2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
33 #include <sys/types.h>
36 #include <urcu/futex.h>
37 #include <urcu/uatomic.h>
42 #include <lttng/lttng.h>
43 #include <common/common.h>
44 #include <common/compat/poll.h>
45 #include <common/compat/socket.h>
46 #include <common/defaults.h>
47 #include <common/futex.h>
48 #include <common/sessiond-comm/sessiond-comm.h>
49 #include <common/sessiond-comm/inet.h>
50 #include <common/sessiond-comm/relayd.h>
51 #include <common/uri.h>
52 #include <common/utils.h>
55 #include "ctf-trace.h"
58 #include "lttng-relayd.h"
61 /* command line options */
62 char *opt_output_path
;
63 static int opt_daemon
;
64 static struct lttng_uri
*control_uri
;
65 static struct lttng_uri
*data_uri
;
66 static struct lttng_uri
*live_uri
;
71 * Quit pipe for all threads. This permits a single cancellation point
72 * for all threads when receiving an event on the pipe.
74 static int thread_quit_pipe
[2] = { -1, -1 };
77 * This pipe is used to inform the worker thread that a command is queued and
78 * ready to be processed.
80 static int relay_cmd_pipe
[2] = { -1, -1 };
82 /* Shared between threads */
83 static int dispatch_thread_exit
;
85 static pthread_t listener_thread
;
86 static pthread_t dispatcher_thread
;
87 static pthread_t worker_thread
;
89 static uint64_t last_relay_stream_id
;
90 static uint64_t last_relay_session_id
;
93 * Relay command queue.
95 * The relay_thread_listener and relay_thread_dispatcher communicate with this
98 static struct relay_cmd_queue relay_cmd_queue
;
100 /* buffer allocated at startup, used to store the trace data */
101 static char *data_buffer
;
102 static unsigned int data_buffer_size
;
104 /* Global hash table that stores relay index object. */
105 static struct lttng_ht
*indexes_ht
;
107 /* We need those values for the file/dir creation. */
108 static uid_t relayd_uid
;
109 static gid_t relayd_gid
;
111 /* Global relay stream hash table. */
112 struct lttng_ht
*relay_streams_ht
;
115 * usage function on stderr
120 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
121 fprintf(stderr
, " -h, --help Display this usage.\n");
122 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
123 fprintf(stderr
, " -C, --control-port URL Control port listening.\n");
124 fprintf(stderr
, " -D, --data-port URL Data port listening.\n");
125 fprintf(stderr
, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
126 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
130 int parse_args(int argc
, char **argv
)
134 char *default_address
;
136 static struct option long_options
[] = {
137 { "control-port", 1, 0, 'C', },
138 { "data-port", 1, 0, 'D', },
139 { "daemonize", 0, 0, 'd', },
140 { "help", 0, 0, 'h', },
141 { "output", 1, 0, 'o', },
142 { "verbose", 0, 0, 'v', },
147 int option_index
= 0;
148 c
= getopt_long(argc
, argv
, "dhv" "C:D:o:",
149 long_options
, &option_index
);
156 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
158 fprintf(stderr
, " with arg %s\n", optarg
);
162 ret
= uri_parse(optarg
, &control_uri
);
164 ERR("Invalid control URI specified");
167 if (control_uri
->port
== 0) {
168 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
172 ret
= uri_parse(optarg
, &data_uri
);
174 ERR("Invalid data URI specified");
177 if (data_uri
->port
== 0) {
178 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
188 ret
= asprintf(&opt_output_path
, "%s", optarg
);
190 PERROR("asprintf opt_output_path");
195 /* Verbose level can increase using multiple -v */
196 lttng_opt_verbose
+= 1;
199 /* Unknown option or other error.
200 * Error is printed by getopt, just return */
206 /* assign default values */
207 if (control_uri
== NULL
) {
208 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
209 DEFAULT_NETWORK_CONTROL_PORT
);
211 PERROR("asprintf default data address");
215 ret
= uri_parse(default_address
, &control_uri
);
216 free(default_address
);
218 ERR("Invalid control URI specified");
222 if (data_uri
== NULL
) {
223 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
224 DEFAULT_NETWORK_DATA_PORT
);
226 PERROR("asprintf default data address");
230 ret
= uri_parse(default_address
, &data_uri
);
231 free(default_address
);
233 ERR("Invalid data URI specified");
237 if (live_uri
== NULL
) {
238 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
239 DEFAULT_NETWORK_VIEWER_PORT
);
241 PERROR("asprintf default viewer control address");
245 ret
= uri_parse(default_address
, &live_uri
);
246 free(default_address
);
248 ERR("Invalid viewer control URI specified");
265 /* free the dynamically allocated opt_output_path */
266 free(opt_output_path
);
268 /* Close thread quit pipes */
269 utils_close_pipe(thread_quit_pipe
);
271 uri_free(control_uri
);
276 * Write to writable pipe used to notify a thread.
279 int notify_thread_pipe(int wpipe
)
284 ret
= write(wpipe
, "!", 1);
285 } while (ret
< 0 && errno
== EINTR
);
286 if (ret
< 0 || ret
!= 1) {
287 PERROR("write poll pipe");
294 * Stop all threads by closing the thread quit pipe.
297 void stop_threads(void)
301 /* Stopping all threads */
302 DBG("Terminating all threads");
303 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
305 ERR("write error on thread quit pipe");
308 /* Dispatch thread */
309 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
310 futex_nto1_wake(&relay_cmd_queue
.futex
);
314 * Signal handler for the daemon
316 * Simply stop all worker threads, leaving main() return gracefully after
317 * joining all threads and calling cleanup().
320 void sighandler(int sig
)
324 DBG("SIGPIPE caught");
327 DBG("SIGINT caught");
331 DBG("SIGTERM caught");
340 * Setup signal handler for :
341 * SIGINT, SIGTERM, SIGPIPE
344 int set_signal_handler(void)
350 if ((ret
= sigemptyset(&sigset
)) < 0) {
351 PERROR("sigemptyset");
355 sa
.sa_handler
= sighandler
;
358 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
363 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
368 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
373 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
379 * Init thread quit pipe.
381 * Return -1 on error or 0 if all pipes are created.
384 int init_thread_quit_pipe(void)
388 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
394 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
397 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
401 if (events
== NULL
|| size
== 0) {
406 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
412 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
424 * Check if the thread quit pipe was triggered.
426 * Return 1 if it was triggered else 0;
429 int check_thread_quit_pipe(int fd
, uint32_t events
)
431 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
439 * Create and init socket from uri.
442 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
445 struct lttcomm_sock
*sock
= NULL
;
447 sock
= lttcomm_alloc_sock_from_uri(uri
);
449 ERR("Allocating socket");
453 ret
= lttcomm_create_sock(sock
);
457 DBG("Listening on sock %d", sock
->fd
);
459 ret
= sock
->ops
->bind(sock
);
464 ret
= sock
->ops
->listen(sock
, -1);
474 lttcomm_destroy_sock(sock
);
480 * Return nonzero if stream needs to be closed.
483 int close_stream_check(struct relay_stream
*stream
)
486 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
488 * We are about to close the stream so set the data pending flag to 1
489 * which will make the end data pending command skip the stream which
490 * is now closed and ready. Note that after proceeding to a file close,
491 * the written file is ready for reading.
493 stream
->data_pending_check_done
= 1;
500 * This thread manages the listening for new connections on the network
503 void *relay_thread_listener(void *data
)
505 int i
, ret
, pollfd
, err
= -1;
507 uint32_t revents
, nb_fd
;
508 struct lttng_poll_event events
;
509 struct lttcomm_sock
*control_sock
, *data_sock
;
511 DBG("[thread] Relay listener started");
513 control_sock
= relay_init_sock(control_uri
);
515 goto error_sock_control
;
518 data_sock
= relay_init_sock(data_uri
);
520 goto error_sock_relay
;
524 * Pass 3 as size here for the thread quit pipe, control and data socket.
526 ret
= create_thread_poll_set(&events
, 3);
528 goto error_create_poll
;
531 /* Add the control socket */
532 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
537 /* Add the data socket */
538 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
544 DBG("Listener accepting connections");
547 ret
= lttng_poll_wait(&events
, -1);
550 * Restart interrupted system call.
552 if (errno
== EINTR
) {
560 DBG("Relay new connection received");
561 for (i
= 0; i
< nb_fd
; i
++) {
562 /* Fetch once the poll data */
563 revents
= LTTNG_POLL_GETEV(&events
, i
);
564 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
566 /* Thread quit pipe has been closed. Killing thread. */
567 ret
= check_thread_quit_pipe(pollfd
, revents
);
573 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
574 ERR("socket poll error");
576 } else if (revents
& LPOLLIN
) {
578 * Get allocated in this thread,
579 * enqueued to a global queue, dequeued
580 * and freed in the worker thread.
582 struct relay_command
*relay_cmd
;
583 struct lttcomm_sock
*newsock
;
585 relay_cmd
= zmalloc(sizeof(struct relay_command
));
586 if (relay_cmd
== NULL
) {
587 PERROR("relay command zmalloc");
591 if (pollfd
== data_sock
->fd
) {
592 newsock
= data_sock
->ops
->accept(data_sock
);
594 PERROR("accepting data sock");
598 relay_cmd
->type
= RELAY_DATA
;
599 DBG("Relay data connection accepted, socket %d", newsock
->fd
);
601 assert(pollfd
== control_sock
->fd
);
602 newsock
= control_sock
->ops
->accept(control_sock
);
604 PERROR("accepting control sock");
608 relay_cmd
->type
= RELAY_CONTROL
;
609 DBG("Relay control connection accepted, socket %d", newsock
->fd
);
611 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
,
614 PERROR("setsockopt inet");
615 lttcomm_destroy_sock(newsock
);
619 relay_cmd
->sock
= newsock
;
621 * Lock free enqueue the request.
623 cds_wfq_enqueue(&relay_cmd_queue
.queue
, &relay_cmd
->node
);
626 * Wake the dispatch queue futex. Implicit memory
627 * barrier with the exchange in cds_wfq_enqueue.
629 futex_nto1_wake(&relay_cmd_queue
.futex
);
637 lttng_poll_clean(&events
);
639 if (data_sock
->fd
>= 0) {
640 ret
= data_sock
->ops
->close(data_sock
);
645 lttcomm_destroy_sock(data_sock
);
647 if (control_sock
->fd
>= 0) {
648 ret
= control_sock
->ops
->close(control_sock
);
653 lttcomm_destroy_sock(control_sock
);
656 DBG("Thread exited with error");
658 DBG("Relay listener thread cleanup complete");
664 * This thread manages the dispatching of the requests to worker threads
667 void *relay_thread_dispatcher(void *data
)
670 struct cds_wfq_node
*node
;
671 struct relay_command
*relay_cmd
= NULL
;
673 DBG("[thread] Relay dispatcher started");
675 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
676 /* Atomically prepare the queue futex */
677 futex_nto1_prepare(&relay_cmd_queue
.futex
);
680 /* Dequeue commands */
681 node
= cds_wfq_dequeue_blocking(&relay_cmd_queue
.queue
);
683 DBG("Woken up but nothing in the relay command queue");
684 /* Continue thread execution */
688 relay_cmd
= caa_container_of(node
, struct relay_command
, node
);
689 DBG("Dispatching request waiting on sock %d", relay_cmd
->sock
->fd
);
692 * Inform worker thread of the new request. This
693 * call is blocking so we can be assured that the data will be read
694 * at some point in time or wait to the end of the world :)
697 ret
= write(relay_cmd_pipe
[1], relay_cmd
,
698 sizeof(struct relay_command
));
699 } while (ret
< 0 && errno
== EINTR
);
701 if (ret
< 0 || ret
!= sizeof(struct relay_command
)) {
702 PERROR("write cmd pipe");
705 } while (node
!= NULL
);
707 /* Futex wait on queue. Blocking call on futex() */
708 futex_nto1_wait(&relay_cmd_queue
.futex
);
712 DBG("Dispatch thread dying");
718 * Get stream from stream id.
719 * Need to be called with RCU read-side lock held.
721 struct relay_stream
*relay_stream_find_by_id(uint64_t stream_id
)
723 struct lttng_ht_node_ulong
*node
;
724 struct lttng_ht_iter iter
;
725 struct relay_stream
*ret
;
727 lttng_ht_lookup(relay_streams_ht
,
728 (void *)((unsigned long) stream_id
),
730 node
= lttng_ht_iter_get_node_ulong(&iter
);
732 DBG("Relay stream %" PRIu64
" not found", stream_id
);
737 ret
= caa_container_of(node
, struct relay_stream
, stream_n
);
744 void deferred_free_stream(struct rcu_head
*head
)
746 struct relay_stream
*stream
=
747 caa_container_of(head
, struct relay_stream
, rcu_node
);
749 ctf_trace_try_destroy(stream
->ctf_trace
);
751 free(stream
->path_name
);
752 free(stream
->channel_name
);
757 void deferred_free_session(struct rcu_head
*head
)
759 struct relay_session
*session
=
760 caa_container_of(head
, struct relay_session
, rcu_node
);
764 static void close_stream(struct relay_stream
*stream
,
765 struct lttng_ht
*viewer_streams_ht
, struct lttng_ht
*ctf_traces_ht
)
768 struct relay_viewer_stream
*vstream
;
769 struct lttng_ht_iter iter
;
772 assert(viewer_streams_ht
);
774 delret
= close(stream
->fd
);
776 PERROR("close stream");
779 if (stream
->index_fd
>= 0) {
780 delret
= close(stream
->index_fd
);
782 PERROR("close stream index_fd");
786 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
,
790 * Set the last good value into the viewer stream. This is done
791 * right before the stream gets deleted from the hash table. The
792 * lookup failure on the live thread side of a stream indicates
793 * that the viewer stream index received value should be used.
795 vstream
->total_index_received
= stream
->total_index_received
;
798 iter
.iter
.node
= &stream
->stream_n
.node
;
799 delret
= lttng_ht_del(relay_streams_ht
, &iter
);
801 iter
.iter
.node
= &stream
->ctf_trace_node
.node
;
802 delret
= lttng_ht_del(ctf_traces_ht
, &iter
);
804 call_rcu(&stream
->rcu_node
, deferred_free_stream
);
805 DBG("Closed tracefile %d from close stream", stream
->fd
);
809 * relay_delete_session: Free all memory associated with a session and
813 void relay_delete_session(struct relay_command
*cmd
,
814 struct lttng_ht
*sessions_ht
)
816 struct lttng_ht_iter iter
;
817 struct lttng_ht_node_ulong
*node
;
818 struct relay_stream
*stream
;
825 DBG("Relay deleting session %" PRIu64
, cmd
->session
->id
);
828 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, node
, node
) {
829 node
= lttng_ht_iter_get_node_ulong(&iter
);
831 stream
= caa_container_of(node
,
832 struct relay_stream
, stream_n
);
833 if (stream
->session
== cmd
->session
) {
834 ret
= close(stream
->fd
);
836 PERROR("close stream fd on delete session");
838 ret
= lttng_ht_del(relay_streams_ht
, &iter
);
840 call_rcu(&stream
->rcu_node
,
841 deferred_free_stream
);
843 /* Cleanup index of that stream. */
844 relay_index_destroy_by_stream_id(stream
->stream_handle
,
848 iter
.iter
.node
= &cmd
->session
->session_n
.node
;
849 ret
= lttng_ht_del(sessions_ht
, &iter
);
851 call_rcu(&cmd
->session
->rcu_node
,
852 deferred_free_session
);
857 * Copy index data from the control port to a given index object.
859 static void copy_index_control_data(struct relay_index
*index
,
860 struct lttcomm_relayd_index
*data
)
866 * The index on disk is encoded in big endian, so we don't need to convert
867 * the data received on the network. The data_offset value is NEVER
868 * modified here and is updated by the data thread.
870 index
->index_data
.packet_size
= data
->packet_size
;
871 index
->index_data
.content_size
= data
->content_size
;
872 index
->index_data
.timestamp_begin
= data
->timestamp_begin
;
873 index
->index_data
.timestamp_end
= data
->timestamp_end
;
874 index
->index_data
.events_discarded
= data
->events_discarded
;
875 index
->index_data
.stream_id
= data
->stream_id
;
879 * Handle the RELAYD_CREATE_SESSION command.
881 * On success, send back the session id or else return a negative value.
884 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
885 struct relay_command
*cmd
,
886 struct lttng_ht
*sessions_ht
)
888 int ret
= 0, send_ret
;
889 struct relay_session
*session
;
890 struct lttcomm_relayd_status_session reply
;
895 memset(&reply
, 0, sizeof(reply
));
897 session
= zmalloc(sizeof(struct relay_session
));
898 if (session
== NULL
) {
899 PERROR("relay session zmalloc");
904 session
->id
= ++last_relay_session_id
;
905 session
->sock
= cmd
->sock
;
906 cmd
->session
= session
;
908 reply
.session_id
= htobe64(session
->id
);
910 switch (cmd
->minor
) {
911 case 4: /* LTTng sessiond 2.4 */
913 ret
= cmd_create_session_2_4(cmd
, session
);
917 lttng_ht_node_init_ulong(&session
->session_n
,
918 (unsigned long) session
->id
);
919 lttng_ht_add_unique_ulong(sessions_ht
,
920 &session
->session_n
);
922 DBG("Created session %" PRIu64
, session
->id
);
926 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
928 reply
.ret_code
= htobe32(LTTNG_OK
);
931 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
933 ERR("Relayd sending session id");
941 * relay_add_stream: allocate a new stream for a session
944 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
945 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
947 struct relay_session
*session
= cmd
->session
;
948 struct relay_stream
*stream
= NULL
;
949 struct lttcomm_relayd_status_stream reply
;
952 if (!session
|| cmd
->version_check_done
== 0) {
953 ERR("Trying to add a stream before version check");
958 stream
= zmalloc(sizeof(struct relay_stream
));
959 if (stream
== NULL
) {
960 PERROR("relay stream zmalloc");
965 switch (cmd
->minor
) {
966 case 1: /* LTTng sessiond 2.1 */
967 ret
= cmd_recv_stream_2_1(cmd
, stream
);
969 case 2: /* LTTng sessiond 2.2 */
971 ret
= cmd_recv_stream_2_2(cmd
, stream
);
975 goto err_free_stream
;
979 stream
->stream_handle
= ++last_relay_stream_id
;
980 stream
->prev_seq
= -1ULL;
981 stream
->session
= session
;
982 stream
->index_fd
= -1;
983 stream
->read_index_fd
= -1;
984 stream
->ctf_trace
= NULL
;
985 pthread_mutex_init(&stream
->lock
, NULL
);
987 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
);
989 ERR("relay creating output directory");
994 * No need to use run_as API here because whatever we receives, the relayd
995 * uses its own credentials for the stream files.
997 ret
= utils_create_stream_file(stream
->path_name
, stream
->channel_name
,
998 stream
->tracefile_size
, 0, relayd_uid
, relayd_gid
, NULL
);
1000 ERR("Create output file");
1004 if (stream
->tracefile_size
) {
1005 DBG("Tracefile %s/%s_0 created", stream
->path_name
, stream
->channel_name
);
1007 DBG("Tracefile %s/%s created", stream
->path_name
, stream
->channel_name
);
1010 if (!strncmp(stream
->channel_name
, DEFAULT_METADATA_NAME
, NAME_MAX
)) {
1011 stream
->metadata_flag
= 1;
1013 * When we receive a new metadata stream, we create a new
1014 * ctf_trace and we assign this ctf_trace to all streams with
1017 * If later on we receive a new stream for the same ctf_trace,
1018 * we copy the information from the first hit in the HT to the
1021 stream
->ctf_trace
= ctf_trace_create();
1022 if (!stream
->ctf_trace
) {
1026 stream
->ctf_trace
->refcount
++;
1027 stream
->ctf_trace
->metadata_stream
= stream
;
1029 ctf_trace_assign(cmd
->ctf_traces_ht
, stream
);
1031 lttng_ht_node_init_ulong(&stream
->stream_n
,
1032 (unsigned long) stream
->stream_handle
);
1033 lttng_ht_add_unique_ulong(relay_streams_ht
,
1036 lttng_ht_node_init_str(&stream
->ctf_trace_node
, stream
->path_name
);
1037 lttng_ht_add_str(cmd
->ctf_traces_ht
, &stream
->ctf_trace_node
);
1039 DBG("Relay new stream added %s with ID %" PRIu64
, stream
->channel_name
,
1040 stream
->stream_handle
);
1043 reply
.handle
= htobe64(stream
->stream_handle
);
1044 /* send the session id to the client or a negative return code on error */
1046 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1047 /* stream was not properly added to the ht, so free it */
1050 reply
.ret_code
= htobe32(LTTNG_OK
);
1053 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1054 sizeof(struct lttcomm_relayd_status_stream
), 0);
1056 ERR("Relay sending stream id");
1065 free(stream
->path_name
);
1066 free(stream
->channel_name
);
1072 * relay_close_stream: close a specific stream
1075 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1076 struct relay_command
*cmd
, struct lttng_ht
*viewer_streams_ht
)
1079 struct relay_session
*session
= cmd
->session
;
1080 struct lttcomm_relayd_close_stream stream_info
;
1081 struct lttcomm_relayd_generic_reply reply
;
1082 struct relay_stream
*stream
;
1084 DBG("Close stream received");
1086 if (!session
|| cmd
->version_check_done
== 0) {
1087 ERR("Trying to close a stream before version check");
1089 goto end_no_session
;
1092 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &stream_info
,
1093 sizeof(struct lttcomm_relayd_close_stream
), 0);
1094 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1096 /* Orderly shutdown. Not necessary to print an error. */
1097 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1099 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1102 goto end_no_session
;
1106 stream
= relay_stream_find_by_id(be64toh(stream_info
.stream_id
));
1112 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1113 stream
->close_flag
= 1;
1115 if (close_stream_check(stream
)) {
1116 close_stream(stream
, viewer_streams_ht
, cmd
->ctf_traces_ht
);
1123 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1125 reply
.ret_code
= htobe32(LTTNG_OK
);
1127 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1128 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1130 ERR("Relay sending stream id");
1139 * relay_unknown_command: send -1 if received unknown command
1142 void relay_unknown_command(struct relay_command
*cmd
)
1144 struct lttcomm_relayd_generic_reply reply
;
1147 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1148 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1149 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1151 ERR("Relay sending unknown command");
1156 * relay_start: send an acknowledgment to the client to tell if we are
1157 * ready to receive data. We are ready if a session is established.
1160 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1161 struct relay_command
*cmd
)
1163 int ret
= htobe32(LTTNG_OK
);
1164 struct lttcomm_relayd_generic_reply reply
;
1165 struct relay_session
*session
= cmd
->session
;
1168 DBG("Trying to start the streaming without a session established");
1169 ret
= htobe32(LTTNG_ERR_UNK
);
1172 reply
.ret_code
= ret
;
1173 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1174 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1176 ERR("Relay sending start ack");
1183 * Append padding to the file pointed by the file descriptor fd.
1185 static int write_padding_to_file(int fd
, uint32_t size
)
1194 zeros
= zmalloc(size
);
1195 if (zeros
== NULL
) {
1196 PERROR("zmalloc zeros for padding");
1202 ret
= write(fd
, zeros
, size
);
1203 } while (ret
< 0 && errno
== EINTR
);
1204 if (ret
< 0 || ret
!= size
) {
1205 PERROR("write padding to file");
1215 * relay_recv_metadata: receive the metada for the session.
1218 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1219 struct relay_command
*cmd
)
1221 int ret
= htobe32(LTTNG_OK
);
1222 struct relay_session
*session
= cmd
->session
;
1223 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1224 struct relay_stream
*metadata_stream
;
1225 uint64_t data_size
, payload_size
;
1228 ERR("Metadata sent before version check");
1233 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1234 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1235 ERR("Incorrect data size");
1239 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1241 if (data_buffer_size
< data_size
) {
1242 /* In case the realloc fails, we can free the memory */
1245 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1246 if (!tmp_data_ptr
) {
1247 ERR("Allocating data buffer");
1252 data_buffer
= tmp_data_ptr
;
1253 data_buffer_size
= data_size
;
1255 memset(data_buffer
, 0, data_size
);
1256 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1257 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
1258 if (ret
< 0 || ret
!= data_size
) {
1260 /* Orderly shutdown. Not necessary to print an error. */
1261 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1263 ERR("Relay didn't receive the whole metadata");
1268 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1271 metadata_stream
= relay_stream_find_by_id(
1272 be64toh(metadata_struct
->stream_id
));
1273 if (!metadata_stream
) {
1279 ret
= write(metadata_stream
->fd
, metadata_struct
->payload
,
1281 } while (ret
< 0 && errno
== EINTR
);
1282 if (ret
< 0 || ret
!= payload_size
) {
1283 ERR("Relay error writing metadata on file");
1288 ret
= write_padding_to_file(metadata_stream
->fd
,
1289 be32toh(metadata_struct
->padding_size
));
1293 metadata_stream
->ctf_trace
->metadata_received
+=
1294 payload_size
+ be32toh(metadata_struct
->padding_size
);
1296 DBG2("Relay metadata written");
1305 * relay_send_version: send relayd version number
1308 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1309 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1312 struct lttcomm_relayd_version reply
, msg
;
1316 cmd
->version_check_done
= 1;
1318 /* Get version from the other side. */
1319 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1320 if (ret
< 0 || ret
!= sizeof(msg
)) {
1322 /* Orderly shutdown. Not necessary to print an error. */
1323 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1325 ERR("Relay failed to receive the version values.");
1331 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1332 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1334 /* Major versions must be the same */
1335 if (reply
.major
!= be32toh(msg
.major
)) {
1336 DBG("Incompatible major versions (%u vs %u), deleting session",
1337 reply
.major
, be32toh(msg
.major
));
1338 relay_delete_session(cmd
, sessions_ht
);
1343 cmd
->major
= reply
.major
;
1344 /* We adapt to the lowest compatible version */
1345 if (reply
.minor
<= be32toh(msg
.minor
)) {
1346 cmd
->minor
= reply
.minor
;
1348 cmd
->minor
= be32toh(msg
.minor
);
1351 reply
.major
= htobe32(reply
.major
);
1352 reply
.minor
= htobe32(reply
.minor
);
1353 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1354 sizeof(struct lttcomm_relayd_version
), 0);
1356 ERR("Relay sending version");
1359 DBG("Version check done using protocol %u.%u", cmd
->major
,
1367 * Check for data pending for a given stream id from the session daemon.
1370 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1371 struct relay_command
*cmd
)
1373 struct relay_session
*session
= cmd
->session
;
1374 struct lttcomm_relayd_data_pending msg
;
1375 struct lttcomm_relayd_generic_reply reply
;
1376 struct relay_stream
*stream
;
1378 uint64_t last_net_seq_num
, stream_id
;
1380 DBG("Data pending command received");
1382 if (!session
|| cmd
->version_check_done
== 0) {
1383 ERR("Trying to check for data before version check");
1385 goto end_no_session
;
1388 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1389 if (ret
< sizeof(msg
)) {
1391 /* Orderly shutdown. Not necessary to print an error. */
1392 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1394 ERR("Relay didn't receive valid data_pending struct size : %d",
1398 goto end_no_session
;
1401 stream_id
= be64toh(msg
.stream_id
);
1402 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1405 stream
= relay_stream_find_by_id(stream_id
);
1406 if (stream
== NULL
) {
1411 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1412 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1415 /* Avoid wrapping issue */
1416 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1417 /* Data has in fact been written and is NOT pending */
1420 /* Data still being streamed thus pending */
1424 /* Pending check is now done. */
1425 stream
->data_pending_check_done
= 1;
1430 reply
.ret_code
= htobe32(ret
);
1431 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1433 ERR("Relay data pending ret code failed");
1441 * Wait for the control socket to reach a quiescent state.
1443 * Note that for now, when receiving this command from the session daemon, this
1444 * means that every subsequent commands or data received on the control socket
1445 * has been handled. So, this is why we simply return OK here.
1448 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1449 struct relay_command
*cmd
)
1453 struct relay_stream
*stream
;
1454 struct lttng_ht_iter iter
;
1455 struct lttcomm_relayd_quiescent_control msg
;
1456 struct lttcomm_relayd_generic_reply reply
;
1458 DBG("Checking quiescent state on control socket");
1460 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1461 ERR("Trying to check for data before version check");
1463 goto end_no_session
;
1466 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1467 if (ret
< sizeof(msg
)) {
1469 /* Orderly shutdown. Not necessary to print an error. */
1470 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1472 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1476 goto end_no_session
;
1479 stream_id
= be64toh(msg
.stream_id
);
1482 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1484 if (stream
->stream_handle
== stream_id
) {
1485 stream
->data_pending_check_done
= 1;
1486 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1493 reply
.ret_code
= htobe32(LTTNG_OK
);
1494 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1496 ERR("Relay data quiescent control ret code failed");
1504 * Initialize a data pending command. This means that a client is about to ask
1505 * for data pending for each stream he/she holds. Simply iterate over all
1506 * streams of a session and set the data_pending_check_done flag.
1508 * This command returns to the client a LTTNG_OK code.
1511 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1512 struct relay_command
*cmd
)
1515 struct lttng_ht_iter iter
;
1516 struct lttcomm_relayd_begin_data_pending msg
;
1517 struct lttcomm_relayd_generic_reply reply
;
1518 struct relay_stream
*stream
;
1519 uint64_t session_id
;
1524 DBG("Init streams for data pending");
1526 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1527 ERR("Trying to check for data before version check");
1529 goto end_no_session
;
1532 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1533 if (ret
< sizeof(msg
)) {
1535 /* Orderly shutdown. Not necessary to print an error. */
1536 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1538 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1542 goto end_no_session
;
1545 session_id
= be64toh(msg
.session_id
);
1548 * Iterate over all streams to set the begin data pending flag. For now, the
1549 * streams are indexed by stream handle so we have to iterate over all
1550 * streams to find the one associated with the right session_id.
1553 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1555 if (stream
->session
->id
== session_id
) {
1556 stream
->data_pending_check_done
= 0;
1557 DBG("Set begin data pending flag to stream %" PRIu64
,
1558 stream
->stream_handle
);
1563 /* All good, send back reply. */
1564 reply
.ret_code
= htobe32(LTTNG_OK
);
1566 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1568 ERR("Relay begin data pending send reply failed");
1576 * End data pending command. This will check, for a given session id, if each
1577 * stream associated with it has its data_pending_check_done flag set. If not,
1578 * this means that the client lost track of the stream but the data is still
1579 * being streamed on our side. In this case, we inform the client that data is
1582 * Return to the client if there is data in flight or not with a ret_code.
1585 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1586 struct relay_command
*cmd
)
1589 struct lttng_ht_iter iter
;
1590 struct lttcomm_relayd_end_data_pending msg
;
1591 struct lttcomm_relayd_generic_reply reply
;
1592 struct relay_stream
*stream
;
1593 uint64_t session_id
;
1594 uint32_t is_data_inflight
= 0;
1599 DBG("End data pending command");
1601 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1602 ERR("Trying to check for data before version check");
1604 goto end_no_session
;
1607 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1608 if (ret
< sizeof(msg
)) {
1610 /* Orderly shutdown. Not necessary to print an error. */
1611 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1613 ERR("Relay didn't receive valid end data_pending struct size: %d",
1617 goto end_no_session
;
1620 session_id
= be64toh(msg
.session_id
);
1622 /* Iterate over all streams to see if the begin data pending flag is set. */
1624 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1626 if (stream
->session
->id
== session_id
&&
1627 !stream
->data_pending_check_done
) {
1628 is_data_inflight
= 1;
1629 DBG("Data is still in flight for stream %" PRIu64
,
1630 stream
->stream_handle
);
1636 /* All good, send back reply. */
1637 reply
.ret_code
= htobe32(is_data_inflight
);
1639 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1641 ERR("Relay end data pending send reply failed");
1649 * Receive an index for a specific stream.
1651 * Return 0 on success else a negative value.
1654 int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1655 struct relay_command
*cmd
, struct lttng_ht
*indexes_ht
)
1657 int ret
, send_ret
, index_created
= 0;
1658 struct relay_session
*session
= cmd
->session
;
1659 struct lttcomm_relayd_index index_info
;
1660 struct relay_index
*index
, *wr_index
= NULL
;
1661 struct lttcomm_relayd_generic_reply reply
;
1662 struct relay_stream
*stream
;
1663 uint64_t net_seq_num
;
1668 DBG("Relay receiving index");
1670 if (!session
|| cmd
->version_check_done
== 0) {
1671 ERR("Trying to close a stream before version check");
1673 goto end_no_session
;
1676 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &index_info
,
1677 sizeof(index_info
), 0);
1678 if (ret
< sizeof(index_info
)) {
1680 /* Orderly shutdown. Not necessary to print an error. */
1681 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1683 ERR("Relay didn't receive valid index struct size : %d", ret
);
1686 goto end_no_session
;
1689 net_seq_num
= be64toh(index_info
.net_seq_num
);
1692 stream
= relay_stream_find_by_id(be64toh(index_info
.relay_stream_id
));
1695 goto end_rcu_unlock
;
1698 /* Live beacon handling */
1699 if (index_info
.packet_size
== 0) {
1700 DBG("Received live beacon for stream %" PRIu64
, stream
->stream_handle
);
1703 * Only flag a stream inactive when it has already received data.
1705 if (stream
->total_index_received
> 0) {
1706 stream
->beacon_ts_end
= be64toh(index_info
.timestamp_end
);
1709 goto end_rcu_unlock
;
1711 stream
->beacon_ts_end
= -1ULL;
1714 index
= relay_index_find(stream
->stream_handle
, net_seq_num
, indexes_ht
);
1716 /* A successful creation will add the object to the HT. */
1717 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
1719 goto end_rcu_unlock
;
1724 copy_index_control_data(index
, &index_info
);
1726 if (index_created
) {
1728 * Try to add the relay index object to the hash table. If an object
1729 * already exist, destroy back the index created, set the data in this
1730 * object and write it on disk.
1732 relay_index_add(index
, indexes_ht
, &wr_index
);
1734 copy_index_control_data(wr_index
, &index_info
);
1738 /* The index already exists so write it on disk. */
1742 /* Do we have a writable ready index to write on disk. */
1744 /* Starting at 2.4, create the index file if none available. */
1745 if (cmd
->minor
>= 4 && stream
->index_fd
< 0) {
1746 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
1747 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
1748 stream
->tracefile_count_current
);
1750 goto end_rcu_unlock
;
1752 stream
->index_fd
= ret
;
1755 ret
= relay_index_write(wr_index
->fd
, wr_index
, indexes_ht
);
1757 goto end_rcu_unlock
;
1759 stream
->total_index_received
++;
1766 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1768 reply
.ret_code
= htobe32(LTTNG_OK
);
1770 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1772 ERR("Relay sending close index id reply");
1781 * Process the commands received on the control socket
1784 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1785 struct relay_command
*cmd
, struct relay_local_data
*ctx
)
1789 switch (be32toh(recv_hdr
->cmd
)) {
1790 case RELAYD_CREATE_SESSION
:
1791 ret
= relay_create_session(recv_hdr
, cmd
, ctx
->sessions_ht
);
1793 case RELAYD_ADD_STREAM
:
1794 ret
= relay_add_stream(recv_hdr
, cmd
, ctx
->sessions_ht
);
1796 case RELAYD_START_DATA
:
1797 ret
= relay_start(recv_hdr
, cmd
);
1799 case RELAYD_SEND_METADATA
:
1800 ret
= relay_recv_metadata(recv_hdr
, cmd
);
1802 case RELAYD_VERSION
:
1803 ret
= relay_send_version(recv_hdr
, cmd
, ctx
->sessions_ht
);
1805 case RELAYD_CLOSE_STREAM
:
1806 ret
= relay_close_stream(recv_hdr
, cmd
, ctx
->viewer_streams_ht
);
1808 case RELAYD_DATA_PENDING
:
1809 ret
= relay_data_pending(recv_hdr
, cmd
);
1811 case RELAYD_QUIESCENT_CONTROL
:
1812 ret
= relay_quiescent_control(recv_hdr
, cmd
);
1814 case RELAYD_BEGIN_DATA_PENDING
:
1815 ret
= relay_begin_data_pending(recv_hdr
, cmd
);
1817 case RELAYD_END_DATA_PENDING
:
1818 ret
= relay_end_data_pending(recv_hdr
, cmd
);
1820 case RELAYD_SEND_INDEX
:
1821 ret
= relay_recv_index(recv_hdr
, cmd
, indexes_ht
);
1823 case RELAYD_UPDATE_SYNC_INFO
:
1825 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
1826 relay_unknown_command(cmd
);
1836 * relay_process_data: Process the data received on the data socket
1839 int relay_process_data(struct relay_command
*cmd
,
1840 struct lttng_ht
*indexes_ht
, struct lttng_ht
*viewer_streams_ht
)
1842 int ret
= 0, rotate_index
= 0, index_created
= 0;
1843 struct relay_stream
*stream
;
1844 struct relay_index
*index
, *wr_index
= NULL
;
1845 struct lttcomm_relayd_data_hdr data_hdr
;
1846 uint64_t stream_id
, data_offset
;
1847 uint64_t net_seq_num
;
1850 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &data_hdr
,
1851 sizeof(struct lttcomm_relayd_data_hdr
), 0);
1854 /* Orderly shutdown. Not necessary to print an error. */
1855 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1857 ERR("Unable to receive data header on sock %d", cmd
->sock
->fd
);
1863 stream_id
= be64toh(data_hdr
.stream_id
);
1866 stream
= relay_stream_find_by_id(stream_id
);
1869 goto end_rcu_unlock
;
1872 data_size
= be32toh(data_hdr
.data_size
);
1873 if (data_buffer_size
< data_size
) {
1876 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1877 if (!tmp_data_ptr
) {
1878 ERR("Allocating data buffer");
1881 goto end_rcu_unlock
;
1883 data_buffer
= tmp_data_ptr
;
1884 data_buffer_size
= data_size
;
1886 memset(data_buffer
, 0, data_size
);
1888 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
1890 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
1891 data_size
, stream_id
, net_seq_num
);
1892 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
1895 /* Orderly shutdown. Not necessary to print an error. */
1896 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1899 goto end_rcu_unlock
;
1902 /* Check if a rotation is needed. */
1903 if (stream
->tracefile_size
> 0 &&
1904 (stream
->tracefile_size_current
+ data_size
) >
1905 stream
->tracefile_size
) {
1906 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
1907 stream
->tracefile_size
, stream
->tracefile_count
,
1908 relayd_uid
, relayd_gid
, stream
->fd
,
1909 &(stream
->tracefile_count_current
), &stream
->fd
);
1911 ERR("Rotating stream output file");
1912 goto end_rcu_unlock
;
1914 /* Reset current size because we just perform a stream rotation. */
1915 stream
->tracefile_size_current
= 0;
1919 /* Get data offset because we are about to update the index. */
1920 data_offset
= htobe64(stream
->tracefile_size_current
);
1923 * Lookup for an existing index for that stream id/sequence number. If on
1924 * exists, the control thread already received the data for it thus we need
1925 * to write it on disk.
1927 index
= relay_index_find(stream_id
, net_seq_num
, indexes_ht
);
1929 /* A successful creation will add the object to the HT. */
1930 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
1932 goto end_rcu_unlock
;
1937 if (rotate_index
|| stream
->index_fd
< 0) {
1938 index
->to_close_fd
= stream
->index_fd
;
1939 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
1940 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
1941 stream
->tracefile_count_current
);
1943 /* This will close the stream's index fd if one. */
1944 relay_index_free_safe(index
);
1945 goto end_rcu_unlock
;
1947 stream
->index_fd
= ret
;
1949 index
->fd
= stream
->index_fd
;
1950 index
->index_data
.offset
= data_offset
;
1952 if (index_created
) {
1954 * Try to add the relay index object to the hash table. If an object
1955 * already exist, destroy back the index created and set the data.
1957 relay_index_add(index
, indexes_ht
, &wr_index
);
1959 /* Copy back data from the created index. */
1960 wr_index
->fd
= index
->fd
;
1961 wr_index
->to_close_fd
= index
->to_close_fd
;
1962 wr_index
->index_data
.offset
= data_offset
;
1966 /* The index already exists so write it on disk. */
1970 /* Do we have a writable ready index to write on disk. */
1972 /* Starting at 2.4, create the index file if none available. */
1973 if (cmd
->minor
>= 4 && stream
->index_fd
< 0) {
1974 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
1975 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
1976 stream
->tracefile_count_current
);
1978 goto end_rcu_unlock
;
1980 stream
->index_fd
= ret
;
1983 ret
= relay_index_write(wr_index
->fd
, wr_index
, indexes_ht
);
1985 goto end_rcu_unlock
;
1987 stream
->total_index_received
++;
1991 ret
= write(stream
->fd
, data_buffer
, data_size
);
1992 } while (ret
< 0 && errno
== EINTR
);
1993 if (ret
< 0 || ret
!= data_size
) {
1994 ERR("Relay error writing data to file");
1996 goto end_rcu_unlock
;
1999 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
2000 ret
, stream
->stream_handle
);
2002 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
2004 goto end_rcu_unlock
;
2006 stream
->tracefile_size_current
+= data_size
+ be32toh(data_hdr
.padding_size
);
2008 stream
->prev_seq
= net_seq_num
;
2010 /* Check if we need to close the FD */
2011 if (close_stream_check(stream
)) {
2012 close_stream(stream
, viewer_streams_ht
, cmd
->ctf_traces_ht
);
2022 void relay_cleanup_poll_connection(struct lttng_poll_event
*events
, int pollfd
)
2026 lttng_poll_del(events
, pollfd
);
2028 ret
= close(pollfd
);
2030 ERR("Closing pollfd %d", pollfd
);
2035 int relay_add_connection(int fd
, struct lttng_poll_event
*events
,
2036 struct lttng_ht
*relay_connections_ht
)
2038 struct relay_command
*relay_connection
;
2041 relay_connection
= zmalloc(sizeof(struct relay_command
));
2042 if (relay_connection
== NULL
) {
2043 PERROR("Relay command zmalloc");
2047 ret
= read(fd
, relay_connection
, sizeof(struct relay_command
));
2048 } while (ret
< 0 && errno
== EINTR
);
2049 if (ret
< 0 || ret
< sizeof(struct relay_command
)) {
2050 PERROR("read relay cmd pipe");
2054 relay_connection
->ctf_traces_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_STRING
);
2055 if (!relay_connection
->ctf_traces_ht
) {
2059 lttng_ht_node_init_ulong(&relay_connection
->sock_n
,
2060 (unsigned long) relay_connection
->sock
->fd
);
2062 lttng_ht_add_unique_ulong(relay_connections_ht
,
2063 &relay_connection
->sock_n
);
2065 return lttng_poll_add(events
,
2066 relay_connection
->sock
->fd
,
2067 LPOLLIN
| LPOLLRDHUP
);
2070 free(relay_connection
);
2076 void deferred_free_connection(struct rcu_head
*head
)
2078 struct relay_command
*relay_connection
=
2079 caa_container_of(head
, struct relay_command
, rcu_node
);
2081 lttng_ht_destroy(relay_connection
->ctf_traces_ht
);
2082 lttcomm_destroy_sock(relay_connection
->sock
);
2083 free(relay_connection
);
2087 void relay_del_connection(struct lttng_ht
*relay_connections_ht
,
2088 struct lttng_ht_iter
*iter
, struct relay_command
*relay_connection
,
2089 struct lttng_ht
*sessions_ht
)
2093 ret
= lttng_ht_del(relay_connections_ht
, iter
);
2095 if (relay_connection
->type
== RELAY_CONTROL
) {
2096 relay_delete_session(relay_connection
, sessions_ht
);
2099 call_rcu(&relay_connection
->rcu_node
,
2100 deferred_free_connection
);
2104 * This thread does the actual work
2107 void *relay_thread_worker(void *data
)
2109 int ret
, err
= -1, last_seen_data_fd
= -1;
2111 struct relay_command
*relay_connection
;
2112 struct lttng_poll_event events
;
2113 struct lttng_ht
*relay_connections_ht
;
2114 struct lttng_ht_node_ulong
*node
;
2115 struct lttng_ht_iter iter
;
2116 struct lttcomm_relayd_hdr recv_hdr
;
2117 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
2118 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
2120 DBG("[thread] Relay worker started");
2122 rcu_register_thread();
2124 /* table of connections indexed on socket */
2125 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2126 if (!relay_connections_ht
) {
2127 goto relay_connections_ht_error
;
2130 /* Tables of received indexes indexed by index handle and net_seq_num. */
2131 indexes_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64
);
2133 goto indexes_ht_error
;
2136 ret
= create_thread_poll_set(&events
, 2);
2138 goto error_poll_create
;
2141 ret
= lttng_poll_add(&events
, relay_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2148 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2150 /* Infinite blocking call, waiting for transmission */
2151 DBG3("Relayd worker thread polling...");
2152 ret
= lttng_poll_wait(&events
, -1);
2155 * Restart interrupted system call.
2157 if (errno
== EINTR
) {
2166 * Process control. The control connection is prioritised so we don't
2167 * starve it with high throughout put tracing data on the data
2170 for (i
= 0; i
< nb_fd
; i
++) {
2171 /* Fetch once the poll data */
2172 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2173 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2175 /* Thread quit pipe has been closed. Killing thread. */
2176 ret
= check_thread_quit_pipe(pollfd
, revents
);
2182 /* Inspect the relay cmd pipe for new connection */
2183 if (pollfd
== relay_cmd_pipe
[0]) {
2184 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2185 ERR("Relay pipe error");
2187 } else if (revents
& LPOLLIN
) {
2188 DBG("Relay command received");
2189 ret
= relay_add_connection(relay_cmd_pipe
[0],
2190 &events
, relay_connections_ht
);
2195 } else if (revents
) {
2197 lttng_ht_lookup(relay_connections_ht
,
2198 (void *)((unsigned long) pollfd
),
2200 node
= lttng_ht_iter_get_node_ulong(&iter
);
2202 DBG2("Relay sock %d not found", pollfd
);
2206 relay_connection
= caa_container_of(node
,
2207 struct relay_command
, sock_n
);
2209 if (revents
& (LPOLLERR
)) {
2211 relay_cleanup_poll_connection(&events
, pollfd
);
2212 relay_del_connection(relay_connections_ht
,
2213 &iter
, relay_connection
, sessions_ht
);
2214 if (last_seen_data_fd
== pollfd
) {
2215 last_seen_data_fd
= last_notdel_data_fd
;
2217 } else if (revents
& (LPOLLHUP
| LPOLLRDHUP
)) {
2218 DBG("Socket %d hung up", pollfd
);
2219 relay_cleanup_poll_connection(&events
, pollfd
);
2220 relay_del_connection(relay_connections_ht
,
2221 &iter
, relay_connection
, sessions_ht
);
2222 if (last_seen_data_fd
== pollfd
) {
2223 last_seen_data_fd
= last_notdel_data_fd
;
2225 } else if (revents
& LPOLLIN
) {
2226 /* control socket */
2227 if (relay_connection
->type
== RELAY_CONTROL
) {
2228 ret
= relay_connection
->sock
->ops
->recvmsg(
2229 relay_connection
->sock
, &recv_hdr
,
2230 sizeof(struct lttcomm_relayd_hdr
), 0);
2231 /* connection closed */
2233 relay_cleanup_poll_connection(&events
, pollfd
);
2234 relay_del_connection(relay_connections_ht
,
2235 &iter
, relay_connection
, sessions_ht
);
2236 DBG("Control connection closed with %d", pollfd
);
2238 if (relay_connection
->session
) {
2239 DBG2("Relay worker receiving data for session : %" PRIu64
,
2240 relay_connection
->session
->id
);
2242 ret
= relay_process_control(&recv_hdr
,
2243 relay_connection
, relay_ctx
);
2245 /* Clear the session on error. */
2246 relay_cleanup_poll_connection(&events
, pollfd
);
2247 relay_del_connection(relay_connections_ht
,
2248 &iter
, relay_connection
, sessions_ht
);
2249 DBG("Connection closed with %d", pollfd
);
2255 * Flag the last seen data fd not deleted. It will be
2256 * used as the last seen fd if any fd gets deleted in
2259 last_notdel_data_fd
= pollfd
;
2267 * The last loop handled a control request, go back to poll to make
2268 * sure we prioritise the control socket.
2274 if (last_seen_data_fd
>= 0) {
2275 for (i
= 0; i
< nb_fd
; i
++) {
2276 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2277 if (last_seen_data_fd
== pollfd
) {
2284 /* Process data connection. */
2285 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2286 /* Fetch the poll data. */
2287 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2288 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2290 /* Skip the command pipe. It's handled in the first loop. */
2291 if (pollfd
== relay_cmd_pipe
[0]) {
2297 lttng_ht_lookup(relay_connections_ht
,
2298 (void *)((unsigned long) pollfd
),
2300 node
= lttng_ht_iter_get_node_ulong(&iter
);
2302 /* Skip it. Might be removed before. */
2306 relay_connection
= caa_container_of(node
,
2307 struct relay_command
, sock_n
);
2309 if (revents
& LPOLLIN
) {
2310 if (relay_connection
->type
!= RELAY_DATA
) {
2314 ret
= relay_process_data(relay_connection
, indexes_ht
,
2315 relay_ctx
->viewer_streams_ht
);
2316 /* connection closed */
2318 relay_cleanup_poll_connection(&events
, pollfd
);
2319 relay_del_connection(relay_connections_ht
,
2320 &iter
, relay_connection
, sessions_ht
);
2321 DBG("Data connection closed with %d", pollfd
);
2323 * Every goto restart call sets the last seen fd where
2324 * here we don't really care since we gracefully
2325 * continue the loop after the connection is deleted.
2328 /* Keep last seen port. */
2329 last_seen_data_fd
= pollfd
;
2337 last_seen_data_fd
= -1;
2342 lttng_poll_clean(&events
);
2344 /* empty the hash table and free the memory */
2346 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, node
, node
) {
2347 node
= lttng_ht_iter_get_node_ulong(&iter
);
2349 relay_connection
= caa_container_of(node
,
2350 struct relay_command
, sock_n
);
2351 relay_del_connection(relay_connections_ht
,
2352 &iter
, relay_connection
, sessions_ht
);
2357 struct relay_index
*index
;
2358 cds_lfht_for_each_entry(indexes_ht
->ht
, &iter
.iter
, index
, index_n
.node
) {
2359 relay_index_delete(index
, indexes_ht
);
2361 lttng_ht_destroy(indexes_ht
);
2365 lttng_ht_destroy(relay_connections_ht
);
2366 relay_connections_ht_error
:
2367 /* Close relay cmd pipes */
2368 utils_close_pipe(relay_cmd_pipe
);
2370 DBG("Thread exited with error");
2372 DBG("Worker thread cleanup complete");
2375 rcu_unregister_thread();
2380 * Create the relay command pipe to wake thread_manage_apps.
2381 * Closed in cleanup().
2383 static int create_relay_cmd_pipe(void)
2387 ret
= utils_create_pipe_cloexec(relay_cmd_pipe
);
2395 int main(int argc
, char **argv
)
2399 struct relay_local_data
*relay_ctx
;
2401 /* Create thread quit pipe */
2402 if ((ret
= init_thread_quit_pipe()) < 0) {
2406 /* Parse arguments */
2408 if ((ret
= parse_args(argc
, argv
)) < 0) {
2412 if ((ret
= set_signal_handler()) < 0) {
2416 /* Try to create directory if -o, --output is specified. */
2417 if (opt_output_path
) {
2418 if (*opt_output_path
!= '/') {
2419 ERR("Please specify an absolute path for -o, --output PATH");
2423 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2425 ERR("Unable to create %s", opt_output_path
);
2439 /* We need those values for the file/dir creation. */
2440 relayd_uid
= getuid();
2441 relayd_gid
= getgid();
2443 /* Check if daemon is UID = 0 */
2444 if (relayd_uid
== 0) {
2445 if (control_uri
->port
< 1024 || data_uri
->port
< 1024) {
2446 ERR("Need to be root to use ports < 1024");
2452 /* Setup the thread apps communication pipe. */
2453 if ((ret
= create_relay_cmd_pipe()) < 0) {
2457 /* Init relay command queue. */
2458 cds_wfq_init(&relay_cmd_queue
.queue
);
2460 /* Set up max poll set size */
2461 lttng_poll_set_max_size();
2463 /* Initialize communication library */
2466 relay_ctx
= zmalloc(sizeof(struct relay_local_data
));
2468 PERROR("relay_ctx");
2472 /* tables of sessions indexed by session ID */
2473 relay_ctx
->sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2474 if (!relay_ctx
->sessions_ht
) {
2475 goto exit_relay_ctx_sessions
;
2478 /* tables of streams indexed by stream ID */
2479 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2480 if (!relay_streams_ht
) {
2481 goto exit_relay_ctx_streams
;
2484 /* tables of streams indexed by stream ID */
2485 relay_ctx
->viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2486 if (!relay_ctx
->viewer_streams_ht
) {
2487 goto exit_relay_ctx_viewer_streams
;
2490 /* Setup the dispatcher thread */
2491 ret
= pthread_create(&dispatcher_thread
, NULL
,
2492 relay_thread_dispatcher
, (void *) NULL
);
2494 PERROR("pthread_create dispatcher");
2495 goto exit_dispatcher
;
2498 /* Setup the worker thread */
2499 ret
= pthread_create(&worker_thread
, NULL
,
2500 relay_thread_worker
, (void *) relay_ctx
);
2502 PERROR("pthread_create worker");
2506 /* Setup the listener thread */
2507 ret
= pthread_create(&listener_thread
, NULL
,
2508 relay_thread_listener
, (void *) NULL
);
2510 PERROR("pthread_create listener");
2514 ret
= live_start_threads(live_uri
, relay_ctx
);
2516 ERR("Starting live viewer threads");
2520 ret
= pthread_join(listener_thread
, &status
);
2522 PERROR("pthread_join");
2523 goto error
; /* join error, exit without cleanup */
2527 ret
= pthread_join(worker_thread
, &status
);
2529 PERROR("pthread_join");
2530 goto error
; /* join error, exit without cleanup */
2534 ret
= pthread_join(dispatcher_thread
, &status
);
2536 PERROR("pthread_join");
2537 goto error
; /* join error, exit without cleanup */
2539 lttng_ht_destroy(relay_ctx
->viewer_streams_ht
);
2541 exit_relay_ctx_viewer_streams
:
2542 lttng_ht_destroy(relay_streams_ht
);
2544 exit_relay_ctx_streams
:
2545 lttng_ht_destroy(relay_ctx
->sessions_ht
);
2547 exit_relay_ctx_sessions
:
2551 live_stop_threads();