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/compat/endian.h>
48 #include <common/defaults.h>
49 #include <common/daemonize.h>
50 #include <common/futex.h>
51 #include <common/sessiond-comm/sessiond-comm.h>
52 #include <common/sessiond-comm/inet.h>
53 #include <common/sessiond-comm/relayd.h>
54 #include <common/uri.h>
55 #include <common/utils.h>
56 #include <common/config/config.h>
59 #include "ctf-trace.h"
62 #include "lttng-relayd.h"
64 #include "health-relayd.h"
65 #include "testpoint.h"
66 #include "viewer-stream.h"
69 #include "connection.h"
71 /* command line options */
72 char *opt_output_path
;
73 static int opt_daemon
, opt_background
;
76 * We need to wait for listener and live listener threads, as well as
77 * health check thread, before being ready to signal readiness.
79 #define NR_LTTNG_RELAY_READY 3
80 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
81 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
82 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
84 static struct lttng_uri
*control_uri
;
85 static struct lttng_uri
*data_uri
;
86 static struct lttng_uri
*live_uri
;
90 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
91 static int tracing_group_name_override
;
93 const char * const config_section_name
= "relayd";
96 * Quit pipe for all threads. This permits a single cancellation point
97 * for all threads when receiving an event on the pipe.
99 int thread_quit_pipe
[2] = { -1, -1 };
102 * This pipe is used to inform the worker thread that a command is queued and
103 * ready to be processed.
105 static int relay_conn_pipe
[2] = { -1, -1 };
107 /* Shared between threads */
108 static int dispatch_thread_exit
;
110 static pthread_t listener_thread
;
111 static pthread_t dispatcher_thread
;
112 static pthread_t worker_thread
;
113 static pthread_t health_thread
;
115 static uint64_t last_relay_stream_id
;
118 * Relay command queue.
120 * The relay_thread_listener and relay_thread_dispatcher communicate with this
123 static struct relay_conn_queue relay_conn_queue
;
125 /* buffer allocated at startup, used to store the trace data */
126 static char *data_buffer
;
127 static unsigned int data_buffer_size
;
129 /* We need those values for the file/dir creation. */
130 static uid_t relayd_uid
;
131 static gid_t relayd_gid
;
133 /* Global relay stream hash table. */
134 struct lttng_ht
*relay_streams_ht
;
136 /* Global relay viewer stream hash table. */
137 struct lttng_ht
*viewer_streams_ht
;
139 /* Global hash table that stores relay index object. */
140 struct lttng_ht
*indexes_ht
;
142 /* Relayd health monitoring */
143 struct health_app
*health_relayd
;
145 static struct option long_options
[] = {
146 { "control-port", 1, 0, 'C', },
147 { "data-port", 1, 0, 'D', },
148 { "live-port", 1, 0, 'L', },
149 { "daemonize", 0, 0, 'd', },
150 { "background", 0, 0, 'b', },
151 { "group", 1, 0, 'g', },
152 { "help", 0, 0, 'h', },
153 { "output", 1, 0, 'o', },
154 { "verbose", 0, 0, 'v', },
155 { "config", 1, 0, 'f' },
159 static const char *config_ignore_options
[] = { "help", "config" };
162 * usage function on stderr
167 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
168 fprintf(stderr
, " -h, --help Display this usage.\n");
169 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
170 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
171 fprintf(stderr
, " -C, --control-port URL Control port listening.\n");
172 fprintf(stderr
, " -D, --data-port URL Data port listening.\n");
173 fprintf(stderr
, " -L, --live-port URL Live view port listening.\n");
174 fprintf(stderr
, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
175 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
176 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
177 fprintf(stderr
, " -f --config Load daemon configuration file\n");
181 * Take an option from the getopt output and set it in the right variable to be
184 * Return 0 on success else a negative value.
187 int set_option(int opt
, const char *arg
, const char *optname
)
193 fprintf(stderr
, "option %s", optname
);
195 fprintf(stderr
, " with arg %s\n", arg
);
199 ret
= uri_parse(arg
, &control_uri
);
201 ERR("Invalid control URI specified");
204 if (control_uri
->port
== 0) {
205 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
209 ret
= uri_parse(arg
, &data_uri
);
211 ERR("Invalid data URI specified");
214 if (data_uri
->port
== 0) {
215 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
219 ret
= uri_parse(arg
, &live_uri
);
221 ERR("Invalid live URI specified");
224 if (live_uri
->port
== 0) {
225 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
235 tracing_group_name
= strdup(arg
);
236 tracing_group_name_override
= 1;
242 ret
= asprintf(&opt_output_path
, "%s", arg
);
245 PERROR("asprintf opt_output_path");
250 /* Verbose level can increase using multiple -v */
252 lttng_opt_verbose
= config_parse_value(arg
);
254 lttng_opt_verbose
+= 1;
258 /* Unknown option or other error.
259 * Error is printed by getopt, just return */
272 * config_entry_handler_cb used to handle options read from a config file.
273 * See config_entry_handler_cb comment in common/config/config.h for the
274 * return value conventions.
277 int config_entry_handler(const struct config_entry
*entry
, void *unused
)
281 if (!entry
|| !entry
->name
|| !entry
->value
) {
286 /* Check if the option is to be ignored */
287 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
288 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
293 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
294 /* Ignore if entry name is not fully matched. */
295 if (strcmp(entry
->name
, long_options
[i
].name
)) {
300 * If the option takes no argument on the command line, we have to
301 * check if the value is "true". We support non-zero numeric values,
304 if (!long_options
[i
].has_arg
) {
305 ret
= config_parse_value(entry
->value
);
308 WARN("Invalid configuration value \"%s\" for option %s",
309 entry
->value
, entry
->name
);
311 /* False, skip boolean config option. */
316 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
320 WARN("Unrecognized option \"%s\" in daemon configuration file.",
328 int set_options(int argc
, char **argv
)
330 int c
, ret
= 0, option_index
= 0;
331 int orig_optopt
= optopt
, orig_optind
= optind
;
332 char *default_address
, *optstring
;
333 const char *config_path
= NULL
;
335 optstring
= utils_generate_optstring(long_options
,
336 sizeof(long_options
) / sizeof(struct option
));
342 /* Check for the --config option */
344 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
345 &option_index
)) != -1) {
349 } else if (c
!= 'f') {
353 config_path
= utils_expand_path(optarg
);
355 ERR("Failed to resolve path: %s", optarg
);
359 ret
= config_get_section_entries(config_path
, config_section_name
,
360 config_entry_handler
, NULL
);
363 ERR("Invalid configuration option at line %i", ret
);
369 /* Reset getopt's global state */
370 optopt
= orig_optopt
;
371 optind
= orig_optind
;
373 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
378 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
384 /* assign default values */
385 if (control_uri
== NULL
) {
386 ret
= asprintf(&default_address
,
387 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
388 DEFAULT_NETWORK_CONTROL_PORT
);
390 PERROR("asprintf default data address");
394 ret
= uri_parse(default_address
, &control_uri
);
395 free(default_address
);
397 ERR("Invalid control URI specified");
401 if (data_uri
== NULL
) {
402 ret
= asprintf(&default_address
,
403 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
404 DEFAULT_NETWORK_DATA_PORT
);
406 PERROR("asprintf default data address");
410 ret
= uri_parse(default_address
, &data_uri
);
411 free(default_address
);
413 ERR("Invalid data URI specified");
417 if (live_uri
== NULL
) {
418 ret
= asprintf(&default_address
,
419 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
420 DEFAULT_NETWORK_VIEWER_PORT
);
422 PERROR("asprintf default viewer control address");
426 ret
= uri_parse(default_address
, &live_uri
);
427 free(default_address
);
429 ERR("Invalid viewer control URI specified");
447 /* free the dynamically allocated opt_output_path */
448 free(opt_output_path
);
450 /* Close thread quit pipes */
451 utils_close_pipe(thread_quit_pipe
);
453 uri_free(control_uri
);
455 /* Live URI is freed in the live thread. */
457 if (tracing_group_name_override
) {
458 free((void *) tracing_group_name
);
463 * Write to writable pipe used to notify a thread.
466 int notify_thread_pipe(int wpipe
)
470 ret
= lttng_write(wpipe
, "!", 1);
472 PERROR("write poll pipe");
478 static void notify_health_quit_pipe(int *pipe
)
482 ret
= lttng_write(pipe
[1], "4", 1);
484 PERROR("write relay health quit");
489 * Stop all threads by closing the thread quit pipe.
492 void stop_threads(void)
496 /* Stopping all threads */
497 DBG("Terminating all threads");
498 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
500 ERR("write error on thread quit pipe");
503 notify_health_quit_pipe(health_quit_pipe
);
505 /* Dispatch thread */
506 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
507 futex_nto1_wake(&relay_conn_queue
.futex
);
511 * Signal handler for the daemon
513 * Simply stop all worker threads, leaving main() return gracefully after
514 * joining all threads and calling cleanup().
517 void sighandler(int sig
)
521 DBG("SIGPIPE caught");
524 DBG("SIGINT caught");
528 DBG("SIGTERM caught");
532 CMM_STORE_SHARED(recv_child_signal
, 1);
540 * Setup signal handler for :
541 * SIGINT, SIGTERM, SIGPIPE
544 int set_signal_handler(void)
550 if ((ret
= sigemptyset(&sigset
)) < 0) {
551 PERROR("sigemptyset");
555 sa
.sa_handler
= sighandler
;
558 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
563 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
568 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
573 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
578 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
583 void lttng_relay_notify_ready(void)
585 /* Notify the parent of the fork() process that we are ready. */
586 if (opt_daemon
|| opt_background
) {
587 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
588 kill(child_ppid
, SIGUSR1
);
594 * Init thread quit pipe.
596 * Return -1 on error or 0 if all pipes are created.
599 int init_thread_quit_pipe(void)
603 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
609 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
612 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
616 if (events
== NULL
|| size
== 0) {
621 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
627 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
639 * Check if the thread quit pipe was triggered.
641 * Return 1 if it was triggered else 0;
644 int check_thread_quit_pipe(int fd
, uint32_t events
)
646 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
654 * Create and init socket from uri.
657 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
660 struct lttcomm_sock
*sock
= NULL
;
662 sock
= lttcomm_alloc_sock_from_uri(uri
);
664 ERR("Allocating socket");
668 ret
= lttcomm_create_sock(sock
);
672 DBG("Listening on sock %d", sock
->fd
);
674 ret
= sock
->ops
->bind(sock
);
679 ret
= sock
->ops
->listen(sock
, -1);
689 lttcomm_destroy_sock(sock
);
695 * Return nonzero if stream needs to be closed.
698 int close_stream_check(struct relay_stream
*stream
)
700 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
702 * We are about to close the stream so set the data pending flag to 1
703 * which will make the end data pending command skip the stream which
704 * is now closed and ready. Note that after proceeding to a file close,
705 * the written file is ready for reading.
707 stream
->data_pending_check_done
= 1;
713 static void try_close_stream(struct relay_session
*session
,
714 struct relay_stream
*stream
)
717 struct ctf_trace
*ctf_trace
;
722 if (!close_stream_check(stream
)) {
723 /* Can't close it, not ready for that. */
727 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
,
731 pthread_mutex_lock(&session
->viewer_ready_lock
);
732 ctf_trace
->invalid_flag
= 1;
733 pthread_mutex_unlock(&session
->viewer_ready_lock
);
735 ret
= stream_close(session
, stream
);
736 if (ret
|| session
->snapshot
) {
737 /* Already close thus the ctf trace is being or has been destroyed. */
741 ctf_trace_try_destroy(session
, ctf_trace
);
748 * This thread manages the listening for new connections on the network
751 void *relay_thread_listener(void *data
)
753 int i
, ret
, pollfd
, err
= -1;
754 uint32_t revents
, nb_fd
;
755 struct lttng_poll_event events
;
756 struct lttcomm_sock
*control_sock
, *data_sock
;
758 DBG("[thread] Relay listener started");
760 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
762 health_code_update();
764 control_sock
= relay_init_sock(control_uri
);
766 goto error_sock_control
;
769 data_sock
= relay_init_sock(data_uri
);
771 goto error_sock_relay
;
775 * Pass 3 as size here for the thread quit pipe, control and data socket.
777 ret
= create_thread_poll_set(&events
, 3);
779 goto error_create_poll
;
782 /* Add the control socket */
783 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
788 /* Add the data socket */
789 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
794 lttng_relay_notify_ready();
796 if (testpoint(relayd_thread_listener
)) {
797 goto error_testpoint
;
801 health_code_update();
803 DBG("Listener accepting connections");
807 ret
= lttng_poll_wait(&events
, -1);
811 * Restart interrupted system call.
813 if (errno
== EINTR
) {
821 DBG("Relay new connection received");
822 for (i
= 0; i
< nb_fd
; i
++) {
823 health_code_update();
825 /* Fetch once the poll data */
826 revents
= LTTNG_POLL_GETEV(&events
, i
);
827 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
829 /* Thread quit pipe has been closed. Killing thread. */
830 ret
= check_thread_quit_pipe(pollfd
, revents
);
836 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
837 ERR("socket poll error");
839 } else if (revents
& LPOLLIN
) {
841 * Get allocated in this thread, enqueued to a global queue,
842 * dequeued and freed in the worker thread.
845 struct relay_connection
*new_conn
;
846 struct lttcomm_sock
*newsock
;
848 new_conn
= connection_create();
853 if (pollfd
== data_sock
->fd
) {
854 new_conn
->type
= RELAY_DATA
;
855 newsock
= data_sock
->ops
->accept(data_sock
);
856 DBG("Relay data connection accepted, socket %d",
859 assert(pollfd
== control_sock
->fd
);
860 new_conn
->type
= RELAY_CONTROL
;
861 newsock
= control_sock
->ops
->accept(control_sock
);
862 DBG("Relay control connection accepted, socket %d",
866 PERROR("accepting sock");
867 connection_free(new_conn
);
871 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
874 PERROR("setsockopt inet");
875 lttcomm_destroy_sock(newsock
);
876 connection_free(new_conn
);
879 new_conn
->sock
= newsock
;
881 /* Enqueue request for the dispatcher thread. */
882 cds_wfq_enqueue(&relay_conn_queue
.queue
, &new_conn
->qnode
);
885 * Wake the dispatch queue futex. Implicit memory barrier with
886 * the exchange in cds_wfq_enqueue.
888 futex_nto1_wake(&relay_conn_queue
.futex
);
897 lttng_poll_clean(&events
);
899 if (data_sock
->fd
>= 0) {
900 ret
= data_sock
->ops
->close(data_sock
);
905 lttcomm_destroy_sock(data_sock
);
907 if (control_sock
->fd
>= 0) {
908 ret
= control_sock
->ops
->close(control_sock
);
913 lttcomm_destroy_sock(control_sock
);
917 ERR("Health error occurred in %s", __func__
);
919 health_unregister(health_relayd
);
920 DBG("Relay listener thread cleanup complete");
926 * This thread manages the dispatching of the requests to worker threads
929 void *relay_thread_dispatcher(void *data
)
933 struct cds_wfq_node
*node
;
934 struct relay_connection
*new_conn
= NULL
;
936 DBG("[thread] Relay dispatcher started");
938 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
940 if (testpoint(relayd_thread_dispatcher
)) {
941 goto error_testpoint
;
944 health_code_update();
946 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
947 health_code_update();
949 /* Atomically prepare the queue futex */
950 futex_nto1_prepare(&relay_conn_queue
.futex
);
953 health_code_update();
955 /* Dequeue commands */
956 node
= cds_wfq_dequeue_blocking(&relay_conn_queue
.queue
);
958 DBG("Woken up but nothing in the relay command queue");
959 /* Continue thread execution */
962 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
964 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
967 * Inform worker thread of the new request. This call is blocking
968 * so we can be assured that the data will be read at some point in
969 * time or wait to the end of the world :)
971 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
973 PERROR("write connection pipe");
974 connection_destroy(new_conn
);
977 } while (node
!= NULL
);
979 /* Futex wait on queue. Blocking call on futex() */
981 futex_nto1_wait(&relay_conn_queue
.futex
);
985 /* Normal exit, no error */
992 ERR("Health error occurred in %s", __func__
);
994 health_unregister(health_relayd
);
995 DBG("Dispatch thread dying");
1000 static void try_close_streams(struct relay_session
*session
)
1002 struct ctf_trace
*ctf_trace
;
1003 struct lttng_ht_iter iter
;
1007 pthread_mutex_lock(&session
->viewer_ready_lock
);
1009 cds_lfht_for_each_entry(session
->ctf_traces_ht
->ht
, &iter
.iter
, ctf_trace
,
1011 struct relay_stream
*stream
;
1013 /* Close streams. */
1014 cds_list_for_each_entry(stream
, &ctf_trace
->stream_list
, trace_list
) {
1015 stream_close(session
, stream
);
1018 ctf_trace
->invalid_flag
= 1;
1019 ctf_trace_try_destroy(session
, ctf_trace
);
1022 pthread_mutex_unlock(&session
->viewer_ready_lock
);
1026 * Try to destroy a session within a connection.
1028 static void destroy_session(struct relay_session
*session
,
1029 struct lttng_ht
*sessions_ht
)
1032 assert(sessions_ht
);
1034 /* Indicate that this session can be destroyed from now on. */
1035 session
->close_flag
= 1;
1037 try_close_streams(session
);
1040 * This will try to delete and destroy the session if no viewer is attached
1041 * to it meaning the refcount is down to zero.
1043 session_try_destroy(sessions_ht
, session
);
1047 * Copy index data from the control port to a given index object.
1049 static void copy_index_control_data(struct relay_index
*index
,
1050 struct lttcomm_relayd_index
*data
)
1056 * The index on disk is encoded in big endian, so we don't need to convert
1057 * the data received on the network. The data_offset value is NEVER
1058 * modified here and is updated by the data thread.
1060 index
->index_data
.packet_size
= data
->packet_size
;
1061 index
->index_data
.content_size
= data
->content_size
;
1062 index
->index_data
.timestamp_begin
= data
->timestamp_begin
;
1063 index
->index_data
.timestamp_end
= data
->timestamp_end
;
1064 index
->index_data
.events_discarded
= data
->events_discarded
;
1065 index
->index_data
.stream_id
= data
->stream_id
;
1069 * Handle the RELAYD_CREATE_SESSION command.
1071 * On success, send back the session id or else return a negative value.
1074 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
1075 struct relay_connection
*conn
)
1077 int ret
= 0, send_ret
;
1078 struct relay_session
*session
;
1079 struct lttcomm_relayd_status_session reply
;
1084 memset(&reply
, 0, sizeof(reply
));
1086 session
= session_create();
1091 session
->minor
= conn
->minor
;
1092 session
->major
= conn
->major
;
1093 conn
->session_id
= session
->id
;
1094 conn
->session
= session
;
1096 reply
.session_id
= htobe64(session
->id
);
1098 switch (conn
->minor
) {
1103 case 4: /* LTTng sessiond 2.4 */
1105 ret
= cmd_create_session_2_4(conn
, session
);
1108 lttng_ht_add_unique_u64(conn
->sessions_ht
, &session
->session_n
);
1109 DBG("Created session %" PRIu64
, session
->id
);
1113 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1115 reply
.ret_code
= htobe32(LTTNG_OK
);
1118 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1120 ERR("Relayd sending session id");
1128 * When we have received all the streams and the metadata for a channel,
1129 * we make them visible to the viewer threads.
1132 void set_viewer_ready_flag(struct relay_connection
*conn
)
1134 struct relay_stream
*stream
, *tmp_stream
;
1136 pthread_mutex_lock(&conn
->session
->viewer_ready_lock
);
1137 cds_list_for_each_entry_safe(stream
, tmp_stream
, &conn
->recv_head
,
1139 stream
->viewer_ready
= 1;
1140 cds_list_del(&stream
->recv_list
);
1142 pthread_mutex_unlock(&conn
->session
->viewer_ready_lock
);
1147 * Add a recv handle node to the connection recv list with the given stream
1148 * handle. A new node is allocated thus must be freed when the node is deleted
1151 static void queue_stream(struct relay_stream
*stream
,
1152 struct relay_connection
*conn
)
1157 cds_list_add(&stream
->recv_list
, &conn
->recv_head
);
1161 * relay_add_stream: allocate a new stream for a session
1164 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1165 struct relay_connection
*conn
)
1168 struct relay_session
*session
= conn
->session
;
1169 struct relay_stream
*stream
= NULL
;
1170 struct lttcomm_relayd_status_stream reply
;
1171 struct ctf_trace
*trace
;
1173 if (!session
|| conn
->version_check_done
== 0) {
1174 ERR("Trying to add a stream before version check");
1176 goto end_no_session
;
1179 stream
= zmalloc(sizeof(struct relay_stream
));
1180 if (stream
== NULL
) {
1181 PERROR("relay stream zmalloc");
1183 goto end_no_session
;
1186 switch (conn
->minor
) {
1187 case 1: /* LTTng sessiond 2.1 */
1188 ret
= cmd_recv_stream_2_1(conn
, stream
);
1190 case 2: /* LTTng sessiond 2.2 */
1192 ret
= cmd_recv_stream_2_2(conn
, stream
);
1196 goto err_free_stream
;
1200 stream
->stream_handle
= ++last_relay_stream_id
;
1201 stream
->prev_seq
= -1ULL;
1202 stream
->session_id
= session
->id
;
1203 stream
->index_fd
= -1;
1204 stream
->read_index_fd
= -1;
1205 lttng_ht_node_init_u64(&stream
->node
, stream
->stream_handle
);
1206 pthread_mutex_init(&stream
->lock
, NULL
);
1208 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
);
1210 ERR("relay creating output directory");
1215 * No need to use run_as API here because whatever we receives, the relayd
1216 * uses its own credentials for the stream files.
1218 ret
= utils_create_stream_file(stream
->path_name
, stream
->channel_name
,
1219 stream
->tracefile_size
, 0, relayd_uid
, relayd_gid
, NULL
);
1221 ERR("Create output file");
1225 if (stream
->tracefile_size
) {
1226 DBG("Tracefile %s/%s_0 created", stream
->path_name
, stream
->channel_name
);
1228 DBG("Tracefile %s/%s created", stream
->path_name
, stream
->channel_name
);
1231 trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
, stream
->path_name
);
1233 trace
= ctf_trace_create(stream
->path_name
);
1238 ctf_trace_add(session
->ctf_traces_ht
, trace
);
1240 ctf_trace_get_ref(trace
);
1242 if (!strncmp(stream
->channel_name
, DEFAULT_METADATA_NAME
, NAME_MAX
)) {
1243 stream
->metadata_flag
= 1;
1244 /* Assign quick reference to the metadata stream in the trace. */
1245 trace
->metadata_stream
= stream
;
1249 * Add the stream in the recv list of the connection. Once the end stream
1250 * message is received, this list is emptied and streams are set with the
1251 * viewer ready flag.
1253 queue_stream(stream
, conn
);
1256 * Both in the ctf_trace object and the global stream ht since the data
1257 * side of the relayd does not have the concept of session.
1259 lttng_ht_add_unique_u64(relay_streams_ht
, &stream
->node
);
1260 cds_list_add_tail(&stream
->trace_list
, &trace
->stream_list
);
1262 session
->stream_count
++;
1264 DBG("Relay new stream added %s with ID %" PRIu64
, stream
->channel_name
,
1265 stream
->stream_handle
);
1268 memset(&reply
, 0, sizeof(reply
));
1269 reply
.handle
= htobe64(stream
->stream_handle
);
1270 /* send the session id to the client or a negative return code on error */
1272 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1273 /* stream was not properly added to the ht, so free it */
1276 reply
.ret_code
= htobe32(LTTNG_OK
);
1279 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1280 sizeof(struct lttcomm_relayd_status_stream
), 0);
1282 ERR("Relay sending stream id");
1291 free(stream
->path_name
);
1292 free(stream
->channel_name
);
1298 * relay_close_stream: close a specific stream
1301 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1302 struct relay_connection
*conn
)
1305 struct relay_session
*session
= conn
->session
;
1306 struct lttcomm_relayd_close_stream stream_info
;
1307 struct lttcomm_relayd_generic_reply reply
;
1308 struct relay_stream
*stream
;
1310 DBG("Close stream received");
1312 if (!session
|| conn
->version_check_done
== 0) {
1313 ERR("Trying to close a stream before version check");
1315 goto end_no_session
;
1318 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
1319 sizeof(struct lttcomm_relayd_close_stream
), 0);
1320 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1322 /* Orderly shutdown. Not necessary to print an error. */
1323 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1325 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1328 goto end_no_session
;
1332 stream
= stream_find_by_id(relay_streams_ht
,
1333 be64toh(stream_info
.stream_id
));
1339 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1340 stream
->close_flag
= 1;
1341 session
->stream_count
--;
1342 assert(session
->stream_count
>= 0);
1344 /* Check if we can close it or else the data will do it. */
1345 try_close_stream(session
, stream
);
1350 memset(&reply
, 0, sizeof(reply
));
1352 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1354 reply
.ret_code
= htobe32(LTTNG_OK
);
1356 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1357 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1359 ERR("Relay sending stream id");
1368 * relay_unknown_command: send -1 if received unknown command
1371 void relay_unknown_command(struct relay_connection
*conn
)
1373 struct lttcomm_relayd_generic_reply reply
;
1376 memset(&reply
, 0, sizeof(reply
));
1377 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1378 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1379 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1381 ERR("Relay sending unknown command");
1386 * relay_start: send an acknowledgment to the client to tell if we are
1387 * ready to receive data. We are ready if a session is established.
1390 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1391 struct relay_connection
*conn
)
1393 int ret
= htobe32(LTTNG_OK
);
1394 struct lttcomm_relayd_generic_reply reply
;
1395 struct relay_session
*session
= conn
->session
;
1398 DBG("Trying to start the streaming without a session established");
1399 ret
= htobe32(LTTNG_ERR_UNK
);
1402 memset(&reply
, 0, sizeof(reply
));
1403 reply
.ret_code
= ret
;
1404 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1405 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1407 ERR("Relay sending start ack");
1414 * Append padding to the file pointed by the file descriptor fd.
1416 static int write_padding_to_file(int fd
, uint32_t size
)
1425 zeros
= zmalloc(size
);
1426 if (zeros
== NULL
) {
1427 PERROR("zmalloc zeros for padding");
1432 ret
= lttng_write(fd
, zeros
, size
);
1434 PERROR("write padding to file");
1444 * relay_recv_metadata: receive the metada for the session.
1447 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1448 struct relay_connection
*conn
)
1450 int ret
= htobe32(LTTNG_OK
);
1452 struct relay_session
*session
= conn
->session
;
1453 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1454 struct relay_stream
*metadata_stream
;
1455 uint64_t data_size
, payload_size
;
1456 struct ctf_trace
*ctf_trace
;
1459 ERR("Metadata sent before version check");
1464 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1465 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1466 ERR("Incorrect data size");
1470 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1472 if (data_buffer_size
< data_size
) {
1473 /* In case the realloc fails, we can free the memory */
1476 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1477 if (!tmp_data_ptr
) {
1478 ERR("Allocating data buffer");
1483 data_buffer
= tmp_data_ptr
;
1484 data_buffer_size
= data_size
;
1486 memset(data_buffer
, 0, data_size
);
1487 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1488 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
1489 if (ret
< 0 || ret
!= data_size
) {
1491 /* Orderly shutdown. Not necessary to print an error. */
1492 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1494 ERR("Relay didn't receive the whole metadata");
1499 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1502 metadata_stream
= stream_find_by_id(relay_streams_ht
,
1503 be64toh(metadata_struct
->stream_id
));
1504 if (!metadata_stream
) {
1509 size_ret
= lttng_write(metadata_stream
->fd
, metadata_struct
->payload
,
1511 if (size_ret
< payload_size
) {
1512 ERR("Relay error writing metadata on file");
1517 ret
= write_padding_to_file(metadata_stream
->fd
,
1518 be32toh(metadata_struct
->padding_size
));
1523 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
,
1524 metadata_stream
->path_name
);
1526 ctf_trace
->metadata_received
+=
1527 payload_size
+ be32toh(metadata_struct
->padding_size
);
1529 DBG2("Relay metadata written");
1538 * relay_send_version: send relayd version number
1541 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1542 struct relay_connection
*conn
)
1545 struct lttcomm_relayd_version reply
, msg
;
1549 conn
->version_check_done
= 1;
1551 /* Get version from the other side. */
1552 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1553 if (ret
< 0 || ret
!= sizeof(msg
)) {
1555 /* Orderly shutdown. Not necessary to print an error. */
1556 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1558 ERR("Relay failed to receive the version values.");
1564 memset(&reply
, 0, sizeof(reply
));
1565 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1566 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1568 /* Major versions must be the same */
1569 if (reply
.major
!= be32toh(msg
.major
)) {
1570 DBG("Incompatible major versions (%u vs %u), deleting session",
1571 reply
.major
, be32toh(msg
.major
));
1572 destroy_session(conn
->session
, conn
->sessions_ht
);
1577 conn
->major
= reply
.major
;
1578 /* We adapt to the lowest compatible version */
1579 if (reply
.minor
<= be32toh(msg
.minor
)) {
1580 conn
->minor
= reply
.minor
;
1582 conn
->minor
= be32toh(msg
.minor
);
1585 reply
.major
= htobe32(reply
.major
);
1586 reply
.minor
= htobe32(reply
.minor
);
1587 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1588 sizeof(struct lttcomm_relayd_version
), 0);
1590 ERR("Relay sending version");
1593 DBG("Version check done using protocol %u.%u", conn
->major
,
1601 * Check for data pending for a given stream id from the session daemon.
1604 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1605 struct relay_connection
*conn
)
1607 struct relay_session
*session
= conn
->session
;
1608 struct lttcomm_relayd_data_pending msg
;
1609 struct lttcomm_relayd_generic_reply reply
;
1610 struct relay_stream
*stream
;
1612 uint64_t last_net_seq_num
, stream_id
;
1614 DBG("Data pending command received");
1616 if (!session
|| conn
->version_check_done
== 0) {
1617 ERR("Trying to check for data before version check");
1619 goto end_no_session
;
1622 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1623 if (ret
< sizeof(msg
)) {
1625 /* Orderly shutdown. Not necessary to print an error. */
1626 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1628 ERR("Relay didn't receive valid data_pending struct size : %d",
1632 goto end_no_session
;
1635 stream_id
= be64toh(msg
.stream_id
);
1636 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1639 stream
= stream_find_by_id(relay_streams_ht
, stream_id
);
1640 if (stream
== NULL
) {
1645 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1646 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1649 /* Avoid wrapping issue */
1650 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1651 /* Data has in fact been written and is NOT pending */
1654 /* Data still being streamed thus pending */
1658 /* Pending check is now done. */
1659 stream
->data_pending_check_done
= 1;
1664 memset(&reply
, 0, sizeof(reply
));
1665 reply
.ret_code
= htobe32(ret
);
1666 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1668 ERR("Relay data pending ret code failed");
1676 * Wait for the control socket to reach a quiescent state.
1678 * Note that for now, when receiving this command from the session daemon, this
1679 * means that every subsequent commands or data received on the control socket
1680 * has been handled. So, this is why we simply return OK here.
1683 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1684 struct relay_connection
*conn
)
1688 struct relay_stream
*stream
;
1689 struct lttng_ht_iter iter
;
1690 struct lttcomm_relayd_quiescent_control msg
;
1691 struct lttcomm_relayd_generic_reply reply
;
1693 DBG("Checking quiescent state on control socket");
1695 if (!conn
->session
|| conn
->version_check_done
== 0) {
1696 ERR("Trying to check for data before version check");
1698 goto end_no_session
;
1701 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1702 if (ret
< sizeof(msg
)) {
1704 /* Orderly shutdown. Not necessary to print an error. */
1705 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1707 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1711 goto end_no_session
;
1714 stream_id
= be64toh(msg
.stream_id
);
1717 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1719 if (stream
->stream_handle
== stream_id
) {
1720 stream
->data_pending_check_done
= 1;
1721 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1728 memset(&reply
, 0, sizeof(reply
));
1729 reply
.ret_code
= htobe32(LTTNG_OK
);
1730 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1732 ERR("Relay data quiescent control ret code failed");
1740 * Initialize a data pending command. This means that a client is about to ask
1741 * for data pending for each stream he/she holds. Simply iterate over all
1742 * streams of a session and set the data_pending_check_done flag.
1744 * This command returns to the client a LTTNG_OK code.
1747 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1748 struct relay_connection
*conn
)
1751 struct lttng_ht_iter iter
;
1752 struct lttcomm_relayd_begin_data_pending msg
;
1753 struct lttcomm_relayd_generic_reply reply
;
1754 struct relay_stream
*stream
;
1755 uint64_t session_id
;
1760 DBG("Init streams for data pending");
1762 if (!conn
->session
|| conn
->version_check_done
== 0) {
1763 ERR("Trying to check for data before version check");
1765 goto end_no_session
;
1768 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1769 if (ret
< sizeof(msg
)) {
1771 /* Orderly shutdown. Not necessary to print an error. */
1772 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1774 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1778 goto end_no_session
;
1781 session_id
= be64toh(msg
.session_id
);
1784 * Iterate over all streams to set the begin data pending flag. For now, the
1785 * streams are indexed by stream handle so we have to iterate over all
1786 * streams to find the one associated with the right session_id.
1789 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1791 if (stream
->session_id
== session_id
) {
1792 stream
->data_pending_check_done
= 0;
1793 DBG("Set begin data pending flag to stream %" PRIu64
,
1794 stream
->stream_handle
);
1799 memset(&reply
, 0, sizeof(reply
));
1800 /* All good, send back reply. */
1801 reply
.ret_code
= htobe32(LTTNG_OK
);
1803 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1805 ERR("Relay begin data pending send reply failed");
1813 * End data pending command. This will check, for a given session id, if each
1814 * stream associated with it has its data_pending_check_done flag set. If not,
1815 * this means that the client lost track of the stream but the data is still
1816 * being streamed on our side. In this case, we inform the client that data is
1819 * Return to the client if there is data in flight or not with a ret_code.
1822 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1823 struct relay_connection
*conn
)
1826 struct lttng_ht_iter iter
;
1827 struct lttcomm_relayd_end_data_pending msg
;
1828 struct lttcomm_relayd_generic_reply reply
;
1829 struct relay_stream
*stream
;
1830 uint64_t session_id
;
1831 uint32_t is_data_inflight
= 0;
1836 DBG("End data pending command");
1838 if (!conn
->session
|| conn
->version_check_done
== 0) {
1839 ERR("Trying to check for data before version check");
1841 goto end_no_session
;
1844 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1845 if (ret
< sizeof(msg
)) {
1847 /* Orderly shutdown. Not necessary to print an error. */
1848 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1850 ERR("Relay didn't receive valid end data_pending struct size: %d",
1854 goto end_no_session
;
1857 session_id
= be64toh(msg
.session_id
);
1859 /* Iterate over all streams to see if the begin data pending flag is set. */
1861 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1863 if (stream
->session_id
== session_id
&&
1864 !stream
->data_pending_check_done
&& !stream
->terminated_flag
) {
1865 is_data_inflight
= 1;
1866 DBG("Data is still in flight for stream %" PRIu64
,
1867 stream
->stream_handle
);
1873 memset(&reply
, 0, sizeof(reply
));
1874 /* All good, send back reply. */
1875 reply
.ret_code
= htobe32(is_data_inflight
);
1877 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1879 ERR("Relay end data pending send reply failed");
1887 * Receive an index for a specific stream.
1889 * Return 0 on success else a negative value.
1892 int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1893 struct relay_connection
*conn
)
1895 int ret
, send_ret
, index_created
= 0;
1896 struct relay_session
*session
= conn
->session
;
1897 struct lttcomm_relayd_index index_info
;
1898 struct relay_index
*index
, *wr_index
= NULL
;
1899 struct lttcomm_relayd_generic_reply reply
;
1900 struct relay_stream
*stream
;
1901 uint64_t net_seq_num
;
1905 DBG("Relay receiving index");
1907 if (!session
|| conn
->version_check_done
== 0) {
1908 ERR("Trying to close a stream before version check");
1910 goto end_no_session
;
1913 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &index_info
,
1914 sizeof(index_info
), 0);
1915 if (ret
< sizeof(index_info
)) {
1917 /* Orderly shutdown. Not necessary to print an error. */
1918 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1920 ERR("Relay didn't receive valid index struct size : %d", ret
);
1923 goto end_no_session
;
1926 net_seq_num
= be64toh(index_info
.net_seq_num
);
1929 stream
= stream_find_by_id(relay_streams_ht
,
1930 be64toh(index_info
.relay_stream_id
));
1933 goto end_rcu_unlock
;
1936 /* Live beacon handling */
1937 if (index_info
.packet_size
== 0) {
1938 DBG("Received live beacon for stream %" PRIu64
, stream
->stream_handle
);
1941 * Only flag a stream inactive when it has already received data.
1943 if (stream
->total_index_received
> 0) {
1944 stream
->beacon_ts_end
= be64toh(index_info
.timestamp_end
);
1947 goto end_rcu_unlock
;
1949 stream
->beacon_ts_end
= -1ULL;
1952 index
= relay_index_find(stream
->stream_handle
, net_seq_num
);
1954 /* A successful creation will add the object to the HT. */
1955 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
1957 goto end_rcu_unlock
;
1962 copy_index_control_data(index
, &index_info
);
1964 if (index_created
) {
1966 * Try to add the relay index object to the hash table. If an object
1967 * already exist, destroy back the index created, set the data in this
1968 * object and write it on disk.
1970 relay_index_add(index
, &wr_index
);
1972 copy_index_control_data(wr_index
, &index_info
);
1976 /* The index already exists so write it on disk. */
1980 /* Do we have a writable ready index to write on disk. */
1982 ret
= relay_index_write(wr_index
->fd
, wr_index
);
1984 goto end_rcu_unlock
;
1986 stream
->total_index_received
++;
1992 memset(&reply
, 0, sizeof(reply
));
1994 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1996 reply
.ret_code
= htobe32(LTTNG_OK
);
1998 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2000 ERR("Relay sending close index id reply");
2009 * Receive the streams_sent message.
2011 * Return 0 on success else a negative value.
2014 int relay_streams_sent(struct lttcomm_relayd_hdr
*recv_hdr
,
2015 struct relay_connection
*conn
)
2018 struct lttcomm_relayd_generic_reply reply
;
2022 DBG("Relay receiving streams_sent");
2024 if (!conn
->session
|| conn
->version_check_done
== 0) {
2025 ERR("Trying to close a stream before version check");
2027 goto end_no_session
;
2031 * Flag every pending stream in the connection recv list that they are
2032 * ready to be used by the viewer.
2034 set_viewer_ready_flag(conn
);
2037 * Inform the viewer that there are new streams in the session.
2039 if (conn
->session
->viewer_refcount
) {
2040 uatomic_set(&conn
->session
->new_streams
, 1);
2043 memset(&reply
, 0, sizeof(reply
));
2044 reply
.ret_code
= htobe32(LTTNG_OK
);
2045 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2047 ERR("Relay sending sent_stream reply");
2059 * Process the commands received on the control socket
2062 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
2063 struct relay_connection
*conn
)
2067 switch (be32toh(recv_hdr
->cmd
)) {
2068 case RELAYD_CREATE_SESSION
:
2069 ret
= relay_create_session(recv_hdr
, conn
);
2071 case RELAYD_ADD_STREAM
:
2072 ret
= relay_add_stream(recv_hdr
, conn
);
2074 case RELAYD_START_DATA
:
2075 ret
= relay_start(recv_hdr
, conn
);
2077 case RELAYD_SEND_METADATA
:
2078 ret
= relay_recv_metadata(recv_hdr
, conn
);
2080 case RELAYD_VERSION
:
2081 ret
= relay_send_version(recv_hdr
, conn
);
2083 case RELAYD_CLOSE_STREAM
:
2084 ret
= relay_close_stream(recv_hdr
, conn
);
2086 case RELAYD_DATA_PENDING
:
2087 ret
= relay_data_pending(recv_hdr
, conn
);
2089 case RELAYD_QUIESCENT_CONTROL
:
2090 ret
= relay_quiescent_control(recv_hdr
, conn
);
2092 case RELAYD_BEGIN_DATA_PENDING
:
2093 ret
= relay_begin_data_pending(recv_hdr
, conn
);
2095 case RELAYD_END_DATA_PENDING
:
2096 ret
= relay_end_data_pending(recv_hdr
, conn
);
2098 case RELAYD_SEND_INDEX
:
2099 ret
= relay_recv_index(recv_hdr
, conn
);
2101 case RELAYD_STREAMS_SENT
:
2102 ret
= relay_streams_sent(recv_hdr
, conn
);
2104 case RELAYD_UPDATE_SYNC_INFO
:
2106 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
2107 relay_unknown_command(conn
);
2117 * Handle index for a data stream.
2119 * RCU read side lock MUST be acquired.
2121 * Return 0 on success else a negative value.
2123 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2126 int ret
= 0, index_created
= 0;
2127 uint64_t stream_id
, data_offset
;
2128 struct relay_index
*index
, *wr_index
= NULL
;
2132 stream_id
= stream
->stream_handle
;
2133 /* Get data offset because we are about to update the index. */
2134 data_offset
= htobe64(stream
->tracefile_size_current
);
2137 * Lookup for an existing index for that stream id/sequence number. If on
2138 * exists, the control thread already received the data for it thus we need
2139 * to write it on disk.
2141 index
= relay_index_find(stream_id
, net_seq_num
);
2143 /* A successful creation will add the object to the HT. */
2144 index
= relay_index_create(stream_id
, net_seq_num
);
2152 if (rotate_index
|| stream
->index_fd
< 0) {
2153 index
->to_close_fd
= stream
->index_fd
;
2154 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
2155 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
2156 stream
->tracefile_count_current
);
2158 /* This will close the stream's index fd if one. */
2159 relay_index_free_safe(index
);
2162 stream
->index_fd
= ret
;
2164 index
->fd
= stream
->index_fd
;
2165 index
->index_data
.offset
= data_offset
;
2167 if (index_created
) {
2169 * Try to add the relay index object to the hash table. If an object
2170 * already exist, destroy back the index created and set the data.
2172 relay_index_add(index
, &wr_index
);
2174 /* Copy back data from the created index. */
2175 wr_index
->fd
= index
->fd
;
2176 wr_index
->to_close_fd
= index
->to_close_fd
;
2177 wr_index
->index_data
.offset
= data_offset
;
2181 /* The index already exists so write it on disk. */
2185 /* Do we have a writable ready index to write on disk. */
2187 ret
= relay_index_write(wr_index
->fd
, wr_index
);
2191 stream
->total_index_received
++;
2199 * relay_process_data: Process the data received on the data socket
2202 int relay_process_data(struct relay_connection
*conn
)
2204 int ret
= 0, rotate_index
= 0;
2206 struct relay_stream
*stream
;
2207 struct lttcomm_relayd_data_hdr data_hdr
;
2209 uint64_t net_seq_num
;
2211 struct relay_session
*session
;
2215 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &data_hdr
,
2216 sizeof(struct lttcomm_relayd_data_hdr
), 0);
2219 /* Orderly shutdown. Not necessary to print an error. */
2220 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2222 ERR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2228 stream_id
= be64toh(data_hdr
.stream_id
);
2231 stream
= stream_find_by_id(relay_streams_ht
, stream_id
);
2234 goto end_rcu_unlock
;
2237 session
= session_find_by_id(conn
->sessions_ht
, stream
->session_id
);
2240 data_size
= be32toh(data_hdr
.data_size
);
2241 if (data_buffer_size
< data_size
) {
2244 tmp_data_ptr
= realloc(data_buffer
, data_size
);
2245 if (!tmp_data_ptr
) {
2246 ERR("Allocating data buffer");
2249 goto end_rcu_unlock
;
2251 data_buffer
= tmp_data_ptr
;
2252 data_buffer_size
= data_size
;
2254 memset(data_buffer
, 0, data_size
);
2256 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2258 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2259 data_size
, stream_id
, net_seq_num
);
2260 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
2263 /* Orderly shutdown. Not necessary to print an error. */
2264 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2267 goto end_rcu_unlock
;
2270 /* Check if a rotation is needed. */
2271 if (stream
->tracefile_size
> 0 &&
2272 (stream
->tracefile_size_current
+ data_size
) >
2273 stream
->tracefile_size
) {
2274 struct relay_viewer_stream
*vstream
;
2277 new_id
= (stream
->tracefile_count_current
+ 1) %
2278 stream
->tracefile_count
;
2280 * When we wrap-around back to 0, we start overwriting old
2283 if (!stream
->tracefile_overwrite
&& new_id
== 0) {
2284 stream
->tracefile_overwrite
= 1;
2286 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
2287 if (stream
->tracefile_overwrite
) {
2288 stream
->oldest_tracefile_id
=
2289 (stream
->oldest_tracefile_id
+ 1) %
2290 stream
->tracefile_count
;
2292 vstream
= viewer_stream_find_by_id(stream
->stream_handle
);
2295 * The viewer is reading a file about to be
2296 * overwritten. Close the FDs it is
2297 * currently using and let it handle the fault.
2299 if (vstream
->tracefile_count_current
== new_id
) {
2300 pthread_mutex_lock(&vstream
->overwrite_lock
);
2301 vstream
->abort_flag
= 1;
2302 pthread_mutex_unlock(&vstream
->overwrite_lock
);
2303 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2304 stream
->channel_name
, new_id
);
2305 } else if (vstream
->tracefile_count_current
==
2306 stream
->tracefile_count_current
) {
2308 * The reader and writer were in the
2309 * same trace file, inform the viewer
2310 * that no new index will ever be added
2313 vstream
->close_write_flag
= 1;
2316 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
2317 stream
->tracefile_size
, stream
->tracefile_count
,
2318 relayd_uid
, relayd_gid
, stream
->fd
,
2319 &(stream
->tracefile_count_current
), &stream
->fd
);
2320 stream
->total_index_received
= 0;
2321 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
2323 ERR("Rotating stream output file");
2324 goto end_rcu_unlock
;
2326 /* Reset current size because we just perform a stream rotation. */
2327 stream
->tracefile_size_current
= 0;
2332 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2333 * index are NOT supported.
2335 if (session
->minor
>= 4 && !session
->snapshot
) {
2336 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2338 goto end_rcu_unlock
;
2342 /* Write data to stream output fd. */
2343 size_ret
= lttng_write(stream
->fd
, data_buffer
, data_size
);
2344 if (size_ret
< data_size
) {
2345 ERR("Relay error writing data to file");
2347 goto end_rcu_unlock
;
2350 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
2351 ret
, stream
->stream_handle
);
2353 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
2355 goto end_rcu_unlock
;
2357 stream
->tracefile_size_current
+= data_size
+ be32toh(data_hdr
.padding_size
);
2359 stream
->prev_seq
= net_seq_num
;
2361 try_close_stream(session
, stream
);
2370 void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2376 (void) lttng_poll_del(events
, pollfd
);
2378 ret
= close(pollfd
);
2380 ERR("Closing pollfd %d", pollfd
);
2384 static void destroy_connection(struct lttng_ht
*relay_connections_ht
,
2385 struct relay_connection
*conn
)
2387 assert(relay_connections_ht
);
2390 connection_delete(relay_connections_ht
, conn
);
2392 /* For the control socket, we try to destroy the session. */
2393 if (conn
->type
== RELAY_CONTROL
&& conn
->session
) {
2394 destroy_session(conn
->session
, conn
->sessions_ht
);
2397 connection_destroy(conn
);
2401 * This thread does the actual work
2404 void *relay_thread_worker(void *data
)
2406 int ret
, err
= -1, last_seen_data_fd
= -1;
2408 struct relay_connection
*conn
;
2409 struct lttng_poll_event events
;
2410 struct lttng_ht
*relay_connections_ht
;
2411 struct lttng_ht_iter iter
;
2412 struct lttcomm_relayd_hdr recv_hdr
;
2413 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
2414 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
2416 DBG("[thread] Relay worker started");
2418 rcu_register_thread();
2420 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2422 if (testpoint(relayd_thread_worker
)) {
2423 goto error_testpoint
;
2426 health_code_update();
2428 /* table of connections indexed on socket */
2429 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2430 if (!relay_connections_ht
) {
2431 goto relay_connections_ht_error
;
2434 /* Tables of received indexes indexed by index handle and net_seq_num. */
2435 indexes_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64
);
2437 goto indexes_ht_error
;
2440 ret
= create_thread_poll_set(&events
, 2);
2442 goto error_poll_create
;
2445 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2452 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2454 health_code_update();
2456 /* Infinite blocking call, waiting for transmission */
2457 DBG3("Relayd worker thread polling...");
2458 health_poll_entry();
2459 ret
= lttng_poll_wait(&events
, -1);
2463 * Restart interrupted system call.
2465 if (errno
== EINTR
) {
2474 * Process control. The control connection is prioritised so we don't
2475 * starve it with high throughout put tracing data on the data
2478 for (i
= 0; i
< nb_fd
; i
++) {
2479 /* Fetch once the poll data */
2480 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2481 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2483 health_code_update();
2485 /* Thread quit pipe has been closed. Killing thread. */
2486 ret
= check_thread_quit_pipe(pollfd
, revents
);
2492 /* Inspect the relay conn pipe for new connection */
2493 if (pollfd
== relay_conn_pipe
[0]) {
2494 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2495 ERR("Relay connection pipe error");
2497 } else if (revents
& LPOLLIN
) {
2498 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
2502 conn
->sessions_ht
= sessions_ht
;
2503 connection_init(conn
);
2504 lttng_poll_add(&events
, conn
->sock
->fd
,
2505 LPOLLIN
| LPOLLRDHUP
);
2507 lttng_ht_add_unique_ulong(relay_connections_ht
,
2510 DBG("Connection socket %d added", conn
->sock
->fd
);
2514 conn
= connection_find_by_sock(relay_connections_ht
, pollfd
);
2515 /* If not found, there is a synchronization issue. */
2518 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2519 cleanup_connection_pollfd(&events
, pollfd
);
2520 destroy_connection(relay_connections_ht
, conn
);
2521 if (last_seen_data_fd
== pollfd
) {
2522 last_seen_data_fd
= last_notdel_data_fd
;
2524 } else if (revents
& LPOLLIN
) {
2525 if (conn
->type
== RELAY_CONTROL
) {
2526 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &recv_hdr
,
2527 sizeof(recv_hdr
), 0);
2529 /* Connection closed */
2530 cleanup_connection_pollfd(&events
, pollfd
);
2531 destroy_connection(relay_connections_ht
, conn
);
2532 DBG("Control connection closed with %d", pollfd
);
2534 ret
= relay_process_control(&recv_hdr
, conn
);
2536 /* Clear the session on error. */
2537 cleanup_connection_pollfd(&events
, pollfd
);
2538 destroy_connection(relay_connections_ht
, conn
);
2539 DBG("Connection closed with %d", pollfd
);
2545 * Flag the last seen data fd not deleted. It will be
2546 * used as the last seen fd if any fd gets deleted in
2549 last_notdel_data_fd
= pollfd
;
2552 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
2559 * The last loop handled a control request, go back to poll to make
2560 * sure we prioritise the control socket.
2566 if (last_seen_data_fd
>= 0) {
2567 for (i
= 0; i
< nb_fd
; i
++) {
2568 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2570 health_code_update();
2572 if (last_seen_data_fd
== pollfd
) {
2579 /* Process data connection. */
2580 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2581 /* Fetch the poll data. */
2582 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2583 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2585 health_code_update();
2587 /* Skip the command pipe. It's handled in the first loop. */
2588 if (pollfd
== relay_conn_pipe
[0]) {
2594 conn
= connection_find_by_sock(relay_connections_ht
, pollfd
);
2596 /* Skip it. Might be removed before. */
2601 if (revents
& LPOLLIN
) {
2602 if (conn
->type
!= RELAY_DATA
) {
2606 ret
= relay_process_data(conn
);
2607 /* Connection closed */
2609 cleanup_connection_pollfd(&events
, pollfd
);
2610 destroy_connection(relay_connections_ht
, conn
);
2611 DBG("Data connection closed with %d", pollfd
);
2613 * Every goto restart call sets the last seen fd where
2614 * here we don't really care since we gracefully
2615 * continue the loop after the connection is deleted.
2618 /* Keep last seen port. */
2619 last_seen_data_fd
= pollfd
;
2627 last_seen_data_fd
= -1;
2630 /* Normal exit, no error */
2635 lttng_poll_clean(&events
);
2637 /* Cleanup reamaining connection object. */
2639 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, conn
,
2641 health_code_update();
2642 destroy_connection(relay_connections_ht
, conn
);
2646 lttng_ht_destroy(indexes_ht
);
2648 lttng_ht_destroy(relay_connections_ht
);
2649 relay_connections_ht_error
:
2650 /* Close relay conn pipes */
2651 utils_close_pipe(relay_conn_pipe
);
2653 DBG("Thread exited with error");
2655 DBG("Worker thread cleanup complete");
2660 ERR("Health error occurred in %s", __func__
);
2662 health_unregister(health_relayd
);
2663 rcu_unregister_thread();
2669 * Create the relay command pipe to wake thread_manage_apps.
2670 * Closed in cleanup().
2672 static int create_relay_conn_pipe(void)
2676 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
2684 int main(int argc
, char **argv
)
2688 struct relay_local_data
*relay_ctx
;
2690 /* Parse arguments */
2692 if ((ret
= set_options(argc
, argv
)) < 0) {
2696 if ((ret
= set_signal_handler()) < 0) {
2700 /* Try to create directory if -o, --output is specified. */
2701 if (opt_output_path
) {
2702 if (*opt_output_path
!= '/') {
2703 ERR("Please specify an absolute path for -o, --output PATH");
2707 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2709 ERR("Unable to create %s", opt_output_path
);
2715 if (opt_daemon
|| opt_background
) {
2718 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
2725 * We are in the child. Make sure all other file
2726 * descriptors are closed, in case we are called with
2727 * more opened file descriptors than the standard ones.
2729 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
2734 /* Create thread quit pipe */
2735 if ((ret
= init_thread_quit_pipe()) < 0) {
2739 /* We need those values for the file/dir creation. */
2740 relayd_uid
= getuid();
2741 relayd_gid
= getgid();
2743 /* Check if daemon is UID = 0 */
2744 if (relayd_uid
== 0) {
2745 if (control_uri
->port
< 1024 || data_uri
->port
< 1024 || live_uri
->port
< 1024) {
2746 ERR("Need to be root to use ports < 1024");
2752 /* Setup the thread apps communication pipe. */
2753 if ((ret
= create_relay_conn_pipe()) < 0) {
2757 /* Init relay command queue. */
2758 cds_wfq_init(&relay_conn_queue
.queue
);
2760 /* Set up max poll set size */
2761 lttng_poll_set_max_size();
2763 /* Initialize communication library */
2765 lttcomm_inet_init();
2767 relay_ctx
= zmalloc(sizeof(struct relay_local_data
));
2769 PERROR("relay_ctx");
2773 /* tables of sessions indexed by session ID */
2774 relay_ctx
->sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2775 if (!relay_ctx
->sessions_ht
) {
2776 goto exit_relay_ctx_sessions
;
2779 /* tables of streams indexed by stream ID */
2780 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2781 if (!relay_streams_ht
) {
2782 goto exit_relay_ctx_streams
;
2785 /* tables of streams indexed by stream ID */
2786 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2787 if (!viewer_streams_ht
) {
2788 goto exit_relay_ctx_viewer_streams
;
2791 /* Initialize thread health monitoring */
2792 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
2793 if (!health_relayd
) {
2794 PERROR("health_app_create error");
2795 goto exit_health_app_create
;
2798 ret
= utils_create_pipe(health_quit_pipe
);
2800 goto error_health_pipe
;
2803 /* Create thread to manage the client socket */
2804 ret
= pthread_create(&health_thread
, NULL
,
2805 thread_manage_health
, (void *) NULL
);
2807 PERROR("pthread_create health");
2811 /* Setup the dispatcher thread */
2812 ret
= pthread_create(&dispatcher_thread
, NULL
,
2813 relay_thread_dispatcher
, (void *) NULL
);
2815 PERROR("pthread_create dispatcher");
2816 goto exit_dispatcher
;
2819 /* Setup the worker thread */
2820 ret
= pthread_create(&worker_thread
, NULL
,
2821 relay_thread_worker
, (void *) relay_ctx
);
2823 PERROR("pthread_create worker");
2827 /* Setup the listener thread */
2828 ret
= pthread_create(&listener_thread
, NULL
,
2829 relay_thread_listener
, (void *) NULL
);
2831 PERROR("pthread_create listener");
2835 ret
= live_start_threads(live_uri
, relay_ctx
);
2837 ERR("Starting live viewer threads");
2842 ret
= pthread_join(listener_thread
, &status
);
2844 PERROR("pthread_join");
2845 goto error
; /* join error, exit without cleanup */
2849 ret
= pthread_join(worker_thread
, &status
);
2851 PERROR("pthread_join");
2852 goto error
; /* join error, exit without cleanup */
2856 ret
= pthread_join(dispatcher_thread
, &status
);
2858 PERROR("pthread_join");
2859 goto error
; /* join error, exit without cleanup */
2863 ret
= pthread_join(health_thread
, &status
);
2865 PERROR("pthread_join health thread");
2866 goto error
; /* join error, exit without cleanup */
2870 * Stop live threads only after joining other threads.
2872 live_stop_threads();
2875 utils_close_pipe(health_quit_pipe
);
2878 health_app_destroy(health_relayd
);
2880 exit_health_app_create
:
2881 lttng_ht_destroy(viewer_streams_ht
);
2883 exit_relay_ctx_viewer_streams
:
2884 lttng_ht_destroy(relay_streams_ht
);
2886 exit_relay_ctx_streams
:
2887 lttng_ht_destroy(relay_ctx
->sessions_ht
);
2889 exit_relay_ctx_sessions
: