs %u), deleting session",
2018 reply
.major
, msg
.major
);
2022 conn
->major
= reply
.major
;
2023 /* We adapt to the lowest compatible version */
2024 if (reply
.minor
<= msg
.minor
) {
2025 conn
->minor
= reply
.minor
;
2027 conn
->minor
= msg
.minor
;
2030 reply
.major
= htobe32(reply
.major
);
2031 reply
.minor
= htobe32(reply
.minor
);
2032 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2034 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2035 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2048 DBG("Version check done using protocol %u.%u", conn
->major
,
2056 * Check for data pending for a given stream id from the session daemon.
2058 static int relay_data_pending(
2059 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
2060 struct relay_connection
*conn
,
2061 const struct lttng_buffer_view
*payload
)
2063 struct relay_session
*session
= conn
->session
;
2064 struct lttcomm_relayd_data_pending msg
;
2065 struct lttcomm_relayd_generic_reply reply
;
2066 struct relay_stream
*stream
;
2069 uint64_t stream_seq
;
2071 DBG("Data pending command received");
2073 if (!session
|| !conn
->version_check_done
) {
2074 ERR("Trying to check for data before version check");
2076 goto end_no_session
;
2079 if (payload
->size
< sizeof(msg
)) {
2080 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2081 sizeof(msg
), payload
->size
);
2083 goto end_no_session
;
2085 memcpy(&msg
, payload
->data
, sizeof(msg
));
2086 msg
.stream_id
= be64toh(msg
.stream_id
);
2087 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
2089 stream
= stream_get_by_id(msg
.stream_id
);
2090 if (stream
== NULL
) {
2095 pthread_mutex_lock(&stream
->lock
);
2097 if (session_streams_have_index(session
)) {
2099 * Ensure that both the index and stream data have been
2100 * flushed up to the requested point.
2102 stream_seq
= std::min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2104 stream_seq
= stream
->prev_data_seq
;
2106 DBG("Data pending for stream id %" PRIu64
": prev_data_seq %" PRIu64
2107 ", prev_index_seq %" PRIu64
2108 ", and last_seq %" PRIu64
, msg
.stream_id
,
2109 stream
->prev_data_seq
, stream
->prev_index_seq
,
2110 msg
.last_net_seq_num
);
2112 /* Avoid wrapping issue */
2113 if (((int64_t) (stream_seq
- msg
.last_net_seq_num
)) >= 0) {
2114 /* Data has in fact been written and is NOT pending */
2117 /* Data still being streamed thus pending */
2121 stream
->data_pending_check_done
= true;
2122 pthread_mutex_unlock(&stream
->lock
);
2127 memset(&reply
, 0, sizeof(reply
));
2128 reply
.ret_code
= htobe32(ret
);
2129 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2130 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2131 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2141 * Wait for the control socket to reach a quiescent state.
2143 * Note that for now, when receiving this command from the session
2144 * daemon, this means that every subsequent commands or data received on
2145 * the control socket has been handled. So, this is why we simply return
2148 static int relay_quiescent_control(
2149 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
2150 struct relay_connection
*conn
,
2151 const struct lttng_buffer_view
*payload
)
2155 struct relay_stream
*stream
;
2156 struct lttcomm_relayd_quiescent_control msg
;
2157 struct lttcomm_relayd_generic_reply reply
;
2159 DBG("Checking quiescent state on control socket");
2161 if (!conn
->session
|| !conn
->version_check_done
) {
2162 ERR("Trying to check for data before version check");
2164 goto end_no_session
;
2167 if (payload
->size
< sizeof(msg
)) {
2168 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2169 sizeof(msg
), payload
->size
);
2171 goto end_no_session
;
2173 memcpy(&msg
, payload
->data
, sizeof(msg
));
2174 msg
.stream_id
= be64toh(msg
.stream_id
);
2176 stream
= stream_get_by_id(msg
.stream_id
);
2180 pthread_mutex_lock(&stream
->lock
);
2181 stream
->data_pending_check_done
= true;
2182 pthread_mutex_unlock(&stream
->lock
);
2184 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
2187 memset(&reply
, 0, sizeof(reply
));
2188 reply
.ret_code
= htobe32(LTTNG_OK
);
2189 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2190 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2191 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2203 * Initialize a data pending command. This means that a consumer is about
2204 * to ask for data pending for each stream it holds. Simply iterate over
2205 * all streams of a session and set the data_pending_check_done flag.
2207 * This command returns to the client a LTTNG_OK code.
2209 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2210 struct relay_connection
*conn
,
2211 const struct lttng_buffer_view
*payload
)
2215 struct lttng_ht_iter iter
;
2216 struct lttcomm_relayd_begin_data_pending msg
;
2217 struct lttcomm_relayd_generic_reply reply
;
2218 struct relay_stream
*stream
;
2220 LTTNG_ASSERT(recv_hdr
);
2223 DBG("Init streams for data pending");
2225 if (!conn
->session
|| !conn
->version_check_done
) {
2226 ERR("Trying to check for data before version check");
2228 goto end_no_session
;
2231 if (payload
->size
< sizeof(msg
)) {
2232 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2233 sizeof(msg
), payload
->size
);
2235 goto end_no_session
;
2237 memcpy(&msg
, payload
->data
, sizeof(msg
));
2238 msg
.session_id
= be64toh(msg
.session_id
);
2241 * Iterate over all streams to set the begin data pending flag.
2242 * For now, the streams are indexed by stream handle so we have
2243 * to iterate over all streams to find the one associated with
2244 * the right session_id.
2247 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2249 if (!stream_get(stream
)) {
2252 if (stream
->trace
->session
->id
== msg
.session_id
) {
2253 pthread_mutex_lock(&stream
->lock
);
2254 stream
->data_pending_check_done
= false;
2255 pthread_mutex_unlock(&stream
->lock
);
2256 DBG("Set begin data pending flag to stream %" PRIu64
,
2257 stream
->stream_handle
);
2263 memset(&reply
, 0, sizeof(reply
));
2264 /* All good, send back reply. */
2265 reply
.ret_code
= htobe32(LTTNG_OK
);
2267 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2268 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2269 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2281 * End data pending command. This will check, for a given session id, if
2282 * each stream associated with it has its data_pending_check_done flag
2283 * set. If not, this means that the client lost track of the stream but
2284 * the data is still being streamed on our side. In this case, we inform
2285 * the client that data is in flight.
2287 * Return to the client if there is data in flight or not with a ret_code.
2289 static int relay_end_data_pending(
2290 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
2291 struct relay_connection
*conn
,
2292 const struct lttng_buffer_view
*payload
)
2296 struct lttng_ht_iter iter
;
2297 struct lttcomm_relayd_end_data_pending msg
;
2298 struct lttcomm_relayd_generic_reply reply
;
2299 struct relay_stream
*stream
;
2300 uint32_t is_data_inflight
= 0;
2302 DBG("End data pending command");
2304 if (!conn
->session
|| !conn
->version_check_done
) {
2305 ERR("Trying to check for data before version check");
2307 goto end_no_session
;
2310 if (payload
->size
< sizeof(msg
)) {
2311 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2312 sizeof(msg
), payload
->size
);
2314 goto end_no_session
;
2316 memcpy(&msg
, payload
->data
, sizeof(msg
));
2317 msg
.session_id
= be64toh(msg
.session_id
);
2320 * Iterate over all streams to see if the begin data pending
2324 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2326 if (!stream_get(stream
)) {
2329 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2333 pthread_mutex_lock(&stream
->lock
);
2334 if (!stream
->data_pending_check_done
) {
2335 uint64_t stream_seq
;
2337 if (session_streams_have_index(conn
->session
)) {
2339 * Ensure that both the index and stream data have been
2340 * flushed up to the requested point.
2342 stream_seq
= std::min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2344 stream_seq
= stream
->prev_data_seq
;
2346 if (!stream
->closed
|| !(((int64_t) (stream_seq
- stream
->last_net_seq_num
)) >= 0)) {
2347 is_data_inflight
= 1;
2348 DBG("Data is still in flight for stream %" PRIu64
,
2349 stream
->stream_handle
);
2350 pthread_mutex_unlock(&stream
->lock
);
2355 pthread_mutex_unlock(&stream
->lock
);
2360 memset(&reply
, 0, sizeof(reply
));
2361 /* All good, send back reply. */
2362 reply
.ret_code
= htobe32(is_data_inflight
);
2364 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2365 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2366 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2378 * Receive an index for a specific stream.
2380 * Return 0 on success else a negative value.
2382 static int relay_recv_index(
2383 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
2384 struct relay_connection
*conn
,
2385 const struct lttng_buffer_view
*payload
)
2389 struct relay_session
*session
= conn
->session
;
2390 struct lttcomm_relayd_index index_info
;
2391 struct lttcomm_relayd_generic_reply reply
;
2392 struct relay_stream
*stream
;
2397 DBG("Relay receiving index");
2399 if (!session
|| !conn
->version_check_done
) {
2400 ERR("Trying to close a stream before version check");
2402 goto end_no_session
;
2405 msg_len
= lttcomm_relayd_index_len(
2406 lttng_to_index_major(conn
->major
, conn
->minor
),
2407 lttng_to_index_minor(conn
->major
, conn
->minor
));
2408 if (payload
->size
< msg_len
) {
2409 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2410 msg_len
, payload
->size
);
2412 goto end_no_session
;
2414 memcpy(&index_info
, payload
->data
, msg_len
);
2415 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2416 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2417 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2418 index_info
.content_size
= be64toh(index_info
.content_size
);
2419 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2420 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2421 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2422 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2424 if (conn
->minor
>= 8) {
2425 index_info
.stream_instance_id
=
2426 be64toh(index_info
.stream_instance_id
);
2427 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2429 index_info
.stream_instance_id
= -1ULL;
2430 index_info
.packet_seq_num
= -1ULL;
2433 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2435 ERR("stream_get_by_id not found");
2440 pthread_mutex_lock(&stream
->lock
);
2441 ret
= stream_add_index(stream
, &index_info
);
2442 pthread_mutex_unlock(&stream
->lock
);
2444 goto end_stream_put
;
2450 memset(&reply
, 0, sizeof(reply
));
2452 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2454 reply
.ret_code
= htobe32(LTTNG_OK
);
2456 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2457 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2458 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2467 * Receive the streams_sent message.
2469 * Return 0 on success else a negative value.
2471 static int relay_streams_sent(
2472 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
2473 struct relay_connection
*conn
,
2474 const struct lttng_buffer_view
*payload
__attribute__((unused
)))
2478 struct lttcomm_relayd_generic_reply reply
;
2482 DBG("Relay receiving streams_sent");
2484 if (!conn
->session
|| !conn
->version_check_done
) {
2485 ERR("Trying to close a stream before version check");
2487 goto end_no_session
;
2491 * Publish every pending stream in the connection recv list which are
2492 * now ready to be used by the viewer.
2494 publish_connection_local_streams(conn
);
2496 memset(&reply
, 0, sizeof(reply
));
2497 reply
.ret_code
= htobe32(LTTNG_OK
);
2498 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2499 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2500 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2512 static ssize_t
relay_unpack_rotate_streams_header(
2513 const struct lttng_buffer_view
*payload
,
2514 struct lttcomm_relayd_rotate_streams
*_rotate_streams
)
2516 struct lttcomm_relayd_rotate_streams rotate_streams
;
2518 * Set to the smallest version (packed) of `lttcomm_relayd_rotate_streams`.
2519 * This is the smallest version of this structure, but it can be larger;
2520 * this variable is updated once the proper size of the structure is known.
2522 * See comment at the declaration of this structure for more information.
2524 ssize_t header_len
= sizeof(struct lttcomm_relayd_rotate_streams_packed
);
2525 size_t expected_payload_size_no_padding
,
2526 expected_payload_size_3_bytes_padding
,
2527 expected_payload_size_7_bytes_padding
;
2529 if (payload
->size
< header_len
) {
2530 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2531 header_len
, payload
->size
);
2536 * Some versions incorrectly omitted the LTTNG_PACKED annotation on the
2537 * `new_chunk_id` optional field of struct lttcomm_relayd_rotate_streams.
2539 * We start by "unpacking" `stream_count` to figure out the padding length
2540 * emited by our peer.
2542 memcpy(&rotate_streams
.stream_count
, payload
->data
,
2543 sizeof(rotate_streams
.stream_count
));
2544 rotate_streams
= (typeof(rotate_streams
)) {
2545 .stream_count
= be32toh(rotate_streams
.stream_count
),
2546 .new_chunk_id
= LTTNG_OPTIONAL_INIT_UNSET
,
2550 * Payload size expected given the possible padding lengths in
2551 * `struct lttcomm_relayd_rotate_streams`.
2553 expected_payload_size_no_padding
= (rotate_streams
.stream_count
*
2554 sizeof(*rotate_streams
.rotation_positions
)) +
2555 sizeof(lttcomm_relayd_rotate_streams_packed
);
2556 expected_payload_size_3_bytes_padding
= (rotate_streams
.stream_count
*
2557 sizeof(*rotate_streams
.rotation_positions
)) +
2558 sizeof(lttcomm_relayd_rotate_streams_3_bytes_padding
);
2559 expected_payload_size_7_bytes_padding
= (rotate_streams
.stream_count
*
2560 sizeof(*rotate_streams
.rotation_positions
)) +
2561 sizeof(lttcomm_relayd_rotate_streams_7_bytes_padding
);
2563 if (payload
->size
== expected_payload_size_no_padding
) {
2564 struct lttcomm_relayd_rotate_streams_packed packed_rotate_streams
;
2567 * This handles cases where someone might build with
2568 * -fpack-struct or any other toolchain that wouldn't produce
2569 * padding to align `value`.
2571 DBG("Received `struct lttcomm_relayd_rotate_streams` with no padding");
2573 header_len
= sizeof(packed_rotate_streams
);
2574 memcpy(&packed_rotate_streams
, payload
->data
, header_len
);
2576 /* Unpack the packed structure to the natively-packed version. */
2577 *_rotate_streams
= (typeof(*_rotate_streams
)) {
2578 .stream_count
= be32toh(packed_rotate_streams
.stream_count
),
2579 .new_chunk_id
= (typeof(_rotate_streams
->new_chunk_id
)) {
2580 .is_set
= !!packed_rotate_streams
.new_chunk_id
.is_set
,
2581 .value
= be64toh(packed_rotate_streams
.new_chunk_id
.value
),
2584 } else if (payload
->size
== expected_payload_size_3_bytes_padding
) {
2585 struct lttcomm_relayd_rotate_streams_3_bytes_padding padded_rotate_streams
;
2587 DBG("Received `struct lttcomm_relayd_rotate_streams` with 3 bytes of padding (4-byte aligned peer)");
2589 header_len
= sizeof(padded_rotate_streams
);
2590 memcpy(&padded_rotate_streams
, payload
->data
, header_len
);
2592 /* Unpack the 3-byte padded structure to the natively-packed version. */
2593 *_rotate_streams
= (typeof(*_rotate_streams
)) {
2594 .stream_count
= be32toh(padded_rotate_streams
.stream_count
),
2595 .new_chunk_id
= (typeof(_rotate_streams
->new_chunk_id
)) {
2596 .is_set
= !!padded_rotate_streams
.new_chunk_id
.is_set
,
2597 .value
= be64toh(padded_rotate_streams
.new_chunk_id
.value
),
2600 } else if (payload
->size
== expected_payload_size_7_bytes_padding
) {
2601 struct lttcomm_relayd_rotate_streams_7_bytes_padding padded_rotate_streams
;
2603 DBG("Received `struct lttcomm_relayd_rotate_streams` with 7 bytes of padding (8-byte aligned peer)");
2605 header_len
= sizeof(padded_rotate_streams
);
2606 memcpy(&padded_rotate_streams
, payload
->data
, header_len
);
2608 /* Unpack the 7-byte padded structure to the natively-packed version. */
2609 *_rotate_streams
= (typeof(*_rotate_streams
)) {
2610 .stream_count
= be32toh(padded_rotate_streams
.stream_count
),
2611 .new_chunk_id
= (typeof(_rotate_streams
->new_chunk_id
)) {
2612 .is_set
= !!padded_rotate_streams
.new_chunk_id
.is_set
,
2613 .value
= be64toh(padded_rotate_streams
.new_chunk_id
.value
),
2617 header_len
= sizeof(padded_rotate_streams
);
2619 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected %zu, %zu or %zu bytes, got %zu bytes",
2620 expected_payload_size_no_padding
,
2621 expected_payload_size_3_bytes_padding
,
2622 expected_payload_size_7_bytes_padding
,
2633 * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2634 * session rotation feature (not the tracefile rotation feature).
2636 static int relay_rotate_session_streams(
2637 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
2638 struct relay_connection
*conn
,
2639 const struct lttng_buffer_view
*payload
)
2644 enum lttng_error_code reply_code
= LTTNG_ERR_UNK
;
2645 struct relay_session
*session
= conn
->session
;
2646 struct lttcomm_relayd_rotate_streams rotate_streams
;
2647 struct lttcomm_relayd_generic_reply reply
= {};
2648 struct relay_stream
*stream
= NULL
;
2649 struct lttng_trace_chunk
*next_trace_chunk
= NULL
;
2650 struct lttng_buffer_view stream_positions
;
2651 char chunk_id_buf
[MAX_INT_DEC_LEN(uint64_t)];
2652 const char *chunk_id_str
= "none";
2655 if (!session
|| !conn
->version_check_done
) {
2656 ERR("Trying to rotate a stream before version check");
2661 if (session
->major
== 2 && session
->minor
< 11) {
2662 ERR("Unsupported feature before 2.11");
2667 header_len
= relay_unpack_rotate_streams_header(payload
, &rotate_streams
);
2668 if (header_len
< 0) {
2673 if (rotate_streams
.new_chunk_id
.is_set
) {
2675 * Retrieve the trace chunk the stream must transition to. As
2676 * per the protocol, this chunk should have been created
2677 * before this command is received.
2679 next_trace_chunk
= sessiond_trace_chunk_registry_get_chunk(
2680 sessiond_trace_chunk_registry
,
2681 session
->sessiond_uuid
,
2682 conn
->session
->id_sessiond
.is_set
?
2683 conn
->session
->id_sessiond
.value
:
2685 rotate_streams
.new_chunk_id
.value
);
2686 if (!next_trace_chunk
) {
2687 char uuid_str
[LTTNG_UUID_STR_LEN
];
2689 lttng_uuid_to_str(session
->sessiond_uuid
, uuid_str
);
2690 ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2691 ", trace_chunk_id = %" PRIu64
,
2692 uuid_str
, session
->id
,
2693 rotate_streams
.new_chunk_id
.value
);
2694 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2699 ret
= snprintf(chunk_id_buf
, sizeof(chunk_id_buf
), "%" PRIu64
,
2700 rotate_streams
.new_chunk_id
.value
);
2701 if (ret
< 0 || ret
>= sizeof(chunk_id_buf
)) {
2702 chunk_id_str
= "formatting error";
2704 chunk_id_str
= chunk_id_buf
;
2708 DBG("Rotate %" PRIu32
" streams of session \"%s\" to chunk \"%s\"",
2709 rotate_streams
.stream_count
, session
->session_name
,
2712 stream_positions
= lttng_buffer_view_from_view(payload
,
2714 if (!stream_positions
.data
||
2715 stream_positions
.size
<
2716 (rotate_streams
.stream_count
*
2717 sizeof(struct lttcomm_relayd_stream_rotation_position
))) {
2718 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2723 for (i
= 0; i
< rotate_streams
.stream_count
; i
++) {
2724 struct lttcomm_relayd_stream_rotation_position
*position_comm
=
2725 &((typeof(position_comm
)) stream_positions
.data
)[i
];
2726 const struct lttcomm_relayd_stream_rotation_position pos
= {
2727 .stream_id
= be64toh(position_comm
->stream_id
),
2728 .rotate_at_seq_num
= be64toh(
2729 position_comm
->rotate_at_seq_num
),
2732 stream
= stream_get_by_id(pos
.stream_id
);
2734 reply_code
= LTTNG_ERR_INVALID
;
2739 pthread_mutex_lock(&stream
->lock
);
2740 ret
= stream_set_pending_rotation(stream
, next_trace_chunk
,
2741 pos
.rotate_at_seq_num
);
2742 pthread_mutex_unlock(&stream
->lock
);
2744 reply_code
= LTTNG_ERR_FILE_CREATION_ERROR
;
2752 reply_code
= LTTNG_OK
;
2759 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2760 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2761 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2762 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2763 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2768 lttng_trace_chunk_put(next_trace_chunk
);
2773 * relay_create_trace_chunk: create a new trace chunk
2775 static int relay_create_trace_chunk(
2776 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
2777 struct relay_connection
*conn
,
2778 const struct lttng_buffer_view
*payload
)
2782 struct relay_session
*session
= conn
->session
;
2783 struct lttcomm_relayd_create_trace_chunk
*msg
;
2784 struct lttcomm_relayd_generic_reply reply
= {};
2785 struct lttng_buffer_view header_view
;
2786 struct lttng_trace_chunk
*chunk
= NULL
, *published_chunk
= NULL
;
2787 enum lttng_error_code reply_code
= LTTNG_OK
;
2788 enum lttng_trace_chunk_status chunk_status
;
2789 const char *new_path
;
2791 if (!session
|| !conn
->version_check_done
) {
2792 ERR("Trying to create a trace chunk before version check");
2797 if (session
->major
== 2 && session
->minor
< 11) {
2798 ERR("Chunk creation command is unsupported before 2.11");
2803 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2804 if (!lttng_buffer_view_is_valid(&header_view
)) {
2805 ERR("Failed to receive payload of chunk creation command");
2810 /* Convert to host endianness. */
2811 msg
= (typeof(msg
)) header_view
.data
;
2812 msg
->chunk_id
= be64toh(msg
->chunk_id
);
2813 msg
->creation_timestamp
= be64toh(msg
->creation_timestamp
);
2814 msg
->override_name_length
= be32toh(msg
->override_name_length
);
2816 pthread_mutex_lock(&conn
->session
->lock
);
2817 session
->ongoing_rotation
= true;
2818 if (session
->current_trace_chunk
&&
2819 !lttng_trace_chunk_get_name_overridden(session
->current_trace_chunk
)) {
2820 chunk_status
= lttng_trace_chunk_rename_path(session
->current_trace_chunk
,
2821 DEFAULT_CHUNK_TMP_OLD_DIRECTORY
);
2822 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2823 ERR("Failed to rename old chunk");
2825 reply_code
= LTTNG_ERR_UNK
;
2829 if (!session
->current_trace_chunk
) {
2830 if (!session
->has_rotated
) {
2836 new_path
= DEFAULT_CHUNK_TMP_NEW_DIRECTORY
;
2838 chunk
= lttng_trace_chunk_create(
2839 msg
->chunk_id
, msg
->creation_timestamp
, new_path
);
2841 ERR("Failed to create trace chunk in trace chunk creation command");
2843 reply_code
= LTTNG_ERR_NOMEM
;
2846 lttng_trace_chunk_set_fd_tracker(chunk
, the_fd_tracker
);
2848 if (msg
->override_name_length
) {
2850 const struct lttng_buffer_view chunk_name_view
=
2851 lttng_buffer_view_from_view(payload
,
2853 msg
->override_name_length
);
2855 if (!lttng_buffer_view_is_valid(&chunk_name_view
)) {
2856 ERR("Invalid payload of chunk creation command (protocol error): buffer too short for expected name length");
2858 reply_code
= LTTNG_ERR_INVALID
;
2862 name
= chunk_name_view
.data
;
2863 if (name
[msg
->override_name_length
- 1]) {
2864 ERR("Invalid payload of chunk creation command (protocol error): name is not null-terminated");
2866 reply_code
= LTTNG_ERR_INVALID
;
2870 chunk_status
= lttng_trace_chunk_override_name(
2871 chunk
, chunk_name_view
.data
);
2872 switch (chunk_status
) {
2873 case LTTNG_TRACE_CHUNK_STATUS_OK
:
2875 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT
:
2876 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2877 reply_code
= LTTNG_ERR_INVALID
;
2881 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2882 reply_code
= LTTNG_ERR_UNK
;
2888 chunk_status
= lttng_trace_chunk_set_credentials_current_user(chunk
);
2889 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2890 reply_code
= LTTNG_ERR_UNK
;
2895 LTTNG_ASSERT(conn
->session
->output_directory
);
2896 chunk_status
= lttng_trace_chunk_set_as_owner(chunk
,
2897 conn
->session
->output_directory
);
2898 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2899 reply_code
= LTTNG_ERR_UNK
;
2904 published_chunk
= sessiond_trace_chunk_registry_publish_chunk(
2905 sessiond_trace_chunk_registry
,
2906 conn
->session
->sessiond_uuid
,
2907 conn
->session
->id_sessiond
.is_set
?
2908 conn
->session
->id_sessiond
.value
:
2911 if (!published_chunk
) {
2912 char uuid_str
[LTTNG_UUID_STR_LEN
];
2914 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2915 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2920 reply_code
= LTTNG_ERR_NOMEM
;
2924 if (conn
->session
->pending_closure_trace_chunk
) {
2926 * Invalid; this means a second create_trace_chunk command was
2927 * received before a close_trace_chunk.
2929 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2930 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2934 conn
->session
->pending_closure_trace_chunk
=
2935 conn
->session
->current_trace_chunk
;
2936 conn
->session
->current_trace_chunk
= published_chunk
;
2937 published_chunk
= NULL
;
2938 if (!conn
->session
->pending_closure_trace_chunk
) {
2939 session
->ongoing_rotation
= false;
2942 pthread_mutex_unlock(&conn
->session
->lock
);
2943 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2944 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2946 sizeof(struct lttcomm_relayd_generic_reply
),
2948 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2949 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2954 lttng_trace_chunk_put(chunk
);
2955 lttng_trace_chunk_put(published_chunk
);
2960 * relay_close_trace_chunk: close a trace chunk
2962 static int relay_close_trace_chunk(
2963 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
2964 struct relay_connection
*conn
,
2965 const struct lttng_buffer_view
*payload
)
2967 int ret
= 0, buf_ret
;
2969 struct relay_session
*session
= conn
->session
;
2970 struct lttcomm_relayd_close_trace_chunk
*msg
;
2971 struct lttcomm_relayd_close_trace_chunk_reply reply
= {};
2972 struct lttng_buffer_view header_view
;
2973 struct lttng_trace_chunk
*chunk
= NULL
;
2974 enum lttng_error_code reply_code
= LTTNG_OK
;
2975 enum lttng_trace_chunk_status chunk_status
;
2977 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type
) close_command
= {};
2978 time_t close_timestamp
;
2979 char closed_trace_chunk_path
[LTTNG_PATH_MAX
];
2980 size_t path_length
= 0;
2981 const char *chunk_name
= NULL
;
2982 struct lttng_dynamic_buffer reply_payload
;
2983 const char *new_path
;
2985 lttng_dynamic_buffer_init(&reply_payload
);
2987 if (!session
|| !conn
->version_check_done
) {
2988 ERR("Trying to close a trace chunk before version check");
2993 if (session
->major
== 2 && session
->minor
< 11) {
2994 ERR("Chunk close command is unsupported before 2.11");
2999 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
3000 if (!lttng_buffer_view_is_valid(&header_view
)) {
3001 ERR("Failed to receive payload of chunk close command");
3006 /* Convert to host endianness. */
3007 msg
= (typeof(msg
)) header_view
.data
;
3008 chunk_id
= be64toh(msg
->chunk_id
);
3009 close_timestamp
= (time_t) be64toh(msg
->close_timestamp
);
3010 close_command
.value
= (lttng_trace_chunk_command_type
) be32toh(msg
->close_command
.value
);
3011 close_command
.is_set
= msg
->close_command
.is_set
;
3013 chunk
= sessiond_trace_chunk_registry_get_chunk(
3014 sessiond_trace_chunk_registry
,
3015 conn
->session
->sessiond_uuid
,
3016 conn
->session
->id_sessiond
.is_set
?
3017 conn
->session
->id_sessiond
.value
:
3021 char uuid_str
[LTTNG_UUID_STR_LEN
];
3023 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
3024 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
3029 reply_code
= LTTNG_ERR_NOMEM
;
3033 pthread_mutex_lock(&session
->lock
);
3034 if (close_command
.is_set
&&
3035 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE
) {
3037 * Clear command. It is a protocol error to ask for a
3038 * clear on a relay which does not allow it. Querying
3039 * the configuration allows figuring out whether
3040 * clearing is allowed before doing the clear.
3042 if (!opt_allow_clear
) {
3044 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
3045 goto end_unlock_session
;
3048 if (session
->pending_closure_trace_chunk
&&
3049 session
->pending_closure_trace_chunk
!= chunk
) {
3050 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
3051 session
->session_name
);
3052 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
3054 goto end_unlock_session
;
3057 if (session
->current_trace_chunk
&& session
->current_trace_chunk
!= chunk
&&
3058 !lttng_trace_chunk_get_name_overridden(session
->current_trace_chunk
)) {
3059 if (close_command
.is_set
&&
3060 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE
&&
3061 !session
->has_rotated
) {
3062 /* New chunk stays in session output directory. */
3065 /* Use chunk name for new chunk. */
3068 /* Rename new chunk path. */
3069 chunk_status
= lttng_trace_chunk_rename_path(session
->current_trace_chunk
,
3071 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
3073 goto end_unlock_session
;
3075 session
->ongoing_rotation
= false;
3077 if ((!close_command
.is_set
||
3078 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION
) &&
3079 !lttng_trace_chunk_get_name_overridden(chunk
)) {
3080 const char *old_path
;
3082 if (!session
->has_rotated
) {
3087 /* We need to move back the .tmp_old_chunk to its rightful place. */
3088 chunk_status
= lttng_trace_chunk_rename_path(chunk
, old_path
);
3089 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
3091 goto end_unlock_session
;
3094 chunk_status
= lttng_trace_chunk_set_close_timestamp(
3095 chunk
, close_timestamp
);
3096 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
3097 ERR("Failed to set trace chunk close timestamp");
3099 reply_code
= LTTNG_ERR_UNK
;
3100 goto end_unlock_session
;
3103 if (close_command
.is_set
) {
3104 chunk_status
= lttng_trace_chunk_set_close_command(
3105 chunk
, close_command
.value
);
3106 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
3108 reply_code
= LTTNG_ERR_INVALID
;
3109 goto end_unlock_session
;
3112 chunk_status
= lttng_trace_chunk_get_name(chunk
, &chunk_name
, NULL
);
3113 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
3114 ERR("Failed to get chunk name");
3116 reply_code
= LTTNG_ERR_UNK
;
3117 goto end_unlock_session
;
3119 if (!session
->has_rotated
&& !session
->snapshot
) {
3120 ret
= lttng_strncpy(closed_trace_chunk_path
,
3121 session
->output_path
,
3122 sizeof(closed_trace_chunk_path
));
3124 ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
3125 strlen(session
->output_path
),
3126 sizeof(closed_trace_chunk_path
));
3127 reply_code
= LTTNG_ERR_NOMEM
;
3129 goto end_unlock_session
;
3132 if (session
->snapshot
) {
3133 ret
= snprintf(closed_trace_chunk_path
,
3134 sizeof(closed_trace_chunk_path
),
3135 "%s/%s", session
->output_path
,
3138 ret
= snprintf(closed_trace_chunk_path
,
3139 sizeof(closed_trace_chunk_path
),
3140 "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY
3142 session
->output_path
, chunk_name
);
3144 if (ret
< 0 || ret
== sizeof(closed_trace_chunk_path
)) {
3145 ERR("Failed to format closed trace chunk resulting path");
3146 reply_code
= ret
< 0 ? LTTNG_ERR_UNK
: LTTNG_ERR_NOMEM
;
3148 goto end_unlock_session
;
3151 if (close_command
.is_set
&&
3152 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED
) {
3153 session
->has_rotated
= true;
3155 DBG("Reply chunk path on close: %s", closed_trace_chunk_path
);
3156 path_length
= strlen(closed_trace_chunk_path
) + 1;
3157 if (path_length
> UINT32_MAX
) {
3158 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3160 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
3161 goto end_unlock_session
;
3164 if (session
->current_trace_chunk
== chunk
) {
3166 * After a trace chunk close command, no new streams
3167 * referencing the chunk may be created. Hence, on the
3168 * event that no new trace chunk have been created for
3169 * the session, the reference to the current trace chunk
3170 * is released in order to allow it to be reclaimed when
3171 * the last stream releases its reference to it.
3173 lttng_trace_chunk_put(session
->current_trace_chunk
);
3174 session
->current_trace_chunk
= NULL
;
3176 lttng_trace_chunk_put(session
->pending_closure_trace_chunk
);
3177 session
->pending_closure_trace_chunk
= NULL
;
3179 pthread_mutex_unlock(&session
->lock
);
3182 reply
.generic
.ret_code
= htobe32((uint32_t) reply_code
);
3183 reply
.path_length
= htobe32((uint32_t) path_length
);
3184 buf_ret
= lttng_dynamic_buffer_append(
3185 &reply_payload
, &reply
, sizeof(reply
));
3187 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3191 if (reply_code
== LTTNG_OK
) {
3192 buf_ret
= lttng_dynamic_buffer_append(&reply_payload
,
3193 closed_trace_chunk_path
, path_length
);
3195 ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3200 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
3204 if (send_ret
< reply_payload
.size
) {
3205 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3206 reply_payload
.size
, send_ret
);
3211 lttng_trace_chunk_put(chunk
);
3212 lttng_dynamic_buffer_reset(&reply_payload
);
3217 * relay_trace_chunk_exists: check if a trace chunk exists
3219 static int relay_trace_chunk_exists(
3220 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
3221 struct relay_connection
*conn
,
3222 const struct lttng_buffer_view
*payload
)
3226 struct relay_session
*session
= conn
->session
;
3227 struct lttcomm_relayd_trace_chunk_exists
*msg
;
3228 struct lttcomm_relayd_trace_chunk_exists_reply reply
= {};
3229 struct lttng_buffer_view header_view
;
3233 if (!session
|| !conn
->version_check_done
) {
3234 ERR("Trying to check for the presence of a trace chunk before version check");
3239 if (session
->major
== 2 && session
->minor
< 11) {
3240 ERR("Chunk exists command is unsupported before 2.11");
3245 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
3246 if (!lttng_buffer_view_is_valid(&header_view
)) {
3247 ERR("Failed to receive payload of chunk exists command");
3252 /* Convert to host endianness. */
3253 msg
= (typeof(msg
)) header_view
.data
;
3254 chunk_id
= be64toh(msg
->chunk_id
);
3256 ret
= sessiond_trace_chunk_registry_chunk_exists(
3257 sessiond_trace_chunk_registry
,
3258 conn
->session
->sessiond_uuid
,
3260 chunk_id
, &chunk_exists
);
3262 * If ret is not 0, send the reply and report the error to the caller.
3263 * It is a protocol (or internal) error and the session/connection
3264 * should be torn down.
3266 reply
.generic
.ret_code
= htobe32((uint32_t) (ret
== 0 ? LTTNG_OK
: LTTNG_ERR_INVALID_PROTOCOL
));
3267 reply
.trace_chunk_exists
= ret
== 0 ? chunk_exists
: 0;
3269 send_ret
= conn
->sock
->ops
->sendmsg(
3270 conn
->sock
, &reply
, sizeof(reply
), 0);
3271 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3272 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
3281 * relay_get_configuration: query whether feature is available
3283 static int relay_get_configuration(
3284 const struct lttcomm_relayd_hdr
*recv_hdr
__attribute__((unused
)),
3285 struct relay_connection
*conn
,
3286 const struct lttng_buffer_view
*payload
)
3290 struct lttcomm_relayd_get_configuration
*msg
;
3291 struct lttcomm_relayd_get_configuration_reply reply
= {};
3292 struct lttng_buffer_view header_view
;
3293 uint64_t query_flags
= 0;
3294 uint64_t result_flags
= 0;
3296 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
3297 if (!lttng_buffer_view_is_valid(&header_view
)) {
3298 ERR("Failed to receive payload of chunk close command");
3303 /* Convert to host endianness. */
3304 msg
= (typeof(msg
)) header_view
.data
;
3305 query_flags
= be64toh(msg
->query_flags
);
3308 ret
= LTTNG_ERR_INVALID_PROTOCOL
;
3311 if (opt_allow_clear
) {
3312 result_flags
|= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED
;
3316 reply
.generic
.ret_code
= htobe32((uint32_t) (ret
== 0 ? LTTNG_OK
: LTTNG_ERR_INVALID_PROTOCOL
));
3317 reply
.relayd_configuration_flags
= htobe64(result_flags
);
3319 send_ret
= conn
->sock
->ops
->sendmsg(
3320 conn
->sock
, &reply
, sizeof(reply
), 0);
3321 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3322 ERR("Failed to send \"get configuration\" command reply (ret = %zd)",
3330 static int relay_process_control_command(struct relay_connection
*conn
,
3331 const struct lttcomm_relayd_hdr
*header
,
3332 const struct lttng_buffer_view
*payload
)
3336 DBG3("Processing \"%s\" command for socket %i",
3337 lttcomm_relayd_command_str((lttcomm_relayd_command
) header
->cmd
),
3339 switch (header
->cmd
) {
3340 case RELAYD_CREATE_SESSION
:
3341 ret
= relay_create_session(header
, conn
, payload
);
3343 case RELAYD_ADD_STREAM
:
3344 ret
= relay_add_stream(header
, conn
, payload
);
3346 case RELAYD_START_DATA
:
3347 ret
= relay_start(header
, conn
, payload
);
3349 case RELAYD_SEND_METADATA
:
3350 ret
= relay_recv_metadata(header
, conn
, payload
);
3352 case RELAYD_VERSION
:
3353 ret
= relay_send_version(header
, conn
, payload
);
3355 case RELAYD_CLOSE_STREAM
:
3356 ret
= relay_close_stream(header
, conn
, payload
);
3358 case RELAYD_DATA_PENDING
:
3359 ret
= relay_data_pending(header
, conn
, payload
);
3361 case RELAYD_QUIESCENT_CONTROL
:
3362 ret
= relay_quiescent_control(header
, conn
, payload
);
3364 case RELAYD_BEGIN_DATA_PENDING
:
3365 ret
= relay_begin_data_pending(header
, conn
, payload
);
3367 case RELAYD_END_DATA_PENDING
:
3368 ret
= relay_end_data_pending(header
, conn
, payload
);
3370 case RELAYD_SEND_INDEX
:
3371 ret
= relay_recv_index(header
, conn
, payload
);
3373 case RELAYD_STREAMS_SENT
:
3374 ret
= relay_streams_sent(header
, conn
, payload
);
3376 case RELAYD_RESET_METADATA
:
3377 ret
= relay_reset_metadata(header
, conn
, payload
);
3379 case RELAYD_ROTATE_STREAMS
:
3380 ret
= relay_rotate_session_streams(header
, conn
, payload
);
3382 case RELAYD_CREATE_TRACE_CHUNK
:
3383 ret
= relay_create_trace_chunk(header
, conn
, payload
);
3385 case RELAYD_CLOSE_TRACE_CHUNK
:
3386 ret
= relay_close_trace_chunk(header
, conn
, payload
);
3388 case RELAYD_TRACE_CHUNK_EXISTS
:
3389 ret
= relay_trace_chunk_exists(header
, conn
, payload
);
3391 case RELAYD_GET_CONFIGURATION
:
3392 ret
= relay_get_configuration(header
, conn
, payload
);
3394 case RELAYD_UPDATE_SYNC_INFO
:
3396 ERR("Received unknown command (%u)", header
->cmd
);
3397 relay_unknown_command(conn
);
3406 static enum relay_connection_status
relay_process_control_receive_payload(
3407 struct relay_connection
*conn
)
3410 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3411 struct lttng_dynamic_buffer
*reception_buffer
=
3412 &conn
->protocol
.ctrl
.reception_buffer
;
3413 struct ctrl_connection_state_receive_payload
*state
=
3414 &conn
->protocol
.ctrl
.state
.receive_payload
;
3415 struct lttng_buffer_view payload_view
;
3417 if (state
->left_to_receive
== 0) {
3418 /* Short-circuit for payload-less commands. */
3419 goto reception_complete
;
3422 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3423 reception_buffer
->data
+ state
->received
,
3424 state
->left_to_receive
, MSG_DONTWAIT
);
3427 DIAGNOSTIC_IGNORE_LOGICAL_OP
3428 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3430 PERROR("Unable to receive command payload on sock %d",
3432 status
= RELAY_CONNECTION_STATUS_ERROR
;
3435 } else if (ret
== 0) {
3436 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3437 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3441 LTTNG_ASSERT(ret
> 0);
3442 LTTNG_ASSERT(ret
<= state
->left_to_receive
);
3444 state
->left_to_receive
-= ret
;
3445 state
->received
+= ret
;
3447 if (state
->left_to_receive
> 0) {
3449 * Can't transition to the protocol's next state, wait to
3450 * receive the rest of the header.
3452 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3453 state
->received
, state
->left_to_receive
,
3459 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
3460 conn
->sock
->fd
, state
->received
);
3462 * The payload required to process the command has been received.
3463 * A view to the reception buffer is forwarded to the various
3464 * commands and the state of the control is reset on success.
3466 * Commands are responsible for sending their reply to the peer.
3468 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
3470 ret
= relay_process_control_command(conn
,
3471 &state
->header
, &payload_view
);
3473 status
= RELAY_CONNECTION_STATUS_ERROR
;
3477 ret
= connection_reset_protocol_state(conn
);
3479 status
= RELAY_CONNECTION_STATUS_ERROR
;
3485 static enum relay_connection_status
relay_process_control_receive_header(
3486 struct relay_connection
*conn
)
3489 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3490 struct lttcomm_relayd_hdr header
;
3491 struct lttng_dynamic_buffer
*reception_buffer
=
3492 &conn
->protocol
.ctrl
.reception_buffer
;
3493 struct ctrl_connection_state_receive_header
*state
=
3494 &conn
->protocol
.ctrl
.state
.receive_header
;
3496 LTTNG_ASSERT(state
->left_to_receive
!= 0);
3498 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3499 reception_buffer
->data
+ state
->received
,
3500 state
->left_to_receive
, MSG_DONTWAIT
);
3503 DIAGNOSTIC_IGNORE_LOGICAL_OP
3504 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3506 PERROR("Unable to receive control command header on sock %d",
3508 status
= RELAY_CONNECTION_STATUS_ERROR
;
3511 } else if (ret
== 0) {
3512 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3513 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3517 LTTNG_ASSERT(ret
> 0);
3518 LTTNG_ASSERT(ret
<= state
->left_to_receive
);
3520 state
->left_to_receive
-= ret
;
3521 state
->received
+= ret
;
3523 if (state
->left_to_receive
> 0) {
3525 * Can't transition to the protocol's next state, wait to
3526 * receive the rest of the header.
3528 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3529 state
->received
, state
->left_to_receive
,
3534 /* Transition to next state: receiving the command's payload. */
3535 conn
->protocol
.ctrl
.state_id
=
3536 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3537 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
3538 header
.circuit_id
= be64toh(header
.circuit_id
);
3539 header
.data_size
= be64toh(header
.data_size
);
3540 header
.cmd
= be32toh(header
.cmd
);
3541 header
.cmd_version
= be32toh(header
.cmd_version
);
3542 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
3543 &header
, sizeof(header
));
3545 DBG("Done receiving control command header: fd = %i, cmd = %s, cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
3546 conn
->sock
->fd
, lttcomm_relayd_command_str((enum lttcomm_relayd_command
) header
.cmd
),
3547 header
.cmd_version
, header
.data_size
);
3549 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
3550 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
3552 status
= RELAY_CONNECTION_STATUS_ERROR
;
3556 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
3558 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
3559 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
3562 status
= RELAY_CONNECTION_STATUS_ERROR
;
3566 if (header
.data_size
== 0) {
3568 * Manually invoke the next state as the poll loop
3569 * will not wake-up to allow us to proceed further.
3571 status
= relay_process_control_receive_payload(conn
);
3578 * Process the commands received on the control socket
3580 static enum relay_connection_status
relay_process_control(
3581 struct relay_connection
*conn
)
3583 enum relay_connection_status status
;
3585 switch (conn
->protocol
.ctrl
.state_id
) {
3586 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
3587 status
= relay_process_control_receive_header(conn
);
3589 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3590 status
= relay_process_control_receive_payload(conn
);
3593 ERR("Unknown control connection protocol state encountered.");
3600 static enum relay_connection_status
relay_process_data_receive_header(
3601 struct relay_connection
*conn
)
3604 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3605 struct data_connection_state_receive_header
*state
=
3606 &conn
->protocol
.data
.state
.receive_header
;
3607 struct lttcomm_relayd_data_hdr header
;
3608 struct relay_stream
*stream
;
3610 LTTNG_ASSERT(state
->left_to_receive
!= 0);
3612 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3613 state
->header_reception_buffer
+ state
->received
,
3614 state
->left_to_receive
, MSG_DONTWAIT
);
3617 DIAGNOSTIC_IGNORE_LOGICAL_OP
3618 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3620 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
3621 status
= RELAY_CONNECTION_STATUS_ERROR
;
3624 } else if (ret
== 0) {
3625 /* Orderly shutdown. Not necessary to print an error. */
3626 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3627 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3631 LTTNG_ASSERT(ret
> 0);
3632 LTTNG_ASSERT(ret
<= state
->left_to_receive
);
3634 state
->left_to_receive
-= ret
;
3635 state
->received
+= ret
;
3637 if (state
->left_to_receive
> 0) {
3639 * Can't transition to the protocol's next state, wait to
3640 * receive the rest of the header.
3642 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3643 state
->received
, state
->left_to_receive
,
3648 /* Transition to next state: receiving the payload. */
3649 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3651 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3652 header
.circuit_id
= be64toh(header
.circuit_id
);
3653 header
.stream_id
= be64toh(header
.stream_id
);
3654 header
.data_size
= be32toh(header
.data_size
);
3655 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3656 header
.padding_size
= be32toh(header
.padding_size
);
3657 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3659 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3661 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3662 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3664 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3665 conn
->sock
->fd
, header
.circuit_id
,
3666 header
.stream_id
, header
.data_size
,
3667 header
.net_seq_num
, header
.padding_size
);
3669 stream
= stream_get_by_id(header
.stream_id
);
3671 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3673 /* Protocol error. */
3674 status
= RELAY_CONNECTION_STATUS_ERROR
;
3678 pthread_mutex_lock(&stream
->lock
);
3679 /* Prepare stream for the reception of a new packet. */
3680 ret
= stream_init_packet(stream
, header
.data_size
,
3681 &conn
->protocol
.data
.state
.receive_payload
.rotate_index
);
3682 pthread_mutex_unlock(&stream
->lock
);
3684 ERR("Failed to rotate stream output file");
3685 status
= RELAY_CONNECTION_STATUS_ERROR
;
3686 goto end_stream_unlock
;
3695 static enum relay_connection_status
relay_process_data_receive_payload(
3696 struct relay_connection
*conn
)
3699 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3700 struct relay_stream
*stream
;
3701 struct data_connection_state_receive_payload
*state
=
3702 &conn
->protocol
.data
.state
.receive_payload
;
3703 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3704 char data_buffer
[chunk_size
];
3705 bool partial_recv
= false;
3706 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3707 uint64_t left_to_receive
= state
->left_to_receive
;
3708 struct relay_session
*session
;
3710 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3711 state
->header
.stream_id
, state
->header
.net_seq_num
,
3712 state
->received
, left_to_receive
);
3714 stream
= stream_get_by_id(state
->header
.stream_id
);
3716 /* Protocol error. */
3717 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3718 state
->header
.stream_id
);
3719 status
= RELAY_CONNECTION_STATUS_ERROR
;
3723 pthread_mutex_lock(&stream
->lock
);
3724 session
= stream
->trace
->session
;
3725 if (!conn
->session
) {
3726 ret
= connection_set_session(conn
, session
);
3728 status
= RELAY_CONNECTION_STATUS_ERROR
;
3729 goto end_stream_unlock
;
3734 * The size of the "chunk" received on any iteration is bounded by:
3735 * - the data left to receive,
3736 * - the data immediately available on the socket,
3737 * - the on-stack data buffer
3739 while (left_to_receive
> 0 && !partial_recv
) {
3740 size_t recv_size
= std::min
<uint64_t>(left_to_receive
, chunk_size
);
3741 struct lttng_buffer_view packet_chunk
;
3743 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3744 recv_size
, MSG_DONTWAIT
);
3747 DIAGNOSTIC_IGNORE_LOGICAL_OP
3748 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3750 PERROR("Socket %d error", conn
->sock
->fd
);
3751 status
= RELAY_CONNECTION_STATUS_ERROR
;
3753 goto end_stream_unlock
;
3754 } else if (ret
== 0) {
3755 /* No more data ready to be consumed on socket. */
3756 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3757 state
->header
.stream_id
);
3758 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3760 } else if (ret
< (int) recv_size
) {
3762 * All the data available on the socket has been
3765 partial_recv
= true;
3769 packet_chunk
= lttng_buffer_view_init(data_buffer
,
3771 LTTNG_ASSERT(packet_chunk
.data
);
3773 ret
= stream_write(stream
, &packet_chunk
, 0);
3775 ERR("Relay error writing data to file");
3776 status
= RELAY_CONNECTION_STATUS_ERROR
;
3777 goto end_stream_unlock
;
3780 left_to_receive
-= recv_size
;
3781 state
->received
+= recv_size
;
3782 state
->left_to_receive
= left_to_receive
;
3785 if (state
->left_to_receive
> 0) {
3787 * Did not receive all the data expected, wait for more data to
3788 * become available on the socket.
3790 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3791 state
->header
.stream_id
, state
->received
,
3792 state
->left_to_receive
);
3793 goto end_stream_unlock
;
3796 ret
= stream_write(stream
, NULL
, state
->header
.padding_size
);
3798 status
= RELAY_CONNECTION_STATUS_ERROR
;
3799 goto end_stream_unlock
;
3802 if (session_streams_have_index(session
)) {
3803 ret
= stream_update_index(stream
, state
->header
.net_seq_num
,
3804 state
->rotate_index
, &index_flushed
,
3805 state
->header
.data_size
+ state
->header
.padding_size
);
3807 ERR("Failed to update index: stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3808 stream
->stream_handle
,
3809 state
->header
.net_seq_num
, ret
);
3810 status
= RELAY_CONNECTION_STATUS_ERROR
;
3811 goto end_stream_unlock
;
3815 if (stream
->prev_data_seq
== -1ULL) {
3819 ret
= stream_complete_packet(stream
, state
->header
.data_size
+
3820 state
->header
.padding_size
, state
->header
.net_seq_num
,
3823 status
= RELAY_CONNECTION_STATUS_ERROR
;
3824 goto end_stream_unlock
;
3828 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3829 * contents of *state which are aliased (union) to the same location as
3830 * the new state. Don't use it beyond this point.
3832 connection_reset_protocol_state(conn
);
3836 close_requested
= stream
->close_requested
;
3837 pthread_mutex_unlock(&stream
->lock
);
3838 if (close_requested
&& left_to_receive
== 0) {
3839 try_stream_close(stream
);
3843 pthread_mutex_lock(&session
->lock
);
3844 uatomic_set(&session
->new_streams
, 1);
3845 pthread_mutex_unlock(&session
->lock
);
3854 * relay_process_data: Process the data received on the data socket
3856 static enum relay_connection_status
relay_process_data(
3857 struct relay_connection
*conn
)
3859 enum relay_connection_status status
;
3861 switch (conn
->protocol
.data
.state_id
) {
3862 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3863 status
= relay_process_data_receive_header(conn
);
3865 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3866 status
= relay_process_data_receive_payload(conn
);
3869 ERR("Unexpected data connection communication state.");
3876 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3880 (void) lttng_poll_del(events
, pollfd
);
3882 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
, &pollfd
, 1,
3883 fd_tracker_util_close_fd
, NULL
);
3885 ERR("Closing pollfd %d", pollfd
);
3889 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3890 int pollfd
, struct relay_connection
*conn
)
3892 const char *type_str
;
3894 switch (conn
->type
) {
3899 type_str
= "Control";
3901 case RELAY_VIEWER_COMMAND
:
3902 type_str
= "Viewer Command";
3904 case RELAY_VIEWER_NOTIFICATION
:
3905 type_str
= "Viewer Notification";
3908 type_str
= "Unknown";
3910 cleanup_connection_pollfd(events
, pollfd
);
3911 connection_put(conn
);
3912 DBG("%s connection closed with %d", type_str
, pollfd
);
3916 * This thread does the actual work
3918 static void *relay_thread_worker(void *data
__attribute__((unused
)))
3920 int ret
, err
= -1, last_seen_data_fd
= -1;
3922 struct lttng_poll_event events
;
3923 struct lttng_ht
*relay_connections_ht
;
3924 struct lttng_ht_iter iter
;
3925 struct relay_connection
*destroy_conn
= NULL
;
3927 DBG("[thread] Relay worker started");
3929 rcu_register_thread();
3931 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3933 if (testpoint(relayd_thread_worker
)) {
3934 goto error_testpoint
;
3937 health_code_update();
3939 /* table of connections indexed on socket */
3940 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3941 if (!relay_connections_ht
) {
3942 goto relay_connections_ht_error
;
3945 ret
= create_named_thread_poll_set(&events
, 2, "Worker thread epoll");
3947 goto error_poll_create
;
3950 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3957 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3959 health_code_update();
3961 /* Infinite blocking call, waiting for transmission */
3962 DBG3("Relayd worker thread polling...");
3963 health_poll_entry();
3964 ret
= lttng_poll_wait(&events
, -1);
3968 * Restart interrupted system call.
3970 if (errno
== EINTR
) {
3979 * Process control. The control connection is
3980 * prioritized so we don't starve it with high
3981 * throughput tracing data on the data connection.
3983 for (i
= 0; i
< nb_fd
; i
++) {
3984 /* Fetch once the poll data */
3985 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3986 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3988 health_code_update();
3990 /* Thread quit pipe has been closed. Killing thread. */
3991 ret
= check_thread_quit_pipe(pollfd
, revents
);
3997 /* Inspect the relay conn pipe for new connection */
3998 if (pollfd
== relay_conn_pipe
[0]) {
3999 if (revents
& LPOLLIN
) {
4000 struct relay_connection
*conn
;
4002 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
4006 ret
= lttng_poll_add(&events
,
4008 LPOLLIN
| LPOLLRDHUP
);
4010 ERR("Failed to add new connection file descriptor to poll set");
4013 connection_ht_add(relay_connections_ht
, conn
);
4014 DBG("Connection socket %d added", conn
->sock
->fd
);
4015 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4016 ERR("Relay connection pipe error");
4019 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4023 struct relay_connection
*ctrl_conn
;
4025 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
4026 /* If not found, there is a synchronization issue. */
4027 LTTNG_ASSERT(ctrl_conn
);
4029 if (ctrl_conn
->type
== RELAY_DATA
) {
4030 if (revents
& LPOLLIN
) {
4032 * Flag the last seen data fd not deleted. It will be
4033 * used as the last seen fd if any fd gets deleted in
4036 last_notdel_data_fd
= pollfd
;
4038 goto put_ctrl_connection
;
4040 LTTNG_ASSERT(ctrl_conn
->type
== RELAY_CONTROL
);
4042 if (revents
& LPOLLIN
) {
4043 enum relay_connection_status status
;
4045 status
= relay_process_control(ctrl_conn
);
4046 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
4048 * On socket error flag the session as aborted to force
4049 * the cleanup of its stream otherwise it can leak
4050 * during the lifetime of the relayd.
4052 * This prevents situations in which streams can be
4053 * left opened because an index was received, the
4054 * control connection is closed, and the data
4055 * connection is closed (uncleanly) before the packet's
4058 * Since the control connection encountered an error,
4059 * it is okay to be conservative and close the
4060 * session right now as we can't rely on the protocol
4061 * being respected anymore.
4063 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
4064 session_abort(ctrl_conn
->session
);
4067 /* Clear the connection on error or close. */
4068 relay_thread_close_connection(&events
,
4073 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4074 relay_thread_close_connection(&events
,
4076 if (last_seen_data_fd
== pollfd
) {
4077 last_seen_data_fd
= last_notdel_data_fd
;
4080 ERR("Unexpected poll events %u for control sock %d",
4082 connection_put(ctrl_conn
);
4085 put_ctrl_connection
:
4086 connection_put(ctrl_conn
);
4091 * The last loop handled a control request, go back to poll to make
4092 * sure we prioritise the control socket.
4098 if (last_seen_data_fd
>= 0) {
4099 for (i
= 0; i
< nb_fd
; i
++) {
4100 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4102 health_code_update();
4104 if (last_seen_data_fd
== pollfd
) {
4111 /* Process data connection. */
4112 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
4113 /* Fetch the poll data. */
4114 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
4115 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4116 struct relay_connection
*data_conn
;
4118 health_code_update();
4121 /* No activity for this FD (poll implementation). */
4125 /* Skip the command pipe. It's handled in the first loop. */
4126 if (pollfd
== relay_conn_pipe
[0]) {
4130 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
4132 /* Skip it. Might be removed before. */
4135 if (data_conn
->type
== RELAY_CONTROL
) {
4136 goto put_data_connection
;
4138 LTTNG_ASSERT(data_conn
->type
== RELAY_DATA
);
4140 if (revents
& LPOLLIN
) {
4141 enum relay_connection_status status
;
4143 status
= relay_process_data(data_conn
);
4144 /* Connection closed or error. */
4145 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
4147 * On socket error flag the session as aborted to force
4148 * the cleanup of its stream otherwise it can leak
4149 * during the lifetime of the relayd.
4151 * This prevents situations in which streams can be
4152 * left opened because an index was received, the
4153 * control connection is closed, and the data
4154 * connection is closed (uncleanly) before the packet's
4157 * Since the data connection encountered an error,
4158 * it is okay to be conservative and close the
4159 * session right now as we can't rely on the protocol
4160 * being respected anymore.
4162 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
4163 session_abort(data_conn
->session
);
4165 relay_thread_close_connection(&events
, pollfd
,
4168 * Every goto restart call sets the last seen fd where
4169 * here we don't really care since we gracefully
4170 * continue the loop after the connection is deleted.
4173 /* Keep last seen port. */
4174 last_seen_data_fd
= pollfd
;
4175 connection_put(data_conn
);
4178 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4179 relay_thread_close_connection(&events
, pollfd
,
4182 ERR("Unknown poll events %u for data sock %d",
4185 put_data_connection
:
4186 connection_put(data_conn
);
4188 last_seen_data_fd
= -1;
4191 /* Normal exit, no error */
4196 /* Cleanup remaining connection object. */
4198 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
4201 health_code_update();
4203 session_abort(destroy_conn
->session
);
4206 * No need to grab another ref, because we own
4209 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
4214 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
4216 lttng_ht_destroy(relay_connections_ht
);
4217 relay_connections_ht_error
:
4218 /* Close relay conn pipes */
4219 (void) fd_tracker_util_pipe_close(the_fd_tracker
,
4222 DBG("Thread exited with error");
4224 DBG("Worker thread cleanup complete");
4228 ERR("Health error occurred in %s", __func__
);
4230 health_unregister(health_relayd
);
4231 rcu_unregister_thread();
4232 lttng_relay_stop_threads();
4237 * Create the relay command pipe to wake thread_manage_apps.
4238 * Closed in cleanup().
4240 static int create_relay_conn_pipe(void)
4242 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
4243 "Relayd connection pipe", relay_conn_pipe
);
4246 static int stdio_open(void *data
__attribute__((unused
)), int *fds
)
4248 fds
[0] = fileno(stdout
);
4249 fds
[1] = fileno(stderr
);
4253 static int track_stdio(void)
4256 const char *names
[] = { "stdout", "stderr" };
4258 return fd_tracker_open_unsuspendable_fd(the_fd_tracker
, fds
,
4259 names
, 2, stdio_open
, NULL
);
4265 int main(int argc
, char **argv
)
4267 bool thread_is_rcu_registered
= false;
4268 int ret
= 0, retval
= 0;
4270 char *unlinked_file_directory_path
= NULL
, *output_path
= NULL
;
4272 /* Parse environment variables */
4273 ret
= parse_env_options();
4281 * Command line arguments overwrite environment.
4284 if (set_options(argc
, argv
)) {
4289 if (set_signal_handler()) {
4294 relayd_config_log();
4296 if (opt_print_version
) {
4302 ret
= fclose(stdin
);
4304 PERROR("Failed to close stdin");
4308 DBG("Clear command %s", opt_allow_clear
? "allowed" : "disallowed");
4310 /* Try to create directory if -o, --output is specified. */
4311 if (opt_output_path
) {
4312 if (*opt_output_path
!= '/') {
4313 ERR("Please specify an absolute path for -o, --output PATH");
4318 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
4321 ERR("Unable to create %s", opt_output_path
);
4328 if (opt_daemon
|| opt_background
) {
4329 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4337 if (opt_working_directory
) {
4338 ret
= utils_change_working_directory(opt_working_directory
);
4340 /* All errors are already logged. */
4345 sessiond_trace_chunk_registry
= sessiond_trace_chunk_registry_create();
4346 if (!sessiond_trace_chunk_registry
) {
4347 ERR("Failed to initialize session daemon trace chunk registry");
4353 * The RCU thread registration (and use, through the fd-tracker's
4354 * creation) is done after the daemonization to allow us to not
4355 * deal with liburcu's fork() management as the call RCU needs to
4358 rcu_register_thread();
4359 thread_is_rcu_registered
= true;
4361 output_path
= create_output_path("");
4363 ERR("Failed to get output path");
4367 ret
= asprintf(&unlinked_file_directory_path
, "%s/%s", output_path
,
4368 DEFAULT_UNLINKED_FILES_DIRECTORY
);
4371 ERR("Failed to format unlinked file directory path");
4375 the_fd_tracker
= fd_tracker_create(
4376 unlinked_file_directory_path
, lttng_opt_fd_pool_size
);
4377 free(unlinked_file_directory_path
);
4378 if (!the_fd_tracker
) {
4383 ret
= track_stdio();
4389 /* Initialize thread health monitoring */
4390 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
4391 if (!health_relayd
) {
4392 PERROR("health_app_create error");
4397 /* Create thread quit pipe */
4398 if (init_thread_quit_pipe()) {
4403 /* Setup the thread apps communication pipe. */
4404 if (create_relay_conn_pipe()) {
4409 /* Init relay command queue. */
4410 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
4412 /* Initialize communication library */
4414 lttcomm_inet_init();
4416 /* tables of sessions indexed by session ID */
4417 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4423 /* tables of streams indexed by stream ID */
4424 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4425 if (!relay_streams_ht
) {
4430 /* tables of streams indexed by stream ID */
4431 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4432 if (!viewer_streams_ht
) {
4437 ret
= init_health_quit_pipe();
4443 /* Create thread to manage the client socket */
4444 ret
= pthread_create(&health_thread
, default_pthread_attr(),
4445 thread_manage_health_relayd
, (void *) NULL
);
4448 PERROR("pthread_create health");
4453 /* Setup the dispatcher thread */
4454 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
4455 relay_thread_dispatcher
, (void *) NULL
);
4458 PERROR("pthread_create dispatcher");
4460 goto exit_dispatcher_thread
;
4463 /* Setup the worker thread */
4464 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
4465 relay_thread_worker
, NULL
);
4468 PERROR("pthread_create worker");
4470 goto exit_worker_thread
;
4473 /* Setup the listener thread */
4474 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
4475 relay_thread_listener
, (void *) NULL
);
4478 PERROR("pthread_create listener");
4480 goto exit_listener_thread
;
4483 ret
= relayd_live_create(live_uri
);
4485 ERR("Starting live viewer threads");
4491 * This is where we start awaiting program completion (e.g. through
4492 * signal that asks threads to teardown).
4495 ret
= relayd_live_join();
4501 ret
= pthread_join(listener_thread
, &status
);
4504 PERROR("pthread_join listener_thread");
4508 exit_listener_thread
:
4509 ret
= pthread_join(worker_thread
, &status
);
4512 PERROR("pthread_join worker_thread");
4517 ret
= pthread_join(dispatcher_thread
, &status
);
4520 PERROR("pthread_join dispatcher_thread");
4523 exit_dispatcher_thread
:
4525 ret
= pthread_join(health_thread
, &status
);
4528 PERROR("pthread_join health_thread");
4533 * Wait for all pending call_rcu work to complete before tearing
4534 * down data structures. call_rcu worker may be trying to
4535 * perform lookups in those structures.
4540 /* Ensure all prior call_rcu are done. */
4543 if (thread_is_rcu_registered
) {
4544 rcu_unregister_thread();