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 /* Only 3 level of verbosity (-vvv). */
255 if (lttng_opt_verbose
< 3) {
256 lttng_opt_verbose
+= 1;
261 /* Unknown option or other error.
262 * Error is printed by getopt, just return */
275 * config_entry_handler_cb used to handle options read from a config file.
276 * See config_entry_handler_cb comment in common/config/config.h for the
277 * return value conventions.
280 int config_entry_handler(const struct config_entry
*entry
, void *unused
)
284 if (!entry
|| !entry
->name
|| !entry
->value
) {
289 /* Check if the option is to be ignored */
290 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
291 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
296 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
297 /* Ignore if entry name is not fully matched. */
298 if (strcmp(entry
->name
, long_options
[i
].name
)) {
303 * If the option takes no argument on the command line, we have to
304 * check if the value is "true". We support non-zero numeric values,
307 if (!long_options
[i
].has_arg
) {
308 ret
= config_parse_value(entry
->value
);
311 WARN("Invalid configuration value \"%s\" for option %s",
312 entry
->value
, entry
->name
);
314 /* False, skip boolean config option. */
319 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
323 WARN("Unrecognized option \"%s\" in daemon configuration file.",
331 int set_options(int argc
, char **argv
)
333 int c
, ret
= 0, option_index
= 0;
334 int orig_optopt
= optopt
, orig_optind
= optind
;
335 char *default_address
, *optstring
;
336 const char *config_path
= NULL
;
338 optstring
= utils_generate_optstring(long_options
,
339 sizeof(long_options
) / sizeof(struct option
));
345 /* Check for the --config option */
347 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
348 &option_index
)) != -1) {
352 } else if (c
!= 'f') {
356 config_path
= utils_expand_path(optarg
);
358 ERR("Failed to resolve path: %s", optarg
);
362 ret
= config_get_section_entries(config_path
, config_section_name
,
363 config_entry_handler
, NULL
);
366 ERR("Invalid configuration option at line %i", ret
);
372 /* Reset getopt's global state */
373 optopt
= orig_optopt
;
374 optind
= orig_optind
;
376 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
381 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
387 /* assign default values */
388 if (control_uri
== NULL
) {
389 ret
= asprintf(&default_address
,
390 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
391 DEFAULT_NETWORK_CONTROL_PORT
);
393 PERROR("asprintf default data address");
397 ret
= uri_parse(default_address
, &control_uri
);
398 free(default_address
);
400 ERR("Invalid control URI specified");
404 if (data_uri
== NULL
) {
405 ret
= asprintf(&default_address
,
406 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
407 DEFAULT_NETWORK_DATA_PORT
);
409 PERROR("asprintf default data address");
413 ret
= uri_parse(default_address
, &data_uri
);
414 free(default_address
);
416 ERR("Invalid data URI specified");
420 if (live_uri
== NULL
) {
421 ret
= asprintf(&default_address
,
422 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
423 DEFAULT_NETWORK_VIEWER_PORT
);
425 PERROR("asprintf default viewer control address");
429 ret
= uri_parse(default_address
, &live_uri
);
430 free(default_address
);
432 ERR("Invalid viewer control URI specified");
450 /* free the dynamically allocated opt_output_path */
451 free(opt_output_path
);
453 /* Close thread quit pipes */
454 utils_close_pipe(thread_quit_pipe
);
456 uri_free(control_uri
);
458 /* Live URI is freed in the live thread. */
460 if (tracing_group_name_override
) {
461 free((void *) tracing_group_name
);
466 * Write to writable pipe used to notify a thread.
469 int notify_thread_pipe(int wpipe
)
473 ret
= lttng_write(wpipe
, "!", 1);
475 PERROR("write poll pipe");
481 static void notify_health_quit_pipe(int *pipe
)
485 ret
= lttng_write(pipe
[1], "4", 1);
487 PERROR("write relay health quit");
492 * Stop all threads by closing the thread quit pipe.
495 void stop_threads(void)
499 /* Stopping all threads */
500 DBG("Terminating all threads");
501 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
503 ERR("write error on thread quit pipe");
506 notify_health_quit_pipe(health_quit_pipe
);
508 /* Dispatch thread */
509 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
510 futex_nto1_wake(&relay_conn_queue
.futex
);
514 * Signal handler for the daemon
516 * Simply stop all worker threads, leaving main() return gracefully after
517 * joining all threads and calling cleanup().
520 void sighandler(int sig
)
524 DBG("SIGPIPE caught");
527 DBG("SIGINT caught");
531 DBG("SIGTERM caught");
535 CMM_STORE_SHARED(recv_child_signal
, 1);
543 * Setup signal handler for :
544 * SIGINT, SIGTERM, SIGPIPE
547 int set_signal_handler(void)
553 if ((ret
= sigemptyset(&sigset
)) < 0) {
554 PERROR("sigemptyset");
558 sa
.sa_handler
= sighandler
;
561 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
566 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
571 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
576 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
581 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
586 void lttng_relay_notify_ready(void)
588 /* Notify the parent of the fork() process that we are ready. */
589 if (opt_daemon
|| opt_background
) {
590 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
591 kill(child_ppid
, SIGUSR1
);
597 * Init thread quit pipe.
599 * Return -1 on error or 0 if all pipes are created.
602 int init_thread_quit_pipe(void)
606 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
612 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
615 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
619 if (events
== NULL
|| size
== 0) {
624 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
630 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
642 * Check if the thread quit pipe was triggered.
644 * Return 1 if it was triggered else 0;
647 int check_thread_quit_pipe(int fd
, uint32_t events
)
649 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
657 * Create and init socket from uri.
660 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
663 struct lttcomm_sock
*sock
= NULL
;
665 sock
= lttcomm_alloc_sock_from_uri(uri
);
667 ERR("Allocating socket");
671 ret
= lttcomm_create_sock(sock
);
675 DBG("Listening on sock %d", sock
->fd
);
677 ret
= sock
->ops
->bind(sock
);
682 ret
= sock
->ops
->listen(sock
, -1);
692 lttcomm_destroy_sock(sock
);
698 * Return nonzero if stream needs to be closed.
701 int close_stream_check(struct relay_stream
*stream
)
703 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
705 * We are about to close the stream so set the data pending flag to 1
706 * which will make the end data pending command skip the stream which
707 * is now closed and ready. Note that after proceeding to a file close,
708 * the written file is ready for reading.
710 stream
->data_pending_check_done
= 1;
716 static void try_close_stream(struct relay_session
*session
,
717 struct relay_stream
*stream
)
720 struct ctf_trace
*ctf_trace
;
725 if (!close_stream_check(stream
)) {
726 /* Can't close it, not ready for that. */
730 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
,
734 pthread_mutex_lock(&session
->viewer_ready_lock
);
735 ctf_trace
->invalid_flag
= 1;
736 pthread_mutex_unlock(&session
->viewer_ready_lock
);
738 ret
= stream_close(session
, stream
);
739 if (ret
|| session
->snapshot
) {
740 /* Already close thus the ctf trace is being or has been destroyed. */
744 ctf_trace_try_destroy(session
, ctf_trace
);
751 * This thread manages the listening for new connections on the network
754 void *relay_thread_listener(void *data
)
756 int i
, ret
, pollfd
, err
= -1;
757 uint32_t revents
, nb_fd
;
758 struct lttng_poll_event events
;
759 struct lttcomm_sock
*control_sock
, *data_sock
;
761 DBG("[thread] Relay listener started");
763 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
765 health_code_update();
767 control_sock
= relay_init_sock(control_uri
);
769 goto error_sock_control
;
772 data_sock
= relay_init_sock(data_uri
);
774 goto error_sock_relay
;
778 * Pass 3 as size here for the thread quit pipe, control and data socket.
780 ret
= create_thread_poll_set(&events
, 3);
782 goto error_create_poll
;
785 /* Add the control socket */
786 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
791 /* Add the data socket */
792 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
797 lttng_relay_notify_ready();
799 if (testpoint(relayd_thread_listener
)) {
800 goto error_testpoint
;
804 health_code_update();
806 DBG("Listener accepting connections");
810 ret
= lttng_poll_wait(&events
, -1);
814 * Restart interrupted system call.
816 if (errno
== EINTR
) {
824 DBG("Relay new connection received");
825 for (i
= 0; i
< nb_fd
; i
++) {
826 health_code_update();
828 /* Fetch once the poll data */
829 revents
= LTTNG_POLL_GETEV(&events
, i
);
830 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
832 /* Thread quit pipe has been closed. Killing thread. */
833 ret
= check_thread_quit_pipe(pollfd
, revents
);
839 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
840 ERR("socket poll error");
842 } else if (revents
& LPOLLIN
) {
844 * Get allocated in this thread, enqueued to a global queue,
845 * dequeued and freed in the worker thread.
848 struct relay_connection
*new_conn
;
849 struct lttcomm_sock
*newsock
;
851 new_conn
= connection_create();
856 if (pollfd
== data_sock
->fd
) {
857 new_conn
->type
= RELAY_DATA
;
858 newsock
= data_sock
->ops
->accept(data_sock
);
859 DBG("Relay data connection accepted, socket %d",
862 assert(pollfd
== control_sock
->fd
);
863 new_conn
->type
= RELAY_CONTROL
;
864 newsock
= control_sock
->ops
->accept(control_sock
);
865 DBG("Relay control connection accepted, socket %d",
869 PERROR("accepting sock");
870 connection_free(new_conn
);
874 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
877 PERROR("setsockopt inet");
878 lttcomm_destroy_sock(newsock
);
879 connection_free(new_conn
);
882 new_conn
->sock
= newsock
;
884 /* Enqueue request for the dispatcher thread. */
885 cds_wfq_enqueue(&relay_conn_queue
.queue
, &new_conn
->qnode
);
888 * Wake the dispatch queue futex. Implicit memory barrier with
889 * the exchange in cds_wfq_enqueue.
891 futex_nto1_wake(&relay_conn_queue
.futex
);
900 lttng_poll_clean(&events
);
902 if (data_sock
->fd
>= 0) {
903 ret
= data_sock
->ops
->close(data_sock
);
908 lttcomm_destroy_sock(data_sock
);
910 if (control_sock
->fd
>= 0) {
911 ret
= control_sock
->ops
->close(control_sock
);
916 lttcomm_destroy_sock(control_sock
);
920 ERR("Health error occurred in %s", __func__
);
922 health_unregister(health_relayd
);
923 DBG("Relay listener thread cleanup complete");
929 * This thread manages the dispatching of the requests to worker threads
932 void *relay_thread_dispatcher(void *data
)
936 struct cds_wfq_node
*node
;
937 struct relay_connection
*new_conn
= NULL
;
939 DBG("[thread] Relay dispatcher started");
941 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
943 if (testpoint(relayd_thread_dispatcher
)) {
944 goto error_testpoint
;
947 health_code_update();
949 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
950 health_code_update();
952 /* Atomically prepare the queue futex */
953 futex_nto1_prepare(&relay_conn_queue
.futex
);
956 health_code_update();
958 /* Dequeue commands */
959 node
= cds_wfq_dequeue_blocking(&relay_conn_queue
.queue
);
961 DBG("Woken up but nothing in the relay command queue");
962 /* Continue thread execution */
965 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
967 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
970 * Inform worker thread of the new request. This call is blocking
971 * so we can be assured that the data will be read at some point in
972 * time or wait to the end of the world :)
974 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
976 PERROR("write connection pipe");
977 connection_destroy(new_conn
);
980 } while (node
!= NULL
);
982 /* Futex wait on queue. Blocking call on futex() */
984 futex_nto1_wait(&relay_conn_queue
.futex
);
988 /* Normal exit, no error */
995 ERR("Health error occurred in %s", __func__
);
997 health_unregister(health_relayd
);
998 DBG("Dispatch thread dying");
1003 static void try_close_streams(struct relay_session
*session
)
1005 struct ctf_trace
*ctf_trace
;
1006 struct lttng_ht_iter iter
;
1010 pthread_mutex_lock(&session
->viewer_ready_lock
);
1012 cds_lfht_for_each_entry(session
->ctf_traces_ht
->ht
, &iter
.iter
, ctf_trace
,
1014 struct relay_stream
*stream
;
1016 /* Close streams. */
1017 cds_list_for_each_entry(stream
, &ctf_trace
->stream_list
, trace_list
) {
1018 stream_close(session
, stream
);
1021 ctf_trace
->invalid_flag
= 1;
1022 ctf_trace_try_destroy(session
, ctf_trace
);
1025 pthread_mutex_unlock(&session
->viewer_ready_lock
);
1029 * Try to destroy a session within a connection.
1031 static void destroy_session(struct relay_session
*session
,
1032 struct lttng_ht
*sessions_ht
)
1035 assert(sessions_ht
);
1037 /* Indicate that this session can be destroyed from now on. */
1038 session
->close_flag
= 1;
1040 try_close_streams(session
);
1043 * This will try to delete and destroy the session if no viewer is attached
1044 * to it meaning the refcount is down to zero.
1046 session_try_destroy(sessions_ht
, session
);
1050 * Copy index data from the control port to a given index object.
1052 static void copy_index_control_data(struct relay_index
*index
,
1053 struct lttcomm_relayd_index
*data
)
1059 * The index on disk is encoded in big endian, so we don't need to convert
1060 * the data received on the network. The data_offset value is NEVER
1061 * modified here and is updated by the data thread.
1063 index
->index_data
.packet_size
= data
->packet_size
;
1064 index
->index_data
.content_size
= data
->content_size
;
1065 index
->index_data
.timestamp_begin
= data
->timestamp_begin
;
1066 index
->index_data
.timestamp_end
= data
->timestamp_end
;
1067 index
->index_data
.events_discarded
= data
->events_discarded
;
1068 index
->index_data
.stream_id
= data
->stream_id
;
1072 * Handle the RELAYD_CREATE_SESSION command.
1074 * On success, send back the session id or else return a negative value.
1077 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
1078 struct relay_connection
*conn
)
1080 int ret
= 0, send_ret
;
1081 struct relay_session
*session
;
1082 struct lttcomm_relayd_status_session reply
;
1087 memset(&reply
, 0, sizeof(reply
));
1089 session
= session_create();
1094 session
->minor
= conn
->minor
;
1095 session
->major
= conn
->major
;
1096 conn
->session_id
= session
->id
;
1097 conn
->session
= session
;
1099 reply
.session_id
= htobe64(session
->id
);
1101 switch (conn
->minor
) {
1106 case 4: /* LTTng sessiond 2.4 */
1108 ret
= cmd_create_session_2_4(conn
, session
);
1111 lttng_ht_add_unique_u64(conn
->sessions_ht
, &session
->session_n
);
1112 DBG("Created session %" PRIu64
, session
->id
);
1116 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1118 reply
.ret_code
= htobe32(LTTNG_OK
);
1121 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1123 ERR("Relayd sending session id");
1131 * When we have received all the streams and the metadata for a channel,
1132 * we make them visible to the viewer threads.
1135 void set_viewer_ready_flag(struct relay_connection
*conn
)
1137 struct relay_stream
*stream
, *tmp_stream
;
1139 pthread_mutex_lock(&conn
->session
->viewer_ready_lock
);
1140 cds_list_for_each_entry_safe(stream
, tmp_stream
, &conn
->recv_head
,
1142 stream
->viewer_ready
= 1;
1143 cds_list_del(&stream
->recv_list
);
1145 pthread_mutex_unlock(&conn
->session
->viewer_ready_lock
);
1150 * Add a recv handle node to the connection recv list with the given stream
1151 * handle. A new node is allocated thus must be freed when the node is deleted
1154 static void queue_stream(struct relay_stream
*stream
,
1155 struct relay_connection
*conn
)
1160 cds_list_add(&stream
->recv_list
, &conn
->recv_head
);
1164 * relay_add_stream: allocate a new stream for a session
1167 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1168 struct relay_connection
*conn
)
1171 struct relay_session
*session
= conn
->session
;
1172 struct relay_stream
*stream
= NULL
;
1173 struct lttcomm_relayd_status_stream reply
;
1174 struct ctf_trace
*trace
;
1176 if (!session
|| conn
->version_check_done
== 0) {
1177 ERR("Trying to add a stream before version check");
1179 goto end_no_session
;
1182 stream
= zmalloc(sizeof(struct relay_stream
));
1183 if (stream
== NULL
) {
1184 PERROR("relay stream zmalloc");
1186 goto end_no_session
;
1189 switch (conn
->minor
) {
1190 case 1: /* LTTng sessiond 2.1 */
1191 ret
= cmd_recv_stream_2_1(conn
, stream
);
1193 case 2: /* LTTng sessiond 2.2 */
1195 ret
= cmd_recv_stream_2_2(conn
, stream
);
1199 goto err_free_stream
;
1203 stream
->stream_handle
= ++last_relay_stream_id
;
1204 stream
->prev_seq
= -1ULL;
1205 stream
->session_id
= session
->id
;
1206 stream
->index_fd
= -1;
1207 stream
->read_index_fd
= -1;
1208 stream
->ctf_stream_id
= -1ULL;
1209 lttng_ht_node_init_u64(&stream
->node
, stream
->stream_handle
);
1210 pthread_mutex_init(&stream
->lock
, NULL
);
1212 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
);
1214 ERR("relay creating output directory");
1219 * No need to use run_as API here because whatever we receives, the relayd
1220 * uses its own credentials for the stream files.
1222 ret
= utils_create_stream_file(stream
->path_name
, stream
->channel_name
,
1223 stream
->tracefile_size
, 0, relayd_uid
, relayd_gid
, NULL
);
1225 ERR("Create output file");
1229 if (stream
->tracefile_size
) {
1230 DBG("Tracefile %s/%s_0 created", stream
->path_name
, stream
->channel_name
);
1232 DBG("Tracefile %s/%s created", stream
->path_name
, stream
->channel_name
);
1235 trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
, stream
->path_name
);
1237 trace
= ctf_trace_create(stream
->path_name
);
1242 ctf_trace_add(session
->ctf_traces_ht
, trace
);
1244 ctf_trace_get_ref(trace
);
1246 if (!strncmp(stream
->channel_name
, DEFAULT_METADATA_NAME
, NAME_MAX
)) {
1247 stream
->metadata_flag
= 1;
1248 /* Assign quick reference to the metadata stream in the trace. */
1249 trace
->metadata_stream
= stream
;
1253 * Add the stream in the recv list of the connection. Once the end stream
1254 * message is received, this list is emptied and streams are set with the
1255 * viewer ready flag.
1257 queue_stream(stream
, conn
);
1260 * Both in the ctf_trace object and the global stream ht since the data
1261 * side of the relayd does not have the concept of session.
1263 lttng_ht_add_unique_u64(relay_streams_ht
, &stream
->node
);
1264 cds_list_add_tail(&stream
->trace_list
, &trace
->stream_list
);
1266 session
->stream_count
++;
1268 DBG("Relay new stream added %s with ID %" PRIu64
, stream
->channel_name
,
1269 stream
->stream_handle
);
1272 memset(&reply
, 0, sizeof(reply
));
1273 reply
.handle
= htobe64(stream
->stream_handle
);
1274 /* send the session id to the client or a negative return code on error */
1276 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1277 /* stream was not properly added to the ht, so free it */
1280 reply
.ret_code
= htobe32(LTTNG_OK
);
1283 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1284 sizeof(struct lttcomm_relayd_status_stream
), 0);
1286 ERR("Relay sending stream id");
1295 free(stream
->path_name
);
1296 free(stream
->channel_name
);
1302 * relay_close_stream: close a specific stream
1305 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1306 struct relay_connection
*conn
)
1309 struct relay_session
*session
= conn
->session
;
1310 struct lttcomm_relayd_close_stream stream_info
;
1311 struct lttcomm_relayd_generic_reply reply
;
1312 struct relay_stream
*stream
;
1314 DBG("Close stream received");
1316 if (!session
|| conn
->version_check_done
== 0) {
1317 ERR("Trying to close a stream before version check");
1319 goto end_no_session
;
1322 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
1323 sizeof(struct lttcomm_relayd_close_stream
), 0);
1324 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1326 /* Orderly shutdown. Not necessary to print an error. */
1327 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1329 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1332 goto end_no_session
;
1336 stream
= stream_find_by_id(relay_streams_ht
,
1337 be64toh(stream_info
.stream_id
));
1343 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1344 stream
->close_flag
= 1;
1345 session
->stream_count
--;
1346 assert(session
->stream_count
>= 0);
1348 /* Check if we can close it or else the data will do it. */
1349 try_close_stream(session
, stream
);
1354 memset(&reply
, 0, sizeof(reply
));
1356 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1358 reply
.ret_code
= htobe32(LTTNG_OK
);
1360 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1361 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1363 ERR("Relay sending stream id");
1372 * relay_unknown_command: send -1 if received unknown command
1375 void relay_unknown_command(struct relay_connection
*conn
)
1377 struct lttcomm_relayd_generic_reply reply
;
1380 memset(&reply
, 0, sizeof(reply
));
1381 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1382 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1383 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1385 ERR("Relay sending unknown command");
1390 * relay_start: send an acknowledgment to the client to tell if we are
1391 * ready to receive data. We are ready if a session is established.
1394 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1395 struct relay_connection
*conn
)
1397 int ret
= htobe32(LTTNG_OK
);
1398 struct lttcomm_relayd_generic_reply reply
;
1399 struct relay_session
*session
= conn
->session
;
1402 DBG("Trying to start the streaming without a session established");
1403 ret
= htobe32(LTTNG_ERR_UNK
);
1406 memset(&reply
, 0, sizeof(reply
));
1407 reply
.ret_code
= ret
;
1408 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1409 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1411 ERR("Relay sending start ack");
1418 * Append padding to the file pointed by the file descriptor fd.
1420 static int write_padding_to_file(int fd
, uint32_t size
)
1429 zeros
= zmalloc(size
);
1430 if (zeros
== NULL
) {
1431 PERROR("zmalloc zeros for padding");
1436 ret
= lttng_write(fd
, zeros
, size
);
1438 PERROR("write padding to file");
1448 * relay_recv_metadata: receive the metada for the session.
1451 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1452 struct relay_connection
*conn
)
1454 int ret
= htobe32(LTTNG_OK
);
1456 struct relay_session
*session
= conn
->session
;
1457 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1458 struct relay_stream
*metadata_stream
;
1459 uint64_t data_size
, payload_size
;
1460 struct ctf_trace
*ctf_trace
;
1463 ERR("Metadata sent before version check");
1468 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1469 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1470 ERR("Incorrect data size");
1474 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1476 if (data_buffer_size
< data_size
) {
1477 /* In case the realloc fails, we can free the memory */
1480 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1481 if (!tmp_data_ptr
) {
1482 ERR("Allocating data buffer");
1487 data_buffer
= tmp_data_ptr
;
1488 data_buffer_size
= data_size
;
1490 memset(data_buffer
, 0, data_size
);
1491 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1492 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
1493 if (ret
< 0 || ret
!= data_size
) {
1495 /* Orderly shutdown. Not necessary to print an error. */
1496 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1498 ERR("Relay didn't receive the whole metadata");
1503 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1506 metadata_stream
= stream_find_by_id(relay_streams_ht
,
1507 be64toh(metadata_struct
->stream_id
));
1508 if (!metadata_stream
) {
1513 size_ret
= lttng_write(metadata_stream
->fd
, metadata_struct
->payload
,
1515 if (size_ret
< payload_size
) {
1516 ERR("Relay error writing metadata on file");
1521 ret
= write_padding_to_file(metadata_stream
->fd
,
1522 be32toh(metadata_struct
->padding_size
));
1527 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
,
1528 metadata_stream
->path_name
);
1530 ctf_trace
->metadata_received
+=
1531 payload_size
+ be32toh(metadata_struct
->padding_size
);
1533 DBG2("Relay metadata written");
1542 * relay_send_version: send relayd version number
1545 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1546 struct relay_connection
*conn
)
1549 struct lttcomm_relayd_version reply
, msg
;
1553 conn
->version_check_done
= 1;
1555 /* Get version from the other side. */
1556 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1557 if (ret
< 0 || ret
!= sizeof(msg
)) {
1559 /* Orderly shutdown. Not necessary to print an error. */
1560 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1562 ERR("Relay failed to receive the version values.");
1568 memset(&reply
, 0, sizeof(reply
));
1569 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1570 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1572 /* Major versions must be the same */
1573 if (reply
.major
!= be32toh(msg
.major
)) {
1574 DBG("Incompatible major versions (%u vs %u), deleting session",
1575 reply
.major
, be32toh(msg
.major
));
1576 destroy_session(conn
->session
, conn
->sessions_ht
);
1581 conn
->major
= reply
.major
;
1582 /* We adapt to the lowest compatible version */
1583 if (reply
.minor
<= be32toh(msg
.minor
)) {
1584 conn
->minor
= reply
.minor
;
1586 conn
->minor
= be32toh(msg
.minor
);
1589 reply
.major
= htobe32(reply
.major
);
1590 reply
.minor
= htobe32(reply
.minor
);
1591 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1592 sizeof(struct lttcomm_relayd_version
), 0);
1594 ERR("Relay sending version");
1597 DBG("Version check done using protocol %u.%u", conn
->major
,
1605 * Check for data pending for a given stream id from the session daemon.
1608 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1609 struct relay_connection
*conn
)
1611 struct relay_session
*session
= conn
->session
;
1612 struct lttcomm_relayd_data_pending msg
;
1613 struct lttcomm_relayd_generic_reply reply
;
1614 struct relay_stream
*stream
;
1616 uint64_t last_net_seq_num
, stream_id
;
1618 DBG("Data pending command received");
1620 if (!session
|| conn
->version_check_done
== 0) {
1621 ERR("Trying to check for data before version check");
1623 goto end_no_session
;
1626 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1627 if (ret
< sizeof(msg
)) {
1629 /* Orderly shutdown. Not necessary to print an error. */
1630 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1632 ERR("Relay didn't receive valid data_pending struct size : %d",
1636 goto end_no_session
;
1639 stream_id
= be64toh(msg
.stream_id
);
1640 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1643 stream
= stream_find_by_id(relay_streams_ht
, stream_id
);
1644 if (stream
== NULL
) {
1649 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1650 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1653 /* Avoid wrapping issue */
1654 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1655 /* Data has in fact been written and is NOT pending */
1658 /* Data still being streamed thus pending */
1662 /* Pending check is now done. */
1663 stream
->data_pending_check_done
= 1;
1668 memset(&reply
, 0, sizeof(reply
));
1669 reply
.ret_code
= htobe32(ret
);
1670 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1672 ERR("Relay data pending ret code failed");
1680 * Wait for the control socket to reach a quiescent state.
1682 * Note that for now, when receiving this command from the session daemon, this
1683 * means that every subsequent commands or data received on the control socket
1684 * has been handled. So, this is why we simply return OK here.
1687 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1688 struct relay_connection
*conn
)
1692 struct relay_stream
*stream
;
1693 struct lttng_ht_iter iter
;
1694 struct lttcomm_relayd_quiescent_control msg
;
1695 struct lttcomm_relayd_generic_reply reply
;
1697 DBG("Checking quiescent state on control socket");
1699 if (!conn
->session
|| conn
->version_check_done
== 0) {
1700 ERR("Trying to check for data before version check");
1702 goto end_no_session
;
1705 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1706 if (ret
< sizeof(msg
)) {
1708 /* Orderly shutdown. Not necessary to print an error. */
1709 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1711 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1715 goto end_no_session
;
1718 stream_id
= be64toh(msg
.stream_id
);
1721 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1723 if (stream
->stream_handle
== stream_id
) {
1724 stream
->data_pending_check_done
= 1;
1725 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1732 memset(&reply
, 0, sizeof(reply
));
1733 reply
.ret_code
= htobe32(LTTNG_OK
);
1734 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1736 ERR("Relay data quiescent control ret code failed");
1744 * Initialize a data pending command. This means that a client is about to ask
1745 * for data pending for each stream he/she holds. Simply iterate over all
1746 * streams of a session and set the data_pending_check_done flag.
1748 * This command returns to the client a LTTNG_OK code.
1751 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1752 struct relay_connection
*conn
)
1755 struct lttng_ht_iter iter
;
1756 struct lttcomm_relayd_begin_data_pending msg
;
1757 struct lttcomm_relayd_generic_reply reply
;
1758 struct relay_stream
*stream
;
1759 uint64_t session_id
;
1764 DBG("Init streams for data pending");
1766 if (!conn
->session
|| conn
->version_check_done
== 0) {
1767 ERR("Trying to check for data before version check");
1769 goto end_no_session
;
1772 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1773 if (ret
< sizeof(msg
)) {
1775 /* Orderly shutdown. Not necessary to print an error. */
1776 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1778 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1782 goto end_no_session
;
1785 session_id
= be64toh(msg
.session_id
);
1788 * Iterate over all streams to set the begin data pending flag. For now, the
1789 * streams are indexed by stream handle so we have to iterate over all
1790 * streams to find the one associated with the right session_id.
1793 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1795 if (stream
->session_id
== session_id
) {
1796 stream
->data_pending_check_done
= 0;
1797 DBG("Set begin data pending flag to stream %" PRIu64
,
1798 stream
->stream_handle
);
1803 memset(&reply
, 0, sizeof(reply
));
1804 /* All good, send back reply. */
1805 reply
.ret_code
= htobe32(LTTNG_OK
);
1807 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1809 ERR("Relay begin data pending send reply failed");
1817 * End data pending command. This will check, for a given session id, if each
1818 * stream associated with it has its data_pending_check_done flag set. If not,
1819 * this means that the client lost track of the stream but the data is still
1820 * being streamed on our side. In this case, we inform the client that data is
1823 * Return to the client if there is data in flight or not with a ret_code.
1826 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1827 struct relay_connection
*conn
)
1830 struct lttng_ht_iter iter
;
1831 struct lttcomm_relayd_end_data_pending msg
;
1832 struct lttcomm_relayd_generic_reply reply
;
1833 struct relay_stream
*stream
;
1834 uint64_t session_id
;
1835 uint32_t is_data_inflight
= 0;
1840 DBG("End data pending command");
1842 if (!conn
->session
|| conn
->version_check_done
== 0) {
1843 ERR("Trying to check for data before version check");
1845 goto end_no_session
;
1848 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1849 if (ret
< sizeof(msg
)) {
1851 /* Orderly shutdown. Not necessary to print an error. */
1852 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1854 ERR("Relay didn't receive valid end data_pending struct size: %d",
1858 goto end_no_session
;
1861 session_id
= be64toh(msg
.session_id
);
1863 /* Iterate over all streams to see if the begin data pending flag is set. */
1865 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1867 if (stream
->session_id
== session_id
&&
1868 !stream
->data_pending_check_done
&& !stream
->terminated_flag
) {
1869 is_data_inflight
= 1;
1870 DBG("Data is still in flight for stream %" PRIu64
,
1871 stream
->stream_handle
);
1877 memset(&reply
, 0, sizeof(reply
));
1878 /* All good, send back reply. */
1879 reply
.ret_code
= htobe32(is_data_inflight
);
1881 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1883 ERR("Relay end data pending send reply failed");
1891 * Receive an index for a specific stream.
1893 * Return 0 on success else a negative value.
1896 int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1897 struct relay_connection
*conn
)
1899 int ret
, send_ret
, index_created
= 0;
1900 struct relay_session
*session
= conn
->session
;
1901 struct lttcomm_relayd_index index_info
;
1902 struct relay_index
*index
, *wr_index
= NULL
;
1903 struct lttcomm_relayd_generic_reply reply
;
1904 struct relay_stream
*stream
;
1905 uint64_t net_seq_num
;
1909 DBG("Relay receiving index");
1911 if (!session
|| conn
->version_check_done
== 0) {
1912 ERR("Trying to close a stream before version check");
1914 goto end_no_session
;
1917 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &index_info
,
1918 sizeof(index_info
), 0);
1919 if (ret
< sizeof(index_info
)) {
1921 /* Orderly shutdown. Not necessary to print an error. */
1922 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1924 ERR("Relay didn't receive valid index struct size : %d", ret
);
1927 goto end_no_session
;
1930 net_seq_num
= be64toh(index_info
.net_seq_num
);
1933 stream
= stream_find_by_id(relay_streams_ht
,
1934 be64toh(index_info
.relay_stream_id
));
1937 goto end_rcu_unlock
;
1940 /* Live beacon handling */
1941 if (index_info
.packet_size
== 0) {
1942 DBG("Received live beacon for stream %" PRIu64
, stream
->stream_handle
);
1945 * Only flag a stream inactive when it has already received data.
1947 if (stream
->total_index_received
> 0) {
1948 stream
->beacon_ts_end
= be64toh(index_info
.timestamp_end
);
1951 goto end_rcu_unlock
;
1953 stream
->beacon_ts_end
= -1ULL;
1956 index
= relay_index_find(stream
->stream_handle
, net_seq_num
);
1958 /* A successful creation will add the object to the HT. */
1959 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
1961 goto end_rcu_unlock
;
1966 copy_index_control_data(index
, &index_info
);
1967 if (stream
->ctf_stream_id
== -1ULL) {
1968 stream
->ctf_stream_id
= be64toh(index_info
.stream_id
);
1971 if (index_created
) {
1973 * Try to add the relay index object to the hash table. If an object
1974 * already exist, destroy back the index created, set the data in this
1975 * object and write it on disk.
1977 relay_index_add(index
, &wr_index
);
1979 copy_index_control_data(wr_index
, &index_info
);
1983 /* The index already exists so write it on disk. */
1987 /* Do we have a writable ready index to write on disk. */
1989 ret
= relay_index_write(wr_index
->fd
, wr_index
);
1991 goto end_rcu_unlock
;
1993 stream
->total_index_received
++;
1999 memset(&reply
, 0, sizeof(reply
));
2001 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2003 reply
.ret_code
= htobe32(LTTNG_OK
);
2005 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2007 ERR("Relay sending close index id reply");
2016 * Receive the streams_sent message.
2018 * Return 0 on success else a negative value.
2021 int relay_streams_sent(struct lttcomm_relayd_hdr
*recv_hdr
,
2022 struct relay_connection
*conn
)
2025 struct lttcomm_relayd_generic_reply reply
;
2029 DBG("Relay receiving streams_sent");
2031 if (!conn
->session
|| conn
->version_check_done
== 0) {
2032 ERR("Trying to close a stream before version check");
2034 goto end_no_session
;
2038 * Flag every pending stream in the connection recv list that they are
2039 * ready to be used by the viewer.
2041 set_viewer_ready_flag(conn
);
2044 * Inform the viewer that there are new streams in the session.
2046 if (conn
->session
->viewer_refcount
) {
2047 uatomic_set(&conn
->session
->new_streams
, 1);
2050 memset(&reply
, 0, sizeof(reply
));
2051 reply
.ret_code
= htobe32(LTTNG_OK
);
2052 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2054 ERR("Relay sending sent_stream reply");
2066 * Process the commands received on the control socket
2069 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
2070 struct relay_connection
*conn
)
2074 switch (be32toh(recv_hdr
->cmd
)) {
2075 case RELAYD_CREATE_SESSION
:
2076 ret
= relay_create_session(recv_hdr
, conn
);
2078 case RELAYD_ADD_STREAM
:
2079 ret
= relay_add_stream(recv_hdr
, conn
);
2081 case RELAYD_START_DATA
:
2082 ret
= relay_start(recv_hdr
, conn
);
2084 case RELAYD_SEND_METADATA
:
2085 ret
= relay_recv_metadata(recv_hdr
, conn
);
2087 case RELAYD_VERSION
:
2088 ret
= relay_send_version(recv_hdr
, conn
);
2090 case RELAYD_CLOSE_STREAM
:
2091 ret
= relay_close_stream(recv_hdr
, conn
);
2093 case RELAYD_DATA_PENDING
:
2094 ret
= relay_data_pending(recv_hdr
, conn
);
2096 case RELAYD_QUIESCENT_CONTROL
:
2097 ret
= relay_quiescent_control(recv_hdr
, conn
);
2099 case RELAYD_BEGIN_DATA_PENDING
:
2100 ret
= relay_begin_data_pending(recv_hdr
, conn
);
2102 case RELAYD_END_DATA_PENDING
:
2103 ret
= relay_end_data_pending(recv_hdr
, conn
);
2105 case RELAYD_SEND_INDEX
:
2106 ret
= relay_recv_index(recv_hdr
, conn
);
2108 case RELAYD_STREAMS_SENT
:
2109 ret
= relay_streams_sent(recv_hdr
, conn
);
2111 case RELAYD_UPDATE_SYNC_INFO
:
2113 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
2114 relay_unknown_command(conn
);
2124 * Handle index for a data stream.
2126 * RCU read side lock MUST be acquired.
2128 * Return 0 on success else a negative value.
2130 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2133 int ret
= 0, index_created
= 0;
2134 uint64_t stream_id
, data_offset
;
2135 struct relay_index
*index
, *wr_index
= NULL
;
2139 stream_id
= stream
->stream_handle
;
2140 /* Get data offset because we are about to update the index. */
2141 data_offset
= htobe64(stream
->tracefile_size_current
);
2144 * Lookup for an existing index for that stream id/sequence number. If on
2145 * exists, the control thread already received the data for it thus we need
2146 * to write it on disk.
2148 index
= relay_index_find(stream_id
, net_seq_num
);
2150 /* A successful creation will add the object to the HT. */
2151 index
= relay_index_create(stream_id
, net_seq_num
);
2159 if (rotate_index
|| stream
->index_fd
< 0) {
2160 index
->to_close_fd
= stream
->index_fd
;
2161 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
2162 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
2163 stream
->tracefile_count_current
);
2165 /* This will close the stream's index fd if one. */
2166 relay_index_free_safe(index
);
2169 stream
->index_fd
= ret
;
2171 index
->fd
= stream
->index_fd
;
2172 index
->index_data
.offset
= data_offset
;
2174 if (index_created
) {
2176 * Try to add the relay index object to the hash table. If an object
2177 * already exist, destroy back the index created and set the data.
2179 relay_index_add(index
, &wr_index
);
2181 /* Copy back data from the created index. */
2182 wr_index
->fd
= index
->fd
;
2183 wr_index
->to_close_fd
= index
->to_close_fd
;
2184 wr_index
->index_data
.offset
= data_offset
;
2188 /* The index already exists so write it on disk. */
2192 /* Do we have a writable ready index to write on disk. */
2194 ret
= relay_index_write(wr_index
->fd
, wr_index
);
2198 stream
->total_index_received
++;
2206 * relay_process_data: Process the data received on the data socket
2209 int relay_process_data(struct relay_connection
*conn
)
2211 int ret
= 0, rotate_index
= 0;
2213 struct relay_stream
*stream
;
2214 struct lttcomm_relayd_data_hdr data_hdr
;
2216 uint64_t net_seq_num
;
2218 struct relay_session
*session
;
2222 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &data_hdr
,
2223 sizeof(struct lttcomm_relayd_data_hdr
), 0);
2226 /* Orderly shutdown. Not necessary to print an error. */
2227 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2229 ERR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2235 stream_id
= be64toh(data_hdr
.stream_id
);
2238 stream
= stream_find_by_id(relay_streams_ht
, stream_id
);
2241 goto end_rcu_unlock
;
2244 session
= session_find_by_id(conn
->sessions_ht
, stream
->session_id
);
2247 data_size
= be32toh(data_hdr
.data_size
);
2248 if (data_buffer_size
< data_size
) {
2251 tmp_data_ptr
= realloc(data_buffer
, data_size
);
2252 if (!tmp_data_ptr
) {
2253 ERR("Allocating data buffer");
2256 goto end_rcu_unlock
;
2258 data_buffer
= tmp_data_ptr
;
2259 data_buffer_size
= data_size
;
2261 memset(data_buffer
, 0, data_size
);
2263 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2265 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2266 data_size
, stream_id
, net_seq_num
);
2267 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
2270 /* Orderly shutdown. Not necessary to print an error. */
2271 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2274 goto end_rcu_unlock
;
2277 /* Check if a rotation is needed. */
2278 if (stream
->tracefile_size
> 0 &&
2279 (stream
->tracefile_size_current
+ data_size
) >
2280 stream
->tracefile_size
) {
2281 struct relay_viewer_stream
*vstream
;
2284 new_id
= (stream
->tracefile_count_current
+ 1) %
2285 stream
->tracefile_count
;
2287 * When we wrap-around back to 0, we start overwriting old
2290 if (!stream
->tracefile_overwrite
&& new_id
== 0) {
2291 stream
->tracefile_overwrite
= 1;
2293 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
2294 if (stream
->tracefile_overwrite
) {
2295 stream
->oldest_tracefile_id
=
2296 (stream
->oldest_tracefile_id
+ 1) %
2297 stream
->tracefile_count
;
2299 vstream
= viewer_stream_find_by_id(stream
->stream_handle
);
2302 * The viewer is reading a file about to be
2303 * overwritten. Close the FDs it is
2304 * currently using and let it handle the fault.
2306 if (vstream
->tracefile_count_current
== new_id
) {
2307 pthread_mutex_lock(&vstream
->overwrite_lock
);
2308 vstream
->abort_flag
= 1;
2309 pthread_mutex_unlock(&vstream
->overwrite_lock
);
2310 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2311 stream
->channel_name
, new_id
);
2312 } else if (vstream
->tracefile_count_current
==
2313 stream
->tracefile_count_current
) {
2315 * The reader and writer were in the
2316 * same trace file, inform the viewer
2317 * that no new index will ever be added
2320 vstream
->close_write_flag
= 1;
2323 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
2324 stream
->tracefile_size
, stream
->tracefile_count
,
2325 relayd_uid
, relayd_gid
, stream
->fd
,
2326 &(stream
->tracefile_count_current
), &stream
->fd
);
2327 stream
->total_index_received
= 0;
2328 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
2330 ERR("Rotating stream output file");
2331 goto end_rcu_unlock
;
2333 /* Reset current size because we just perform a stream rotation. */
2334 stream
->tracefile_size_current
= 0;
2339 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2340 * index are NOT supported.
2342 if (session
->minor
>= 4 && !session
->snapshot
) {
2343 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2345 goto end_rcu_unlock
;
2349 /* Write data to stream output fd. */
2350 size_ret
= lttng_write(stream
->fd
, data_buffer
, data_size
);
2351 if (size_ret
< data_size
) {
2352 ERR("Relay error writing data to file");
2354 goto end_rcu_unlock
;
2357 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
2358 ret
, stream
->stream_handle
);
2360 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
2362 goto end_rcu_unlock
;
2364 stream
->tracefile_size_current
+= data_size
+ be32toh(data_hdr
.padding_size
);
2366 stream
->prev_seq
= net_seq_num
;
2368 try_close_stream(session
, stream
);
2377 void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2383 (void) lttng_poll_del(events
, pollfd
);
2385 ret
= close(pollfd
);
2387 ERR("Closing pollfd %d", pollfd
);
2391 static void destroy_connection(struct lttng_ht
*relay_connections_ht
,
2392 struct relay_connection
*conn
)
2394 assert(relay_connections_ht
);
2397 connection_delete(relay_connections_ht
, conn
);
2399 /* For the control socket, we try to destroy the session. */
2400 if (conn
->type
== RELAY_CONTROL
&& conn
->session
) {
2401 destroy_session(conn
->session
, conn
->sessions_ht
);
2404 connection_destroy(conn
);
2408 * This thread does the actual work
2411 void *relay_thread_worker(void *data
)
2413 int ret
, err
= -1, last_seen_data_fd
= -1;
2415 struct relay_connection
*conn
;
2416 struct lttng_poll_event events
;
2417 struct lttng_ht
*relay_connections_ht
;
2418 struct lttng_ht_iter iter
;
2419 struct lttcomm_relayd_hdr recv_hdr
;
2420 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
2421 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
2423 DBG("[thread] Relay worker started");
2425 rcu_register_thread();
2427 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2429 if (testpoint(relayd_thread_worker
)) {
2430 goto error_testpoint
;
2433 health_code_update();
2435 /* table of connections indexed on socket */
2436 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2437 if (!relay_connections_ht
) {
2438 goto relay_connections_ht_error
;
2441 /* Tables of received indexes indexed by index handle and net_seq_num. */
2442 indexes_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64
);
2444 goto indexes_ht_error
;
2447 ret
= create_thread_poll_set(&events
, 2);
2449 goto error_poll_create
;
2452 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2459 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2461 health_code_update();
2463 /* Infinite blocking call, waiting for transmission */
2464 DBG3("Relayd worker thread polling...");
2465 health_poll_entry();
2466 ret
= lttng_poll_wait(&events
, -1);
2470 * Restart interrupted system call.
2472 if (errno
== EINTR
) {
2481 * Process control. The control connection is prioritised so we don't
2482 * starve it with high throughout put tracing data on the data
2485 for (i
= 0; i
< nb_fd
; i
++) {
2486 /* Fetch once the poll data */
2487 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2488 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2490 health_code_update();
2492 /* Thread quit pipe has been closed. Killing thread. */
2493 ret
= check_thread_quit_pipe(pollfd
, revents
);
2499 /* Inspect the relay conn pipe for new connection */
2500 if (pollfd
== relay_conn_pipe
[0]) {
2501 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2502 ERR("Relay connection pipe error");
2504 } else if (revents
& LPOLLIN
) {
2505 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
2509 conn
->sessions_ht
= sessions_ht
;
2510 connection_init(conn
);
2511 lttng_poll_add(&events
, conn
->sock
->fd
,
2512 LPOLLIN
| LPOLLRDHUP
);
2514 lttng_ht_add_unique_ulong(relay_connections_ht
,
2517 DBG("Connection socket %d added", conn
->sock
->fd
);
2521 conn
= connection_find_by_sock(relay_connections_ht
, pollfd
);
2522 /* If not found, there is a synchronization issue. */
2525 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2526 cleanup_connection_pollfd(&events
, pollfd
);
2527 destroy_connection(relay_connections_ht
, conn
);
2528 if (last_seen_data_fd
== pollfd
) {
2529 last_seen_data_fd
= last_notdel_data_fd
;
2531 } else if (revents
& LPOLLIN
) {
2532 if (conn
->type
== RELAY_CONTROL
) {
2533 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &recv_hdr
,
2534 sizeof(recv_hdr
), 0);
2536 /* Connection closed */
2537 cleanup_connection_pollfd(&events
, pollfd
);
2538 destroy_connection(relay_connections_ht
, conn
);
2539 DBG("Control connection closed with %d", pollfd
);
2541 ret
= relay_process_control(&recv_hdr
, conn
);
2543 /* Clear the session on error. */
2544 cleanup_connection_pollfd(&events
, pollfd
);
2545 destroy_connection(relay_connections_ht
, conn
);
2546 DBG("Connection closed with %d", pollfd
);
2552 * Flag the last seen data fd not deleted. It will be
2553 * used as the last seen fd if any fd gets deleted in
2556 last_notdel_data_fd
= pollfd
;
2559 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
2566 * The last loop handled a control request, go back to poll to make
2567 * sure we prioritise the control socket.
2573 if (last_seen_data_fd
>= 0) {
2574 for (i
= 0; i
< nb_fd
; i
++) {
2575 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2577 health_code_update();
2579 if (last_seen_data_fd
== pollfd
) {
2586 /* Process data connection. */
2587 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2588 /* Fetch the poll data. */
2589 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2590 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2592 health_code_update();
2594 /* Skip the command pipe. It's handled in the first loop. */
2595 if (pollfd
== relay_conn_pipe
[0]) {
2601 conn
= connection_find_by_sock(relay_connections_ht
, pollfd
);
2603 /* Skip it. Might be removed before. */
2608 if (revents
& LPOLLIN
) {
2609 if (conn
->type
!= RELAY_DATA
) {
2613 ret
= relay_process_data(conn
);
2614 /* Connection closed */
2616 cleanup_connection_pollfd(&events
, pollfd
);
2617 destroy_connection(relay_connections_ht
, conn
);
2618 DBG("Data connection closed with %d", pollfd
);
2620 * Every goto restart call sets the last seen fd where
2621 * here we don't really care since we gracefully
2622 * continue the loop after the connection is deleted.
2625 /* Keep last seen port. */
2626 last_seen_data_fd
= pollfd
;
2634 last_seen_data_fd
= -1;
2637 /* Normal exit, no error */
2642 lttng_poll_clean(&events
);
2644 /* Cleanup reamaining connection object. */
2646 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, conn
,
2648 health_code_update();
2649 destroy_connection(relay_connections_ht
, conn
);
2653 lttng_ht_destroy(indexes_ht
);
2655 lttng_ht_destroy(relay_connections_ht
);
2656 relay_connections_ht_error
:
2657 /* Close relay conn pipes */
2658 utils_close_pipe(relay_conn_pipe
);
2660 DBG("Thread exited with error");
2662 DBG("Worker thread cleanup complete");
2667 ERR("Health error occurred in %s", __func__
);
2669 health_unregister(health_relayd
);
2670 rcu_unregister_thread();
2676 * Create the relay command pipe to wake thread_manage_apps.
2677 * Closed in cleanup().
2679 static int create_relay_conn_pipe(void)
2683 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
2691 int main(int argc
, char **argv
)
2695 struct relay_local_data
*relay_ctx
;
2697 /* Parse arguments */
2699 if ((ret
= set_options(argc
, argv
)) < 0) {
2703 if ((ret
= set_signal_handler()) < 0) {
2707 /* Try to create directory if -o, --output is specified. */
2708 if (opt_output_path
) {
2709 if (*opt_output_path
!= '/') {
2710 ERR("Please specify an absolute path for -o, --output PATH");
2714 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2716 ERR("Unable to create %s", opt_output_path
);
2722 if (opt_daemon
|| opt_background
) {
2725 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
2732 * We are in the child. Make sure all other file
2733 * descriptors are closed, in case we are called with
2734 * more opened file descriptors than the standard ones.
2736 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
2741 /* Create thread quit pipe */
2742 if ((ret
= init_thread_quit_pipe()) < 0) {
2746 /* We need those values for the file/dir creation. */
2747 relayd_uid
= getuid();
2748 relayd_gid
= getgid();
2750 /* Check if daemon is UID = 0 */
2751 if (relayd_uid
== 0) {
2752 if (control_uri
->port
< 1024 || data_uri
->port
< 1024 || live_uri
->port
< 1024) {
2753 ERR("Need to be root to use ports < 1024");
2759 /* Setup the thread apps communication pipe. */
2760 if ((ret
= create_relay_conn_pipe()) < 0) {
2764 /* Init relay command queue. */
2765 cds_wfq_init(&relay_conn_queue
.queue
);
2767 /* Set up max poll set size */
2768 lttng_poll_set_max_size();
2770 /* Initialize communication library */
2772 lttcomm_inet_init();
2774 relay_ctx
= zmalloc(sizeof(struct relay_local_data
));
2776 PERROR("relay_ctx");
2780 /* tables of sessions indexed by session ID */
2781 relay_ctx
->sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2782 if (!relay_ctx
->sessions_ht
) {
2783 goto exit_relay_ctx_sessions
;
2786 /* tables of streams indexed by stream ID */
2787 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2788 if (!relay_streams_ht
) {
2789 goto exit_relay_ctx_streams
;
2792 /* tables of streams indexed by stream ID */
2793 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2794 if (!viewer_streams_ht
) {
2795 goto exit_relay_ctx_viewer_streams
;
2798 /* Initialize thread health monitoring */
2799 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
2800 if (!health_relayd
) {
2801 PERROR("health_app_create error");
2802 goto exit_health_app_create
;
2805 ret
= utils_create_pipe(health_quit_pipe
);
2807 goto error_health_pipe
;
2810 /* Create thread to manage the client socket */
2811 ret
= pthread_create(&health_thread
, NULL
,
2812 thread_manage_health
, (void *) NULL
);
2814 PERROR("pthread_create health");
2818 /* Setup the dispatcher thread */
2819 ret
= pthread_create(&dispatcher_thread
, NULL
,
2820 relay_thread_dispatcher
, (void *) NULL
);
2822 PERROR("pthread_create dispatcher");
2823 goto exit_dispatcher
;
2826 /* Setup the worker thread */
2827 ret
= pthread_create(&worker_thread
, NULL
,
2828 relay_thread_worker
, (void *) relay_ctx
);
2830 PERROR("pthread_create worker");
2834 /* Setup the listener thread */
2835 ret
= pthread_create(&listener_thread
, NULL
,
2836 relay_thread_listener
, (void *) NULL
);
2838 PERROR("pthread_create listener");
2842 ret
= live_start_threads(live_uri
, relay_ctx
);
2844 ERR("Starting live viewer threads");
2849 ret
= pthread_join(listener_thread
, &status
);
2851 PERROR("pthread_join");
2852 goto error
; /* join error, exit without cleanup */
2856 ret
= pthread_join(worker_thread
, &status
);
2858 PERROR("pthread_join");
2859 goto error
; /* join error, exit without cleanup */
2863 ret
= pthread_join(dispatcher_thread
, &status
);
2865 PERROR("pthread_join");
2866 goto error
; /* join error, exit without cleanup */
2870 ret
= pthread_join(health_thread
, &status
);
2872 PERROR("pthread_join health thread");
2873 goto error
; /* join error, exit without cleanup */
2877 * Stop live threads only after joining other threads.
2879 live_stop_threads();
2882 utils_close_pipe(health_quit_pipe
);
2885 health_app_destroy(health_relayd
);
2887 exit_health_app_create
:
2888 lttng_ht_destroy(viewer_streams_ht
);
2890 exit_relay_ctx_viewer_streams
:
2891 lttng_ht_destroy(relay_streams_ht
);
2893 exit_relay_ctx_streams
:
2894 lttng_ht_destroy(relay_ctx
->sessions_ht
);
2896 exit_relay_ctx_sessions
: