2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
35 #include <sys/types.h>
38 #include <urcu/futex.h>
39 #include <urcu/uatomic.h>
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/compat/endian.h>
48 #include <common/compat/getenv.h>
49 #include <common/defaults.h>
50 #include <common/daemonize.h>
51 #include <common/futex.h>
52 #include <common/sessiond-comm/sessiond-comm.h>
53 #include <common/sessiond-comm/inet.h>
54 #include <common/sessiond-comm/relayd.h>
55 #include <common/uri.h>
56 #include <common/utils.h>
57 #include <common/align.h>
58 #include <common/config/session-config.h>
59 #include <common/dynamic-buffer.h>
60 #include <common/buffer-view.h>
61 #include <urcu/rculist.h>
64 #include "ctf-trace.h"
67 #include "lttng-relayd.h"
69 #include "health-relayd.h"
70 #include "testpoint.h"
71 #include "viewer-stream.h"
74 #include "connection.h"
75 #include "tracefile-array.h"
76 #include "tcp_keep_alive.h"
77 #include "sessiond-trace-chunks.h"
79 static const char *help_msg
=
80 #ifdef LTTNG_EMBED_HELP
81 #include <lttng-relayd.8.h>
87 enum relay_connection_status
{
88 RELAY_CONNECTION_STATUS_OK
,
89 /* An error occurred while processing an event on the connection. */
90 RELAY_CONNECTION_STATUS_ERROR
,
91 /* Connection closed/shutdown cleanly. */
92 RELAY_CONNECTION_STATUS_CLOSED
,
95 /* command line options */
96 char *opt_output_path
;
97 static int opt_daemon
, opt_background
;
100 * We need to wait for listener and live listener threads, as well as
101 * health check thread, before being ready to signal readiness.
103 #define NR_LTTNG_RELAY_READY 3
104 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
106 /* Size of receive buffer. */
107 #define RECV_DATA_BUFFER_SIZE 65536
108 #define FILE_COPY_BUFFER_SIZE 65536
110 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
111 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
113 static struct lttng_uri
*control_uri
;
114 static struct lttng_uri
*data_uri
;
115 static struct lttng_uri
*live_uri
;
117 const char *progname
;
119 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
120 static int tracing_group_name_override
;
122 const char * const config_section_name
= "relayd";
125 * Quit pipe for all threads. This permits a single cancellation point
126 * for all threads when receiving an event on the pipe.
128 int thread_quit_pipe
[2] = { -1, -1 };
131 * This pipe is used to inform the worker thread that a command is queued and
132 * ready to be processed.
134 static int relay_conn_pipe
[2] = { -1, -1 };
136 /* Shared between threads */
137 static int dispatch_thread_exit
;
139 static pthread_t listener_thread
;
140 static pthread_t dispatcher_thread
;
141 static pthread_t worker_thread
;
142 static pthread_t health_thread
;
145 * last_relay_stream_id_lock protects last_relay_stream_id increment
146 * atomicity on 32-bit architectures.
148 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
149 static uint64_t last_relay_stream_id
;
152 * Relay command queue.
154 * The relay_thread_listener and relay_thread_dispatcher communicate with this
157 static struct relay_conn_queue relay_conn_queue
;
159 /* Global relay stream hash table. */
160 struct lttng_ht
*relay_streams_ht
;
162 /* Global relay viewer stream hash table. */
163 struct lttng_ht
*viewer_streams_ht
;
165 /* Global relay sessions hash table. */
166 struct lttng_ht
*sessions_ht
;
168 /* Relayd health monitoring */
169 struct health_app
*health_relayd
;
171 struct sessiond_trace_chunk_registry
*sessiond_trace_chunk_registry
;
173 static struct option long_options
[] = {
174 { "control-port", 1, 0, 'C', },
175 { "data-port", 1, 0, 'D', },
176 { "live-port", 1, 0, 'L', },
177 { "daemonize", 0, 0, 'd', },
178 { "background", 0, 0, 'b', },
179 { "group", 1, 0, 'g', },
180 { "help", 0, 0, 'h', },
181 { "output", 1, 0, 'o', },
182 { "verbose", 0, 0, 'v', },
183 { "config", 1, 0, 'f' },
184 { "version", 0, 0, 'V' },
188 static const char *config_ignore_options
[] = { "help", "config", "version" };
191 * Take an option from the getopt output and set it in the right variable to be
194 * Return 0 on success else a negative value.
196 static int set_option(int opt
, const char *arg
, const char *optname
)
202 fprintf(stderr
, "option %s", optname
);
204 fprintf(stderr
, " with arg %s\n", arg
);
208 if (lttng_is_setuid_setgid()) {
209 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
210 "-C, --control-port");
212 ret
= uri_parse(arg
, &control_uri
);
214 ERR("Invalid control URI specified");
217 if (control_uri
->port
== 0) {
218 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
223 if (lttng_is_setuid_setgid()) {
224 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
227 ret
= uri_parse(arg
, &data_uri
);
229 ERR("Invalid data URI specified");
232 if (data_uri
->port
== 0) {
233 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
238 if (lttng_is_setuid_setgid()) {
239 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
242 ret
= uri_parse(arg
, &live_uri
);
244 ERR("Invalid live URI specified");
247 if (live_uri
->port
== 0) {
248 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
259 if (lttng_is_setuid_setgid()) {
260 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
263 tracing_group_name
= strdup(arg
);
264 if (tracing_group_name
== NULL
) {
269 tracing_group_name_override
= 1;
273 ret
= utils_show_help(8, "lttng-relayd", help_msg
);
275 ERR("Cannot show --help for `lttng-relayd`");
280 fprintf(stdout
, "%s\n", VERSION
);
283 if (lttng_is_setuid_setgid()) {
284 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
287 ret
= asprintf(&opt_output_path
, "%s", arg
);
290 PERROR("asprintf opt_output_path");
296 /* Verbose level can increase using multiple -v */
298 lttng_opt_verbose
= config_parse_value(arg
);
300 /* Only 3 level of verbosity (-vvv). */
301 if (lttng_opt_verbose
< 3) {
302 lttng_opt_verbose
+= 1;
307 /* Unknown option or other error.
308 * Error is printed by getopt, just return */
321 * config_entry_handler_cb used to handle options read from a config file.
322 * See config_entry_handler_cb comment in common/config/session-config.h for the
323 * return value conventions.
325 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
329 if (!entry
|| !entry
->name
|| !entry
->value
) {
334 /* Check if the option is to be ignored */
335 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
336 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
341 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
342 /* Ignore if entry name is not fully matched. */
343 if (strcmp(entry
->name
, long_options
[i
].name
)) {
348 * If the option takes no argument on the command line,
349 * we have to check if the value is "true". We support
350 * non-zero numeric values, true, on and yes.
352 if (!long_options
[i
].has_arg
) {
353 ret
= config_parse_value(entry
->value
);
356 WARN("Invalid configuration value \"%s\" for option %s",
357 entry
->value
, entry
->name
);
359 /* False, skip boolean config option. */
364 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
368 WARN("Unrecognized option \"%s\" in daemon configuration file.",
375 static int set_options(int argc
, char **argv
)
377 int c
, ret
= 0, option_index
= 0, retval
= 0;
378 int orig_optopt
= optopt
, orig_optind
= optind
;
379 char *default_address
, *optstring
;
380 const char *config_path
= NULL
;
382 optstring
= utils_generate_optstring(long_options
,
383 sizeof(long_options
) / sizeof(struct option
));
389 /* Check for the --config option */
391 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
392 &option_index
)) != -1) {
396 } else if (c
!= 'f') {
400 if (lttng_is_setuid_setgid()) {
401 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
404 config_path
= utils_expand_path(optarg
);
406 ERR("Failed to resolve path: %s", optarg
);
411 ret
= config_get_section_entries(config_path
, config_section_name
,
412 config_entry_handler
, NULL
);
415 ERR("Invalid configuration option at line %i", ret
);
421 /* Reset getopt's global state */
422 optopt
= orig_optopt
;
423 optind
= orig_optind
;
425 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
430 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
437 /* assign default values */
438 if (control_uri
== NULL
) {
439 ret
= asprintf(&default_address
,
440 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
441 DEFAULT_NETWORK_CONTROL_PORT
);
443 PERROR("asprintf default data address");
448 ret
= uri_parse(default_address
, &control_uri
);
449 free(default_address
);
451 ERR("Invalid control URI specified");
456 if (data_uri
== NULL
) {
457 ret
= asprintf(&default_address
,
458 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
459 DEFAULT_NETWORK_DATA_PORT
);
461 PERROR("asprintf default data address");
466 ret
= uri_parse(default_address
, &data_uri
);
467 free(default_address
);
469 ERR("Invalid data URI specified");
474 if (live_uri
== NULL
) {
475 ret
= asprintf(&default_address
,
476 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
477 DEFAULT_NETWORK_VIEWER_PORT
);
479 PERROR("asprintf default viewer control address");
484 ret
= uri_parse(default_address
, &live_uri
);
485 free(default_address
);
487 ERR("Invalid viewer control URI specified");
498 static void print_global_objects(void)
500 rcu_register_thread();
502 print_viewer_streams();
503 print_relay_streams();
506 rcu_unregister_thread();
512 static void relayd_cleanup(void)
514 print_global_objects();
518 if (viewer_streams_ht
)
519 lttng_ht_destroy(viewer_streams_ht
);
520 if (relay_streams_ht
)
521 lttng_ht_destroy(relay_streams_ht
);
523 lttng_ht_destroy(sessions_ht
);
525 /* free the dynamically allocated opt_output_path */
526 free(opt_output_path
);
528 /* Close thread quit pipes */
529 utils_close_pipe(thread_quit_pipe
);
531 uri_free(control_uri
);
533 /* Live URI is freed in the live thread. */
535 if (tracing_group_name_override
) {
536 free((void *) tracing_group_name
);
541 * Write to writable pipe used to notify a thread.
543 static int notify_thread_pipe(int wpipe
)
547 ret
= lttng_write(wpipe
, "!", 1);
549 PERROR("write poll pipe");
557 static int notify_health_quit_pipe(int *pipe
)
561 ret
= lttng_write(pipe
[1], "4", 1);
563 PERROR("write relay health quit");
572 * Stop all relayd and relayd-live threads.
574 int lttng_relay_stop_threads(void)
578 /* Stopping all threads */
579 DBG("Terminating all threads");
580 if (notify_thread_pipe(thread_quit_pipe
[1])) {
581 ERR("write error on thread quit pipe");
585 if (notify_health_quit_pipe(health_quit_pipe
)) {
586 ERR("write error on health quit pipe");
589 /* Dispatch thread */
590 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
591 futex_nto1_wake(&relay_conn_queue
.futex
);
593 if (relayd_live_stop()) {
594 ERR("Error stopping live threads");
601 * Signal handler for the daemon
603 * Simply stop all worker threads, leaving main() return gracefully after
604 * joining all threads and calling cleanup().
606 static void sighandler(int sig
)
610 DBG("SIGINT caught");
611 if (lttng_relay_stop_threads()) {
612 ERR("Error stopping threads");
616 DBG("SIGTERM caught");
617 if (lttng_relay_stop_threads()) {
618 ERR("Error stopping threads");
622 CMM_STORE_SHARED(recv_child_signal
, 1);
630 * Setup signal handler for :
631 * SIGINT, SIGTERM, SIGPIPE
633 static int set_signal_handler(void)
639 if ((ret
= sigemptyset(&sigset
)) < 0) {
640 PERROR("sigemptyset");
647 sa
.sa_handler
= sighandler
;
648 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
653 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
658 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
663 sa
.sa_handler
= SIG_IGN
;
664 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
669 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
674 void lttng_relay_notify_ready(void)
676 /* Notify the parent of the fork() process that we are ready. */
677 if (opt_daemon
|| opt_background
) {
678 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
679 kill(child_ppid
, SIGUSR1
);
685 * Init thread quit pipe.
687 * Return -1 on error or 0 if all pipes are created.
689 static int init_thread_quit_pipe(void)
693 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
699 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
701 static int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
705 if (events
== NULL
|| size
== 0) {
710 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
716 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
728 * Check if the thread quit pipe was triggered.
730 * Return 1 if it was triggered else 0;
732 static int check_thread_quit_pipe(int fd
, uint32_t events
)
734 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
742 * Create and init socket from uri.
744 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
)
747 struct lttcomm_sock
*sock
= NULL
;
749 sock
= lttcomm_alloc_sock_from_uri(uri
);
751 ERR("Allocating socket");
755 ret
= lttcomm_create_sock(sock
);
759 DBG("Listening on sock %d", sock
->fd
);
761 ret
= sock
->ops
->bind(sock
);
763 PERROR("Failed to bind socket");
767 ret
= sock
->ops
->listen(sock
, -1);
777 lttcomm_destroy_sock(sock
);
783 * This thread manages the listening for new connections on the network
785 static void *relay_thread_listener(void *data
)
787 int i
, ret
, pollfd
, err
= -1;
788 uint32_t revents
, nb_fd
;
789 struct lttng_poll_event events
;
790 struct lttcomm_sock
*control_sock
, *data_sock
;
792 DBG("[thread] Relay listener started");
794 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
796 health_code_update();
798 control_sock
= relay_socket_create(control_uri
);
800 goto error_sock_control
;
803 data_sock
= relay_socket_create(data_uri
);
805 goto error_sock_relay
;
809 * Pass 3 as size here for the thread quit pipe, control and
812 ret
= create_thread_poll_set(&events
, 3);
814 goto error_create_poll
;
817 /* Add the control socket */
818 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
823 /* Add the data socket */
824 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
829 lttng_relay_notify_ready();
831 if (testpoint(relayd_thread_listener
)) {
832 goto error_testpoint
;
836 health_code_update();
838 DBG("Listener accepting connections");
842 ret
= lttng_poll_wait(&events
, -1);
846 * Restart interrupted system call.
848 if (errno
== EINTR
) {
856 DBG("Relay new connection received");
857 for (i
= 0; i
< nb_fd
; i
++) {
858 health_code_update();
860 /* Fetch once the poll data */
861 revents
= LTTNG_POLL_GETEV(&events
, i
);
862 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
864 /* Thread quit pipe has been closed. Killing thread. */
865 ret
= check_thread_quit_pipe(pollfd
, revents
);
871 if (revents
& LPOLLIN
) {
873 * A new connection is requested, therefore a
874 * sessiond/consumerd connection is allocated in
875 * this thread, enqueued to a global queue and
876 * dequeued (and freed) in the worker thread.
879 struct relay_connection
*new_conn
;
880 struct lttcomm_sock
*newsock
;
881 enum connection_type type
;
883 if (pollfd
== data_sock
->fd
) {
885 newsock
= data_sock
->ops
->accept(data_sock
);
886 DBG("Relay data connection accepted, socket %d",
889 assert(pollfd
== control_sock
->fd
);
890 type
= RELAY_CONTROL
;
891 newsock
= control_sock
->ops
->accept(control_sock
);
892 DBG("Relay control connection accepted, socket %d",
896 PERROR("accepting sock");
900 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
903 PERROR("setsockopt inet");
904 lttcomm_destroy_sock(newsock
);
908 ret
= socket_apply_keep_alive_config(newsock
->fd
);
910 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
912 lttcomm_destroy_sock(newsock
);
916 new_conn
= connection_create(newsock
, type
);
918 lttcomm_destroy_sock(newsock
);
922 /* Enqueue request for the dispatcher thread. */
923 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
927 * Wake the dispatch queue futex.
928 * Implicit memory barrier with the
929 * exchange in cds_wfcq_enqueue.
931 futex_nto1_wake(&relay_conn_queue
.futex
);
932 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
933 ERR("socket poll error");
936 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
946 lttng_poll_clean(&events
);
948 if (data_sock
->fd
>= 0) {
949 ret
= data_sock
->ops
->close(data_sock
);
954 lttcomm_destroy_sock(data_sock
);
956 if (control_sock
->fd
>= 0) {
957 ret
= control_sock
->ops
->close(control_sock
);
962 lttcomm_destroy_sock(control_sock
);
966 ERR("Health error occurred in %s", __func__
);
968 health_unregister(health_relayd
);
969 DBG("Relay listener thread cleanup complete");
970 lttng_relay_stop_threads();
975 * This thread manages the dispatching of the requests to worker threads
977 static void *relay_thread_dispatcher(void *data
)
981 struct cds_wfcq_node
*node
;
982 struct relay_connection
*new_conn
= NULL
;
984 DBG("[thread] Relay dispatcher started");
986 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
988 if (testpoint(relayd_thread_dispatcher
)) {
989 goto error_testpoint
;
992 health_code_update();
995 health_code_update();
997 /* Atomically prepare the queue futex */
998 futex_nto1_prepare(&relay_conn_queue
.futex
);
1000 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1005 health_code_update();
1007 /* Dequeue commands */
1008 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1009 &relay_conn_queue
.tail
);
1011 DBG("Woken up but nothing in the relay command queue");
1012 /* Continue thread execution */
1015 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1017 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1020 * Inform worker thread of the new request. This
1021 * call is blocking so we can be assured that
1022 * the data will be read at some point in time
1023 * or wait to the end of the world :)
1025 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1027 PERROR("write connection pipe");
1028 connection_put(new_conn
);
1031 } while (node
!= NULL
);
1033 /* Futex wait on queue. Blocking call on futex() */
1034 health_poll_entry();
1035 futex_nto1_wait(&relay_conn_queue
.futex
);
1039 /* Normal exit, no error */
1046 ERR("Health error occurred in %s", __func__
);
1048 health_unregister(health_relayd
);
1049 DBG("Dispatch thread dying");
1050 lttng_relay_stop_threads();
1055 * Set index data from the control port to a given index object.
1057 static int set_index_control_data(struct relay_index
*index
,
1058 struct lttcomm_relayd_index
*data
,
1059 struct relay_connection
*conn
)
1061 struct ctf_packet_index index_data
;
1064 * The index on disk is encoded in big endian.
1066 index_data
.packet_size
= htobe64(data
->packet_size
);
1067 index_data
.content_size
= htobe64(data
->content_size
);
1068 index_data
.timestamp_begin
= htobe64(data
->timestamp_begin
);
1069 index_data
.timestamp_end
= htobe64(data
->timestamp_end
);
1070 index_data
.events_discarded
= htobe64(data
->events_discarded
);
1071 index_data
.stream_id
= htobe64(data
->stream_id
);
1073 if (conn
->minor
>= 8) {
1074 index
->index_data
.stream_instance_id
= htobe64(data
->stream_instance_id
);
1075 index
->index_data
.packet_seq_num
= htobe64(data
->packet_seq_num
);
1078 return relay_index_set_data(index
, &index_data
);
1081 static bool session_streams_have_index(const struct relay_session
*session
)
1083 return session
->minor
>= 4 && !session
->snapshot
;
1087 * Handle the RELAYD_CREATE_SESSION command.
1089 * On success, send back the session id or else return a negative value.
1091 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1092 struct relay_connection
*conn
,
1093 const struct lttng_buffer_view
*payload
)
1097 struct relay_session
*session
= NULL
;
1098 struct lttcomm_relayd_status_session reply
= {};
1099 char session_name
[LTTNG_NAME_MAX
] = {};
1100 char hostname
[LTTNG_HOST_NAME_MAX
] = {};
1101 uint32_t live_timer
= 0;
1102 bool snapshot
= false;
1103 /* Left nil for peers < 2.11. */
1104 lttng_uuid sessiond_uuid
= {};
1105 LTTNG_OPTIONAL(uint64_t) id_sessiond
= {};
1106 LTTNG_OPTIONAL(uint64_t) current_chunk_id
= {};
1107 LTTNG_OPTIONAL(time_t) creation_time
= {};
1109 if (conn
->minor
< 4) {
1110 /* From 2.1 to 2.3 */
1112 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1113 /* From 2.4 to 2.10 */
1114 ret
= cmd_create_session_2_4(payload
, session_name
,
1115 hostname
, &live_timer
, &snapshot
);
1117 bool has_current_chunk
;
1118 uint64_t current_chunk_id_value
;
1119 time_t creation_time_value
;
1120 uint64_t id_sessiond_value
;
1122 /* From 2.11 to ... */
1123 ret
= cmd_create_session_2_11(payload
, session_name
, hostname
,
1124 &live_timer
, &snapshot
, &id_sessiond_value
,
1125 sessiond_uuid
, &has_current_chunk
,
1126 ¤t_chunk_id_value
, &creation_time_value
);
1127 if (lttng_uuid_is_nil(sessiond_uuid
)) {
1128 /* The nil UUID is reserved for pre-2.11 clients. */
1129 ERR("Illegal nil UUID announced by peer in create session command");
1133 LTTNG_OPTIONAL_SET(&id_sessiond
, id_sessiond_value
);
1134 LTTNG_OPTIONAL_SET(&creation_time
, creation_time_value
);
1135 if (has_current_chunk
) {
1136 LTTNG_OPTIONAL_SET(¤t_chunk_id
,
1137 current_chunk_id_value
);
1145 session
= session_create(session_name
, hostname
, live_timer
,
1146 snapshot
, sessiond_uuid
,
1147 id_sessiond
.is_set
? &id_sessiond
.value
: NULL
,
1148 current_chunk_id
.is_set
? ¤t_chunk_id
.value
: NULL
,
1149 creation_time
.is_set
? &creation_time
.value
: NULL
,
1150 conn
->major
, conn
->minor
);
1155 assert(!conn
->session
);
1156 conn
->session
= session
;
1157 DBG("Created session %" PRIu64
, session
->id
);
1159 reply
.session_id
= htobe64(session
->id
);
1163 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1165 reply
.ret_code
= htobe32(LTTNG_OK
);
1168 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1169 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1170 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1174 if (ret
< 0 && session
) {
1175 session_put(session
);
1181 * When we have received all the streams and the metadata for a channel,
1182 * we make them visible to the viewer threads.
1184 static void publish_connection_local_streams(struct relay_connection
*conn
)
1186 struct relay_stream
*stream
;
1187 struct relay_session
*session
= conn
->session
;
1190 * We publish all streams belonging to a session atomically wrt
1193 pthread_mutex_lock(&session
->lock
);
1195 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1197 stream_publish(stream
);
1202 * Inform the viewer that there are new streams in the session.
1204 if (session
->viewer_attached
) {
1205 uatomic_set(&session
->new_streams
, 1);
1207 pthread_mutex_unlock(&session
->lock
);
1211 * relay_add_stream: allocate a new stream for a session
1213 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1214 struct relay_connection
*conn
,
1215 const struct lttng_buffer_view
*payload
)
1219 struct relay_session
*session
= conn
->session
;
1220 struct relay_stream
*stream
= NULL
;
1221 struct lttcomm_relayd_status_stream reply
;
1222 struct ctf_trace
*trace
= NULL
;
1223 uint64_t stream_handle
= -1ULL;
1224 char *path_name
= NULL
, *channel_name
= NULL
;
1225 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1226 struct relay_stream_chunk_id stream_chunk_id
= { 0 };
1228 if (!session
|| !conn
->version_check_done
) {
1229 ERR("Trying to add a stream before version check");
1231 goto end_no_session
;
1234 if (session
->minor
== 1) {
1236 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1238 } else if (session
->minor
> 1 && session
->minor
< 11) {
1239 /* From 2.2 to 2.10 */
1240 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1241 &channel_name
, &tracefile_size
, &tracefile_count
);
1243 /* From 2.11 to ... */
1244 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1245 &channel_name
, &tracefile_size
, &tracefile_count
,
1246 &stream_chunk_id
.value
);
1247 stream_chunk_id
.is_set
= true;
1254 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1258 /* This stream here has one reference on the trace. */
1260 pthread_mutex_lock(&last_relay_stream_id_lock
);
1261 stream_handle
= ++last_relay_stream_id
;
1262 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1264 /* We pass ownership of path_name and channel_name. */
1265 stream
= stream_create(trace
, stream_handle
, path_name
,
1266 channel_name
, tracefile_size
, tracefile_count
,
1269 channel_name
= NULL
;
1272 * Streams are the owners of their trace. Reference to trace is
1273 * kept within stream_create().
1275 ctf_trace_put(trace
);
1278 memset(&reply
, 0, sizeof(reply
));
1279 reply
.handle
= htobe64(stream_handle
);
1281 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1283 reply
.ret_code
= htobe32(LTTNG_OK
);
1286 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1287 sizeof(struct lttcomm_relayd_status_stream
), 0);
1288 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1289 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1301 * relay_close_stream: close a specific stream
1303 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1304 struct relay_connection
*conn
,
1305 const struct lttng_buffer_view
*payload
)
1309 struct relay_session
*session
= conn
->session
;
1310 struct lttcomm_relayd_close_stream stream_info
;
1311 struct lttcomm_relayd_generic_reply reply
;
1312 struct relay_stream
*stream
;
1314 DBG("Close stream received");
1316 if (!session
|| !conn
->version_check_done
) {
1317 ERR("Trying to close a stream before version check");
1319 goto end_no_session
;
1322 if (payload
->size
< sizeof(stream_info
)) {
1323 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1324 sizeof(stream_info
), payload
->size
);
1326 goto end_no_session
;
1328 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1329 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1330 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1332 stream
= stream_get_by_id(stream_info
.stream_id
);
1339 * Set last_net_seq_num before the close flag. Required by data
1342 pthread_mutex_lock(&stream
->lock
);
1343 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1344 pthread_mutex_unlock(&stream
->lock
);
1347 * This is one of the conditions which may trigger a stream close
1348 * with the others being:
1349 * 1) A close command is received for a stream
1350 * 2) The control connection owning the stream is closed
1351 * 3) We have received all of the stream's data _after_ a close
1354 try_stream_close(stream
);
1355 if (stream
->is_metadata
) {
1356 struct relay_viewer_stream
*vstream
;
1358 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1360 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1362 * Since all the metadata has been sent to the
1363 * viewer and that we have a request to close
1364 * its stream, we can safely teardown the
1365 * corresponding metadata viewer stream.
1367 viewer_stream_put(vstream
);
1369 /* Put local reference. */
1370 viewer_stream_put(vstream
);
1377 memset(&reply
, 0, sizeof(reply
));
1379 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1381 reply
.ret_code
= htobe32(LTTNG_OK
);
1383 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1384 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1385 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1386 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1396 * relay_reset_metadata: reset a metadata stream
1399 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1400 struct relay_connection
*conn
,
1401 const struct lttng_buffer_view
*payload
)
1405 struct relay_session
*session
= conn
->session
;
1406 struct lttcomm_relayd_reset_metadata stream_info
;
1407 struct lttcomm_relayd_generic_reply reply
;
1408 struct relay_stream
*stream
;
1410 DBG("Reset metadata received");
1412 if (!session
|| !conn
->version_check_done
) {
1413 ERR("Trying to reset a metadata stream before version check");
1415 goto end_no_session
;
1418 if (payload
->size
< sizeof(stream_info
)) {
1419 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1420 sizeof(stream_info
), payload
->size
);
1422 goto end_no_session
;
1424 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1425 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1426 stream_info
.version
= be64toh(stream_info
.version
);
1428 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1430 /* Unsupported for live sessions for now. */
1431 if (session
->live_timer
!= 0) {
1436 stream
= stream_get_by_id(stream_info
.stream_id
);
1441 pthread_mutex_lock(&stream
->lock
);
1442 if (!stream
->is_metadata
) {
1447 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
1448 0, 0, -1, -1, stream
->stream_fd
->fd
, NULL
,
1449 &stream
->stream_fd
->fd
);
1451 ERR("Failed to rotate metadata file %s of channel %s",
1452 stream
->path_name
, stream
->channel_name
);
1457 pthread_mutex_unlock(&stream
->lock
);
1461 memset(&reply
, 0, sizeof(reply
));
1463 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1465 reply
.ret_code
= htobe32(LTTNG_OK
);
1467 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1468 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1469 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1470 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1480 * relay_unknown_command: send -1 if received unknown command
1482 static void relay_unknown_command(struct relay_connection
*conn
)
1484 struct lttcomm_relayd_generic_reply reply
;
1487 memset(&reply
, 0, sizeof(reply
));
1488 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1489 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1490 if (send_ret
< sizeof(reply
)) {
1491 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1496 * relay_start: send an acknowledgment to the client to tell if we are
1497 * ready to receive data. We are ready if a session is established.
1499 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1500 struct relay_connection
*conn
,
1501 const struct lttng_buffer_view
*payload
)
1505 struct lttcomm_relayd_generic_reply reply
;
1506 struct relay_session
*session
= conn
->session
;
1509 DBG("Trying to start the streaming without a session established");
1510 ret
= htobe32(LTTNG_ERR_UNK
);
1513 memset(&reply
, 0, sizeof(reply
));
1514 reply
.ret_code
= htobe32(LTTNG_OK
);
1515 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1517 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1518 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1527 * Append padding to the file pointed by the file descriptor fd.
1529 static int write_padding_to_file(int fd
, uint32_t size
)
1538 zeros
= zmalloc(size
);
1539 if (zeros
== NULL
) {
1540 PERROR("zmalloc zeros for padding");
1545 ret
= lttng_write(fd
, zeros
, size
);
1547 PERROR("write padding to file");
1557 * Close the current index file if it is open, and create a new one.
1559 * Return 0 on success, -1 on error.
1562 int create_rotate_index_file(struct relay_stream
*stream
,
1563 const char *stream_path
)
1566 uint32_t major
, minor
;
1568 /* Put ref on previous index_file. */
1569 if (stream
->index_file
) {
1570 lttng_index_file_put(stream
->index_file
);
1571 stream
->index_file
= NULL
;
1573 major
= stream
->trace
->session
->major
;
1574 minor
= stream
->trace
->session
->minor
;
1575 stream
->index_file
= lttng_index_file_create(stream_path
,
1576 stream
->channel_name
,
1577 -1, -1, stream
->tracefile_size
,
1578 tracefile_array_get_file_index_head(stream
->tfa
),
1579 lttng_to_index_major(major
, minor
),
1580 lttng_to_index_minor(major
, minor
));
1581 if (!stream
->index_file
) {
1593 int do_rotate_stream_data(struct relay_stream
*stream
)
1597 DBG("Rotating stream %" PRIu64
" data file",
1598 stream
->stream_handle
);
1599 /* Perform the stream rotation. */
1600 ret
= utils_rotate_stream_file(stream
->path_name
,
1601 stream
->channel_name
, stream
->tracefile_size
,
1602 stream
->tracefile_count
, -1,
1603 -1, stream
->stream_fd
->fd
,
1604 NULL
, &stream
->stream_fd
->fd
);
1606 ERR("Rotating stream output file");
1609 stream
->tracefile_size_current
= 0;
1610 stream
->pos_after_last_complete_data_index
= 0;
1611 stream
->data_rotated
= true;
1613 if (stream
->data_rotated
&& stream
->index_rotated
) {
1614 /* Rotation completed; reset its state. */
1615 DBG("Rotation completed for stream %" PRIu64
,
1616 stream
->stream_handle
);
1617 stream
->rotate_at_seq_num
= -1ULL;
1618 stream
->data_rotated
= false;
1619 stream
->index_rotated
= false;
1626 * If too much data has been written in a tracefile before we received the
1627 * rotation command, we have to move the excess data to the new tracefile and
1628 * perform the rotation. This can happen because the control and data
1629 * connections are separate, the indexes as well as the commands arrive from
1630 * the control connection and we have no control over the order so we could be
1631 * in a situation where too much data has been received on the data connection
1632 * before the rotation command on the control connection arrives.
1635 int rotate_truncate_stream(struct relay_stream
*stream
)
1639 uint64_t diff
, pos
= 0;
1640 char buf
[FILE_COPY_BUFFER_SIZE
];
1642 assert(!stream
->is_metadata
);
1644 assert(stream
->tracefile_size_current
>
1645 stream
->pos_after_last_complete_data_index
);
1646 diff
= stream
->tracefile_size_current
-
1647 stream
->pos_after_last_complete_data_index
;
1649 /* Create the new tracefile. */
1650 new_fd
= utils_create_stream_file(stream
->path_name
,
1651 stream
->channel_name
,
1652 stream
->tracefile_size
, stream
->tracefile_count
,
1653 /* uid */ -1, /* gid */ -1, /* suffix */ NULL
);
1655 ERR("Failed to create new stream file at path %s for channel %s",
1656 stream
->path_name
, stream
->channel_name
);
1662 * Rewind the current tracefile to the position at which the rotation
1663 * should have occurred.
1665 lseek_ret
= lseek(stream
->stream_fd
->fd
,
1666 stream
->pos_after_last_complete_data_index
, SEEK_SET
);
1667 if (lseek_ret
< 0) {
1668 PERROR("seek truncate stream");
1673 /* Move data from the old file to the new file. */
1674 while (pos
< diff
) {
1675 uint64_t count
, bytes_left
;
1678 bytes_left
= diff
- pos
;
1679 count
= bytes_left
> sizeof(buf
) ? sizeof(buf
) : bytes_left
;
1680 assert(count
<= SIZE_MAX
);
1682 io_ret
= lttng_read(stream
->stream_fd
->fd
, buf
, count
);
1683 if (io_ret
< (ssize_t
) count
) {
1684 char error_string
[256];
1686 snprintf(error_string
, sizeof(error_string
),
1687 "Failed to read %" PRIu64
" bytes from fd %i in rotate_truncate_stream(), returned %zi",
1688 count
, stream
->stream_fd
->fd
, io_ret
);
1690 PERROR("%s", error_string
);
1692 ERR("%s", error_string
);
1698 io_ret
= lttng_write(new_fd
, buf
, count
);
1699 if (io_ret
< (ssize_t
) count
) {
1700 char error_string
[256];
1702 snprintf(error_string
, sizeof(error_string
),
1703 "Failed to write %" PRIu64
" bytes from fd %i in rotate_truncate_stream(), returned %zi",
1704 count
, new_fd
, io_ret
);
1706 PERROR("%s", error_string
);
1708 ERR("%s", error_string
);
1717 /* Truncate the file to get rid of the excess data. */
1718 ret
= ftruncate(stream
->stream_fd
->fd
,
1719 stream
->pos_after_last_complete_data_index
);
1721 PERROR("ftruncate");
1725 ret
= close(stream
->stream_fd
->fd
);
1727 PERROR("Closing tracefile");
1732 * Update the offset and FD of all the eventual indexes created by the
1733 * data connection before the rotation command arrived.
1735 ret
= relay_index_switch_all_files(stream
);
1737 ERR("Failed to rotate index file");
1741 stream
->stream_fd
->fd
= new_fd
;
1742 stream
->tracefile_size_current
= diff
;
1743 stream
->pos_after_last_complete_data_index
= 0;
1744 stream
->rotate_at_seq_num
= -1ULL;
1753 * Check if a stream's index file should be rotated (for session rotation).
1754 * Must be called with the stream lock held.
1756 * Return 0 on success, a negative value on error.
1759 int try_rotate_stream_index(struct relay_stream
*stream
)
1763 if (stream
->rotate_at_seq_num
== -1ULL) {
1764 /* No rotation expected. */
1768 if (stream
->index_rotated
) {
1769 /* Rotation of the index has already occurred. */
1773 if (stream
->prev_index_seq
== -1ULL ||
1774 stream
->prev_index_seq
< stream
->rotate_at_seq_num
) {
1775 DBG("Stream %" PRIu64
" index not yet ready for rotation (rotate_at_seq_num = %" PRIu64
", prev_index_seq = %" PRIu64
")",
1776 stream
->stream_handle
,
1777 stream
->rotate_at_seq_num
,
1778 stream
->prev_index_seq
);
1780 } else if (stream
->prev_index_seq
!= stream
->rotate_at_seq_num
) {
1782 * Unexpected, protocol error/bug.
1783 * It could mean that we received a rotation position
1784 * that is in the past.
1786 ERR("Stream %" PRIu64
" index is in an inconsistent state (rotate_at_seq_num = %" PRIu64
", prev_data_seq = %" PRIu64
", prev_index_seq = %" PRIu64
")",
1787 stream
->stream_handle
,
1788 stream
->rotate_at_seq_num
,
1789 stream
->prev_data_seq
,
1790 stream
->prev_index_seq
);
1794 DBG("Rotating stream %" PRIu64
" index file",
1795 stream
->stream_handle
);
1796 ret
= create_rotate_index_file(stream
, stream
->path_name
);
1797 stream
->index_rotated
= true;
1799 if (stream
->data_rotated
&& stream
->index_rotated
) {
1800 /* Rotation completed; reset its state. */
1801 DBG("Rotation completed for stream %" PRIu64
,
1802 stream
->stream_handle
);
1803 stream
->rotate_at_seq_num
= -1ULL;
1804 stream
->data_rotated
= false;
1805 stream
->index_rotated
= false;
1814 * Check if a stream's data file (as opposed to index) should be rotated
1815 * (for session rotation).
1816 * Must be called with the stream lock held.
1818 * Return 0 on success, a negative value on error.
1821 int try_rotate_stream_data(struct relay_stream
*stream
)
1825 if (stream
->rotate_at_seq_num
== -1ULL) {
1826 /* No rotation expected. */
1830 if (stream
->data_rotated
) {
1831 /* Rotation of the data file has already occurred. */
1835 if (stream
->prev_data_seq
== -1ULL ||
1836 stream
->prev_data_seq
< stream
->rotate_at_seq_num
) {
1837 DBG("Stream %" PRIu64
" not yet ready for rotation (rotate_at_seq_num = %" PRIu64
", prev_data_seq = %" PRIu64
")",
1838 stream
->stream_handle
,
1839 stream
->rotate_at_seq_num
,
1840 stream
->prev_data_seq
);
1842 } else if (stream
->prev_data_seq
> stream
->rotate_at_seq_num
) {
1844 * prev_data_seq is checked here since indexes and rotation
1845 * commands are serialized with respect to each other.
1847 DBG("Rotation after too much data has been written in tracefile "
1848 "for stream %" PRIu64
", need to truncate before "
1849 "rotating", stream
->stream_handle
);
1850 ret
= rotate_truncate_stream(stream
);
1852 ERR("Failed to truncate stream");
1855 } else if (stream
->prev_data_seq
!= stream
->rotate_at_seq_num
) {
1857 * Unexpected, protocol error/bug.
1858 * It could mean that we received a rotation position
1859 * that is in the past.
1861 ERR("Stream %" PRIu64
" data is in an inconsistent state (rotate_at_seq_num = %" PRIu64
", prev_data_seq = %" PRIu64
")",
1862 stream
->stream_handle
,
1863 stream
->rotate_at_seq_num
,
1864 stream
->prev_data_seq
);
1868 ret
= do_rotate_stream_data(stream
);
1876 * relay_recv_metadata: receive the metadata for the session.
1878 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1879 struct relay_connection
*conn
,
1880 const struct lttng_buffer_view
*payload
)
1884 struct relay_session
*session
= conn
->session
;
1885 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1886 struct relay_stream
*metadata_stream
;
1887 uint64_t metadata_payload_size
;
1890 ERR("Metadata sent before version check");
1895 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1896 ERR("Incorrect data size");
1900 metadata_payload_size
= recv_hdr
->data_size
-
1901 sizeof(struct lttcomm_relayd_metadata_payload
);
1903 memcpy(&metadata_payload_header
, payload
->data
,
1904 sizeof(metadata_payload_header
));
1905 metadata_payload_header
.stream_id
= be64toh(
1906 metadata_payload_header
.stream_id
);
1907 metadata_payload_header
.padding_size
= be32toh(
1908 metadata_payload_header
.padding_size
);
1910 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1911 if (!metadata_stream
) {
1916 pthread_mutex_lock(&metadata_stream
->lock
);
1918 size_ret
= lttng_write(metadata_stream
->stream_fd
->fd
,
1919 payload
->data
+ sizeof(metadata_payload_header
),
1920 metadata_payload_size
);
1921 if (size_ret
< metadata_payload_size
) {
1922 ERR("Relay error writing metadata on file");
1927 size_ret
= write_padding_to_file(metadata_stream
->stream_fd
->fd
,
1928 metadata_payload_header
.padding_size
);
1929 if (size_ret
< (int64_t) metadata_payload_header
.padding_size
) {
1934 metadata_stream
->metadata_received
+=
1935 metadata_payload_size
+ metadata_payload_header
.padding_size
;
1936 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1937 metadata_stream
->metadata_received
);
1939 ret
= try_rotate_stream_data(metadata_stream
);
1945 pthread_mutex_unlock(&metadata_stream
->lock
);
1946 stream_put(metadata_stream
);
1952 * relay_send_version: send relayd version number
1954 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1955 struct relay_connection
*conn
,
1956 const struct lttng_buffer_view
*payload
)
1960 struct lttcomm_relayd_version reply
, msg
;
1961 bool compatible
= true;
1963 conn
->version_check_done
= true;
1965 /* Get version from the other side. */
1966 if (payload
->size
< sizeof(msg
)) {
1967 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1968 sizeof(msg
), payload
->size
);
1973 memcpy(&msg
, payload
->data
, sizeof(msg
));
1974 msg
.major
= be32toh(msg
.major
);
1975 msg
.minor
= be32toh(msg
.minor
);
1977 memset(&reply
, 0, sizeof(reply
));
1978 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1979 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1981 /* Major versions must be the same */
1982 if (reply
.major
!= msg
.major
) {
1983 DBG("Incompatible major versions (%u vs %u), deleting session",
1984 reply
.major
, msg
.major
);
1988 conn
->major
= reply
.major
;
1989 /* We adapt to the lowest compatible version */
1990 if (reply
.minor
<= msg
.minor
) {
1991 conn
->minor
= reply
.minor
;
1993 conn
->minor
= msg
.minor
;
1996 reply
.major
= htobe32(reply
.major
);
1997 reply
.minor
= htobe32(reply
.minor
);
1998 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2000 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2001 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2014 DBG("Version check done using protocol %u.%u", conn
->major
,
2022 * Check for data pending for a given stream id from the session daemon.
2024 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2025 struct relay_connection
*conn
,
2026 const struct lttng_buffer_view
*payload
)
2028 struct relay_session
*session
= conn
->session
;
2029 struct lttcomm_relayd_data_pending msg
;
2030 struct lttcomm_relayd_generic_reply reply
;
2031 struct relay_stream
*stream
;
2034 uint64_t stream_seq
;
2036 DBG("Data pending command received");
2038 if (!session
|| !conn
->version_check_done
) {
2039 ERR("Trying to check for data before version check");
2041 goto end_no_session
;
2044 if (payload
->size
< sizeof(msg
)) {
2045 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2046 sizeof(msg
), payload
->size
);
2048 goto end_no_session
;
2050 memcpy(&msg
, payload
->data
, sizeof(msg
));
2051 msg
.stream_id
= be64toh(msg
.stream_id
);
2052 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
2054 stream
= stream_get_by_id(msg
.stream_id
);
2055 if (stream
== NULL
) {
2060 pthread_mutex_lock(&stream
->lock
);
2062 if (session_streams_have_index(session
)) {
2064 * Ensure that both the index and stream data have been
2065 * flushed up to the requested point.
2067 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2069 stream_seq
= stream
->prev_data_seq
;
2071 DBG("Data pending for stream id %" PRIu64
": prev_data_seq %" PRIu64
2072 ", prev_index_seq %" PRIu64
2073 ", and last_seq %" PRIu64
, msg
.stream_id
,
2074 stream
->prev_data_seq
, stream
->prev_index_seq
,
2075 msg
.last_net_seq_num
);
2077 /* Avoid wrapping issue */
2078 if (((int64_t) (stream_seq
- msg
.last_net_seq_num
)) >= 0) {
2079 /* Data has in fact been written and is NOT pending */
2082 /* Data still being streamed thus pending */
2086 stream
->data_pending_check_done
= true;
2087 pthread_mutex_unlock(&stream
->lock
);
2092 memset(&reply
, 0, sizeof(reply
));
2093 reply
.ret_code
= htobe32(ret
);
2094 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2095 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2096 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2106 * Wait for the control socket to reach a quiescent state.
2108 * Note that for now, when receiving this command from the session
2109 * daemon, this means that every subsequent commands or data received on
2110 * the control socket has been handled. So, this is why we simply return
2113 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
2114 struct relay_connection
*conn
,
2115 const struct lttng_buffer_view
*payload
)
2119 struct relay_stream
*stream
;
2120 struct lttcomm_relayd_quiescent_control msg
;
2121 struct lttcomm_relayd_generic_reply reply
;
2123 DBG("Checking quiescent state on control socket");
2125 if (!conn
->session
|| !conn
->version_check_done
) {
2126 ERR("Trying to check for data before version check");
2128 goto end_no_session
;
2131 if (payload
->size
< sizeof(msg
)) {
2132 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2133 sizeof(msg
), payload
->size
);
2135 goto end_no_session
;
2137 memcpy(&msg
, payload
->data
, sizeof(msg
));
2138 msg
.stream_id
= be64toh(msg
.stream_id
);
2140 stream
= stream_get_by_id(msg
.stream_id
);
2144 pthread_mutex_lock(&stream
->lock
);
2145 stream
->data_pending_check_done
= true;
2146 pthread_mutex_unlock(&stream
->lock
);
2148 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
2151 memset(&reply
, 0, sizeof(reply
));
2152 reply
.ret_code
= htobe32(LTTNG_OK
);
2153 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2154 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2155 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2167 * Initialize a data pending command. This means that a consumer is about
2168 * to ask for data pending for each stream it holds. Simply iterate over
2169 * all streams of a session and set the data_pending_check_done flag.
2171 * This command returns to the client a LTTNG_OK code.
2173 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2174 struct relay_connection
*conn
,
2175 const struct lttng_buffer_view
*payload
)
2179 struct lttng_ht_iter iter
;
2180 struct lttcomm_relayd_begin_data_pending msg
;
2181 struct lttcomm_relayd_generic_reply reply
;
2182 struct relay_stream
*stream
;
2187 DBG("Init streams for data pending");
2189 if (!conn
->session
|| !conn
->version_check_done
) {
2190 ERR("Trying to check for data before version check");
2192 goto end_no_session
;
2195 if (payload
->size
< sizeof(msg
)) {
2196 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2197 sizeof(msg
), payload
->size
);
2199 goto end_no_session
;
2201 memcpy(&msg
, payload
->data
, sizeof(msg
));
2202 msg
.session_id
= be64toh(msg
.session_id
);
2205 * Iterate over all streams to set the begin data pending flag.
2206 * For now, the streams are indexed by stream handle so we have
2207 * to iterate over all streams to find the one associated with
2208 * the right session_id.
2211 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2213 if (!stream_get(stream
)) {
2216 if (stream
->trace
->session
->id
== msg
.session_id
) {
2217 pthread_mutex_lock(&stream
->lock
);
2218 stream
->data_pending_check_done
= false;
2219 pthread_mutex_unlock(&stream
->lock
);
2220 DBG("Set begin data pending flag to stream %" PRIu64
,
2221 stream
->stream_handle
);
2227 memset(&reply
, 0, sizeof(reply
));
2228 /* All good, send back reply. */
2229 reply
.ret_code
= htobe32(LTTNG_OK
);
2231 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2232 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2233 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2245 * End data pending command. This will check, for a given session id, if
2246 * each stream associated with it has its data_pending_check_done flag
2247 * set. If not, this means that the client lost track of the stream but
2248 * the data is still being streamed on our side. In this case, we inform
2249 * the client that data is in flight.
2251 * Return to the client if there is data in flight or not with a ret_code.
2253 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2254 struct relay_connection
*conn
,
2255 const struct lttng_buffer_view
*payload
)
2259 struct lttng_ht_iter iter
;
2260 struct lttcomm_relayd_end_data_pending msg
;
2261 struct lttcomm_relayd_generic_reply reply
;
2262 struct relay_stream
*stream
;
2263 uint32_t is_data_inflight
= 0;
2265 DBG("End data pending command");
2267 if (!conn
->session
|| !conn
->version_check_done
) {
2268 ERR("Trying to check for data before version check");
2270 goto end_no_session
;
2273 if (payload
->size
< sizeof(msg
)) {
2274 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2275 sizeof(msg
), payload
->size
);
2277 goto end_no_session
;
2279 memcpy(&msg
, payload
->data
, sizeof(msg
));
2280 msg
.session_id
= be64toh(msg
.session_id
);
2283 * Iterate over all streams to see if the begin data pending
2287 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2289 if (!stream_get(stream
)) {
2292 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2296 pthread_mutex_lock(&stream
->lock
);
2297 if (!stream
->data_pending_check_done
) {
2298 uint64_t stream_seq
;
2300 if (session_streams_have_index(conn
->session
)) {
2302 * Ensure that both the index and stream data have been
2303 * flushed up to the requested point.
2305 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2307 stream_seq
= stream
->prev_data_seq
;
2309 if (!stream
->closed
|| !(((int64_t) (stream_seq
- stream
->last_net_seq_num
)) >= 0)) {
2310 is_data_inflight
= 1;
2311 DBG("Data is still in flight for stream %" PRIu64
,
2312 stream
->stream_handle
);
2313 pthread_mutex_unlock(&stream
->lock
);
2318 pthread_mutex_unlock(&stream
->lock
);
2323 memset(&reply
, 0, sizeof(reply
));
2324 /* All good, send back reply. */
2325 reply
.ret_code
= htobe32(is_data_inflight
);
2327 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2328 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2329 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2341 * Receive an index for a specific stream.
2343 * Return 0 on success else a negative value.
2345 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2346 struct relay_connection
*conn
,
2347 const struct lttng_buffer_view
*payload
)
2351 struct relay_session
*session
= conn
->session
;
2352 struct lttcomm_relayd_index index_info
;
2353 struct relay_index
*index
;
2354 struct lttcomm_relayd_generic_reply reply
;
2355 struct relay_stream
*stream
;
2360 DBG("Relay receiving index");
2362 if (!session
|| !conn
->version_check_done
) {
2363 ERR("Trying to close a stream before version check");
2365 goto end_no_session
;
2368 msg_len
= lttcomm_relayd_index_len(
2369 lttng_to_index_major(conn
->major
, conn
->minor
),
2370 lttng_to_index_minor(conn
->major
, conn
->minor
));
2371 if (payload
->size
< msg_len
) {
2372 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2373 msg_len
, payload
->size
);
2375 goto end_no_session
;
2377 memcpy(&index_info
, payload
->data
, msg_len
);
2378 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2379 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2380 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2381 index_info
.content_size
= be64toh(index_info
.content_size
);
2382 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2383 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2384 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2385 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2387 if (conn
->minor
>= 8) {
2388 index_info
.stream_instance_id
=
2389 be64toh(index_info
.stream_instance_id
);
2390 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2393 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2395 ERR("stream_get_by_id not found");
2399 pthread_mutex_lock(&stream
->lock
);
2401 /* Live beacon handling */
2402 if (index_info
.packet_size
== 0) {
2403 DBG("Received live beacon for stream %" PRIu64
,
2404 stream
->stream_handle
);
2407 * Only flag a stream inactive when it has already
2408 * received data and no indexes are in flight.
2410 if (stream
->index_received_seqcount
> 0
2411 && stream
->indexes_in_flight
== 0) {
2412 stream
->beacon_ts_end
= index_info
.timestamp_end
;
2415 goto end_stream_put
;
2417 stream
->beacon_ts_end
= -1ULL;
2420 if (stream
->ctf_stream_id
== -1ULL) {
2421 stream
->ctf_stream_id
= index_info
.stream_id
;
2423 index
= relay_index_get_by_id_or_create(stream
, index_info
.net_seq_num
);
2426 ERR("relay_index_get_by_id_or_create index NULL");
2427 goto end_stream_put
;
2429 if (set_index_control_data(index
, &index_info
, conn
)) {
2430 ERR("set_index_control_data error");
2431 relay_index_put(index
);
2433 goto end_stream_put
;
2435 ret
= relay_index_try_flush(index
);
2437 tracefile_array_commit_seq(stream
->tfa
);
2438 stream
->index_received_seqcount
++;
2439 stream
->pos_after_last_complete_data_index
+= index
->total_size
;
2440 stream
->prev_index_seq
= index_info
.net_seq_num
;
2442 ret
= try_rotate_stream_index(stream
);
2444 goto end_stream_put
;
2446 } else if (ret
> 0) {
2453 * relay_index_try_flush is responsible for the self-reference
2454 * put of the index object on error.
2456 ERR("relay_index_try_flush error %d", ret
);
2461 pthread_mutex_unlock(&stream
->lock
);
2466 memset(&reply
, 0, sizeof(reply
));
2468 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2470 reply
.ret_code
= htobe32(LTTNG_OK
);
2472 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2473 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2474 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2483 * Receive the streams_sent message.
2485 * Return 0 on success else a negative value.
2487 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2488 struct relay_connection
*conn
,
2489 const struct lttng_buffer_view
*payload
)
2493 struct lttcomm_relayd_generic_reply reply
;
2497 DBG("Relay receiving streams_sent");
2499 if (!conn
->session
|| !conn
->version_check_done
) {
2500 ERR("Trying to close a stream before version check");
2502 goto end_no_session
;
2506 * Publish every pending stream in the connection recv list which are
2507 * now ready to be used by the viewer.
2509 publish_connection_local_streams(conn
);
2511 memset(&reply
, 0, sizeof(reply
));
2512 reply
.ret_code
= htobe32(LTTNG_OK
);
2513 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2514 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2515 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2528 * relay_rotate_session_stream: rotate a stream to a new tracefile for the session
2529 * rotation feature (not the tracefile rotation feature).
2531 static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
2532 struct relay_connection
*conn
,
2533 const struct lttng_buffer_view
*payload
)
2537 struct relay_session
*session
= conn
->session
;
2538 struct lttcomm_relayd_rotate_stream stream_info
;
2539 struct lttcomm_relayd_generic_reply reply
;
2540 struct relay_stream
*stream
;
2543 struct lttng_buffer_view new_path_view
;
2545 if (!session
|| !conn
->version_check_done
) {
2546 ERR("Trying to rotate a stream before version check");
2551 if (session
->major
== 2 && session
->minor
< 11) {
2552 ERR("Unsupported feature before 2.11");
2557 header_len
= sizeof(struct lttcomm_relayd_rotate_stream
);
2559 if (payload
->size
< header_len
) {
2560 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2561 header_len
, payload
->size
);
2566 memcpy(&stream_info
, payload
->data
, header_len
);
2568 /* Convert to host */
2569 stream_info
.pathname_length
= be32toh(stream_info
.pathname_length
);
2570 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
2571 stream_info
.new_chunk_id
= be64toh(stream_info
.new_chunk_id
);
2572 stream_info
.rotate_at_seq_num
= be64toh(stream_info
.rotate_at_seq_num
);
2574 path_len
= stream_info
.pathname_length
;
2575 if (payload
->size
< header_len
+ path_len
) {
2576 ERR("Unexpected payload size in \"relay_rotate_session_stream\" including path: expected >= %zu bytes, got %zu bytes",
2577 header_len
+ path_len
, payload
->size
);
2582 /* Ensure it fits in local filename length. */
2583 if (path_len
>= LTTNG_PATH_MAX
) {
2584 ret
= -ENAMETOOLONG
;
2585 ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes",
2586 path_len
, LTTNG_PATH_MAX
);
2590 new_path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2591 stream_info
.pathname_length
);
2593 stream
= stream_get_by_id(stream_info
.stream_id
);
2599 pthread_mutex_lock(&stream
->lock
);
2602 * Update the trace path (just the folder, the stream name does not
2605 free(stream
->prev_path_name
);
2606 stream
->prev_path_name
= stream
->path_name
;
2607 stream
->path_name
= create_output_path(new_path_view
.data
);
2608 if (!stream
->path_name
) {
2609 ERR("Failed to create a new output path");
2611 goto end_stream_unlock
;
2613 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
,
2616 ERR("relay creating output directory");
2618 goto end_stream_unlock
;
2621 assert(stream
->current_chunk_id
.is_set
);
2622 stream
->current_chunk_id
.value
= stream_info
.new_chunk_id
;
2624 if (stream
->is_metadata
) {
2626 * Metadata streams have no index; consider its rotation
2629 stream
->index_rotated
= true;
2631 * The metadata stream is sent only over the control connection
2632 * so we know we have all the data to perform the stream
2635 ret
= do_rotate_stream_data(stream
);
2637 stream
->rotate_at_seq_num
= stream_info
.rotate_at_seq_num
;
2638 ret
= try_rotate_stream_data(stream
);
2640 goto end_stream_unlock
;
2643 ret
= try_rotate_stream_index(stream
);
2645 goto end_stream_unlock
;
2650 pthread_mutex_unlock(&stream
->lock
);
2653 memset(&reply
, 0, sizeof(reply
));
2655 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2657 reply
.ret_code
= htobe32(LTTNG_OK
);
2659 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2660 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2661 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2662 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2671 static int init_session_output_directory_handle(struct relay_session
*session
,
2672 struct lttng_directory_handle
*handle
)
2675 /* hostname/session_name */
2676 char *session_directory
= NULL
;
2678 * base path + session_directory
2679 * e.g. /home/user/lttng-traces/hostname/session_name
2681 char *full_session_path
= NULL
;
2683 pthread_mutex_lock(&session
->lock
);
2684 ret
= asprintf(&session_directory
, "%s/%s", session
->hostname
,
2685 session
->session_name
);
2686 pthread_mutex_unlock(&session
->lock
);
2688 PERROR("Failed to format session directory name");
2692 full_session_path
= create_output_path(session_directory
);
2693 if (!full_session_path
) {
2698 ret
= utils_mkdir_recursive(
2699 full_session_path
, S_IRWXU
| S_IRWXG
, -1, -1);
2701 ERR("Failed to create session output path \"%s\"",
2706 ret
= lttng_directory_handle_init(handle
, full_session_path
);
2711 free(session_directory
);
2712 free(full_session_path
);
2717 * relay_create_trace_chunk: create a new trace chunk
2719 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2720 struct relay_connection
*conn
,
2721 const struct lttng_buffer_view
*payload
)
2725 struct relay_session
*session
= conn
->session
;
2726 struct lttcomm_relayd_create_trace_chunk
*msg
;
2727 struct lttcomm_relayd_generic_reply reply
= {};
2728 struct lttng_buffer_view header_view
;
2729 struct lttng_buffer_view chunk_name_view
;
2730 struct lttng_trace_chunk
*chunk
= NULL
, *published_chunk
= NULL
;
2731 enum lttng_error_code reply_code
= LTTNG_OK
;
2732 enum lttng_trace_chunk_status chunk_status
;
2733 struct lttng_directory_handle session_output
;
2735 if (!session
|| !conn
->version_check_done
) {
2736 ERR("Trying to create a trace chunk before version check");
2741 if (session
->major
== 2 && session
->minor
< 11) {
2742 ERR("Chunk creation command is unsupported before 2.11");
2747 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2748 if (!header_view
.data
) {
2749 ERR("Failed to receive payload of chunk creation command");
2754 /* Convert to host endianness. */
2755 msg
= (typeof(msg
)) header_view
.data
;
2756 msg
->chunk_id
= be64toh(msg
->chunk_id
);
2757 msg
->creation_timestamp
= be64toh(msg
->creation_timestamp
);
2758 msg
->override_name_length
= be32toh(msg
->override_name_length
);
2760 chunk
= lttng_trace_chunk_create(
2761 msg
->chunk_id
, msg
->creation_timestamp
);
2763 ERR("Failed to create trace chunk in trace chunk creation command");
2765 reply_code
= LTTNG_ERR_NOMEM
;
2769 if (msg
->override_name_length
) {
2772 chunk_name_view
= lttng_buffer_view_from_view(payload
,
2774 msg
->override_name_length
);
2775 name
= chunk_name_view
.data
;
2776 if (!name
|| name
[msg
->override_name_length
- 1]) {
2777 ERR("Failed to receive payload of chunk creation command");
2779 reply_code
= LTTNG_ERR_INVALID
;
2783 chunk_status
= lttng_trace_chunk_override_name(
2784 chunk
, chunk_name_view
.data
);
2785 switch (chunk_status
) {
2786 case LTTNG_TRACE_CHUNK_STATUS_OK
:
2788 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT
:
2789 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2790 reply_code
= LTTNG_ERR_INVALID
;
2794 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2795 reply_code
= LTTNG_ERR_UNK
;
2801 ret
= init_session_output_directory_handle(
2802 conn
->session
, &session_output
);
2804 reply_code
= LTTNG_ERR_CREATE_DIR_FAIL
;
2808 chunk_status
= lttng_trace_chunk_set_credentials_current_user(chunk
);
2809 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2810 reply_code
= LTTNG_ERR_UNK
;
2815 chunk_status
= lttng_trace_chunk_set_as_owner(chunk
, &session_output
);
2816 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2817 reply_code
= LTTNG_ERR_UNK
;
2822 published_chunk
= sessiond_trace_chunk_registry_publish_chunk(
2823 sessiond_trace_chunk_registry
,
2824 conn
->session
->sessiond_uuid
,
2827 if (!published_chunk
) {
2828 char uuid_str
[UUID_STR_LEN
];
2830 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2831 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2836 reply_code
= LTTNG_ERR_NOMEM
;
2840 pthread_mutex_lock(&conn
->session
->lock
);
2841 lttng_trace_chunk_put(conn
->session
->current_trace_chunk
);
2842 conn
->session
->current_trace_chunk
= published_chunk
;
2843 pthread_mutex_unlock(&conn
->session
->lock
);
2844 published_chunk
= NULL
;
2847 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2848 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2850 sizeof(struct lttcomm_relayd_generic_reply
),
2852 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2853 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2858 lttng_trace_chunk_put(chunk
);
2859 lttng_trace_chunk_put(published_chunk
);
2860 lttng_directory_handle_fini(&session_output
);
2865 * relay_close_trace_chunk: close a trace chunk
2867 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2868 struct relay_connection
*conn
,
2869 const struct lttng_buffer_view
*payload
)
2873 struct relay_session
*session
= conn
->session
;
2874 struct lttcomm_relayd_close_trace_chunk
*msg
;
2875 struct lttcomm_relayd_generic_reply reply
= {};
2876 struct lttng_buffer_view header_view
;
2877 struct lttng_trace_chunk
*chunk
= NULL
;
2878 enum lttng_error_code reply_code
= LTTNG_OK
;
2879 enum lttng_trace_chunk_status chunk_status
;
2881 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type
) close_command
;
2882 time_t close_timestamp
;
2884 if (!session
|| !conn
->version_check_done
) {
2885 ERR("Trying to close a trace chunk before version check");
2890 if (session
->major
== 2 && session
->minor
< 11) {
2891 ERR("Chunk close command is unsupported before 2.11");
2896 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2897 if (!header_view
.data
) {
2898 ERR("Failed to receive payload of chunk close command");
2903 /* Convert to host endianness. */
2904 msg
= (typeof(msg
)) header_view
.data
;
2905 chunk_id
= be64toh(msg
->chunk_id
);
2906 close_timestamp
= (time_t) be64toh(msg
->close_timestamp
);
2907 close_command
= (typeof(close_command
)){
2908 .value
= be32toh(msg
->close_command
.value
),
2909 .is_set
= msg
->close_command
.is_set
,
2912 chunk
= sessiond_trace_chunk_registry_get_chunk(
2913 sessiond_trace_chunk_registry
,
2914 conn
->session
->sessiond_uuid
,
2918 char uuid_str
[UUID_STR_LEN
];
2920 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2921 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2926 reply_code
= LTTNG_ERR_NOMEM
;
2930 chunk_status
= lttng_trace_chunk_set_close_timestamp(
2931 chunk
, close_timestamp
);
2932 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2933 ERR("Failed to set trace chunk close timestamp");
2935 reply_code
= LTTNG_ERR_UNK
;
2939 if (close_command
.is_set
) {
2940 chunk_status
= lttng_trace_chunk_set_close_command(
2941 chunk
, close_command
.value
);
2942 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2944 reply_code
= LTTNG_ERR_INVALID
;
2950 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2951 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2953 sizeof(struct lttcomm_relayd_generic_reply
),
2955 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2956 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2961 lttng_trace_chunk_put(chunk
);
2965 #define DBG_CMD(cmd_name, conn) \
2966 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2968 static int relay_process_control_command(struct relay_connection
*conn
,
2969 const struct lttcomm_relayd_hdr
*header
,
2970 const struct lttng_buffer_view
*payload
)
2974 switch (header
->cmd
) {
2975 case RELAYD_CREATE_SESSION
:
2976 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
2977 ret
= relay_create_session(header
, conn
, payload
);
2979 case RELAYD_ADD_STREAM
:
2980 DBG_CMD("RELAYD_ADD_STREAM", conn
);
2981 ret
= relay_add_stream(header
, conn
, payload
);
2983 case RELAYD_START_DATA
:
2984 DBG_CMD("RELAYD_START_DATA", conn
);
2985 ret
= relay_start(header
, conn
, payload
);
2987 case RELAYD_SEND_METADATA
:
2988 DBG_CMD("RELAYD_SEND_METADATA", conn
);
2989 ret
= relay_recv_metadata(header
, conn
, payload
);
2991 case RELAYD_VERSION
:
2992 DBG_CMD("RELAYD_VERSION", conn
);
2993 ret
= relay_send_version(header
, conn
, payload
);
2995 case RELAYD_CLOSE_STREAM
:
2996 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
2997 ret
= relay_close_stream(header
, conn
, payload
);
2999 case RELAYD_DATA_PENDING
:
3000 DBG_CMD("RELAYD_DATA_PENDING", conn
);
3001 ret
= relay_data_pending(header
, conn
, payload
);
3003 case RELAYD_QUIESCENT_CONTROL
:
3004 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
3005 ret
= relay_quiescent_control(header
, conn
, payload
);
3007 case RELAYD_BEGIN_DATA_PENDING
:
3008 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
3009 ret
= relay_begin_data_pending(header
, conn
, payload
);
3011 case RELAYD_END_DATA_PENDING
:
3012 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
3013 ret
= relay_end_data_pending(header
, conn
, payload
);
3015 case RELAYD_SEND_INDEX
:
3016 DBG_CMD("RELAYD_SEND_INDEX", conn
);
3017 ret
= relay_recv_index(header
, conn
, payload
);
3019 case RELAYD_STREAMS_SENT
:
3020 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
3021 ret
= relay_streams_sent(header
, conn
, payload
);
3023 case RELAYD_RESET_METADATA
:
3024 DBG_CMD("RELAYD_RESET_METADATA", conn
);
3025 ret
= relay_reset_metadata(header
, conn
, payload
);
3027 case RELAYD_ROTATE_STREAM
:
3028 DBG_CMD("RELAYD_ROTATE_STREAM", conn
);
3029 ret
= relay_rotate_session_stream(header
, conn
, payload
);
3031 case RELAYD_CREATE_TRACE_CHUNK
:
3032 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn
);
3033 ret
= relay_create_trace_chunk(header
, conn
, payload
);
3035 case RELAYD_CLOSE_TRACE_CHUNK
:
3036 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn
);
3037 ret
= relay_close_trace_chunk(header
, conn
, payload
);
3039 case RELAYD_UPDATE_SYNC_INFO
:
3041 ERR("Received unknown command (%u)", header
->cmd
);
3042 relay_unknown_command(conn
);
3051 static enum relay_connection_status
relay_process_control_receive_payload(
3052 struct relay_connection
*conn
)
3055 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3056 struct lttng_dynamic_buffer
*reception_buffer
=
3057 &conn
->protocol
.ctrl
.reception_buffer
;
3058 struct ctrl_connection_state_receive_payload
*state
=
3059 &conn
->protocol
.ctrl
.state
.receive_payload
;
3060 struct lttng_buffer_view payload_view
;
3062 if (state
->left_to_receive
== 0) {
3063 /* Short-circuit for payload-less commands. */
3064 goto reception_complete
;
3067 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3068 reception_buffer
->data
+ state
->received
,
3069 state
->left_to_receive
, MSG_DONTWAIT
);
3071 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3072 PERROR("Unable to receive command payload on sock %d",
3074 status
= RELAY_CONNECTION_STATUS_ERROR
;
3077 } else if (ret
== 0) {
3078 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3079 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3084 assert(ret
<= state
->left_to_receive
);
3086 state
->left_to_receive
-= ret
;
3087 state
->received
+= ret
;
3089 if (state
->left_to_receive
> 0) {
3091 * Can't transition to the protocol's next state, wait to
3092 * receive the rest of the header.
3094 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3095 state
->received
, state
->left_to_receive
,
3101 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
3102 conn
->sock
->fd
, state
->received
);
3104 * The payload required to process the command has been received.
3105 * A view to the reception buffer is forwarded to the various
3106 * commands and the state of the control is reset on success.
3108 * Commands are responsible for sending their reply to the peer.
3110 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
3112 ret
= relay_process_control_command(conn
,
3113 &state
->header
, &payload_view
);
3115 status
= RELAY_CONNECTION_STATUS_ERROR
;
3119 ret
= connection_reset_protocol_state(conn
);
3121 status
= RELAY_CONNECTION_STATUS_ERROR
;
3127 static enum relay_connection_status
relay_process_control_receive_header(
3128 struct relay_connection
*conn
)
3131 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3132 struct lttcomm_relayd_hdr header
;
3133 struct lttng_dynamic_buffer
*reception_buffer
=
3134 &conn
->protocol
.ctrl
.reception_buffer
;
3135 struct ctrl_connection_state_receive_header
*state
=
3136 &conn
->protocol
.ctrl
.state
.receive_header
;
3138 assert(state
->left_to_receive
!= 0);
3140 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3141 reception_buffer
->data
+ state
->received
,
3142 state
->left_to_receive
, MSG_DONTWAIT
);
3144 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3145 PERROR("Unable to receive control command header on sock %d",
3147 status
= RELAY_CONNECTION_STATUS_ERROR
;
3150 } else if (ret
== 0) {
3151 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3152 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3157 assert(ret
<= state
->left_to_receive
);
3159 state
->left_to_receive
-= ret
;
3160 state
->received
+= ret
;
3162 if (state
->left_to_receive
> 0) {
3164 * Can't transition to the protocol's next state, wait to
3165 * receive the rest of the header.
3167 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3168 state
->received
, state
->left_to_receive
,
3173 /* Transition to next state: receiving the command's payload. */
3174 conn
->protocol
.ctrl
.state_id
=
3175 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3176 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
3177 header
.circuit_id
= be64toh(header
.circuit_id
);
3178 header
.data_size
= be64toh(header
.data_size
);
3179 header
.cmd
= be32toh(header
.cmd
);
3180 header
.cmd_version
= be32toh(header
.cmd_version
);
3181 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
3182 &header
, sizeof(header
));
3184 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
3185 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
3188 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
3189 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
3191 status
= RELAY_CONNECTION_STATUS_ERROR
;
3195 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
3197 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
3198 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
3201 status
= RELAY_CONNECTION_STATUS_ERROR
;
3205 if (header
.data_size
== 0) {
3207 * Manually invoke the next state as the poll loop
3208 * will not wake-up to allow us to proceed further.
3210 status
= relay_process_control_receive_payload(conn
);
3217 * Process the commands received on the control socket
3219 static enum relay_connection_status
relay_process_control(
3220 struct relay_connection
*conn
)
3222 enum relay_connection_status status
;
3224 switch (conn
->protocol
.ctrl
.state_id
) {
3225 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
3226 status
= relay_process_control_receive_header(conn
);
3228 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3229 status
= relay_process_control_receive_payload(conn
);
3232 ERR("Unknown control connection protocol state encountered.");
3240 * Handle index for a data stream.
3242 * Called with the stream lock held.
3244 * Return 0 on success else a negative value.
3246 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
3247 bool rotate_index
, bool *flushed
, uint64_t total_size
)
3250 uint64_t data_offset
;
3251 struct relay_index
*index
;
3253 /* Get data offset because we are about to update the index. */
3254 data_offset
= htobe64(stream
->tracefile_size_current
);
3256 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
3257 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
3260 * Lookup for an existing index for that stream id/sequence
3261 * number. If it exists, the control thread has already received the
3262 * data for it, thus we need to write it to disk.
3264 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
3270 if (rotate_index
|| !stream
->index_file
) {
3271 const char *stream_path
;
3274 * The data connection creates the stream's first index file.
3276 * This can happen _after_ a ROTATE_STREAM command. In
3277 * other words, the data of the first packet of this stream
3278 * can be received after a ROTATE_STREAM command.
3280 * The ROTATE_STREAM command changes the stream's path_name
3281 * to point to the "next" chunk. If a rotation is pending for
3282 * this stream, as indicated by "rotate_at_seq_num != -1ULL",
3283 * it means that we are still receiving data that belongs in the
3284 * stream's former path.
3286 * In this very specific case, we must ensure that the index
3287 * file is created in the streams's former path,
3290 * All other rotations beyond the first one are not affected
3291 * by this problem since the actual rotation operation creates
3292 * the new chunk's index file.
3294 stream_path
= stream
->rotate_at_seq_num
== -1ULL ?
3296 stream
->prev_path_name
;
3298 ret
= create_rotate_index_file(stream
, stream_path
);
3300 ERR("Failed to rotate index");
3301 /* Put self-ref for this index due to error. */
3302 relay_index_put(index
);
3308 if (relay_index_set_file(index
, stream
->index_file
, data_offset
)) {
3310 /* Put self-ref for this index due to error. */
3311 relay_index_put(index
);
3316 ret
= relay_index_try_flush(index
);
3318 tracefile_array_commit_seq(stream
->tfa
);
3319 stream
->index_received_seqcount
++;
3321 } else if (ret
> 0) {
3322 index
->total_size
= total_size
;
3329 * relay_index_try_flush is responsible for the self-reference
3330 * put of the index object on error.
3332 ERR("relay_index_try_flush error %d", ret
);
3339 static enum relay_connection_status
relay_process_data_receive_header(
3340 struct relay_connection
*conn
)
3343 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3344 struct data_connection_state_receive_header
*state
=
3345 &conn
->protocol
.data
.state
.receive_header
;
3346 struct lttcomm_relayd_data_hdr header
;
3347 struct relay_stream
*stream
;
3349 assert(state
->left_to_receive
!= 0);
3351 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3352 state
->header_reception_buffer
+ state
->received
,
3353 state
->left_to_receive
, MSG_DONTWAIT
);
3355 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3356 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
3357 status
= RELAY_CONNECTION_STATUS_ERROR
;
3360 } else if (ret
== 0) {
3361 /* Orderly shutdown. Not necessary to print an error. */
3362 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3363 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3368 assert(ret
<= state
->left_to_receive
);
3370 state
->left_to_receive
-= ret
;
3371 state
->received
+= ret
;
3373 if (state
->left_to_receive
> 0) {
3375 * Can't transition to the protocol's next state, wait to
3376 * receive the rest of the header.
3378 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3379 state
->received
, state
->left_to_receive
,
3384 /* Transition to next state: receiving the payload. */
3385 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3387 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3388 header
.circuit_id
= be64toh(header
.circuit_id
);
3389 header
.stream_id
= be64toh(header
.stream_id
);
3390 header
.data_size
= be32toh(header
.data_size
);
3391 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3392 header
.padding_size
= be32toh(header
.padding_size
);
3393 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3395 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3397 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3398 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3400 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3401 conn
->sock
->fd
, header
.circuit_id
,
3402 header
.stream_id
, header
.data_size
,
3403 header
.net_seq_num
, header
.padding_size
);
3405 stream
= stream_get_by_id(header
.stream_id
);
3407 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3409 /* Protocol error. */
3410 status
= RELAY_CONNECTION_STATUS_ERROR
;
3414 pthread_mutex_lock(&stream
->lock
);
3416 /* Check if a rotation is needed. */
3417 if (stream
->tracefile_size
> 0 &&
3418 (stream
->tracefile_size_current
+ header
.data_size
) >
3419 stream
->tracefile_size
) {
3420 uint64_t old_id
, new_id
;
3422 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
3423 tracefile_array_file_rotate(stream
->tfa
);
3425 /* new_id is updated by utils_rotate_stream_file. */
3428 ret
= utils_rotate_stream_file(stream
->path_name
,
3429 stream
->channel_name
, stream
->tracefile_size
,
3430 stream
->tracefile_count
, -1,
3431 -1, stream
->stream_fd
->fd
,
3432 &new_id
, &stream
->stream_fd
->fd
);
3434 ERR("Failed to rotate stream output file");
3435 status
= RELAY_CONNECTION_STATUS_ERROR
;
3436 goto end_stream_unlock
;
3440 * Reset current size because we just performed a stream
3443 stream
->tracefile_size_current
= 0;
3444 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= true;
3448 pthread_mutex_unlock(&stream
->lock
);
3454 static enum relay_connection_status
relay_process_data_receive_payload(
3455 struct relay_connection
*conn
)
3458 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3459 struct relay_stream
*stream
;
3460 struct data_connection_state_receive_payload
*state
=
3461 &conn
->protocol
.data
.state
.receive_payload
;
3462 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3463 char data_buffer
[chunk_size
];
3464 bool partial_recv
= false;
3465 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3466 uint64_t left_to_receive
= state
->left_to_receive
;
3467 struct relay_session
*session
;
3469 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3470 state
->header
.stream_id
, state
->header
.net_seq_num
,
3471 state
->received
, left_to_receive
);
3473 stream
= stream_get_by_id(state
->header
.stream_id
);
3475 /* Protocol error. */
3476 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3477 state
->header
.stream_id
);
3478 status
= RELAY_CONNECTION_STATUS_ERROR
;
3482 pthread_mutex_lock(&stream
->lock
);
3483 session
= stream
->trace
->session
;
3484 if (!conn
->session
) {
3485 ret
= connection_set_session(conn
, session
);
3487 status
= RELAY_CONNECTION_STATUS_ERROR
;
3488 goto end_stream_unlock
;
3493 * The size of the "chunk" received on any iteration is bounded by:
3494 * - the data left to receive,
3495 * - the data immediately available on the socket,
3496 * - the on-stack data buffer
3498 while (left_to_receive
> 0 && !partial_recv
) {
3500 size_t recv_size
= min(left_to_receive
, chunk_size
);
3502 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3503 recv_size
, MSG_DONTWAIT
);
3505 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3506 PERROR("Socket %d error", conn
->sock
->fd
);
3507 status
= RELAY_CONNECTION_STATUS_ERROR
;
3509 goto end_stream_unlock
;
3510 } else if (ret
== 0) {
3511 /* No more data ready to be consumed on socket. */
3512 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3513 state
->header
.stream_id
);
3514 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3516 } else if (ret
< (int) recv_size
) {
3518 * All the data available on the socket has been
3521 partial_recv
= true;
3526 /* Write data to stream output fd. */
3527 write_ret
= lttng_write(stream
->stream_fd
->fd
, data_buffer
,
3529 if (write_ret
< (ssize_t
) recv_size
) {
3530 ERR("Relay error writing data to file");
3531 status
= RELAY_CONNECTION_STATUS_ERROR
;
3532 goto end_stream_unlock
;
3535 left_to_receive
-= recv_size
;
3536 state
->received
+= recv_size
;
3537 state
->left_to_receive
= left_to_receive
;
3539 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
3540 write_ret
, stream
->stream_handle
);
3543 if (state
->left_to_receive
> 0) {
3545 * Did not receive all the data expected, wait for more data to
3546 * become available on the socket.
3548 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3549 state
->header
.stream_id
, state
->received
,
3550 state
->left_to_receive
);
3551 goto end_stream_unlock
;
3554 ret
= write_padding_to_file(stream
->stream_fd
->fd
,
3555 state
->header
.padding_size
);
3556 if ((int64_t) ret
< (int64_t) state
->header
.padding_size
) {
3557 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3558 stream
->stream_handle
,
3559 state
->header
.net_seq_num
, ret
);
3560 status
= RELAY_CONNECTION_STATUS_ERROR
;
3561 goto end_stream_unlock
;
3565 if (session_streams_have_index(session
)) {
3566 ret
= handle_index_data(stream
, state
->header
.net_seq_num
,
3567 state
->rotate_index
, &index_flushed
, state
->header
.data_size
+ state
->header
.padding_size
);
3569 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3570 stream
->stream_handle
,
3571 state
->header
.net_seq_num
, ret
);
3572 status
= RELAY_CONNECTION_STATUS_ERROR
;
3573 goto end_stream_unlock
;
3577 stream
->tracefile_size_current
+= state
->header
.data_size
+
3578 state
->header
.padding_size
;
3580 if (stream
->prev_data_seq
== -1ULL) {
3583 if (index_flushed
) {
3584 stream
->pos_after_last_complete_data_index
=
3585 stream
->tracefile_size_current
;
3586 stream
->prev_index_seq
= state
->header
.net_seq_num
;
3587 ret
= try_rotate_stream_index(stream
);
3589 goto end_stream_unlock
;
3593 stream
->prev_data_seq
= state
->header
.net_seq_num
;
3596 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3597 * contents of *state which are aliased (union) to the same location as
3598 * the new state. Don't use it beyond this point.
3600 connection_reset_protocol_state(conn
);
3603 ret
= try_rotate_stream_data(stream
);
3605 status
= RELAY_CONNECTION_STATUS_ERROR
;
3606 goto end_stream_unlock
;
3610 close_requested
= stream
->close_requested
;
3611 pthread_mutex_unlock(&stream
->lock
);
3612 if (close_requested
&& left_to_receive
== 0) {
3613 try_stream_close(stream
);
3617 pthread_mutex_lock(&session
->lock
);
3618 uatomic_set(&session
->new_streams
, 1);
3619 pthread_mutex_unlock(&session
->lock
);
3628 * relay_process_data: Process the data received on the data socket
3630 static enum relay_connection_status
relay_process_data(
3631 struct relay_connection
*conn
)
3633 enum relay_connection_status status
;
3635 switch (conn
->protocol
.data
.state_id
) {
3636 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3637 status
= relay_process_data_receive_header(conn
);
3639 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3640 status
= relay_process_data_receive_payload(conn
);
3643 ERR("Unexpected data connection communication state.");
3650 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3654 (void) lttng_poll_del(events
, pollfd
);
3656 ret
= close(pollfd
);
3658 ERR("Closing pollfd %d", pollfd
);
3662 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3663 int pollfd
, struct relay_connection
*conn
)
3665 const char *type_str
;
3667 switch (conn
->type
) {
3672 type_str
= "Control";
3674 case RELAY_VIEWER_COMMAND
:
3675 type_str
= "Viewer Command";
3677 case RELAY_VIEWER_NOTIFICATION
:
3678 type_str
= "Viewer Notification";
3681 type_str
= "Unknown";
3683 cleanup_connection_pollfd(events
, pollfd
);
3684 connection_put(conn
);
3685 DBG("%s connection closed with %d", type_str
, pollfd
);
3689 * This thread does the actual work
3691 static void *relay_thread_worker(void *data
)
3693 int ret
, err
= -1, last_seen_data_fd
= -1;
3695 struct lttng_poll_event events
;
3696 struct lttng_ht
*relay_connections_ht
;
3697 struct lttng_ht_iter iter
;
3698 struct relay_connection
*destroy_conn
= NULL
;
3700 DBG("[thread] Relay worker started");
3702 rcu_register_thread();
3704 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3706 if (testpoint(relayd_thread_worker
)) {
3707 goto error_testpoint
;
3710 health_code_update();
3712 /* table of connections indexed on socket */
3713 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3714 if (!relay_connections_ht
) {
3715 goto relay_connections_ht_error
;
3718 ret
= create_thread_poll_set(&events
, 2);
3720 goto error_poll_create
;
3723 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3730 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3732 health_code_update();
3734 /* Infinite blocking call, waiting for transmission */
3735 DBG3("Relayd worker thread polling...");
3736 health_poll_entry();
3737 ret
= lttng_poll_wait(&events
, -1);
3741 * Restart interrupted system call.
3743 if (errno
== EINTR
) {
3752 * Process control. The control connection is
3753 * prioritized so we don't starve it with high
3754 * throughput tracing data on the data connection.
3756 for (i
= 0; i
< nb_fd
; i
++) {
3757 /* Fetch once the poll data */
3758 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3759 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3761 health_code_update();
3763 /* Thread quit pipe has been closed. Killing thread. */
3764 ret
= check_thread_quit_pipe(pollfd
, revents
);
3770 /* Inspect the relay conn pipe for new connection */
3771 if (pollfd
== relay_conn_pipe
[0]) {
3772 if (revents
& LPOLLIN
) {
3773 struct relay_connection
*conn
;
3775 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3779 lttng_poll_add(&events
, conn
->sock
->fd
,
3780 LPOLLIN
| LPOLLRDHUP
);
3781 connection_ht_add(relay_connections_ht
, conn
);
3782 DBG("Connection socket %d added", conn
->sock
->fd
);
3783 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3784 ERR("Relay connection pipe error");
3787 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3791 struct relay_connection
*ctrl_conn
;
3793 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3794 /* If not found, there is a synchronization issue. */
3797 if (ctrl_conn
->type
== RELAY_DATA
) {
3798 if (revents
& LPOLLIN
) {
3800 * Flag the last seen data fd not deleted. It will be
3801 * used as the last seen fd if any fd gets deleted in
3804 last_notdel_data_fd
= pollfd
;
3806 goto put_ctrl_connection
;
3808 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3810 if (revents
& LPOLLIN
) {
3811 enum relay_connection_status status
;
3813 status
= relay_process_control(ctrl_conn
);
3814 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3816 * On socket error flag the session as aborted to force
3817 * the cleanup of its stream otherwise it can leak
3818 * during the lifetime of the relayd.
3820 * This prevents situations in which streams can be
3821 * left opened because an index was received, the
3822 * control connection is closed, and the data
3823 * connection is closed (uncleanly) before the packet's
3826 * Since the control connection encountered an error,
3827 * it is okay to be conservative and close the
3828 * session right now as we can't rely on the protocol
3829 * being respected anymore.
3831 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3832 session_abort(ctrl_conn
->session
);
3835 /* Clear the connection on error or close. */
3836 relay_thread_close_connection(&events
,
3841 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3842 relay_thread_close_connection(&events
,
3844 if (last_seen_data_fd
== pollfd
) {
3845 last_seen_data_fd
= last_notdel_data_fd
;
3848 ERR("Unexpected poll events %u for control sock %d",
3850 connection_put(ctrl_conn
);
3853 put_ctrl_connection
:
3854 connection_put(ctrl_conn
);
3859 * The last loop handled a control request, go back to poll to make
3860 * sure we prioritise the control socket.
3866 if (last_seen_data_fd
>= 0) {
3867 for (i
= 0; i
< nb_fd
; i
++) {
3868 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3870 health_code_update();
3872 if (last_seen_data_fd
== pollfd
) {
3879 /* Process data connection. */
3880 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3881 /* Fetch the poll data. */
3882 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3883 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3884 struct relay_connection
*data_conn
;
3886 health_code_update();
3889 /* No activity for this FD (poll implementation). */
3893 /* Skip the command pipe. It's handled in the first loop. */
3894 if (pollfd
== relay_conn_pipe
[0]) {
3898 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3900 /* Skip it. Might be removed before. */
3903 if (data_conn
->type
== RELAY_CONTROL
) {
3904 goto put_data_connection
;
3906 assert(data_conn
->type
== RELAY_DATA
);
3908 if (revents
& LPOLLIN
) {
3909 enum relay_connection_status status
;
3911 status
= relay_process_data(data_conn
);
3912 /* Connection closed or error. */
3913 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3915 * On socket error flag the session as aborted to force
3916 * the cleanup of its stream otherwise it can leak
3917 * during the lifetime of the relayd.
3919 * This prevents situations in which streams can be
3920 * left opened because an index was received, the
3921 * control connection is closed, and the data
3922 * connection is closed (uncleanly) before the packet's
3925 * Since the data connection encountered an error,
3926 * it is okay to be conservative and close the
3927 * session right now as we can't rely on the protocol
3928 * being respected anymore.
3930 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3931 session_abort(data_conn
->session
);
3933 relay_thread_close_connection(&events
, pollfd
,
3936 * Every goto restart call sets the last seen fd where
3937 * here we don't really care since we gracefully
3938 * continue the loop after the connection is deleted.
3941 /* Keep last seen port. */
3942 last_seen_data_fd
= pollfd
;
3943 connection_put(data_conn
);
3946 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3947 relay_thread_close_connection(&events
, pollfd
,
3950 ERR("Unknown poll events %u for data sock %d",
3953 put_data_connection
:
3954 connection_put(data_conn
);
3956 last_seen_data_fd
= -1;
3959 /* Normal exit, no error */
3964 /* Cleanup remaining connection object. */
3966 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3969 health_code_update();
3971 session_abort(destroy_conn
->session
);
3974 * No need to grab another ref, because we own
3977 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
3982 lttng_poll_clean(&events
);
3984 lttng_ht_destroy(relay_connections_ht
);
3985 relay_connections_ht_error
:
3986 /* Close relay conn pipes */
3987 utils_close_pipe(relay_conn_pipe
);
3989 DBG("Thread exited with error");
3991 DBG("Worker thread cleanup complete");
3995 ERR("Health error occurred in %s", __func__
);
3997 health_unregister(health_relayd
);
3998 rcu_unregister_thread();
3999 lttng_relay_stop_threads();
4004 * Create the relay command pipe to wake thread_manage_apps.
4005 * Closed in cleanup().
4007 static int create_relay_conn_pipe(void)
4011 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
4019 int main(int argc
, char **argv
)
4021 int ret
= 0, retval
= 0;
4024 /* Parse arguments */
4026 if (set_options(argc
, argv
)) {
4031 if (set_signal_handler()) {
4036 /* Try to create directory if -o, --output is specified. */
4037 if (opt_output_path
) {
4038 if (*opt_output_path
!= '/') {
4039 ERR("Please specify an absolute path for -o, --output PATH");
4044 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
4047 ERR("Unable to create %s", opt_output_path
);
4054 if (opt_daemon
|| opt_background
) {
4057 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4065 * We are in the child. Make sure all other file
4066 * descriptors are closed, in case we are called with
4067 * more opened file descriptors than the standard ones.
4069 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4074 sessiond_trace_chunk_registry
= sessiond_trace_chunk_registry_create();
4075 if (!sessiond_trace_chunk_registry
) {
4076 ERR("Failed to initialize session daemon trace chunk registry");
4078 goto exit_sessiond_trace_chunk_registry
;
4081 /* Initialize thread health monitoring */
4082 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
4083 if (!health_relayd
) {
4084 PERROR("health_app_create error");
4086 goto exit_health_app_create
;
4089 /* Create thread quit pipe */
4090 if (init_thread_quit_pipe()) {
4092 goto exit_init_data
;
4095 /* Setup the thread apps communication pipe. */
4096 if (create_relay_conn_pipe()) {
4098 goto exit_init_data
;
4101 /* Init relay command queue. */
4102 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
4104 /* Initialize communication library */
4106 lttcomm_inet_init();
4108 /* tables of sessions indexed by session ID */
4109 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4112 goto exit_init_data
;
4115 /* tables of streams indexed by stream ID */
4116 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4117 if (!relay_streams_ht
) {
4119 goto exit_init_data
;
4122 /* tables of streams indexed by stream ID */
4123 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4124 if (!viewer_streams_ht
) {
4126 goto exit_init_data
;
4129 ret
= utils_create_pipe(health_quit_pipe
);
4132 goto exit_health_quit_pipe
;
4135 /* Create thread to manage the client socket */
4136 ret
= pthread_create(&health_thread
, default_pthread_attr(),
4137 thread_manage_health
, (void *) NULL
);
4140 PERROR("pthread_create health");
4142 goto exit_health_thread
;
4145 /* Setup the dispatcher thread */
4146 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
4147 relay_thread_dispatcher
, (void *) NULL
);
4150 PERROR("pthread_create dispatcher");
4152 goto exit_dispatcher_thread
;
4155 /* Setup the worker thread */
4156 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
4157 relay_thread_worker
, NULL
);
4160 PERROR("pthread_create worker");
4162 goto exit_worker_thread
;
4165 /* Setup the listener thread */
4166 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
4167 relay_thread_listener
, (void *) NULL
);
4170 PERROR("pthread_create listener");
4172 goto exit_listener_thread
;
4175 ret
= relayd_live_create(live_uri
);
4177 ERR("Starting live viewer threads");
4183 * This is where we start awaiting program completion (e.g. through
4184 * signal that asks threads to teardown).
4187 ret
= relayd_live_join();
4193 ret
= pthread_join(listener_thread
, &status
);
4196 PERROR("pthread_join listener_thread");
4200 exit_listener_thread
:
4201 ret
= pthread_join(worker_thread
, &status
);
4204 PERROR("pthread_join worker_thread");
4209 ret
= pthread_join(dispatcher_thread
, &status
);
4212 PERROR("pthread_join dispatcher_thread");
4215 exit_dispatcher_thread
:
4217 ret
= pthread_join(health_thread
, &status
);
4220 PERROR("pthread_join health_thread");
4225 utils_close_pipe(health_quit_pipe
);
4226 exit_health_quit_pipe
:
4229 health_app_destroy(health_relayd
);
4230 sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry
);
4231 exit_health_app_create
:
4232 exit_sessiond_trace_chunk_registry
:
4235 * Wait for all pending call_rcu work to complete before tearing
4236 * down data structures. call_rcu worker may be trying to
4237 * perform lookups in those structures.
4242 /* Ensure all prior call_rcu are done. */