2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
34 #include <sys/types.h>
37 #include <urcu/futex.h>
38 #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/defaults.h>
48 #include <common/daemonize.h>
49 #include <common/futex.h>
50 #include <common/sessiond-comm/sessiond-comm.h>
51 #include <common/sessiond-comm/inet.h>
52 #include <common/sessiond-comm/relayd.h>
53 #include <common/uri.h>
54 #include <common/utils.h>
55 #include <common/config/config.h>
58 #include "ctf-trace.h"
61 #include "lttng-relayd.h"
63 #include "health-relayd.h"
64 #include "testpoint.h"
65 #include "viewer-stream.h"
68 #include "connection.h"
70 /* command line options */
71 char *opt_output_path
;
72 static int opt_daemon
, opt_background
;
75 * We need to wait for listener and live listener threads, as well as
76 * health check thread, before being ready to signal readiness.
78 #define NR_LTTNG_RELAY_READY 3
79 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
80 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
81 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
83 static struct lttng_uri
*control_uri
;
84 static struct lttng_uri
*data_uri
;
85 static struct lttng_uri
*live_uri
;
89 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
90 static int tracing_group_name_override
;
92 const char * const config_section_name
= "relayd";
95 * Quit pipe for all threads. This permits a single cancellation point
96 * for all threads when receiving an event on the pipe.
98 int thread_quit_pipe
[2] = { -1, -1 };
101 * This pipe is used to inform the worker thread that a command is queued and
102 * ready to be processed.
104 static int relay_conn_pipe
[2] = { -1, -1 };
106 /* Shared between threads */
107 static int dispatch_thread_exit
;
109 static pthread_t listener_thread
;
110 static pthread_t dispatcher_thread
;
111 static pthread_t worker_thread
;
112 static pthread_t health_thread
;
114 static uint64_t last_relay_stream_id
;
117 * Relay command queue.
119 * The relay_thread_listener and relay_thread_dispatcher communicate with this
122 static struct relay_conn_queue relay_conn_queue
;
124 /* buffer allocated at startup, used to store the trace data */
125 static char *data_buffer
;
126 static unsigned int data_buffer_size
;
128 /* We need those values for the file/dir creation. */
129 static uid_t relayd_uid
;
130 static gid_t relayd_gid
;
132 /* Global relay stream hash table. */
133 struct lttng_ht
*relay_streams_ht
;
135 /* Global relay viewer stream hash table. */
136 struct lttng_ht
*viewer_streams_ht
;
138 /* Global hash table that stores relay index object. */
139 struct lttng_ht
*indexes_ht
;
141 /* Relayd health monitoring */
142 struct health_app
*health_relayd
;
144 static struct option long_options
[] = {
145 { "control-port", 1, 0, 'C', },
146 { "data-port", 1, 0, 'D', },
147 { "live-port", 1, 0, 'L', },
148 { "daemonize", 0, 0, 'd', },
149 { "background", 0, 0, 'b', },
150 { "group", 1, 0, 'g', },
151 { "help", 0, 0, 'h', },
152 { "output", 1, 0, 'o', },
153 { "verbose", 0, 0, 'v', },
154 { "config", 1, 0, 'f' },
158 static const char *config_ignore_options
[] = { "help", "config" };
161 * usage function on stderr
166 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
167 fprintf(stderr
, " -h, --help Display this usage.\n");
168 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
169 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
170 fprintf(stderr
, " -C, --control-port URL Control port listening.\n");
171 fprintf(stderr
, " -D, --data-port URL Data port listening.\n");
172 fprintf(stderr
, " -L, --live-port URL Live view port listening.\n");
173 fprintf(stderr
, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
174 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
175 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
176 fprintf(stderr
, " -f --config Load daemon configuration file\n");
180 * Take an option from the getopt output and set it in the right variable to be
183 * Return 0 on success else a negative value.
186 int set_option(int opt
, const char *arg
, const char *optname
)
192 fprintf(stderr
, "option %s", optname
);
194 fprintf(stderr
, " with arg %s\n", arg
);
198 ret
= uri_parse(arg
, &control_uri
);
200 ERR("Invalid control URI specified");
203 if (control_uri
->port
== 0) {
204 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
208 ret
= uri_parse(arg
, &data_uri
);
210 ERR("Invalid data URI specified");
213 if (data_uri
->port
== 0) {
214 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
218 ret
= uri_parse(arg
, &live_uri
);
220 ERR("Invalid live URI specified");
223 if (live_uri
->port
== 0) {
224 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
234 tracing_group_name
= strdup(arg
);
235 tracing_group_name_override
= 1;
241 ret
= asprintf(&opt_output_path
, "%s", arg
);
244 PERROR("asprintf opt_output_path");
249 /* Verbose level can increase using multiple -v */
251 lttng_opt_verbose
= config_parse_value(arg
);
253 lttng_opt_verbose
+= 1;
257 /* Unknown option or other error.
258 * Error is printed by getopt, just return */
271 * config_entry_handler_cb used to handle options read from a config file.
272 * See config_entry_handler_cb comment in common/config/config.h for the
273 * return value conventions.
276 int config_entry_handler(const struct config_entry
*entry
, void *unused
)
280 if (!entry
|| !entry
->name
|| !entry
->value
) {
285 /* Check if the option is to be ignored */
286 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
287 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
292 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
293 /* Ignore if entry name is not fully matched. */
294 if (strcmp(entry
->name
, long_options
[i
].name
)) {
299 * If the option takes no argument on the command line, we have to
300 * check if the value is "true". We support non-zero numeric values,
303 if (!long_options
[i
].has_arg
) {
304 ret
= config_parse_value(entry
->value
);
307 WARN("Invalid configuration value \"%s\" for option %s",
308 entry
->value
, entry
->name
);
310 /* False, skip boolean config option. */
315 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
319 WARN("Unrecognized option \"%s\" in daemon configuration file.",
327 int set_options(int argc
, char **argv
)
329 int c
, ret
= 0, option_index
= 0;
330 int orig_optopt
= optopt
, orig_optind
= optind
;
331 char *default_address
, *optstring
;
332 const char *config_path
= NULL
;
334 optstring
= utils_generate_optstring(long_options
,
335 sizeof(long_options
) / sizeof(struct option
));
341 /* Check for the --config option */
343 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
344 &option_index
)) != -1) {
348 } else if (c
!= 'f') {
352 config_path
= utils_expand_path(optarg
);
354 ERR("Failed to resolve path: %s", optarg
);
358 ret
= config_get_section_entries(config_path
, config_section_name
,
359 config_entry_handler
, NULL
);
362 ERR("Invalid configuration option at line %i", ret
);
368 /* Reset getopt's global state */
369 optopt
= orig_optopt
;
370 optind
= orig_optind
;
372 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
377 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
383 /* assign default values */
384 if (control_uri
== NULL
) {
385 ret
= asprintf(&default_address
,
386 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
387 DEFAULT_NETWORK_CONTROL_PORT
);
389 PERROR("asprintf default data address");
393 ret
= uri_parse(default_address
, &control_uri
);
394 free(default_address
);
396 ERR("Invalid control URI specified");
400 if (data_uri
== NULL
) {
401 ret
= asprintf(&default_address
,
402 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
403 DEFAULT_NETWORK_DATA_PORT
);
405 PERROR("asprintf default data address");
409 ret
= uri_parse(default_address
, &data_uri
);
410 free(default_address
);
412 ERR("Invalid data URI specified");
416 if (live_uri
== NULL
) {
417 ret
= asprintf(&default_address
,
418 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
419 DEFAULT_NETWORK_VIEWER_PORT
);
421 PERROR("asprintf default viewer control address");
425 ret
= uri_parse(default_address
, &live_uri
);
426 free(default_address
);
428 ERR("Invalid viewer control URI specified");
446 /* free the dynamically allocated opt_output_path */
447 free(opt_output_path
);
449 /* Close thread quit pipes */
450 utils_close_pipe(thread_quit_pipe
);
452 uri_free(control_uri
);
454 /* Live URI is freed in the live thread. */
456 if (tracing_group_name_override
) {
457 free((void *) tracing_group_name
);
462 * Write to writable pipe used to notify a thread.
465 int notify_thread_pipe(int wpipe
)
469 ret
= lttng_write(wpipe
, "!", 1);
471 PERROR("write poll pipe");
477 static void notify_health_quit_pipe(int *pipe
)
481 ret
= lttng_write(pipe
[1], "4", 1);
483 PERROR("write relay health quit");
488 * Stop all threads by closing the thread quit pipe.
491 void stop_threads(void)
495 /* Stopping all threads */
496 DBG("Terminating all threads");
497 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
499 ERR("write error on thread quit pipe");
502 notify_health_quit_pipe(health_quit_pipe
);
504 /* Dispatch thread */
505 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
506 futex_nto1_wake(&relay_conn_queue
.futex
);
510 * Signal handler for the daemon
512 * Simply stop all worker threads, leaving main() return gracefully after
513 * joining all threads and calling cleanup().
516 void sighandler(int sig
)
520 DBG("SIGPIPE caught");
523 DBG("SIGINT caught");
527 DBG("SIGTERM caught");
531 CMM_STORE_SHARED(recv_child_signal
, 1);
539 * Setup signal handler for :
540 * SIGINT, SIGTERM, SIGPIPE
543 int set_signal_handler(void)
549 if ((ret
= sigemptyset(&sigset
)) < 0) {
550 PERROR("sigemptyset");
554 sa
.sa_handler
= sighandler
;
557 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
562 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
567 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
572 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
577 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
582 void lttng_relay_notify_ready(void)
584 /* Notify the parent of the fork() process that we are ready. */
585 if (opt_daemon
|| opt_background
) {
586 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
587 kill(child_ppid
, SIGUSR1
);
593 * Init thread quit pipe.
595 * Return -1 on error or 0 if all pipes are created.
598 int init_thread_quit_pipe(void)
602 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
608 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
611 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
615 if (events
== NULL
|| size
== 0) {
620 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
626 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
638 * Check if the thread quit pipe was triggered.
640 * Return 1 if it was triggered else 0;
643 int check_thread_quit_pipe(int fd
, uint32_t events
)
645 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
653 * Create and init socket from uri.
656 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
659 struct lttcomm_sock
*sock
= NULL
;
661 sock
= lttcomm_alloc_sock_from_uri(uri
);
663 ERR("Allocating socket");
667 ret
= lttcomm_create_sock(sock
);
671 DBG("Listening on sock %d", sock
->fd
);
673 ret
= sock
->ops
->bind(sock
);
678 ret
= sock
->ops
->listen(sock
, -1);
688 lttcomm_destroy_sock(sock
);
694 * Return nonzero if stream needs to be closed.
697 int close_stream_check(struct relay_stream
*stream
)
699 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
701 * We are about to close the stream so set the data pending flag to 1
702 * which will make the end data pending command skip the stream which
703 * is now closed and ready. Note that after proceeding to a file close,
704 * the written file is ready for reading.
706 stream
->data_pending_check_done
= 1;
712 static void try_close_stream(struct relay_session
*session
,
713 struct relay_stream
*stream
)
716 struct ctf_trace
*ctf_trace
;
721 if (!close_stream_check(stream
)) {
722 /* Can't close it, not ready for that. */
726 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
,
730 pthread_mutex_lock(&session
->viewer_ready_lock
);
731 ctf_trace
->invalid_flag
= 1;
732 pthread_mutex_unlock(&session
->viewer_ready_lock
);
734 ret
= stream_close(session
, stream
);
735 if (ret
|| session
->snapshot
) {
736 /* Already close thus the ctf trace is being or has been destroyed. */
740 ctf_trace_try_destroy(session
, ctf_trace
);
747 * This thread manages the listening for new connections on the network
750 void *relay_thread_listener(void *data
)
752 int i
, ret
, pollfd
, err
= -1;
753 uint32_t revents
, nb_fd
;
754 struct lttng_poll_event events
;
755 struct lttcomm_sock
*control_sock
, *data_sock
;
757 DBG("[thread] Relay listener started");
759 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
761 health_code_update();
763 control_sock
= relay_init_sock(control_uri
);
765 goto error_sock_control
;
768 data_sock
= relay_init_sock(data_uri
);
770 goto error_sock_relay
;
774 * Pass 3 as size here for the thread quit pipe, control and data socket.
776 ret
= create_thread_poll_set(&events
, 3);
778 goto error_create_poll
;
781 /* Add the control socket */
782 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
787 /* Add the data socket */
788 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
793 lttng_relay_notify_ready();
795 if (testpoint(relayd_thread_listener
)) {
796 goto error_testpoint
;
800 health_code_update();
802 DBG("Listener accepting connections");
806 ret
= lttng_poll_wait(&events
, -1);
810 * Restart interrupted system call.
812 if (errno
== EINTR
) {
820 DBG("Relay new connection received");
821 for (i
= 0; i
< nb_fd
; i
++) {
822 health_code_update();
824 /* Fetch once the poll data */
825 revents
= LTTNG_POLL_GETEV(&events
, i
);
826 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
828 /* Thread quit pipe has been closed. Killing thread. */
829 ret
= check_thread_quit_pipe(pollfd
, revents
);
835 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
836 ERR("socket poll error");
838 } else if (revents
& LPOLLIN
) {
840 * Get allocated in this thread, enqueued to a global queue,
841 * dequeued and freed in the worker thread.
844 struct relay_connection
*new_conn
;
845 struct lttcomm_sock
*newsock
;
847 new_conn
= connection_create();
852 if (pollfd
== data_sock
->fd
) {
853 new_conn
->type
= RELAY_DATA
;
854 newsock
= data_sock
->ops
->accept(data_sock
);
855 DBG("Relay data connection accepted, socket %d",
858 assert(pollfd
== control_sock
->fd
);
859 new_conn
->type
= RELAY_CONTROL
;
860 newsock
= control_sock
->ops
->accept(control_sock
);
861 DBG("Relay control connection accepted, socket %d",
865 PERROR("accepting sock");
866 connection_free(new_conn
);
870 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
873 PERROR("setsockopt inet");
874 lttcomm_destroy_sock(newsock
);
875 connection_free(new_conn
);
878 new_conn
->sock
= newsock
;
880 /* Enqueue request for the dispatcher thread. */
881 cds_wfq_enqueue(&relay_conn_queue
.queue
, &new_conn
->qnode
);
884 * Wake the dispatch queue futex. Implicit memory barrier with
885 * the exchange in cds_wfq_enqueue.
887 futex_nto1_wake(&relay_conn_queue
.futex
);
896 lttng_poll_clean(&events
);
898 if (data_sock
->fd
>= 0) {
899 ret
= data_sock
->ops
->close(data_sock
);
904 lttcomm_destroy_sock(data_sock
);
906 if (control_sock
->fd
>= 0) {
907 ret
= control_sock
->ops
->close(control_sock
);
912 lttcomm_destroy_sock(control_sock
);
916 ERR("Health error occurred in %s", __func__
);
918 health_unregister(health_relayd
);
919 DBG("Relay listener thread cleanup complete");
925 * This thread manages the dispatching of the requests to worker threads
928 void *relay_thread_dispatcher(void *data
)
932 struct cds_wfq_node
*node
;
933 struct relay_connection
*new_conn
= NULL
;
935 DBG("[thread] Relay dispatcher started");
937 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
939 if (testpoint(relayd_thread_dispatcher
)) {
940 goto error_testpoint
;
943 health_code_update();
945 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
946 health_code_update();
948 /* Atomically prepare the queue futex */
949 futex_nto1_prepare(&relay_conn_queue
.futex
);
952 health_code_update();
954 /* Dequeue commands */
955 node
= cds_wfq_dequeue_blocking(&relay_conn_queue
.queue
);
957 DBG("Woken up but nothing in the relay command queue");
958 /* Continue thread execution */
961 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
963 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
966 * Inform worker thread of the new request. This call is blocking
967 * so we can be assured that the data will be read at some point in
968 * time or wait to the end of the world :)
970 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
972 PERROR("write connection pipe");
973 connection_destroy(new_conn
);
976 } while (node
!= NULL
);
978 /* Futex wait on queue. Blocking call on futex() */
980 futex_nto1_wait(&relay_conn_queue
.futex
);
984 /* Normal exit, no error */
991 ERR("Health error occurred in %s", __func__
);
993 health_unregister(health_relayd
);
994 DBG("Dispatch thread dying");
999 static void try_close_streams(struct relay_session
*session
)
1001 struct ctf_trace
*ctf_trace
;
1002 struct lttng_ht_iter iter
;
1006 pthread_mutex_lock(&session
->viewer_ready_lock
);
1008 cds_lfht_for_each_entry(session
->ctf_traces_ht
->ht
, &iter
.iter
, ctf_trace
,
1010 struct relay_stream
*stream
;
1012 /* Close streams. */
1013 cds_list_for_each_entry(stream
, &ctf_trace
->stream_list
, trace_list
) {
1014 stream_close(session
, stream
);
1017 ctf_trace
->invalid_flag
= 1;
1018 ctf_trace_try_destroy(session
, ctf_trace
);
1021 pthread_mutex_unlock(&session
->viewer_ready_lock
);
1025 * Try to destroy a session within a connection.
1027 static void destroy_session(struct relay_session
*session
,
1028 struct lttng_ht
*sessions_ht
)
1031 assert(sessions_ht
);
1033 /* Indicate that this session can be destroyed from now on. */
1034 session
->close_flag
= 1;
1036 try_close_streams(session
);
1039 * This will try to delete and destroy the session if no viewer is attached
1040 * to it meaning the refcount is down to zero.
1042 session_try_destroy(sessions_ht
, session
);
1046 * Copy index data from the control port to a given index object.
1048 static void copy_index_control_data(struct relay_index
*index
,
1049 struct lttcomm_relayd_index
*data
)
1055 * The index on disk is encoded in big endian, so we don't need to convert
1056 * the data received on the network. The data_offset value is NEVER
1057 * modified here and is updated by the data thread.
1059 index
->index_data
.packet_size
= data
->packet_size
;
1060 index
->index_data
.content_size
= data
->content_size
;
1061 index
->index_data
.timestamp_begin
= data
->timestamp_begin
;
1062 index
->index_data
.timestamp_end
= data
->timestamp_end
;
1063 index
->index_data
.events_discarded
= data
->events_discarded
;
1064 index
->index_data
.stream_id
= data
->stream_id
;
1068 * Handle the RELAYD_CREATE_SESSION command.
1070 * On success, send back the session id or else return a negative value.
1073 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
1074 struct relay_connection
*conn
)
1076 int ret
= 0, send_ret
;
1077 struct relay_session
*session
;
1078 struct lttcomm_relayd_status_session reply
;
1083 memset(&reply
, 0, sizeof(reply
));
1085 session
= session_create();
1090 session
->minor
= conn
->minor
;
1091 session
->major
= conn
->major
;
1092 conn
->session_id
= session
->id
;
1093 conn
->session
= session
;
1095 reply
.session_id
= htobe64(session
->id
);
1097 switch (conn
->minor
) {
1102 case 4: /* LTTng sessiond 2.4 */
1104 ret
= cmd_create_session_2_4(conn
, session
);
1107 lttng_ht_add_unique_u64(conn
->sessions_ht
, &session
->session_n
);
1108 DBG("Created session %" PRIu64
, session
->id
);
1112 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1114 reply
.ret_code
= htobe32(LTTNG_OK
);
1117 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1119 ERR("Relayd sending session id");
1127 * When we have received all the streams and the metadata for a channel,
1128 * we make them visible to the viewer threads.
1131 void set_viewer_ready_flag(struct relay_connection
*conn
)
1133 struct relay_stream
*stream
, *tmp_stream
;
1135 pthread_mutex_lock(&conn
->session
->viewer_ready_lock
);
1136 cds_list_for_each_entry_safe(stream
, tmp_stream
, &conn
->recv_head
,
1138 stream
->viewer_ready
= 1;
1139 cds_list_del(&stream
->recv_list
);
1141 pthread_mutex_unlock(&conn
->session
->viewer_ready_lock
);
1146 * Add a recv handle node to the connection recv list with the given stream
1147 * handle. A new node is allocated thus must be freed when the node is deleted
1150 static void queue_stream(struct relay_stream
*stream
,
1151 struct relay_connection
*conn
)
1156 cds_list_add(&stream
->recv_list
, &conn
->recv_head
);
1160 * relay_add_stream: allocate a new stream for a session
1163 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1164 struct relay_connection
*conn
)
1167 struct relay_session
*session
= conn
->session
;
1168 struct relay_stream
*stream
= NULL
;
1169 struct lttcomm_relayd_status_stream reply
;
1170 struct ctf_trace
*trace
;
1172 if (!session
|| conn
->version_check_done
== 0) {
1173 ERR("Trying to add a stream before version check");
1175 goto end_no_session
;
1178 stream
= zmalloc(sizeof(struct relay_stream
));
1179 if (stream
== NULL
) {
1180 PERROR("relay stream zmalloc");
1182 goto end_no_session
;
1185 switch (conn
->minor
) {
1186 case 1: /* LTTng sessiond 2.1 */
1187 ret
= cmd_recv_stream_2_1(conn
, stream
);
1189 case 2: /* LTTng sessiond 2.2 */
1191 ret
= cmd_recv_stream_2_2(conn
, stream
);
1195 goto err_free_stream
;
1199 stream
->stream_handle
= ++last_relay_stream_id
;
1200 stream
->prev_seq
= -1ULL;
1201 stream
->session_id
= session
->id
;
1202 stream
->index_fd
= -1;
1203 stream
->read_index_fd
= -1;
1204 lttng_ht_node_init_u64(&stream
->node
, stream
->stream_handle
);
1205 pthread_mutex_init(&stream
->lock
, NULL
);
1207 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
);
1209 ERR("relay creating output directory");
1214 * No need to use run_as API here because whatever we receives, the relayd
1215 * uses its own credentials for the stream files.
1217 ret
= utils_create_stream_file(stream
->path_name
, stream
->channel_name
,
1218 stream
->tracefile_size
, 0, relayd_uid
, relayd_gid
, NULL
);
1220 ERR("Create output file");
1224 if (stream
->tracefile_size
) {
1225 DBG("Tracefile %s/%s_0 created", stream
->path_name
, stream
->channel_name
);
1227 DBG("Tracefile %s/%s created", stream
->path_name
, stream
->channel_name
);
1230 trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
, stream
->path_name
);
1232 trace
= ctf_trace_create(stream
->path_name
);
1237 ctf_trace_add(session
->ctf_traces_ht
, trace
);
1239 ctf_trace_get_ref(trace
);
1241 if (!strncmp(stream
->channel_name
, DEFAULT_METADATA_NAME
, NAME_MAX
)) {
1242 stream
->metadata_flag
= 1;
1243 /* Assign quick reference to the metadata stream in the trace. */
1244 trace
->metadata_stream
= stream
;
1248 * Add the stream in the recv list of the connection. Once the end stream
1249 * message is received, this list is emptied and streams are set with the
1250 * viewer ready flag.
1252 queue_stream(stream
, conn
);
1255 * Both in the ctf_trace object and the global stream ht since the data
1256 * side of the relayd does not have the concept of session.
1258 lttng_ht_add_unique_u64(relay_streams_ht
, &stream
->node
);
1259 cds_list_add_tail(&stream
->trace_list
, &trace
->stream_list
);
1261 session
->stream_count
++;
1263 DBG("Relay new stream added %s with ID %" PRIu64
, stream
->channel_name
,
1264 stream
->stream_handle
);
1267 memset(&reply
, 0, sizeof(reply
));
1268 reply
.handle
= htobe64(stream
->stream_handle
);
1269 /* send the session id to the client or a negative return code on error */
1271 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1272 /* stream was not properly added to the ht, so free it */
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);
1281 ERR("Relay sending stream id");
1290 free(stream
->path_name
);
1291 free(stream
->channel_name
);
1297 * relay_close_stream: close a specific stream
1300 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1301 struct relay_connection
*conn
)
1304 struct relay_session
*session
= conn
->session
;
1305 struct lttcomm_relayd_close_stream stream_info
;
1306 struct lttcomm_relayd_generic_reply reply
;
1307 struct relay_stream
*stream
;
1309 DBG("Close stream received");
1311 if (!session
|| conn
->version_check_done
== 0) {
1312 ERR("Trying to close a stream before version check");
1314 goto end_no_session
;
1317 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
1318 sizeof(struct lttcomm_relayd_close_stream
), 0);
1319 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1321 /* Orderly shutdown. Not necessary to print an error. */
1322 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1324 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1327 goto end_no_session
;
1331 stream
= stream_find_by_id(relay_streams_ht
,
1332 be64toh(stream_info
.stream_id
));
1338 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1339 stream
->close_flag
= 1;
1340 session
->stream_count
--;
1341 assert(session
->stream_count
>= 0);
1343 /* Check if we can close it or else the data will do it. */
1344 try_close_stream(session
, stream
);
1349 memset(&reply
, 0, sizeof(reply
));
1351 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1353 reply
.ret_code
= htobe32(LTTNG_OK
);
1355 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1356 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1358 ERR("Relay sending stream id");
1367 * relay_unknown_command: send -1 if received unknown command
1370 void relay_unknown_command(struct relay_connection
*conn
)
1372 struct lttcomm_relayd_generic_reply reply
;
1375 memset(&reply
, 0, sizeof(reply
));
1376 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1377 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1378 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1380 ERR("Relay sending unknown command");
1385 * relay_start: send an acknowledgment to the client to tell if we are
1386 * ready to receive data. We are ready if a session is established.
1389 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1390 struct relay_connection
*conn
)
1392 int ret
= htobe32(LTTNG_OK
);
1393 struct lttcomm_relayd_generic_reply reply
;
1394 struct relay_session
*session
= conn
->session
;
1397 DBG("Trying to start the streaming without a session established");
1398 ret
= htobe32(LTTNG_ERR_UNK
);
1401 memset(&reply
, 0, sizeof(reply
));
1402 reply
.ret_code
= ret
;
1403 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1404 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1406 ERR("Relay sending start ack");
1413 * Append padding to the file pointed by the file descriptor fd.
1415 static int write_padding_to_file(int fd
, uint32_t size
)
1424 zeros
= zmalloc(size
);
1425 if (zeros
== NULL
) {
1426 PERROR("zmalloc zeros for padding");
1431 ret
= lttng_write(fd
, zeros
, size
);
1433 PERROR("write padding to file");
1443 * relay_recv_metadata: receive the metada for the session.
1446 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1447 struct relay_connection
*conn
)
1449 int ret
= htobe32(LTTNG_OK
);
1451 struct relay_session
*session
= conn
->session
;
1452 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1453 struct relay_stream
*metadata_stream
;
1454 uint64_t data_size
, payload_size
;
1455 struct ctf_trace
*ctf_trace
;
1458 ERR("Metadata sent before version check");
1463 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1464 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1465 ERR("Incorrect data size");
1469 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1471 if (data_buffer_size
< data_size
) {
1472 /* In case the realloc fails, we can free the memory */
1475 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1476 if (!tmp_data_ptr
) {
1477 ERR("Allocating data buffer");
1482 data_buffer
= tmp_data_ptr
;
1483 data_buffer_size
= data_size
;
1485 memset(data_buffer
, 0, data_size
);
1486 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1487 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
1488 if (ret
< 0 || ret
!= data_size
) {
1490 /* Orderly shutdown. Not necessary to print an error. */
1491 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1493 ERR("Relay didn't receive the whole metadata");
1498 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1501 metadata_stream
= stream_find_by_id(relay_streams_ht
,
1502 be64toh(metadata_struct
->stream_id
));
1503 if (!metadata_stream
) {
1508 size_ret
= lttng_write(metadata_stream
->fd
, metadata_struct
->payload
,
1510 if (size_ret
< payload_size
) {
1511 ERR("Relay error writing metadata on file");
1516 ret
= write_padding_to_file(metadata_stream
->fd
,
1517 be32toh(metadata_struct
->padding_size
));
1522 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
,
1523 metadata_stream
->path_name
);
1525 ctf_trace
->metadata_received
+=
1526 payload_size
+ be32toh(metadata_struct
->padding_size
);
1528 DBG2("Relay metadata written");
1537 * relay_send_version: send relayd version number
1540 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1541 struct relay_connection
*conn
)
1544 struct lttcomm_relayd_version reply
, msg
;
1548 conn
->version_check_done
= 1;
1550 /* Get version from the other side. */
1551 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1552 if (ret
< 0 || ret
!= sizeof(msg
)) {
1554 /* Orderly shutdown. Not necessary to print an error. */
1555 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1557 ERR("Relay failed to receive the version values.");
1563 memset(&reply
, 0, sizeof(reply
));
1564 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1565 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1567 /* Major versions must be the same */
1568 if (reply
.major
!= be32toh(msg
.major
)) {
1569 DBG("Incompatible major versions (%u vs %u), deleting session",
1570 reply
.major
, be32toh(msg
.major
));
1571 destroy_session(conn
->session
, conn
->sessions_ht
);
1576 conn
->major
= reply
.major
;
1577 /* We adapt to the lowest compatible version */
1578 if (reply
.minor
<= be32toh(msg
.minor
)) {
1579 conn
->minor
= reply
.minor
;
1581 conn
->minor
= be32toh(msg
.minor
);
1584 reply
.major
= htobe32(reply
.major
);
1585 reply
.minor
= htobe32(reply
.minor
);
1586 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1587 sizeof(struct lttcomm_relayd_version
), 0);
1589 ERR("Relay sending version");
1592 DBG("Version check done using protocol %u.%u", conn
->major
,
1600 * Check for data pending for a given stream id from the session daemon.
1603 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1604 struct relay_connection
*conn
)
1606 struct relay_session
*session
= conn
->session
;
1607 struct lttcomm_relayd_data_pending msg
;
1608 struct lttcomm_relayd_generic_reply reply
;
1609 struct relay_stream
*stream
;
1611 uint64_t last_net_seq_num
, stream_id
;
1613 DBG("Data pending command received");
1615 if (!session
|| conn
->version_check_done
== 0) {
1616 ERR("Trying to check for data before version check");
1618 goto end_no_session
;
1621 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1622 if (ret
< sizeof(msg
)) {
1624 /* Orderly shutdown. Not necessary to print an error. */
1625 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1627 ERR("Relay didn't receive valid data_pending struct size : %d",
1631 goto end_no_session
;
1634 stream_id
= be64toh(msg
.stream_id
);
1635 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1638 stream
= stream_find_by_id(relay_streams_ht
, stream_id
);
1639 if (stream
== NULL
) {
1644 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1645 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1648 /* Avoid wrapping issue */
1649 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1650 /* Data has in fact been written and is NOT pending */
1653 /* Data still being streamed thus pending */
1657 /* Pending check is now done. */
1658 stream
->data_pending_check_done
= 1;
1663 memset(&reply
, 0, sizeof(reply
));
1664 reply
.ret_code
= htobe32(ret
);
1665 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1667 ERR("Relay data pending ret code failed");
1675 * Wait for the control socket to reach a quiescent state.
1677 * Note that for now, when receiving this command from the session daemon, this
1678 * means that every subsequent commands or data received on the control socket
1679 * has been handled. So, this is why we simply return OK here.
1682 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1683 struct relay_connection
*conn
)
1687 struct relay_stream
*stream
;
1688 struct lttng_ht_iter iter
;
1689 struct lttcomm_relayd_quiescent_control msg
;
1690 struct lttcomm_relayd_generic_reply reply
;
1692 DBG("Checking quiescent state on control socket");
1694 if (!conn
->session
|| conn
->version_check_done
== 0) {
1695 ERR("Trying to check for data before version check");
1697 goto end_no_session
;
1700 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1701 if (ret
< sizeof(msg
)) {
1703 /* Orderly shutdown. Not necessary to print an error. */
1704 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1706 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1710 goto end_no_session
;
1713 stream_id
= be64toh(msg
.stream_id
);
1716 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1718 if (stream
->stream_handle
== stream_id
) {
1719 stream
->data_pending_check_done
= 1;
1720 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1727 memset(&reply
, 0, sizeof(reply
));
1728 reply
.ret_code
= htobe32(LTTNG_OK
);
1729 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1731 ERR("Relay data quiescent control ret code failed");
1739 * Initialize a data pending command. This means that a client is about to ask
1740 * for data pending for each stream he/she holds. Simply iterate over all
1741 * streams of a session and set the data_pending_check_done flag.
1743 * This command returns to the client a LTTNG_OK code.
1746 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1747 struct relay_connection
*conn
)
1750 struct lttng_ht_iter iter
;
1751 struct lttcomm_relayd_begin_data_pending msg
;
1752 struct lttcomm_relayd_generic_reply reply
;
1753 struct relay_stream
*stream
;
1754 uint64_t session_id
;
1759 DBG("Init streams for data pending");
1761 if (!conn
->session
|| conn
->version_check_done
== 0) {
1762 ERR("Trying to check for data before version check");
1764 goto end_no_session
;
1767 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1768 if (ret
< sizeof(msg
)) {
1770 /* Orderly shutdown. Not necessary to print an error. */
1771 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1773 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1777 goto end_no_session
;
1780 session_id
= be64toh(msg
.session_id
);
1783 * Iterate over all streams to set the begin data pending flag. For now, the
1784 * streams are indexed by stream handle so we have to iterate over all
1785 * streams to find the one associated with the right session_id.
1788 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1790 if (stream
->session_id
== session_id
) {
1791 stream
->data_pending_check_done
= 0;
1792 DBG("Set begin data pending flag to stream %" PRIu64
,
1793 stream
->stream_handle
);
1798 memset(&reply
, 0, sizeof(reply
));
1799 /* All good, send back reply. */
1800 reply
.ret_code
= htobe32(LTTNG_OK
);
1802 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1804 ERR("Relay begin data pending send reply failed");
1812 * End data pending command. This will check, for a given session id, if each
1813 * stream associated with it has its data_pending_check_done flag set. If not,
1814 * this means that the client lost track of the stream but the data is still
1815 * being streamed on our side. In this case, we inform the client that data is
1818 * Return to the client if there is data in flight or not with a ret_code.
1821 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1822 struct relay_connection
*conn
)
1825 struct lttng_ht_iter iter
;
1826 struct lttcomm_relayd_end_data_pending msg
;
1827 struct lttcomm_relayd_generic_reply reply
;
1828 struct relay_stream
*stream
;
1829 uint64_t session_id
;
1830 uint32_t is_data_inflight
= 0;
1835 DBG("End data pending command");
1837 if (!conn
->session
|| conn
->version_check_done
== 0) {
1838 ERR("Trying to check for data before version check");
1840 goto end_no_session
;
1843 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1844 if (ret
< sizeof(msg
)) {
1846 /* Orderly shutdown. Not necessary to print an error. */
1847 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1849 ERR("Relay didn't receive valid end data_pending struct size: %d",
1853 goto end_no_session
;
1856 session_id
= be64toh(msg
.session_id
);
1858 /* Iterate over all streams to see if the begin data pending flag is set. */
1860 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1862 if (stream
->session_id
== session_id
&&
1863 !stream
->data_pending_check_done
&& !stream
->terminated_flag
) {
1864 is_data_inflight
= 1;
1865 DBG("Data is still in flight for stream %" PRIu64
,
1866 stream
->stream_handle
);
1872 memset(&reply
, 0, sizeof(reply
));
1873 /* All good, send back reply. */
1874 reply
.ret_code
= htobe32(is_data_inflight
);
1876 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1878 ERR("Relay end data pending send reply failed");
1886 * Receive an index for a specific stream.
1888 * Return 0 on success else a negative value.
1891 int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1892 struct relay_connection
*conn
)
1894 int ret
, send_ret
, index_created
= 0;
1895 struct relay_session
*session
= conn
->session
;
1896 struct lttcomm_relayd_index index_info
;
1897 struct relay_index
*index
, *wr_index
= NULL
;
1898 struct lttcomm_relayd_generic_reply reply
;
1899 struct relay_stream
*stream
;
1900 uint64_t net_seq_num
;
1904 DBG("Relay receiving index");
1906 if (!session
|| conn
->version_check_done
== 0) {
1907 ERR("Trying to close a stream before version check");
1909 goto end_no_session
;
1912 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &index_info
,
1913 sizeof(index_info
), 0);
1914 if (ret
< sizeof(index_info
)) {
1916 /* Orderly shutdown. Not necessary to print an error. */
1917 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1919 ERR("Relay didn't receive valid index struct size : %d", ret
);
1922 goto end_no_session
;
1925 net_seq_num
= be64toh(index_info
.net_seq_num
);
1928 stream
= stream_find_by_id(relay_streams_ht
,
1929 be64toh(index_info
.relay_stream_id
));
1932 goto end_rcu_unlock
;
1935 /* Live beacon handling */
1936 if (index_info
.packet_size
== 0) {
1937 DBG("Received live beacon for stream %" PRIu64
, stream
->stream_handle
);
1940 * Only flag a stream inactive when it has already received data.
1942 if (stream
->total_index_received
> 0) {
1943 stream
->beacon_ts_end
= be64toh(index_info
.timestamp_end
);
1946 goto end_rcu_unlock
;
1948 stream
->beacon_ts_end
= -1ULL;
1951 index
= relay_index_find(stream
->stream_handle
, net_seq_num
);
1953 /* A successful creation will add the object to the HT. */
1954 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
1956 goto end_rcu_unlock
;
1961 copy_index_control_data(index
, &index_info
);
1963 if (index_created
) {
1965 * Try to add the relay index object to the hash table. If an object
1966 * already exist, destroy back the index created, set the data in this
1967 * object and write it on disk.
1969 relay_index_add(index
, &wr_index
);
1971 copy_index_control_data(wr_index
, &index_info
);
1975 /* The index already exists so write it on disk. */
1979 /* Do we have a writable ready index to write on disk. */
1981 ret
= relay_index_write(wr_index
->fd
, wr_index
);
1983 goto end_rcu_unlock
;
1985 stream
->total_index_received
++;
1991 memset(&reply
, 0, sizeof(reply
));
1993 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1995 reply
.ret_code
= htobe32(LTTNG_OK
);
1997 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1999 ERR("Relay sending close index id reply");
2008 * Receive the streams_sent message.
2010 * Return 0 on success else a negative value.
2013 int relay_streams_sent(struct lttcomm_relayd_hdr
*recv_hdr
,
2014 struct relay_connection
*conn
)
2017 struct lttcomm_relayd_generic_reply reply
;
2021 DBG("Relay receiving streams_sent");
2023 if (!conn
->session
|| conn
->version_check_done
== 0) {
2024 ERR("Trying to close a stream before version check");
2026 goto end_no_session
;
2030 * Flag every pending stream in the connection recv list that they are
2031 * ready to be used by the viewer.
2033 set_viewer_ready_flag(conn
);
2036 * Inform the viewer that there are new streams in the session.
2038 if (conn
->session
->viewer_refcount
) {
2039 uatomic_set(&conn
->session
->new_streams
, 1);
2042 memset(&reply
, 0, sizeof(reply
));
2043 reply
.ret_code
= htobe32(LTTNG_OK
);
2044 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2046 ERR("Relay sending sent_stream reply");
2058 * Process the commands received on the control socket
2061 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
2062 struct relay_connection
*conn
)
2066 switch (be32toh(recv_hdr
->cmd
)) {
2067 case RELAYD_CREATE_SESSION
:
2068 ret
= relay_create_session(recv_hdr
, conn
);
2070 case RELAYD_ADD_STREAM
:
2071 ret
= relay_add_stream(recv_hdr
, conn
);
2073 case RELAYD_START_DATA
:
2074 ret
= relay_start(recv_hdr
, conn
);
2076 case RELAYD_SEND_METADATA
:
2077 ret
= relay_recv_metadata(recv_hdr
, conn
);
2079 case RELAYD_VERSION
:
2080 ret
= relay_send_version(recv_hdr
, conn
);
2082 case RELAYD_CLOSE_STREAM
:
2083 ret
= relay_close_stream(recv_hdr
, conn
);
2085 case RELAYD_DATA_PENDING
:
2086 ret
= relay_data_pending(recv_hdr
, conn
);
2088 case RELAYD_QUIESCENT_CONTROL
:
2089 ret
= relay_quiescent_control(recv_hdr
, conn
);
2091 case RELAYD_BEGIN_DATA_PENDING
:
2092 ret
= relay_begin_data_pending(recv_hdr
, conn
);
2094 case RELAYD_END_DATA_PENDING
:
2095 ret
= relay_end_data_pending(recv_hdr
, conn
);
2097 case RELAYD_SEND_INDEX
:
2098 ret
= relay_recv_index(recv_hdr
, conn
);
2100 case RELAYD_STREAMS_SENT
:
2101 ret
= relay_streams_sent(recv_hdr
, conn
);
2103 case RELAYD_UPDATE_SYNC_INFO
:
2105 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
2106 relay_unknown_command(conn
);
2116 * Handle index for a data stream.
2118 * RCU read side lock MUST be acquired.
2120 * Return 0 on success else a negative value.
2122 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2125 int ret
= 0, index_created
= 0;
2126 uint64_t stream_id
, data_offset
;
2127 struct relay_index
*index
, *wr_index
= NULL
;
2131 stream_id
= stream
->stream_handle
;
2132 /* Get data offset because we are about to update the index. */
2133 data_offset
= htobe64(stream
->tracefile_size_current
);
2136 * Lookup for an existing index for that stream id/sequence number. If on
2137 * exists, the control thread already received the data for it thus we need
2138 * to write it on disk.
2140 index
= relay_index_find(stream_id
, net_seq_num
);
2142 /* A successful creation will add the object to the HT. */
2143 index
= relay_index_create(stream_id
, net_seq_num
);
2151 if (rotate_index
|| stream
->index_fd
< 0) {
2152 index
->to_close_fd
= stream
->index_fd
;
2153 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
2154 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
2155 stream
->tracefile_count_current
);
2157 /* This will close the stream's index fd if one. */
2158 relay_index_free_safe(index
);
2161 stream
->index_fd
= ret
;
2163 index
->fd
= stream
->index_fd
;
2164 index
->index_data
.offset
= data_offset
;
2166 if (index_created
) {
2168 * Try to add the relay index object to the hash table. If an object
2169 * already exist, destroy back the index created and set the data.
2171 relay_index_add(index
, &wr_index
);
2173 /* Copy back data from the created index. */
2174 wr_index
->fd
= index
->fd
;
2175 wr_index
->to_close_fd
= index
->to_close_fd
;
2176 wr_index
->index_data
.offset
= data_offset
;
2180 /* The index already exists so write it on disk. */
2184 /* Do we have a writable ready index to write on disk. */
2186 ret
= relay_index_write(wr_index
->fd
, wr_index
);
2190 stream
->total_index_received
++;
2198 * relay_process_data: Process the data received on the data socket
2201 int relay_process_data(struct relay_connection
*conn
)
2203 int ret
= 0, rotate_index
= 0;
2205 struct relay_stream
*stream
;
2206 struct lttcomm_relayd_data_hdr data_hdr
;
2208 uint64_t net_seq_num
;
2210 struct relay_session
*session
;
2214 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &data_hdr
,
2215 sizeof(struct lttcomm_relayd_data_hdr
), 0);
2218 /* Orderly shutdown. Not necessary to print an error. */
2219 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2221 ERR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2227 stream_id
= be64toh(data_hdr
.stream_id
);
2230 stream
= stream_find_by_id(relay_streams_ht
, stream_id
);
2233 goto end_rcu_unlock
;
2236 session
= session_find_by_id(conn
->sessions_ht
, stream
->session_id
);
2239 data_size
= be32toh(data_hdr
.data_size
);
2240 if (data_buffer_size
< data_size
) {
2243 tmp_data_ptr
= realloc(data_buffer
, data_size
);
2244 if (!tmp_data_ptr
) {
2245 ERR("Allocating data buffer");
2248 goto end_rcu_unlock
;
2250 data_buffer
= tmp_data_ptr
;
2251 data_buffer_size
= data_size
;
2253 memset(data_buffer
, 0, data_size
);
2255 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2257 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2258 data_size
, stream_id
, net_seq_num
);
2259 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
2262 /* Orderly shutdown. Not necessary to print an error. */
2263 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2266 goto end_rcu_unlock
;
2269 /* Check if a rotation is needed. */
2270 if (stream
->tracefile_size
> 0 &&
2271 (stream
->tracefile_size_current
+ data_size
) >
2272 stream
->tracefile_size
) {
2273 struct relay_viewer_stream
*vstream
;
2276 new_id
= (stream
->tracefile_count_current
+ 1) %
2277 stream
->tracefile_count
;
2279 * When we wrap-around back to 0, we start overwriting old
2282 if (!stream
->tracefile_overwrite
&& new_id
== 0) {
2283 stream
->tracefile_overwrite
= 1;
2285 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
2286 if (stream
->tracefile_overwrite
) {
2287 stream
->oldest_tracefile_id
=
2288 (stream
->oldest_tracefile_id
+ 1) %
2289 stream
->tracefile_count
;
2291 vstream
= viewer_stream_find_by_id(stream
->stream_handle
);
2294 * The viewer is reading a file about to be
2295 * overwritten. Close the FDs it is
2296 * currently using and let it handle the fault.
2298 if (vstream
->tracefile_count_current
== new_id
) {
2299 pthread_mutex_lock(&vstream
->overwrite_lock
);
2300 vstream
->abort_flag
= 1;
2301 pthread_mutex_unlock(&vstream
->overwrite_lock
);
2302 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2303 stream
->channel_name
, new_id
);
2304 } else if (vstream
->tracefile_count_current
==
2305 stream
->tracefile_count_current
) {
2307 * The reader and writer were in the
2308 * same trace file, inform the viewer
2309 * that no new index will ever be added
2312 vstream
->close_write_flag
= 1;
2315 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
2316 stream
->tracefile_size
, stream
->tracefile_count
,
2317 relayd_uid
, relayd_gid
, stream
->fd
,
2318 &(stream
->tracefile_count_current
), &stream
->fd
);
2319 stream
->total_index_received
= 0;
2320 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
2322 ERR("Rotating stream output file");
2323 goto end_rcu_unlock
;
2325 /* Reset current size because we just perform a stream rotation. */
2326 stream
->tracefile_size_current
= 0;
2331 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2332 * index are NOT supported.
2334 if (session
->minor
>= 4 && !session
->snapshot
) {
2335 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2337 goto end_rcu_unlock
;
2341 /* Write data to stream output fd. */
2342 size_ret
= lttng_write(stream
->fd
, data_buffer
, data_size
);
2343 if (size_ret
< data_size
) {
2344 ERR("Relay error writing data to file");
2346 goto end_rcu_unlock
;
2349 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
2350 ret
, stream
->stream_handle
);
2352 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
2354 goto end_rcu_unlock
;
2356 stream
->tracefile_size_current
+= data_size
+ be32toh(data_hdr
.padding_size
);
2358 stream
->prev_seq
= net_seq_num
;
2360 try_close_stream(session
, stream
);
2369 void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2375 (void) lttng_poll_del(events
, pollfd
);
2377 ret
= close(pollfd
);
2379 ERR("Closing pollfd %d", pollfd
);
2383 static void destroy_connection(struct lttng_ht
*relay_connections_ht
,
2384 struct relay_connection
*conn
)
2386 assert(relay_connections_ht
);
2389 connection_delete(relay_connections_ht
, conn
);
2391 /* For the control socket, we try to destroy the session. */
2392 if (conn
->type
== RELAY_CONTROL
&& conn
->session
) {
2393 destroy_session(conn
->session
, conn
->sessions_ht
);
2396 connection_destroy(conn
);
2400 * This thread does the actual work
2403 void *relay_thread_worker(void *data
)
2405 int ret
, err
= -1, last_seen_data_fd
= -1;
2407 struct relay_connection
*conn
;
2408 struct lttng_poll_event events
;
2409 struct lttng_ht
*relay_connections_ht
;
2410 struct lttng_ht_iter iter
;
2411 struct lttcomm_relayd_hdr recv_hdr
;
2412 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
2413 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
2415 DBG("[thread] Relay worker started");
2417 rcu_register_thread();
2419 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2421 if (testpoint(relayd_thread_worker
)) {
2422 goto error_testpoint
;
2425 health_code_update();
2427 /* table of connections indexed on socket */
2428 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2429 if (!relay_connections_ht
) {
2430 goto relay_connections_ht_error
;
2433 /* Tables of received indexes indexed by index handle and net_seq_num. */
2434 indexes_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64
);
2436 goto indexes_ht_error
;
2439 ret
= create_thread_poll_set(&events
, 2);
2441 goto error_poll_create
;
2444 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2451 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2453 health_code_update();
2455 /* Infinite blocking call, waiting for transmission */
2456 DBG3("Relayd worker thread polling...");
2457 health_poll_entry();
2458 ret
= lttng_poll_wait(&events
, -1);
2462 * Restart interrupted system call.
2464 if (errno
== EINTR
) {
2473 * Process control. The control connection is prioritised so we don't
2474 * starve it with high throughout put tracing data on the data
2477 for (i
= 0; i
< nb_fd
; i
++) {
2478 /* Fetch once the poll data */
2479 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2480 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2482 health_code_update();
2484 /* Thread quit pipe has been closed. Killing thread. */
2485 ret
= check_thread_quit_pipe(pollfd
, revents
);
2491 /* Inspect the relay conn pipe for new connection */
2492 if (pollfd
== relay_conn_pipe
[0]) {
2493 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2494 ERR("Relay connection pipe error");
2496 } else if (revents
& LPOLLIN
) {
2497 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
2501 conn
->sessions_ht
= sessions_ht
;
2502 connection_init(conn
);
2503 lttng_poll_add(&events
, conn
->sock
->fd
,
2504 LPOLLIN
| LPOLLRDHUP
);
2506 lttng_ht_add_unique_ulong(relay_connections_ht
,
2509 DBG("Connection socket %d added", conn
->sock
->fd
);
2513 conn
= connection_find_by_sock(relay_connections_ht
, pollfd
);
2514 /* If not found, there is a synchronization issue. */
2517 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2518 cleanup_connection_pollfd(&events
, pollfd
);
2519 destroy_connection(relay_connections_ht
, conn
);
2520 if (last_seen_data_fd
== pollfd
) {
2521 last_seen_data_fd
= last_notdel_data_fd
;
2523 } else if (revents
& LPOLLIN
) {
2524 if (conn
->type
== RELAY_CONTROL
) {
2525 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &recv_hdr
,
2526 sizeof(recv_hdr
), 0);
2528 /* Connection closed */
2529 cleanup_connection_pollfd(&events
, pollfd
);
2530 destroy_connection(relay_connections_ht
, conn
);
2531 DBG("Control connection closed with %d", pollfd
);
2533 ret
= relay_process_control(&recv_hdr
, conn
);
2535 /* Clear the session on error. */
2536 cleanup_connection_pollfd(&events
, pollfd
);
2537 destroy_connection(relay_connections_ht
, conn
);
2538 DBG("Connection closed with %d", pollfd
);
2544 * Flag the last seen data fd not deleted. It will be
2545 * used as the last seen fd if any fd gets deleted in
2548 last_notdel_data_fd
= pollfd
;
2551 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
2558 * The last loop handled a control request, go back to poll to make
2559 * sure we prioritise the control socket.
2565 if (last_seen_data_fd
>= 0) {
2566 for (i
= 0; i
< nb_fd
; i
++) {
2567 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2569 health_code_update();
2571 if (last_seen_data_fd
== pollfd
) {
2578 /* Process data connection. */
2579 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2580 /* Fetch the poll data. */
2581 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2582 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2584 health_code_update();
2586 /* Skip the command pipe. It's handled in the first loop. */
2587 if (pollfd
== relay_conn_pipe
[0]) {
2593 conn
= connection_find_by_sock(relay_connections_ht
, pollfd
);
2595 /* Skip it. Might be removed before. */
2600 if (revents
& LPOLLIN
) {
2601 if (conn
->type
!= RELAY_DATA
) {
2605 ret
= relay_process_data(conn
);
2606 /* Connection closed */
2608 cleanup_connection_pollfd(&events
, pollfd
);
2609 destroy_connection(relay_connections_ht
, conn
);
2610 DBG("Data connection closed with %d", pollfd
);
2612 * Every goto restart call sets the last seen fd where
2613 * here we don't really care since we gracefully
2614 * continue the loop after the connection is deleted.
2617 /* Keep last seen port. */
2618 last_seen_data_fd
= pollfd
;
2626 last_seen_data_fd
= -1;
2629 /* Normal exit, no error */
2634 lttng_poll_clean(&events
);
2636 /* Cleanup reamaining connection object. */
2638 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, conn
,
2640 health_code_update();
2641 destroy_connection(relay_connections_ht
, conn
);
2645 lttng_ht_destroy(indexes_ht
);
2647 lttng_ht_destroy(relay_connections_ht
);
2648 relay_connections_ht_error
:
2649 /* Close relay conn pipes */
2650 utils_close_pipe(relay_conn_pipe
);
2652 DBG("Thread exited with error");
2654 DBG("Worker thread cleanup complete");
2659 ERR("Health error occurred in %s", __func__
);
2661 health_unregister(health_relayd
);
2662 rcu_unregister_thread();
2668 * Create the relay command pipe to wake thread_manage_apps.
2669 * Closed in cleanup().
2671 static int create_relay_conn_pipe(void)
2675 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
2683 int main(int argc
, char **argv
)
2687 struct relay_local_data
*relay_ctx
;
2689 /* Parse arguments */
2691 if ((ret
= set_options(argc
, argv
)) < 0) {
2695 if ((ret
= set_signal_handler()) < 0) {
2699 /* Try to create directory if -o, --output is specified. */
2700 if (opt_output_path
) {
2701 if (*opt_output_path
!= '/') {
2702 ERR("Please specify an absolute path for -o, --output PATH");
2706 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2708 ERR("Unable to create %s", opt_output_path
);
2714 if (opt_daemon
|| opt_background
) {
2717 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
2724 * We are in the child. Make sure all other file
2725 * descriptors are closed, in case we are called with
2726 * more opened file descriptors than the standard ones.
2728 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
2733 /* Create thread quit pipe */
2734 if ((ret
= init_thread_quit_pipe()) < 0) {
2738 /* We need those values for the file/dir creation. */
2739 relayd_uid
= getuid();
2740 relayd_gid
= getgid();
2742 /* Check if daemon is UID = 0 */
2743 if (relayd_uid
== 0) {
2744 if (control_uri
->port
< 1024 || data_uri
->port
< 1024 || live_uri
->port
< 1024) {
2745 ERR("Need to be root to use ports < 1024");
2751 /* Setup the thread apps communication pipe. */
2752 if ((ret
= create_relay_conn_pipe()) < 0) {
2756 /* Init relay command queue. */
2757 cds_wfq_init(&relay_conn_queue
.queue
);
2759 /* Set up max poll set size */
2760 lttng_poll_set_max_size();
2762 /* Initialize communication library */
2764 lttcomm_inet_init();
2766 relay_ctx
= zmalloc(sizeof(struct relay_local_data
));
2768 PERROR("relay_ctx");
2772 /* tables of sessions indexed by session ID */
2773 relay_ctx
->sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2774 if (!relay_ctx
->sessions_ht
) {
2775 goto exit_relay_ctx_sessions
;
2778 /* tables of streams indexed by stream ID */
2779 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2780 if (!relay_streams_ht
) {
2781 goto exit_relay_ctx_streams
;
2784 /* tables of streams indexed by stream ID */
2785 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2786 if (!viewer_streams_ht
) {
2787 goto exit_relay_ctx_viewer_streams
;
2790 /* Initialize thread health monitoring */
2791 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
2792 if (!health_relayd
) {
2793 PERROR("health_app_create error");
2794 goto exit_health_app_create
;
2797 ret
= utils_create_pipe(health_quit_pipe
);
2799 goto error_health_pipe
;
2802 /* Create thread to manage the client socket */
2803 ret
= pthread_create(&health_thread
, NULL
,
2804 thread_manage_health
, (void *) NULL
);
2806 PERROR("pthread_create health");
2810 /* Setup the dispatcher thread */
2811 ret
= pthread_create(&dispatcher_thread
, NULL
,
2812 relay_thread_dispatcher
, (void *) NULL
);
2814 PERROR("pthread_create dispatcher");
2815 goto exit_dispatcher
;
2818 /* Setup the worker thread */
2819 ret
= pthread_create(&worker_thread
, NULL
,
2820 relay_thread_worker
, (void *) relay_ctx
);
2822 PERROR("pthread_create worker");
2826 /* Setup the listener thread */
2827 ret
= pthread_create(&listener_thread
, NULL
,
2828 relay_thread_listener
, (void *) NULL
);
2830 PERROR("pthread_create listener");
2834 ret
= live_start_threads(live_uri
, relay_ctx
);
2836 ERR("Starting live viewer threads");
2841 ret
= pthread_join(listener_thread
, &status
);
2843 PERROR("pthread_join");
2844 goto error
; /* join error, exit without cleanup */
2848 ret
= pthread_join(worker_thread
, &status
);
2850 PERROR("pthread_join");
2851 goto error
; /* join error, exit without cleanup */
2855 ret
= pthread_join(dispatcher_thread
, &status
);
2857 PERROR("pthread_join");
2858 goto error
; /* join error, exit without cleanup */
2862 ret
= pthread_join(health_thread
, &status
);
2864 PERROR("pthread_join health thread");
2865 goto error
; /* join error, exit without cleanup */
2869 * Stop live threads only after joining other threads.
2871 live_stop_threads();
2874 utils_close_pipe(health_quit_pipe
);
2877 health_app_destroy(health_relayd
);
2879 exit_health_app_create
:
2880 lttng_ht_destroy(viewer_streams_ht
);
2882 exit_relay_ctx_viewer_streams
:
2883 lttng_ht_destroy(relay_streams_ht
);
2885 exit_relay_ctx_streams
:
2886 lttng_ht_destroy(relay_ctx
->sessions_ht
);
2888 exit_relay_ctx_sessions
: