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"
60 #include "health-relayd.h"
62 /* command line options */
63 char *opt_output_path
;
64 static int opt_daemon
;
65 static struct lttng_uri
*control_uri
;
66 static struct lttng_uri
*data_uri
;
67 static struct lttng_uri
*live_uri
;
71 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
74 * Quit pipe for all threads. This permits a single cancellation point
75 * for all threads when receiving an event on the pipe.
77 static int thread_quit_pipe
[2] = { -1, -1 };
80 * This pipe is used to inform the worker thread that a command is queued and
81 * ready to be processed.
83 static int relay_cmd_pipe
[2] = { -1, -1 };
85 /* Shared between threads */
86 static int dispatch_thread_exit
;
88 static pthread_t listener_thread
;
89 static pthread_t dispatcher_thread
;
90 static pthread_t worker_thread
;
91 static pthread_t health_thread
;
93 static uint64_t last_relay_stream_id
;
94 static uint64_t last_relay_session_id
;
97 * Relay command queue.
99 * The relay_thread_listener and relay_thread_dispatcher communicate with this
102 static struct relay_cmd_queue relay_cmd_queue
;
104 /* buffer allocated at startup, used to store the trace data */
105 static char *data_buffer
;
106 static unsigned int data_buffer_size
;
108 /* We need those values for the file/dir creation. */
109 static uid_t relayd_uid
;
110 static gid_t relayd_gid
;
112 /* Global relay stream hash table. */
113 struct lttng_ht
*relay_streams_ht
;
115 /* Global relay viewer stream hash table. */
116 struct lttng_ht
*viewer_streams_ht
;
118 /* Global hash table that stores relay index object. */
119 struct lttng_ht
*indexes_ht
;
121 /* Relayd health monitoring */
122 struct health_app
*health_relayd
;
125 * usage function on stderr
130 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
131 fprintf(stderr
, " -h, --help Display this usage.\n");
132 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
133 fprintf(stderr
, " -C, --control-port URL Control port listening.\n");
134 fprintf(stderr
, " -D, --data-port URL Data port listening.\n");
135 fprintf(stderr
, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
136 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
137 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
141 int parse_args(int argc
, char **argv
)
145 char *default_address
;
147 static struct option long_options
[] = {
148 { "control-port", 1, 0, 'C', },
149 { "data-port", 1, 0, 'D', },
150 { "daemonize", 0, 0, 'd', },
151 { "group", 1, 0, 'g', },
152 { "help", 0, 0, 'h', },
153 { "output", 1, 0, 'o', },
154 { "verbose", 0, 0, 'v', },
159 int option_index
= 0;
160 c
= getopt_long(argc
, argv
, "dhv" "C:D:o:g:",
161 long_options
, &option_index
);
168 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
170 fprintf(stderr
, " with arg %s\n", optarg
);
174 ret
= uri_parse(optarg
, &control_uri
);
176 ERR("Invalid control URI specified");
179 if (control_uri
->port
== 0) {
180 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
184 ret
= uri_parse(optarg
, &data_uri
);
186 ERR("Invalid data URI specified");
189 if (data_uri
->port
== 0) {
190 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
197 tracing_group_name
= optarg
;
203 ret
= asprintf(&opt_output_path
, "%s", optarg
);
205 PERROR("asprintf opt_output_path");
210 /* Verbose level can increase using multiple -v */
211 lttng_opt_verbose
+= 1;
214 /* Unknown option or other error.
215 * Error is printed by getopt, just return */
221 /* assign default values */
222 if (control_uri
== NULL
) {
223 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
224 DEFAULT_NETWORK_CONTROL_PORT
);
226 PERROR("asprintf default data address");
230 ret
= uri_parse(default_address
, &control_uri
);
231 free(default_address
);
233 ERR("Invalid control URI specified");
237 if (data_uri
== NULL
) {
238 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
239 DEFAULT_NETWORK_DATA_PORT
);
241 PERROR("asprintf default data address");
245 ret
= uri_parse(default_address
, &data_uri
);
246 free(default_address
);
248 ERR("Invalid data URI specified");
252 if (live_uri
== NULL
) {
253 ret
= asprintf(&default_address
, "tcp://0.0.0.0:%d",
254 DEFAULT_NETWORK_VIEWER_PORT
);
256 PERROR("asprintf default viewer control address");
260 ret
= uri_parse(default_address
, &live_uri
);
261 free(default_address
);
263 ERR("Invalid viewer control URI specified");
280 /* free the dynamically allocated opt_output_path */
281 free(opt_output_path
);
283 /* Close thread quit pipes */
284 utils_close_pipe(thread_quit_pipe
);
286 uri_free(control_uri
);
291 * Write to writable pipe used to notify a thread.
294 int notify_thread_pipe(int wpipe
)
298 ret
= lttng_write(wpipe
, "!", 1);
300 PERROR("write poll pipe");
306 static void notify_health_quit_pipe(int *pipe
)
310 ret
= lttng_write(pipe
[1], "4", 1);
312 PERROR("write relay health quit");
317 * Stop all threads by closing the thread quit pipe.
320 void stop_threads(void)
324 /* Stopping all threads */
325 DBG("Terminating all threads");
326 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
328 ERR("write error on thread quit pipe");
331 notify_health_quit_pipe(health_quit_pipe
);
333 /* Dispatch thread */
334 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
335 futex_nto1_wake(&relay_cmd_queue
.futex
);
339 * Signal handler for the daemon
341 * Simply stop all worker threads, leaving main() return gracefully after
342 * joining all threads and calling cleanup().
345 void sighandler(int sig
)
349 DBG("SIGPIPE caught");
352 DBG("SIGINT caught");
356 DBG("SIGTERM caught");
365 * Setup signal handler for :
366 * SIGINT, SIGTERM, SIGPIPE
369 int set_signal_handler(void)
375 if ((ret
= sigemptyset(&sigset
)) < 0) {
376 PERROR("sigemptyset");
380 sa
.sa_handler
= sighandler
;
383 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
388 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
393 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
398 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
404 * Init thread quit pipe.
406 * Return -1 on error or 0 if all pipes are created.
409 int init_thread_quit_pipe(void)
413 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
419 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
422 int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
426 if (events
== NULL
|| size
== 0) {
431 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
437 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
449 * Check if the thread quit pipe was triggered.
451 * Return 1 if it was triggered else 0;
454 int check_thread_quit_pipe(int fd
, uint32_t events
)
456 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
464 * Create and init socket from uri.
467 struct lttcomm_sock
*relay_init_sock(struct lttng_uri
*uri
)
470 struct lttcomm_sock
*sock
= NULL
;
472 sock
= lttcomm_alloc_sock_from_uri(uri
);
474 ERR("Allocating socket");
478 ret
= lttcomm_create_sock(sock
);
482 DBG("Listening on sock %d", sock
->fd
);
484 ret
= sock
->ops
->bind(sock
);
489 ret
= sock
->ops
->listen(sock
, -1);
499 lttcomm_destroy_sock(sock
);
505 * Return nonzero if stream needs to be closed.
508 int close_stream_check(struct relay_stream
*stream
)
511 if (stream
->close_flag
&& stream
->prev_seq
== stream
->last_net_seq_num
) {
513 * We are about to close the stream so set the data pending flag to 1
514 * which will make the end data pending command skip the stream which
515 * is now closed and ready. Note that after proceeding to a file close,
516 * the written file is ready for reading.
518 stream
->data_pending_check_done
= 1;
525 * This thread manages the listening for new connections on the network
528 void *relay_thread_listener(void *data
)
530 int i
, ret
, pollfd
, err
= -1;
532 uint32_t revents
, nb_fd
;
533 struct lttng_poll_event events
;
534 struct lttcomm_sock
*control_sock
, *data_sock
;
536 DBG("[thread] Relay listener started");
538 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
540 health_code_update();
542 control_sock
= relay_init_sock(control_uri
);
544 goto error_sock_control
;
547 data_sock
= relay_init_sock(data_uri
);
549 goto error_sock_relay
;
553 * Pass 3 as size here for the thread quit pipe, control and data socket.
555 ret
= create_thread_poll_set(&events
, 3);
557 goto error_create_poll
;
560 /* Add the control socket */
561 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
566 /* Add the data socket */
567 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
573 health_code_update();
575 DBG("Listener accepting connections");
579 ret
= lttng_poll_wait(&events
, -1);
583 * Restart interrupted system call.
585 if (errno
== EINTR
) {
593 DBG("Relay new connection received");
594 for (i
= 0; i
< nb_fd
; i
++) {
595 health_code_update();
597 /* Fetch once the poll data */
598 revents
= LTTNG_POLL_GETEV(&events
, i
);
599 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
601 /* Thread quit pipe has been closed. Killing thread. */
602 ret
= check_thread_quit_pipe(pollfd
, revents
);
608 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
609 ERR("socket poll error");
611 } else if (revents
& LPOLLIN
) {
613 * Get allocated in this thread,
614 * enqueued to a global queue, dequeued
615 * and freed in the worker thread.
617 struct relay_command
*relay_cmd
;
618 struct lttcomm_sock
*newsock
;
620 relay_cmd
= zmalloc(sizeof(struct relay_command
));
621 if (relay_cmd
== NULL
) {
622 PERROR("relay command zmalloc");
626 if (pollfd
== data_sock
->fd
) {
627 newsock
= data_sock
->ops
->accept(data_sock
);
629 PERROR("accepting data sock");
633 relay_cmd
->type
= RELAY_DATA
;
634 DBG("Relay data connection accepted, socket %d", newsock
->fd
);
636 assert(pollfd
== control_sock
->fd
);
637 newsock
= control_sock
->ops
->accept(control_sock
);
639 PERROR("accepting control sock");
643 relay_cmd
->type
= RELAY_CONTROL
;
644 DBG("Relay control connection accepted, socket %d", newsock
->fd
);
646 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
,
649 PERROR("setsockopt inet");
650 lttcomm_destroy_sock(newsock
);
654 relay_cmd
->sock
= newsock
;
656 * Lock free enqueue the request.
658 cds_wfq_enqueue(&relay_cmd_queue
.queue
, &relay_cmd
->node
);
661 * Wake the dispatch queue futex. Implicit memory
662 * barrier with the exchange in cds_wfq_enqueue.
664 futex_nto1_wake(&relay_cmd_queue
.futex
);
672 lttng_poll_clean(&events
);
674 if (data_sock
->fd
>= 0) {
675 ret
= data_sock
->ops
->close(data_sock
);
680 lttcomm_destroy_sock(data_sock
);
682 if (control_sock
->fd
>= 0) {
683 ret
= control_sock
->ops
->close(control_sock
);
688 lttcomm_destroy_sock(control_sock
);
692 ERR("Health error occurred in %s", __func__
);
694 health_unregister(health_relayd
);
695 DBG("Relay listener thread cleanup complete");
701 * This thread manages the dispatching of the requests to worker threads
704 void *relay_thread_dispatcher(void *data
)
708 struct cds_wfq_node
*node
;
709 struct relay_command
*relay_cmd
= NULL
;
711 DBG("[thread] Relay dispatcher started");
713 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
715 health_code_update();
717 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
718 health_code_update();
720 /* Atomically prepare the queue futex */
721 futex_nto1_prepare(&relay_cmd_queue
.futex
);
724 health_code_update();
726 /* Dequeue commands */
727 node
= cds_wfq_dequeue_blocking(&relay_cmd_queue
.queue
);
729 DBG("Woken up but nothing in the relay command queue");
730 /* Continue thread execution */
734 relay_cmd
= caa_container_of(node
, struct relay_command
, node
);
735 DBG("Dispatching request waiting on sock %d", relay_cmd
->sock
->fd
);
738 * Inform worker thread of the new request. This
739 * call is blocking so we can be assured that the data will be read
740 * at some point in time or wait to the end of the world :)
742 ret
= lttng_write(relay_cmd_pipe
[1], relay_cmd
,
743 sizeof(struct relay_command
));
745 if (ret
< sizeof(struct relay_command
)) {
746 PERROR("write cmd pipe");
749 } while (node
!= NULL
);
751 /* Futex wait on queue. Blocking call on futex() */
753 futex_nto1_wait(&relay_cmd_queue
.futex
);
757 /* Normal exit, no error */
763 ERR("Health error occurred in %s", __func__
);
765 health_unregister(health_relayd
);
766 DBG("Dispatch thread dying");
772 * Get stream from stream id.
773 * Need to be called with RCU read-side lock held.
775 struct relay_stream
*relay_stream_find_by_id(uint64_t stream_id
)
777 struct lttng_ht_node_ulong
*node
;
778 struct lttng_ht_iter iter
;
779 struct relay_stream
*ret
;
781 lttng_ht_lookup(relay_streams_ht
,
782 (void *)((unsigned long) stream_id
),
784 node
= lttng_ht_iter_get_node_ulong(&iter
);
786 DBG("Relay stream %" PRIu64
" not found", stream_id
);
791 ret
= caa_container_of(node
, struct relay_stream
, stream_n
);
798 void deferred_free_stream(struct rcu_head
*head
)
800 struct relay_stream
*stream
=
801 caa_container_of(head
, struct relay_stream
, rcu_node
);
803 ctf_trace_try_destroy(stream
->ctf_trace
);
805 free(stream
->path_name
);
806 free(stream
->channel_name
);
811 void deferred_free_session(struct rcu_head
*head
)
813 struct relay_session
*session
=
814 caa_container_of(head
, struct relay_session
, rcu_node
);
819 * Close a given stream. The stream is freed using a call RCU.
821 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
822 * BEFORE the stream lock MUST be acquired.
824 static void destroy_stream(struct relay_stream
*stream
)
827 struct relay_viewer_stream
*vstream
;
828 struct lttng_ht_iter iter
;
832 delret
= close(stream
->fd
);
834 PERROR("close stream");
837 if (stream
->index_fd
>= 0) {
838 delret
= close(stream
->index_fd
);
840 PERROR("close stream index_fd");
844 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
);
847 * Set the last good value into the viewer stream. This is done
848 * right before the stream gets deleted from the hash table. The
849 * lookup failure on the live thread side of a stream indicates
850 * that the viewer stream index received value should be used.
852 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
853 vstream
->total_index_received
= stream
->total_index_received
;
854 vstream
->close_write_flag
= 1;
855 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
858 /* Cleanup index of that stream. */
859 relay_index_destroy_by_stream_id(stream
->stream_handle
);
861 iter
.iter
.node
= &stream
->stream_n
.node
;
862 delret
= lttng_ht_del(relay_streams_ht
, &iter
);
864 iter
.iter
.node
= &stream
->ctf_trace_node
.node
;
865 delret
= lttng_ht_del(stream
->ctf_traces_ht
, &iter
);
867 call_rcu(&stream
->rcu_node
, deferred_free_stream
);
868 DBG("Closed tracefile %d from close stream", stream
->fd
);
872 * relay_delete_session: Free all memory associated with a session and
876 void relay_delete_session(struct relay_command
*cmd
,
877 struct lttng_ht
*sessions_ht
)
879 struct lttng_ht_iter iter
;
880 struct lttng_ht_node_ulong
*node
;
881 struct relay_stream
*stream
;
888 DBG("Relay deleting session %" PRIu64
, cmd
->session
->id
);
891 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, node
, node
) {
892 node
= lttng_ht_iter_get_node_ulong(&iter
);
896 stream
= caa_container_of(node
, struct relay_stream
, stream_n
);
897 if (stream
->session
== cmd
->session
) {
898 destroy_stream(stream
);
899 cmd
->session
->stream_count
--;
900 assert(cmd
->session
->stream_count
>= 0);
904 /* Make this session not visible anymore. */
905 iter
.iter
.node
= &cmd
->session
->session_n
.node
;
906 ret
= lttng_ht_del(sessions_ht
, &iter
);
908 call_rcu(&cmd
->session
->rcu_node
, deferred_free_session
);
913 * Copy index data from the control port to a given index object.
915 static void copy_index_control_data(struct relay_index
*index
,
916 struct lttcomm_relayd_index
*data
)
922 * The index on disk is encoded in big endian, so we don't need to convert
923 * the data received on the network. The data_offset value is NEVER
924 * modified here and is updated by the data thread.
926 index
->index_data
.packet_size
= data
->packet_size
;
927 index
->index_data
.content_size
= data
->content_size
;
928 index
->index_data
.timestamp_begin
= data
->timestamp_begin
;
929 index
->index_data
.timestamp_end
= data
->timestamp_end
;
930 index
->index_data
.events_discarded
= data
->events_discarded
;
931 index
->index_data
.stream_id
= data
->stream_id
;
935 * Handle the RELAYD_CREATE_SESSION command.
937 * On success, send back the session id or else return a negative value.
940 int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
941 struct relay_command
*cmd
,
942 struct lttng_ht
*sessions_ht
)
944 int ret
= 0, send_ret
;
945 struct relay_session
*session
;
946 struct lttcomm_relayd_status_session reply
;
951 memset(&reply
, 0, sizeof(reply
));
953 session
= zmalloc(sizeof(struct relay_session
));
954 if (session
== NULL
) {
955 PERROR("relay session zmalloc");
960 session
->id
= ++last_relay_session_id
;
961 session
->sock
= cmd
->sock
;
962 session
->minor
= cmd
->minor
;
963 session
->major
= cmd
->major
;
964 cmd
->session
= session
;
966 reply
.session_id
= htobe64(session
->id
);
968 switch (cmd
->minor
) {
969 case 4: /* LTTng sessiond 2.4 */
971 ret
= cmd_create_session_2_4(cmd
, session
);
975 lttng_ht_node_init_ulong(&session
->session_n
,
976 (unsigned long) session
->id
);
977 lttng_ht_add_unique_ulong(sessions_ht
,
978 &session
->session_n
);
980 DBG("Created session %" PRIu64
, session
->id
);
984 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
986 reply
.ret_code
= htobe32(LTTNG_OK
);
989 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
991 ERR("Relayd sending session id");
999 * relay_add_stream: allocate a new stream for a session
1002 int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1003 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1005 struct relay_session
*session
= cmd
->session
;
1006 struct relay_stream
*stream
= NULL
;
1007 struct lttcomm_relayd_status_stream reply
;
1010 if (!session
|| cmd
->version_check_done
== 0) {
1011 ERR("Trying to add a stream before version check");
1013 goto end_no_session
;
1016 stream
= zmalloc(sizeof(struct relay_stream
));
1017 if (stream
== NULL
) {
1018 PERROR("relay stream zmalloc");
1020 goto end_no_session
;
1023 switch (cmd
->minor
) {
1024 case 1: /* LTTng sessiond 2.1 */
1025 ret
= cmd_recv_stream_2_1(cmd
, stream
);
1027 case 2: /* LTTng sessiond 2.2 */
1029 ret
= cmd_recv_stream_2_2(cmd
, stream
);
1033 goto err_free_stream
;
1037 stream
->stream_handle
= ++last_relay_stream_id
;
1038 stream
->prev_seq
= -1ULL;
1039 stream
->session
= session
;
1040 stream
->index_fd
= -1;
1041 stream
->read_index_fd
= -1;
1042 stream
->ctf_trace
= NULL
;
1043 pthread_mutex_init(&stream
->lock
, NULL
);
1045 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
);
1047 ERR("relay creating output directory");
1052 * No need to use run_as API here because whatever we receives, the relayd
1053 * uses its own credentials for the stream files.
1055 ret
= utils_create_stream_file(stream
->path_name
, stream
->channel_name
,
1056 stream
->tracefile_size
, 0, relayd_uid
, relayd_gid
, NULL
);
1058 ERR("Create output file");
1062 if (stream
->tracefile_size
) {
1063 DBG("Tracefile %s/%s_0 created", stream
->path_name
, stream
->channel_name
);
1065 DBG("Tracefile %s/%s created", stream
->path_name
, stream
->channel_name
);
1068 if (!strncmp(stream
->channel_name
, DEFAULT_METADATA_NAME
, NAME_MAX
)) {
1069 stream
->metadata_flag
= 1;
1071 * When we receive a new metadata stream, we create a new
1072 * ctf_trace and we assign this ctf_trace to all streams with
1075 * If later on we receive a new stream for the same ctf_trace,
1076 * we copy the information from the first hit in the HT to the
1079 stream
->ctf_trace
= ctf_trace_create();
1080 if (!stream
->ctf_trace
) {
1084 stream
->ctf_trace
->refcount
++;
1085 stream
->ctf_trace
->metadata_stream
= stream
;
1087 ctf_trace_assign(cmd
->ctf_traces_ht
, stream
);
1088 stream
->ctf_traces_ht
= cmd
->ctf_traces_ht
;
1090 lttng_ht_node_init_ulong(&stream
->stream_n
,
1091 (unsigned long) stream
->stream_handle
);
1092 lttng_ht_add_unique_ulong(relay_streams_ht
,
1095 lttng_ht_node_init_str(&stream
->ctf_trace_node
, stream
->path_name
);
1096 lttng_ht_add_str(cmd
->ctf_traces_ht
, &stream
->ctf_trace_node
);
1097 session
->stream_count
++;
1099 DBG("Relay new stream added %s with ID %" PRIu64
, stream
->channel_name
,
1100 stream
->stream_handle
);
1103 reply
.handle
= htobe64(stream
->stream_handle
);
1104 /* send the session id to the client or a negative return code on error */
1106 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1107 /* stream was not properly added to the ht, so free it */
1110 reply
.ret_code
= htobe32(LTTNG_OK
);
1113 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1114 sizeof(struct lttcomm_relayd_status_stream
), 0);
1116 ERR("Relay sending stream id");
1125 free(stream
->path_name
);
1126 free(stream
->channel_name
);
1132 * relay_close_stream: close a specific stream
1135 int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1136 struct relay_command
*cmd
)
1139 struct relay_session
*session
= cmd
->session
;
1140 struct lttcomm_relayd_close_stream stream_info
;
1141 struct lttcomm_relayd_generic_reply reply
;
1142 struct relay_stream
*stream
;
1144 DBG("Close stream received");
1146 if (!session
|| cmd
->version_check_done
== 0) {
1147 ERR("Trying to close a stream before version check");
1149 goto end_no_session
;
1152 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &stream_info
,
1153 sizeof(struct lttcomm_relayd_close_stream
), 0);
1154 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1156 /* Orderly shutdown. Not necessary to print an error. */
1157 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1159 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1162 goto end_no_session
;
1166 stream
= relay_stream_find_by_id(be64toh(stream_info
.stream_id
));
1172 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1173 stream
->close_flag
= 1;
1174 session
->stream_count
--;
1175 assert(session
->stream_count
>= 0);
1177 if (close_stream_check(stream
)) {
1178 destroy_stream(stream
);
1185 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1187 reply
.ret_code
= htobe32(LTTNG_OK
);
1189 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1190 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1192 ERR("Relay sending stream id");
1201 * relay_unknown_command: send -1 if received unknown command
1204 void relay_unknown_command(struct relay_command
*cmd
)
1206 struct lttcomm_relayd_generic_reply reply
;
1209 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1210 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1211 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1213 ERR("Relay sending unknown command");
1218 * relay_start: send an acknowledgment to the client to tell if we are
1219 * ready to receive data. We are ready if a session is established.
1222 int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1223 struct relay_command
*cmd
)
1225 int ret
= htobe32(LTTNG_OK
);
1226 struct lttcomm_relayd_generic_reply reply
;
1227 struct relay_session
*session
= cmd
->session
;
1230 DBG("Trying to start the streaming without a session established");
1231 ret
= htobe32(LTTNG_ERR_UNK
);
1234 reply
.ret_code
= ret
;
1235 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1236 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1238 ERR("Relay sending start ack");
1245 * Append padding to the file pointed by the file descriptor fd.
1247 static int write_padding_to_file(int fd
, uint32_t size
)
1256 zeros
= zmalloc(size
);
1257 if (zeros
== NULL
) {
1258 PERROR("zmalloc zeros for padding");
1263 ret
= lttng_write(fd
, zeros
, size
);
1265 PERROR("write padding to file");
1275 * relay_recv_metadata: receive the metada for the session.
1278 int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1279 struct relay_command
*cmd
)
1281 int ret
= htobe32(LTTNG_OK
);
1283 struct relay_session
*session
= cmd
->session
;
1284 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1285 struct relay_stream
*metadata_stream
;
1286 uint64_t data_size
, payload_size
;
1289 ERR("Metadata sent before version check");
1294 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1295 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1296 ERR("Incorrect data size");
1300 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1302 if (data_buffer_size
< data_size
) {
1303 /* In case the realloc fails, we can free the memory */
1306 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1307 if (!tmp_data_ptr
) {
1308 ERR("Allocating data buffer");
1313 data_buffer
= tmp_data_ptr
;
1314 data_buffer_size
= data_size
;
1316 memset(data_buffer
, 0, data_size
);
1317 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1318 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
1319 if (ret
< 0 || ret
!= data_size
) {
1321 /* Orderly shutdown. Not necessary to print an error. */
1322 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1324 ERR("Relay didn't receive the whole metadata");
1329 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1332 metadata_stream
= relay_stream_find_by_id(
1333 be64toh(metadata_struct
->stream_id
));
1334 if (!metadata_stream
) {
1339 size_ret
= lttng_write(metadata_stream
->fd
, metadata_struct
->payload
,
1341 if (size_ret
< payload_size
) {
1342 ERR("Relay error writing metadata on file");
1347 ret
= write_padding_to_file(metadata_stream
->fd
,
1348 be32toh(metadata_struct
->padding_size
));
1352 metadata_stream
->ctf_trace
->metadata_received
+=
1353 payload_size
+ be32toh(metadata_struct
->padding_size
);
1355 DBG2("Relay metadata written");
1364 * relay_send_version: send relayd version number
1367 int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1368 struct relay_command
*cmd
, struct lttng_ht
*sessions_ht
)
1371 struct lttcomm_relayd_version reply
, msg
;
1375 cmd
->version_check_done
= 1;
1377 /* Get version from the other side. */
1378 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1379 if (ret
< 0 || ret
!= sizeof(msg
)) {
1381 /* Orderly shutdown. Not necessary to print an error. */
1382 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1384 ERR("Relay failed to receive the version values.");
1390 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1391 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1393 /* Major versions must be the same */
1394 if (reply
.major
!= be32toh(msg
.major
)) {
1395 DBG("Incompatible major versions (%u vs %u), deleting session",
1396 reply
.major
, be32toh(msg
.major
));
1397 relay_delete_session(cmd
, sessions_ht
);
1402 cmd
->major
= reply
.major
;
1403 /* We adapt to the lowest compatible version */
1404 if (reply
.minor
<= be32toh(msg
.minor
)) {
1405 cmd
->minor
= reply
.minor
;
1407 cmd
->minor
= be32toh(msg
.minor
);
1410 reply
.major
= htobe32(reply
.major
);
1411 reply
.minor
= htobe32(reply
.minor
);
1412 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
,
1413 sizeof(struct lttcomm_relayd_version
), 0);
1415 ERR("Relay sending version");
1418 DBG("Version check done using protocol %u.%u", cmd
->major
,
1426 * Check for data pending for a given stream id from the session daemon.
1429 int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1430 struct relay_command
*cmd
)
1432 struct relay_session
*session
= cmd
->session
;
1433 struct lttcomm_relayd_data_pending msg
;
1434 struct lttcomm_relayd_generic_reply reply
;
1435 struct relay_stream
*stream
;
1437 uint64_t last_net_seq_num
, stream_id
;
1439 DBG("Data pending command received");
1441 if (!session
|| cmd
->version_check_done
== 0) {
1442 ERR("Trying to check for data before version check");
1444 goto end_no_session
;
1447 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1448 if (ret
< sizeof(msg
)) {
1450 /* Orderly shutdown. Not necessary to print an error. */
1451 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1453 ERR("Relay didn't receive valid data_pending struct size : %d",
1457 goto end_no_session
;
1460 stream_id
= be64toh(msg
.stream_id
);
1461 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1464 stream
= relay_stream_find_by_id(stream_id
);
1465 if (stream
== NULL
) {
1470 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1471 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1474 /* Avoid wrapping issue */
1475 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1476 /* Data has in fact been written and is NOT pending */
1479 /* Data still being streamed thus pending */
1483 /* Pending check is now done. */
1484 stream
->data_pending_check_done
= 1;
1489 reply
.ret_code
= htobe32(ret
);
1490 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1492 ERR("Relay data pending ret code failed");
1500 * Wait for the control socket to reach a quiescent state.
1502 * Note that for now, when receiving this command from the session daemon, this
1503 * means that every subsequent commands or data received on the control socket
1504 * has been handled. So, this is why we simply return OK here.
1507 int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1508 struct relay_command
*cmd
)
1512 struct relay_stream
*stream
;
1513 struct lttng_ht_iter iter
;
1514 struct lttcomm_relayd_quiescent_control msg
;
1515 struct lttcomm_relayd_generic_reply reply
;
1517 DBG("Checking quiescent state on control socket");
1519 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1520 ERR("Trying to check for data before version check");
1522 goto end_no_session
;
1525 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1526 if (ret
< sizeof(msg
)) {
1528 /* Orderly shutdown. Not necessary to print an error. */
1529 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1531 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1535 goto end_no_session
;
1538 stream_id
= be64toh(msg
.stream_id
);
1541 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1543 if (stream
->stream_handle
== stream_id
) {
1544 stream
->data_pending_check_done
= 1;
1545 DBG("Relay quiescent control pending flag set to %" PRIu64
,
1552 reply
.ret_code
= htobe32(LTTNG_OK
);
1553 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1555 ERR("Relay data quiescent control ret code failed");
1563 * Initialize a data pending command. This means that a client is about to ask
1564 * for data pending for each stream he/she holds. Simply iterate over all
1565 * streams of a session and set the data_pending_check_done flag.
1567 * This command returns to the client a LTTNG_OK code.
1570 int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1571 struct relay_command
*cmd
)
1574 struct lttng_ht_iter iter
;
1575 struct lttcomm_relayd_begin_data_pending msg
;
1576 struct lttcomm_relayd_generic_reply reply
;
1577 struct relay_stream
*stream
;
1578 uint64_t session_id
;
1583 DBG("Init streams for data pending");
1585 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1586 ERR("Trying to check for data before version check");
1588 goto end_no_session
;
1591 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1592 if (ret
< sizeof(msg
)) {
1594 /* Orderly shutdown. Not necessary to print an error. */
1595 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1597 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1601 goto end_no_session
;
1604 session_id
= be64toh(msg
.session_id
);
1607 * Iterate over all streams to set the begin data pending flag. For now, the
1608 * streams are indexed by stream handle so we have to iterate over all
1609 * streams to find the one associated with the right session_id.
1612 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1614 if (stream
->session
->id
== session_id
) {
1615 stream
->data_pending_check_done
= 0;
1616 DBG("Set begin data pending flag to stream %" PRIu64
,
1617 stream
->stream_handle
);
1622 /* All good, send back reply. */
1623 reply
.ret_code
= htobe32(LTTNG_OK
);
1625 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1627 ERR("Relay begin data pending send reply failed");
1635 * End data pending command. This will check, for a given session id, if each
1636 * stream associated with it has its data_pending_check_done flag set. If not,
1637 * this means that the client lost track of the stream but the data is still
1638 * being streamed on our side. In this case, we inform the client that data is
1641 * Return to the client if there is data in flight or not with a ret_code.
1644 int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1645 struct relay_command
*cmd
)
1648 struct lttng_ht_iter iter
;
1649 struct lttcomm_relayd_end_data_pending msg
;
1650 struct lttcomm_relayd_generic_reply reply
;
1651 struct relay_stream
*stream
;
1652 uint64_t session_id
;
1653 uint32_t is_data_inflight
= 0;
1658 DBG("End data pending command");
1660 if (!cmd
->session
|| cmd
->version_check_done
== 0) {
1661 ERR("Trying to check for data before version check");
1663 goto end_no_session
;
1666 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &msg
, sizeof(msg
), 0);
1667 if (ret
< sizeof(msg
)) {
1669 /* Orderly shutdown. Not necessary to print an error. */
1670 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1672 ERR("Relay didn't receive valid end data_pending struct size: %d",
1676 goto end_no_session
;
1679 session_id
= be64toh(msg
.session_id
);
1681 /* Iterate over all streams to see if the begin data pending flag is set. */
1683 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1685 if (stream
->session
->id
== session_id
&&
1686 !stream
->data_pending_check_done
) {
1687 is_data_inflight
= 1;
1688 DBG("Data is still in flight for stream %" PRIu64
,
1689 stream
->stream_handle
);
1695 /* All good, send back reply. */
1696 reply
.ret_code
= htobe32(is_data_inflight
);
1698 ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1700 ERR("Relay end data pending send reply failed");
1708 * Receive an index for a specific stream.
1710 * Return 0 on success else a negative value.
1713 int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1714 struct relay_command
*cmd
)
1716 int ret
, send_ret
, index_created
= 0;
1717 struct relay_session
*session
= cmd
->session
;
1718 struct lttcomm_relayd_index index_info
;
1719 struct relay_index
*index
, *wr_index
= NULL
;
1720 struct lttcomm_relayd_generic_reply reply
;
1721 struct relay_stream
*stream
;
1722 uint64_t net_seq_num
;
1726 DBG("Relay receiving index");
1728 if (!session
|| cmd
->version_check_done
== 0) {
1729 ERR("Trying to close a stream before version check");
1731 goto end_no_session
;
1734 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &index_info
,
1735 sizeof(index_info
), 0);
1736 if (ret
< sizeof(index_info
)) {
1738 /* Orderly shutdown. Not necessary to print an error. */
1739 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1741 ERR("Relay didn't receive valid index struct size : %d", ret
);
1744 goto end_no_session
;
1747 net_seq_num
= be64toh(index_info
.net_seq_num
);
1750 stream
= relay_stream_find_by_id(be64toh(index_info
.relay_stream_id
));
1753 goto end_rcu_unlock
;
1756 /* Live beacon handling */
1757 if (index_info
.packet_size
== 0) {
1758 DBG("Received live beacon for stream %" PRIu64
, stream
->stream_handle
);
1761 * Only flag a stream inactive when it has already received data.
1763 if (stream
->total_index_received
> 0) {
1764 stream
->beacon_ts_end
= be64toh(index_info
.timestamp_end
);
1767 goto end_rcu_unlock
;
1769 stream
->beacon_ts_end
= -1ULL;
1772 index
= relay_index_find(stream
->stream_handle
, net_seq_num
);
1774 /* A successful creation will add the object to the HT. */
1775 index
= relay_index_create(stream
->stream_handle
, net_seq_num
);
1777 goto end_rcu_unlock
;
1782 copy_index_control_data(index
, &index_info
);
1784 if (index_created
) {
1786 * Try to add the relay index object to the hash table. If an object
1787 * already exist, destroy back the index created, set the data in this
1788 * object and write it on disk.
1790 relay_index_add(index
, &wr_index
);
1792 copy_index_control_data(wr_index
, &index_info
);
1796 /* The index already exists so write it on disk. */
1800 /* Do we have a writable ready index to write on disk. */
1802 /* Starting at 2.4, create the index file if none available. */
1803 if (cmd
->minor
>= 4 && stream
->index_fd
< 0) {
1804 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
1805 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
1806 stream
->tracefile_count_current
);
1808 goto end_rcu_unlock
;
1810 stream
->index_fd
= ret
;
1813 ret
= relay_index_write(wr_index
->fd
, wr_index
);
1815 goto end_rcu_unlock
;
1817 stream
->total_index_received
++;
1824 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1826 reply
.ret_code
= htobe32(LTTNG_OK
);
1828 send_ret
= cmd
->sock
->ops
->sendmsg(cmd
->sock
, &reply
, sizeof(reply
), 0);
1830 ERR("Relay sending close index id reply");
1839 * Process the commands received on the control socket
1842 int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1843 struct relay_command
*cmd
, struct relay_local_data
*ctx
)
1847 switch (be32toh(recv_hdr
->cmd
)) {
1848 case RELAYD_CREATE_SESSION
:
1849 ret
= relay_create_session(recv_hdr
, cmd
, ctx
->sessions_ht
);
1851 case RELAYD_ADD_STREAM
:
1852 ret
= relay_add_stream(recv_hdr
, cmd
, ctx
->sessions_ht
);
1854 case RELAYD_START_DATA
:
1855 ret
= relay_start(recv_hdr
, cmd
);
1857 case RELAYD_SEND_METADATA
:
1858 ret
= relay_recv_metadata(recv_hdr
, cmd
);
1860 case RELAYD_VERSION
:
1861 ret
= relay_send_version(recv_hdr
, cmd
, ctx
->sessions_ht
);
1863 case RELAYD_CLOSE_STREAM
:
1864 ret
= relay_close_stream(recv_hdr
, cmd
);
1866 case RELAYD_DATA_PENDING
:
1867 ret
= relay_data_pending(recv_hdr
, cmd
);
1869 case RELAYD_QUIESCENT_CONTROL
:
1870 ret
= relay_quiescent_control(recv_hdr
, cmd
);
1872 case RELAYD_BEGIN_DATA_PENDING
:
1873 ret
= relay_begin_data_pending(recv_hdr
, cmd
);
1875 case RELAYD_END_DATA_PENDING
:
1876 ret
= relay_end_data_pending(recv_hdr
, cmd
);
1878 case RELAYD_SEND_INDEX
:
1879 ret
= relay_recv_index(recv_hdr
, cmd
);
1881 case RELAYD_UPDATE_SYNC_INFO
:
1883 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
1884 relay_unknown_command(cmd
);
1894 * Handle index for a data stream.
1896 * RCU read side lock MUST be acquired.
1898 * Return 0 on success else a negative value.
1900 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
1903 int ret
= 0, index_created
= 0;
1904 uint64_t stream_id
, data_offset
;
1905 struct relay_index
*index
, *wr_index
= NULL
;
1909 stream_id
= stream
->stream_handle
;
1910 /* Get data offset because we are about to update the index. */
1911 data_offset
= htobe64(stream
->tracefile_size_current
);
1914 * Lookup for an existing index for that stream id/sequence number. If on
1915 * exists, the control thread already received the data for it thus we need
1916 * to write it on disk.
1918 index
= relay_index_find(stream_id
, net_seq_num
);
1920 /* A successful creation will add the object to the HT. */
1921 index
= relay_index_create(stream_id
, net_seq_num
);
1929 if (rotate_index
|| stream
->index_fd
< 0) {
1930 index
->to_close_fd
= stream
->index_fd
;
1931 ret
= index_create_file(stream
->path_name
, stream
->channel_name
,
1932 relayd_uid
, relayd_gid
, stream
->tracefile_size
,
1933 stream
->tracefile_count_current
);
1935 /* This will close the stream's index fd if one. */
1936 relay_index_free_safe(index
);
1939 stream
->index_fd
= ret
;
1941 index
->fd
= stream
->index_fd
;
1942 index
->index_data
.offset
= data_offset
;
1944 if (index_created
) {
1946 * Try to add the relay index object to the hash table. If an object
1947 * already exist, destroy back the index created and set the data.
1949 relay_index_add(index
, &wr_index
);
1951 /* Copy back data from the created index. */
1952 wr_index
->fd
= index
->fd
;
1953 wr_index
->to_close_fd
= index
->to_close_fd
;
1954 wr_index
->index_data
.offset
= data_offset
;
1958 /* The index already exists so write it on disk. */
1962 /* Do we have a writable ready index to write on disk. */
1964 ret
= relay_index_write(wr_index
->fd
, wr_index
);
1968 stream
->total_index_received
++;
1976 * relay_process_data: Process the data received on the data socket
1979 int relay_process_data(struct relay_command
*cmd
)
1981 int ret
= 0, rotate_index
= 0;
1983 struct relay_stream
*stream
;
1984 struct lttcomm_relayd_data_hdr data_hdr
;
1986 uint64_t net_seq_num
;
1989 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, &data_hdr
,
1990 sizeof(struct lttcomm_relayd_data_hdr
), 0);
1993 /* Orderly shutdown. Not necessary to print an error. */
1994 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
1996 ERR("Unable to receive data header on sock %d", cmd
->sock
->fd
);
2002 stream_id
= be64toh(data_hdr
.stream_id
);
2005 stream
= relay_stream_find_by_id(stream_id
);
2008 goto end_rcu_unlock
;
2011 data_size
= be32toh(data_hdr
.data_size
);
2012 if (data_buffer_size
< data_size
) {
2015 tmp_data_ptr
= realloc(data_buffer
, data_size
);
2016 if (!tmp_data_ptr
) {
2017 ERR("Allocating data buffer");
2020 goto end_rcu_unlock
;
2022 data_buffer
= tmp_data_ptr
;
2023 data_buffer_size
= data_size
;
2025 memset(data_buffer
, 0, data_size
);
2027 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2029 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2030 data_size
, stream_id
, net_seq_num
);
2031 ret
= cmd
->sock
->ops
->recvmsg(cmd
->sock
, data_buffer
, data_size
, 0);
2034 /* Orderly shutdown. Not necessary to print an error. */
2035 DBG("Socket %d did an orderly shutdown", cmd
->sock
->fd
);
2038 goto end_rcu_unlock
;
2041 /* Check if a rotation is needed. */
2042 if (stream
->tracefile_size
> 0 &&
2043 (stream
->tracefile_size_current
+ data_size
) >
2044 stream
->tracefile_size
) {
2045 struct relay_viewer_stream
*vstream
;
2048 new_id
= (stream
->tracefile_count_current
+ 1) %
2049 stream
->tracefile_count
;
2051 * When we wrap-around back to 0, we start overwriting old
2054 if (!stream
->tracefile_overwrite
&& new_id
== 0) {
2055 stream
->tracefile_overwrite
= 1;
2057 pthread_mutex_lock(&stream
->viewer_stream_rotation_lock
);
2058 if (stream
->tracefile_overwrite
) {
2059 stream
->oldest_tracefile_id
=
2060 (stream
->oldest_tracefile_id
+ 1) %
2061 stream
->tracefile_count
;
2063 vstream
= live_find_viewer_stream_by_id(stream
->stream_handle
);
2066 * The viewer is reading a file about to be
2067 * overwritten. Close the FDs it is
2068 * currently using and let it handle the fault.
2070 if (vstream
->tracefile_count_current
== new_id
) {
2071 pthread_mutex_lock(&vstream
->overwrite_lock
);
2072 vstream
->abort_flag
= 1;
2073 pthread_mutex_unlock(&vstream
->overwrite_lock
);
2074 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2075 stream
->channel_name
, new_id
);
2076 } else if (vstream
->tracefile_count_current
==
2077 stream
->tracefile_count_current
) {
2079 * The reader and writer were in the
2080 * same trace file, inform the viewer
2081 * that no new index will ever be added
2084 vstream
->close_write_flag
= 1;
2087 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
2088 stream
->tracefile_size
, stream
->tracefile_count
,
2089 relayd_uid
, relayd_gid
, stream
->fd
,
2090 &(stream
->tracefile_count_current
), &stream
->fd
);
2091 stream
->total_index_received
= 0;
2092 pthread_mutex_unlock(&stream
->viewer_stream_rotation_lock
);
2094 ERR("Rotating stream output file");
2095 goto end_rcu_unlock
;
2097 /* Reset current size because we just perform a stream rotation. */
2098 stream
->tracefile_size_current
= 0;
2103 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2104 * index are NOT supported.
2106 if (stream
->session
->minor
>= 4 && !stream
->session
->snapshot
) {
2107 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2109 goto end_rcu_unlock
;
2113 /* Write data to stream output fd. */
2114 size_ret
= lttng_write(stream
->fd
, data_buffer
, data_size
);
2115 if (size_ret
< data_size
) {
2116 ERR("Relay error writing data to file");
2118 goto end_rcu_unlock
;
2121 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64
,
2122 ret
, stream
->stream_handle
);
2124 ret
= write_padding_to_file(stream
->fd
, be32toh(data_hdr
.padding_size
));
2126 goto end_rcu_unlock
;
2128 stream
->tracefile_size_current
+= data_size
+ be32toh(data_hdr
.padding_size
);
2130 stream
->prev_seq
= net_seq_num
;
2132 /* Check if we need to close the FD */
2133 if (close_stream_check(stream
)) {
2134 destroy_stream(stream
);
2144 void relay_cleanup_poll_connection(struct lttng_poll_event
*events
, int pollfd
)
2148 lttng_poll_del(events
, pollfd
);
2150 ret
= close(pollfd
);
2152 ERR("Closing pollfd %d", pollfd
);
2157 int relay_add_connection(int fd
, struct lttng_poll_event
*events
,
2158 struct lttng_ht
*relay_connections_ht
)
2160 struct relay_command
*relay_connection
;
2163 relay_connection
= zmalloc(sizeof(struct relay_command
));
2164 if (relay_connection
== NULL
) {
2165 PERROR("Relay command zmalloc");
2168 ret
= lttng_read(fd
, relay_connection
, sizeof(struct relay_command
));
2169 if (ret
< sizeof(struct relay_command
)) {
2170 PERROR("read relay cmd pipe");
2175 * Only used by the control side and the reference is copied inside each
2176 * stream from that connection. Thus a destroy HT must be done after every
2177 * stream has been destroyed.
2179 if (relay_connection
->type
== RELAY_CONTROL
) {
2180 relay_connection
->ctf_traces_ht
= lttng_ht_new(0,
2181 LTTNG_HT_TYPE_STRING
);
2182 if (!relay_connection
->ctf_traces_ht
) {
2187 lttng_ht_node_init_ulong(&relay_connection
->sock_n
,
2188 (unsigned long) relay_connection
->sock
->fd
);
2190 lttng_ht_add_unique_ulong(relay_connections_ht
,
2191 &relay_connection
->sock_n
);
2193 return lttng_poll_add(events
,
2194 relay_connection
->sock
->fd
,
2195 LPOLLIN
| LPOLLRDHUP
);
2198 free(relay_connection
);
2204 void deferred_free_connection(struct rcu_head
*head
)
2206 struct relay_command
*relay_connection
=
2207 caa_container_of(head
, struct relay_command
, rcu_node
);
2209 lttcomm_destroy_sock(relay_connection
->sock
);
2210 free(relay_connection
);
2214 void relay_del_connection(struct lttng_ht
*relay_connections_ht
,
2215 struct lttng_ht_iter
*iter
, struct relay_command
*relay_connection
,
2216 struct lttng_ht
*sessions_ht
)
2220 ret
= lttng_ht_del(relay_connections_ht
, iter
);
2223 if (relay_connection
->type
== RELAY_CONTROL
) {
2224 relay_delete_session(relay_connection
, sessions_ht
);
2225 lttng_ht_destroy(relay_connection
->ctf_traces_ht
);
2228 call_rcu(&relay_connection
->rcu_node
, deferred_free_connection
);
2232 * This thread does the actual work
2235 void *relay_thread_worker(void *data
)
2237 int ret
, err
= -1, last_seen_data_fd
= -1;
2239 struct relay_command
*relay_connection
;
2240 struct lttng_poll_event events
;
2241 struct lttng_ht
*relay_connections_ht
;
2242 struct lttng_ht_node_ulong
*node
;
2243 struct lttng_ht_iter iter
;
2244 struct lttcomm_relayd_hdr recv_hdr
;
2245 struct relay_local_data
*relay_ctx
= (struct relay_local_data
*) data
;
2246 struct lttng_ht
*sessions_ht
= relay_ctx
->sessions_ht
;
2248 DBG("[thread] Relay worker started");
2250 rcu_register_thread();
2252 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2254 health_code_update();
2256 /* table of connections indexed on socket */
2257 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2258 if (!relay_connections_ht
) {
2259 goto relay_connections_ht_error
;
2262 /* Tables of received indexes indexed by index handle and net_seq_num. */
2263 indexes_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64
);
2265 goto indexes_ht_error
;
2268 ret
= create_thread_poll_set(&events
, 2);
2270 goto error_poll_create
;
2273 ret
= lttng_poll_add(&events
, relay_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2280 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2282 health_code_update();
2284 /* Infinite blocking call, waiting for transmission */
2285 DBG3("Relayd worker thread polling...");
2286 health_poll_entry();
2287 ret
= lttng_poll_wait(&events
, -1);
2291 * Restart interrupted system call.
2293 if (errno
== EINTR
) {
2302 * Process control. The control connection is prioritised so we don't
2303 * starve it with high throughout put tracing data on the data
2306 for (i
= 0; i
< nb_fd
; i
++) {
2307 /* Fetch once the poll data */
2308 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2309 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2311 health_code_update();
2313 /* Thread quit pipe has been closed. Killing thread. */
2314 ret
= check_thread_quit_pipe(pollfd
, revents
);
2320 /* Inspect the relay cmd pipe for new connection */
2321 if (pollfd
== relay_cmd_pipe
[0]) {
2322 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2323 ERR("Relay pipe error");
2325 } else if (revents
& LPOLLIN
) {
2326 DBG("Relay command received");
2327 ret
= relay_add_connection(relay_cmd_pipe
[0],
2328 &events
, relay_connections_ht
);
2333 } else if (revents
) {
2335 lttng_ht_lookup(relay_connections_ht
,
2336 (void *)((unsigned long) pollfd
),
2338 node
= lttng_ht_iter_get_node_ulong(&iter
);
2340 DBG2("Relay sock %d not found", pollfd
);
2344 relay_connection
= caa_container_of(node
,
2345 struct relay_command
, sock_n
);
2347 if (revents
& (LPOLLERR
)) {
2349 relay_cleanup_poll_connection(&events
, pollfd
);
2350 relay_del_connection(relay_connections_ht
,
2351 &iter
, relay_connection
, sessions_ht
);
2352 if (last_seen_data_fd
== pollfd
) {
2353 last_seen_data_fd
= last_notdel_data_fd
;
2355 } else if (revents
& (LPOLLHUP
| LPOLLRDHUP
)) {
2356 DBG("Socket %d hung up", pollfd
);
2357 relay_cleanup_poll_connection(&events
, pollfd
);
2358 relay_del_connection(relay_connections_ht
,
2359 &iter
, relay_connection
, sessions_ht
);
2360 if (last_seen_data_fd
== pollfd
) {
2361 last_seen_data_fd
= last_notdel_data_fd
;
2363 } else if (revents
& LPOLLIN
) {
2364 /* control socket */
2365 if (relay_connection
->type
== RELAY_CONTROL
) {
2366 ret
= relay_connection
->sock
->ops
->recvmsg(
2367 relay_connection
->sock
, &recv_hdr
,
2368 sizeof(struct lttcomm_relayd_hdr
), 0);
2369 /* connection closed */
2371 relay_cleanup_poll_connection(&events
, pollfd
);
2372 relay_del_connection(relay_connections_ht
,
2373 &iter
, relay_connection
, sessions_ht
);
2374 DBG("Control connection closed with %d", pollfd
);
2376 if (relay_connection
->session
) {
2377 DBG2("Relay worker receiving data for session : %" PRIu64
,
2378 relay_connection
->session
->id
);
2380 ret
= relay_process_control(&recv_hdr
,
2381 relay_connection
, relay_ctx
);
2383 /* Clear the session on error. */
2384 relay_cleanup_poll_connection(&events
, pollfd
);
2385 relay_del_connection(relay_connections_ht
,
2386 &iter
, relay_connection
, sessions_ht
);
2387 DBG("Connection closed with %d", pollfd
);
2393 * Flag the last seen data fd not deleted. It will be
2394 * used as the last seen fd if any fd gets deleted in
2397 last_notdel_data_fd
= pollfd
;
2405 * The last loop handled a control request, go back to poll to make
2406 * sure we prioritise the control socket.
2412 if (last_seen_data_fd
>= 0) {
2413 for (i
= 0; i
< nb_fd
; i
++) {
2414 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2416 health_code_update();
2418 if (last_seen_data_fd
== pollfd
) {
2425 /* Process data connection. */
2426 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2427 /* Fetch the poll data. */
2428 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2429 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2431 health_code_update();
2433 /* Skip the command pipe. It's handled in the first loop. */
2434 if (pollfd
== relay_cmd_pipe
[0]) {
2440 lttng_ht_lookup(relay_connections_ht
,
2441 (void *)((unsigned long) pollfd
),
2443 node
= lttng_ht_iter_get_node_ulong(&iter
);
2445 /* Skip it. Might be removed before. */
2449 relay_connection
= caa_container_of(node
,
2450 struct relay_command
, sock_n
);
2452 if (revents
& LPOLLIN
) {
2453 if (relay_connection
->type
!= RELAY_DATA
) {
2457 ret
= relay_process_data(relay_connection
);
2458 /* connection closed */
2460 relay_cleanup_poll_connection(&events
, pollfd
);
2461 relay_del_connection(relay_connections_ht
,
2462 &iter
, relay_connection
, sessions_ht
);
2463 DBG("Data connection closed with %d", pollfd
);
2465 * Every goto restart call sets the last seen fd where
2466 * here we don't really care since we gracefully
2467 * continue the loop after the connection is deleted.
2470 /* Keep last seen port. */
2471 last_seen_data_fd
= pollfd
;
2479 last_seen_data_fd
= -1;
2482 /* Normal exit, no error */
2487 lttng_poll_clean(&events
);
2489 /* empty the hash table and free the memory */
2491 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
, node
, node
) {
2492 health_code_update();
2494 node
= lttng_ht_iter_get_node_ulong(&iter
);
2496 relay_connection
= caa_container_of(node
,
2497 struct relay_command
, sock_n
);
2498 relay_del_connection(relay_connections_ht
,
2499 &iter
, relay_connection
, sessions_ht
);
2504 lttng_ht_destroy(indexes_ht
);
2506 lttng_ht_destroy(relay_connections_ht
);
2507 relay_connections_ht_error
:
2508 /* Close relay cmd pipes */
2509 utils_close_pipe(relay_cmd_pipe
);
2511 DBG("Thread exited with error");
2513 DBG("Worker thread cleanup complete");
2517 ERR("Health error occurred in %s", __func__
);
2519 health_unregister(health_relayd
);
2520 rcu_unregister_thread();
2526 * Create the relay command pipe to wake thread_manage_apps.
2527 * Closed in cleanup().
2529 static int create_relay_cmd_pipe(void)
2533 ret
= utils_create_pipe_cloexec(relay_cmd_pipe
);
2541 int main(int argc
, char **argv
)
2545 struct relay_local_data
*relay_ctx
;
2547 /* Create thread quit pipe */
2548 if ((ret
= init_thread_quit_pipe()) < 0) {
2552 /* Parse arguments */
2554 if ((ret
= parse_args(argc
, argv
)) < 0) {
2558 if ((ret
= set_signal_handler()) < 0) {
2562 /* Try to create directory if -o, --output is specified. */
2563 if (opt_output_path
) {
2564 if (*opt_output_path
!= '/') {
2565 ERR("Please specify an absolute path for -o, --output PATH");
2569 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
);
2571 ERR("Unable to create %s", opt_output_path
);
2585 /* We need those values for the file/dir creation. */
2586 relayd_uid
= getuid();
2587 relayd_gid
= getgid();
2589 /* Check if daemon is UID = 0 */
2590 if (relayd_uid
== 0) {
2591 if (control_uri
->port
< 1024 || data_uri
->port
< 1024) {
2592 ERR("Need to be root to use ports < 1024");
2598 /* Setup the thread apps communication pipe. */
2599 if ((ret
= create_relay_cmd_pipe()) < 0) {
2603 /* Init relay command queue. */
2604 cds_wfq_init(&relay_cmd_queue
.queue
);
2606 /* Set up max poll set size */
2607 lttng_poll_set_max_size();
2609 /* Initialize communication library */
2612 relay_ctx
= zmalloc(sizeof(struct relay_local_data
));
2614 PERROR("relay_ctx");
2618 /* tables of sessions indexed by session ID */
2619 relay_ctx
->sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2620 if (!relay_ctx
->sessions_ht
) {
2621 goto exit_relay_ctx_sessions
;
2624 /* tables of streams indexed by stream ID */
2625 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2626 if (!relay_streams_ht
) {
2627 goto exit_relay_ctx_streams
;
2630 /* tables of streams indexed by stream ID */
2631 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2632 if (!viewer_streams_ht
) {
2633 goto exit_relay_ctx_viewer_streams
;
2636 /* Initialize thread health monitoring */
2637 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
2638 if (!health_relayd
) {
2639 PERROR("health_app_create error");
2640 goto exit_health_app_create
;
2643 ret
= utils_create_pipe(health_quit_pipe
);
2645 goto error_health_pipe
;
2648 /* Create thread to manage the client socket */
2649 ret
= pthread_create(&health_thread
, NULL
,
2650 thread_manage_health
, (void *) NULL
);
2652 PERROR("pthread_create health");
2656 /* Setup the dispatcher thread */
2657 ret
= pthread_create(&dispatcher_thread
, NULL
,
2658 relay_thread_dispatcher
, (void *) NULL
);
2660 PERROR("pthread_create dispatcher");
2661 goto exit_dispatcher
;
2664 /* Setup the worker thread */
2665 ret
= pthread_create(&worker_thread
, NULL
,
2666 relay_thread_worker
, (void *) relay_ctx
);
2668 PERROR("pthread_create worker");
2672 /* Setup the listener thread */
2673 ret
= pthread_create(&listener_thread
, NULL
,
2674 relay_thread_listener
, (void *) NULL
);
2676 PERROR("pthread_create listener");
2680 ret
= live_start_threads(live_uri
, relay_ctx
, thread_quit_pipe
);
2682 ERR("Starting live viewer threads");
2687 ret
= pthread_join(listener_thread
, &status
);
2689 PERROR("pthread_join");
2690 goto error
; /* join error, exit without cleanup */
2694 ret
= pthread_join(worker_thread
, &status
);
2696 PERROR("pthread_join");
2697 goto error
; /* join error, exit without cleanup */
2701 ret
= pthread_join(dispatcher_thread
, &status
);
2703 PERROR("pthread_join");
2704 goto error
; /* join error, exit without cleanup */
2708 ret
= pthread_join(health_thread
, &status
);
2710 PERROR("pthread_join health thread");
2711 goto error
; /* join error, exit without cleanup */
2715 * Stop live threads only after joining other threads.
2717 live_stop_threads();
2720 utils_close_pipe(health_quit_pipe
);
2723 health_app_destroy(health_relayd
);
2725 exit_health_app_create
:
2726 lttng_ht_destroy(viewer_streams_ht
);
2728 exit_relay_ctx_viewer_streams
:
2729 lttng_ht_destroy(relay_streams_ht
);
2731 exit_relay_ctx_streams
:
2732 lttng_ht_destroy(relay_ctx
->sessions_ht
);
2734 exit_relay_ctx_sessions
: