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
= {};
1108 if (conn
->minor
< 4) {
1109 /* From 2.1 to 2.3 */
1111 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1112 /* From 2.4 to 2.10 */
1113 ret
= cmd_create_session_2_4(payload
, session_name
,
1114 hostname
, &live_timer
, &snapshot
);
1116 bool has_current_chunk
;
1118 /* From 2.11 to ... */
1119 ret
= cmd_create_session_2_11(payload
, session_name
,
1120 hostname
, &live_timer
, &snapshot
,
1121 &id_sessiond
.value
, sessiond_uuid
,
1123 ¤t_chunk_id
.value
);
1124 if (lttng_uuid_is_nil(sessiond_uuid
)) {
1125 /* The nil UUID is reserved for pre-2.11 clients. */
1126 ERR("Illegal nil UUID announced by peer in create session command");
1130 id_sessiond
.is_set
= true;
1131 current_chunk_id
.is_set
= has_current_chunk
;
1138 session
= session_create(session_name
, hostname
, live_timer
,
1139 snapshot
, sessiond_uuid
,
1140 id_sessiond
.is_set
? &id_sessiond
.value
: NULL
,
1141 current_chunk_id
.is_set
? ¤t_chunk_id
.value
: NULL
,
1142 conn
->major
, conn
->minor
);
1147 assert(!conn
->session
);
1148 conn
->session
= session
;
1149 DBG("Created session %" PRIu64
, session
->id
);
1151 reply
.session_id
= htobe64(session
->id
);
1155 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1157 reply
.ret_code
= htobe32(LTTNG_OK
);
1160 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1161 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1162 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1166 if (ret
< 0 && session
) {
1167 session_put(session
);
1173 * When we have received all the streams and the metadata for a channel,
1174 * we make them visible to the viewer threads.
1176 static void publish_connection_local_streams(struct relay_connection
*conn
)
1178 struct relay_stream
*stream
;
1179 struct relay_session
*session
= conn
->session
;
1182 * We publish all streams belonging to a session atomically wrt
1185 pthread_mutex_lock(&session
->lock
);
1187 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1189 stream_publish(stream
);
1194 * Inform the viewer that there are new streams in the session.
1196 if (session
->viewer_attached
) {
1197 uatomic_set(&session
->new_streams
, 1);
1199 pthread_mutex_unlock(&session
->lock
);
1203 * relay_add_stream: allocate a new stream for a session
1205 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1206 struct relay_connection
*conn
,
1207 const struct lttng_buffer_view
*payload
)
1211 struct relay_session
*session
= conn
->session
;
1212 struct relay_stream
*stream
= NULL
;
1213 struct lttcomm_relayd_status_stream reply
;
1214 struct ctf_trace
*trace
= NULL
;
1215 uint64_t stream_handle
= -1ULL;
1216 char *path_name
= NULL
, *channel_name
= NULL
;
1217 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1218 struct relay_stream_chunk_id stream_chunk_id
= { 0 };
1220 if (!session
|| !conn
->version_check_done
) {
1221 ERR("Trying to add a stream before version check");
1223 goto end_no_session
;
1226 if (session
->minor
== 1) {
1228 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1230 } else if (session
->minor
> 1 && session
->minor
< 11) {
1231 /* From 2.2 to 2.10 */
1232 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1233 &channel_name
, &tracefile_size
, &tracefile_count
);
1235 /* From 2.11 to ... */
1236 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1237 &channel_name
, &tracefile_size
, &tracefile_count
,
1238 &stream_chunk_id
.value
);
1239 stream_chunk_id
.is_set
= true;
1246 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1250 /* This stream here has one reference on the trace. */
1252 pthread_mutex_lock(&last_relay_stream_id_lock
);
1253 stream_handle
= ++last_relay_stream_id
;
1254 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1256 /* We pass ownership of path_name and channel_name. */
1257 stream
= stream_create(trace
, stream_handle
, path_name
,
1258 channel_name
, tracefile_size
, tracefile_count
,
1261 channel_name
= NULL
;
1264 * Streams are the owners of their trace. Reference to trace is
1265 * kept within stream_create().
1267 ctf_trace_put(trace
);
1270 memset(&reply
, 0, sizeof(reply
));
1271 reply
.handle
= htobe64(stream_handle
);
1273 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1275 reply
.ret_code
= htobe32(LTTNG_OK
);
1278 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1279 sizeof(struct lttcomm_relayd_status_stream
), 0);
1280 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1281 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1293 * relay_close_stream: close a specific stream
1295 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1296 struct relay_connection
*conn
,
1297 const struct lttng_buffer_view
*payload
)
1301 struct relay_session
*session
= conn
->session
;
1302 struct lttcomm_relayd_close_stream stream_info
;
1303 struct lttcomm_relayd_generic_reply reply
;
1304 struct relay_stream
*stream
;
1306 DBG("Close stream received");
1308 if (!session
|| !conn
->version_check_done
) {
1309 ERR("Trying to close a stream before version check");
1311 goto end_no_session
;
1314 if (payload
->size
< sizeof(stream_info
)) {
1315 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1316 sizeof(stream_info
), payload
->size
);
1318 goto end_no_session
;
1320 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1321 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1322 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1324 stream
= stream_get_by_id(stream_info
.stream_id
);
1331 * Set last_net_seq_num before the close flag. Required by data
1334 pthread_mutex_lock(&stream
->lock
);
1335 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1336 pthread_mutex_unlock(&stream
->lock
);
1339 * This is one of the conditions which may trigger a stream close
1340 * with the others being:
1341 * 1) A close command is received for a stream
1342 * 2) The control connection owning the stream is closed
1343 * 3) We have received all of the stream's data _after_ a close
1346 try_stream_close(stream
);
1347 if (stream
->is_metadata
) {
1348 struct relay_viewer_stream
*vstream
;
1350 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1352 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1354 * Since all the metadata has been sent to the
1355 * viewer and that we have a request to close
1356 * its stream, we can safely teardown the
1357 * corresponding metadata viewer stream.
1359 viewer_stream_put(vstream
);
1361 /* Put local reference. */
1362 viewer_stream_put(vstream
);
1369 memset(&reply
, 0, sizeof(reply
));
1371 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1373 reply
.ret_code
= htobe32(LTTNG_OK
);
1375 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1376 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1377 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1378 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1388 * relay_reset_metadata: reset a metadata stream
1391 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1392 struct relay_connection
*conn
,
1393 const struct lttng_buffer_view
*payload
)
1397 struct relay_session
*session
= conn
->session
;
1398 struct lttcomm_relayd_reset_metadata stream_info
;
1399 struct lttcomm_relayd_generic_reply reply
;
1400 struct relay_stream
*stream
;
1402 DBG("Reset metadata received");
1404 if (!session
|| !conn
->version_check_done
) {
1405 ERR("Trying to reset a metadata stream before version check");
1407 goto end_no_session
;
1410 if (payload
->size
< sizeof(stream_info
)) {
1411 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1412 sizeof(stream_info
), payload
->size
);
1414 goto end_no_session
;
1416 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1417 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1418 stream_info
.version
= be64toh(stream_info
.version
);
1420 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1422 /* Unsupported for live sessions for now. */
1423 if (session
->live_timer
!= 0) {
1428 stream
= stream_get_by_id(stream_info
.stream_id
);
1433 pthread_mutex_lock(&stream
->lock
);
1434 if (!stream
->is_metadata
) {
1439 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
1440 0, 0, -1, -1, stream
->stream_fd
->fd
, NULL
,
1441 &stream
->stream_fd
->fd
);
1443 ERR("Failed to rotate metadata file %s of channel %s",
1444 stream
->path_name
, stream
->channel_name
);
1449 pthread_mutex_unlock(&stream
->lock
);
1453 memset(&reply
, 0, sizeof(reply
));
1455 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1457 reply
.ret_code
= htobe32(LTTNG_OK
);
1459 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1460 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1461 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1462 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1472 * relay_unknown_command: send -1 if received unknown command
1474 static void relay_unknown_command(struct relay_connection
*conn
)
1476 struct lttcomm_relayd_generic_reply reply
;
1479 memset(&reply
, 0, sizeof(reply
));
1480 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1481 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1482 if (send_ret
< sizeof(reply
)) {
1483 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1488 * relay_start: send an acknowledgment to the client to tell if we are
1489 * ready to receive data. We are ready if a session is established.
1491 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1492 struct relay_connection
*conn
,
1493 const struct lttng_buffer_view
*payload
)
1497 struct lttcomm_relayd_generic_reply reply
;
1498 struct relay_session
*session
= conn
->session
;
1501 DBG("Trying to start the streaming without a session established");
1502 ret
= htobe32(LTTNG_ERR_UNK
);
1505 memset(&reply
, 0, sizeof(reply
));
1506 reply
.ret_code
= htobe32(LTTNG_OK
);
1507 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1509 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1510 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1519 * Append padding to the file pointed by the file descriptor fd.
1521 static int write_padding_to_file(int fd
, uint32_t size
)
1530 zeros
= zmalloc(size
);
1531 if (zeros
== NULL
) {
1532 PERROR("zmalloc zeros for padding");
1537 ret
= lttng_write(fd
, zeros
, size
);
1539 PERROR("write padding to file");
1549 * Close the current index file if it is open, and create a new one.
1551 * Return 0 on success, -1 on error.
1554 int create_rotate_index_file(struct relay_stream
*stream
,
1555 const char *stream_path
)
1558 uint32_t major
, minor
;
1560 /* Put ref on previous index_file. */
1561 if (stream
->index_file
) {
1562 lttng_index_file_put(stream
->index_file
);
1563 stream
->index_file
= NULL
;
1565 major
= stream
->trace
->session
->major
;
1566 minor
= stream
->trace
->session
->minor
;
1567 stream
->index_file
= lttng_index_file_create(stream_path
,
1568 stream
->channel_name
,
1569 -1, -1, stream
->tracefile_size
,
1570 tracefile_array_get_file_index_head(stream
->tfa
),
1571 lttng_to_index_major(major
, minor
),
1572 lttng_to_index_minor(major
, minor
));
1573 if (!stream
->index_file
) {
1585 int do_rotate_stream_data(struct relay_stream
*stream
)
1589 DBG("Rotating stream %" PRIu64
" data file",
1590 stream
->stream_handle
);
1591 /* Perform the stream rotation. */
1592 ret
= utils_rotate_stream_file(stream
->path_name
,
1593 stream
->channel_name
, stream
->tracefile_size
,
1594 stream
->tracefile_count
, -1,
1595 -1, stream
->stream_fd
->fd
,
1596 NULL
, &stream
->stream_fd
->fd
);
1598 ERR("Rotating stream output file");
1601 stream
->tracefile_size_current
= 0;
1602 stream
->pos_after_last_complete_data_index
= 0;
1603 stream
->data_rotated
= true;
1605 if (stream
->data_rotated
&& stream
->index_rotated
) {
1606 /* Rotation completed; reset its state. */
1607 DBG("Rotation completed for stream %" PRIu64
,
1608 stream
->stream_handle
);
1609 stream
->rotate_at_seq_num
= -1ULL;
1610 stream
->data_rotated
= false;
1611 stream
->index_rotated
= false;
1618 * If too much data has been written in a tracefile before we received the
1619 * rotation command, we have to move the excess data to the new tracefile and
1620 * perform the rotation. This can happen because the control and data
1621 * connections are separate, the indexes as well as the commands arrive from
1622 * the control connection and we have no control over the order so we could be
1623 * in a situation where too much data has been received on the data connection
1624 * before the rotation command on the control connection arrives.
1627 int rotate_truncate_stream(struct relay_stream
*stream
)
1631 uint64_t diff
, pos
= 0;
1632 char buf
[FILE_COPY_BUFFER_SIZE
];
1634 assert(!stream
->is_metadata
);
1636 assert(stream
->tracefile_size_current
>
1637 stream
->pos_after_last_complete_data_index
);
1638 diff
= stream
->tracefile_size_current
-
1639 stream
->pos_after_last_complete_data_index
;
1641 /* Create the new tracefile. */
1642 new_fd
= utils_create_stream_file(stream
->path_name
,
1643 stream
->channel_name
,
1644 stream
->tracefile_size
, stream
->tracefile_count
,
1645 /* uid */ -1, /* gid */ -1, /* suffix */ NULL
);
1647 ERR("Failed to create new stream file at path %s for channel %s",
1648 stream
->path_name
, stream
->channel_name
);
1654 * Rewind the current tracefile to the position at which the rotation
1655 * should have occurred.
1657 lseek_ret
= lseek(stream
->stream_fd
->fd
,
1658 stream
->pos_after_last_complete_data_index
, SEEK_SET
);
1659 if (lseek_ret
< 0) {
1660 PERROR("seek truncate stream");
1665 /* Move data from the old file to the new file. */
1666 while (pos
< diff
) {
1667 uint64_t count
, bytes_left
;
1670 bytes_left
= diff
- pos
;
1671 count
= bytes_left
> sizeof(buf
) ? sizeof(buf
) : bytes_left
;
1672 assert(count
<= SIZE_MAX
);
1674 io_ret
= lttng_read(stream
->stream_fd
->fd
, buf
, count
);
1675 if (io_ret
< (ssize_t
) count
) {
1676 char error_string
[256];
1678 snprintf(error_string
, sizeof(error_string
),
1679 "Failed to read %" PRIu64
" bytes from fd %i in rotate_truncate_stream(), returned %zi",
1680 count
, stream
->stream_fd
->fd
, io_ret
);
1682 PERROR("%s", error_string
);
1684 ERR("%s", error_string
);
1690 io_ret
= lttng_write(new_fd
, buf
, count
);
1691 if (io_ret
< (ssize_t
) count
) {
1692 char error_string
[256];
1694 snprintf(error_string
, sizeof(error_string
),
1695 "Failed to write %" PRIu64
" bytes from fd %i in rotate_truncate_stream(), returned %zi",
1696 count
, new_fd
, io_ret
);
1698 PERROR("%s", error_string
);
1700 ERR("%s", error_string
);
1709 /* Truncate the file to get rid of the excess data. */
1710 ret
= ftruncate(stream
->stream_fd
->fd
,
1711 stream
->pos_after_last_complete_data_index
);
1713 PERROR("ftruncate");
1717 ret
= close(stream
->stream_fd
->fd
);
1719 PERROR("Closing tracefile");
1724 * Update the offset and FD of all the eventual indexes created by the
1725 * data connection before the rotation command arrived.
1727 ret
= relay_index_switch_all_files(stream
);
1729 ERR("Failed to rotate index file");
1733 stream
->stream_fd
->fd
= new_fd
;
1734 stream
->tracefile_size_current
= diff
;
1735 stream
->pos_after_last_complete_data_index
= 0;
1736 stream
->rotate_at_seq_num
= -1ULL;
1745 * Check if a stream's index file should be rotated (for session rotation).
1746 * Must be called with the stream lock held.
1748 * Return 0 on success, a negative value on error.
1751 int try_rotate_stream_index(struct relay_stream
*stream
)
1755 if (stream
->rotate_at_seq_num
== -1ULL) {
1756 /* No rotation expected. */
1760 if (stream
->index_rotated
) {
1761 /* Rotation of the index has already occurred. */
1765 if (stream
->prev_index_seq
== -1ULL ||
1766 stream
->prev_index_seq
< stream
->rotate_at_seq_num
) {
1767 DBG("Stream %" PRIu64
" index not yet ready for rotation (rotate_at_seq_num = %" PRIu64
", prev_index_seq = %" PRIu64
")",
1768 stream
->stream_handle
,
1769 stream
->rotate_at_seq_num
,
1770 stream
->prev_index_seq
);
1772 } else if (stream
->prev_index_seq
!= stream
->rotate_at_seq_num
) {
1774 * Unexpected, protocol error/bug.
1775 * It could mean that we received a rotation position
1776 * that is in the past.
1778 ERR("Stream %" PRIu64
" index is in an inconsistent state (rotate_at_seq_num = %" PRIu64
", prev_data_seq = %" PRIu64
", prev_index_seq = %" PRIu64
")",
1779 stream
->stream_handle
,
1780 stream
->rotate_at_seq_num
,
1781 stream
->prev_data_seq
,
1782 stream
->prev_index_seq
);
1786 DBG("Rotating stream %" PRIu64
" index file",
1787 stream
->stream_handle
);
1788 ret
= create_rotate_index_file(stream
, stream
->path_name
);
1789 stream
->index_rotated
= true;
1791 if (stream
->data_rotated
&& stream
->index_rotated
) {
1792 /* Rotation completed; reset its state. */
1793 DBG("Rotation completed for stream %" PRIu64
,
1794 stream
->stream_handle
);
1795 stream
->rotate_at_seq_num
= -1ULL;
1796 stream
->data_rotated
= false;
1797 stream
->index_rotated
= false;
1806 * Check if a stream's data file (as opposed to index) should be rotated
1807 * (for session rotation).
1808 * Must be called with the stream lock held.
1810 * Return 0 on success, a negative value on error.
1813 int try_rotate_stream_data(struct relay_stream
*stream
)
1817 if (stream
->rotate_at_seq_num
== -1ULL) {
1818 /* No rotation expected. */
1822 if (stream
->data_rotated
) {
1823 /* Rotation of the data file has already occurred. */
1827 if (stream
->prev_data_seq
== -1ULL ||
1828 stream
->prev_data_seq
< stream
->rotate_at_seq_num
) {
1829 DBG("Stream %" PRIu64
" not yet ready for rotation (rotate_at_seq_num = %" PRIu64
", prev_data_seq = %" PRIu64
")",
1830 stream
->stream_handle
,
1831 stream
->rotate_at_seq_num
,
1832 stream
->prev_data_seq
);
1834 } else if (stream
->prev_data_seq
> stream
->rotate_at_seq_num
) {
1836 * prev_data_seq is checked here since indexes and rotation
1837 * commands are serialized with respect to each other.
1839 DBG("Rotation after too much data has been written in tracefile "
1840 "for stream %" PRIu64
", need to truncate before "
1841 "rotating", stream
->stream_handle
);
1842 ret
= rotate_truncate_stream(stream
);
1844 ERR("Failed to truncate stream");
1847 } else if (stream
->prev_data_seq
!= stream
->rotate_at_seq_num
) {
1849 * Unexpected, protocol error/bug.
1850 * It could mean that we received a rotation position
1851 * that is in the past.
1853 ERR("Stream %" PRIu64
" data is in an inconsistent state (rotate_at_seq_num = %" PRIu64
", prev_data_seq = %" PRIu64
")",
1854 stream
->stream_handle
,
1855 stream
->rotate_at_seq_num
,
1856 stream
->prev_data_seq
);
1860 ret
= do_rotate_stream_data(stream
);
1868 * relay_recv_metadata: receive the metadata for the session.
1870 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1871 struct relay_connection
*conn
,
1872 const struct lttng_buffer_view
*payload
)
1876 struct relay_session
*session
= conn
->session
;
1877 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1878 struct relay_stream
*metadata_stream
;
1879 uint64_t metadata_payload_size
;
1882 ERR("Metadata sent before version check");
1887 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1888 ERR("Incorrect data size");
1892 metadata_payload_size
= recv_hdr
->data_size
-
1893 sizeof(struct lttcomm_relayd_metadata_payload
);
1895 memcpy(&metadata_payload_header
, payload
->data
,
1896 sizeof(metadata_payload_header
));
1897 metadata_payload_header
.stream_id
= be64toh(
1898 metadata_payload_header
.stream_id
);
1899 metadata_payload_header
.padding_size
= be32toh(
1900 metadata_payload_header
.padding_size
);
1902 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1903 if (!metadata_stream
) {
1908 pthread_mutex_lock(&metadata_stream
->lock
);
1910 size_ret
= lttng_write(metadata_stream
->stream_fd
->fd
,
1911 payload
->data
+ sizeof(metadata_payload_header
),
1912 metadata_payload_size
);
1913 if (size_ret
< metadata_payload_size
) {
1914 ERR("Relay error writing metadata on file");
1919 size_ret
= write_padding_to_file(metadata_stream
->stream_fd
->fd
,
1920 metadata_payload_header
.padding_size
);
1921 if (size_ret
< (int64_t) metadata_payload_header
.padding_size
) {
1926 metadata_stream
->metadata_received
+=
1927 metadata_payload_size
+ metadata_payload_header
.padding_size
;
1928 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1929 metadata_stream
->metadata_received
);
1931 ret
= try_rotate_stream_data(metadata_stream
);
1937 pthread_mutex_unlock(&metadata_stream
->lock
);
1938 stream_put(metadata_stream
);
1944 * relay_send_version: send relayd version number
1946 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1947 struct relay_connection
*conn
,
1948 const struct lttng_buffer_view
*payload
)
1952 struct lttcomm_relayd_version reply
, msg
;
1953 bool compatible
= true;
1955 conn
->version_check_done
= true;
1957 /* Get version from the other side. */
1958 if (payload
->size
< sizeof(msg
)) {
1959 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1960 sizeof(msg
), payload
->size
);
1965 memcpy(&msg
, payload
->data
, sizeof(msg
));
1966 msg
.major
= be32toh(msg
.major
);
1967 msg
.minor
= be32toh(msg
.minor
);
1969 memset(&reply
, 0, sizeof(reply
));
1970 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1971 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1973 /* Major versions must be the same */
1974 if (reply
.major
!= msg
.major
) {
1975 DBG("Incompatible major versions (%u vs %u), deleting session",
1976 reply
.major
, msg
.major
);
1980 conn
->major
= reply
.major
;
1981 /* We adapt to the lowest compatible version */
1982 if (reply
.minor
<= msg
.minor
) {
1983 conn
->minor
= reply
.minor
;
1985 conn
->minor
= msg
.minor
;
1988 reply
.major
= htobe32(reply
.major
);
1989 reply
.minor
= htobe32(reply
.minor
);
1990 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1992 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1993 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2006 DBG("Version check done using protocol %u.%u", conn
->major
,
2014 * Check for data pending for a given stream id from the session daemon.
2016 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2017 struct relay_connection
*conn
,
2018 const struct lttng_buffer_view
*payload
)
2020 struct relay_session
*session
= conn
->session
;
2021 struct lttcomm_relayd_data_pending msg
;
2022 struct lttcomm_relayd_generic_reply reply
;
2023 struct relay_stream
*stream
;
2026 uint64_t stream_seq
;
2028 DBG("Data pending command received");
2030 if (!session
|| !conn
->version_check_done
) {
2031 ERR("Trying to check for data before version check");
2033 goto end_no_session
;
2036 if (payload
->size
< sizeof(msg
)) {
2037 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2038 sizeof(msg
), payload
->size
);
2040 goto end_no_session
;
2042 memcpy(&msg
, payload
->data
, sizeof(msg
));
2043 msg
.stream_id
= be64toh(msg
.stream_id
);
2044 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
2046 stream
= stream_get_by_id(msg
.stream_id
);
2047 if (stream
== NULL
) {
2052 pthread_mutex_lock(&stream
->lock
);
2054 if (session_streams_have_index(session
)) {
2056 * Ensure that both the index and stream data have been
2057 * flushed up to the requested point.
2059 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2061 stream_seq
= stream
->prev_data_seq
;
2063 DBG("Data pending for stream id %" PRIu64
": prev_data_seq %" PRIu64
2064 ", prev_index_seq %" PRIu64
2065 ", and last_seq %" PRIu64
, msg
.stream_id
,
2066 stream
->prev_data_seq
, stream
->prev_index_seq
,
2067 msg
.last_net_seq_num
);
2069 /* Avoid wrapping issue */
2070 if (((int64_t) (stream_seq
- msg
.last_net_seq_num
)) >= 0) {
2071 /* Data has in fact been written and is NOT pending */
2074 /* Data still being streamed thus pending */
2078 stream
->data_pending_check_done
= true;
2079 pthread_mutex_unlock(&stream
->lock
);
2084 memset(&reply
, 0, sizeof(reply
));
2085 reply
.ret_code
= htobe32(ret
);
2086 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2087 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2088 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2098 * Wait for the control socket to reach a quiescent state.
2100 * Note that for now, when receiving this command from the session
2101 * daemon, this means that every subsequent commands or data received on
2102 * the control socket has been handled. So, this is why we simply return
2105 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
2106 struct relay_connection
*conn
,
2107 const struct lttng_buffer_view
*payload
)
2111 struct relay_stream
*stream
;
2112 struct lttcomm_relayd_quiescent_control msg
;
2113 struct lttcomm_relayd_generic_reply reply
;
2115 DBG("Checking quiescent state on control socket");
2117 if (!conn
->session
|| !conn
->version_check_done
) {
2118 ERR("Trying to check for data before version check");
2120 goto end_no_session
;
2123 if (payload
->size
< sizeof(msg
)) {
2124 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2125 sizeof(msg
), payload
->size
);
2127 goto end_no_session
;
2129 memcpy(&msg
, payload
->data
, sizeof(msg
));
2130 msg
.stream_id
= be64toh(msg
.stream_id
);
2132 stream
= stream_get_by_id(msg
.stream_id
);
2136 pthread_mutex_lock(&stream
->lock
);
2137 stream
->data_pending_check_done
= true;
2138 pthread_mutex_unlock(&stream
->lock
);
2140 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
2143 memset(&reply
, 0, sizeof(reply
));
2144 reply
.ret_code
= htobe32(LTTNG_OK
);
2145 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2146 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2147 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2159 * Initialize a data pending command. This means that a consumer is about
2160 * to ask for data pending for each stream it holds. Simply iterate over
2161 * all streams of a session and set the data_pending_check_done flag.
2163 * This command returns to the client a LTTNG_OK code.
2165 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2166 struct relay_connection
*conn
,
2167 const struct lttng_buffer_view
*payload
)
2171 struct lttng_ht_iter iter
;
2172 struct lttcomm_relayd_begin_data_pending msg
;
2173 struct lttcomm_relayd_generic_reply reply
;
2174 struct relay_stream
*stream
;
2179 DBG("Init streams for data pending");
2181 if (!conn
->session
|| !conn
->version_check_done
) {
2182 ERR("Trying to check for data before version check");
2184 goto end_no_session
;
2187 if (payload
->size
< sizeof(msg
)) {
2188 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2189 sizeof(msg
), payload
->size
);
2191 goto end_no_session
;
2193 memcpy(&msg
, payload
->data
, sizeof(msg
));
2194 msg
.session_id
= be64toh(msg
.session_id
);
2197 * Iterate over all streams to set the begin data pending flag.
2198 * For now, the streams are indexed by stream handle so we have
2199 * to iterate over all streams to find the one associated with
2200 * the right session_id.
2203 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2205 if (!stream_get(stream
)) {
2208 if (stream
->trace
->session
->id
== msg
.session_id
) {
2209 pthread_mutex_lock(&stream
->lock
);
2210 stream
->data_pending_check_done
= false;
2211 pthread_mutex_unlock(&stream
->lock
);
2212 DBG("Set begin data pending flag to stream %" PRIu64
,
2213 stream
->stream_handle
);
2219 memset(&reply
, 0, sizeof(reply
));
2220 /* All good, send back reply. */
2221 reply
.ret_code
= htobe32(LTTNG_OK
);
2223 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2224 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2225 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2237 * End data pending command. This will check, for a given session id, if
2238 * each stream associated with it has its data_pending_check_done flag
2239 * set. If not, this means that the client lost track of the stream but
2240 * the data is still being streamed on our side. In this case, we inform
2241 * the client that data is in flight.
2243 * Return to the client if there is data in flight or not with a ret_code.
2245 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2246 struct relay_connection
*conn
,
2247 const struct lttng_buffer_view
*payload
)
2251 struct lttng_ht_iter iter
;
2252 struct lttcomm_relayd_end_data_pending msg
;
2253 struct lttcomm_relayd_generic_reply reply
;
2254 struct relay_stream
*stream
;
2255 uint32_t is_data_inflight
= 0;
2257 DBG("End data pending command");
2259 if (!conn
->session
|| !conn
->version_check_done
) {
2260 ERR("Trying to check for data before version check");
2262 goto end_no_session
;
2265 if (payload
->size
< sizeof(msg
)) {
2266 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2267 sizeof(msg
), payload
->size
);
2269 goto end_no_session
;
2271 memcpy(&msg
, payload
->data
, sizeof(msg
));
2272 msg
.session_id
= be64toh(msg
.session_id
);
2275 * Iterate over all streams to see if the begin data pending
2279 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2281 if (!stream_get(stream
)) {
2284 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2288 pthread_mutex_lock(&stream
->lock
);
2289 if (!stream
->data_pending_check_done
) {
2290 uint64_t stream_seq
;
2292 if (session_streams_have_index(conn
->session
)) {
2294 * Ensure that both the index and stream data have been
2295 * flushed up to the requested point.
2297 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2299 stream_seq
= stream
->prev_data_seq
;
2301 if (!stream
->closed
|| !(((int64_t) (stream_seq
- stream
->last_net_seq_num
)) >= 0)) {
2302 is_data_inflight
= 1;
2303 DBG("Data is still in flight for stream %" PRIu64
,
2304 stream
->stream_handle
);
2305 pthread_mutex_unlock(&stream
->lock
);
2310 pthread_mutex_unlock(&stream
->lock
);
2315 memset(&reply
, 0, sizeof(reply
));
2316 /* All good, send back reply. */
2317 reply
.ret_code
= htobe32(is_data_inflight
);
2319 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2320 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2321 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2333 * Receive an index for a specific stream.
2335 * Return 0 on success else a negative value.
2337 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2338 struct relay_connection
*conn
,
2339 const struct lttng_buffer_view
*payload
)
2343 struct relay_session
*session
= conn
->session
;
2344 struct lttcomm_relayd_index index_info
;
2345 struct relay_index
*index
;
2346 struct lttcomm_relayd_generic_reply reply
;
2347 struct relay_stream
*stream
;
2352 DBG("Relay receiving index");
2354 if (!session
|| !conn
->version_check_done
) {
2355 ERR("Trying to close a stream before version check");
2357 goto end_no_session
;
2360 msg_len
= lttcomm_relayd_index_len(
2361 lttng_to_index_major(conn
->major
, conn
->minor
),
2362 lttng_to_index_minor(conn
->major
, conn
->minor
));
2363 if (payload
->size
< msg_len
) {
2364 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2365 msg_len
, payload
->size
);
2367 goto end_no_session
;
2369 memcpy(&index_info
, payload
->data
, msg_len
);
2370 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2371 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2372 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2373 index_info
.content_size
= be64toh(index_info
.content_size
);
2374 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2375 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2376 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2377 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2379 if (conn
->minor
>= 8) {
2380 index_info
.stream_instance_id
=
2381 be64toh(index_info
.stream_instance_id
);
2382 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2385 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2387 ERR("stream_get_by_id not found");
2391 pthread_mutex_lock(&stream
->lock
);
2393 /* Live beacon handling */
2394 if (index_info
.packet_size
== 0) {
2395 DBG("Received live beacon for stream %" PRIu64
,
2396 stream
->stream_handle
);
2399 * Only flag a stream inactive when it has already
2400 * received data and no indexes are in flight.
2402 if (stream
->index_received_seqcount
> 0
2403 && stream
->indexes_in_flight
== 0) {
2404 stream
->beacon_ts_end
= index_info
.timestamp_end
;
2407 goto end_stream_put
;
2409 stream
->beacon_ts_end
= -1ULL;
2412 if (stream
->ctf_stream_id
== -1ULL) {
2413 stream
->ctf_stream_id
= index_info
.stream_id
;
2415 index
= relay_index_get_by_id_or_create(stream
, index_info
.net_seq_num
);
2418 ERR("relay_index_get_by_id_or_create index NULL");
2419 goto end_stream_put
;
2421 if (set_index_control_data(index
, &index_info
, conn
)) {
2422 ERR("set_index_control_data error");
2423 relay_index_put(index
);
2425 goto end_stream_put
;
2427 ret
= relay_index_try_flush(index
);
2429 tracefile_array_commit_seq(stream
->tfa
);
2430 stream
->index_received_seqcount
++;
2431 stream
->pos_after_last_complete_data_index
+= index
->total_size
;
2432 stream
->prev_index_seq
= index_info
.net_seq_num
;
2434 ret
= try_rotate_stream_index(stream
);
2436 goto end_stream_put
;
2438 } else if (ret
> 0) {
2445 * relay_index_try_flush is responsible for the self-reference
2446 * put of the index object on error.
2448 ERR("relay_index_try_flush error %d", ret
);
2453 pthread_mutex_unlock(&stream
->lock
);
2458 memset(&reply
, 0, sizeof(reply
));
2460 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2462 reply
.ret_code
= htobe32(LTTNG_OK
);
2464 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2465 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2466 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2475 * Receive the streams_sent message.
2477 * Return 0 on success else a negative value.
2479 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2480 struct relay_connection
*conn
,
2481 const struct lttng_buffer_view
*payload
)
2485 struct lttcomm_relayd_generic_reply reply
;
2489 DBG("Relay receiving streams_sent");
2491 if (!conn
->session
|| !conn
->version_check_done
) {
2492 ERR("Trying to close a stream before version check");
2494 goto end_no_session
;
2498 * Publish every pending stream in the connection recv list which are
2499 * now ready to be used by the viewer.
2501 publish_connection_local_streams(conn
);
2503 memset(&reply
, 0, sizeof(reply
));
2504 reply
.ret_code
= htobe32(LTTNG_OK
);
2505 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2506 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2507 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2520 * relay_rotate_session_stream: rotate a stream to a new tracefile for the session
2521 * rotation feature (not the tracefile rotation feature).
2523 static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
2524 struct relay_connection
*conn
,
2525 const struct lttng_buffer_view
*payload
)
2529 struct relay_session
*session
= conn
->session
;
2530 struct lttcomm_relayd_rotate_stream stream_info
;
2531 struct lttcomm_relayd_generic_reply reply
;
2532 struct relay_stream
*stream
;
2535 struct lttng_buffer_view new_path_view
;
2537 if (!session
|| !conn
->version_check_done
) {
2538 ERR("Trying to rotate a stream before version check");
2543 if (session
->major
== 2 && session
->minor
< 11) {
2544 ERR("Unsupported feature before 2.11");
2549 header_len
= sizeof(struct lttcomm_relayd_rotate_stream
);
2551 if (payload
->size
< header_len
) {
2552 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2553 header_len
, payload
->size
);
2558 memcpy(&stream_info
, payload
->data
, header_len
);
2560 /* Convert to host */
2561 stream_info
.pathname_length
= be32toh(stream_info
.pathname_length
);
2562 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
2563 stream_info
.new_chunk_id
= be64toh(stream_info
.new_chunk_id
);
2564 stream_info
.rotate_at_seq_num
= be64toh(stream_info
.rotate_at_seq_num
);
2566 path_len
= stream_info
.pathname_length
;
2567 if (payload
->size
< header_len
+ path_len
) {
2568 ERR("Unexpected payload size in \"relay_rotate_session_stream\" including path: expected >= %zu bytes, got %zu bytes",
2569 header_len
+ path_len
, payload
->size
);
2574 /* Ensure it fits in local filename length. */
2575 if (path_len
>= LTTNG_PATH_MAX
) {
2576 ret
= -ENAMETOOLONG
;
2577 ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes",
2578 path_len
, LTTNG_PATH_MAX
);
2582 new_path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2583 stream_info
.pathname_length
);
2585 stream
= stream_get_by_id(stream_info
.stream_id
);
2591 pthread_mutex_lock(&stream
->lock
);
2594 * Update the trace path (just the folder, the stream name does not
2597 free(stream
->prev_path_name
);
2598 stream
->prev_path_name
= stream
->path_name
;
2599 stream
->path_name
= create_output_path(new_path_view
.data
);
2600 if (!stream
->path_name
) {
2601 ERR("Failed to create a new output path");
2603 goto end_stream_unlock
;
2605 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
,
2608 ERR("relay creating output directory");
2610 goto end_stream_unlock
;
2613 assert(stream
->current_chunk_id
.is_set
);
2614 stream
->current_chunk_id
.value
= stream_info
.new_chunk_id
;
2616 if (stream
->is_metadata
) {
2618 * Metadata streams have no index; consider its rotation
2621 stream
->index_rotated
= true;
2623 * The metadata stream is sent only over the control connection
2624 * so we know we have all the data to perform the stream
2627 ret
= do_rotate_stream_data(stream
);
2629 stream
->rotate_at_seq_num
= stream_info
.rotate_at_seq_num
;
2630 ret
= try_rotate_stream_data(stream
);
2632 goto end_stream_unlock
;
2635 ret
= try_rotate_stream_index(stream
);
2637 goto end_stream_unlock
;
2642 pthread_mutex_unlock(&stream
->lock
);
2645 memset(&reply
, 0, sizeof(reply
));
2647 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2649 reply
.ret_code
= htobe32(LTTNG_OK
);
2651 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2652 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2653 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2654 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2663 static int init_session_output_directory_handle(struct relay_session
*session
,
2664 struct lttng_directory_handle
*handle
)
2667 /* hostname/session_name */
2668 char *session_directory
= NULL
;
2670 * base path + session_directory
2671 * e.g. /home/user/lttng-traces/hostname/session_name
2673 char *full_session_path
= NULL
;
2675 pthread_mutex_lock(&session
->lock
);
2676 ret
= asprintf(&session_directory
, "%s/%s", session
->hostname
,
2677 session
->session_name
);
2678 pthread_mutex_unlock(&session
->lock
);
2680 PERROR("Failed to format session directory name");
2684 full_session_path
= create_output_path(session_directory
);
2685 if (!full_session_path
) {
2690 ret
= utils_mkdir_recursive(
2691 full_session_path
, S_IRWXU
| S_IRWXG
, -1, -1);
2693 ERR("Failed to create session output path \"%s\"",
2698 ret
= lttng_directory_handle_init(handle
, full_session_path
);
2703 free(session_directory
);
2704 free(full_session_path
);
2709 * relay_create_trace_chunk: create a new trace chunk
2711 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2712 struct relay_connection
*conn
,
2713 const struct lttng_buffer_view
*payload
)
2717 struct relay_session
*session
= conn
->session
;
2718 struct lttcomm_relayd_create_trace_chunk
*msg
;
2719 struct lttcomm_relayd_generic_reply reply
= {};
2720 struct lttng_buffer_view header_view
;
2721 struct lttng_buffer_view chunk_name_view
;
2722 struct lttng_trace_chunk
*chunk
= NULL
, *published_chunk
= NULL
;
2723 enum lttng_error_code reply_code
= LTTNG_OK
;
2724 enum lttng_trace_chunk_status chunk_status
;
2725 struct lttng_directory_handle session_output
;
2727 if (!session
|| !conn
->version_check_done
) {
2728 ERR("Trying to create a trace chunk before version check");
2733 if (session
->major
== 2 && session
->minor
< 11) {
2734 ERR("Chunk creation command is unsupported before 2.11");
2739 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2740 if (!header_view
.data
) {
2741 ERR("Failed to receive payload of chunk creation command");
2746 /* Convert to host endianness. */
2747 msg
= (typeof(msg
)) header_view
.data
;
2748 msg
->chunk_id
= be64toh(msg
->chunk_id
);
2749 msg
->creation_timestamp
= be64toh(msg
->creation_timestamp
);
2750 msg
->override_name_length
= be32toh(msg
->override_name_length
);
2752 chunk
= lttng_trace_chunk_create(
2753 msg
->chunk_id
, msg
->creation_timestamp
);
2755 ERR("Failed to create trace chunk in trace chunk creation command");
2757 reply_code
= LTTNG_ERR_NOMEM
;
2761 if (msg
->override_name_length
) {
2764 chunk_name_view
= lttng_buffer_view_from_view(payload
,
2766 msg
->override_name_length
);
2767 name
= chunk_name_view
.data
;
2768 if (!name
|| name
[msg
->override_name_length
- 1]) {
2769 ERR("Failed to receive payload of chunk creation command");
2771 reply_code
= LTTNG_ERR_INVALID
;
2775 chunk_status
= lttng_trace_chunk_override_name(
2776 chunk
, chunk_name_view
.data
);
2777 switch (chunk_status
) {
2778 case LTTNG_TRACE_CHUNK_STATUS_OK
:
2780 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT
:
2781 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2782 reply_code
= LTTNG_ERR_INVALID
;
2786 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2787 reply_code
= LTTNG_ERR_UNK
;
2793 ret
= init_session_output_directory_handle(
2794 conn
->session
, &session_output
);
2796 reply_code
= LTTNG_ERR_CREATE_DIR_FAIL
;
2800 chunk_status
= lttng_trace_chunk_set_credentials_current_user(chunk
);
2801 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2802 reply_code
= LTTNG_ERR_UNK
;
2807 chunk_status
= lttng_trace_chunk_set_as_owner(chunk
, &session_output
);
2808 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2809 reply_code
= LTTNG_ERR_UNK
;
2814 published_chunk
= sessiond_trace_chunk_registry_publish_chunk(
2815 sessiond_trace_chunk_registry
,
2816 conn
->session
->sessiond_uuid
,
2819 if (!published_chunk
) {
2820 char uuid_str
[UUID_STR_LEN
];
2822 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2823 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2828 reply_code
= LTTNG_ERR_NOMEM
;
2832 pthread_mutex_lock(&conn
->session
->lock
);
2833 lttng_trace_chunk_put(conn
->session
->current_trace_chunk
);
2834 conn
->session
->current_trace_chunk
= published_chunk
;
2835 pthread_mutex_unlock(&conn
->session
->lock
);
2836 published_chunk
= NULL
;
2839 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2840 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2842 sizeof(struct lttcomm_relayd_generic_reply
),
2844 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2845 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2850 lttng_trace_chunk_put(chunk
);
2851 lttng_trace_chunk_put(published_chunk
);
2852 lttng_directory_handle_fini(&session_output
);
2857 * relay_close_trace_chunk: close a trace chunk
2859 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2860 struct relay_connection
*conn
,
2861 const struct lttng_buffer_view
*payload
)
2865 struct relay_session
*session
= conn
->session
;
2866 struct lttcomm_relayd_close_trace_chunk
*msg
;
2867 struct lttcomm_relayd_generic_reply reply
= {};
2868 struct lttng_buffer_view header_view
;
2869 struct lttng_trace_chunk
*chunk
= NULL
;
2870 enum lttng_error_code reply_code
= LTTNG_OK
;
2871 enum lttng_trace_chunk_status chunk_status
;
2873 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type
) close_command
;
2874 time_t close_timestamp
;
2876 if (!session
|| !conn
->version_check_done
) {
2877 ERR("Trying to close a trace chunk before version check");
2882 if (session
->major
== 2 && session
->minor
< 11) {
2883 ERR("Chunk close command is unsupported before 2.11");
2888 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2889 if (!header_view
.data
) {
2890 ERR("Failed to receive payload of chunk close command");
2895 /* Convert to host endianness. */
2896 msg
= (typeof(msg
)) header_view
.data
;
2897 chunk_id
= be64toh(msg
->chunk_id
);
2898 close_timestamp
= (time_t) be64toh(msg
->close_timestamp
);
2899 close_command
= (typeof(close_command
)){
2900 .value
= be32toh(msg
->close_command
.value
),
2901 .is_set
= msg
->close_command
.is_set
,
2904 chunk
= sessiond_trace_chunk_registry_get_chunk(
2905 sessiond_trace_chunk_registry
,
2906 conn
->session
->sessiond_uuid
,
2910 char uuid_str
[UUID_STR_LEN
];
2912 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2913 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2918 reply_code
= LTTNG_ERR_NOMEM
;
2922 chunk_status
= lttng_trace_chunk_set_close_timestamp(
2923 chunk
, close_timestamp
);
2924 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2925 ERR("Failed to set trace chunk close timestamp");
2927 reply_code
= LTTNG_ERR_UNK
;
2931 if (close_command
.is_set
) {
2932 chunk_status
= lttng_trace_chunk_set_close_command(
2933 chunk
, close_command
.value
);
2934 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2936 reply_code
= LTTNG_ERR_INVALID
;
2942 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2943 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2945 sizeof(struct lttcomm_relayd_generic_reply
),
2947 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2948 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2953 lttng_trace_chunk_put(chunk
);
2957 #define DBG_CMD(cmd_name, conn) \
2958 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2960 static int relay_process_control_command(struct relay_connection
*conn
,
2961 const struct lttcomm_relayd_hdr
*header
,
2962 const struct lttng_buffer_view
*payload
)
2966 switch (header
->cmd
) {
2967 case RELAYD_CREATE_SESSION
:
2968 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
2969 ret
= relay_create_session(header
, conn
, payload
);
2971 case RELAYD_ADD_STREAM
:
2972 DBG_CMD("RELAYD_ADD_STREAM", conn
);
2973 ret
= relay_add_stream(header
, conn
, payload
);
2975 case RELAYD_START_DATA
:
2976 DBG_CMD("RELAYD_START_DATA", conn
);
2977 ret
= relay_start(header
, conn
, payload
);
2979 case RELAYD_SEND_METADATA
:
2980 DBG_CMD("RELAYD_SEND_METADATA", conn
);
2981 ret
= relay_recv_metadata(header
, conn
, payload
);
2983 case RELAYD_VERSION
:
2984 DBG_CMD("RELAYD_VERSION", conn
);
2985 ret
= relay_send_version(header
, conn
, payload
);
2987 case RELAYD_CLOSE_STREAM
:
2988 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
2989 ret
= relay_close_stream(header
, conn
, payload
);
2991 case RELAYD_DATA_PENDING
:
2992 DBG_CMD("RELAYD_DATA_PENDING", conn
);
2993 ret
= relay_data_pending(header
, conn
, payload
);
2995 case RELAYD_QUIESCENT_CONTROL
:
2996 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
2997 ret
= relay_quiescent_control(header
, conn
, payload
);
2999 case RELAYD_BEGIN_DATA_PENDING
:
3000 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
3001 ret
= relay_begin_data_pending(header
, conn
, payload
);
3003 case RELAYD_END_DATA_PENDING
:
3004 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
3005 ret
= relay_end_data_pending(header
, conn
, payload
);
3007 case RELAYD_SEND_INDEX
:
3008 DBG_CMD("RELAYD_SEND_INDEX", conn
);
3009 ret
= relay_recv_index(header
, conn
, payload
);
3011 case RELAYD_STREAMS_SENT
:
3012 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
3013 ret
= relay_streams_sent(header
, conn
, payload
);
3015 case RELAYD_RESET_METADATA
:
3016 DBG_CMD("RELAYD_RESET_METADATA", conn
);
3017 ret
= relay_reset_metadata(header
, conn
, payload
);
3019 case RELAYD_ROTATE_STREAM
:
3020 DBG_CMD("RELAYD_ROTATE_STREAM", conn
);
3021 ret
= relay_rotate_session_stream(header
, conn
, payload
);
3023 case RELAYD_CREATE_TRACE_CHUNK
:
3024 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn
);
3025 ret
= relay_create_trace_chunk(header
, conn
, payload
);
3027 case RELAYD_CLOSE_TRACE_CHUNK
:
3028 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn
);
3029 ret
= relay_close_trace_chunk(header
, conn
, payload
);
3031 case RELAYD_UPDATE_SYNC_INFO
:
3033 ERR("Received unknown command (%u)", header
->cmd
);
3034 relay_unknown_command(conn
);
3043 static enum relay_connection_status
relay_process_control_receive_payload(
3044 struct relay_connection
*conn
)
3047 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3048 struct lttng_dynamic_buffer
*reception_buffer
=
3049 &conn
->protocol
.ctrl
.reception_buffer
;
3050 struct ctrl_connection_state_receive_payload
*state
=
3051 &conn
->protocol
.ctrl
.state
.receive_payload
;
3052 struct lttng_buffer_view payload_view
;
3054 if (state
->left_to_receive
== 0) {
3055 /* Short-circuit for payload-less commands. */
3056 goto reception_complete
;
3059 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3060 reception_buffer
->data
+ state
->received
,
3061 state
->left_to_receive
, MSG_DONTWAIT
);
3063 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3064 PERROR("Unable to receive command payload on sock %d",
3066 status
= RELAY_CONNECTION_STATUS_ERROR
;
3069 } else if (ret
== 0) {
3070 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3071 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3076 assert(ret
<= state
->left_to_receive
);
3078 state
->left_to_receive
-= ret
;
3079 state
->received
+= ret
;
3081 if (state
->left_to_receive
> 0) {
3083 * Can't transition to the protocol's next state, wait to
3084 * receive the rest of the header.
3086 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3087 state
->received
, state
->left_to_receive
,
3093 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
3094 conn
->sock
->fd
, state
->received
);
3096 * The payload required to process the command has been received.
3097 * A view to the reception buffer is forwarded to the various
3098 * commands and the state of the control is reset on success.
3100 * Commands are responsible for sending their reply to the peer.
3102 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
3104 ret
= relay_process_control_command(conn
,
3105 &state
->header
, &payload_view
);
3107 status
= RELAY_CONNECTION_STATUS_ERROR
;
3111 ret
= connection_reset_protocol_state(conn
);
3113 status
= RELAY_CONNECTION_STATUS_ERROR
;
3119 static enum relay_connection_status
relay_process_control_receive_header(
3120 struct relay_connection
*conn
)
3123 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3124 struct lttcomm_relayd_hdr header
;
3125 struct lttng_dynamic_buffer
*reception_buffer
=
3126 &conn
->protocol
.ctrl
.reception_buffer
;
3127 struct ctrl_connection_state_receive_header
*state
=
3128 &conn
->protocol
.ctrl
.state
.receive_header
;
3130 assert(state
->left_to_receive
!= 0);
3132 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3133 reception_buffer
->data
+ state
->received
,
3134 state
->left_to_receive
, MSG_DONTWAIT
);
3136 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3137 PERROR("Unable to receive control command header on sock %d",
3139 status
= RELAY_CONNECTION_STATUS_ERROR
;
3142 } else if (ret
== 0) {
3143 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3144 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3149 assert(ret
<= state
->left_to_receive
);
3151 state
->left_to_receive
-= ret
;
3152 state
->received
+= ret
;
3154 if (state
->left_to_receive
> 0) {
3156 * Can't transition to the protocol's next state, wait to
3157 * receive the rest of the header.
3159 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3160 state
->received
, state
->left_to_receive
,
3165 /* Transition to next state: receiving the command's payload. */
3166 conn
->protocol
.ctrl
.state_id
=
3167 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3168 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
3169 header
.circuit_id
= be64toh(header
.circuit_id
);
3170 header
.data_size
= be64toh(header
.data_size
);
3171 header
.cmd
= be32toh(header
.cmd
);
3172 header
.cmd_version
= be32toh(header
.cmd_version
);
3173 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
3174 &header
, sizeof(header
));
3176 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
3177 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
3180 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
3181 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
3183 status
= RELAY_CONNECTION_STATUS_ERROR
;
3187 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
3189 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
3190 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
3193 status
= RELAY_CONNECTION_STATUS_ERROR
;
3197 if (header
.data_size
== 0) {
3199 * Manually invoke the next state as the poll loop
3200 * will not wake-up to allow us to proceed further.
3202 status
= relay_process_control_receive_payload(conn
);
3209 * Process the commands received on the control socket
3211 static enum relay_connection_status
relay_process_control(
3212 struct relay_connection
*conn
)
3214 enum relay_connection_status status
;
3216 switch (conn
->protocol
.ctrl
.state_id
) {
3217 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
3218 status
= relay_process_control_receive_header(conn
);
3220 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3221 status
= relay_process_control_receive_payload(conn
);
3224 ERR("Unknown control connection protocol state encountered.");
3232 * Handle index for a data stream.
3234 * Called with the stream lock held.
3236 * Return 0 on success else a negative value.
3238 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
3239 bool rotate_index
, bool *flushed
, uint64_t total_size
)
3242 uint64_t data_offset
;
3243 struct relay_index
*index
;
3245 /* Get data offset because we are about to update the index. */
3246 data_offset
= htobe64(stream
->tracefile_size_current
);
3248 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
3249 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
3252 * Lookup for an existing index for that stream id/sequence
3253 * number. If it exists, the control thread has already received the
3254 * data for it, thus we need to write it to disk.
3256 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
3262 if (rotate_index
|| !stream
->index_file
) {
3263 const char *stream_path
;
3266 * The data connection creates the stream's first index file.
3268 * This can happen _after_ a ROTATE_STREAM command. In
3269 * other words, the data of the first packet of this stream
3270 * can be received after a ROTATE_STREAM command.
3272 * The ROTATE_STREAM command changes the stream's path_name
3273 * to point to the "next" chunk. If a rotation is pending for
3274 * this stream, as indicated by "rotate_at_seq_num != -1ULL",
3275 * it means that we are still receiving data that belongs in the
3276 * stream's former path.
3278 * In this very specific case, we must ensure that the index
3279 * file is created in the streams's former path,
3282 * All other rotations beyond the first one are not affected
3283 * by this problem since the actual rotation operation creates
3284 * the new chunk's index file.
3286 stream_path
= stream
->rotate_at_seq_num
== -1ULL ?
3288 stream
->prev_path_name
;
3290 ret
= create_rotate_index_file(stream
, stream_path
);
3292 ERR("Failed to rotate index");
3293 /* Put self-ref for this index due to error. */
3294 relay_index_put(index
);
3300 if (relay_index_set_file(index
, stream
->index_file
, data_offset
)) {
3302 /* Put self-ref for this index due to error. */
3303 relay_index_put(index
);
3308 ret
= relay_index_try_flush(index
);
3310 tracefile_array_commit_seq(stream
->tfa
);
3311 stream
->index_received_seqcount
++;
3313 } else if (ret
> 0) {
3314 index
->total_size
= total_size
;
3321 * relay_index_try_flush is responsible for the self-reference
3322 * put of the index object on error.
3324 ERR("relay_index_try_flush error %d", ret
);
3331 static enum relay_connection_status
relay_process_data_receive_header(
3332 struct relay_connection
*conn
)
3335 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3336 struct data_connection_state_receive_header
*state
=
3337 &conn
->protocol
.data
.state
.receive_header
;
3338 struct lttcomm_relayd_data_hdr header
;
3339 struct relay_stream
*stream
;
3341 assert(state
->left_to_receive
!= 0);
3343 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3344 state
->header_reception_buffer
+ state
->received
,
3345 state
->left_to_receive
, MSG_DONTWAIT
);
3347 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3348 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
3349 status
= RELAY_CONNECTION_STATUS_ERROR
;
3352 } else if (ret
== 0) {
3353 /* Orderly shutdown. Not necessary to print an error. */
3354 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3355 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3360 assert(ret
<= state
->left_to_receive
);
3362 state
->left_to_receive
-= ret
;
3363 state
->received
+= ret
;
3365 if (state
->left_to_receive
> 0) {
3367 * Can't transition to the protocol's next state, wait to
3368 * receive the rest of the header.
3370 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3371 state
->received
, state
->left_to_receive
,
3376 /* Transition to next state: receiving the payload. */
3377 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3379 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3380 header
.circuit_id
= be64toh(header
.circuit_id
);
3381 header
.stream_id
= be64toh(header
.stream_id
);
3382 header
.data_size
= be32toh(header
.data_size
);
3383 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3384 header
.padding_size
= be32toh(header
.padding_size
);
3385 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3387 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3389 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3390 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3392 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3393 conn
->sock
->fd
, header
.circuit_id
,
3394 header
.stream_id
, header
.data_size
,
3395 header
.net_seq_num
, header
.padding_size
);
3397 stream
= stream_get_by_id(header
.stream_id
);
3399 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3401 /* Protocol error. */
3402 status
= RELAY_CONNECTION_STATUS_ERROR
;
3406 pthread_mutex_lock(&stream
->lock
);
3408 /* Check if a rotation is needed. */
3409 if (stream
->tracefile_size
> 0 &&
3410 (stream
->tracefile_size_current
+ header
.data_size
) >
3411 stream
->tracefile_size
) {
3412 uint64_t old_id
, new_id
;
3414 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
3415 tracefile_array_file_rotate(stream
->tfa
);
3417 /* new_id is updated by utils_rotate_stream_file. */
3420 ret
= utils_rotate_stream_file(stream
->path_name
,
3421 stream
->channel_name
, stream
->tracefile_size
,
3422 stream
->tracefile_count
, -1,
3423 -1, stream
->stream_fd
->fd
,
3424 &new_id
, &stream
->stream_fd
->fd
);
3426 ERR("Failed to rotate stream output file");
3427 status
= RELAY_CONNECTION_STATUS_ERROR
;
3428 goto end_stream_unlock
;
3432 * Reset current size because we just performed a stream
3435 stream
->tracefile_size_current
= 0;
3436 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= true;
3440 pthread_mutex_unlock(&stream
->lock
);
3446 static enum relay_connection_status
relay_process_data_receive_payload(
3447 struct relay_connection
*conn
)
3450 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3451 struct relay_stream
*stream
;
3452 struct data_connection_state_receive_payload
*state
=
3453 &conn
->protocol
.data
.state
.receive_payload
;
3454 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3455 char data_buffer
[chunk_size
];
3456 bool partial_recv
= false;
3457 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3458 uint64_t left_to_receive
= state
->left_to_receive
;
3459 struct relay_session
*session
;
3461 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3462 state
->header
.stream_id
, state
->header
.net_seq_num
,
3463 state
->received
, left_to_receive
);
3465 stream
= stream_get_by_id(state
->header
.stream_id
);
3467 /* Protocol error. */
3468 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3469 state
->header
.stream_id
);
3470 status
= RELAY_CONNECTION_STATUS_ERROR
;
3474 pthread_mutex_lock(&stream
->lock
);
3475 session
= stream
->trace
->session
;
3476 if (!conn
->session
) {
3477 ret
= connection_set_session(conn
, session
);
3479 status
= RELAY_CONNECTION_STATUS_ERROR
;
3480 goto end_stream_unlock
;
3485 * The size of the "chunk" received on any iteration is bounded by:
3486 * - the data left to receive,
3487 * - the data immediately available on the socket,
3488 * - the on-stack data buffer
3490 while (left_to_receive
> 0 && !partial_recv
) {
3492 size_t recv_size
= min(left_to_receive
, chunk_size
);
3494 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3495 recv_size
, MSG_DONTWAIT
);
3497 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3498 PERROR("Socket %d error", conn
->sock
->fd
);
3499 status
= RELAY_CONNECTION_STATUS_ERROR
;
3501 goto end_stream_unlock
;
3502 } else if (ret
== 0) {
3503 /* No more data ready to be consumed on socket. */
3504 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3505 state
->header
.stream_id
);
3506 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3508 } else if (ret
< (int) recv_size
) {
3510 * All the data available on the socket has been
3513 partial_recv
= true;
3518 /* Write data to stream output fd. */
3519 write_ret
= lttng_write(stream
->stream_fd
->fd
, data_buffer
,
3521 if (write_ret
< (ssize_t
) recv_size
) {
3522 ERR("Relay error writing data to file");
3523 status
= RELAY_CONNECTION_STATUS_ERROR
;
3524 goto end_stream_unlock
;
3527 left_to_receive
-= recv_size
;
3528 state
->received
+= recv_size
;
3529 state
->left_to_receive
= left_to_receive
;
3531 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
3532 write_ret
, stream
->stream_handle
);
3535 if (state
->left_to_receive
> 0) {
3537 * Did not receive all the data expected, wait for more data to
3538 * become available on the socket.
3540 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3541 state
->header
.stream_id
, state
->received
,
3542 state
->left_to_receive
);
3543 goto end_stream_unlock
;
3546 ret
= write_padding_to_file(stream
->stream_fd
->fd
,
3547 state
->header
.padding_size
);
3548 if ((int64_t) ret
< (int64_t) state
->header
.padding_size
) {
3549 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3550 stream
->stream_handle
,
3551 state
->header
.net_seq_num
, ret
);
3552 status
= RELAY_CONNECTION_STATUS_ERROR
;
3553 goto end_stream_unlock
;
3557 if (session_streams_have_index(session
)) {
3558 ret
= handle_index_data(stream
, state
->header
.net_seq_num
,
3559 state
->rotate_index
, &index_flushed
, state
->header
.data_size
+ state
->header
.padding_size
);
3561 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3562 stream
->stream_handle
,
3563 state
->header
.net_seq_num
, ret
);
3564 status
= RELAY_CONNECTION_STATUS_ERROR
;
3565 goto end_stream_unlock
;
3569 stream
->tracefile_size_current
+= state
->header
.data_size
+
3570 state
->header
.padding_size
;
3572 if (stream
->prev_data_seq
== -1ULL) {
3575 if (index_flushed
) {
3576 stream
->pos_after_last_complete_data_index
=
3577 stream
->tracefile_size_current
;
3578 stream
->prev_index_seq
= state
->header
.net_seq_num
;
3579 ret
= try_rotate_stream_index(stream
);
3581 goto end_stream_unlock
;
3585 stream
->prev_data_seq
= state
->header
.net_seq_num
;
3588 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3589 * contents of *state which are aliased (union) to the same location as
3590 * the new state. Don't use it beyond this point.
3592 connection_reset_protocol_state(conn
);
3595 ret
= try_rotate_stream_data(stream
);
3597 status
= RELAY_CONNECTION_STATUS_ERROR
;
3598 goto end_stream_unlock
;
3602 close_requested
= stream
->close_requested
;
3603 pthread_mutex_unlock(&stream
->lock
);
3604 if (close_requested
&& left_to_receive
== 0) {
3605 try_stream_close(stream
);
3609 pthread_mutex_lock(&session
->lock
);
3610 uatomic_set(&session
->new_streams
, 1);
3611 pthread_mutex_unlock(&session
->lock
);
3620 * relay_process_data: Process the data received on the data socket
3622 static enum relay_connection_status
relay_process_data(
3623 struct relay_connection
*conn
)
3625 enum relay_connection_status status
;
3627 switch (conn
->protocol
.data
.state_id
) {
3628 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3629 status
= relay_process_data_receive_header(conn
);
3631 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3632 status
= relay_process_data_receive_payload(conn
);
3635 ERR("Unexpected data connection communication state.");
3642 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3646 (void) lttng_poll_del(events
, pollfd
);
3648 ret
= close(pollfd
);
3650 ERR("Closing pollfd %d", pollfd
);
3654 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3655 int pollfd
, struct relay_connection
*conn
)
3657 const char *type_str
;
3659 switch (conn
->type
) {
3664 type_str
= "Control";
3666 case RELAY_VIEWER_COMMAND
:
3667 type_str
= "Viewer Command";
3669 case RELAY_VIEWER_NOTIFICATION
:
3670 type_str
= "Viewer Notification";
3673 type_str
= "Unknown";
3675 cleanup_connection_pollfd(events
, pollfd
);
3676 connection_put(conn
);
3677 DBG("%s connection closed with %d", type_str
, pollfd
);
3681 * This thread does the actual work
3683 static void *relay_thread_worker(void *data
)
3685 int ret
, err
= -1, last_seen_data_fd
= -1;
3687 struct lttng_poll_event events
;
3688 struct lttng_ht
*relay_connections_ht
;
3689 struct lttng_ht_iter iter
;
3690 struct relay_connection
*destroy_conn
= NULL
;
3692 DBG("[thread] Relay worker started");
3694 rcu_register_thread();
3696 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3698 if (testpoint(relayd_thread_worker
)) {
3699 goto error_testpoint
;
3702 health_code_update();
3704 /* table of connections indexed on socket */
3705 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3706 if (!relay_connections_ht
) {
3707 goto relay_connections_ht_error
;
3710 ret
= create_thread_poll_set(&events
, 2);
3712 goto error_poll_create
;
3715 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3722 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3724 health_code_update();
3726 /* Infinite blocking call, waiting for transmission */
3727 DBG3("Relayd worker thread polling...");
3728 health_poll_entry();
3729 ret
= lttng_poll_wait(&events
, -1);
3733 * Restart interrupted system call.
3735 if (errno
== EINTR
) {
3744 * Process control. The control connection is
3745 * prioritized so we don't starve it with high
3746 * throughput tracing data on the data connection.
3748 for (i
= 0; i
< nb_fd
; i
++) {
3749 /* Fetch once the poll data */
3750 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3751 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3753 health_code_update();
3755 /* Thread quit pipe has been closed. Killing thread. */
3756 ret
= check_thread_quit_pipe(pollfd
, revents
);
3762 /* Inspect the relay conn pipe for new connection */
3763 if (pollfd
== relay_conn_pipe
[0]) {
3764 if (revents
& LPOLLIN
) {
3765 struct relay_connection
*conn
;
3767 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3771 lttng_poll_add(&events
, conn
->sock
->fd
,
3772 LPOLLIN
| LPOLLRDHUP
);
3773 connection_ht_add(relay_connections_ht
, conn
);
3774 DBG("Connection socket %d added", conn
->sock
->fd
);
3775 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3776 ERR("Relay connection pipe error");
3779 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3783 struct relay_connection
*ctrl_conn
;
3785 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3786 /* If not found, there is a synchronization issue. */
3789 if (ctrl_conn
->type
== RELAY_DATA
) {
3790 if (revents
& LPOLLIN
) {
3792 * Flag the last seen data fd not deleted. It will be
3793 * used as the last seen fd if any fd gets deleted in
3796 last_notdel_data_fd
= pollfd
;
3798 goto put_ctrl_connection
;
3800 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3802 if (revents
& LPOLLIN
) {
3803 enum relay_connection_status status
;
3805 status
= relay_process_control(ctrl_conn
);
3806 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3808 * On socket error flag the session as aborted to force
3809 * the cleanup of its stream otherwise it can leak
3810 * during the lifetime of the relayd.
3812 * This prevents situations in which streams can be
3813 * left opened because an index was received, the
3814 * control connection is closed, and the data
3815 * connection is closed (uncleanly) before the packet's
3818 * Since the control connection encountered an error,
3819 * it is okay to be conservative and close the
3820 * session right now as we can't rely on the protocol
3821 * being respected anymore.
3823 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3824 session_abort(ctrl_conn
->session
);
3827 /* Clear the connection on error or close. */
3828 relay_thread_close_connection(&events
,
3833 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3834 relay_thread_close_connection(&events
,
3836 if (last_seen_data_fd
== pollfd
) {
3837 last_seen_data_fd
= last_notdel_data_fd
;
3840 ERR("Unexpected poll events %u for control sock %d",
3842 connection_put(ctrl_conn
);
3845 put_ctrl_connection
:
3846 connection_put(ctrl_conn
);
3851 * The last loop handled a control request, go back to poll to make
3852 * sure we prioritise the control socket.
3858 if (last_seen_data_fd
>= 0) {
3859 for (i
= 0; i
< nb_fd
; i
++) {
3860 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3862 health_code_update();
3864 if (last_seen_data_fd
== pollfd
) {
3871 /* Process data connection. */
3872 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3873 /* Fetch the poll data. */
3874 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3875 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3876 struct relay_connection
*data_conn
;
3878 health_code_update();
3881 /* No activity for this FD (poll implementation). */
3885 /* Skip the command pipe. It's handled in the first loop. */
3886 if (pollfd
== relay_conn_pipe
[0]) {
3890 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3892 /* Skip it. Might be removed before. */
3895 if (data_conn
->type
== RELAY_CONTROL
) {
3896 goto put_data_connection
;
3898 assert(data_conn
->type
== RELAY_DATA
);
3900 if (revents
& LPOLLIN
) {
3901 enum relay_connection_status status
;
3903 status
= relay_process_data(data_conn
);
3904 /* Connection closed or error. */
3905 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3907 * On socket error flag the session as aborted to force
3908 * the cleanup of its stream otherwise it can leak
3909 * during the lifetime of the relayd.
3911 * This prevents situations in which streams can be
3912 * left opened because an index was received, the
3913 * control connection is closed, and the data
3914 * connection is closed (uncleanly) before the packet's
3917 * Since the data connection encountered an error,
3918 * it is okay to be conservative and close the
3919 * session right now as we can't rely on the protocol
3920 * being respected anymore.
3922 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3923 session_abort(data_conn
->session
);
3925 relay_thread_close_connection(&events
, pollfd
,
3928 * Every goto restart call sets the last seen fd where
3929 * here we don't really care since we gracefully
3930 * continue the loop after the connection is deleted.
3933 /* Keep last seen port. */
3934 last_seen_data_fd
= pollfd
;
3935 connection_put(data_conn
);
3938 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3939 relay_thread_close_connection(&events
, pollfd
,
3942 ERR("Unknown poll events %u for data sock %d",
3945 put_data_connection
:
3946 connection_put(data_conn
);
3948 last_seen_data_fd
= -1;
3951 /* Normal exit, no error */
3956 /* Cleanup remaining connection object. */
3958 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
3961 health_code_update();
3963 session_abort(destroy_conn
->session
);
3966 * No need to grab another ref, because we own
3969 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
3974 lttng_poll_clean(&events
);
3976 lttng_ht_destroy(relay_connections_ht
);
3977 relay_connections_ht_error
:
3978 /* Close relay conn pipes */
3979 utils_close_pipe(relay_conn_pipe
);
3981 DBG("Thread exited with error");
3983 DBG("Worker thread cleanup complete");
3987 ERR("Health error occurred in %s", __func__
);
3989 health_unregister(health_relayd
);
3990 rcu_unregister_thread();
3991 lttng_relay_stop_threads();
3996 * Create the relay command pipe to wake thread_manage_apps.
3997 * Closed in cleanup().
3999 static int create_relay_conn_pipe(void)
4003 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
4011 int main(int argc
, char **argv
)
4013 int ret
= 0, retval
= 0;
4016 /* Parse arguments */
4018 if (set_options(argc
, argv
)) {
4023 if (set_signal_handler()) {
4028 /* Try to create directory if -o, --output is specified. */
4029 if (opt_output_path
) {
4030 if (*opt_output_path
!= '/') {
4031 ERR("Please specify an absolute path for -o, --output PATH");
4036 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
4039 ERR("Unable to create %s", opt_output_path
);
4046 if (opt_daemon
|| opt_background
) {
4049 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4057 * We are in the child. Make sure all other file
4058 * descriptors are closed, in case we are called with
4059 * more opened file descriptors than the standard ones.
4061 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4066 sessiond_trace_chunk_registry
= sessiond_trace_chunk_registry_create();
4067 if (!sessiond_trace_chunk_registry
) {
4068 ERR("Failed to initialize session daemon trace chunk registry");
4070 goto exit_sessiond_trace_chunk_registry
;
4073 /* Initialize thread health monitoring */
4074 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
4075 if (!health_relayd
) {
4076 PERROR("health_app_create error");
4078 goto exit_health_app_create
;
4081 /* Create thread quit pipe */
4082 if (init_thread_quit_pipe()) {
4084 goto exit_init_data
;
4087 /* Setup the thread apps communication pipe. */
4088 if (create_relay_conn_pipe()) {
4090 goto exit_init_data
;
4093 /* Init relay command queue. */
4094 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
4096 /* Initialize communication library */
4098 lttcomm_inet_init();
4100 /* tables of sessions indexed by session ID */
4101 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4104 goto exit_init_data
;
4107 /* tables of streams indexed by stream ID */
4108 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4109 if (!relay_streams_ht
) {
4111 goto exit_init_data
;
4114 /* tables of streams indexed by stream ID */
4115 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4116 if (!viewer_streams_ht
) {
4118 goto exit_init_data
;
4121 ret
= utils_create_pipe(health_quit_pipe
);
4124 goto exit_health_quit_pipe
;
4127 /* Create thread to manage the client socket */
4128 ret
= pthread_create(&health_thread
, default_pthread_attr(),
4129 thread_manage_health
, (void *) NULL
);
4132 PERROR("pthread_create health");
4134 goto exit_health_thread
;
4137 /* Setup the dispatcher thread */
4138 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
4139 relay_thread_dispatcher
, (void *) NULL
);
4142 PERROR("pthread_create dispatcher");
4144 goto exit_dispatcher_thread
;
4147 /* Setup the worker thread */
4148 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
4149 relay_thread_worker
, NULL
);
4152 PERROR("pthread_create worker");
4154 goto exit_worker_thread
;
4157 /* Setup the listener thread */
4158 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
4159 relay_thread_listener
, (void *) NULL
);
4162 PERROR("pthread_create listener");
4164 goto exit_listener_thread
;
4167 ret
= relayd_live_create(live_uri
);
4169 ERR("Starting live viewer threads");
4175 * This is where we start awaiting program completion (e.g. through
4176 * signal that asks threads to teardown).
4179 ret
= relayd_live_join();
4185 ret
= pthread_join(listener_thread
, &status
);
4188 PERROR("pthread_join listener_thread");
4192 exit_listener_thread
:
4193 ret
= pthread_join(worker_thread
, &status
);
4196 PERROR("pthread_join worker_thread");
4201 ret
= pthread_join(dispatcher_thread
, &status
);
4204 PERROR("pthread_join dispatcher_thread");
4207 exit_dispatcher_thread
:
4209 ret
= pthread_join(health_thread
, &status
);
4212 PERROR("pthread_join health_thread");
4217 utils_close_pipe(health_quit_pipe
);
4218 exit_health_quit_pipe
:
4221 health_app_destroy(health_relayd
);
4222 sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry
);
4223 exit_health_app_create
:
4224 exit_sessiond_trace_chunk_registry
:
4227 * Wait for all pending call_rcu work to complete before tearing
4228 * down data structures. call_rcu worker may be trying to
4229 * perform lookups in those structures.
4234 /* Ensure all prior call_rcu are done. */