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>
37 #include <lttng/lttng.h>
38 #include <common/common.h>
39 #include <common/compat/poll.h>
40 #include <common/compat/socket.h>
41 #include <common/compat/endian.h>
42 #include <common/compat/getenv.h>
43 #include <common/defaults.h>
44 #include <common/daemonize.h>
45 #include <common/futex.h>
46 #include <common/sessiond-comm/sessiond-comm.h>
47 #include <common/sessiond-comm/inet.h>
48 #include <common/sessiond-comm/relayd.h>
49 #include <common/uri.h>
50 #include <common/utils.h>
51 #include <common/align.h>
52 #include <common/config/session-config.h>
53 #include <common/dynamic-buffer.h>
54 #include <common/buffer-view.h>
55 #include <common/string-utils/format.h>
56 #include <common/fd-tracker/fd-tracker.h>
57 #include <common/fd-tracker/utils.h>
59 #include "backward-compatibility-group-by.h"
61 #include "connection.h"
62 #include "ctf-trace.h"
63 #include "health-relayd.h"
66 #include "lttng-relayd.h"
68 #include "sessiond-trace-chunks.h"
70 #include "tcp_keep_alive.h"
71 #include "testpoint.h"
72 #include "tracefile-array.h"
75 #include "viewer-stream.h"
77 static const char *help_msg
=
78 #ifdef LTTNG_EMBED_HELP
79 #include <lttng-relayd.8.h>
85 enum relay_connection_status
{
86 RELAY_CONNECTION_STATUS_OK
,
87 /* An error occurred while processing an event on the connection. */
88 RELAY_CONNECTION_STATUS_ERROR
,
89 /* Connection closed/shutdown cleanly. */
90 RELAY_CONNECTION_STATUS_CLOSED
,
93 /* command line options */
94 char *opt_output_path
, *opt_working_directory
;
95 static int opt_daemon
, opt_background
, opt_print_version
, opt_allow_clear
= 1;
96 enum relay_group_output_by opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
;
99 * We need to wait for listener and live listener threads, as well as
100 * health check thread, before being ready to signal readiness.
102 #define NR_LTTNG_RELAY_READY 3
103 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
105 /* Size of receive buffer. */
106 #define RECV_DATA_BUFFER_SIZE 65536
108 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
109 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
111 static struct lttng_uri
*control_uri
;
112 static struct lttng_uri
*data_uri
;
113 static struct lttng_uri
*live_uri
;
115 const char *progname
;
117 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
118 static int tracing_group_name_override
;
120 const char * const config_section_name
= "relayd";
123 * Quit pipe for all threads. This permits a single cancellation point
124 * for all threads when receiving an event on the pipe.
126 int thread_quit_pipe
[2] = { -1, -1 };
129 * This pipe is used to inform the worker thread that a command is queued and
130 * ready to be processed.
132 static int relay_conn_pipe
[2] = { -1, -1 };
134 /* Shared between threads */
135 static int dispatch_thread_exit
;
137 static pthread_t listener_thread
;
138 static pthread_t dispatcher_thread
;
139 static pthread_t worker_thread
;
140 static pthread_t health_thread
;
143 * last_relay_stream_id_lock protects last_relay_stream_id increment
144 * atomicity on 32-bit architectures.
146 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
147 static uint64_t last_relay_stream_id
;
150 * Relay command queue.
152 * The relay_thread_listener and relay_thread_dispatcher communicate with this
155 static struct relay_conn_queue relay_conn_queue
;
157 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
158 static unsigned int lttng_opt_fd_pool_size
= -1;
160 /* Global relay stream hash table. */
161 struct lttng_ht
*relay_streams_ht
;
163 /* Global relay viewer stream hash table. */
164 struct lttng_ht
*viewer_streams_ht
;
166 /* Global relay sessions hash table. */
167 struct lttng_ht
*sessions_ht
;
169 /* Relayd health monitoring */
170 struct health_app
*health_relayd
;
172 struct sessiond_trace_chunk_registry
*sessiond_trace_chunk_registry
;
174 /* Global fd tracker. */
175 struct fd_tracker
*the_fd_tracker
;
177 static struct option long_options
[] = {
178 { "control-port", 1, 0, 'C', },
179 { "data-port", 1, 0, 'D', },
180 { "live-port", 1, 0, 'L', },
181 { "daemonize", 0, 0, 'd', },
182 { "background", 0, 0, 'b', },
183 { "group", 1, 0, 'g', },
184 { "fd-pool-size", 1, 0, '\0', },
185 { "help", 0, 0, 'h', },
186 { "output", 1, 0, 'o', },
187 { "verbose", 0, 0, 'v', },
188 { "config", 1, 0, 'f' },
189 { "version", 0, 0, 'V' },
190 { "working-directory", 1, 0, 'w', },
191 { "group-output-by-session", 0, 0, 's', },
192 { "group-output-by-host", 0, 0, 'p', },
193 { "disallow-clear", 0, 0, 'x' },
197 static const char *config_ignore_options
[] = { "help", "config", "version" };
199 static void print_version(void) {
200 fprintf(stdout
, "%s\n", VERSION
);
203 static void relayd_config_log(void)
205 DBG("LTTng-relayd " VERSION
" - " VERSION_NAME
"%s%s",
206 GIT_VERSION
[0] == '\0' ? "" : " - " GIT_VERSION
,
207 EXTRA_VERSION_NAME
[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME
);
208 if (EXTRA_VERSION_DESCRIPTION
[0] != '\0') {
209 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION
"\n");
211 if (EXTRA_VERSION_PATCHES
[0] != '\0') {
212 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES
"\n");
217 * Take an option from the getopt output and set it in the right variable to be
220 * Return 0 on success else a negative value.
222 static int set_option(int opt
, const char *arg
, const char *optname
)
228 if (!strcmp(optname
, "fd-pool-size")) {
232 v
= strtoul(arg
, NULL
, 0);
233 if (errno
!= 0 || !isdigit((unsigned char) arg
[0])) {
234 ERR("Wrong value in --fd-pool-size parameter: %s", arg
);
239 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s", arg
);
243 lttng_opt_fd_pool_size
= (unsigned int) v
;
245 fprintf(stderr
, "unknown option %s", optname
);
247 fprintf(stderr
, " with arg %s\n", arg
);
252 if (lttng_is_setuid_setgid()) {
253 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
254 "-C, --control-port");
256 ret
= uri_parse(arg
, &control_uri
);
258 ERR("Invalid control URI specified");
261 if (control_uri
->port
== 0) {
262 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
267 if (lttng_is_setuid_setgid()) {
268 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
271 ret
= uri_parse(arg
, &data_uri
);
273 ERR("Invalid data URI specified");
276 if (data_uri
->port
== 0) {
277 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
282 if (lttng_is_setuid_setgid()) {
283 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
286 ret
= uri_parse(arg
, &live_uri
);
288 ERR("Invalid live URI specified");
291 if (live_uri
->port
== 0) {
292 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
303 if (lttng_is_setuid_setgid()) {
304 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
307 tracing_group_name
= strdup(arg
);
308 if (tracing_group_name
== NULL
) {
313 tracing_group_name_override
= 1;
317 ret
= utils_show_help(8, "lttng-relayd", help_msg
);
319 ERR("Cannot show --help for `lttng-relayd`");
324 opt_print_version
= 1;
327 if (lttng_is_setuid_setgid()) {
328 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
331 ret
= asprintf(&opt_output_path
, "%s", arg
);
334 PERROR("asprintf opt_output_path");
340 if (lttng_is_setuid_setgid()) {
341 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
342 "-w, --working-directory");
344 ret
= asprintf(&opt_working_directory
, "%s", arg
);
347 PERROR("asprintf opt_working_directory");
354 /* Verbose level can increase using multiple -v */
356 lttng_opt_verbose
= config_parse_value(arg
);
358 /* Only 3 level of verbosity (-vvv). */
359 if (lttng_opt_verbose
< 3) {
360 lttng_opt_verbose
+= 1;
365 if (opt_group_output_by
!= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
366 ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
369 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_SESSION
;
372 if (opt_group_output_by
!= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
373 ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
376 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_HOST
;
383 /* Unknown option or other error.
384 * Error is printed by getopt, just return */
397 * config_entry_handler_cb used to handle options read from a config file.
398 * See config_entry_handler_cb comment in common/config/session-config.h for the
399 * return value conventions.
401 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
405 if (!entry
|| !entry
->name
|| !entry
->value
) {
410 /* Check if the option is to be ignored */
411 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
412 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
417 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
418 /* Ignore if entry name is not fully matched. */
419 if (strcmp(entry
->name
, long_options
[i
].name
)) {
424 * If the option takes no argument on the command line,
425 * we have to check if the value is "true". We support
426 * non-zero numeric values, true, on and yes.
428 if (!long_options
[i
].has_arg
) {
429 ret
= config_parse_value(entry
->value
);
432 WARN("Invalid configuration value \"%s\" for option %s",
433 entry
->value
, entry
->name
);
435 /* False, skip boolean config option. */
440 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
444 WARN("Unrecognized option \"%s\" in daemon configuration file.",
451 static int parse_env_options(void)
456 value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV
);
458 opt_working_directory
= strdup(value
);
459 if (!opt_working_directory
) {
460 ERR("Failed to allocate working directory string (\"%s\")",
468 static int set_fd_pool_size(void)
471 struct rlimit rlimit
;
473 ret
= getrlimit(RLIMIT_NOFILE
, &rlimit
);
475 PERROR("Failed to get file descriptor limit");
480 DBG("File descriptor count limits are %" PRIu64
" (soft) and %" PRIu64
" (hard)",
481 (uint64_t) rlimit
.rlim_cur
,
482 (uint64_t) rlimit
.rlim_max
);
483 if (lttng_opt_fd_pool_size
== -1) {
484 /* Use default value (soft limit - reserve). */
485 if (rlimit
.rlim_cur
< DEFAULT_RELAYD_MIN_FD_POOL_SIZE
) {
486 ERR("The process' file number limit is too low (%" PRIu64
"). The process' file number limit must be set to at least %i.",
487 (uint64_t) rlimit
.rlim_cur
, DEFAULT_RELAYD_MIN_FD_POOL_SIZE
);
491 lttng_opt_fd_pool_size
= rlimit
.rlim_cur
-
492 DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
;
496 if (lttng_opt_fd_pool_size
< DEFAULT_RELAYD_MIN_FD_POOL_SIZE
) {
497 ERR("File descriptor pool size must be set to at least %d",
498 DEFAULT_RELAYD_MIN_FD_POOL_SIZE
);
503 if (lttng_opt_fd_pool_size
> rlimit
.rlim_cur
) {
504 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64
").",
505 lttng_opt_fd_pool_size
, (uint64_t) rlimit
.rlim_cur
);
510 DBG("File descriptor pool size argument (%u) adjusted to %u to accommodates transient fd uses",
511 lttng_opt_fd_pool_size
,
512 lttng_opt_fd_pool_size
- DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
);
513 lttng_opt_fd_pool_size
-= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
;
518 static int set_options(int argc
, char **argv
)
520 int c
, ret
= 0, option_index
= 0, retval
= 0;
521 int orig_optopt
= optopt
, orig_optind
= optind
;
522 char *default_address
, *optstring
;
523 const char *config_path
= NULL
;
525 optstring
= utils_generate_optstring(long_options
,
526 sizeof(long_options
) / sizeof(struct option
));
532 /* Check for the --config option */
534 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
535 &option_index
)) != -1) {
539 } else if (c
!= 'f') {
543 if (lttng_is_setuid_setgid()) {
544 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
547 config_path
= utils_expand_path(optarg
);
549 ERR("Failed to resolve path: %s", optarg
);
554 ret
= config_get_section_entries(config_path
, config_section_name
,
555 config_entry_handler
, NULL
);
558 ERR("Invalid configuration option at line %i", ret
);
564 /* Reset getopt's global state */
565 optopt
= orig_optopt
;
566 optind
= orig_optind
;
568 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
573 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
580 /* assign default values */
581 if (control_uri
== NULL
) {
582 ret
= asprintf(&default_address
,
583 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
584 DEFAULT_NETWORK_CONTROL_PORT
);
586 PERROR("asprintf default data address");
591 ret
= uri_parse(default_address
, &control_uri
);
592 free(default_address
);
594 ERR("Invalid control URI specified");
599 if (data_uri
== NULL
) {
600 ret
= asprintf(&default_address
,
601 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
602 DEFAULT_NETWORK_DATA_PORT
);
604 PERROR("asprintf default data address");
609 ret
= uri_parse(default_address
, &data_uri
);
610 free(default_address
);
612 ERR("Invalid data URI specified");
617 if (live_uri
== NULL
) {
618 ret
= asprintf(&default_address
,
619 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
620 DEFAULT_NETWORK_VIEWER_PORT
);
622 PERROR("asprintf default viewer control address");
627 ret
= uri_parse(default_address
, &live_uri
);
628 free(default_address
);
630 ERR("Invalid viewer control URI specified");
635 ret
= set_fd_pool_size();
641 if (opt_group_output_by
== RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
642 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_HOST
;
644 if (opt_allow_clear
) {
645 /* Check if env variable exists. */
646 const char *value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV
);
648 ret
= config_parse_value(value
);
650 ERR("Invalid value for %s specified", DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV
);
654 opt_allow_clear
= !ret
;
663 static void print_global_objects(void)
665 print_viewer_streams();
666 print_relay_streams();
670 static int noop_close(void *data
, int *fds
)
675 static void untrack_stdio(void)
677 int fds
[] = { fileno(stdout
), fileno(stderr
) };
680 * noop_close is used since we don't really want to close
681 * the stdio output fds; we merely want to stop tracking them.
683 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
684 fds
, 2, noop_close
, NULL
);
690 static void relayd_cleanup(void)
692 print_global_objects();
696 if (viewer_streams_ht
)
697 lttng_ht_destroy(viewer_streams_ht
);
698 if (relay_streams_ht
)
699 lttng_ht_destroy(relay_streams_ht
);
701 lttng_ht_destroy(sessions_ht
);
703 free(opt_output_path
);
704 free(opt_working_directory
);
707 health_app_destroy(health_relayd
);
709 /* Close thread quit pipes */
710 if (health_quit_pipe
[0] != -1) {
711 (void) fd_tracker_util_pipe_close(
712 the_fd_tracker
, health_quit_pipe
);
714 if (thread_quit_pipe
[0] != -1) {
715 (void) fd_tracker_util_pipe_close(
716 the_fd_tracker
, thread_quit_pipe
);
718 if (sessiond_trace_chunk_registry
) {
719 sessiond_trace_chunk_registry_destroy(
720 sessiond_trace_chunk_registry
);
722 if (the_fd_tracker
) {
725 * fd_tracker_destroy() will log the contents of the fd-tracker
726 * if a leak is detected.
728 fd_tracker_destroy(the_fd_tracker
);
731 uri_free(control_uri
);
733 /* Live URI is freed in the live thread. */
735 if (tracing_group_name_override
) {
736 free((void *) tracing_group_name
);
741 * Write to writable pipe used to notify a thread.
743 static int notify_thread_pipe(int wpipe
)
747 ret
= lttng_write(wpipe
, "!", 1);
749 PERROR("write poll pipe");
757 static int notify_health_quit_pipe(int *pipe
)
761 ret
= lttng_write(pipe
[1], "4", 1);
763 PERROR("write relay health quit");
772 * Stop all relayd and relayd-live threads.
774 int lttng_relay_stop_threads(void)
778 /* Stopping all threads */
779 DBG("Terminating all threads");
780 if (notify_thread_pipe(thread_quit_pipe
[1])) {
781 ERR("write error on thread quit pipe");
785 if (notify_health_quit_pipe(health_quit_pipe
)) {
786 ERR("write error on health quit pipe");
789 /* Dispatch thread */
790 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
791 futex_nto1_wake(&relay_conn_queue
.futex
);
793 if (relayd_live_stop()) {
794 ERR("Error stopping live threads");
801 * Signal handler for the daemon
803 * Simply stop all worker threads, leaving main() return gracefully after
804 * joining all threads and calling cleanup().
806 static void sighandler(int sig
)
810 DBG("SIGINT caught");
811 if (lttng_relay_stop_threads()) {
812 ERR("Error stopping threads");
816 DBG("SIGTERM caught");
817 if (lttng_relay_stop_threads()) {
818 ERR("Error stopping threads");
822 CMM_STORE_SHARED(recv_child_signal
, 1);
830 * Setup signal handler for :
831 * SIGINT, SIGTERM, SIGPIPE
833 static int set_signal_handler(void)
839 if ((ret
= sigemptyset(&sigset
)) < 0) {
840 PERROR("sigemptyset");
847 sa
.sa_handler
= sighandler
;
848 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
853 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
858 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
863 sa
.sa_handler
= SIG_IGN
;
864 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
869 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
874 void lttng_relay_notify_ready(void)
876 /* Notify the parent of the fork() process that we are ready. */
877 if (opt_daemon
|| opt_background
) {
878 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
879 kill(child_ppid
, SIGUSR1
);
885 * Init thread quit pipe.
887 * Return -1 on error or 0 if all pipes are created.
889 static int init_thread_quit_pipe(void)
891 return fd_tracker_util_pipe_open_cloexec(
892 the_fd_tracker
, "Quit pipe", thread_quit_pipe
);
896 * Init health quit pipe.
898 * Return -1 on error or 0 if all pipes are created.
900 static int init_health_quit_pipe(void)
902 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
903 "Health quit pipe", health_quit_pipe
);
907 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
909 static int create_named_thread_poll_set(struct lttng_poll_event
*events
,
910 int size
, const char *name
)
914 if (events
== NULL
|| size
== 0) {
919 ret
= fd_tracker_util_poll_create(the_fd_tracker
,
920 name
, events
, 1, LTTNG_CLOEXEC
);
922 PERROR("Failed to create \"%s\" poll file descriptor", name
);
927 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
939 * Check if the thread quit pipe was triggered.
941 * Return 1 if it was triggered else 0;
943 static int check_thread_quit_pipe(int fd
, uint32_t events
)
945 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
952 static int create_sock(void *data
, int *out_fd
)
955 struct lttcomm_sock
*sock
= data
;
957 ret
= lttcomm_create_sock(sock
);
967 static int close_sock(void *data
, int *in_fd
)
969 struct lttcomm_sock
*sock
= data
;
971 return sock
->ops
->close(sock
);
974 static int accept_sock(void *data
, int *out_fd
)
977 /* Socks is an array of in_sock, out_sock. */
978 struct lttcomm_sock
**socks
= data
;
979 struct lttcomm_sock
*in_sock
= socks
[0];
981 socks
[1] = in_sock
->ops
->accept(in_sock
);
986 *out_fd
= socks
[1]->fd
;
992 * Create and init socket from uri.
994 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
,
998 struct lttcomm_sock
*sock
= NULL
;
999 char uri_str
[PATH_MAX
];
1000 char *formated_name
= NULL
;
1002 sock
= lttcomm_alloc_sock_from_uri(uri
);
1004 ERR("Allocating socket");
1009 * Don't fail to create the socket if the name can't be built as it is
1010 * only used for debugging purposes.
1012 ret
= uri_to_str_url(uri
, uri_str
, sizeof(uri_str
));
1013 uri_str
[sizeof(uri_str
) - 1] = '\0';
1015 ret
= asprintf(&formated_name
, "%s socket @ %s", name
,
1018 formated_name
= NULL
;
1022 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &sock_fd
,
1023 (const char **) (formated_name
? &formated_name
: NULL
),
1024 1, create_sock
, sock
);
1026 PERROR("Failed to open \"%s\" relay socket",
1027 formated_name
?: "Unknown");
1030 DBG("Listening on %s socket %d", name
, sock
->fd
);
1032 ret
= sock
->ops
->bind(sock
);
1034 PERROR("Failed to bind socket");
1038 ret
= sock
->ops
->listen(sock
, -1);
1044 free(formated_name
);
1049 lttcomm_destroy_sock(sock
);
1051 free(formated_name
);
1056 struct lttcomm_sock
*accept_relayd_sock(struct lttcomm_sock
*listening_sock
,
1060 struct lttcomm_sock
*socks
[2] = { listening_sock
, NULL
};
1061 struct lttcomm_sock
*new_sock
= NULL
;
1063 ret
= fd_tracker_open_unsuspendable_fd(
1064 the_fd_tracker
, &out_fd
,
1065 (const char **) &name
,
1066 1, accept_sock
, &socks
);
1070 new_sock
= socks
[1];
1071 DBG("%s accepted, socket %d", name
, new_sock
->fd
);
1077 * This thread manages the listening for new connections on the network
1079 static void *relay_thread_listener(void *data
)
1081 int i
, ret
, pollfd
, err
= -1;
1082 uint32_t revents
, nb_fd
;
1083 struct lttng_poll_event events
;
1084 struct lttcomm_sock
*control_sock
, *data_sock
;
1086 DBG("[thread] Relay listener started");
1088 rcu_register_thread();
1089 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
1091 health_code_update();
1093 control_sock
= relay_socket_create(control_uri
, "Control listener");
1094 if (!control_sock
) {
1095 goto error_sock_control
;
1098 data_sock
= relay_socket_create(data_uri
, "Data listener");
1100 goto error_sock_relay
;
1104 * Pass 3 as size here for the thread quit pipe, control and
1107 ret
= create_named_thread_poll_set(&events
, 3, "Listener thread epoll");
1109 goto error_create_poll
;
1112 /* Add the control socket */
1113 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1115 goto error_poll_add
;
1118 /* Add the data socket */
1119 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1121 goto error_poll_add
;
1124 lttng_relay_notify_ready();
1126 if (testpoint(relayd_thread_listener
)) {
1127 goto error_testpoint
;
1131 health_code_update();
1133 DBG("Listener accepting connections");
1136 health_poll_entry();
1137 ret
= lttng_poll_wait(&events
, -1);
1141 * Restart interrupted system call.
1143 if (errno
== EINTR
) {
1151 DBG("Relay new connection received");
1152 for (i
= 0; i
< nb_fd
; i
++) {
1153 health_code_update();
1155 /* Fetch once the poll data */
1156 revents
= LTTNG_POLL_GETEV(&events
, i
);
1157 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1159 /* Thread quit pipe has been closed. Killing thread. */
1160 ret
= check_thread_quit_pipe(pollfd
, revents
);
1166 if (revents
& LPOLLIN
) {
1168 * A new connection is requested, therefore a
1169 * sessiond/consumerd connection is allocated in
1170 * this thread, enqueued to a global queue and
1171 * dequeued (and freed) in the worker thread.
1174 struct relay_connection
*new_conn
;
1175 struct lttcomm_sock
*newsock
= NULL
;
1176 enum connection_type type
;
1178 if (pollfd
== data_sock
->fd
) {
1180 newsock
= accept_relayd_sock(data_sock
,
1181 "Data socket to relayd");
1183 assert(pollfd
== control_sock
->fd
);
1184 type
= RELAY_CONTROL
;
1185 newsock
= accept_relayd_sock(control_sock
,
1186 "Control socket to relayd");
1189 PERROR("accepting sock");
1193 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
1196 PERROR("setsockopt inet");
1197 lttcomm_destroy_sock(newsock
);
1201 ret
= socket_apply_keep_alive_config(newsock
->fd
);
1203 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1205 lttcomm_destroy_sock(newsock
);
1209 new_conn
= connection_create(newsock
, type
);
1211 lttcomm_destroy_sock(newsock
);
1215 /* Enqueue request for the dispatcher thread. */
1216 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
1220 * Wake the dispatch queue futex.
1221 * Implicit memory barrier with the
1222 * exchange in cds_wfcq_enqueue.
1224 futex_nto1_wake(&relay_conn_queue
.futex
);
1225 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1226 ERR("socket poll error");
1229 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1239 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
1241 if (data_sock
->fd
>= 0) {
1242 int data_sock_fd
= data_sock
->fd
;
1244 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1245 &data_sock_fd
, 1, close_sock
,
1248 PERROR("Failed to close the data listener socket file descriptor");
1252 lttcomm_destroy_sock(data_sock
);
1254 if (control_sock
->fd
>= 0) {
1255 int control_sock_fd
= control_sock
->fd
;
1257 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1258 &control_sock_fd
, 1, close_sock
,
1261 PERROR("Failed to close the control listener socket file descriptor");
1263 control_sock
->fd
= -1;
1265 lttcomm_destroy_sock(control_sock
);
1269 ERR("Health error occurred in %s", __func__
);
1271 health_unregister(health_relayd
);
1272 rcu_unregister_thread();
1273 DBG("Relay listener thread cleanup complete");
1274 lttng_relay_stop_threads();
1279 * This thread manages the dispatching of the requests to worker threads
1281 static void *relay_thread_dispatcher(void *data
)
1285 struct cds_wfcq_node
*node
;
1286 struct relay_connection
*new_conn
= NULL
;
1288 DBG("[thread] Relay dispatcher started");
1290 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
1292 if (testpoint(relayd_thread_dispatcher
)) {
1293 goto error_testpoint
;
1296 health_code_update();
1299 health_code_update();
1301 /* Atomically prepare the queue futex */
1302 futex_nto1_prepare(&relay_conn_queue
.futex
);
1304 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1309 health_code_update();
1311 /* Dequeue commands */
1312 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1313 &relay_conn_queue
.tail
);
1315 DBG("Woken up but nothing in the relay command queue");
1316 /* Continue thread execution */
1319 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1321 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1324 * Inform worker thread of the new request. This
1325 * call is blocking so we can be assured that
1326 * the data will be read at some point in time
1327 * or wait to the end of the world :)
1329 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1331 PERROR("write connection pipe");
1332 connection_put(new_conn
);
1335 } while (node
!= NULL
);
1337 /* Futex wait on queue. Blocking call on futex() */
1338 health_poll_entry();
1339 futex_nto1_wait(&relay_conn_queue
.futex
);
1343 /* Normal exit, no error */
1350 ERR("Health error occurred in %s", __func__
);
1352 health_unregister(health_relayd
);
1353 DBG("Dispatch thread dying");
1354 lttng_relay_stop_threads();
1358 static bool session_streams_have_index(const struct relay_session
*session
)
1360 return session
->minor
>= 4 && !session
->snapshot
;
1364 * Handle the RELAYD_CREATE_SESSION command.
1366 * On success, send back the session id or else return a negative value.
1368 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1369 struct relay_connection
*conn
,
1370 const struct lttng_buffer_view
*payload
)
1374 struct relay_session
*session
= NULL
;
1375 struct lttcomm_relayd_create_session_reply_2_11 reply
= {};
1376 char session_name
[LTTNG_NAME_MAX
] = {};
1377 char hostname
[LTTNG_HOST_NAME_MAX
] = {};
1378 uint32_t live_timer
= 0;
1379 bool snapshot
= false;
1380 bool session_name_contains_creation_timestamp
= false;
1381 /* Left nil for peers < 2.11. */
1382 char base_path
[LTTNG_PATH_MAX
] = {};
1383 lttng_uuid sessiond_uuid
= {};
1384 LTTNG_OPTIONAL(uint64_t) id_sessiond
= {};
1385 LTTNG_OPTIONAL(uint64_t) current_chunk_id
= {};
1386 LTTNG_OPTIONAL(time_t) creation_time
= {};
1387 struct lttng_dynamic_buffer reply_payload
;
1389 lttng_dynamic_buffer_init(&reply_payload
);
1391 if (conn
->minor
< 4) {
1392 /* From 2.1 to 2.3 */
1394 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1395 /* From 2.4 to 2.10 */
1396 ret
= cmd_create_session_2_4(payload
, session_name
,
1397 hostname
, &live_timer
, &snapshot
);
1399 bool has_current_chunk
;
1400 uint64_t current_chunk_id_value
;
1401 time_t creation_time_value
;
1402 uint64_t id_sessiond_value
;
1404 /* From 2.11 to ... */
1405 ret
= cmd_create_session_2_11(payload
, session_name
, hostname
,
1406 base_path
, &live_timer
, &snapshot
, &id_sessiond_value
,
1407 sessiond_uuid
, &has_current_chunk
,
1408 ¤t_chunk_id_value
, &creation_time_value
,
1409 &session_name_contains_creation_timestamp
);
1410 if (lttng_uuid_is_nil(sessiond_uuid
)) {
1411 /* The nil UUID is reserved for pre-2.11 clients. */
1412 ERR("Illegal nil UUID announced by peer in create session command");
1416 LTTNG_OPTIONAL_SET(&id_sessiond
, id_sessiond_value
);
1417 LTTNG_OPTIONAL_SET(&creation_time
, creation_time_value
);
1418 if (has_current_chunk
) {
1419 LTTNG_OPTIONAL_SET(¤t_chunk_id
,
1420 current_chunk_id_value
);
1428 session
= session_create(session_name
, hostname
, base_path
, live_timer
,
1429 snapshot
, sessiond_uuid
,
1430 id_sessiond
.is_set
? &id_sessiond
.value
: NULL
,
1431 current_chunk_id
.is_set
? ¤t_chunk_id
.value
: NULL
,
1432 creation_time
.is_set
? &creation_time
.value
: NULL
,
1433 conn
->major
, conn
->minor
,
1434 session_name_contains_creation_timestamp
);
1439 assert(!conn
->session
);
1440 conn
->session
= session
;
1441 DBG("Created session %" PRIu64
, session
->id
);
1443 reply
.generic
.session_id
= htobe64(session
->id
);
1447 reply
.generic
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1449 reply
.generic
.ret_code
= htobe32(LTTNG_OK
);
1452 if (conn
->minor
< 11) {
1453 /* From 2.1 to 2.10 */
1454 ret
= lttng_dynamic_buffer_append(&reply_payload
,
1455 &reply
.generic
, sizeof(reply
.generic
));
1457 ERR("Failed to append \"create session\" command reply header to payload buffer");
1462 const uint32_t output_path_length
=
1463 session
? strlen(session
->output_path
) + 1 : 0;
1465 reply
.output_path_length
= htobe32(output_path_length
);
1466 ret
= lttng_dynamic_buffer_append(
1467 &reply_payload
, &reply
, sizeof(reply
));
1469 ERR("Failed to append \"create session\" command reply header to payload buffer");
1473 if (output_path_length
) {
1474 ret
= lttng_dynamic_buffer_append(&reply_payload
,
1475 session
->output_path
,
1476 output_path_length
);
1478 ERR("Failed to append \"create session\" command reply path to payload buffer");
1484 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, reply_payload
.data
,
1485 reply_payload
.size
, 0);
1486 if (send_ret
< (ssize_t
) reply_payload
.size
) {
1487 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1488 reply_payload
.size
, send_ret
);
1492 if (ret
< 0 && session
) {
1493 session_put(session
);
1495 lttng_dynamic_buffer_reset(&reply_payload
);
1500 * When we have received all the streams and the metadata for a channel,
1501 * we make them visible to the viewer threads.
1503 static void publish_connection_local_streams(struct relay_connection
*conn
)
1505 struct relay_stream
*stream
;
1506 struct relay_session
*session
= conn
->session
;
1509 * We publish all streams belonging to a session atomically wrt
1512 pthread_mutex_lock(&session
->lock
);
1514 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1516 stream_publish(stream
);
1521 * Inform the viewer that there are new streams in the session.
1523 if (session
->viewer_attached
) {
1524 uatomic_set(&session
->new_streams
, 1);
1526 pthread_mutex_unlock(&session
->lock
);
1529 static int conform_channel_path(char *channel_path
)
1533 if (strstr("../", channel_path
)) {
1534 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1540 if (*channel_path
== '/') {
1541 const size_t len
= strlen(channel_path
);
1544 * Channel paths from peers prior to 2.11 are expressed as an
1545 * absolute path that is, in reality, relative to the relay
1546 * daemon's output directory. Remove the leading slash so it
1547 * is correctly interpreted as a relative path later on.
1549 * len (and not len - 1) is used to copy the trailing NULL.
1551 bcopy(channel_path
+ 1, channel_path
, len
);
1558 * relay_add_stream: allocate a new stream for a session
1560 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1561 struct relay_connection
*conn
,
1562 const struct lttng_buffer_view
*payload
)
1566 struct relay_session
*session
= conn
->session
;
1567 struct relay_stream
*stream
= NULL
;
1568 struct lttcomm_relayd_status_stream reply
;
1569 struct ctf_trace
*trace
= NULL
;
1570 uint64_t stream_handle
= -1ULL;
1571 char *path_name
= NULL
, *channel_name
= NULL
;
1572 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1573 LTTNG_OPTIONAL(uint64_t) stream_chunk_id
= {};
1575 if (!session
|| !conn
->version_check_done
) {
1576 ERR("Trying to add a stream before version check");
1578 goto end_no_session
;
1581 if (session
->minor
== 1) {
1583 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1585 } else if (session
->minor
> 1 && session
->minor
< 11) {
1586 /* From 2.2 to 2.10 */
1587 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1588 &channel_name
, &tracefile_size
, &tracefile_count
);
1590 /* From 2.11 to ... */
1591 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1592 &channel_name
, &tracefile_size
, &tracefile_count
,
1593 &stream_chunk_id
.value
);
1594 stream_chunk_id
.is_set
= true;
1601 if (conform_channel_path(path_name
)) {
1606 * Backward compatibility for --group-output-by-session.
1607 * Prior to lttng 2.11, the complete path is passed by the stream.
1608 * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1609 * >=2.11 the chunk is responsible for the output path. When dealing
1610 * with producer < 2.11 the chunk output_path is the root output path
1611 * and the stream carries the complete path (path_name).
1612 * To support --group-output-by-session with older producer (<2.11), we
1613 * need to craft the path based on the stream path.
1615 if (opt_group_output_by
== RELAYD_GROUP_OUTPUT_BY_SESSION
) {
1616 if (conn
->minor
< 4) {
1618 * From 2.1 to 2.3, the session_name is not passed on
1619 * the RELAYD_CREATE_SESSION command. The session name
1620 * is necessary to detect the presence of a base_path
1621 * inside the stream path. Without it we cannot perform
1622 * a valid group-output-by-session transformation.
1624 WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1625 " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1626 session
->id
, path_name
);
1627 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1628 char *group_by_session_path_name
;
1630 assert(session
->session_name
[0] != '\0');
1632 group_by_session_path_name
=
1633 backward_compat_group_by_session(
1635 session
->session_name
);
1636 if (!group_by_session_path_name
) {
1637 ERR("Failed to apply group by session to stream of session %" PRIu64
,
1642 DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1643 path_name
, group_by_session_path_name
);
1646 path_name
= group_by_session_path_name
;
1650 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1655 /* This stream here has one reference on the trace. */
1656 pthread_mutex_lock(&last_relay_stream_id_lock
);
1657 stream_handle
= ++last_relay_stream_id
;
1658 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1660 /* We pass ownership of path_name and channel_name. */
1661 stream
= stream_create(trace
, stream_handle
, path_name
,
1662 channel_name
, tracefile_size
, tracefile_count
);
1664 channel_name
= NULL
;
1667 * Streams are the owners of their trace. Reference to trace is
1668 * kept within stream_create().
1670 ctf_trace_put(trace
);
1673 memset(&reply
, 0, sizeof(reply
));
1674 reply
.handle
= htobe64(stream_handle
);
1676 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1678 reply
.ret_code
= htobe32(LTTNG_OK
);
1681 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1682 sizeof(struct lttcomm_relayd_status_stream
), 0);
1683 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1684 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1696 * relay_close_stream: close a specific stream
1698 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1699 struct relay_connection
*conn
,
1700 const struct lttng_buffer_view
*payload
)
1704 struct relay_session
*session
= conn
->session
;
1705 struct lttcomm_relayd_close_stream stream_info
;
1706 struct lttcomm_relayd_generic_reply reply
;
1707 struct relay_stream
*stream
;
1709 DBG("Close stream received");
1711 if (!session
|| !conn
->version_check_done
) {
1712 ERR("Trying to close a stream before version check");
1714 goto end_no_session
;
1717 if (payload
->size
< sizeof(stream_info
)) {
1718 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1719 sizeof(stream_info
), payload
->size
);
1721 goto end_no_session
;
1723 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1724 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1725 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1727 stream
= stream_get_by_id(stream_info
.stream_id
);
1734 * Set last_net_seq_num before the close flag. Required by data
1737 pthread_mutex_lock(&stream
->lock
);
1738 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1739 pthread_mutex_unlock(&stream
->lock
);
1742 * This is one of the conditions which may trigger a stream close
1743 * with the others being:
1744 * 1) A close command is received for a stream
1745 * 2) The control connection owning the stream is closed
1746 * 3) We have received all of the stream's data _after_ a close
1749 try_stream_close(stream
);
1750 if (stream
->is_metadata
) {
1751 struct relay_viewer_stream
*vstream
;
1753 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1755 if (stream
->no_new_metadata_notified
) {
1757 * Since all the metadata has been sent to the
1758 * viewer and that we have a request to close
1759 * its stream, we can safely teardown the
1760 * corresponding metadata viewer stream.
1762 viewer_stream_put(vstream
);
1764 /* Put local reference. */
1765 viewer_stream_put(vstream
);
1772 memset(&reply
, 0, sizeof(reply
));
1774 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1776 reply
.ret_code
= htobe32(LTTNG_OK
);
1778 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1779 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1780 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1781 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1791 * relay_reset_metadata: reset a metadata stream
1794 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
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(const struct lttcomm_relayd_hdr
*recv_hdr
,
1894 struct relay_connection
*conn
,
1895 const struct lttng_buffer_view
*payload
)
1899 struct lttcomm_relayd_generic_reply reply
;
1900 struct relay_session
*session
= conn
->session
;
1903 DBG("Trying to start the streaming without a session established");
1904 ret
= htobe32(LTTNG_ERR_UNK
);
1907 memset(&reply
, 0, sizeof(reply
));
1908 reply
.ret_code
= htobe32(LTTNG_OK
);
1909 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1911 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1912 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1921 * relay_recv_metadata: receive the metadata for the session.
1923 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1924 struct relay_connection
*conn
,
1925 const struct lttng_buffer_view
*payload
)
1928 struct relay_session
*session
= conn
->session
;
1929 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1930 struct relay_stream
*metadata_stream
;
1931 uint64_t metadata_payload_size
;
1932 struct lttng_buffer_view packet_view
;
1935 ERR("Metadata sent before version check");
1940 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1941 ERR("Incorrect data size");
1945 metadata_payload_size
= recv_hdr
->data_size
-
1946 sizeof(struct lttcomm_relayd_metadata_payload
);
1948 memcpy(&metadata_payload_header
, payload
->data
,
1949 sizeof(metadata_payload_header
));
1950 metadata_payload_header
.stream_id
= be64toh(
1951 metadata_payload_header
.stream_id
);
1952 metadata_payload_header
.padding_size
= be32toh(
1953 metadata_payload_header
.padding_size
);
1955 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1956 if (!metadata_stream
) {
1961 packet_view
= lttng_buffer_view_from_view(payload
,
1962 sizeof(metadata_payload_header
), metadata_payload_size
);
1963 if (!packet_view
.data
) {
1964 ERR("Invalid metadata packet length announced by header");
1969 pthread_mutex_lock(&metadata_stream
->lock
);
1970 ret
= stream_write(metadata_stream
, &packet_view
,
1971 metadata_payload_header
.padding_size
);
1972 pthread_mutex_unlock(&metadata_stream
->lock
);
1978 stream_put(metadata_stream
);
1984 * relay_send_version: send relayd version number
1986 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1987 struct relay_connection
*conn
,
1988 const struct lttng_buffer_view
*payload
)
1992 struct lttcomm_relayd_version reply
, msg
;
1993 bool compatible
= true;
1995 conn
->version_check_done
= true;
1997 /* Get version from the other side. */
1998 if (payload
->size
< sizeof(msg
)) {
1999 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
2000 sizeof(msg
), payload
->size
);
2005 memcpy(&msg
, payload
->data
, sizeof(msg
));
2006 msg
.major
= be32toh(msg
.major
);
2007 msg
.minor
= be32toh(msg
.minor
);
2009 memset(&reply
, 0, sizeof(reply
));
2010 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
2011 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
2013 /* Major versions must be the same */
2014 if (reply
.major
!= msg
.major
) {
2015 DBG("Incompatible major versions (%u vs %u), deleting session",
2016 reply
.major
, msg
.major
);
2020 conn
->major
= reply
.major
;
2021 /* We adapt to the lowest compatible version */
2022 if (reply
.minor
<= msg
.minor
) {
2023 conn
->minor
= reply
.minor
;
2025 conn
->minor
= msg
.minor
;
2028 reply
.major
= htobe32(reply
.major
);
2029 reply
.minor
= htobe32(reply
.minor
);
2030 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2032 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2033 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2046 DBG("Version check done using protocol %u.%u", conn
->major
,
2054 * Check for data pending for a given stream id from the session daemon.
2056 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2057 struct relay_connection
*conn
,
2058 const struct lttng_buffer_view
*payload
)
2060 struct relay_session
*session
= conn
->session
;
2061 struct lttcomm_relayd_data_pending msg
;
2062 struct lttcomm_relayd_generic_reply reply
;
2063 struct relay_stream
*stream
;
2066 uint64_t stream_seq
;
2068 DBG("Data pending command received");
2070 if (!session
|| !conn
->version_check_done
) {
2071 ERR("Trying to check for data before version check");
2073 goto end_no_session
;
2076 if (payload
->size
< sizeof(msg
)) {
2077 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2078 sizeof(msg
), payload
->size
);
2080 goto end_no_session
;
2082 memcpy(&msg
, payload
->data
, sizeof(msg
));
2083 msg
.stream_id
= be64toh(msg
.stream_id
);
2084 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
2086 stream
= stream_get_by_id(msg
.stream_id
);
2087 if (stream
== NULL
) {
2092 pthread_mutex_lock(&stream
->lock
);
2094 if (session_streams_have_index(session
)) {
2096 * Ensure that both the index and stream data have been
2097 * flushed up to the requested point.
2099 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2101 stream_seq
= stream
->prev_data_seq
;
2103 DBG("Data pending for stream id %" PRIu64
": prev_data_seq %" PRIu64
2104 ", prev_index_seq %" PRIu64
2105 ", and last_seq %" PRIu64
, msg
.stream_id
,
2106 stream
->prev_data_seq
, stream
->prev_index_seq
,
2107 msg
.last_net_seq_num
);
2109 /* Avoid wrapping issue */
2110 if (((int64_t) (stream_seq
- msg
.last_net_seq_num
)) >= 0) {
2111 /* Data has in fact been written and is NOT pending */
2114 /* Data still being streamed thus pending */
2118 stream
->data_pending_check_done
= true;
2119 pthread_mutex_unlock(&stream
->lock
);
2124 memset(&reply
, 0, sizeof(reply
));
2125 reply
.ret_code
= htobe32(ret
);
2126 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2127 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2128 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2138 * Wait for the control socket to reach a quiescent state.
2140 * Note that for now, when receiving this command from the session
2141 * daemon, this means that every subsequent commands or data received on
2142 * the control socket has been handled. So, this is why we simply return
2145 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
2146 struct relay_connection
*conn
,
2147 const struct lttng_buffer_view
*payload
)
2151 struct relay_stream
*stream
;
2152 struct lttcomm_relayd_quiescent_control msg
;
2153 struct lttcomm_relayd_generic_reply reply
;
2155 DBG("Checking quiescent state on control socket");
2157 if (!conn
->session
|| !conn
->version_check_done
) {
2158 ERR("Trying to check for data before version check");
2160 goto end_no_session
;
2163 if (payload
->size
< sizeof(msg
)) {
2164 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2165 sizeof(msg
), payload
->size
);
2167 goto end_no_session
;
2169 memcpy(&msg
, payload
->data
, sizeof(msg
));
2170 msg
.stream_id
= be64toh(msg
.stream_id
);
2172 stream
= stream_get_by_id(msg
.stream_id
);
2176 pthread_mutex_lock(&stream
->lock
);
2177 stream
->data_pending_check_done
= true;
2178 pthread_mutex_unlock(&stream
->lock
);
2180 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
2183 memset(&reply
, 0, sizeof(reply
));
2184 reply
.ret_code
= htobe32(LTTNG_OK
);
2185 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2186 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2187 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2199 * Initialize a data pending command. This means that a consumer is about
2200 * to ask for data pending for each stream it holds. Simply iterate over
2201 * all streams of a session and set the data_pending_check_done flag.
2203 * This command returns to the client a LTTNG_OK code.
2205 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2206 struct relay_connection
*conn
,
2207 const struct lttng_buffer_view
*payload
)
2211 struct lttng_ht_iter iter
;
2212 struct lttcomm_relayd_begin_data_pending msg
;
2213 struct lttcomm_relayd_generic_reply reply
;
2214 struct relay_stream
*stream
;
2219 DBG("Init streams for data pending");
2221 if (!conn
->session
|| !conn
->version_check_done
) {
2222 ERR("Trying to check for data before version check");
2224 goto end_no_session
;
2227 if (payload
->size
< sizeof(msg
)) {
2228 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2229 sizeof(msg
), payload
->size
);
2231 goto end_no_session
;
2233 memcpy(&msg
, payload
->data
, sizeof(msg
));
2234 msg
.session_id
= be64toh(msg
.session_id
);
2237 * Iterate over all streams to set the begin data pending flag.
2238 * For now, the streams are indexed by stream handle so we have
2239 * to iterate over all streams to find the one associated with
2240 * the right session_id.
2243 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2245 if (!stream_get(stream
)) {
2248 if (stream
->trace
->session
->id
== msg
.session_id
) {
2249 pthread_mutex_lock(&stream
->lock
);
2250 stream
->data_pending_check_done
= false;
2251 pthread_mutex_unlock(&stream
->lock
);
2252 DBG("Set begin data pending flag to stream %" PRIu64
,
2253 stream
->stream_handle
);
2259 memset(&reply
, 0, sizeof(reply
));
2260 /* All good, send back reply. */
2261 reply
.ret_code
= htobe32(LTTNG_OK
);
2263 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2264 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2265 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2277 * End data pending command. This will check, for a given session id, if
2278 * each stream associated with it has its data_pending_check_done flag
2279 * set. If not, this means that the client lost track of the stream but
2280 * the data is still being streamed on our side. In this case, we inform
2281 * the client that data is in flight.
2283 * Return to the client if there is data in flight or not with a ret_code.
2285 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2286 struct relay_connection
*conn
,
2287 const struct lttng_buffer_view
*payload
)
2291 struct lttng_ht_iter iter
;
2292 struct lttcomm_relayd_end_data_pending msg
;
2293 struct lttcomm_relayd_generic_reply reply
;
2294 struct relay_stream
*stream
;
2295 uint32_t is_data_inflight
= 0;
2297 DBG("End data pending command");
2299 if (!conn
->session
|| !conn
->version_check_done
) {
2300 ERR("Trying to check for data before version check");
2302 goto end_no_session
;
2305 if (payload
->size
< sizeof(msg
)) {
2306 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2307 sizeof(msg
), payload
->size
);
2309 goto end_no_session
;
2311 memcpy(&msg
, payload
->data
, sizeof(msg
));
2312 msg
.session_id
= be64toh(msg
.session_id
);
2315 * Iterate over all streams to see if the begin data pending
2319 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2321 if (!stream_get(stream
)) {
2324 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2328 pthread_mutex_lock(&stream
->lock
);
2329 if (!stream
->data_pending_check_done
) {
2330 uint64_t stream_seq
;
2332 if (session_streams_have_index(conn
->session
)) {
2334 * Ensure that both the index and stream data have been
2335 * flushed up to the requested point.
2337 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2339 stream_seq
= stream
->prev_data_seq
;
2341 if (!stream
->closed
|| !(((int64_t) (stream_seq
- stream
->last_net_seq_num
)) >= 0)) {
2342 is_data_inflight
= 1;
2343 DBG("Data is still in flight for stream %" PRIu64
,
2344 stream
->stream_handle
);
2345 pthread_mutex_unlock(&stream
->lock
);
2350 pthread_mutex_unlock(&stream
->lock
);
2355 memset(&reply
, 0, sizeof(reply
));
2356 /* All good, send back reply. */
2357 reply
.ret_code
= htobe32(is_data_inflight
);
2359 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2360 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2361 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2373 * Receive an index for a specific stream.
2375 * Return 0 on success else a negative value.
2377 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2378 struct relay_connection
*conn
,
2379 const struct lttng_buffer_view
*payload
)
2383 struct relay_session
*session
= conn
->session
;
2384 struct lttcomm_relayd_index index_info
;
2385 struct lttcomm_relayd_generic_reply reply
;
2386 struct relay_stream
*stream
;
2391 DBG("Relay receiving index");
2393 if (!session
|| !conn
->version_check_done
) {
2394 ERR("Trying to close a stream before version check");
2396 goto end_no_session
;
2399 msg_len
= lttcomm_relayd_index_len(
2400 lttng_to_index_major(conn
->major
, conn
->minor
),
2401 lttng_to_index_minor(conn
->major
, conn
->minor
));
2402 if (payload
->size
< msg_len
) {
2403 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2404 msg_len
, payload
->size
);
2406 goto end_no_session
;
2408 memcpy(&index_info
, payload
->data
, msg_len
);
2409 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2410 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2411 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2412 index_info
.content_size
= be64toh(index_info
.content_size
);
2413 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2414 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2415 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2416 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2418 if (conn
->minor
>= 8) {
2419 index_info
.stream_instance_id
=
2420 be64toh(index_info
.stream_instance_id
);
2421 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2423 index_info
.stream_instance_id
= -1ULL;
2424 index_info
.packet_seq_num
= -1ULL;
2427 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2429 ERR("stream_get_by_id not found");
2434 pthread_mutex_lock(&stream
->lock
);
2435 ret
= stream_add_index(stream
, &index_info
);
2436 pthread_mutex_unlock(&stream
->lock
);
2438 goto end_stream_put
;
2444 memset(&reply
, 0, sizeof(reply
));
2446 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2448 reply
.ret_code
= htobe32(LTTNG_OK
);
2450 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2451 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2452 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2461 * Receive the streams_sent message.
2463 * Return 0 on success else a negative value.
2465 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2466 struct relay_connection
*conn
,
2467 const struct lttng_buffer_view
*payload
)
2471 struct lttcomm_relayd_generic_reply reply
;
2475 DBG("Relay receiving streams_sent");
2477 if (!conn
->session
|| !conn
->version_check_done
) {
2478 ERR("Trying to close a stream before version check");
2480 goto end_no_session
;
2484 * Publish every pending stream in the connection recv list which are
2485 * now ready to be used by the viewer.
2487 publish_connection_local_streams(conn
);
2489 memset(&reply
, 0, sizeof(reply
));
2490 reply
.ret_code
= htobe32(LTTNG_OK
);
2491 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2492 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2493 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2506 * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2507 * session rotation feature (not the tracefile rotation feature).
2509 static int relay_rotate_session_streams(
2510 const struct lttcomm_relayd_hdr
*recv_hdr
,
2511 struct relay_connection
*conn
,
2512 const struct lttng_buffer_view
*payload
)
2517 enum lttng_error_code reply_code
= LTTNG_ERR_UNK
;
2518 struct relay_session
*session
= conn
->session
;
2519 struct lttcomm_relayd_rotate_streams rotate_streams
;
2520 struct lttcomm_relayd_generic_reply reply
= {};
2521 struct relay_stream
*stream
= NULL
;
2522 const size_t header_len
= sizeof(struct lttcomm_relayd_rotate_streams
);
2523 struct lttng_trace_chunk
*next_trace_chunk
= NULL
;
2524 struct lttng_buffer_view stream_positions
;
2525 char chunk_id_buf
[MAX_INT_DEC_LEN(uint64_t)];
2526 const char *chunk_id_str
= "none";
2528 if (!session
|| !conn
->version_check_done
) {
2529 ERR("Trying to rotate a stream before version check");
2534 if (session
->major
== 2 && session
->minor
< 11) {
2535 ERR("Unsupported feature before 2.11");
2540 if (payload
->size
< header_len
) {
2541 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2542 header_len
, payload
->size
);
2547 memcpy(&rotate_streams
, payload
->data
, header_len
);
2549 /* Convert header to host endianness. */
2550 rotate_streams
= (typeof(rotate_streams
)) {
2551 .stream_count
= be32toh(rotate_streams
.stream_count
),
2552 .new_chunk_id
= (typeof(rotate_streams
.new_chunk_id
)) {
2553 .is_set
= !!rotate_streams
.new_chunk_id
.is_set
,
2554 .value
= be64toh(rotate_streams
.new_chunk_id
.value
),
2558 if (rotate_streams
.new_chunk_id
.is_set
) {
2560 * Retrieve the trace chunk the stream must transition to. As
2561 * per the protocol, this chunk should have been created
2562 * before this command is received.
2564 next_trace_chunk
= sessiond_trace_chunk_registry_get_chunk(
2565 sessiond_trace_chunk_registry
,
2566 session
->sessiond_uuid
, session
->id
,
2567 rotate_streams
.new_chunk_id
.value
);
2568 if (!next_trace_chunk
) {
2569 char uuid_str
[LTTNG_UUID_STR_LEN
];
2571 lttng_uuid_to_str(session
->sessiond_uuid
, uuid_str
);
2572 ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2573 ", trace_chunk_id = %" PRIu64
,
2574 uuid_str
, session
->id
,
2575 rotate_streams
.new_chunk_id
.value
);
2576 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2581 ret
= snprintf(chunk_id_buf
, sizeof(chunk_id_buf
), "%" PRIu64
,
2582 rotate_streams
.new_chunk_id
.value
);
2583 if (ret
< 0 || ret
>= sizeof(chunk_id_buf
)) {
2584 chunk_id_str
= "formatting error";
2586 chunk_id_str
= chunk_id_buf
;
2590 DBG("Rotate %" PRIu32
" streams of session \"%s\" to chunk \"%s\"",
2591 rotate_streams
.stream_count
, session
->session_name
,
2594 stream_positions
= lttng_buffer_view_from_view(payload
,
2595 sizeof(rotate_streams
), -1);
2596 if (!stream_positions
.data
||
2597 stream_positions
.size
<
2598 (rotate_streams
.stream_count
*
2599 sizeof(struct lttcomm_relayd_stream_rotation_position
))) {
2600 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2605 for (i
= 0; i
< rotate_streams
.stream_count
; i
++) {
2606 struct lttcomm_relayd_stream_rotation_position
*position_comm
=
2607 &((typeof(position_comm
)) stream_positions
.data
)[i
];
2608 const struct lttcomm_relayd_stream_rotation_position pos
= {
2609 .stream_id
= be64toh(position_comm
->stream_id
),
2610 .rotate_at_seq_num
= be64toh(
2611 position_comm
->rotate_at_seq_num
),
2614 stream
= stream_get_by_id(pos
.stream_id
);
2616 reply_code
= LTTNG_ERR_INVALID
;
2621 pthread_mutex_lock(&stream
->lock
);
2622 ret
= stream_set_pending_rotation(stream
, next_trace_chunk
,
2623 pos
.rotate_at_seq_num
);
2624 pthread_mutex_unlock(&stream
->lock
);
2626 reply_code
= LTTNG_ERR_FILE_CREATION_ERROR
;
2634 reply_code
= LTTNG_OK
;
2641 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2642 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2643 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2644 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2645 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2650 lttng_trace_chunk_put(next_trace_chunk
);
2657 * relay_create_trace_chunk: create a new trace chunk
2659 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2660 struct relay_connection
*conn
,
2661 const struct lttng_buffer_view
*payload
)
2665 struct relay_session
*session
= conn
->session
;
2666 struct lttcomm_relayd_create_trace_chunk
*msg
;
2667 struct lttcomm_relayd_generic_reply reply
= {};
2668 struct lttng_buffer_view header_view
;
2669 struct lttng_buffer_view chunk_name_view
;
2670 struct lttng_trace_chunk
*chunk
= NULL
, *published_chunk
= NULL
;
2671 enum lttng_error_code reply_code
= LTTNG_OK
;
2672 enum lttng_trace_chunk_status chunk_status
;
2673 const char *new_path
;
2675 if (!session
|| !conn
->version_check_done
) {
2676 ERR("Trying to create a trace chunk before version check");
2681 if (session
->major
== 2 && session
->minor
< 11) {
2682 ERR("Chunk creation command is unsupported before 2.11");
2687 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2688 if (!header_view
.data
) {
2689 ERR("Failed to receive payload of chunk creation command");
2694 /* Convert to host endianness. */
2695 msg
= (typeof(msg
)) header_view
.data
;
2696 msg
->chunk_id
= be64toh(msg
->chunk_id
);
2697 msg
->creation_timestamp
= be64toh(msg
->creation_timestamp
);
2698 msg
->override_name_length
= be32toh(msg
->override_name_length
);
2700 if (session
->current_trace_chunk
&&
2701 !lttng_trace_chunk_get_name_overridden(session
->current_trace_chunk
)) {
2702 chunk_status
= lttng_trace_chunk_rename_path(session
->current_trace_chunk
,
2703 DEFAULT_CHUNK_TMP_OLD_DIRECTORY
);
2704 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2705 ERR("Failed to rename old chunk");
2707 reply_code
= LTTNG_ERR_UNK
;
2711 session
->ongoing_rotation
= true;
2712 if (!session
->current_trace_chunk
) {
2713 if (!session
->has_rotated
) {
2719 new_path
= DEFAULT_CHUNK_TMP_NEW_DIRECTORY
;
2721 chunk
= lttng_trace_chunk_create(
2722 msg
->chunk_id
, msg
->creation_timestamp
, new_path
);
2724 ERR("Failed to create trace chunk in trace chunk creation command");
2726 reply_code
= LTTNG_ERR_NOMEM
;
2729 lttng_trace_chunk_set_fd_tracker(chunk
, the_fd_tracker
);
2731 if (msg
->override_name_length
) {
2734 chunk_name_view
= lttng_buffer_view_from_view(payload
,
2736 msg
->override_name_length
);
2737 name
= chunk_name_view
.data
;
2738 if (!name
|| name
[msg
->override_name_length
- 1]) {
2739 ERR("Failed to receive payload of chunk creation command");
2741 reply_code
= LTTNG_ERR_INVALID
;
2745 chunk_status
= lttng_trace_chunk_override_name(
2746 chunk
, chunk_name_view
.data
);
2747 switch (chunk_status
) {
2748 case LTTNG_TRACE_CHUNK_STATUS_OK
:
2750 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT
:
2751 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2752 reply_code
= LTTNG_ERR_INVALID
;
2756 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2757 reply_code
= LTTNG_ERR_UNK
;
2763 chunk_status
= lttng_trace_chunk_set_credentials_current_user(chunk
);
2764 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2765 reply_code
= LTTNG_ERR_UNK
;
2770 assert(conn
->session
->output_directory
);
2771 chunk_status
= lttng_trace_chunk_set_as_owner(chunk
,
2772 conn
->session
->output_directory
);
2773 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2774 reply_code
= LTTNG_ERR_UNK
;
2779 published_chunk
= sessiond_trace_chunk_registry_publish_chunk(
2780 sessiond_trace_chunk_registry
,
2781 conn
->session
->sessiond_uuid
,
2784 if (!published_chunk
) {
2785 char uuid_str
[LTTNG_UUID_STR_LEN
];
2787 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2788 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2793 reply_code
= LTTNG_ERR_NOMEM
;
2797 pthread_mutex_lock(&conn
->session
->lock
);
2798 if (conn
->session
->pending_closure_trace_chunk
) {
2800 * Invalid; this means a second create_trace_chunk command was
2801 * received before a close_trace_chunk.
2803 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2804 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2806 goto end_unlock_session
;
2808 conn
->session
->pending_closure_trace_chunk
=
2809 conn
->session
->current_trace_chunk
;
2810 conn
->session
->current_trace_chunk
= published_chunk
;
2811 published_chunk
= NULL
;
2812 if (!conn
->session
->pending_closure_trace_chunk
) {
2813 session
->ongoing_rotation
= false;
2816 pthread_mutex_unlock(&conn
->session
->lock
);
2818 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2819 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2821 sizeof(struct lttcomm_relayd_generic_reply
),
2823 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2824 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2829 lttng_trace_chunk_put(chunk
);
2830 lttng_trace_chunk_put(published_chunk
);
2835 * relay_close_trace_chunk: close a trace chunk
2837 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2838 struct relay_connection
*conn
,
2839 const struct lttng_buffer_view
*payload
)
2841 int ret
= 0, buf_ret
;
2843 struct relay_session
*session
= conn
->session
;
2844 struct lttcomm_relayd_close_trace_chunk
*msg
;
2845 struct lttcomm_relayd_close_trace_chunk_reply reply
= {};
2846 struct lttng_buffer_view header_view
;
2847 struct lttng_trace_chunk
*chunk
= NULL
;
2848 enum lttng_error_code reply_code
= LTTNG_OK
;
2849 enum lttng_trace_chunk_status chunk_status
;
2851 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type
) close_command
= {};
2852 time_t close_timestamp
;
2853 char closed_trace_chunk_path
[LTTNG_PATH_MAX
];
2854 size_t path_length
= 0;
2855 const char *chunk_name
= NULL
;
2856 struct lttng_dynamic_buffer reply_payload
;
2857 const char *new_path
;
2859 lttng_dynamic_buffer_init(&reply_payload
);
2861 if (!session
|| !conn
->version_check_done
) {
2862 ERR("Trying to close a trace chunk before version check");
2867 if (session
->major
== 2 && session
->minor
< 11) {
2868 ERR("Chunk close command is unsupported before 2.11");
2873 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2874 if (!header_view
.data
) {
2875 ERR("Failed to receive payload of chunk close command");
2880 /* Convert to host endianness. */
2881 msg
= (typeof(msg
)) header_view
.data
;
2882 chunk_id
= be64toh(msg
->chunk_id
);
2883 close_timestamp
= (time_t) be64toh(msg
->close_timestamp
);
2884 close_command
= (typeof(close_command
)){
2885 .value
= be32toh(msg
->close_command
.value
),
2886 .is_set
= msg
->close_command
.is_set
,
2889 chunk
= sessiond_trace_chunk_registry_get_chunk(
2890 sessiond_trace_chunk_registry
,
2891 conn
->session
->sessiond_uuid
,
2895 char uuid_str
[LTTNG_UUID_STR_LEN
];
2897 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2898 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2903 reply_code
= LTTNG_ERR_NOMEM
;
2907 pthread_mutex_lock(&session
->lock
);
2908 if (close_command
.is_set
&&
2909 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE
) {
2911 * Clear command. It is a protocol error to ask for a
2912 * clear on a relay which does not allow it. Querying
2913 * the configuration allows figuring out whether
2914 * clearing is allowed before doing the clear.
2916 if (!opt_allow_clear
) {
2918 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2919 goto end_unlock_session
;
2922 if (session
->pending_closure_trace_chunk
&&
2923 session
->pending_closure_trace_chunk
!= chunk
) {
2924 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
2925 session
->session_name
);
2926 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2928 goto end_unlock_session
;
2931 if (session
->current_trace_chunk
&& session
->current_trace_chunk
!= chunk
&&
2932 !lttng_trace_chunk_get_name_overridden(session
->current_trace_chunk
)) {
2933 if (close_command
.is_set
&&
2934 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE
&&
2935 !session
->has_rotated
) {
2936 /* New chunk stays in session output directory. */
2939 /* Use chunk name for new chunk. */
2942 /* Rename new chunk path. */
2943 chunk_status
= lttng_trace_chunk_rename_path(session
->current_trace_chunk
,
2945 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2947 goto end_unlock_session
;
2949 session
->ongoing_rotation
= false;
2951 if ((!close_command
.is_set
||
2952 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION
) &&
2953 !lttng_trace_chunk_get_name_overridden(chunk
)) {
2954 const char *old_path
;
2956 if (!session
->has_rotated
) {
2961 /* We need to move back the .tmp_old_chunk to its rightful place. */
2962 chunk_status
= lttng_trace_chunk_rename_path(chunk
, old_path
);
2963 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2965 goto end_unlock_session
;
2968 chunk_status
= lttng_trace_chunk_set_close_timestamp(
2969 chunk
, close_timestamp
);
2970 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2971 ERR("Failed to set trace chunk close timestamp");
2973 reply_code
= LTTNG_ERR_UNK
;
2974 goto end_unlock_session
;
2977 if (close_command
.is_set
) {
2978 chunk_status
= lttng_trace_chunk_set_close_command(
2979 chunk
, close_command
.value
);
2980 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2982 reply_code
= LTTNG_ERR_INVALID
;
2983 goto end_unlock_session
;
2986 chunk_status
= lttng_trace_chunk_get_name(chunk
, &chunk_name
, NULL
);
2987 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2988 ERR("Failed to get chunk name");
2990 reply_code
= LTTNG_ERR_UNK
;
2991 goto end_unlock_session
;
2993 if (!session
->has_rotated
&& !session
->snapshot
) {
2994 ret
= lttng_strncpy(closed_trace_chunk_path
,
2995 session
->output_path
,
2996 sizeof(closed_trace_chunk_path
));
2998 ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
2999 strlen(session
->output_path
),
3000 sizeof(closed_trace_chunk_path
));
3001 reply_code
= LTTNG_ERR_NOMEM
;
3003 goto end_unlock_session
;
3006 if (session
->snapshot
) {
3007 ret
= snprintf(closed_trace_chunk_path
,
3008 sizeof(closed_trace_chunk_path
),
3009 "%s/%s", session
->output_path
,
3012 ret
= snprintf(closed_trace_chunk_path
,
3013 sizeof(closed_trace_chunk_path
),
3014 "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY
3016 session
->output_path
, chunk_name
);
3018 if (ret
< 0 || ret
== sizeof(closed_trace_chunk_path
)) {
3019 ERR("Failed to format closed trace chunk resulting path");
3020 reply_code
= ret
< 0 ? LTTNG_ERR_UNK
: LTTNG_ERR_NOMEM
;
3022 goto end_unlock_session
;
3025 if (close_command
.is_set
&&
3026 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED
) {
3027 session
->has_rotated
= true;
3029 DBG("Reply chunk path on close: %s", closed_trace_chunk_path
);
3030 path_length
= strlen(closed_trace_chunk_path
) + 1;
3031 if (path_length
> UINT32_MAX
) {
3032 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3034 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
3035 goto end_unlock_session
;
3038 if (session
->current_trace_chunk
== chunk
) {
3040 * After a trace chunk close command, no new streams
3041 * referencing the chunk may be created. Hence, on the
3042 * event that no new trace chunk have been created for
3043 * the session, the reference to the current trace chunk
3044 * is released in order to allow it to be reclaimed when
3045 * the last stream releases its reference to it.
3047 lttng_trace_chunk_put(session
->current_trace_chunk
);
3048 session
->current_trace_chunk
= NULL
;
3050 lttng_trace_chunk_put(session
->pending_closure_trace_chunk
);
3051 session
->pending_closure_trace_chunk
= NULL
;
3053 pthread_mutex_unlock(&session
->lock
);
3056 reply
.generic
.ret_code
= htobe32((uint32_t) reply_code
);
3057 reply
.path_length
= htobe32((uint32_t) path_length
);
3058 buf_ret
= lttng_dynamic_buffer_append(
3059 &reply_payload
, &reply
, sizeof(reply
));
3061 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3065 if (reply_code
== LTTNG_OK
) {
3066 buf_ret
= lttng_dynamic_buffer_append(&reply_payload
,
3067 closed_trace_chunk_path
, path_length
);
3069 ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3074 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
3078 if (send_ret
< reply_payload
.size
) {
3079 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3080 reply_payload
.size
, send_ret
);
3085 lttng_trace_chunk_put(chunk
);
3086 lttng_dynamic_buffer_reset(&reply_payload
);
3091 * relay_trace_chunk_exists: check if a trace chunk exists
3093 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr
*recv_hdr
,
3094 struct relay_connection
*conn
,
3095 const struct lttng_buffer_view
*payload
)
3099 struct relay_session
*session
= conn
->session
;
3100 struct lttcomm_relayd_trace_chunk_exists
*msg
;
3101 struct lttcomm_relayd_trace_chunk_exists_reply reply
= {};
3102 struct lttng_buffer_view header_view
;
3106 if (!session
|| !conn
->version_check_done
) {
3107 ERR("Trying to close a trace chunk before version check");
3112 if (session
->major
== 2 && session
->minor
< 11) {
3113 ERR("Chunk close command is unsupported before 2.11");
3118 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
3119 if (!header_view
.data
) {
3120 ERR("Failed to receive payload of chunk close command");
3125 /* Convert to host endianness. */
3126 msg
= (typeof(msg
)) header_view
.data
;
3127 chunk_id
= be64toh(msg
->chunk_id
);
3129 ret
= sessiond_trace_chunk_registry_chunk_exists(
3130 sessiond_trace_chunk_registry
,
3131 conn
->session
->sessiond_uuid
,
3133 chunk_id
, &chunk_exists
);
3135 * If ret is not 0, send the reply and report the error to the caller.
3136 * It is a protocol (or internal) error and the session/connection
3137 * should be torn down.
3139 reply
= (typeof(reply
)){
3140 .generic
.ret_code
= htobe32((uint32_t)
3141 (ret
== 0 ? LTTNG_OK
: LTTNG_ERR_INVALID_PROTOCOL
)),
3142 .trace_chunk_exists
= ret
== 0 ? chunk_exists
: 0,
3144 send_ret
= conn
->sock
->ops
->sendmsg(
3145 conn
->sock
, &reply
, sizeof(reply
), 0);
3146 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3147 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
3156 * relay_get_configuration: query whether feature is available
3158 static int relay_get_configuration(const struct lttcomm_relayd_hdr
*recv_hdr
,
3159 struct relay_connection
*conn
,
3160 const struct lttng_buffer_view
*payload
)
3164 struct lttcomm_relayd_get_configuration
*msg
;
3165 struct lttcomm_relayd_get_configuration_reply reply
= {};
3166 struct lttng_buffer_view header_view
;
3167 uint64_t query_flags
= 0;
3168 uint64_t result_flags
= 0;
3170 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
3171 if (!header_view
.data
) {
3172 ERR("Failed to receive payload of chunk close command");
3177 /* Convert to host endianness. */
3178 msg
= (typeof(msg
)) header_view
.data
;
3179 query_flags
= be64toh(msg
->query_flags
);
3182 ret
= LTTNG_ERR_INVALID_PROTOCOL
;
3185 if (opt_allow_clear
) {
3186 result_flags
|= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED
;
3190 reply
= (typeof(reply
)){
3191 .generic
.ret_code
= htobe32((uint32_t)
3192 (ret
== 0 ? LTTNG_OK
: LTTNG_ERR_INVALID_PROTOCOL
)),
3193 .relayd_configuration_flags
= htobe64(result_flags
),
3195 send_ret
= conn
->sock
->ops
->sendmsg(
3196 conn
->sock
, &reply
, sizeof(reply
), 0);
3197 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3198 ERR("Failed to send \"get configuration\" command reply (ret = %zd)",
3206 #define DBG_CMD(cmd_name, conn) \
3207 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
3209 static int relay_process_control_command(struct relay_connection
*conn
,
3210 const struct lttcomm_relayd_hdr
*header
,
3211 const struct lttng_buffer_view
*payload
)
3215 switch (header
->cmd
) {
3216 case RELAYD_CREATE_SESSION
:
3217 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
3218 ret
= relay_create_session(header
, conn
, payload
);
3220 case RELAYD_ADD_STREAM
:
3221 DBG_CMD("RELAYD_ADD_STREAM", conn
);
3222 ret
= relay_add_stream(header
, conn
, payload
);
3224 case RELAYD_START_DATA
:
3225 DBG_CMD("RELAYD_START_DATA", conn
);
3226 ret
= relay_start(header
, conn
, payload
);
3228 case RELAYD_SEND_METADATA
:
3229 DBG_CMD("RELAYD_SEND_METADATA", conn
);
3230 ret
= relay_recv_metadata(header
, conn
, payload
);
3232 case RELAYD_VERSION
:
3233 DBG_CMD("RELAYD_VERSION", conn
);
3234 ret
= relay_send_version(header
, conn
, payload
);
3236 case RELAYD_CLOSE_STREAM
:
3237 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
3238 ret
= relay_close_stream(header
, conn
, payload
);
3240 case RELAYD_DATA_PENDING
:
3241 DBG_CMD("RELAYD_DATA_PENDING", conn
);
3242 ret
= relay_data_pending(header
, conn
, payload
);
3244 case RELAYD_QUIESCENT_CONTROL
:
3245 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
3246 ret
= relay_quiescent_control(header
, conn
, payload
);
3248 case RELAYD_BEGIN_DATA_PENDING
:
3249 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
3250 ret
= relay_begin_data_pending(header
, conn
, payload
);
3252 case RELAYD_END_DATA_PENDING
:
3253 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
3254 ret
= relay_end_data_pending(header
, conn
, payload
);
3256 case RELAYD_SEND_INDEX
:
3257 DBG_CMD("RELAYD_SEND_INDEX", conn
);
3258 ret
= relay_recv_index(header
, conn
, payload
);
3260 case RELAYD_STREAMS_SENT
:
3261 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
3262 ret
= relay_streams_sent(header
, conn
, payload
);
3264 case RELAYD_RESET_METADATA
:
3265 DBG_CMD("RELAYD_RESET_METADATA", conn
);
3266 ret
= relay_reset_metadata(header
, conn
, payload
);
3268 case RELAYD_ROTATE_STREAMS
:
3269 DBG_CMD("RELAYD_ROTATE_STREAMS", conn
);
3270 ret
= relay_rotate_session_streams(header
, conn
, payload
);
3272 case RELAYD_CREATE_TRACE_CHUNK
:
3273 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn
);
3274 ret
= relay_create_trace_chunk(header
, conn
, payload
);
3276 case RELAYD_CLOSE_TRACE_CHUNK
:
3277 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn
);
3278 ret
= relay_close_trace_chunk(header
, conn
, payload
);
3280 case RELAYD_TRACE_CHUNK_EXISTS
:
3281 DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn
);
3282 ret
= relay_trace_chunk_exists(header
, conn
, payload
);
3284 case RELAYD_GET_CONFIGURATION
:
3285 DBG_CMD("RELAYD_GET_CONFIGURATION", conn
);
3286 ret
= relay_get_configuration(header
, conn
, payload
);
3288 case RELAYD_UPDATE_SYNC_INFO
:
3290 ERR("Received unknown command (%u)", header
->cmd
);
3291 relay_unknown_command(conn
);
3300 static enum relay_connection_status
relay_process_control_receive_payload(
3301 struct relay_connection
*conn
)
3304 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3305 struct lttng_dynamic_buffer
*reception_buffer
=
3306 &conn
->protocol
.ctrl
.reception_buffer
;
3307 struct ctrl_connection_state_receive_payload
*state
=
3308 &conn
->protocol
.ctrl
.state
.receive_payload
;
3309 struct lttng_buffer_view payload_view
;
3311 if (state
->left_to_receive
== 0) {
3312 /* Short-circuit for payload-less commands. */
3313 goto reception_complete
;
3316 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3317 reception_buffer
->data
+ state
->received
,
3318 state
->left_to_receive
, MSG_DONTWAIT
);
3320 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3321 PERROR("Unable to receive command payload on sock %d",
3323 status
= RELAY_CONNECTION_STATUS_ERROR
;
3326 } else if (ret
== 0) {
3327 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3328 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3333 assert(ret
<= state
->left_to_receive
);
3335 state
->left_to_receive
-= ret
;
3336 state
->received
+= ret
;
3338 if (state
->left_to_receive
> 0) {
3340 * Can't transition to the protocol's next state, wait to
3341 * receive the rest of the header.
3343 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3344 state
->received
, state
->left_to_receive
,
3350 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
3351 conn
->sock
->fd
, state
->received
);
3353 * The payload required to process the command has been received.
3354 * A view to the reception buffer is forwarded to the various
3355 * commands and the state of the control is reset on success.
3357 * Commands are responsible for sending their reply to the peer.
3359 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
3361 ret
= relay_process_control_command(conn
,
3362 &state
->header
, &payload_view
);
3364 status
= RELAY_CONNECTION_STATUS_ERROR
;
3368 ret
= connection_reset_protocol_state(conn
);
3370 status
= RELAY_CONNECTION_STATUS_ERROR
;
3376 static enum relay_connection_status
relay_process_control_receive_header(
3377 struct relay_connection
*conn
)
3380 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3381 struct lttcomm_relayd_hdr header
;
3382 struct lttng_dynamic_buffer
*reception_buffer
=
3383 &conn
->protocol
.ctrl
.reception_buffer
;
3384 struct ctrl_connection_state_receive_header
*state
=
3385 &conn
->protocol
.ctrl
.state
.receive_header
;
3387 assert(state
->left_to_receive
!= 0);
3389 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3390 reception_buffer
->data
+ state
->received
,
3391 state
->left_to_receive
, MSG_DONTWAIT
);
3393 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3394 PERROR("Unable to receive control command header on sock %d",
3396 status
= RELAY_CONNECTION_STATUS_ERROR
;
3399 } else if (ret
== 0) {
3400 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3401 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3406 assert(ret
<= state
->left_to_receive
);
3408 state
->left_to_receive
-= ret
;
3409 state
->received
+= ret
;
3411 if (state
->left_to_receive
> 0) {
3413 * Can't transition to the protocol's next state, wait to
3414 * receive the rest of the header.
3416 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3417 state
->received
, state
->left_to_receive
,
3422 /* Transition to next state: receiving the command's payload. */
3423 conn
->protocol
.ctrl
.state_id
=
3424 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3425 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
3426 header
.circuit_id
= be64toh(header
.circuit_id
);
3427 header
.data_size
= be64toh(header
.data_size
);
3428 header
.cmd
= be32toh(header
.cmd
);
3429 header
.cmd_version
= be32toh(header
.cmd_version
);
3430 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
3431 &header
, sizeof(header
));
3433 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
3434 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
3437 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
3438 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
3440 status
= RELAY_CONNECTION_STATUS_ERROR
;
3444 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
3446 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
3447 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
3450 status
= RELAY_CONNECTION_STATUS_ERROR
;
3454 if (header
.data_size
== 0) {
3456 * Manually invoke the next state as the poll loop
3457 * will not wake-up to allow us to proceed further.
3459 status
= relay_process_control_receive_payload(conn
);
3466 * Process the commands received on the control socket
3468 static enum relay_connection_status
relay_process_control(
3469 struct relay_connection
*conn
)
3471 enum relay_connection_status status
;
3473 switch (conn
->protocol
.ctrl
.state_id
) {
3474 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
3475 status
= relay_process_control_receive_header(conn
);
3477 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3478 status
= relay_process_control_receive_payload(conn
);
3481 ERR("Unknown control connection protocol state encountered.");
3488 static enum relay_connection_status
relay_process_data_receive_header(
3489 struct relay_connection
*conn
)
3492 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3493 struct data_connection_state_receive_header
*state
=
3494 &conn
->protocol
.data
.state
.receive_header
;
3495 struct lttcomm_relayd_data_hdr header
;
3496 struct relay_stream
*stream
;
3498 assert(state
->left_to_receive
!= 0);
3500 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3501 state
->header_reception_buffer
+ state
->received
,
3502 state
->left_to_receive
, MSG_DONTWAIT
);
3504 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3505 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
3506 status
= RELAY_CONNECTION_STATUS_ERROR
;
3509 } else if (ret
== 0) {
3510 /* Orderly shutdown. Not necessary to print an error. */
3511 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3512 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3517 assert(ret
<= state
->left_to_receive
);
3519 state
->left_to_receive
-= ret
;
3520 state
->received
+= ret
;
3522 if (state
->left_to_receive
> 0) {
3524 * Can't transition to the protocol's next state, wait to
3525 * receive the rest of the header.
3527 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3528 state
->received
, state
->left_to_receive
,
3533 /* Transition to next state: receiving the payload. */
3534 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3536 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3537 header
.circuit_id
= be64toh(header
.circuit_id
);
3538 header
.stream_id
= be64toh(header
.stream_id
);
3539 header
.data_size
= be32toh(header
.data_size
);
3540 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3541 header
.padding_size
= be32toh(header
.padding_size
);
3542 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3544 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3546 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3547 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3549 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3550 conn
->sock
->fd
, header
.circuit_id
,
3551 header
.stream_id
, header
.data_size
,
3552 header
.net_seq_num
, header
.padding_size
);
3554 stream
= stream_get_by_id(header
.stream_id
);
3556 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3558 /* Protocol error. */
3559 status
= RELAY_CONNECTION_STATUS_ERROR
;
3563 pthread_mutex_lock(&stream
->lock
);
3564 /* Prepare stream for the reception of a new packet. */
3565 ret
= stream_init_packet(stream
, header
.data_size
,
3566 &conn
->protocol
.data
.state
.receive_payload
.rotate_index
);
3567 pthread_mutex_unlock(&stream
->lock
);
3569 ERR("Failed to rotate stream output file");
3570 status
= RELAY_CONNECTION_STATUS_ERROR
;
3571 goto end_stream_unlock
;
3580 static enum relay_connection_status
relay_process_data_receive_payload(
3581 struct relay_connection
*conn
)
3584 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3585 struct relay_stream
*stream
;
3586 struct data_connection_state_receive_payload
*state
=
3587 &conn
->protocol
.data
.state
.receive_payload
;
3588 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3589 char data_buffer
[chunk_size
];
3590 bool partial_recv
= false;
3591 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3592 uint64_t left_to_receive
= state
->left_to_receive
;
3593 struct relay_session
*session
;
3595 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3596 state
->header
.stream_id
, state
->header
.net_seq_num
,
3597 state
->received
, left_to_receive
);
3599 stream
= stream_get_by_id(state
->header
.stream_id
);
3601 /* Protocol error. */
3602 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3603 state
->header
.stream_id
);
3604 status
= RELAY_CONNECTION_STATUS_ERROR
;
3608 pthread_mutex_lock(&stream
->lock
);
3609 session
= stream
->trace
->session
;
3610 if (!conn
->session
) {
3611 ret
= connection_set_session(conn
, session
);
3613 status
= RELAY_CONNECTION_STATUS_ERROR
;
3614 goto end_stream_unlock
;
3619 * The size of the "chunk" received on any iteration is bounded by:
3620 * - the data left to receive,
3621 * - the data immediately available on the socket,
3622 * - the on-stack data buffer
3624 while (left_to_receive
> 0 && !partial_recv
) {
3625 size_t recv_size
= min(left_to_receive
, chunk_size
);
3626 struct lttng_buffer_view packet_chunk
;
3628 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3629 recv_size
, MSG_DONTWAIT
);
3631 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3632 PERROR("Socket %d error", conn
->sock
->fd
);
3633 status
= RELAY_CONNECTION_STATUS_ERROR
;
3635 goto end_stream_unlock
;
3636 } else if (ret
== 0) {
3637 /* No more data ready to be consumed on socket. */
3638 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3639 state
->header
.stream_id
);
3640 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3642 } else if (ret
< (int) recv_size
) {
3644 * All the data available on the socket has been
3647 partial_recv
= true;
3651 packet_chunk
= lttng_buffer_view_init(data_buffer
,
3653 assert(packet_chunk
.data
);
3655 ret
= stream_write(stream
, &packet_chunk
, 0);
3657 ERR("Relay error writing data to file");
3658 status
= RELAY_CONNECTION_STATUS_ERROR
;
3659 goto end_stream_unlock
;
3662 left_to_receive
-= recv_size
;
3663 state
->received
+= recv_size
;
3664 state
->left_to_receive
= left_to_receive
;
3667 if (state
->left_to_receive
> 0) {
3669 * Did not receive all the data expected, wait for more data to
3670 * become available on the socket.
3672 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3673 state
->header
.stream_id
, state
->received
,
3674 state
->left_to_receive
);
3675 goto end_stream_unlock
;
3678 ret
= stream_write(stream
, NULL
, state
->header
.padding_size
);
3680 status
= RELAY_CONNECTION_STATUS_ERROR
;
3681 goto end_stream_unlock
;
3684 if (session_streams_have_index(session
)) {
3685 ret
= stream_update_index(stream
, state
->header
.net_seq_num
,
3686 state
->rotate_index
, &index_flushed
,
3687 state
->header
.data_size
+ state
->header
.padding_size
);
3689 ERR("Failed to update index: stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3690 stream
->stream_handle
,
3691 state
->header
.net_seq_num
, ret
);
3692 status
= RELAY_CONNECTION_STATUS_ERROR
;
3693 goto end_stream_unlock
;
3697 if (stream
->prev_data_seq
== -1ULL) {
3701 ret
= stream_complete_packet(stream
, state
->header
.data_size
+
3702 state
->header
.padding_size
, state
->header
.net_seq_num
,
3705 status
= RELAY_CONNECTION_STATUS_ERROR
;
3706 goto end_stream_unlock
;
3710 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3711 * contents of *state which are aliased (union) to the same location as
3712 * the new state. Don't use it beyond this point.
3714 connection_reset_protocol_state(conn
);
3718 close_requested
= stream
->close_requested
;
3719 pthread_mutex_unlock(&stream
->lock
);
3720 if (close_requested
&& left_to_receive
== 0) {
3721 try_stream_close(stream
);
3725 pthread_mutex_lock(&session
->lock
);
3726 uatomic_set(&session
->new_streams
, 1);
3727 pthread_mutex_unlock(&session
->lock
);
3736 * relay_process_data: Process the data received on the data socket
3738 static enum relay_connection_status
relay_process_data(
3739 struct relay_connection
*conn
)
3741 enum relay_connection_status status
;
3743 switch (conn
->protocol
.data
.state_id
) {
3744 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3745 status
= relay_process_data_receive_header(conn
);
3747 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3748 status
= relay_process_data_receive_payload(conn
);
3751 ERR("Unexpected data connection communication state.");
3758 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3762 (void) lttng_poll_del(events
, pollfd
);
3764 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
, &pollfd
, 1,
3765 fd_tracker_util_close_fd
, NULL
);
3767 ERR("Closing pollfd %d", pollfd
);
3771 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3772 int pollfd
, struct relay_connection
*conn
)
3774 const char *type_str
;
3776 switch (conn
->type
) {
3781 type_str
= "Control";
3783 case RELAY_VIEWER_COMMAND
:
3784 type_str
= "Viewer Command";
3786 case RELAY_VIEWER_NOTIFICATION
:
3787 type_str
= "Viewer Notification";
3790 type_str
= "Unknown";
3792 cleanup_connection_pollfd(events
, pollfd
);
3793 connection_put(conn
);
3794 DBG("%s connection closed with %d", type_str
, pollfd
);
3798 * This thread does the actual work
3800 static void *relay_thread_worker(void *data
)
3802 int ret
, err
= -1, last_seen_data_fd
= -1;
3804 struct lttng_poll_event events
;
3805 struct lttng_ht
*relay_connections_ht
;
3806 struct lttng_ht_iter iter
;
3807 struct relay_connection
*destroy_conn
= NULL
;
3809 DBG("[thread] Relay worker started");
3811 rcu_register_thread();
3813 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3815 if (testpoint(relayd_thread_worker
)) {
3816 goto error_testpoint
;
3819 health_code_update();
3821 /* table of connections indexed on socket */
3822 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3823 if (!relay_connections_ht
) {
3824 goto relay_connections_ht_error
;
3827 ret
= create_named_thread_poll_set(&events
, 2, "Worker thread epoll");
3829 goto error_poll_create
;
3832 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3839 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3841 health_code_update();
3843 /* Infinite blocking call, waiting for transmission */
3844 DBG3("Relayd worker thread polling...");
3845 health_poll_entry();
3846 ret
= lttng_poll_wait(&events
, -1);
3850 * Restart interrupted system call.
3852 if (errno
== EINTR
) {
3861 * Process control. The control connection is
3862 * prioritized so we don't starve it with high
3863 * throughput tracing data on the data connection.
3865 for (i
= 0; i
< nb_fd
; i
++) {
3866 /* Fetch once the poll data */
3867 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3868 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3870 health_code_update();
3872 /* Thread quit pipe has been closed. Killing thread. */
3873 ret
= check_thread_quit_pipe(pollfd
, revents
);
3879 /* Inspect the relay conn pipe for new connection */
3880 if (pollfd
== relay_conn_pipe
[0]) {
3881 if (revents
& LPOLLIN
) {
3882 struct relay_connection
*conn
;
3884 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3888 ret
= lttng_poll_add(&events
,
3890 LPOLLIN
| LPOLLRDHUP
);
3892 ERR("Failed to add new connection file descriptor to poll set");
3895 connection_ht_add(relay_connections_ht
, conn
);
3896 DBG("Connection socket %d added", conn
->sock
->fd
);
3897 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3898 ERR("Relay connection pipe error");
3901 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3905 struct relay_connection
*ctrl_conn
;
3907 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3908 /* If not found, there is a synchronization issue. */
3911 if (ctrl_conn
->type
== RELAY_DATA
) {
3912 if (revents
& LPOLLIN
) {
3914 * Flag the last seen data fd not deleted. It will be
3915 * used as the last seen fd if any fd gets deleted in
3918 last_notdel_data_fd
= pollfd
;
3920 goto put_ctrl_connection
;
3922 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3924 if (revents
& LPOLLIN
) {
3925 enum relay_connection_status status
;
3927 status
= relay_process_control(ctrl_conn
);
3928 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3930 * On socket error flag the session as aborted to force
3931 * the cleanup of its stream otherwise it can leak
3932 * during the lifetime of the relayd.
3934 * This prevents situations in which streams can be
3935 * left opened because an index was received, the
3936 * control connection is closed, and the data
3937 * connection is closed (uncleanly) before the packet's
3940 * Since the control connection encountered an error,
3941 * it is okay to be conservative and close the
3942 * session right now as we can't rely on the protocol
3943 * being respected anymore.
3945 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3946 session_abort(ctrl_conn
->session
);
3949 /* Clear the connection on error or close. */
3950 relay_thread_close_connection(&events
,
3955 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3956 relay_thread_close_connection(&events
,
3958 if (last_seen_data_fd
== pollfd
) {
3959 last_seen_data_fd
= last_notdel_data_fd
;
3962 ERR("Unexpected poll events %u for control sock %d",
3964 connection_put(ctrl_conn
);
3967 put_ctrl_connection
:
3968 connection_put(ctrl_conn
);
3973 * The last loop handled a control request, go back to poll to make
3974 * sure we prioritise the control socket.
3980 if (last_seen_data_fd
>= 0) {
3981 for (i
= 0; i
< nb_fd
; i
++) {
3982 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3984 health_code_update();
3986 if (last_seen_data_fd
== pollfd
) {
3993 /* Process data connection. */
3994 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3995 /* Fetch the poll data. */
3996 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3997 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3998 struct relay_connection
*data_conn
;
4000 health_code_update();
4003 /* No activity for this FD (poll implementation). */
4007 /* Skip the command pipe. It's handled in the first loop. */
4008 if (pollfd
== relay_conn_pipe
[0]) {
4012 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
4014 /* Skip it. Might be removed before. */
4017 if (data_conn
->type
== RELAY_CONTROL
) {
4018 goto put_data_connection
;
4020 assert(data_conn
->type
== RELAY_DATA
);
4022 if (revents
& LPOLLIN
) {
4023 enum relay_connection_status status
;
4025 status
= relay_process_data(data_conn
);
4026 /* Connection closed or error. */
4027 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
4029 * On socket error flag the session as aborted to force
4030 * the cleanup of its stream otherwise it can leak
4031 * during the lifetime of the relayd.
4033 * This prevents situations in which streams can be
4034 * left opened because an index was received, the
4035 * control connection is closed, and the data
4036 * connection is closed (uncleanly) before the packet's
4039 * Since the data connection encountered an error,
4040 * it is okay to be conservative and close the
4041 * session right now as we can't rely on the protocol
4042 * being respected anymore.
4044 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
4045 session_abort(data_conn
->session
);
4047 relay_thread_close_connection(&events
, pollfd
,
4050 * Every goto restart call sets the last seen fd where
4051 * here we don't really care since we gracefully
4052 * continue the loop after the connection is deleted.
4055 /* Keep last seen port. */
4056 last_seen_data_fd
= pollfd
;
4057 connection_put(data_conn
);
4060 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4061 relay_thread_close_connection(&events
, pollfd
,
4064 ERR("Unknown poll events %u for data sock %d",
4067 put_data_connection
:
4068 connection_put(data_conn
);
4070 last_seen_data_fd
= -1;
4073 /* Normal exit, no error */
4078 /* Cleanup remaining connection object. */
4080 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
4083 health_code_update();
4085 session_abort(destroy_conn
->session
);
4088 * No need to grab another ref, because we own
4091 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
4096 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
4098 lttng_ht_destroy(relay_connections_ht
);
4099 relay_connections_ht_error
:
4100 /* Close relay conn pipes */
4101 (void) fd_tracker_util_pipe_close(the_fd_tracker
,
4104 DBG("Thread exited with error");
4106 DBG("Worker thread cleanup complete");
4110 ERR("Health error occurred in %s", __func__
);
4112 health_unregister(health_relayd
);
4113 rcu_unregister_thread();
4114 lttng_relay_stop_threads();
4119 * Create the relay command pipe to wake thread_manage_apps.
4120 * Closed in cleanup().
4122 static int create_relay_conn_pipe(void)
4124 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
4125 "Relayd connection pipe", relay_conn_pipe
);
4128 static int stdio_open(void *data
, int *fds
)
4130 fds
[0] = fileno(stdout
);
4131 fds
[1] = fileno(stderr
);
4135 static int track_stdio(void)
4138 const char *names
[] = { "stdout", "stderr" };
4140 return fd_tracker_open_unsuspendable_fd(the_fd_tracker
, fds
,
4141 names
, 2, stdio_open
, NULL
);
4147 int main(int argc
, char **argv
)
4149 bool thread_is_rcu_registered
= false;
4150 int ret
= 0, retval
= 0;
4152 char *unlinked_file_directory_path
= NULL
, *output_path
= NULL
;
4154 /* Parse environment variables */
4155 ret
= parse_env_options();
4163 * Command line arguments overwrite environment.
4166 if (set_options(argc
, argv
)) {
4171 if (set_signal_handler()) {
4176 relayd_config_log();
4178 if (opt_print_version
) {
4184 ret
= fclose(stdin
);
4186 PERROR("Failed to close stdin");
4190 DBG("Clear command %s", opt_allow_clear
? "allowed" : "disallowed");
4192 /* Try to create directory if -o, --output is specified. */
4193 if (opt_output_path
) {
4194 if (*opt_output_path
!= '/') {
4195 ERR("Please specify an absolute path for -o, --output PATH");
4200 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
4203 ERR("Unable to create %s", opt_output_path
);
4210 if (opt_daemon
|| opt_background
) {
4211 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4219 if (opt_working_directory
) {
4220 ret
= utils_change_working_directory(opt_working_directory
);
4222 /* All errors are already logged. */
4227 sessiond_trace_chunk_registry
= sessiond_trace_chunk_registry_create();
4228 if (!sessiond_trace_chunk_registry
) {
4229 ERR("Failed to initialize session daemon trace chunk registry");
4235 * The RCU thread registration (and use, through the fd-tracker's
4236 * creation) is done after the daemonization to allow us to not
4237 * deal with liburcu's fork() management as the call RCU needs to
4240 rcu_register_thread();
4241 thread_is_rcu_registered
= true;
4243 output_path
= create_output_path("");
4245 ERR("Failed to get output path");
4249 ret
= asprintf(&unlinked_file_directory_path
, "%s/%s", output_path
,
4250 DEFAULT_UNLINKED_FILES_DIRECTORY
);
4253 ERR("Failed to format unlinked file directory path");
4257 the_fd_tracker
= fd_tracker_create(
4258 unlinked_file_directory_path
, lttng_opt_fd_pool_size
);
4259 free(unlinked_file_directory_path
);
4260 if (!the_fd_tracker
) {
4265 ret
= track_stdio();
4271 /* Initialize thread health monitoring */
4272 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
4273 if (!health_relayd
) {
4274 PERROR("health_app_create error");
4279 /* Create thread quit pipe */
4280 if (init_thread_quit_pipe()) {
4285 /* Setup the thread apps communication pipe. */
4286 if (create_relay_conn_pipe()) {
4291 /* Init relay command queue. */
4292 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
4294 /* Initialize communication library */
4296 lttcomm_inet_init();
4298 /* tables of sessions indexed by session ID */
4299 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4305 /* tables of streams indexed by stream ID */
4306 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4307 if (!relay_streams_ht
) {
4312 /* tables of streams indexed by stream ID */
4313 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4314 if (!viewer_streams_ht
) {
4319 ret
= init_health_quit_pipe();
4325 /* Create thread to manage the client socket */
4326 ret
= pthread_create(&health_thread
, default_pthread_attr(),
4327 thread_manage_health
, (void *) NULL
);
4330 PERROR("pthread_create health");
4335 /* Setup the dispatcher thread */
4336 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
4337 relay_thread_dispatcher
, (void *) NULL
);
4340 PERROR("pthread_create dispatcher");
4342 goto exit_dispatcher_thread
;
4345 /* Setup the worker thread */
4346 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
4347 relay_thread_worker
, NULL
);
4350 PERROR("pthread_create worker");
4352 goto exit_worker_thread
;
4355 /* Setup the listener thread */
4356 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
4357 relay_thread_listener
, (void *) NULL
);
4360 PERROR("pthread_create listener");
4362 goto exit_listener_thread
;
4365 ret
= relayd_live_create(live_uri
);
4367 ERR("Starting live viewer threads");
4373 * This is where we start awaiting program completion (e.g. through
4374 * signal that asks threads to teardown).
4377 ret
= relayd_live_join();
4383 ret
= pthread_join(listener_thread
, &status
);
4386 PERROR("pthread_join listener_thread");
4390 exit_listener_thread
:
4391 ret
= pthread_join(worker_thread
, &status
);
4394 PERROR("pthread_join worker_thread");
4399 ret
= pthread_join(dispatcher_thread
, &status
);
4402 PERROR("pthread_join dispatcher_thread");
4405 exit_dispatcher_thread
:
4407 ret
= pthread_join(health_thread
, &status
);
4410 PERROR("pthread_join health_thread");
4415 * Wait for all pending call_rcu work to complete before tearing
4416 * down data structures. call_rcu worker may be trying to
4417 * perform lookups in those structures.
4422 /* Ensure all prior call_rcu are done. */
4425 if (thread_is_rcu_registered
) {
4426 rcu_unregister_thread();