2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
35 #include <sys/types.h>
38 #include <urcu/futex.h>
39 #include <urcu/uatomic.h>
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/compat/endian.h>
48 #include <common/compat/getenv.h>
49 #include <common/defaults.h>
50 #include <common/daemonize.h>
51 #include <common/futex.h>
52 #include <common/sessiond-comm/sessiond-comm.h>
53 #include <common/sessiond-comm/inet.h>
54 #include <common/sessiond-comm/relayd.h>
55 #include <common/uri.h>
56 #include <common/utils.h>
57 #include <common/config/session-config.h>
58 #include <urcu/rculist.h>
61 #include "ctf-trace.h"
64 #include "lttng-relayd.h"
66 #include "health-relayd.h"
67 #include "testpoint.h"
68 #include "viewer-stream.h"
71 #include "connection.h"
72 #include "tracefile-array.h"
74 /* command line options */
75 char *opt_output_path
;
76 static int opt_daemon
, opt_background
;
79 * We need to wait for listener and live listener threads, as well as
80 * health check thread, before being ready to signal readiness.
82 #define NR_LTTNG_RELAY_READY 3
83 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
84 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
85 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
87 static struct lttng_uri
*control_uri
;
88 static struct lttng_uri
*data_uri
;
89 static struct lttng_uri
*live_uri
;
93 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
94 static int tracing_group_name_override
;
96 const char * const config_section_name
= "relayd";
99 * Quit pipe for all threads. This permits a single cancellation point
100 * for all threads when receiving an event on the pipe.
102 int thread_quit_pipe
[2] = { -1, -1 };
105 * This pipe is used to inform the worker thread that a command is queued and
106 * ready to be processed.
108 static int relay_conn_pipe
[2] = { -1, -1 };
110 /* Shared between threads */
111 static int dispatch_thread_exit
;
113 static pthread_t listener_thread
;
114 static pthread_t dispatcher_thread
;
115 static pthread_t worker_thread
;
116 static pthread_t health_thread
;
119 * last_relay_stream_id_lock protects last_relay_stream_id increment
120 * atomicity on 32-bit architectures.
122 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
123 static uint64_t last_relay_stream_id
;
126 * Relay command queue.
128 * The relay_thread_listener and relay_thread_dispatcher communicate with this
131 static struct relay_conn_queue relay_conn_queue
;
133 /* buffer allocated at startup, used to store the trace data */
134 static char *data_buffer
;
135 static unsigned int data_buffer_size
;
137 /* Global relay stream hash table. */
138 struct lttng_ht
*relay_streams_ht
;
140 /* Global relay viewer stream hash table. */
141 struct lttng_ht
*viewer_streams_ht
;
143 /* Global relay sessions hash table. */
144 struct lttng_ht
*sessions_ht
;
146 /* Relayd health monitoring */
147 struct health_app
*health_relayd
;
149 static struct option long_options
[] = {
150 { "control-port", 1, 0, 'C', },
151 { "data-port", 1, 0, 'D', },
152 { "live-port", 1, 0, 'L', },
153 { "daemonize", 0, 0, 'd', },
154 { "background", 0, 0, 'b', },
155 { "group", 1, 0, 'g', },
156 { "help", 0, 0, 'h', },
157 { "output", 1, 0, 'o', },
158 { "verbose", 0, 0, 'v', },
159 { "config", 1, 0, 'f' },
163 static const char *config_ignore_options
[] = { "help", "config" };
166 * usage function on stderr
168 static void usage(void)
170 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
171 fprintf(stderr
, " -h, --help Display this usage.\n");
172 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
173 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
174 fprintf(stderr
, " -C, --control-port URL Control port listening.\n");
175 fprintf(stderr
, " -D, --data-port URL Data port listening.\n");
176 fprintf(stderr
, " -L, --live-port URL Live view port listening.\n");
177 fprintf(stderr
, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
178 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
179 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
180 fprintf(stderr
, " -f --config Load daemon configuration file\n");
184 * Take an option from the getopt output and set it in the right variable to be
187 * Return 0 on success else a negative value.
189 static int set_option(int opt
, const char *arg
, const char *optname
)
195 fprintf(stderr
, "option %s", optname
);
197 fprintf(stderr
, " with arg %s\n", arg
);
201 if (lttng_is_setuid_setgid()) {
202 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
203 "-C, --control-port");
205 ret
= uri_parse(arg
, &control_uri
);
207 ERR("Invalid control URI specified");
210 if (control_uri
->port
== 0) {
211 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
216 if (lttng_is_setuid_setgid()) {
217 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
220 ret
= uri_parse(arg
, &data_uri
);
222 ERR("Invalid data URI specified");
225 if (data_uri
->port
== 0) {
226 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
231 if (lttng_is_setuid_setgid()) {
232 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
235 ret
= uri_parse(arg
, &live_uri
);
237 ERR("Invalid live URI specified");
240 if (live_uri
->port
== 0) {
241 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
252 if (lttng_is_setuid_setgid()) {
253 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
256 tracing_group_name
= strdup(arg
);
257 if (tracing_group_name
== NULL
) {
262 tracing_group_name_override
= 1;
269 if (lttng_is_setuid_setgid()) {
270 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
273 ret
= asprintf(&opt_output_path
, "%s", arg
);
276 PERROR("asprintf opt_output_path");
282 /* Verbose level can increase using multiple -v */
284 lttng_opt_verbose
= config_parse_value(arg
);
286 /* Only 3 level of verbosity (-vvv). */
287 if (lttng_opt_verbose
< 3) {
288 lttng_opt_verbose
+= 1;
293 /* Unknown option or other error.
294 * Error is printed by getopt, just return */
307 * config_entry_handler_cb used to handle options read from a config file.
308 * See config_entry_handler_cb comment in common/config/session-config.h for the
309 * return value conventions.
311 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
315 if (!entry
|| !entry
->name
|| !entry
->value
) {
320 /* Check if the option is to be ignored */
321 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
322 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
327 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
328 /* Ignore if entry name is not fully matched. */
329 if (strcmp(entry
->name
, long_options
[i
].name
)) {
334 * If the option takes no argument on the command line,
335 * we have to check if the value is "true". We support
336 * non-zero numeric values, true, on and yes.
338 if (!long_options
[i
].has_arg
) {
339 ret
= config_parse_value(entry
->value
);
342 WARN("Invalid configuration value \"%s\" for option %s",
343 entry
->value
, entry
->name
);
345 /* False, skip boolean config option. */
350 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
354 WARN("Unrecognized option \"%s\" in daemon configuration file.",
361 static int set_options(int argc
, char **argv
)
363 int c
, ret
= 0, option_index
= 0, retval
= 0;
364 int orig_optopt
= optopt
, orig_optind
= optind
;
365 char *default_address
, *optstring
;
366 const char *config_path
= NULL
;
368 optstring
= utils_generate_optstring(long_options
,
369 sizeof(long_options
) / sizeof(struct option
));
375 /* Check for the --config option */
377 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
378 &option_index
)) != -1) {
382 } else if (c
!= 'f') {
386 if (lttng_is_setuid_setgid()) {
387 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
390 config_path
= utils_expand_path(optarg
);
392 ERR("Failed to resolve path: %s", optarg
);
397 ret
= config_get_section_entries(config_path
, config_section_name
,
398 config_entry_handler
, NULL
);
401 ERR("Invalid configuration option at line %i", ret
);
407 /* Reset getopt's global state */
408 optopt
= orig_optopt
;
409 optind
= orig_optind
;
411 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
416 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
423 /* assign default values */
424 if (control_uri
== NULL
) {
425 ret
= asprintf(&default_address
,
426 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
427 DEFAULT_NETWORK_CONTROL_PORT
);
429 PERROR("asprintf default data address");
434 ret
= uri_parse(default_address
, &control_uri
);
435 free(default_address
);
437 ERR("Invalid control URI specified");
442 if (data_uri
== NULL
) {
443 ret
= asprintf(&default_address
,
444 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
445 DEFAULT_NETWORK_DATA_PORT
);
447 PERROR("asprintf default data address");
452 ret
= uri_parse(default_address
, &data_uri
);
453 free(default_address
);
455 ERR("Invalid data URI specified");
460 if (live_uri
== NULL
) {
461 ret
= asprintf(&default_address
,
462 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
463 DEFAULT_NETWORK_VIEWER_PORT
);
465 PERROR("asprintf default viewer control address");
470 ret
= uri_parse(default_address
, &live_uri
);
471 free(default_address
);
473 ERR("Invalid viewer control URI specified");
484 static void print_global_objects(void)
486 rcu_register_thread();
488 print_viewer_streams();
489 print_relay_streams();
492 rcu_unregister_thread();
498 static void relayd_cleanup(void)
500 print_global_objects();
504 if (viewer_streams_ht
)
505 lttng_ht_destroy(viewer_streams_ht
);
506 if (relay_streams_ht
)
507 lttng_ht_destroy(relay_streams_ht
);
509 lttng_ht_destroy(sessions_ht
);
511 /* free the dynamically allocated opt_output_path */
512 free(opt_output_path
);
514 /* Close thread quit pipes */
515 utils_close_pipe(thread_quit_pipe
);
517 uri_free(control_uri
);
519 /* Live URI is freed in the live thread. */
521 if (tracing_group_name_override
) {
522 free((void *) tracing_group_name
);
527 * Write to writable pipe used to notify a thread.
529 static int notify_thread_pipe(int wpipe
)
533 ret
= lttng_write(wpipe
, "!", 1);
535 PERROR("write poll pipe");
543 static int notify_health_quit_pipe(int *pipe
)
547 ret
= lttng_write(pipe
[1], "4", 1);
549 PERROR("write relay health quit");
558 * Stop all relayd and relayd-live threads.
560 int lttng_relay_stop_threads(void)
564 /* Stopping all threads */
565 DBG("Terminating all threads");
566 if (notify_thread_pipe(thread_quit_pipe
[1])) {
567 ERR("write error on thread quit pipe");
571 if (notify_health_quit_pipe(health_quit_pipe
)) {
572 ERR("write error on health quit pipe");
575 /* Dispatch thread */
576 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
577 futex_nto1_wake(&relay_conn_queue
.futex
);
579 if (relayd_live_stop()) {
580 ERR("Error stopping live threads");
587 * Signal handler for the daemon
589 * Simply stop all worker threads, leaving main() return gracefully after
590 * joining all threads and calling cleanup().
592 static void sighandler(int sig
)
596 DBG("SIGPIPE caught");
599 DBG("SIGINT caught");
600 if (lttng_relay_stop_threads()) {
601 ERR("Error stopping threads");
605 DBG("SIGTERM caught");
606 if (lttng_relay_stop_threads()) {
607 ERR("Error stopping threads");
611 CMM_STORE_SHARED(recv_child_signal
, 1);
619 * Setup signal handler for :
620 * SIGINT, SIGTERM, SIGPIPE
622 static int set_signal_handler(void)
628 if ((ret
= sigemptyset(&sigset
)) < 0) {
629 PERROR("sigemptyset");
633 sa
.sa_handler
= sighandler
;
636 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
641 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
646 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
651 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
656 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
661 void lttng_relay_notify_ready(void)
663 /* Notify the parent of the fork() process that we are ready. */
664 if (opt_daemon
|| opt_background
) {
665 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
666 kill(child_ppid
, SIGUSR1
);
672 * Init thread quit pipe.
674 * Return -1 on error or 0 if all pipes are created.
676 static int init_thread_quit_pipe(void)
680 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
686 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
688 static int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
692 if (events
== NULL
|| size
== 0) {
697 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
703 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
715 * Check if the thread quit pipe was triggered.
717 * Return 1 if it was triggered else 0;
719 static int check_thread_quit_pipe(int fd
, uint32_t events
)
721 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
729 * Create and init socket from uri.
731 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
)
734 struct lttcomm_sock
*sock
= NULL
;
736 sock
= lttcomm_alloc_sock_from_uri(uri
);
738 ERR("Allocating socket");
742 ret
= lttcomm_create_sock(sock
);
746 DBG("Listening on sock %d", sock
->fd
);
748 ret
= sock
->ops
->bind(sock
);
753 ret
= sock
->ops
->listen(sock
, -1);
763 lttcomm_destroy_sock(sock
);
769 * This thread manages the listening for new connections on the network
771 static void *relay_thread_listener(void *data
)
773 int i
, ret
, pollfd
, err
= -1;
774 uint32_t revents
, nb_fd
;
775 struct lttng_poll_event events
;
776 struct lttcomm_sock
*control_sock
, *data_sock
;
778 DBG("[thread] Relay listener started");
780 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
782 health_code_update();
784 control_sock
= relay_socket_create(control_uri
);
786 goto error_sock_control
;
789 data_sock
= relay_socket_create(data_uri
);
791 goto error_sock_relay
;
795 * Pass 3 as size here for the thread quit pipe, control and
798 ret
= create_thread_poll_set(&events
, 3);
800 goto error_create_poll
;
803 /* Add the control socket */
804 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
809 /* Add the data socket */
810 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
815 lttng_relay_notify_ready();
817 if (testpoint(relayd_thread_listener
)) {
818 goto error_testpoint
;
822 health_code_update();
824 DBG("Listener accepting connections");
828 ret
= lttng_poll_wait(&events
, -1);
832 * Restart interrupted system call.
834 if (errno
== EINTR
) {
842 DBG("Relay new connection received");
843 for (i
= 0; i
< nb_fd
; i
++) {
844 health_code_update();
846 /* Fetch once the poll data */
847 revents
= LTTNG_POLL_GETEV(&events
, i
);
848 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
852 * No activity for this FD (poll
858 /* Thread quit pipe has been closed. Killing thread. */
859 ret
= check_thread_quit_pipe(pollfd
, revents
);
865 if (revents
& LPOLLIN
) {
867 * A new connection is requested, therefore a
868 * sessiond/consumerd connection is allocated in
869 * this thread, enqueued to a global queue and
870 * dequeued (and freed) in the worker thread.
873 struct relay_connection
*new_conn
;
874 struct lttcomm_sock
*newsock
;
875 enum connection_type type
;
877 if (pollfd
== data_sock
->fd
) {
879 newsock
= data_sock
->ops
->accept(data_sock
);
880 DBG("Relay data connection accepted, socket %d",
883 assert(pollfd
== control_sock
->fd
);
884 type
= RELAY_CONTROL
;
885 newsock
= control_sock
->ops
->accept(control_sock
);
886 DBG("Relay control connection accepted, socket %d",
890 PERROR("accepting sock");
894 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
897 PERROR("setsockopt inet");
898 lttcomm_destroy_sock(newsock
);
901 new_conn
= connection_create(newsock
, type
);
903 lttcomm_destroy_sock(newsock
);
907 /* Enqueue request for the dispatcher thread. */
908 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
912 * Wake the dispatch queue futex.
913 * Implicit memory barrier with the
914 * exchange in cds_wfcq_enqueue.
916 futex_nto1_wake(&relay_conn_queue
.futex
);
917 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
918 ERR("socket poll error");
921 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
931 lttng_poll_clean(&events
);
933 if (data_sock
->fd
>= 0) {
934 ret
= data_sock
->ops
->close(data_sock
);
939 lttcomm_destroy_sock(data_sock
);
941 if (control_sock
->fd
>= 0) {
942 ret
= control_sock
->ops
->close(control_sock
);
947 lttcomm_destroy_sock(control_sock
);
951 ERR("Health error occurred in %s", __func__
);
953 health_unregister(health_relayd
);
954 DBG("Relay listener thread cleanup complete");
955 lttng_relay_stop_threads();
960 * This thread manages the dispatching of the requests to worker threads
962 static void *relay_thread_dispatcher(void *data
)
966 struct cds_wfcq_node
*node
;
967 struct relay_connection
*new_conn
= NULL
;
969 DBG("[thread] Relay dispatcher started");
971 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
973 if (testpoint(relayd_thread_dispatcher
)) {
974 goto error_testpoint
;
977 health_code_update();
979 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
980 health_code_update();
982 /* Atomically prepare the queue futex */
983 futex_nto1_prepare(&relay_conn_queue
.futex
);
986 health_code_update();
988 /* Dequeue commands */
989 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
990 &relay_conn_queue
.tail
);
992 DBG("Woken up but nothing in the relay command queue");
993 /* Continue thread execution */
996 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
998 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1001 * Inform worker thread of the new request. This
1002 * call is blocking so we can be assured that
1003 * the data will be read at some point in time
1004 * or wait to the end of the world :)
1006 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1008 PERROR("write connection pipe");
1009 connection_put(new_conn
);
1012 } while (node
!= NULL
);
1014 /* Futex wait on queue. Blocking call on futex() */
1015 health_poll_entry();
1016 futex_nto1_wait(&relay_conn_queue
.futex
);
1020 /* Normal exit, no error */
1027 ERR("Health error occurred in %s", __func__
);
1029 health_unregister(health_relayd
);
1030 DBG("Dispatch thread dying");
1031 lttng_relay_stop_threads();
1036 * Set index data from the control port to a given index object.
1038 static int set_index_control_data(struct relay_index
*index
,
1039 struct lttcomm_relayd_index
*data
,
1040 struct relay_connection
*conn
)
1042 struct ctf_packet_index index_data
;
1045 * The index on disk is encoded in big endian, so we don't need
1046 * to convert the data received on the network. The data_offset
1047 * value is NEVER modified here and is updated by the data
1050 index_data
.packet_size
= data
->packet_size
;
1051 index_data
.content_size
= data
->content_size
;
1052 index_data
.timestamp_begin
= data
->timestamp_begin
;
1053 index_data
.timestamp_end
= data
->timestamp_end
;
1054 index_data
.events_discarded
= data
->events_discarded
;
1055 index_data
.stream_id
= data
->stream_id
;
1057 if (conn
->minor
>= 8) {
1058 index
->index_data
.stream_instance_id
= data
->stream_instance_id
;
1059 index
->index_data
.packet_seq_num
= data
->packet_seq_num
;
1062 return relay_index_set_data(index
, &index_data
);
1066 * Handle the RELAYD_CREATE_SESSION command.
1068 * On success, send back the session id or else return a negative value.
1070 static int relay_create_session(struct lttcomm_relayd_hdr
*recv_hdr
,
1071 struct relay_connection
*conn
)
1073 int ret
= 0, send_ret
;
1074 struct relay_session
*session
;
1075 struct lttcomm_relayd_status_session reply
;
1076 char session_name
[LTTNG_NAME_MAX
];
1077 char hostname
[LTTNG_HOST_NAME_MAX
];
1078 uint32_t live_timer
= 0;
1079 bool snapshot
= false;
1081 memset(session_name
, 0, LTTNG_NAME_MAX
);
1082 memset(hostname
, 0, LTTNG_HOST_NAME_MAX
);
1084 memset(&reply
, 0, sizeof(reply
));
1086 switch (conn
->minor
) {
1091 case 4: /* LTTng sessiond 2.4 */
1093 ret
= cmd_create_session_2_4(conn
, session_name
,
1094 hostname
, &live_timer
, &snapshot
);
1100 session
= session_create(session_name
, hostname
, live_timer
,
1101 snapshot
, conn
->major
, conn
->minor
);
1106 assert(!conn
->session
);
1107 conn
->session
= session
;
1108 DBG("Created session %" PRIu64
, session
->id
);
1110 reply
.session_id
= htobe64(session
->id
);
1114 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1116 reply
.ret_code
= htobe32(LTTNG_OK
);
1119 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1121 ERR("Relayd sending session id");
1129 * When we have received all the streams and the metadata for a channel,
1130 * we make them visible to the viewer threads.
1132 static void publish_connection_local_streams(struct relay_connection
*conn
)
1134 struct relay_stream
*stream
;
1135 struct relay_session
*session
= conn
->session
;
1138 * We publish all streams belonging to a session atomically wrt
1141 pthread_mutex_lock(&session
->lock
);
1143 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1145 stream_publish(stream
);
1150 * Inform the viewer that there are new streams in the session.
1152 if (session
->viewer_attached
) {
1153 uatomic_set(&session
->new_streams
, 1);
1155 pthread_mutex_unlock(&session
->lock
);
1159 * relay_add_stream: allocate a new stream for a session
1161 static int relay_add_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1162 struct relay_connection
*conn
)
1166 struct relay_session
*session
= conn
->session
;
1167 struct relay_stream
*stream
= NULL
;
1168 struct lttcomm_relayd_status_stream reply
;
1169 struct ctf_trace
*trace
= NULL
;
1170 uint64_t stream_handle
= -1ULL;
1171 char *path_name
= NULL
, *channel_name
= NULL
;
1172 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1174 if (!session
|| conn
->version_check_done
== 0) {
1175 ERR("Trying to add a stream before version check");
1177 goto end_no_session
;
1180 switch (session
->minor
) {
1181 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1182 ret
= cmd_recv_stream_2_1(conn
, &path_name
,
1185 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1187 ret
= cmd_recv_stream_2_2(conn
, &path_name
,
1188 &channel_name
, &tracefile_size
, &tracefile_count
);
1195 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1199 /* This stream here has one reference on the trace. */
1201 pthread_mutex_lock(&last_relay_stream_id_lock
);
1202 stream_handle
= ++last_relay_stream_id
;
1203 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1205 /* We pass ownership of path_name and channel_name. */
1206 stream
= stream_create(trace
, stream_handle
, path_name
,
1207 channel_name
, tracefile_size
, tracefile_count
);
1209 channel_name
= NULL
;
1212 * Streams are the owners of their trace. Reference to trace is
1213 * kept within stream_create().
1215 ctf_trace_put(trace
);
1218 memset(&reply
, 0, sizeof(reply
));
1219 reply
.handle
= htobe64(stream_handle
);
1221 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1223 reply
.ret_code
= htobe32(LTTNG_OK
);
1226 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1227 sizeof(struct lttcomm_relayd_status_stream
), 0);
1229 ERR("Relay sending stream id");
1230 ret
= (int) send_ret
;
1240 * relay_close_stream: close a specific stream
1242 static int relay_close_stream(struct lttcomm_relayd_hdr
*recv_hdr
,
1243 struct relay_connection
*conn
)
1246 struct relay_session
*session
= conn
->session
;
1247 struct lttcomm_relayd_close_stream stream_info
;
1248 struct lttcomm_relayd_generic_reply reply
;
1249 struct relay_stream
*stream
;
1251 DBG("Close stream received");
1253 if (!session
|| conn
->version_check_done
== 0) {
1254 ERR("Trying to close a stream before version check");
1256 goto end_no_session
;
1259 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &stream_info
,
1260 sizeof(struct lttcomm_relayd_close_stream
), 0);
1261 if (ret
< sizeof(struct lttcomm_relayd_close_stream
)) {
1263 /* Orderly shutdown. Not necessary to print an error. */
1264 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1266 ERR("Relay didn't receive valid add_stream struct size : %d", ret
);
1269 goto end_no_session
;
1272 stream
= stream_get_by_id(be64toh(stream_info
.stream_id
));
1279 * Set last_net_seq_num before the close flag. Required by data
1282 pthread_mutex_lock(&stream
->lock
);
1283 stream
->last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1284 pthread_mutex_unlock(&stream
->lock
);
1287 * This is one of the conditions which may trigger a stream close
1288 * with the others being:
1289 * 1) A close command is received for a stream
1290 * 2) The control connection owning the stream is closed
1291 * 3) We have received all of the stream's data _after_ a close
1294 try_stream_close(stream
);
1295 if (stream
->is_metadata
) {
1296 struct relay_viewer_stream
*vstream
;
1298 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1300 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1302 * Since all the metadata has been sent to the
1303 * viewer and that we have a request to close
1304 * its stream, we can safely teardown the
1305 * corresponding metadata viewer stream.
1307 viewer_stream_put(vstream
);
1309 /* Put local reference. */
1310 viewer_stream_put(vstream
);
1316 memset(&reply
, 0, sizeof(reply
));
1318 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1320 reply
.ret_code
= htobe32(LTTNG_OK
);
1322 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1323 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1325 ERR("Relay sending stream id");
1334 * relay_unknown_command: send -1 if received unknown command
1336 static void relay_unknown_command(struct relay_connection
*conn
)
1338 struct lttcomm_relayd_generic_reply reply
;
1341 memset(&reply
, 0, sizeof(reply
));
1342 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1343 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1344 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1346 ERR("Relay sending unknown command");
1351 * relay_start: send an acknowledgment to the client to tell if we are
1352 * ready to receive data. We are ready if a session is established.
1354 static int relay_start(struct lttcomm_relayd_hdr
*recv_hdr
,
1355 struct relay_connection
*conn
)
1357 int ret
= htobe32(LTTNG_OK
);
1358 struct lttcomm_relayd_generic_reply reply
;
1359 struct relay_session
*session
= conn
->session
;
1362 DBG("Trying to start the streaming without a session established");
1363 ret
= htobe32(LTTNG_ERR_UNK
);
1366 memset(&reply
, 0, sizeof(reply
));
1367 reply
.ret_code
= ret
;
1368 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1369 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1371 ERR("Relay sending start ack");
1378 * Append padding to the file pointed by the file descriptor fd.
1380 static int write_padding_to_file(int fd
, uint32_t size
)
1389 zeros
= zmalloc(size
);
1390 if (zeros
== NULL
) {
1391 PERROR("zmalloc zeros for padding");
1396 ret
= lttng_write(fd
, zeros
, size
);
1398 PERROR("write padding to file");
1408 * relay_recv_metadata: receive the metadata for the session.
1410 static int relay_recv_metadata(struct lttcomm_relayd_hdr
*recv_hdr
,
1411 struct relay_connection
*conn
)
1415 struct relay_session
*session
= conn
->session
;
1416 struct lttcomm_relayd_metadata_payload
*metadata_struct
;
1417 struct relay_stream
*metadata_stream
;
1418 uint64_t data_size
, payload_size
;
1421 ERR("Metadata sent before version check");
1426 data_size
= payload_size
= be64toh(recv_hdr
->data_size
);
1427 if (data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1428 ERR("Incorrect data size");
1432 payload_size
-= sizeof(struct lttcomm_relayd_metadata_payload
);
1434 if (data_buffer_size
< data_size
) {
1435 /* In case the realloc fails, we can free the memory */
1438 tmp_data_ptr
= realloc(data_buffer
, data_size
);
1439 if (!tmp_data_ptr
) {
1440 ERR("Allocating data buffer");
1445 data_buffer
= tmp_data_ptr
;
1446 data_buffer_size
= data_size
;
1448 memset(data_buffer
, 0, data_size
);
1449 DBG2("Relay receiving metadata, waiting for %" PRIu64
" bytes", data_size
);
1450 size_ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
1451 if (size_ret
< 0 || size_ret
!= data_size
) {
1452 if (size_ret
== 0) {
1453 /* Orderly shutdown. Not necessary to print an error. */
1454 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1456 ERR("Relay didn't receive the whole metadata");
1461 metadata_struct
= (struct lttcomm_relayd_metadata_payload
*) data_buffer
;
1463 metadata_stream
= stream_get_by_id(be64toh(metadata_struct
->stream_id
));
1464 if (!metadata_stream
) {
1469 pthread_mutex_lock(&metadata_stream
->lock
);
1471 size_ret
= lttng_write(metadata_stream
->stream_fd
->fd
, metadata_struct
->payload
,
1473 if (size_ret
< payload_size
) {
1474 ERR("Relay error writing metadata on file");
1479 size_ret
= write_padding_to_file(metadata_stream
->stream_fd
->fd
,
1480 be32toh(metadata_struct
->padding_size
));
1485 metadata_stream
->metadata_received
+=
1486 payload_size
+ be32toh(metadata_struct
->padding_size
);
1487 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1488 metadata_stream
->metadata_received
);
1491 pthread_mutex_unlock(&metadata_stream
->lock
);
1492 stream_put(metadata_stream
);
1498 * relay_send_version: send relayd version number
1500 static int relay_send_version(struct lttcomm_relayd_hdr
*recv_hdr
,
1501 struct relay_connection
*conn
)
1504 struct lttcomm_relayd_version reply
, msg
;
1506 conn
->version_check_done
= 1;
1508 /* Get version from the other side. */
1509 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1510 if (ret
< 0 || ret
!= sizeof(msg
)) {
1512 /* Orderly shutdown. Not necessary to print an error. */
1513 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1515 ERR("Relay failed to receive the version values.");
1521 memset(&reply
, 0, sizeof(reply
));
1522 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1523 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1525 /* Major versions must be the same */
1526 if (reply
.major
!= be32toh(msg
.major
)) {
1527 DBG("Incompatible major versions (%u vs %u), deleting session",
1528 reply
.major
, be32toh(msg
.major
));
1529 connection_put(conn
);
1534 conn
->major
= reply
.major
;
1535 /* We adapt to the lowest compatible version */
1536 if (reply
.minor
<= be32toh(msg
.minor
)) {
1537 conn
->minor
= reply
.minor
;
1539 conn
->minor
= be32toh(msg
.minor
);
1542 reply
.major
= htobe32(reply
.major
);
1543 reply
.minor
= htobe32(reply
.minor
);
1544 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1545 sizeof(struct lttcomm_relayd_version
), 0);
1547 ERR("Relay sending version");
1550 DBG("Version check done using protocol %u.%u", conn
->major
,
1558 * Check for data pending for a given stream id from the session daemon.
1560 static int relay_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1561 struct relay_connection
*conn
)
1563 struct relay_session
*session
= conn
->session
;
1564 struct lttcomm_relayd_data_pending msg
;
1565 struct lttcomm_relayd_generic_reply reply
;
1566 struct relay_stream
*stream
;
1568 uint64_t last_net_seq_num
, stream_id
;
1570 DBG("Data pending command received");
1572 if (!session
|| conn
->version_check_done
== 0) {
1573 ERR("Trying to check for data before version check");
1575 goto end_no_session
;
1578 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1579 if (ret
< sizeof(msg
)) {
1581 /* Orderly shutdown. Not necessary to print an error. */
1582 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1584 ERR("Relay didn't receive valid data_pending struct size : %d",
1588 goto end_no_session
;
1591 stream_id
= be64toh(msg
.stream_id
);
1592 last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
1594 stream
= stream_get_by_id(stream_id
);
1595 if (stream
== NULL
) {
1600 pthread_mutex_lock(&stream
->lock
);
1602 DBG("Data pending for stream id %" PRIu64
" prev_seq %" PRIu64
1603 " and last_seq %" PRIu64
, stream_id
, stream
->prev_seq
,
1606 /* Avoid wrapping issue */
1607 if (((int64_t) (stream
->prev_seq
- last_net_seq_num
)) >= 0) {
1608 /* Data has in fact been written and is NOT pending */
1611 /* Data still being streamed thus pending */
1615 stream
->data_pending_check_done
= true;
1616 pthread_mutex_unlock(&stream
->lock
);
1621 memset(&reply
, 0, sizeof(reply
));
1622 reply
.ret_code
= htobe32(ret
);
1623 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1625 ERR("Relay data pending ret code failed");
1633 * Wait for the control socket to reach a quiescent state.
1635 * Note that for now, when receiving this command from the session
1636 * daemon, this means that every subsequent commands or data received on
1637 * the control socket has been handled. So, this is why we simply return
1640 static int relay_quiescent_control(struct lttcomm_relayd_hdr
*recv_hdr
,
1641 struct relay_connection
*conn
)
1645 struct relay_stream
*stream
;
1646 struct lttcomm_relayd_quiescent_control msg
;
1647 struct lttcomm_relayd_generic_reply reply
;
1649 DBG("Checking quiescent state on control socket");
1651 if (!conn
->session
|| conn
->version_check_done
== 0) {
1652 ERR("Trying to check for data before version check");
1654 goto end_no_session
;
1657 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1658 if (ret
< sizeof(msg
)) {
1660 /* Orderly shutdown. Not necessary to print an error. */
1661 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1663 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1667 goto end_no_session
;
1670 stream_id
= be64toh(msg
.stream_id
);
1671 stream
= stream_get_by_id(stream_id
);
1675 pthread_mutex_lock(&stream
->lock
);
1676 stream
->data_pending_check_done
= true;
1677 pthread_mutex_unlock(&stream
->lock
);
1678 DBG("Relay quiescent control pending flag set to %" PRIu64
, stream_id
);
1681 memset(&reply
, 0, sizeof(reply
));
1682 reply
.ret_code
= htobe32(LTTNG_OK
);
1683 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1685 ERR("Relay data quiescent control ret code failed");
1693 * Initialize a data pending command. This means that a consumer is about
1694 * to ask for data pending for each stream it holds. Simply iterate over
1695 * all streams of a session and set the data_pending_check_done flag.
1697 * This command returns to the client a LTTNG_OK code.
1699 static int relay_begin_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1700 struct relay_connection
*conn
)
1703 struct lttng_ht_iter iter
;
1704 struct lttcomm_relayd_begin_data_pending msg
;
1705 struct lttcomm_relayd_generic_reply reply
;
1706 struct relay_stream
*stream
;
1707 uint64_t session_id
;
1712 DBG("Init streams for data pending");
1714 if (!conn
->session
|| conn
->version_check_done
== 0) {
1715 ERR("Trying to check for data before version check");
1717 goto end_no_session
;
1720 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1721 if (ret
< sizeof(msg
)) {
1723 /* Orderly shutdown. Not necessary to print an error. */
1724 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1726 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1730 goto end_no_session
;
1733 session_id
= be64toh(msg
.session_id
);
1736 * Iterate over all streams to set the begin data pending flag.
1737 * For now, the streams are indexed by stream handle so we have
1738 * to iterate over all streams to find the one associated with
1739 * the right session_id.
1742 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1744 if (!stream_get(stream
)) {
1747 if (stream
->trace
->session
->id
== session_id
) {
1748 pthread_mutex_lock(&stream
->lock
);
1749 stream
->data_pending_check_done
= false;
1750 pthread_mutex_unlock(&stream
->lock
);
1751 DBG("Set begin data pending flag to stream %" PRIu64
,
1752 stream
->stream_handle
);
1758 memset(&reply
, 0, sizeof(reply
));
1759 /* All good, send back reply. */
1760 reply
.ret_code
= htobe32(LTTNG_OK
);
1762 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1764 ERR("Relay begin data pending send reply failed");
1772 * End data pending command. This will check, for a given session id, if
1773 * each stream associated with it has its data_pending_check_done flag
1774 * set. If not, this means that the client lost track of the stream but
1775 * the data is still being streamed on our side. In this case, we inform
1776 * the client that data is in flight.
1778 * Return to the client if there is data in flight or not with a ret_code.
1780 static int relay_end_data_pending(struct lttcomm_relayd_hdr
*recv_hdr
,
1781 struct relay_connection
*conn
)
1784 struct lttng_ht_iter iter
;
1785 struct lttcomm_relayd_end_data_pending msg
;
1786 struct lttcomm_relayd_generic_reply reply
;
1787 struct relay_stream
*stream
;
1788 uint64_t session_id
;
1789 uint32_t is_data_inflight
= 0;
1791 DBG("End data pending command");
1793 if (!conn
->session
|| conn
->version_check_done
== 0) {
1794 ERR("Trying to check for data before version check");
1796 goto end_no_session
;
1799 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &msg
, sizeof(msg
), 0);
1800 if (ret
< sizeof(msg
)) {
1802 /* Orderly shutdown. Not necessary to print an error. */
1803 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1805 ERR("Relay didn't receive valid end data_pending struct size: %d",
1809 goto end_no_session
;
1812 session_id
= be64toh(msg
.session_id
);
1815 * Iterate over all streams to see if the begin data pending
1819 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
1821 if (!stream_get(stream
)) {
1824 if (stream
->trace
->session
->id
!= session_id
) {
1828 pthread_mutex_lock(&stream
->lock
);
1829 if (!stream
->data_pending_check_done
) {
1830 if (!stream
->closed
|| !(((int64_t) (stream
->prev_seq
- stream
->last_net_seq_num
)) >= 0)) {
1831 is_data_inflight
= 1;
1832 DBG("Data is still in flight for stream %" PRIu64
,
1833 stream
->stream_handle
);
1834 pthread_mutex_unlock(&stream
->lock
);
1839 pthread_mutex_unlock(&stream
->lock
);
1844 memset(&reply
, 0, sizeof(reply
));
1845 /* All good, send back reply. */
1846 reply
.ret_code
= htobe32(is_data_inflight
);
1848 ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1850 ERR("Relay end data pending send reply failed");
1858 * Receive an index for a specific stream.
1860 * Return 0 on success else a negative value.
1862 static int relay_recv_index(struct lttcomm_relayd_hdr
*recv_hdr
,
1863 struct relay_connection
*conn
)
1866 struct relay_session
*session
= conn
->session
;
1867 struct lttcomm_relayd_index index_info
;
1868 struct relay_index
*index
;
1869 struct lttcomm_relayd_generic_reply reply
;
1870 struct relay_stream
*stream
;
1871 uint64_t net_seq_num
;
1875 DBG("Relay receiving index");
1877 if (!session
|| conn
->version_check_done
== 0) {
1878 ERR("Trying to close a stream before version check");
1880 goto end_no_session
;
1883 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &index_info
,
1884 sizeof(index_info
), 0);
1885 if (ret
< sizeof(index_info
)) {
1887 /* Orderly shutdown. Not necessary to print an error. */
1888 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
1890 ERR("Relay didn't receive valid index struct size : %d", ret
);
1893 goto end_no_session
;
1896 net_seq_num
= be64toh(index_info
.net_seq_num
);
1898 stream
= stream_get_by_id(be64toh(index_info
.relay_stream_id
));
1900 ERR("stream_get_by_id not found");
1904 pthread_mutex_lock(&stream
->lock
);
1906 /* Live beacon handling */
1907 if (index_info
.packet_size
== 0) {
1908 DBG("Received live beacon for stream %" PRIu64
,
1909 stream
->stream_handle
);
1912 * Only flag a stream inactive when it has already
1913 * received data and no indexes are in flight.
1915 if (stream
->index_received_seqcount
> 0
1916 && stream
->indexes_in_flight
== 0) {
1917 stream
->beacon_ts_end
=
1918 be64toh(index_info
.timestamp_end
);
1921 goto end_stream_put
;
1923 stream
->beacon_ts_end
= -1ULL;
1926 if (stream
->ctf_stream_id
== -1ULL) {
1927 stream
->ctf_stream_id
= be64toh(index_info
.stream_id
);
1929 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
1932 ERR("relay_index_get_by_id_or_create index NULL");
1933 goto end_stream_put
;
1935 if (set_index_control_data(index
, &index_info
, conn
)) {
1936 ERR("set_index_control_data error");
1937 relay_index_put(index
);
1939 goto end_stream_put
;
1941 ret
= relay_index_try_flush(index
);
1943 tracefile_array_commit_seq(stream
->tfa
);
1944 stream
->index_received_seqcount
++;
1945 } else if (ret
> 0) {
1949 ERR("relay_index_try_flush error %d", ret
);
1950 relay_index_put(index
);
1955 pthread_mutex_unlock(&stream
->lock
);
1960 memset(&reply
, 0, sizeof(reply
));
1962 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1964 reply
.ret_code
= htobe32(LTTNG_OK
);
1966 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1968 ERR("Relay sending close index id reply");
1977 * Receive the streams_sent message.
1979 * Return 0 on success else a negative value.
1981 static int relay_streams_sent(struct lttcomm_relayd_hdr
*recv_hdr
,
1982 struct relay_connection
*conn
)
1985 struct lttcomm_relayd_generic_reply reply
;
1989 DBG("Relay receiving streams_sent");
1991 if (!conn
->session
|| conn
->version_check_done
== 0) {
1992 ERR("Trying to close a stream before version check");
1994 goto end_no_session
;
1998 * Publish every pending stream in the connection recv list which are
1999 * now ready to be used by the viewer.
2001 publish_connection_local_streams(conn
);
2003 memset(&reply
, 0, sizeof(reply
));
2004 reply
.ret_code
= htobe32(LTTNG_OK
);
2005 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2007 ERR("Relay sending sent_stream reply");
2019 * Process the commands received on the control socket
2021 static int relay_process_control(struct lttcomm_relayd_hdr
*recv_hdr
,
2022 struct relay_connection
*conn
)
2026 switch (be32toh(recv_hdr
->cmd
)) {
2027 case RELAYD_CREATE_SESSION
:
2028 ret
= relay_create_session(recv_hdr
, conn
);
2030 case RELAYD_ADD_STREAM
:
2031 ret
= relay_add_stream(recv_hdr
, conn
);
2033 case RELAYD_START_DATA
:
2034 ret
= relay_start(recv_hdr
, conn
);
2036 case RELAYD_SEND_METADATA
:
2037 ret
= relay_recv_metadata(recv_hdr
, conn
);
2039 case RELAYD_VERSION
:
2040 ret
= relay_send_version(recv_hdr
, conn
);
2042 case RELAYD_CLOSE_STREAM
:
2043 ret
= relay_close_stream(recv_hdr
, conn
);
2045 case RELAYD_DATA_PENDING
:
2046 ret
= relay_data_pending(recv_hdr
, conn
);
2048 case RELAYD_QUIESCENT_CONTROL
:
2049 ret
= relay_quiescent_control(recv_hdr
, conn
);
2051 case RELAYD_BEGIN_DATA_PENDING
:
2052 ret
= relay_begin_data_pending(recv_hdr
, conn
);
2054 case RELAYD_END_DATA_PENDING
:
2055 ret
= relay_end_data_pending(recv_hdr
, conn
);
2057 case RELAYD_SEND_INDEX
:
2058 ret
= relay_recv_index(recv_hdr
, conn
);
2060 case RELAYD_STREAMS_SENT
:
2061 ret
= relay_streams_sent(recv_hdr
, conn
);
2063 case RELAYD_UPDATE_SYNC_INFO
:
2065 ERR("Received unknown command (%u)", be32toh(recv_hdr
->cmd
));
2066 relay_unknown_command(conn
);
2076 * Handle index for a data stream.
2078 * Called with the stream lock held.
2080 * Return 0 on success else a negative value.
2082 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
2086 uint64_t data_offset
;
2087 struct relay_index
*index
;
2089 /* Get data offset because we are about to update the index. */
2090 data_offset
= htobe64(stream
->tracefile_size_current
);
2092 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
2093 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
2096 * Lookup for an existing index for that stream id/sequence
2097 * number. If it exists, the control thread has already received the
2098 * data for it, thus we need to write it to disk.
2100 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
2106 if (rotate_index
|| !stream
->index_fd
) {
2109 /* Put ref on previous index_fd. */
2110 if (stream
->index_fd
) {
2111 stream_fd_put(stream
->index_fd
);
2112 stream
->index_fd
= NULL
;
2115 fd
= index_create_file(stream
->path_name
, stream
->channel_name
,
2116 -1, -1, stream
->tracefile_size
,
2117 tracefile_array_get_file_index_head(stream
->tfa
));
2120 /* Put self-ref for this index due to error. */
2121 relay_index_put(index
);
2124 stream
->index_fd
= stream_fd_create(fd
);
2125 if (!stream
->index_fd
) {
2128 PERROR("Error closing FD %d", fd
);
2130 /* Put self-ref for this index due to error. */
2131 relay_index_put(index
);
2132 /* Will put the local ref. */
2137 if (relay_index_set_fd(index
, stream
->index_fd
, data_offset
)) {
2139 /* Put self-ref for this index due to error. */
2140 relay_index_put(index
);
2144 ret
= relay_index_try_flush(index
);
2146 tracefile_array_commit_seq(stream
->tfa
);
2147 stream
->index_received_seqcount
++;
2148 } else if (ret
> 0) {
2152 /* Put self-ref for this index due to error. */
2153 relay_index_put(index
);
2161 * relay_process_data: Process the data received on the data socket
2163 static int relay_process_data(struct relay_connection
*conn
)
2165 int ret
= 0, rotate_index
= 0;
2167 struct relay_stream
*stream
;
2168 struct lttcomm_relayd_data_hdr data_hdr
;
2170 uint64_t net_seq_num
;
2172 struct relay_session
*session
;
2173 bool new_stream
= false, close_requested
= false;
2175 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &data_hdr
,
2176 sizeof(struct lttcomm_relayd_data_hdr
), 0);
2179 /* Orderly shutdown. Not necessary to print an error. */
2180 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2182 ERR("Unable to receive data header on sock %d", conn
->sock
->fd
);
2188 stream_id
= be64toh(data_hdr
.stream_id
);
2189 stream
= stream_get_by_id(stream_id
);
2191 ERR("relay_process_data: Cannot find stream %" PRIu64
, stream_id
);
2195 session
= stream
->trace
->session
;
2196 data_size
= be32toh(data_hdr
.data_size
);
2197 if (data_buffer_size
< data_size
) {
2200 tmp_data_ptr
= realloc(data_buffer
, data_size
);
2201 if (!tmp_data_ptr
) {
2202 ERR("Allocating data buffer");
2205 goto end_stream_put
;
2207 data_buffer
= tmp_data_ptr
;
2208 data_buffer_size
= data_size
;
2210 memset(data_buffer
, 0, data_size
);
2212 net_seq_num
= be64toh(data_hdr
.net_seq_num
);
2214 DBG3("Receiving data of size %u for stream id %" PRIu64
" seqnum %" PRIu64
,
2215 data_size
, stream_id
, net_seq_num
);
2216 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
, data_size
, 0);
2219 /* Orderly shutdown. Not necessary to print an error. */
2220 DBG("Socket %d did an orderly shutdown", conn
->sock
->fd
);
2222 ERR("Socket %d error %d", conn
->sock
->fd
, ret
);
2225 goto end_stream_put
;
2228 pthread_mutex_lock(&stream
->lock
);
2230 /* Check if a rotation is needed. */
2231 if (stream
->tracefile_size
> 0 &&
2232 (stream
->tracefile_size_current
+ data_size
) >
2233 stream
->tracefile_size
) {
2234 uint64_t old_id
, new_id
;
2236 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
2237 tracefile_array_file_rotate(stream
->tfa
);
2239 /* new_id is updated by utils_rotate_stream_file. */
2242 ret
= utils_rotate_stream_file(stream
->path_name
,
2243 stream
->channel_name
, stream
->tracefile_size
,
2244 stream
->tracefile_count
, -1,
2245 -1, stream
->stream_fd
->fd
,
2246 &new_id
, &stream
->stream_fd
->fd
);
2248 ERR("Rotating stream output file");
2249 goto end_stream_unlock
;
2252 * Reset current size because we just performed a stream
2255 stream
->tracefile_size_current
= 0;
2260 * Index are handled in protocol version 2.4 and above. Also,
2261 * snapshot and index are NOT supported.
2263 if (session
->minor
>= 4 && !session
->snapshot
) {
2264 ret
= handle_index_data(stream
, net_seq_num
, rotate_index
);
2266 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2267 stream
->stream_handle
, net_seq_num
, ret
);
2268 goto end_stream_unlock
;
2272 /* Write data to stream output fd. */
2273 size_ret
= lttng_write(stream
->stream_fd
->fd
, data_buffer
, data_size
);
2274 if (size_ret
< data_size
) {
2275 ERR("Relay error writing data to file");
2277 goto end_stream_unlock
;
2280 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
2281 size_ret
, stream
->stream_handle
);
2283 ret
= write_padding_to_file(stream
->stream_fd
->fd
,
2284 be32toh(data_hdr
.padding_size
));
2286 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
2287 stream
->stream_handle
, net_seq_num
, ret
);
2288 goto end_stream_unlock
;
2290 stream
->tracefile_size_current
+=
2291 data_size
+ be32toh(data_hdr
.padding_size
);
2292 if (stream
->prev_seq
== -1ULL) {
2296 stream
->prev_seq
= net_seq_num
;
2299 close_requested
= stream
->close_requested
;
2300 pthread_mutex_unlock(&stream
->lock
);
2301 if (close_requested
) {
2302 try_stream_close(stream
);
2306 pthread_mutex_lock(&session
->lock
);
2307 uatomic_set(&session
->new_streams
, 1);
2308 pthread_mutex_unlock(&session
->lock
);
2316 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2320 (void) lttng_poll_del(events
, pollfd
);
2322 ret
= close(pollfd
);
2324 ERR("Closing pollfd %d", pollfd
);
2328 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
2329 int pollfd
, struct relay_connection
*conn
)
2331 const char *type_str
;
2333 switch (conn
->type
) {
2338 type_str
= "Control";
2340 case RELAY_VIEWER_COMMAND
:
2341 type_str
= "Viewer Command";
2343 case RELAY_VIEWER_NOTIFICATION
:
2344 type_str
= "Viewer Notification";
2347 type_str
= "Unknown";
2349 cleanup_connection_pollfd(events
, pollfd
);
2350 connection_put(conn
);
2351 DBG("%s connection closed with %d", type_str
, pollfd
);
2355 * This thread does the actual work
2357 static void *relay_thread_worker(void *data
)
2359 int ret
, err
= -1, last_seen_data_fd
= -1;
2361 struct lttng_poll_event events
;
2362 struct lttng_ht
*relay_connections_ht
;
2363 struct lttng_ht_iter iter
;
2364 struct lttcomm_relayd_hdr recv_hdr
;
2365 struct relay_connection
*destroy_conn
= NULL
;
2367 DBG("[thread] Relay worker started");
2369 rcu_register_thread();
2371 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
2373 if (testpoint(relayd_thread_worker
)) {
2374 goto error_testpoint
;
2377 health_code_update();
2379 /* table of connections indexed on socket */
2380 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2381 if (!relay_connections_ht
) {
2382 goto relay_connections_ht_error
;
2385 ret
= create_thread_poll_set(&events
, 2);
2387 goto error_poll_create
;
2390 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2397 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
2399 health_code_update();
2401 /* Infinite blocking call, waiting for transmission */
2402 DBG3("Relayd worker thread polling...");
2403 health_poll_entry();
2404 ret
= lttng_poll_wait(&events
, -1);
2408 * Restart interrupted system call.
2410 if (errno
== EINTR
) {
2419 * Process control. The control connection is
2420 * prioritized so we don't starve it with high
2421 * throughput tracing data on the data connection.
2423 for (i
= 0; i
< nb_fd
; i
++) {
2424 /* Fetch once the poll data */
2425 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2426 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2428 health_code_update();
2432 * No activity for this FD (poll
2438 /* Thread quit pipe has been closed. Killing thread. */
2439 ret
= check_thread_quit_pipe(pollfd
, revents
);
2445 /* Inspect the relay conn pipe for new connection */
2446 if (pollfd
== relay_conn_pipe
[0]) {
2447 if (revents
& LPOLLIN
) {
2448 struct relay_connection
*conn
;
2450 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
2454 lttng_poll_add(&events
, conn
->sock
->fd
,
2455 LPOLLIN
| LPOLLRDHUP
);
2456 connection_ht_add(relay_connections_ht
, conn
);
2457 DBG("Connection socket %d added", conn
->sock
->fd
);
2458 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2459 ERR("Relay connection pipe error");
2462 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2466 struct relay_connection
*ctrl_conn
;
2468 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
2469 /* If not found, there is a synchronization issue. */
2472 if (ctrl_conn
->type
== RELAY_DATA
) {
2473 if (revents
& LPOLLIN
) {
2475 * Flag the last seen data fd not deleted. It will be
2476 * used as the last seen fd if any fd gets deleted in
2479 last_notdel_data_fd
= pollfd
;
2481 goto put_ctrl_connection
;
2483 assert(ctrl_conn
->type
== RELAY_CONTROL
);
2485 if (revents
& LPOLLIN
) {
2486 ret
= ctrl_conn
->sock
->ops
->recvmsg(ctrl_conn
->sock
,
2487 &recv_hdr
, sizeof(recv_hdr
), 0);
2489 /* Connection closed */
2490 relay_thread_close_connection(&events
, pollfd
,
2493 ret
= relay_process_control(&recv_hdr
, ctrl_conn
);
2495 /* Clear the session on error. */
2496 relay_thread_close_connection(&events
,
2501 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2502 relay_thread_close_connection(&events
,
2504 if (last_seen_data_fd
== pollfd
) {
2505 last_seen_data_fd
= last_notdel_data_fd
;
2508 ERR("Unexpected poll events %u for control sock %d",
2510 connection_put(ctrl_conn
);
2513 put_ctrl_connection
:
2514 connection_put(ctrl_conn
);
2519 * The last loop handled a control request, go back to poll to make
2520 * sure we prioritise the control socket.
2526 if (last_seen_data_fd
>= 0) {
2527 for (i
= 0; i
< nb_fd
; i
++) {
2528 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2530 health_code_update();
2532 if (last_seen_data_fd
== pollfd
) {
2539 /* Process data connection. */
2540 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
2541 /* Fetch the poll data. */
2542 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2543 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2544 struct relay_connection
*data_conn
;
2546 health_code_update();
2549 /* No activity for this FD (poll implementation). */
2553 /* Skip the command pipe. It's handled in the first loop. */
2554 if (pollfd
== relay_conn_pipe
[0]) {
2558 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
2560 /* Skip it. Might be removed before. */
2563 if (data_conn
->type
== RELAY_CONTROL
) {
2564 goto put_data_connection
;
2566 assert(data_conn
->type
== RELAY_DATA
);
2568 if (revents
& LPOLLIN
) {
2569 ret
= relay_process_data(data_conn
);
2570 /* Connection closed */
2572 relay_thread_close_connection(&events
, pollfd
,
2575 * Every goto restart call sets the last seen fd where
2576 * here we don't really care since we gracefully
2577 * continue the loop after the connection is deleted.
2580 /* Keep last seen port. */
2581 last_seen_data_fd
= pollfd
;
2582 connection_put(data_conn
);
2585 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2586 relay_thread_close_connection(&events
, pollfd
,
2589 ERR("Unknown poll events %u for data sock %d",
2592 put_data_connection
:
2593 connection_put(data_conn
);
2595 last_seen_data_fd
= -1;
2598 /* Normal exit, no error */
2603 /* Cleanup reamaining connection object. */
2605 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
2608 health_code_update();
2610 * No need to grab another ref, because we own
2613 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
2618 lttng_poll_clean(&events
);
2620 lttng_ht_destroy(relay_connections_ht
);
2621 relay_connections_ht_error
:
2622 /* Close relay conn pipes */
2623 utils_close_pipe(relay_conn_pipe
);
2625 DBG("Thread exited with error");
2627 DBG("Worker thread cleanup complete");
2632 ERR("Health error occurred in %s", __func__
);
2634 health_unregister(health_relayd
);
2635 rcu_unregister_thread();
2636 lttng_relay_stop_threads();
2641 * Create the relay command pipe to wake thread_manage_apps.
2642 * Closed in cleanup().
2644 static int create_relay_conn_pipe(void)
2648 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
2656 int main(int argc
, char **argv
)
2658 int ret
= 0, retval
= 0;
2661 /* Parse arguments */
2663 if (set_options(argc
, argv
)) {
2668 if (set_signal_handler()) {
2673 /* Try to create directory if -o, --output is specified. */
2674 if (opt_output_path
) {
2675 if (*opt_output_path
!= '/') {
2676 ERR("Please specify an absolute path for -o, --output PATH");
2681 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
2684 ERR("Unable to create %s", opt_output_path
);
2691 if (opt_daemon
|| opt_background
) {
2694 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
2702 * We are in the child. Make sure all other file
2703 * descriptors are closed, in case we are called with
2704 * more opened file descriptors than the standard ones.
2706 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
2712 /* Initialize thread health monitoring */
2713 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
2714 if (!health_relayd
) {
2715 PERROR("health_app_create error");
2717 goto exit_health_app_create
;
2720 /* Create thread quit pipe */
2721 if (init_thread_quit_pipe()) {
2723 goto exit_init_data
;
2726 /* Check if daemon is UID = 0 */
2728 if (control_uri
->port
< 1024 || data_uri
->port
< 1024 || live_uri
->port
< 1024) {
2729 ERR("Need to be root to use ports < 1024");
2731 goto exit_init_data
;
2735 /* Setup the thread apps communication pipe. */
2736 if (create_relay_conn_pipe()) {
2738 goto exit_init_data
;
2741 /* Init relay command queue. */
2742 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
2744 /* Set up max poll set size */
2745 if (lttng_poll_set_max_size()) {
2747 goto exit_init_data
;
2750 /* Initialize communication library */
2752 lttcomm_inet_init();
2754 /* tables of sessions indexed by session ID */
2755 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2758 goto exit_init_data
;
2761 /* tables of streams indexed by stream ID */
2762 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2763 if (!relay_streams_ht
) {
2765 goto exit_init_data
;
2768 /* tables of streams indexed by stream ID */
2769 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2770 if (!viewer_streams_ht
) {
2772 goto exit_init_data
;
2775 ret
= utils_create_pipe(health_quit_pipe
);
2778 goto exit_health_quit_pipe
;
2781 /* Create thread to manage the client socket */
2782 ret
= pthread_create(&health_thread
, NULL
,
2783 thread_manage_health
, (void *) NULL
);
2786 PERROR("pthread_create health");
2788 goto exit_health_thread
;
2791 /* Setup the dispatcher thread */
2792 ret
= pthread_create(&dispatcher_thread
, NULL
,
2793 relay_thread_dispatcher
, (void *) NULL
);
2796 PERROR("pthread_create dispatcher");
2798 goto exit_dispatcher_thread
;
2801 /* Setup the worker thread */
2802 ret
= pthread_create(&worker_thread
, NULL
,
2803 relay_thread_worker
, NULL
);
2806 PERROR("pthread_create worker");
2808 goto exit_worker_thread
;
2811 /* Setup the listener thread */
2812 ret
= pthread_create(&listener_thread
, NULL
,
2813 relay_thread_listener
, (void *) NULL
);
2816 PERROR("pthread_create listener");
2818 goto exit_listener_thread
;
2821 ret
= relayd_live_create(live_uri
);
2823 ERR("Starting live viewer threads");
2829 * This is where we start awaiting program completion (e.g. through
2830 * signal that asks threads to teardown).
2833 ret
= relayd_live_join();
2839 ret
= pthread_join(listener_thread
, &status
);
2842 PERROR("pthread_join listener_thread");
2846 exit_listener_thread
:
2847 ret
= pthread_join(worker_thread
, &status
);
2850 PERROR("pthread_join worker_thread");
2855 ret
= pthread_join(dispatcher_thread
, &status
);
2858 PERROR("pthread_join dispatcher_thread");
2861 exit_dispatcher_thread
:
2863 ret
= pthread_join(health_thread
, &status
);
2866 PERROR("pthread_join health_thread");
2871 utils_close_pipe(health_quit_pipe
);
2872 exit_health_quit_pipe
:
2875 health_app_destroy(health_relayd
);
2876 exit_health_app_create
:
2880 /* Ensure all prior call_rcu are done. */