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 if (tracing_group_name
== NULL
) {
241 tracing_group_name_override
= 1;
247 ret
= asprintf(&opt_output_path
, "%s", arg
);
250 PERROR("asprintf opt_output_path");
255 /* Verbose level can increase using multiple -v */
257 lttng_opt_verbose
= config_parse_value(arg
);
259 /* Only 3 level of verbosity (-vvv). */
260 if (lttng_opt_verbose
< 3) {
261 lttng_opt_verbose
+= 1;
266 /* Unknown option or other error.
267 * Error is printed by getopt, just return */
280 * config_entry_handler_cb used to handle options read from a config file.
281 * See config_entry_handler_cb comment in common/config/config.h for the
282 * return value conventions.
285 int config_entry_handler(const struct config_entry
*entry
, void *unused
)
289 if (!entry
|| !entry
->name
|| !entry
->value
) {
294 /* Check if the option is to be ignored */
295 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
296 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
301 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
302 /* Ignore if entry name is not fully matched. */
303 if (strcmp(entry
->name
, long_options
[i
].name
)) {
308 * If the option takes no argument on the command line, we have to
309 * check if the value is "true". We support non-zero numeric values,
312 if (!long_options
[i
].has_arg
) {
313 ret
= config_parse_value(entry
->value
);
316 WARN("Invalid configuration value \"%s\" for option %s",
317 entry
->value
, entry
->name
);
319 /* False, skip boolean config option. */
324 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
328 WARN("Unrecognized option \"%s\" in daemon configuration file.",
336 int set_options(int argc
, char **argv
)
338 int c
, ret
= 0, option_index
= 0;
339 int orig_optopt
= optopt
, orig_optind
= optind
;
340 char *default_address
, *optstring
;
341 const char *config_path
= NULL
;
343 optstring
= utils_generate_optstring(long_options
,
344 sizeof(long_options
) / sizeof(struct option
));
350 /* Check for the --config option */
352 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
353 &option_index
)) != -1) {
357 } else if (c
!= 'f') {
361 config_path
= utils_expand_path(optarg
);
363 ERR("Failed to resolve path: %s", optarg
);
367 ret
= config_get_section_entries(config_path
, config_section_name
,
368 config_entry_handler
, NULL
);
371 ERR("Invalid configuration option at line %i", ret
);
377 /* Reset getopt's global state */
378 optopt
= orig_optopt
;
379 optind
= orig_optind
;
381 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
386 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
392 /* assign default values */
393 if (control_uri
== NULL
) {
394 ret
= asprintf(&default_address
,
395 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
396 DEFAULT_NETWORK_CONTROL_PORT
);
398 PERROR("asprintf default data address");
402 ret
= uri_parse(default_address
, &control_uri
);
403 free(default_address
);
405 ERR("Invalid control URI specified");
409 if (data_uri
== NULL
) {
410 ret
= asprintf(&default_address
,
411 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
412 DEFAULT_NETWORK_DATA_PORT
);
414 PERROR("asprintf default data address");
418 ret
= uri_parse(default_address
, &data_uri
);
419 free(default_address
);
421 ERR("Invalid data URI specified");
425 if (live_uri
== NULL
) {
426 ret
= asprintf(&default_address
,
427 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
428 DEFAULT_NETWORK_VIEWER_PORT
);
430 PERROR("asprintf default viewer control address");
434 ret
= uri_parse(default_address
, &live_uri
);
435 free(default_address
);
437 ERR("Invalid viewer control URI specified");
455 /* free the dynamically allocated opt_output_path */
456 free(opt_output_path
);
458 /* Close thread quit pipes */
459 utils_close_pipe(thread_quit_pipe
);
461 uri_free(control_uri
);
463 /* Live URI is freed in the live thread. */
465 if (tracing_group_name_override
) {
466 free((void *) tracing_group_name
);
471 * Write to writable pipe used to notify a thread.
474 int notify_thread_pipe(int wpipe
)
478 ret
= lttng_write(wpipe
, "!", 1);
480 PERROR("write poll pipe");
486 static void notify_health_quit_pipe(int *pipe
)
490 ret
= lttng_write(pipe
[1], "4", 1);
492 PERROR("write relay health quit");
497 * Stop all threads by closing the thread quit pipe.
500 void stop_threads(void)
504 /* Stopping all threads */
505 DBG("Terminating all threads");
506 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
508 ERR("write error on thread quit pipe");
511 notify_health_quit_pipe(health_quit_pipe
);
513 /* Dispatch thread */
514 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
515 futex_nto1_wake(&relay_conn_queue
.futex
);
519 * Signal handler for the daemon
521 * Simply stop all worker threads, leaving main() return gracefully after
522 * joining all threads and calling cleanup().
525 void sighandler(int sig
)
529 DBG("SIGPIPE caught");
532 DBG("SIGINT caught");
536 DBG("SIGTERM caught");
540 CMM_STORE_SHARED(recv_child_signal
, 1);
548 * Setup signal handler for :
549 * SIGINT, SIGTERM, SIGPIPE
552 int set_signal_handler(void)
558 if ((ret
= sigemptyset(&sigset
)) < 0) {
559 PERROR("sigemptyset");
563 sa
.sa_handler
= sighandler
;
566 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
571 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
576 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
581 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
586 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
591 void lttng_relay_notify_ready(void)
593 /* Notify the parent of the fork() process that we are ready. */
594 if (opt_daemon
|| opt_background
) {
595 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
596 kill(child_ppid
, SIGUSR1
);
602 * Init thread quit pipe.
604 * Return -1 on error or 0 if all pipes are created.
607 int init_thread_quit_pipe(void)
611 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
617 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
620 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
624 if (events
== NULL
|| size
== 0) {
629 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
635 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
647 * Check if the thread quit pipe was triggered.
649 * Return 1 if it was triggered else 0;
652 int check_thread_quit_pipe(int fd
, uint32_t events
)
654 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
662 * Create and init socket from uri.
665 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
668 struct lttcomm_sock
*sock
= NULL
;
670 sock
= lttcomm_alloc_sock_from_uri(uri
);
672 ERR("Allocating socket");
676 ret
= lttcomm_create_sock(sock
);
680 DBG("Listening on sock %d", sock
->fd
);
682 ret
= sock
->ops
->bind(sock
);
687 ret
= sock
->ops
->listen(sock
, -1);
697 lttcomm_destroy_sock(sock
);
703 * Return nonzero if stream needs to be closed.
706 int close_stream_check(struct relay_stream
*stream
)
708 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
710 * We are about to close the stream so set the data pending flag to 1
711 * which will make the end data pending command skip the stream which
712 * is now closed and ready. Note that after proceeding to a file close,
713 * the written file is ready for reading.
715 stream
->data_pending_check_done
= 1;
721 static void try_close_stream(struct relay_session
*session
,
722 struct relay_stream
*stream
)
725 struct ctf_trace
*ctf_trace
;
730 if (!close_stream_check(stream
)) {
731 /* Can't close it, not ready for that. */
735 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
,
739 pthread_mutex_lock(&session
->viewer_ready_lock
);
740 ctf_trace
->invalid_flag
= 1;
741 pthread_mutex_unlock(&session
->viewer_ready_lock
);
743 ret
= stream_close(session
, stream
);
744 if (ret
|| session
->snapshot
) {
745 /* Already close thus the ctf trace is being or has been destroyed. */
749 ctf_trace_try_destroy(session
, ctf_trace
);
756 * This thread manages the listening for new connections on the network
759 void *relay_thread_listener(void *data
)
761 int i
, ret
, pollfd
, err
= -1;
762 uint32_t revents
, nb_fd
;
763 struct lttng_poll_event events
;
764 struct lttcomm_sock
*control_sock
, *data_sock
;
766 DBG("[thread] Relay listener started");
768 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
770 health_code_update();
772 control_sock
= relay_init_sock(control_uri
);
774 goto error_sock_control
;
777 data_sock
= relay_init_sock(data_uri
);
779 goto error_sock_relay
;
783 * Pass 3 as size here for the thread quit pipe, control and data socket.
785 ret
= create_thread_poll_set(&events
, 3);
787 goto error_create_poll
;
790 /* Add the control socket */
791 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
796 /* Add the data socket */
797 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
802 lttng_relay_notify_ready();
804 if (testpoint(relayd_thread_listener
)) {
805 goto error_testpoint
;
809 health_code_update();
811 DBG("Listener accepting connections");
815 ret
= lttng_poll_wait(&events
, -1);
819 * Restart interrupted system call.
821 if (errno
== EINTR
) {
829 DBG("Relay new connection received");
830 for (i
= 0; i
< nb_fd
; i
++) {
831 health_code_update();
833 /* Fetch once the poll data */
834 revents
= LTTNG_POLL_GETEV(&events
, i
);
835 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
837 /* Thread quit pipe has been closed. Killing thread. */
838 ret
= check_thread_quit_pipe(pollfd
, revents
);
844 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
845 ERR("socket poll error");
847 } else if (revents
& LPOLLIN
) {
849 * Get allocated in this thread, enqueued to a global queue,
850 * dequeued and freed in the worker thread.
853 struct relay_connection
*new_conn
;
854 struct lttcomm_sock
*newsock
;
856 new_conn
= connection_create();
861 if (pollfd
== data_sock
->fd
) {
862 new_conn
->type
= RELAY_DATA
;
863 newsock
= data_sock
->ops
->accept(data_sock
);
864 DBG("Relay data connection accepted, socket %d",
867 assert(pollfd
== control_sock
->fd
);
868 new_conn
->type
= RELAY_CONTROL
;
869 newsock
= control_sock
->ops
->accept(control_sock
);
870 DBG("Relay control connection accepted, socket %d",
874 PERROR("accepting sock");
875 connection_free(new_conn
);
879 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
882 PERROR("setsockopt inet");
883 lttcomm_destroy_sock(newsock
);
884 connection_free(new_conn
);
887 new_conn
->sock
= newsock
;
889 /* Enqueue request for the dispatcher thread. */
890 cds_wfq_enqueue(&relay_conn_queue
.queue
, &new_conn
->qnode
);
893 * Wake the dispatch queue futex. Implicit memory barrier with
894 * the exchange in cds_wfq_enqueue.
896 futex_nto1_wake(&relay_conn_queue
.futex
);
905 lttng_poll_clean(&events
);
907 if (data_sock
->fd
>= 0) {
908 ret
= data_sock
->ops
->close(data_sock
);
913 lttcomm_destroy_sock(data_sock
);
915 if (control_sock
->fd
>= 0) {
916 ret
= control_sock
->ops
->close(control_sock
);
921 lttcomm_destroy_sock(control_sock
);
925 ERR("Health error occurred in %s", __func__
);
927 health_unregister(health_relayd
);
928 DBG("Relay listener thread cleanup complete");
934 * This thread manages the dispatching of the requests to worker threads
937 void *relay_thread_dispatcher(void *data
)
941 struct cds_wfq_node
*node
;
942 struct relay_connection
*new_conn
= NULL
;
944 DBG("[thread] Relay dispatcher started");
946 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
948 if (testpoint(relayd_thread_dispatcher
)) {
949 goto error_testpoint
;
952 health_code_update();
954 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
955 health_code_update();
957 /* Atomically prepare the queue futex */
958 futex_nto1_prepare(&relay_conn_queue
.futex
);
961 health_code_update();
963 /* Dequeue commands */
964 node
= cds_wfq_dequeue_blocking(&relay_conn_queue
.queue
);
966 DBG("Woken up but nothing in the relay command queue");
967 /* Continue thread execution */
970 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
972 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
975 * Inform worker thread of the new request. This call is blocking
976 * so we can be assured that the data will be read at some point in
977 * time or wait to the end of the world :)
979 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
981 PERROR("write connection pipe");
982 connection_destroy(new_conn
);
985 } while (node
!= NULL
);
987 /* Futex wait on queue. Blocking call on futex() */
989 futex_nto1_wait(&relay_conn_queue
.futex
);
993 /* Normal exit, no error */
1000 ERR("Health error occurred in %s", __func__
);
1002 health_unregister(health_relayd
);
1003 DBG("Dispatch thread dying");
1008 static void try_close_streams(struct relay_session
*session
)
1010 struct ctf_trace
*ctf_trace
;
1011 struct lttng_ht_iter iter
;
1015 pthread_mutex_lock(&session
->viewer_ready_lock
);
1017 cds_lfht_for_each_entry(session
->ctf_traces_ht
->ht
, &iter
.iter
, ctf_trace
,
1019 struct relay_stream
*stream
;
1021 /* Close streams. */
1022 cds_list_for_each_entry(stream
, &ctf_trace
->stream_list
, trace_list
) {
1023 stream_close(session
, stream
);
1026 ctf_trace
->invalid_flag
= 1;
1027 ctf_trace_try_destroy(session
, ctf_trace
);
1030 pthread_mutex_unlock(&session
->viewer_ready_lock
);
1034 * Try to destroy a session within a connection.
1036 static void destroy_session(struct relay_session
*session
,
1037 struct lttng_ht
*sessions_ht
)
1040 assert(sessions_ht
);
1042 /* Indicate that this session can be destroyed from now on. */
1043 session
->close_flag
= 1;
1045 try_close_streams(session
);
1048 * This will try to delete and destroy the session if no viewer is attached
1049 * to it meaning the refcount is down to zero.
1051 session_try_destroy(sessions_ht
, session
);
1055 * Copy index data from the control port to a given index object.
1057 static void copy_index_control_data(struct relay_index
*index
,
1058 struct lttcomm_relayd_index
*data
)
1064 * The index on disk is encoded in big endian, so we don't need to convert
1065 * the data received on the network. The data_offset value is NEVER
1066 * modified here and is updated by the data thread.
1068 index
->index_data
.packet_size
= data
->packet_size
;
1069 index
->index_data
.content_size
= data
->content_size
;
1070 index
->index_data
.timestamp_begin
= data
->timestamp_begin
;
1071 index
->index_data
.timestamp_end
= data
->timestamp_end
;
1072 index
->index_data
.events_discarded
= data
->events_discarded
;
1073 index
->index_data
.stream_id
= data
->stream_id
;
1077 * Handle the RELAYD_CREATE_SESSION command.
1079 * On success, send back the session id or else return a negative value.
1082 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
1083 struct relay_connection
*conn
)
1085 int ret
= 0, send_ret
;
1086 struct relay_session
*session
;
1087 struct lttcomm_relayd_status_session reply
;
1092 memset(&reply
, 0, sizeof(reply
));
1094 session
= session_create();
1099 session
->minor
= conn
->minor
;
1100 session
->major
= conn
->major
;
1101 conn
->session_id
= session
->id
;
1102 conn
->session
= session
;
1104 reply
.session_id
= htobe64(session
->id
);
1106 switch (conn
->minor
) {
1111 case 4: /* LTTng sessiond 2.4 */
1113 ret
= cmd_create_session_2_4(conn
, session
);
1116 lttng_ht_add_unique_u64(conn
->sessions_ht
, &session
->session_n
);
1117 DBG("Created session %" PRIu64
, session
->id
);
1121 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1123 reply
.ret_code
= htobe32(LTTNG_OK
);
1126 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1128 ERR("Relayd sending session id");
1136 * When we have received all the streams and the metadata for a channel,
1137 * we make them visible to the viewer threads.
1140 void set_viewer_ready_flag(struct relay_connection
*conn
)
1142 struct relay_stream
*stream
, *tmp_stream
;
1144 pthread_mutex_lock(&conn
->session
->viewer_ready_lock
);
1145 cds_list_for_each_entry_safe(stream
, tmp_stream
, &conn
->recv_head
,
1147 stream
->viewer_ready
= 1;
1148 cds_list_del(&stream
->recv_list
);
1150 pthread_mutex_unlock(&conn
->session
->viewer_ready_lock
);
1155 * Add a recv handle node to the connection recv list with the given stream
1156 * handle. A new node is allocated thus must be freed when the node is deleted
1159 static void queue_stream(struct relay_stream
*stream
,
1160 struct relay_connection
*conn
)
1165 cds_list_add(&stream
->recv_list
, &conn
->recv_head
);
1169 * relay_add_stream: allocate a new stream for a session
1172 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1173 struct relay_connection
*conn
)
1176 struct relay_session
*session
= conn
->session
;
1177 struct relay_stream
*stream
= NULL
;
1178 struct lttcomm_relayd_status_stream reply
;
1179 struct ctf_trace
*trace
;
1181 if (!session
|| conn
->version_check_done
== 0) {
1182 ERR("Trying to add a stream before version check");
1184 goto end_no_session
;
1187 stream
= zmalloc(sizeof(struct relay_stream
));
1188 if (stream
== NULL
) {
1189 PERROR("relay stream zmalloc");
1191 goto end_no_session
;
1194 switch (conn
->minor
) {
1195 case 1: /* LTTng sessiond 2.1 */
1196 ret
= cmd_recv_stream_2_1(conn
, stream
);
1198 case 2: /* LTTng sessiond 2.2 */
1200 ret
= cmd_recv_stream_2_2(conn
, stream
);
1204 goto err_free_stream
;
1208 stream
->stream_handle
= ++last_relay_stream_id
;
1209 stream
->prev_seq
= -1ULL;
1210 stream
->session_id
= session
->id
;
1211 stream
->index_fd
= -1;
1212 stream
->read_index_fd
= -1;
1213 stream
->ctf_stream_id
= -1ULL;
1214 lttng_ht_node_init_u64(&stream
->node
, stream
->stream_handle
);
1215 pthread_mutex_init(&stream
->lock
, NULL
);
1217 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
);
1219 ERR("relay creating output directory");
1224 * No need to use run_as API here because whatever we receives, the relayd
1225 * uses its own credentials for the stream files.
1227 ret
= utils_create_stream_file(stream
->path_name
, stream
->channel_name
,
1228 stream
->tracefile_size
, 0, relayd_uid
, relayd_gid
, NULL
);
1230 ERR("Create output file");
1234 if (stream
->tracefile_size
) {
1235 DBG("Tracefile %s/%s_0 created", stream
->path_name
, stream
->channel_name
);
1237 DBG("Tracefile %s/%s created", stream
->path_name
, stream
->channel_name
);
1240 trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
, stream
->path_name
);
1242 trace
= ctf_trace_create(stream
->path_name
);
1247 ctf_trace_add(session
->ctf_traces_ht
, trace
);
1249 ctf_trace_get_ref(trace
);
1251 if (!strncmp(stream
->channel_name
, DEFAULT_METADATA_NAME
, NAME_MAX
)) {
1252 stream
->metadata_flag
= 1;
1253 /* Assign quick reference to the metadata stream in the trace. */
1254 trace
->metadata_stream
= stream
;
1258 * Add the stream in the recv list of the connection. Once the end stream
1259 * message is received, this list is emptied and streams are set with the
1260 * viewer ready flag.
1262 queue_stream(stream
, conn
);
1265 * Both in the ctf_trace object and the global stream ht since the data
1266 * side of the relayd does not have the concept of session.
1268 lttng_ht_add_unique_u64(relay_streams_ht
, &stream
->node
);
1269 cds_list_add_tail(&stream
->trace_list
, &trace
->stream_list
);
1271 session
->stream_count
++;
1273 DBG("Relay new stream added %s with ID %" PRIu64
, stream
->channel_name
,
1274 stream
->stream_handle
);
1277 memset(&reply
, 0, sizeof(reply
));
1278 reply
.handle
= htobe64(stream
->stream_handle
);
1279 /* send the session id to the client or a negative return code on error */
1281 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1282 /* stream was not properly added to the ht, so free it */
1285 reply
.ret_code
= htobe32(LTTNG_OK
);
1288 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1289 sizeof(struct lttcomm_relayd_status_stream
), 0);
1291 ERR("Relay sending stream id");
1300 free(stream
->path_name
);
1301 free(stream
->channel_name
);
1307 * relay_close_stream: close a specific stream
1310 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1311 struct relay_connection
*conn
)
1314 struct relay_session
*session
= conn
->session
;
1315 struct lttcomm_relayd_close_stream stream_info
;
1316 struct lttcomm_relayd_generic_reply reply
;
1317 struct relay_stream
*stream
;
1319 DBG("Close stream received");
1321 if (!session
|| conn
->version_check_done
== 0) {
1322 ERR("Trying to close a stream before version check");
1324 goto end_no_session
;
1327 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
1328 sizeof(struct lttcomm_relayd_close_stream
), 0);
1329 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1331 /* Orderly shutdown. Not necessary to print an error. */
1332 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1334 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1337 goto end_no_session
;
1341 stream
= stream_find_by_id(relay_streams_ht
,
1342 be64toh(stream_info
.stream_id
));
1348 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1349 stream
->close_flag
= 1;
1350 session
->stream_count
--;
1351 assert(session
->stream_count
>= 0);
1353 /* Check if we can close it or else the data will do it. */
1354 try_close_stream(session
, stream
);
1359 memset(&reply
, 0, sizeof(reply
));
1361 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1363 reply
.ret_code
= htobe32(LTTNG_OK
);
1365 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1366 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1368 ERR("Relay sending stream id");
1377 * relay_unknown_command: send -1 if received unknown command
1380 void relay_unknown_command(struct relay_connection
*conn
)
1382 struct lttcomm_relayd_generic_reply reply
;
1385 memset(&reply
, 0, sizeof(reply
));
1386 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1387 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1388 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1390 ERR("Relay sending unknown command");
1395 * relay_start: send an acknowledgment to the client to tell if we are
1396 * ready to receive data. We are ready if a session is established.
1399 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1400 struct relay_connection
*conn
)
1402 int ret
= htobe32(LTTNG_OK
);
1403 struct lttcomm_relayd_generic_reply reply
;
1404 struct relay_session
*session
= conn
->session
;
1407 DBG("Trying to start the streaming without a session established");
1408 ret
= htobe32(LTTNG_ERR_UNK
);
1411 memset(&reply
, 0, sizeof(reply
));
1412 reply
.ret_code
= ret
;
1413 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1414 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1416 ERR("Relay sending start ack");
1423 * Append padding to the file pointed by the file descriptor fd.
1425 static int write_padding_to_file(int fd
, uint32_t size
)
1434 zeros
= zmalloc(size
);
1435 if (zeros
== NULL
) {
1436 PERROR("zmalloc zeros for padding");
1441 ret
= lttng_write(fd
, zeros
, size
);
1443 PERROR("write padding to file");
1453 * relay_recv_metadata: receive the metada for the session.
1456 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1457 struct relay_connection
*conn
)
1459 int ret
= htobe32(LTTNG_OK
);
1461 struct relay_session
*session
= conn
->session
;
1462 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1463 struct relay_stream
*metadata_stream
;
1464 uint64_t data_size
, payload_size
;
1465 struct ctf_trace
*ctf_trace
;
1468 ERR("Metadata sent before version check");
1473 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1474 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1475 ERR("Incorrect data size");
1479 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1481 if (data_buffer_size
< data_size
) {
1482 /* In case the realloc fails, we can free the memory */
1485 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1486 if (!tmp_data_ptr
) {
1487 ERR("Allocating data buffer");
1492 data_buffer
= tmp_data_ptr
;
1493 data_buffer_size
= data_size
;
1495 memset(data_buffer
, 0, data_size
);
1496 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1497 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
1498 if (ret
< 0 || ret
!= data_size
) {
1500 /* Orderly shutdown. Not necessary to print an error. */
1501 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1503 ERR("Relay didn't receive the whole metadata");
1508 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1511 metadata_stream
= stream_find_by_id(relay_streams_ht
,
1512 be64toh(metadata_struct
->stream_id
));
1513 if (!metadata_stream
) {
1518 size_ret
= lttng_write(metadata_stream
->fd
, metadata_struct
->payload
,
1520 if (size_ret
< payload_size
) {
1521 ERR("Relay error writing metadata on file");
1526 ret
= write_padding_to_file(metadata_stream
->fd
,
1527 be32toh(metadata_struct
->padding_size
));
1532 ctf_trace
= ctf_trace_find_by_path(session
->ctf_traces_ht
,
1533 metadata_stream
->path_name
);
1535 ctf_trace
->metadata_received
+=
1536 payload_size
+ be32toh(metadata_struct
->padding_size
);
1538 DBG2("Relay metadata written");
1547 * relay_send_version: send relayd version number
1550 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1551 struct relay_connection
*conn
)
1554 struct lttcomm_relayd_version reply
, msg
;
1558 conn
->version_check_done
= 1;
1560 /* Get version from the other side. */
1561 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1562 if (ret
< 0 || ret
!= sizeof(msg
)) {
1564 /* Orderly shutdown. Not necessary to print an error. */
1565 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1567 ERR("Relay failed to receive the version values.");
1573 memset(&reply
, 0, sizeof(reply
));
1574 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1575 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1577 /* Major versions must be the same */
1578 if (reply
.major
!= be32toh(msg
.major
)) {
1579 DBG("Incompatible major versions (%u vs %u), deleting session",
1580 reply
.major
, be32toh(msg
.major
));
1581 destroy_session(conn
->session
, conn
->sessions_ht
);
1586 conn
->major
= reply
.major
;
1587 /* We adapt to the lowest compatible version */
1588 if (reply
.minor
<= be32toh(msg
.minor
)) {
1589 conn
->minor
= reply
.minor
;
1591 conn
->minor
= be32toh(msg
.minor
);
1594 reply
.major
= htobe32(reply
.major
);
1595 reply
.minor
= htobe32(reply
.minor
);
1596 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1597 sizeof(struct lttcomm_relayd_version
), 0);
1599 ERR("Relay sending version");
1602 DBG("Version check done using protocol %u.%u", conn
->major
,
1610 * Check for data pending for a given stream id from the session daemon.
1613 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1614 struct relay_connection
*conn
)
1616 struct relay_session
*session
= conn
->session
;
1617 struct lttcomm_relayd_data_pending msg
;
1618 struct lttcomm_relayd_generic_reply reply
;
1619 struct relay_stream
*stream
;
1621 uint64_t last_net_seq_num
, stream_id
;
1623 DBG("Data pending command received");
1625 if (!session
|| conn
->version_check_done
== 0) {
1626 ERR("Trying to check for data before version check");
1628 goto end_no_session
;
1631 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1632 if (ret
< sizeof(msg
)) {
1634 /* Orderly shutdown. Not necessary to print an error. */
1635 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1637 ERR("Relay didn't receive valid data_pending struct size : %d",
1641 goto end_no_session
;
1644 stream_id
= be64toh(msg
.stream_id
);
1645 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1648 stream
= stream_find_by_id(relay_streams_ht
, stream_id
);
1649 if (stream
== NULL
) {
1654 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1655 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1658 /* Avoid wrapping issue */
1659 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1660 /* Data has in fact been written and is NOT pending */
1663 /* Data still being streamed thus pending */
1667 /* Pending check is now done. */
1668 stream
->data_pending_check_done
= 1;
1673 memset(&reply
, 0, sizeof(reply
));
1674 reply
.ret_code
= htobe32(ret
);
1675 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1677 ERR("Relay data pending ret code failed");
1685 * Wait for the control socket to reach a quiescent state.
1687 * Note that for now, when receiving this command from the session daemon, this
1688 * means that every subsequent commands or data received on the control socket
1689 * has been handled. So, this is why we simply return OK here.
1692 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1693 struct relay_connection
*conn
)
1697 struct relay_stream
*stream
;
1698 struct lttng_ht_iter iter
;
1699 struct lttcomm_relayd_quiescent_control msg
;
1700 struct lttcomm_relayd_generic_reply reply
;
1702 DBG("Checking quiescent state on control socket");
1704 if (!conn
->session
|| conn
->version_check_done
== 0) {
1705 ERR("Trying to check for data before version check");
1707 goto end_no_session
;
1710 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1711 if (ret
< sizeof(msg
)) {
1713 /* Orderly shutdown. Not necessary to print an error. */
1714 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1716 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1720 goto end_no_session
;
1723 stream_id
= be64toh(msg
.stream_id
);
1726 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1728 if (stream
->stream_handle
== stream_id
) {
1729 stream
->data_pending_check_done
= 1;
1730 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1737 memset(&reply
, 0, sizeof(reply
));
1738 reply
.ret_code
= htobe32(LTTNG_OK
);
1739 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1741 ERR("Relay data quiescent control ret code failed");
1749 * Initialize a data pending command. This means that a client is about to ask
1750 * for data pending for each stream he/she holds. Simply iterate over all
1751 * streams of a session and set the data_pending_check_done flag.
1753 * This command returns to the client a LTTNG_OK code.
1756 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1757 struct relay_connection
*conn
)
1760 struct lttng_ht_iter iter
;
1761 struct lttcomm_relayd_begin_data_pending msg
;
1762 struct lttcomm_relayd_generic_reply reply
;
1763 struct relay_stream
*stream
;
1764 uint64_t session_id
;
1769 DBG("Init streams for data pending");
1771 if (!conn
->session
|| conn
->version_check_done
== 0) {
1772 ERR("Trying to check for data before version check");
1774 goto end_no_session
;
1777 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1778 if (ret
< sizeof(msg
)) {
1780 /* Orderly shutdown. Not necessary to print an error. */
1781 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1783 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1787 goto end_no_session
;
1790 session_id
= be64toh(msg
.session_id
);
1793 * Iterate over all streams to set the begin data pending flag. For now, the
1794 * streams are indexed by stream handle so we have to iterate over all
1795 * streams to find the one associated with the right session_id.
1798 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1800 if (stream
->session_id
== session_id
) {
1801 stream
->data_pending_check_done
= 0;
1802 DBG("Set begin data pending flag to stream %" PRIu64
,
1803 stream
->stream_handle
);
1808 memset(&reply
, 0, sizeof(reply
));
1809 /* All good, send back reply. */
1810 reply
.ret_code
= htobe32(LTTNG_OK
);
1812 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1814 ERR("Relay begin data pending send reply failed");
1822 * End data pending command. This will check, for a given session id, if each
1823 * stream associated with it has its data_pending_check_done flag set. If not,
1824 * this means that the client lost track of the stream but the data is still
1825 * being streamed on our side. In this case, we inform the client that data is
1828 * Return to the client if there is data in flight or not with a ret_code.
1831 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1832 struct relay_connection
*conn
)
1835 struct lttng_ht_iter iter
;
1836 struct lttcomm_relayd_end_data_pending msg
;
1837 struct lttcomm_relayd_generic_reply reply
;
1838 struct relay_stream
*stream
;
1839 uint64_t session_id
;
1840 uint32_t is_data_inflight
= 0;
1845 DBG("End data pending command");
1847 if (!conn
->session
|| conn
->version_check_done
== 0) {
1848 ERR("Trying to check for data before version check");
1850 goto end_no_session
;
1853 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1854 if (ret
< sizeof(msg
)) {
1856 /* Orderly shutdown. Not necessary to print an error. */
1857 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1859 ERR("Relay didn't receive valid end data_pending struct size: %d",
1863 goto end_no_session
;
1866 session_id
= be64toh(msg
.session_id
);
1868 /* Iterate over all streams to see if the begin data pending flag is set. */
1870 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1872 if (stream
->session_id
== session_id
&&
1873 !stream
->data_pending_check_done
&& !stream
->terminated_flag
) {
1874 is_data_inflight
= 1;
1875 DBG("Data is still in flight for stream %" PRIu64
,
1876 stream
->stream_handle
);
1882 memset(&reply
, 0, sizeof(reply
));
1883 /* All good, send back reply. */
1884 reply
.ret_code
= htobe32(is_data_inflight
);
1886 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1888 ERR("Relay end data pending send reply failed");
1896 * Receive an index for a specific stream.
1898 * Return 0 on success else a negative value.
1901 int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1902 struct relay_connection
*conn
)
1904 int ret
, send_ret
, index_created
= 0;
1905 struct relay_session
*session
= conn
->session
;
1906 struct lttcomm_relayd_index index_info
;
1907 struct relay_index
*index
, *wr_index
= NULL
;
1908 struct lttcomm_relayd_generic_reply reply
;
1909 struct relay_stream
*stream
;
1910 uint64_t net_seq_num
;
1914 DBG("Relay receiving index");
1916 if (!session
|| conn
->version_check_done
== 0) {
1917 ERR("Trying to close a stream before version check");
1919 goto end_no_session
;
1922 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &index_info
,
1923 sizeof(index_info
), 0);
1924 if (ret
< sizeof(index_info
)) {
1926 /* Orderly shutdown. Not necessary to print an error. */
1927 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1929 ERR("Relay didn't receive valid index struct size : %d", ret
);
1932 goto end_no_session
;
1935 net_seq_num
= be64toh(index_info
.net_seq_num
);
1938 stream
= stream_find_by_id(relay_streams_ht
,
1939 be64toh(index_info
.relay_stream_id
));
1942 goto end_rcu_unlock
;
1945 /* Live beacon handling */
1946 if (index_info
.packet_size
== 0) {
1947 DBG("Received live beacon for stream %" PRIu64
, stream
->stream_handle
);
1950 * Only flag a stream inactive when it has already received data
1951 * and no indexes are in flight.
1953 if (stream
->total_index_received
> 0 && stream
->indexes_in_flight
== 0) {
1954 stream
->beacon_ts_end
= be64toh(index_info
.timestamp_end
);
1957 goto end_rcu_unlock
;
1959 stream
->beacon_ts_end
= -1ULL;
1962 index
= relay_index_find(stream
->stream_handle
, net_seq_num
);
1964 /* A successful creation will add the object to the HT. */
1965 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
1967 goto end_rcu_unlock
;
1970 stream
->indexes_in_flight
++;
1973 copy_index_control_data(index
, &index_info
);
1974 if (stream
->ctf_stream_id
== -1ULL) {
1975 stream
->ctf_stream_id
= be64toh(index_info
.stream_id
);
1978 if (index_created
) {
1980 * Try to add the relay index object to the hash table. If an object
1981 * already exist, destroy back the index created, set the data in this
1982 * object and write it on disk.
1984 relay_index_add(index
, &wr_index
);
1986 copy_index_control_data(wr_index
, &index_info
);
1990 /* The index already exists so write it on disk. */
1994 /* Do we have a writable ready index to write on disk. */
1996 ret
= relay_index_write(wr_index
->fd
, wr_index
);
1998 goto end_rcu_unlock
;
2000 stream
->total_index_received
++;
2001 stream
->indexes_in_flight
--;
2002 assert(stream
->indexes_in_flight
>= 0);
2008 memset(&reply
, 0, sizeof(reply
));
2010 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2012 reply
.ret_code
= htobe32(LTTNG_OK
);
2014 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2016 ERR("Relay sending close index id reply");
2025 * Receive the streams_sent message.
2027 * Return 0 on success else a negative value.
2030 int relay_streams_sent(struct lttcomm_relayd_hdr
*recv_hdr
,
2031 struct relay_connection
*conn
)
2034 struct lttcomm_relayd_generic_reply reply
;
2038 DBG("Relay receiving streams_sent");
2040 if (!conn
->session
|| conn
->version_check_done
== 0) {
2041 ERR("Trying to close a stream before version check");
2043 goto end_no_session
;
2047 * Flag every pending stream in the connection recv list that they are
2048 * ready to be used by the viewer.
2050 set_viewer_ready_flag(conn
);
2053 * Inform the viewer that there are new streams in the session.
2055 if (conn
->session
->viewer_refcount
) {
2056 uatomic_set(&conn
->session
->new_streams
, 1);
2059 memset(&reply
, 0, sizeof(reply
));
2060 reply
.ret_code
= htobe32(LTTNG_OK
);
2061 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2063 ERR("Relay sending sent_stream reply");
2075 * Process the commands received on the control socket
2078 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
2079 struct relay_connection
*conn
)
2083 switch (be32toh(recv_hdr
->cmd
)) {
2084 case RELAYD_CREATE_SESSION
:
2085 ret
= relay_create_session(recv_hdr
, conn
);
2087 case RELAYD_ADD_STREAM
:
2088 ret
= relay_add_stream(recv_hdr
, conn
);
2090 case RELAYD_START_DATA
:
2091 ret
= relay_start(recv_hdr
, conn
);
2093 case RELAYD_SEND_METADATA
:
2094 ret
= relay_recv_metadata(recv_hdr
, conn
);
2096 case RELAYD_VERSION
:
2097 ret
= relay_send_version(recv_hdr
, conn
);
2099 case RELAYD_CLOSE_STREAM
:
2100 ret
= relay_close_stream(recv_hdr
, conn
);
2102 case RELAYD_DATA_PENDING
:
2103 ret
= relay_data_pending(recv_hdr
, conn
);
2105 case RELAYD_QUIESCENT_CONTROL
:
2106 ret
= relay_quiescent_control(recv_hdr
, conn
);
2108 case RELAYD_BEGIN_DATA_PENDING
:
2109 ret
= relay_begin_data_pending(recv_hdr
, conn
);
2111 case RELAYD_END_DATA_PENDING
:
2112 ret
= relay_end_data_pending(recv_hdr
, conn
);
2114 case RELAYD_SEND_INDEX
:
2115 ret
= relay_recv_index(recv_hdr
, conn
);
2117 case RELAYD_STREAMS_SENT
:
2118 ret
= relay_streams_sent(recv_hdr
, conn
);
2120 case RELAYD_UPDATE_SYNC_INFO
:
2122 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
2123 relay_unknown_command(conn
);
2133 * Handle index for a data stream.
2135 * RCU read side lock MUST be acquired.
2137 * Return 0 on success else a negative value.
2139 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2142 int ret
= 0, index_created
= 0;
2143 uint64_t stream_id
, data_offset
;
2144 struct relay_index
*index
, *wr_index
= NULL
;
2148 stream_id
= stream
->stream_handle
;
2149 /* Get data offset because we are about to update the index. */
2150 data_offset
= htobe64(stream
->tracefile_size_current
);
2153 * Lookup for an existing index for that stream id/sequence number. If on
2154 * exists, the control thread already received the data for it thus we need
2155 * to write it on disk.
2157 index
= relay_index_find(stream_id
, net_seq_num
);
2159 /* A successful creation will add the object to the HT. */
2160 index
= relay_index_create(stream_id
, net_seq_num
);
2166 stream
->indexes_in_flight
++;
2169 if (rotate_index
|| stream
->index_fd
< 0) {
2170 index
->to_close_fd
= stream
->index_fd
;
2171 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
2172 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
2173 stream
->tracefile_count_current
);
2175 /* This will close the stream's index fd if one. */
2176 relay_index_free_safe(index
);
2179 stream
->index_fd
= ret
;
2181 index
->fd
= stream
->index_fd
;
2182 index
->index_data
.offset
= data_offset
;
2184 if (index_created
) {
2186 * Try to add the relay index object to the hash table. If an object
2187 * already exist, destroy back the index created and set the data.
2189 relay_index_add(index
, &wr_index
);
2191 /* Copy back data from the created index. */
2192 wr_index
->fd
= index
->fd
;
2193 wr_index
->to_close_fd
= index
->to_close_fd
;
2194 wr_index
->index_data
.offset
= data_offset
;
2198 /* The index already exists so write it on disk. */
2202 /* Do we have a writable ready index to write on disk. */
2204 ret
= relay_index_write(wr_index
->fd
, wr_index
);
2208 stream
->total_index_received
++;
2209 stream
->indexes_in_flight
--;
2210 assert(stream
->indexes_in_flight
>= 0);
2218 * relay_process_data: Process the data received on the data socket
2221 int relay_process_data(struct relay_connection
*conn
)
2223 int ret
= 0, rotate_index
= 0;
2225 struct relay_stream
*stream
;
2226 struct lttcomm_relayd_data_hdr data_hdr
;
2228 uint64_t net_seq_num
;
2230 struct relay_session
*session
;
2234 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &data_hdr
,
2235 sizeof(struct lttcomm_relayd_data_hdr
), 0);
2238 /* Orderly shutdown. Not necessary to print an error. */
2239 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2241 ERR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2247 stream_id
= be64toh(data_hdr
.stream_id
);
2250 stream
= stream_find_by_id(relay_streams_ht
, stream_id
);
2253 goto end_rcu_unlock
;
2256 session
= session_find_by_id(conn
->sessions_ht
, stream
->session_id
);
2259 data_size
= be32toh(data_hdr
.data_size
);
2260 if (data_buffer_size
< data_size
) {
2263 tmp_data_ptr
= realloc(data_buffer
, data_size
);
2264 if (!tmp_data_ptr
) {
2265 ERR("Allocating data buffer");
2268 goto end_rcu_unlock
;
2270 data_buffer
= tmp_data_ptr
;
2271 data_buffer_size
= data_size
;
2273 memset(data_buffer
, 0, data_size
);
2275 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2277 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2278 data_size
, stream_id
, net_seq_num
);
2279 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
2282 /* Orderly shutdown. Not necessary to print an error. */
2283 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2286 goto end_rcu_unlock
;
2289 /* Check if a rotation is needed. */
2290 if (stream
->tracefile_size
> 0 &&
2291 (stream
->tracefile_size_current
+ data_size
) >
2292 stream
->tracefile_size
) {
2293 struct relay_viewer_stream
*vstream
;
2296 new_id
= (stream
->tracefile_count_current
+ 1) %
2297 stream
->tracefile_count
;
2299 * When we wrap-around back to 0, we start overwriting old
2302 if (!stream
->tracefile_overwrite
&& new_id
== 0) {
2303 stream
->tracefile_overwrite
= 1;
2305 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
2306 if (stream
->tracefile_overwrite
) {
2307 stream
->oldest_tracefile_id
=
2308 (stream
->oldest_tracefile_id
+ 1) %
2309 stream
->tracefile_count
;
2311 vstream
= viewer_stream_find_by_id(stream
->stream_handle
);
2314 * The viewer is reading a file about to be
2315 * overwritten. Close the FDs it is
2316 * currently using and let it handle the fault.
2318 if (vstream
->tracefile_count_current
== new_id
) {
2319 pthread_mutex_lock(&vstream
->overwrite_lock
);
2320 vstream
->abort_flag
= 1;
2321 pthread_mutex_unlock(&vstream
->overwrite_lock
);
2322 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2323 stream
->channel_name
, new_id
);
2324 } else if (vstream
->tracefile_count_current
==
2325 stream
->tracefile_count_current
) {
2327 * The reader and writer were in the
2328 * same trace file, inform the viewer
2329 * that no new index will ever be added
2332 vstream
->close_write_flag
= 1;
2335 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
2336 stream
->tracefile_size
, stream
->tracefile_count
,
2337 relayd_uid
, relayd_gid
, stream
->fd
,
2338 &(stream
->tracefile_count_current
), &stream
->fd
);
2339 stream
->total_index_received
= 0;
2340 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
2342 ERR("Rotating stream output file");
2343 goto end_rcu_unlock
;
2345 /* Reset current size because we just perform a stream rotation. */
2346 stream
->tracefile_size_current
= 0;
2351 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2352 * index are NOT supported.
2354 if (session
->minor
>= 4 && !session
->snapshot
) {
2355 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2357 goto end_rcu_unlock
;
2361 /* Write data to stream output fd. */
2362 size_ret
= lttng_write(stream
->fd
, data_buffer
, data_size
);
2363 if (size_ret
< data_size
) {
2364 ERR("Relay error writing data to file");
2366 goto end_rcu_unlock
;
2369 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
2370 ret
, stream
->stream_handle
);
2372 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
2374 goto end_rcu_unlock
;
2376 stream
->tracefile_size_current
+= data_size
+ be32toh(data_hdr
.padding_size
);
2378 stream
->prev_seq
= net_seq_num
;
2380 try_close_stream(session
, stream
);
2389 void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2395 (void) lttng_poll_del(events
, pollfd
);
2397 ret
= close(pollfd
);
2399 ERR("Closing pollfd %d", pollfd
);
2403 static void destroy_connection(struct lttng_ht
*relay_connections_ht
,
2404 struct relay_connection
*conn
)
2406 assert(relay_connections_ht
);
2409 connection_delete(relay_connections_ht
, conn
);
2411 /* For the control socket, we try to destroy the session. */
2412 if (conn
->type
== RELAY_CONTROL
&& conn
->session
) {
2413 destroy_session(conn
->session
, conn
->sessions_ht
);
2416 connection_destroy(conn
);
2420 * This thread does the actual work
2423 void *relay_thread_worker(void *data
)
2425 int ret
, err
= -1, last_seen_data_fd
= -1;
2427 struct relay_connection
*conn
;
2428 struct lttng_poll_event events
;
2429 struct lttng_ht
*relay_connections_ht
;
2430 struct lttng_ht_iter iter
;
2431 struct lttcomm_relayd_hdr recv_hdr
;
2432 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
2433 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
2434 struct relay_index
*index
;
2436 DBG("[thread] Relay worker started");
2438 rcu_register_thread();
2440 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2442 if (testpoint(relayd_thread_worker
)) {
2443 goto error_testpoint
;
2446 health_code_update();
2448 /* table of connections indexed on socket */
2449 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2450 if (!relay_connections_ht
) {
2451 goto relay_connections_ht_error
;
2454 /* Tables of received indexes indexed by index handle and net_seq_num. */
2455 indexes_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64
);
2457 goto indexes_ht_error
;
2460 ret
= create_thread_poll_set(&events
, 2);
2462 goto error_poll_create
;
2465 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2472 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2474 health_code_update();
2476 /* Infinite blocking call, waiting for transmission */
2477 DBG3("Relayd worker thread polling...");
2478 health_poll_entry();
2479 ret
= lttng_poll_wait(&events
, -1);
2483 * Restart interrupted system call.
2485 if (errno
== EINTR
) {
2494 * Process control. The control connection is prioritised so we don't
2495 * starve it with high throughout put tracing data on the data
2498 for (i
= 0; i
< nb_fd
; i
++) {
2499 /* Fetch once the poll data */
2500 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2501 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2503 health_code_update();
2505 /* Thread quit pipe has been closed. Killing thread. */
2506 ret
= check_thread_quit_pipe(pollfd
, revents
);
2512 /* Inspect the relay conn pipe for new connection */
2513 if (pollfd
== relay_conn_pipe
[0]) {
2514 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2515 ERR("Relay connection pipe error");
2517 } else if (revents
& LPOLLIN
) {
2518 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
2522 conn
->sessions_ht
= sessions_ht
;
2523 connection_init(conn
);
2524 lttng_poll_add(&events
, conn
->sock
->fd
,
2525 LPOLLIN
| LPOLLRDHUP
);
2527 lttng_ht_add_unique_ulong(relay_connections_ht
,
2530 DBG("Connection socket %d added", conn
->sock
->fd
);
2534 conn
= connection_find_by_sock(relay_connections_ht
, pollfd
);
2535 /* If not found, there is a synchronization issue. */
2538 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2539 cleanup_connection_pollfd(&events
, pollfd
);
2540 destroy_connection(relay_connections_ht
, conn
);
2541 if (last_seen_data_fd
== pollfd
) {
2542 last_seen_data_fd
= last_notdel_data_fd
;
2544 } else if (revents
& LPOLLIN
) {
2545 if (conn
->type
== RELAY_CONTROL
) {
2546 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &recv_hdr
,
2547 sizeof(recv_hdr
), 0);
2549 /* Connection closed */
2550 cleanup_connection_pollfd(&events
, pollfd
);
2551 destroy_connection(relay_connections_ht
, conn
);
2552 DBG("Control connection closed with %d", pollfd
);
2554 ret
= relay_process_control(&recv_hdr
, conn
);
2556 /* Clear the session on error. */
2557 cleanup_connection_pollfd(&events
, pollfd
);
2558 destroy_connection(relay_connections_ht
, conn
);
2559 DBG("Connection closed with %d", pollfd
);
2565 * Flag the last seen data fd not deleted. It will be
2566 * used as the last seen fd if any fd gets deleted in
2569 last_notdel_data_fd
= pollfd
;
2572 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
2579 * The last loop handled a control request, go back to poll to make
2580 * sure we prioritise the control socket.
2586 if (last_seen_data_fd
>= 0) {
2587 for (i
= 0; i
< nb_fd
; i
++) {
2588 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2590 health_code_update();
2592 if (last_seen_data_fd
== pollfd
) {
2599 /* Process data connection. */
2600 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2601 /* Fetch the poll data. */
2602 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2603 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2605 health_code_update();
2607 /* Skip the command pipe. It's handled in the first loop. */
2608 if (pollfd
== relay_conn_pipe
[0]) {
2614 conn
= connection_find_by_sock(relay_connections_ht
, pollfd
);
2616 /* Skip it. Might be removed before. */
2621 if (revents
& LPOLLIN
) {
2622 if (conn
->type
!= RELAY_DATA
) {
2627 ret
= relay_process_data(conn
);
2628 /* Connection closed */
2630 cleanup_connection_pollfd(&events
, pollfd
);
2631 destroy_connection(relay_connections_ht
, conn
);
2632 DBG("Data connection closed with %d", pollfd
);
2634 * Every goto restart call sets the last seen fd where
2635 * here we don't really care since we gracefully
2636 * continue the loop after the connection is deleted.
2639 /* Keep last seen port. */
2640 last_seen_data_fd
= pollfd
;
2648 last_seen_data_fd
= -1;
2651 /* Normal exit, no error */
2656 lttng_poll_clean(&events
);
2658 /* Cleanup reamaining connection object. */
2660 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, conn
,
2662 health_code_update();
2663 destroy_connection(relay_connections_ht
, conn
);
2668 cds_lfht_for_each_entry(indexes_ht
->ht
, &iter
.iter
, index
,
2670 health_code_update();
2671 relay_index_delete(index
);
2672 relay_index_free_safe(index
);
2675 lttng_ht_destroy(indexes_ht
);
2677 lttng_ht_destroy(relay_connections_ht
);
2678 relay_connections_ht_error
:
2679 /* Close relay conn pipes */
2680 utils_close_pipe(relay_conn_pipe
);
2682 DBG("Thread exited with error");
2684 DBG("Worker thread cleanup complete");
2689 ERR("Health error occurred in %s", __func__
);
2691 health_unregister(health_relayd
);
2692 rcu_unregister_thread();
2698 * Create the relay command pipe to wake thread_manage_apps.
2699 * Closed in cleanup().
2701 static int create_relay_conn_pipe(void)
2705 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
2713 int main(int argc
, char **argv
)
2717 struct relay_local_data
*relay_ctx
;
2719 /* Parse arguments */
2721 if ((ret
= set_options(argc
, argv
)) < 0) {
2725 if ((ret
= set_signal_handler()) < 0) {
2729 /* Try to create directory if -o, --output is specified. */
2730 if (opt_output_path
) {
2731 if (*opt_output_path
!= '/') {
2732 ERR("Please specify an absolute path for -o, --output PATH");
2736 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2738 ERR("Unable to create %s", opt_output_path
);
2744 if (opt_daemon
|| opt_background
) {
2747 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
2754 * We are in the child. Make sure all other file
2755 * descriptors are closed, in case we are called with
2756 * more opened file descriptors than the standard ones.
2758 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
2763 /* Create thread quit pipe */
2764 if ((ret
= init_thread_quit_pipe()) < 0) {
2768 /* We need those values for the file/dir creation. */
2769 relayd_uid
= getuid();
2770 relayd_gid
= getgid();
2772 /* Check if daemon is UID = 0 */
2773 if (relayd_uid
== 0) {
2774 if (control_uri
->port
< 1024 || data_uri
->port
< 1024 || live_uri
->port
< 1024) {
2775 ERR("Need to be root to use ports < 1024");
2781 /* Setup the thread apps communication pipe. */
2782 if ((ret
= create_relay_conn_pipe()) < 0) {
2786 /* Init relay command queue. */
2787 cds_wfq_init(&relay_conn_queue
.queue
);
2789 /* Set up max poll set size */
2790 lttng_poll_set_max_size();
2792 /* Initialize communication library */
2794 lttcomm_inet_init();
2796 relay_ctx
= zmalloc(sizeof(struct relay_local_data
));
2798 PERROR("relay_ctx");
2802 /* tables of sessions indexed by session ID */
2803 relay_ctx
->sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2804 if (!relay_ctx
->sessions_ht
) {
2805 goto exit_relay_ctx_sessions
;
2808 /* tables of streams indexed by stream ID */
2809 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2810 if (!relay_streams_ht
) {
2811 goto exit_relay_ctx_streams
;
2814 /* tables of streams indexed by stream ID */
2815 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2816 if (!viewer_streams_ht
) {
2817 goto exit_relay_ctx_viewer_streams
;
2820 /* Initialize thread health monitoring */
2821 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
2822 if (!health_relayd
) {
2823 PERROR("health_app_create error");
2824 goto exit_health_app_create
;
2827 ret
= utils_create_pipe(health_quit_pipe
);
2829 goto error_health_pipe
;
2832 /* Create thread to manage the client socket */
2833 ret
= pthread_create(&health_thread
, NULL
,
2834 thread_manage_health
, (void *) NULL
);
2836 PERROR("pthread_create health");
2840 /* Setup the dispatcher thread */
2841 ret
= pthread_create(&dispatcher_thread
, NULL
,
2842 relay_thread_dispatcher
, (void *) NULL
);
2844 PERROR("pthread_create dispatcher");
2845 goto exit_dispatcher
;
2848 /* Setup the worker thread */
2849 ret
= pthread_create(&worker_thread
, NULL
,
2850 relay_thread_worker
, (void *) relay_ctx
);
2852 PERROR("pthread_create worker");
2856 /* Setup the listener thread */
2857 ret
= pthread_create(&listener_thread
, NULL
,
2858 relay_thread_listener
, (void *) NULL
);
2860 PERROR("pthread_create listener");
2864 ret
= live_start_threads(live_uri
, relay_ctx
);
2866 ERR("Starting live viewer threads");
2871 ret
= pthread_join(listener_thread
, &status
);
2873 PERROR("pthread_join");
2874 goto error
; /* join error, exit without cleanup */
2878 ret
= pthread_join(worker_thread
, &status
);
2880 PERROR("pthread_join");
2881 goto error
; /* join error, exit without cleanup */
2885 ret
= pthread_join(dispatcher_thread
, &status
);
2887 PERROR("pthread_join");
2888 goto error
; /* join error, exit without cleanup */
2892 ret
= pthread_join(health_thread
, &status
);
2894 PERROR("pthread_join health thread");
2895 goto error
; /* join error, exit without cleanup */
2899 * Stop live threads only after joining other threads.
2901 live_stop_threads();
2904 utils_close_pipe(health_quit_pipe
);
2907 health_app_destroy(health_relayd
);
2909 exit_health_app_create
:
2910 lttng_ht_destroy(viewer_streams_ht
);
2912 exit_relay_ctx_viewer_streams
:
2913 lttng_ht_destroy(relay_streams_ht
);
2915 exit_relay_ctx_streams
:
2916 lttng_ht_destroy(relay_ctx
->sessions_ht
);
2918 exit_relay_ctx_sessions
: