2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@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,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
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.
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
34 #include <sys/types.h>
36 #include <urcu/uatomic.h>
40 #include <common/common.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
48 #include "lttng-sessiond.h"
49 #include "buffer-registry.h"
56 #include "kernel-consumer.h"
60 #include "ust-consumer.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
67 #define CONSUMERD_FILE "lttng-consumerd"
70 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
73 const char *opt_tracing_group
;
74 static const char *opt_pidfile
;
75 static int opt_sig_parent
;
76 static int opt_verbose_consumer
;
77 static int opt_daemon
;
78 static int opt_no_kernel
;
79 static int is_root
; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid
; /* Parent PID for --sig-parent option */
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
87 static struct consumer_data kconsumer_data
= {
88 .type
= LTTNG_CONSUMER_KERNEL
,
89 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
90 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
93 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
94 .lock
= PTHREAD_MUTEX_INITIALIZER
,
95 .cond
= PTHREAD_COND_INITIALIZER
,
96 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
98 static struct consumer_data ustconsumer64_data
= {
99 .type
= LTTNG_CONSUMER64_UST
,
100 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
101 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
104 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
105 .lock
= PTHREAD_MUTEX_INITIALIZER
,
106 .cond
= PTHREAD_COND_INITIALIZER
,
107 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
109 static struct consumer_data ustconsumer32_data
= {
110 .type
= LTTNG_CONSUMER32_UST
,
111 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
112 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
115 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
116 .lock
= PTHREAD_MUTEX_INITIALIZER
,
117 .cond
= PTHREAD_COND_INITIALIZER
,
118 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
121 /* Shared between threads */
122 static int dispatch_thread_exit
;
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path
[PATH_MAX
];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path
[PATH_MAX
];
128 /* global wait shm path for UST */
129 static char wait_shm_path
[PATH_MAX
];
130 /* Global health check unix path */
131 static char health_unix_sock_path
[PATH_MAX
];
133 /* Sockets and FDs */
134 static int client_sock
= -1;
135 static int apps_sock
= -1;
136 int kernel_tracer_fd
= -1;
137 static int kernel_poll_pipe
[2] = { -1, -1 };
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
143 static int thread_quit_pipe
[2] = { -1, -1 };
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
149 static int apps_cmd_pipe
[2] = { -1, -1 };
151 int apps_cmd_notify_pipe
[2] = { -1, -1 };
153 /* Pthread, Mutexes and Semaphores */
154 static pthread_t apps_thread
;
155 static pthread_t apps_notify_thread
;
156 static pthread_t reg_apps_thread
;
157 static pthread_t client_thread
;
158 static pthread_t kernel_thread
;
159 static pthread_t dispatch_thread
;
160 static pthread_t health_thread
;
161 static pthread_t ht_cleanup_thread
;
164 * UST registration command queue. This queue is tied with a futex and uses a N
165 * wakers / 1 waiter implemented and detailed in futex.c/.h
167 * The thread_manage_apps and thread_dispatch_ust_registration interact with
168 * this queue and the wait/wake scheme.
170 static struct ust_cmd_queue ust_cmd_queue
;
173 * Pointer initialized before thread creation.
175 * This points to the tracing session list containing the session count and a
176 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
177 * MUST NOT be taken if you call a public function in session.c.
179 * The lock is nested inside the structure: session_list_ptr->lock. Please use
180 * session_lock_list and session_unlock_list for lock acquisition.
182 static struct ltt_session_list
*session_list_ptr
;
184 int ust_consumerd64_fd
= -1;
185 int ust_consumerd32_fd
= -1;
187 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
188 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
189 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
190 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
192 static const char *module_proc_lttng
= "/proc/lttng";
195 * Consumer daemon state which is changed when spawning it, killing it or in
196 * case of a fatal error.
198 enum consumerd_state
{
199 CONSUMER_STARTED
= 1,
200 CONSUMER_STOPPED
= 2,
205 * This consumer daemon state is used to validate if a client command will be
206 * able to reach the consumer. If not, the client is informed. For instance,
207 * doing a "lttng start" when the consumer state is set to ERROR will return an
208 * error to the client.
210 * The following example shows a possible race condition of this scheme:
212 * consumer thread error happens
214 * client cmd checks state -> still OK
215 * consumer thread exit, sets error
216 * client cmd try to talk to consumer
219 * However, since the consumer is a different daemon, we have no way of making
220 * sure the command will reach it safely even with this state flag. This is why
221 * we consider that up to the state validation during command processing, the
222 * command is safe. After that, we can not guarantee the correctness of the
223 * client request vis-a-vis the consumer.
225 static enum consumerd_state ust_consumerd_state
;
226 static enum consumerd_state kernel_consumerd_state
;
229 * Socket timeout for receiving and sending in seconds.
231 static int app_socket_timeout
;
233 /* Set in main() with the current page size. */
237 void setup_consumerd_path(void)
239 const char *bin
, *libdir
;
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin
[0]) {
248 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
250 if (!consumerd32_libdir
[0]) {
251 consumerd32_libdir
= INSTALL_LIB_PATH
;
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin
[0]) {
255 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
257 if (!consumerd64_libdir
[0]) {
258 consumerd64_libdir
= INSTALL_LIB_PATH
;
261 #error "Unknown bitness"
265 * runtime env. var. overrides the build default.
267 bin
= getenv("LTTNG_CONSUMERD32_BIN");
269 consumerd32_bin
= bin
;
271 bin
= getenv("LTTNG_CONSUMERD64_BIN");
273 consumerd64_bin
= bin
;
275 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
277 consumerd32_libdir
= libdir
;
279 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
281 consumerd64_libdir
= libdir
;
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
288 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
294 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
300 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
312 * Check if the thread quit pipe was triggered.
314 * Return 1 if it was triggered else 0;
316 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
318 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
326 * Return group ID of the tracing group or -1 if not found.
328 static gid_t
allowed_group(void)
332 if (opt_tracing_group
) {
333 grp
= getgrnam(opt_tracing_group
);
335 grp
= getgrnam(default_tracing_group
);
345 * Init thread quit pipe.
347 * Return -1 on error or 0 if all pipes are created.
349 static int init_thread_quit_pipe(void)
353 ret
= pipe(thread_quit_pipe
);
355 PERROR("thread quit pipe");
359 for (i
= 0; i
< 2; i
++) {
360 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
372 * Stop all threads by closing the thread quit pipe.
374 static void stop_threads(void)
378 /* Stopping all threads */
379 DBG("Terminating all threads");
380 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
382 ERR("write error on thread quit pipe");
385 /* Dispatch thread */
386 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
387 futex_nto1_wake(&ust_cmd_queue
.futex
);
391 * Close every consumer sockets.
393 static void close_consumer_sockets(void)
397 if (kconsumer_data
.err_sock
>= 0) {
398 ret
= close(kconsumer_data
.err_sock
);
400 PERROR("kernel consumer err_sock close");
403 if (ustconsumer32_data
.err_sock
>= 0) {
404 ret
= close(ustconsumer32_data
.err_sock
);
406 PERROR("UST consumerd32 err_sock close");
409 if (ustconsumer64_data
.err_sock
>= 0) {
410 ret
= close(ustconsumer64_data
.err_sock
);
412 PERROR("UST consumerd64 err_sock close");
415 if (kconsumer_data
.cmd_sock
>= 0) {
416 ret
= close(kconsumer_data
.cmd_sock
);
418 PERROR("kernel consumer cmd_sock close");
421 if (ustconsumer32_data
.cmd_sock
>= 0) {
422 ret
= close(ustconsumer32_data
.cmd_sock
);
424 PERROR("UST consumerd32 cmd_sock close");
427 if (ustconsumer64_data
.cmd_sock
>= 0) {
428 ret
= close(ustconsumer64_data
.cmd_sock
);
430 PERROR("UST consumerd64 cmd_sock close");
438 static void cleanup(void)
442 struct ltt_session
*sess
, *stmp
;
447 * Close the thread quit pipe. It has already done its job,
448 * since we are now called.
450 utils_close_pipe(thread_quit_pipe
);
453 * If opt_pidfile is undefined, the default file will be wiped when
454 * removing the rundir.
457 ret
= remove(opt_pidfile
);
459 PERROR("remove pidfile %s", opt_pidfile
);
463 DBG("Removing %s directory", rundir
);
464 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
466 ERR("asprintf failed. Something is really wrong!");
469 /* Remove lttng run directory */
472 ERR("Unable to clean %s", rundir
);
477 DBG("Cleaning up all sessions");
479 /* Destroy session list mutex */
480 if (session_list_ptr
!= NULL
) {
481 pthread_mutex_destroy(&session_list_ptr
->lock
);
483 /* Cleanup ALL session */
484 cds_list_for_each_entry_safe(sess
, stmp
,
485 &session_list_ptr
->head
, list
) {
486 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
490 DBG("Closing all UST sockets");
491 ust_app_clean_list();
492 buffer_reg_destroy_registries();
494 if (is_root
&& !opt_no_kernel
) {
495 DBG2("Closing kernel fd");
496 if (kernel_tracer_fd
>= 0) {
497 ret
= close(kernel_tracer_fd
);
502 DBG("Unloading kernel modules");
503 modprobe_remove_lttng_all();
506 close_consumer_sockets();
509 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
510 "Matthew, BEET driven development works!%c[%dm",
511 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
516 * Send data on a unix socket using the liblttsessiondcomm API.
518 * Return lttcomm error code.
520 static int send_unix_sock(int sock
, void *buf
, size_t len
)
522 /* Check valid length */
527 return lttcomm_send_unix_sock(sock
, buf
, len
);
531 * Free memory of a command context structure.
533 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
535 DBG("Clean command context structure");
537 if ((*cmd_ctx
)->llm
) {
538 free((*cmd_ctx
)->llm
);
540 if ((*cmd_ctx
)->lsm
) {
541 free((*cmd_ctx
)->lsm
);
549 * Notify UST applications using the shm mmap futex.
551 static int notify_ust_apps(int active
)
555 DBG("Notifying applications of session daemon state: %d", active
);
557 /* See shm.c for this call implying mmap, shm and futex calls */
558 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
559 if (wait_shm_mmap
== NULL
) {
563 /* Wake waiting process */
564 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
566 /* Apps notified successfully */
574 * Setup the outgoing data buffer for the response (llm) by allocating the
575 * right amount of memory and copying the original information from the lsm
578 * Return total size of the buffer pointed by buf.
580 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
586 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
587 if (cmd_ctx
->llm
== NULL
) {
593 /* Copy common data */
594 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
595 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
597 cmd_ctx
->llm
->data_size
= size
;
598 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
607 * Update the kernel poll set of all channel fd available over all tracing
608 * session. Add the wakeup pipe at the end of the set.
610 static int update_kernel_poll(struct lttng_poll_event
*events
)
613 struct ltt_session
*session
;
614 struct ltt_kernel_channel
*channel
;
616 DBG("Updating kernel poll set");
619 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
620 session_lock(session
);
621 if (session
->kernel_session
== NULL
) {
622 session_unlock(session
);
626 cds_list_for_each_entry(channel
,
627 &session
->kernel_session
->channel_list
.head
, list
) {
628 /* Add channel fd to the kernel poll set */
629 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
631 session_unlock(session
);
634 DBG("Channel fd %d added to kernel set", channel
->fd
);
636 session_unlock(session
);
638 session_unlock_list();
643 session_unlock_list();
648 * Find the channel fd from 'fd' over all tracing session. When found, check
649 * for new channel stream and send those stream fds to the kernel consumer.
651 * Useful for CPU hotplug feature.
653 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
656 struct ltt_session
*session
;
657 struct ltt_kernel_session
*ksess
;
658 struct ltt_kernel_channel
*channel
;
660 DBG("Updating kernel streams for channel fd %d", fd
);
663 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
664 session_lock(session
);
665 if (session
->kernel_session
== NULL
) {
666 session_unlock(session
);
669 ksess
= session
->kernel_session
;
671 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
672 if (channel
->fd
== fd
) {
673 DBG("Channel found, updating kernel streams");
674 ret
= kernel_open_channel_stream(channel
);
678 /* Update the stream global counter */
679 ksess
->stream_count_global
+= ret
;
682 * Have we already sent fds to the consumer? If yes, it means
683 * that tracing is started so it is safe to send our updated
686 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
687 struct lttng_ht_iter iter
;
688 struct consumer_socket
*socket
;
691 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
692 &iter
.iter
, socket
, node
.node
) {
693 pthread_mutex_lock(socket
->lock
);
694 ret
= kernel_consumer_send_channel_stream(socket
,
696 session
->output_traces
? 1 : 0);
697 pthread_mutex_unlock(socket
->lock
);
708 session_unlock(session
);
710 session_unlock_list();
714 session_unlock(session
);
715 session_unlock_list();
720 * For each tracing session, update newly registered apps. The session list
721 * lock MUST be acquired before calling this.
723 static void update_ust_app(int app_sock
)
725 struct ltt_session
*sess
, *stmp
;
727 /* Consumer is in an ERROR state. Stop any application update. */
728 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
729 /* Stop the update process since the consumer is dead. */
733 /* For all tracing session(s) */
734 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
736 if (sess
->ust_session
) {
737 ust_app_global_update(sess
->ust_session
, app_sock
);
739 session_unlock(sess
);
744 * This thread manage event coming from the kernel.
746 * Features supported in this thread:
749 static void *thread_manage_kernel(void *data
)
751 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
752 uint32_t revents
, nb_fd
;
754 struct lttng_poll_event events
;
756 DBG("[thread] Thread manage kernel started");
758 health_register(HEALTH_TYPE_KERNEL
);
761 * This first step of the while is to clean this structure which could free
762 * non NULL pointers so initialize it before the loop.
764 lttng_poll_init(&events
);
766 if (testpoint(thread_manage_kernel
)) {
767 goto error_testpoint
;
770 health_code_update();
772 if (testpoint(thread_manage_kernel_before_loop
)) {
773 goto error_testpoint
;
777 health_code_update();
779 if (update_poll_flag
== 1) {
780 /* Clean events object. We are about to populate it again. */
781 lttng_poll_clean(&events
);
783 ret
= sessiond_set_thread_pollset(&events
, 2);
785 goto error_poll_create
;
788 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
793 /* This will add the available kernel channel if any. */
794 ret
= update_kernel_poll(&events
);
798 update_poll_flag
= 0;
801 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
803 /* Poll infinite value of time */
806 ret
= lttng_poll_wait(&events
, -1);
810 * Restart interrupted system call.
812 if (errno
== EINTR
) {
816 } else if (ret
== 0) {
817 /* Should not happen since timeout is infinite */
818 ERR("Return value of poll is 0 with an infinite timeout.\n"
819 "This should not have happened! Continuing...");
825 for (i
= 0; i
< nb_fd
; i
++) {
826 /* Fetch once the poll data */
827 revents
= LTTNG_POLL_GETEV(&events
, i
);
828 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
830 health_code_update();
832 /* Thread quit pipe has been closed. Killing thread. */
833 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
839 /* Check for data on kernel pipe */
840 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
842 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
843 } while (ret
< 0 && errno
== EINTR
);
845 * Ret value is useless here, if this pipe gets any actions an
846 * update is required anyway.
848 update_poll_flag
= 1;
852 * New CPU detected by the kernel. Adding kernel stream to
853 * kernel session and updating the kernel consumer
855 if (revents
& LPOLLIN
) {
856 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
862 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
863 * and unregister kernel stream at this point.
872 lttng_poll_clean(&events
);
875 utils_close_pipe(kernel_poll_pipe
);
876 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
879 ERR("Health error occurred in %s", __func__
);
880 WARN("Kernel thread died unexpectedly. "
881 "Kernel tracing can continue but CPU hotplug is disabled.");
884 DBG("Kernel thread dying");
889 * Signal pthread condition of the consumer data that the thread.
891 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
893 pthread_mutex_lock(&data
->cond_mutex
);
896 * The state is set before signaling. It can be any value, it's the waiter
897 * job to correctly interpret this condition variable associated to the
898 * consumer pthread_cond.
900 * A value of 0 means that the corresponding thread of the consumer data
901 * was not started. 1 indicates that the thread has started and is ready
902 * for action. A negative value means that there was an error during the
905 data
->consumer_thread_is_ready
= state
;
906 (void) pthread_cond_signal(&data
->cond
);
908 pthread_mutex_unlock(&data
->cond_mutex
);
912 * This thread manage the consumer error sent back to the session daemon.
914 static void *thread_manage_consumer(void *data
)
916 int sock
= -1, i
, ret
, pollfd
, err
= -1;
917 uint32_t revents
, nb_fd
;
918 enum lttcomm_return_code code
;
919 struct lttng_poll_event events
;
920 struct consumer_data
*consumer_data
= data
;
922 DBG("[thread] Manage consumer started");
924 health_register(HEALTH_TYPE_CONSUMER
);
926 health_code_update();
929 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
930 * metadata_sock. Nothing more will be added to this poll set.
932 ret
= sessiond_set_thread_pollset(&events
, 3);
938 * The error socket here is already in a listening state which was done
939 * just before spawning this thread to avoid a race between the consumer
940 * daemon exec trying to connect and the listen() call.
942 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
947 health_code_update();
949 /* Infinite blocking call, waiting for transmission */
953 if (testpoint(thread_manage_consumer
)) {
957 ret
= lttng_poll_wait(&events
, -1);
961 * Restart interrupted system call.
963 if (errno
== EINTR
) {
971 for (i
= 0; i
< nb_fd
; i
++) {
972 /* Fetch once the poll data */
973 revents
= LTTNG_POLL_GETEV(&events
, i
);
974 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
976 health_code_update();
978 /* Thread quit pipe has been closed. Killing thread. */
979 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
985 /* Event on the registration socket */
986 if (pollfd
== consumer_data
->err_sock
) {
987 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
988 ERR("consumer err socket poll error");
994 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1000 * Set the CLOEXEC flag. Return code is useless because either way, the
1003 (void) utils_set_fd_cloexec(sock
);
1005 health_code_update();
1007 DBG2("Receiving code from consumer err_sock");
1009 /* Getting status code from kconsumerd */
1010 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1011 sizeof(enum lttcomm_return_code
));
1016 health_code_update();
1018 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1019 /* Connect both socket, command and metadata. */
1020 consumer_data
->cmd_sock
=
1021 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1022 consumer_data
->metadata_fd
=
1023 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1024 if (consumer_data
->cmd_sock
< 0
1025 || consumer_data
->metadata_fd
< 0) {
1026 PERROR("consumer connect cmd socket");
1027 /* On error, signal condition and quit. */
1028 signal_consumer_condition(consumer_data
, -1);
1031 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1032 /* Create metadata socket lock. */
1033 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1034 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1035 PERROR("zmalloc pthread mutex");
1039 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1041 signal_consumer_condition(consumer_data
, 1);
1042 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1043 DBG("Consumer metadata socket ready (fd: %d)",
1044 consumer_data
->metadata_fd
);
1046 ERR("consumer error when waiting for SOCK_READY : %s",
1047 lttcomm_get_readable_code(-code
));
1051 /* Remove the consumerd error sock since we've established a connexion */
1052 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1057 /* Add new accepted error socket. */
1058 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1063 /* Add metadata socket that is successfully connected. */
1064 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1065 LPOLLIN
| LPOLLRDHUP
);
1070 health_code_update();
1072 /* Infinite blocking call, waiting for transmission */
1075 health_poll_entry();
1076 ret
= lttng_poll_wait(&events
, -1);
1080 * Restart interrupted system call.
1082 if (errno
== EINTR
) {
1090 for (i
= 0; i
< nb_fd
; i
++) {
1091 /* Fetch once the poll data */
1092 revents
= LTTNG_POLL_GETEV(&events
, i
);
1093 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1095 health_code_update();
1097 /* Thread quit pipe has been closed. Killing thread. */
1098 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1104 if (pollfd
== sock
) {
1105 /* Event on the consumerd socket */
1106 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1107 ERR("consumer err socket second poll error");
1110 health_code_update();
1111 /* Wait for any kconsumerd error */
1112 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1113 sizeof(enum lttcomm_return_code
));
1115 ERR("consumer closed the command socket");
1119 ERR("consumer return code : %s",
1120 lttcomm_get_readable_code(-code
));
1123 } else if (pollfd
== consumer_data
->metadata_fd
) {
1124 /* UST metadata requests */
1125 ret
= ust_consumer_metadata_request(
1126 &consumer_data
->metadata_sock
);
1128 ERR("Handling metadata request");
1133 ERR("Unknown pollfd");
1137 health_code_update();
1143 * We lock here because we are about to close the sockets and some other
1144 * thread might be using them so get exclusive access which will abort all
1145 * other consumer command by other threads.
1147 pthread_mutex_lock(&consumer_data
->lock
);
1149 /* Immediately set the consumerd state to stopped */
1150 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1151 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1152 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1153 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1154 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1156 /* Code flow error... */
1160 if (consumer_data
->err_sock
>= 0) {
1161 ret
= close(consumer_data
->err_sock
);
1165 consumer_data
->err_sock
= -1;
1167 if (consumer_data
->cmd_sock
>= 0) {
1168 ret
= close(consumer_data
->cmd_sock
);
1172 consumer_data
->cmd_sock
= -1;
1174 if (*consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1175 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1188 unlink(consumer_data
->err_unix_sock_path
);
1189 unlink(consumer_data
->cmd_unix_sock_path
);
1190 consumer_data
->pid
= 0;
1191 pthread_mutex_unlock(&consumer_data
->lock
);
1193 /* Cleanup metadata socket mutex. */
1194 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1195 free(consumer_data
->metadata_sock
.lock
);
1197 lttng_poll_clean(&events
);
1201 ERR("Health error occurred in %s", __func__
);
1203 health_unregister();
1204 DBG("consumer thread cleanup completed");
1210 * This thread manage application communication.
1212 static void *thread_manage_apps(void *data
)
1214 int i
, ret
, pollfd
, err
= -1;
1215 uint32_t revents
, nb_fd
;
1216 struct lttng_poll_event events
;
1218 DBG("[thread] Manage application started");
1220 rcu_register_thread();
1221 rcu_thread_online();
1223 health_register(HEALTH_TYPE_APP_MANAGE
);
1225 if (testpoint(thread_manage_apps
)) {
1226 goto error_testpoint
;
1229 health_code_update();
1231 ret
= sessiond_set_thread_pollset(&events
, 2);
1233 goto error_poll_create
;
1236 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1241 if (testpoint(thread_manage_apps_before_loop
)) {
1245 health_code_update();
1248 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1250 /* Inifinite blocking call, waiting for transmission */
1252 health_poll_entry();
1253 ret
= lttng_poll_wait(&events
, -1);
1257 * Restart interrupted system call.
1259 if (errno
== EINTR
) {
1267 for (i
= 0; i
< nb_fd
; i
++) {
1268 /* Fetch once the poll data */
1269 revents
= LTTNG_POLL_GETEV(&events
, i
);
1270 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1272 health_code_update();
1274 /* Thread quit pipe has been closed. Killing thread. */
1275 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1281 /* Inspect the apps cmd pipe */
1282 if (pollfd
== apps_cmd_pipe
[0]) {
1283 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1284 ERR("Apps command pipe error");
1286 } else if (revents
& LPOLLIN
) {
1291 ret
= read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1292 } while (ret
< 0 && errno
== EINTR
);
1293 if (ret
< 0 || ret
< sizeof(sock
)) {
1294 PERROR("read apps cmd pipe");
1298 health_code_update();
1301 * We only monitor the error events of the socket. This
1302 * thread does not handle any incoming data from UST
1305 ret
= lttng_poll_add(&events
, sock
,
1306 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1312 * Set socket timeout for both receiving and ending.
1313 * app_socket_timeout is in seconds, whereas
1314 * lttcomm_setsockopt_rcv_timeout and
1315 * lttcomm_setsockopt_snd_timeout expect msec as
1318 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1319 app_socket_timeout
* 1000);
1320 (void) lttcomm_setsockopt_snd_timeout(sock
,
1321 app_socket_timeout
* 1000);
1323 DBG("Apps with sock %d added to poll set", sock
);
1325 health_code_update();
1331 * At this point, we know that a registered application made
1332 * the event at poll_wait.
1334 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1335 /* Removing from the poll set */
1336 ret
= lttng_poll_del(&events
, pollfd
);
1341 /* Socket closed on remote end. */
1342 ust_app_unregister(pollfd
);
1347 health_code_update();
1353 lttng_poll_clean(&events
);
1356 utils_close_pipe(apps_cmd_pipe
);
1357 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1360 * We don't clean the UST app hash table here since already registered
1361 * applications can still be controlled so let them be until the session
1362 * daemon dies or the applications stop.
1367 ERR("Health error occurred in %s", __func__
);
1369 health_unregister();
1370 DBG("Application communication apps thread cleanup complete");
1371 rcu_thread_offline();
1372 rcu_unregister_thread();
1377 * Send a socket to a thread This is called from the dispatch UST registration
1378 * thread once all sockets are set for the application.
1380 * The sock value can be invalid, we don't really care, the thread will handle
1381 * it and make the necessary cleanup if so.
1383 * On success, return 0 else a negative value being the errno message of the
1386 static int send_socket_to_thread(int fd
, int sock
)
1391 * It's possible that the FD is set as invalid with -1 concurrently just
1392 * before calling this function being a shutdown state of the thread.
1400 ret
= write(fd
, &sock
, sizeof(sock
));
1401 } while (ret
< 0 && errno
== EINTR
);
1402 if (ret
< 0 || ret
!= sizeof(sock
)) {
1403 PERROR("write apps pipe %d", fd
);
1410 /* All good. Don't send back the write positive ret value. */
1417 * Sanitize the wait queue of the dispatch registration thread meaning removing
1418 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1419 * notify socket is never received.
1421 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1423 int ret
, nb_fd
= 0, i
;
1424 unsigned int fd_added
= 0;
1425 struct lttng_poll_event events
;
1426 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1430 lttng_poll_init(&events
);
1432 /* Just skip everything for an empty queue. */
1433 if (!wait_queue
->count
) {
1437 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1442 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1443 &wait_queue
->head
, head
) {
1444 assert(wait_node
->app
);
1445 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1446 LPOLLHUP
| LPOLLERR
);
1459 * Poll but don't block so we can quickly identify the faulty events and
1460 * clean them afterwards from the wait queue.
1462 ret
= lttng_poll_wait(&events
, 0);
1468 for (i
= 0; i
< nb_fd
; i
++) {
1469 /* Get faulty FD. */
1470 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1471 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1473 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1474 &wait_queue
->head
, head
) {
1475 if (pollfd
== wait_node
->app
->sock
&&
1476 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1477 cds_list_del(&wait_node
->head
);
1478 wait_queue
->count
--;
1479 ust_app_destroy(wait_node
->app
);
1487 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1491 lttng_poll_clean(&events
);
1495 lttng_poll_clean(&events
);
1497 ERR("Unable to sanitize wait queue");
1502 * Dispatch request from the registration threads to the application
1503 * communication thread.
1505 static void *thread_dispatch_ust_registration(void *data
)
1508 struct cds_wfq_node
*node
;
1509 struct ust_command
*ust_cmd
= NULL
;
1510 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1511 struct ust_reg_wait_queue wait_queue
= {
1515 health_register(HEALTH_TYPE_APP_REG_DISPATCH
);
1517 health_code_update();
1519 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1521 DBG("[thread] Dispatch UST command started");
1523 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1524 health_code_update();
1526 /* Atomically prepare the queue futex */
1527 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1530 struct ust_app
*app
= NULL
;
1534 * Make sure we don't have node(s) that have hung up before receiving
1535 * the notify socket. This is to clean the list in order to avoid
1536 * memory leaks from notify socket that are never seen.
1538 sanitize_wait_queue(&wait_queue
);
1540 health_code_update();
1541 /* Dequeue command for registration */
1542 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1544 DBG("Woken up but nothing in the UST command queue");
1545 /* Continue thread execution */
1549 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1551 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1552 " gid:%d sock:%d name:%s (version %d.%d)",
1553 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1554 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1555 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1556 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1558 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1559 wait_node
= zmalloc(sizeof(*wait_node
));
1561 PERROR("zmalloc wait_node dispatch");
1562 ret
= close(ust_cmd
->sock
);
1564 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1566 lttng_fd_put(1, LTTNG_FD_APPS
);
1570 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1572 /* Create application object if socket is CMD. */
1573 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1575 if (!wait_node
->app
) {
1576 ret
= close(ust_cmd
->sock
);
1578 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1580 lttng_fd_put(1, LTTNG_FD_APPS
);
1586 * Add application to the wait queue so we can set the notify
1587 * socket before putting this object in the global ht.
1589 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1594 * We have to continue here since we don't have the notify
1595 * socket and the application MUST be added to the hash table
1596 * only at that moment.
1601 * Look for the application in the local wait queue and set the
1602 * notify socket if found.
1604 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1605 &wait_queue
.head
, head
) {
1606 health_code_update();
1607 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1608 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1609 cds_list_del(&wait_node
->head
);
1611 app
= wait_node
->app
;
1613 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1619 * With no application at this stage the received socket is
1620 * basically useless so close it before we free the cmd data
1621 * structure for good.
1624 ret
= close(ust_cmd
->sock
);
1626 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1628 lttng_fd_put(1, LTTNG_FD_APPS
);
1635 * @session_lock_list
1637 * Lock the global session list so from the register up to the
1638 * registration done message, no thread can see the application
1639 * and change its state.
1641 session_lock_list();
1645 * Add application to the global hash table. This needs to be
1646 * done before the update to the UST registry can locate the
1651 /* Set app version. This call will print an error if needed. */
1652 (void) ust_app_version(app
);
1654 /* Send notify socket through the notify pipe. */
1655 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1659 session_unlock_list();
1661 * No notify thread, stop the UST tracing. However, this is
1662 * not an internal error of the this thread thus setting
1663 * the health error code to a normal exit.
1670 * Update newly registered application with the tracing
1671 * registry info already enabled information.
1673 update_ust_app(app
->sock
);
1676 * Don't care about return value. Let the manage apps threads
1677 * handle app unregistration upon socket close.
1679 (void) ust_app_register_done(app
->sock
);
1682 * Even if the application socket has been closed, send the app
1683 * to the thread and unregistration will take place at that
1686 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1689 session_unlock_list();
1691 * No apps. thread, stop the UST tracing. However, this is
1692 * not an internal error of the this thread thus setting
1693 * the health error code to a normal exit.
1700 session_unlock_list();
1702 } while (node
!= NULL
);
1704 health_poll_entry();
1705 /* Futex wait on queue. Blocking call on futex() */
1706 futex_nto1_wait(&ust_cmd_queue
.futex
);
1709 /* Normal exit, no error */
1713 /* Clean up wait queue. */
1714 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1715 &wait_queue
.head
, head
) {
1716 cds_list_del(&wait_node
->head
);
1721 DBG("Dispatch thread dying");
1724 ERR("Health error occurred in %s", __func__
);
1726 health_unregister();
1731 * This thread manage application registration.
1733 static void *thread_registration_apps(void *data
)
1735 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1736 uint32_t revents
, nb_fd
;
1737 struct lttng_poll_event events
;
1739 * Get allocated in this thread, enqueued to a global queue, dequeued and
1740 * freed in the manage apps thread.
1742 struct ust_command
*ust_cmd
= NULL
;
1744 DBG("[thread] Manage application registration started");
1746 health_register(HEALTH_TYPE_APP_REG
);
1748 if (testpoint(thread_registration_apps
)) {
1749 goto error_testpoint
;
1752 ret
= lttcomm_listen_unix_sock(apps_sock
);
1758 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1759 * more will be added to this poll set.
1761 ret
= sessiond_set_thread_pollset(&events
, 2);
1763 goto error_create_poll
;
1766 /* Add the application registration socket */
1767 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1769 goto error_poll_add
;
1772 /* Notify all applications to register */
1773 ret
= notify_ust_apps(1);
1775 ERR("Failed to notify applications or create the wait shared memory.\n"
1776 "Execution continues but there might be problem for already\n"
1777 "running applications that wishes to register.");
1781 DBG("Accepting application registration");
1783 /* Inifinite blocking call, waiting for transmission */
1785 health_poll_entry();
1786 ret
= lttng_poll_wait(&events
, -1);
1790 * Restart interrupted system call.
1792 if (errno
== EINTR
) {
1800 for (i
= 0; i
< nb_fd
; i
++) {
1801 health_code_update();
1803 /* Fetch once the poll data */
1804 revents
= LTTNG_POLL_GETEV(&events
, i
);
1805 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1807 /* Thread quit pipe has been closed. Killing thread. */
1808 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1814 /* Event on the registration socket */
1815 if (pollfd
== apps_sock
) {
1816 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1817 ERR("Register apps socket poll error");
1819 } else if (revents
& LPOLLIN
) {
1820 sock
= lttcomm_accept_unix_sock(apps_sock
);
1826 * Set the CLOEXEC flag. Return code is useless because
1827 * either way, the show must go on.
1829 (void) utils_set_fd_cloexec(sock
);
1831 /* Create UST registration command for enqueuing */
1832 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1833 if (ust_cmd
== NULL
) {
1834 PERROR("ust command zmalloc");
1839 * Using message-based transmissions to ensure we don't
1840 * have to deal with partially received messages.
1842 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1844 ERR("Exhausted file descriptors allowed for applications.");
1854 health_code_update();
1855 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1858 /* Close socket of the application. */
1863 lttng_fd_put(LTTNG_FD_APPS
, 1);
1867 health_code_update();
1869 ust_cmd
->sock
= sock
;
1872 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1873 " gid:%d sock:%d name:%s (version %d.%d)",
1874 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1875 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1876 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1877 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1880 * Lock free enqueue the registration request. The red pill
1881 * has been taken! This apps will be part of the *system*.
1883 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1886 * Wake the registration queue futex. Implicit memory
1887 * barrier with the exchange in cds_wfq_enqueue.
1889 futex_nto1_wake(&ust_cmd_queue
.futex
);
1899 ERR("Health error occurred in %s", __func__
);
1902 /* Notify that the registration thread is gone */
1905 if (apps_sock
>= 0) {
1906 ret
= close(apps_sock
);
1916 lttng_fd_put(LTTNG_FD_APPS
, 1);
1918 unlink(apps_unix_sock_path
);
1921 lttng_poll_clean(&events
);
1925 DBG("UST Registration thread cleanup complete");
1926 health_unregister();
1932 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1933 * exec or it will fails.
1935 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1938 struct timespec timeout
;
1940 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1941 consumer_data
->consumer_thread_is_ready
= 0;
1943 /* Setup pthread condition */
1944 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1947 PERROR("pthread_condattr_init consumer data");
1952 * Set the monotonic clock in order to make sure we DO NOT jump in time
1953 * between the clock_gettime() call and the timedwait call. See bug #324
1954 * for a more details and how we noticed it.
1956 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1959 PERROR("pthread_condattr_setclock consumer data");
1963 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1966 PERROR("pthread_cond_init consumer data");
1970 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1973 PERROR("pthread_create consumer");
1978 /* We are about to wait on a pthread condition */
1979 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1981 /* Get time for sem_timedwait absolute timeout */
1982 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1984 * Set the timeout for the condition timed wait even if the clock gettime
1985 * call fails since we might loop on that call and we want to avoid to
1986 * increment the timeout too many times.
1988 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1991 * The following loop COULD be skipped in some conditions so this is why we
1992 * set ret to 0 in order to make sure at least one round of the loop is
1998 * Loop until the condition is reached or when a timeout is reached. Note
1999 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2000 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2001 * possible. This loop does not take any chances and works with both of
2004 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2005 if (clock_ret
< 0) {
2006 PERROR("clock_gettime spawn consumer");
2007 /* Infinite wait for the consumerd thread to be ready */
2008 ret
= pthread_cond_wait(&consumer_data
->cond
,
2009 &consumer_data
->cond_mutex
);
2011 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2012 &consumer_data
->cond_mutex
, &timeout
);
2016 /* Release the pthread condition */
2017 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2021 if (ret
== ETIMEDOUT
) {
2023 * Call has timed out so we kill the kconsumerd_thread and return
2026 ERR("Condition timed out. The consumer thread was never ready."
2028 ret
= pthread_cancel(consumer_data
->thread
);
2030 PERROR("pthread_cancel consumer thread");
2033 PERROR("pthread_cond_wait failed consumer thread");
2038 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2039 if (consumer_data
->pid
== 0) {
2040 ERR("Consumerd did not start");
2041 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2044 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2053 * Join consumer thread
2055 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2059 /* Consumer pid must be a real one. */
2060 if (consumer_data
->pid
> 0) {
2062 ret
= kill(consumer_data
->pid
, SIGTERM
);
2064 ERR("Error killing consumer daemon");
2067 return pthread_join(consumer_data
->thread
, &status
);
2074 * Fork and exec a consumer daemon (consumerd).
2076 * Return pid if successful else -1.
2078 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2082 const char *consumer_to_use
;
2083 const char *verbosity
;
2086 DBG("Spawning consumerd");
2093 if (opt_verbose_consumer
) {
2094 verbosity
= "--verbose";
2096 verbosity
= "--quiet";
2098 switch (consumer_data
->type
) {
2099 case LTTNG_CONSUMER_KERNEL
:
2101 * Find out which consumerd to execute. We will first try the
2102 * 64-bit path, then the sessiond's installation directory, and
2103 * fallback on the 32-bit one,
2105 DBG3("Looking for a kernel consumer at these locations:");
2106 DBG3(" 1) %s", consumerd64_bin
);
2107 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2108 DBG3(" 3) %s", consumerd32_bin
);
2109 if (stat(consumerd64_bin
, &st
) == 0) {
2110 DBG3("Found location #1");
2111 consumer_to_use
= consumerd64_bin
;
2112 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2113 DBG3("Found location #2");
2114 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2115 } else if (stat(consumerd32_bin
, &st
) == 0) {
2116 DBG3("Found location #3");
2117 consumer_to_use
= consumerd32_bin
;
2119 DBG("Could not find any valid consumerd executable");
2122 DBG("Using kernel consumer at: %s", consumer_to_use
);
2123 execl(consumer_to_use
,
2124 "lttng-consumerd", verbosity
, "-k",
2125 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2126 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2129 case LTTNG_CONSUMER64_UST
:
2131 char *tmpnew
= NULL
;
2133 if (consumerd64_libdir
[0] != '\0') {
2137 tmp
= getenv("LD_LIBRARY_PATH");
2141 tmplen
= strlen("LD_LIBRARY_PATH=")
2142 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2143 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2148 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2149 strcat(tmpnew
, consumerd64_libdir
);
2150 if (tmp
[0] != '\0') {
2151 strcat(tmpnew
, ":");
2152 strcat(tmpnew
, tmp
);
2154 ret
= putenv(tmpnew
);
2161 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2162 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2163 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2164 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2166 if (consumerd64_libdir
[0] != '\0') {
2174 case LTTNG_CONSUMER32_UST
:
2176 char *tmpnew
= NULL
;
2178 if (consumerd32_libdir
[0] != '\0') {
2182 tmp
= getenv("LD_LIBRARY_PATH");
2186 tmplen
= strlen("LD_LIBRARY_PATH=")
2187 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2188 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2193 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2194 strcat(tmpnew
, consumerd32_libdir
);
2195 if (tmp
[0] != '\0') {
2196 strcat(tmpnew
, ":");
2197 strcat(tmpnew
, tmp
);
2199 ret
= putenv(tmpnew
);
2206 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2207 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2208 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2209 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2211 if (consumerd32_libdir
[0] != '\0') {
2220 PERROR("unknown consumer type");
2224 PERROR("kernel start consumer exec");
2227 } else if (pid
> 0) {
2230 PERROR("start consumer fork");
2238 * Spawn the consumerd daemon and session daemon thread.
2240 static int start_consumerd(struct consumer_data
*consumer_data
)
2245 * Set the listen() state on the socket since there is a possible race
2246 * between the exec() of the consumer daemon and this call if place in the
2247 * consumer thread. See bug #366 for more details.
2249 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2254 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2255 if (consumer_data
->pid
!= 0) {
2256 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2260 ret
= spawn_consumerd(consumer_data
);
2262 ERR("Spawning consumerd failed");
2263 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2267 /* Setting up the consumer_data pid */
2268 consumer_data
->pid
= ret
;
2269 DBG2("Consumer pid %d", consumer_data
->pid
);
2270 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2272 DBG2("Spawning consumer control thread");
2273 ret
= spawn_consumer_thread(consumer_data
);
2275 ERR("Fatal error spawning consumer control thread");
2283 /* Cleanup already created sockets on error. */
2284 if (consumer_data
->err_sock
>= 0) {
2287 err
= close(consumer_data
->err_sock
);
2289 PERROR("close consumer data error socket");
2296 * Compute health status of each consumer. If one of them is zero (bad
2297 * state), we return 0.
2299 static int check_consumer_health(void)
2303 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2305 DBG3("Health consumer check %d", ret
);
2311 * Setup necessary data for kernel tracer action.
2313 static int init_kernel_tracer(void)
2317 /* Modprobe lttng kernel modules */
2318 ret
= modprobe_lttng_control();
2323 /* Open debugfs lttng */
2324 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2325 if (kernel_tracer_fd
< 0) {
2326 DBG("Failed to open %s", module_proc_lttng
);
2331 /* Validate kernel version */
2332 ret
= kernel_validate_version(kernel_tracer_fd
);
2337 ret
= modprobe_lttng_data();
2342 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2346 modprobe_remove_lttng_control();
2347 ret
= close(kernel_tracer_fd
);
2351 kernel_tracer_fd
= -1;
2352 return LTTNG_ERR_KERN_VERSION
;
2355 ret
= close(kernel_tracer_fd
);
2361 modprobe_remove_lttng_control();
2364 WARN("No kernel tracer available");
2365 kernel_tracer_fd
= -1;
2367 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2369 return LTTNG_ERR_KERN_NA
;
2375 * Copy consumer output from the tracing session to the domain session. The
2376 * function also applies the right modification on a per domain basis for the
2377 * trace files destination directory.
2379 * Should *NOT* be called with RCU read-side lock held.
2381 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2384 const char *dir_name
;
2385 struct consumer_output
*consumer
;
2388 assert(session
->consumer
);
2391 case LTTNG_DOMAIN_KERNEL
:
2392 DBG3("Copying tracing session consumer output in kernel session");
2394 * XXX: We should audit the session creation and what this function
2395 * does "extra" in order to avoid a destroy since this function is used
2396 * in the domain session creation (kernel and ust) only. Same for UST
2399 if (session
->kernel_session
->consumer
) {
2400 consumer_destroy_output(session
->kernel_session
->consumer
);
2402 session
->kernel_session
->consumer
=
2403 consumer_copy_output(session
->consumer
);
2404 /* Ease our life a bit for the next part */
2405 consumer
= session
->kernel_session
->consumer
;
2406 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2408 case LTTNG_DOMAIN_UST
:
2409 DBG3("Copying tracing session consumer output in UST session");
2410 if (session
->ust_session
->consumer
) {
2411 consumer_destroy_output(session
->ust_session
->consumer
);
2413 session
->ust_session
->consumer
=
2414 consumer_copy_output(session
->consumer
);
2415 /* Ease our life a bit for the next part */
2416 consumer
= session
->ust_session
->consumer
;
2417 dir_name
= DEFAULT_UST_TRACE_DIR
;
2420 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2424 /* Append correct directory to subdir */
2425 strncat(consumer
->subdir
, dir_name
,
2426 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2427 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2436 * Create an UST session and add it to the session ust list.
2438 * Should *NOT* be called with RCU read-side lock held.
2440 static int create_ust_session(struct ltt_session
*session
,
2441 struct lttng_domain
*domain
)
2444 struct ltt_ust_session
*lus
= NULL
;
2448 assert(session
->consumer
);
2450 switch (domain
->type
) {
2451 case LTTNG_DOMAIN_UST
:
2454 ERR("Unknown UST domain on create session %d", domain
->type
);
2455 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2459 DBG("Creating UST session");
2461 lus
= trace_ust_create_session(session
->id
);
2463 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2467 lus
->uid
= session
->uid
;
2468 lus
->gid
= session
->gid
;
2469 lus
->output_traces
= session
->output_traces
;
2470 lus
->snapshot_mode
= session
->snapshot_mode
;
2471 session
->ust_session
= lus
;
2473 /* Copy session output to the newly created UST session */
2474 ret
= copy_session_consumer(domain
->type
, session
);
2475 if (ret
!= LTTNG_OK
) {
2483 session
->ust_session
= NULL
;
2488 * Create a kernel tracer session then create the default channel.
2490 static int create_kernel_session(struct ltt_session
*session
)
2494 DBG("Creating kernel session");
2496 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2498 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2502 /* Code flow safety */
2503 assert(session
->kernel_session
);
2505 /* Copy session output to the newly created Kernel session */
2506 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2507 if (ret
!= LTTNG_OK
) {
2511 /* Create directory(ies) on local filesystem. */
2512 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2513 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2514 ret
= run_as_mkdir_recursive(
2515 session
->kernel_session
->consumer
->dst
.trace_path
,
2516 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2518 if (ret
!= -EEXIST
) {
2519 ERR("Trace directory creation error");
2525 session
->kernel_session
->uid
= session
->uid
;
2526 session
->kernel_session
->gid
= session
->gid
;
2527 session
->kernel_session
->output_traces
= session
->output_traces
;
2528 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2533 trace_kernel_destroy_session(session
->kernel_session
);
2534 session
->kernel_session
= NULL
;
2539 * Count number of session permitted by uid/gid.
2541 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2544 struct ltt_session
*session
;
2546 DBG("Counting number of available session for UID %d GID %d",
2548 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2550 * Only list the sessions the user can control.
2552 if (!session_access_ok(session
, uid
, gid
)) {
2561 * Process the command requested by the lttng client within the command
2562 * context structure. This function make sure that the return structure (llm)
2563 * is set and ready for transmission before returning.
2565 * Return any error encountered or 0 for success.
2567 * "sock" is only used for special-case var. len data.
2569 * Should *NOT* be called with RCU read-side lock held.
2571 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2575 int need_tracing_session
= 1;
2578 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2582 switch (cmd_ctx
->lsm
->cmd_type
) {
2583 case LTTNG_CREATE_SESSION
:
2584 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2585 case LTTNG_DESTROY_SESSION
:
2586 case LTTNG_LIST_SESSIONS
:
2587 case LTTNG_LIST_DOMAINS
:
2588 case LTTNG_START_TRACE
:
2589 case LTTNG_STOP_TRACE
:
2590 case LTTNG_DATA_PENDING
:
2591 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2592 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2593 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2594 case LTTNG_SNAPSHOT_RECORD
:
2601 if (opt_no_kernel
&& need_domain
2602 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2604 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2606 ret
= LTTNG_ERR_KERN_NA
;
2611 /* Deny register consumer if we already have a spawned consumer. */
2612 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2613 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2614 if (kconsumer_data
.pid
> 0) {
2615 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2616 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2619 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2623 * Check for command that don't needs to allocate a returned payload. We do
2624 * this here so we don't have to make the call for no payload at each
2627 switch(cmd_ctx
->lsm
->cmd_type
) {
2628 case LTTNG_LIST_SESSIONS
:
2629 case LTTNG_LIST_TRACEPOINTS
:
2630 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2631 case LTTNG_LIST_DOMAINS
:
2632 case LTTNG_LIST_CHANNELS
:
2633 case LTTNG_LIST_EVENTS
:
2636 /* Setup lttng message with no payload */
2637 ret
= setup_lttng_msg(cmd_ctx
, 0);
2639 /* This label does not try to unlock the session */
2640 goto init_setup_error
;
2644 /* Commands that DO NOT need a session. */
2645 switch (cmd_ctx
->lsm
->cmd_type
) {
2646 case LTTNG_CREATE_SESSION
:
2647 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2648 case LTTNG_CALIBRATE
:
2649 case LTTNG_LIST_SESSIONS
:
2650 case LTTNG_LIST_TRACEPOINTS
:
2651 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2652 need_tracing_session
= 0;
2655 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2657 * We keep the session list lock across _all_ commands
2658 * for now, because the per-session lock does not
2659 * handle teardown properly.
2661 session_lock_list();
2662 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2663 if (cmd_ctx
->session
== NULL
) {
2664 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2667 /* Acquire lock for the session */
2668 session_lock(cmd_ctx
->session
);
2678 * Check domain type for specific "pre-action".
2680 switch (cmd_ctx
->lsm
->domain
.type
) {
2681 case LTTNG_DOMAIN_KERNEL
:
2683 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2687 /* Kernel tracer check */
2688 if (kernel_tracer_fd
== -1) {
2689 /* Basically, load kernel tracer modules */
2690 ret
= init_kernel_tracer();
2696 /* Consumer is in an ERROR state. Report back to client */
2697 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2698 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2702 /* Need a session for kernel command */
2703 if (need_tracing_session
) {
2704 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2705 ret
= create_kernel_session(cmd_ctx
->session
);
2707 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2712 /* Start the kernel consumer daemon */
2713 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2714 if (kconsumer_data
.pid
== 0 &&
2715 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2716 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2717 ret
= start_consumerd(&kconsumer_data
);
2719 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2722 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2724 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2728 * The consumer was just spawned so we need to add the socket to
2729 * the consumer output of the session if exist.
2731 ret
= consumer_create_socket(&kconsumer_data
,
2732 cmd_ctx
->session
->kernel_session
->consumer
);
2739 case LTTNG_DOMAIN_UST
:
2741 if (!ust_app_supported()) {
2742 ret
= LTTNG_ERR_NO_UST
;
2745 /* Consumer is in an ERROR state. Report back to client */
2746 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2747 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2751 if (need_tracing_session
) {
2752 /* Create UST session if none exist. */
2753 if (cmd_ctx
->session
->ust_session
== NULL
) {
2754 ret
= create_ust_session(cmd_ctx
->session
,
2755 &cmd_ctx
->lsm
->domain
);
2756 if (ret
!= LTTNG_OK
) {
2761 /* Start the UST consumer daemons */
2763 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2764 if (consumerd64_bin
[0] != '\0' &&
2765 ustconsumer64_data
.pid
== 0 &&
2766 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2767 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2768 ret
= start_consumerd(&ustconsumer64_data
);
2770 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2771 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2775 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2776 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2778 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2782 * Setup socket for consumer 64 bit. No need for atomic access
2783 * since it was set above and can ONLY be set in this thread.
2785 ret
= consumer_create_socket(&ustconsumer64_data
,
2786 cmd_ctx
->session
->ust_session
->consumer
);
2792 if (consumerd32_bin
[0] != '\0' &&
2793 ustconsumer32_data
.pid
== 0 &&
2794 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2795 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2796 ret
= start_consumerd(&ustconsumer32_data
);
2798 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2799 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2803 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2804 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2806 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2810 * Setup socket for consumer 64 bit. No need for atomic access
2811 * since it was set above and can ONLY be set in this thread.
2813 ret
= consumer_create_socket(&ustconsumer32_data
,
2814 cmd_ctx
->session
->ust_session
->consumer
);
2826 /* Validate consumer daemon state when start/stop trace command */
2827 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2828 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2829 switch (cmd_ctx
->lsm
->domain
.type
) {
2830 case LTTNG_DOMAIN_UST
:
2831 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2832 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2836 case LTTNG_DOMAIN_KERNEL
:
2837 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2838 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2846 * Check that the UID or GID match that of the tracing session.
2847 * The root user can interact with all sessions.
2849 if (need_tracing_session
) {
2850 if (!session_access_ok(cmd_ctx
->session
,
2851 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2852 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2853 ret
= LTTNG_ERR_EPERM
;
2859 * Send relayd information to consumer as soon as we have a domain and a
2862 if (cmd_ctx
->session
&& need_domain
) {
2864 * Setup relayd if not done yet. If the relayd information was already
2865 * sent to the consumer, this call will gracefully return.
2867 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2868 if (ret
!= LTTNG_OK
) {
2873 /* Process by command type */
2874 switch (cmd_ctx
->lsm
->cmd_type
) {
2875 case LTTNG_ADD_CONTEXT
:
2877 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2878 cmd_ctx
->lsm
->u
.context
.channel_name
,
2879 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2882 case LTTNG_DISABLE_CHANNEL
:
2884 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2885 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2888 case LTTNG_DISABLE_EVENT
:
2890 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2891 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2892 cmd_ctx
->lsm
->u
.disable
.name
);
2895 case LTTNG_DISABLE_ALL_EVENT
:
2897 DBG("Disabling all events");
2899 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2900 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2903 case LTTNG_ENABLE_CHANNEL
:
2905 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2906 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2909 case LTTNG_ENABLE_EVENT
:
2911 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2912 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2913 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2916 case LTTNG_ENABLE_ALL_EVENT
:
2918 DBG("Enabling all events");
2920 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2921 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2922 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2925 case LTTNG_LIST_TRACEPOINTS
:
2927 struct lttng_event
*events
;
2930 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2931 if (nb_events
< 0) {
2932 /* Return value is a negative lttng_error_code. */
2938 * Setup lttng message with payload size set to the event list size in
2939 * bytes and then copy list into the llm payload.
2941 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2947 /* Copy event list into message payload */
2948 memcpy(cmd_ctx
->llm
->payload
, events
,
2949 sizeof(struct lttng_event
) * nb_events
);
2956 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2958 struct lttng_event_field
*fields
;
2961 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2963 if (nb_fields
< 0) {
2964 /* Return value is a negative lttng_error_code. */
2970 * Setup lttng message with payload size set to the event list size in
2971 * bytes and then copy list into the llm payload.
2973 ret
= setup_lttng_msg(cmd_ctx
,
2974 sizeof(struct lttng_event_field
) * nb_fields
);
2980 /* Copy event list into message payload */
2981 memcpy(cmd_ctx
->llm
->payload
, fields
,
2982 sizeof(struct lttng_event_field
) * nb_fields
);
2989 case LTTNG_SET_CONSUMER_URI
:
2992 struct lttng_uri
*uris
;
2994 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2995 len
= nb_uri
* sizeof(struct lttng_uri
);
2998 ret
= LTTNG_ERR_INVALID
;
3002 uris
= zmalloc(len
);
3004 ret
= LTTNG_ERR_FATAL
;
3008 /* Receive variable len data */
3009 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3010 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3012 DBG("No URIs received from client... continuing");
3014 ret
= LTTNG_ERR_SESSION_FAIL
;
3019 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3021 if (ret
!= LTTNG_OK
) {
3027 * XXX: 0 means that this URI should be applied on the session. Should
3028 * be a DOMAIN enuam.
3030 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3031 /* Add the URI for the UST session if a consumer is present. */
3032 if (cmd_ctx
->session
->ust_session
&&
3033 cmd_ctx
->session
->ust_session
->consumer
) {
3034 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3036 } else if (cmd_ctx
->session
->kernel_session
&&
3037 cmd_ctx
->session
->kernel_session
->consumer
) {
3038 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3039 cmd_ctx
->session
, nb_uri
, uris
);
3047 case LTTNG_START_TRACE
:
3049 ret
= cmd_start_trace(cmd_ctx
->session
);
3052 case LTTNG_STOP_TRACE
:
3054 ret
= cmd_stop_trace(cmd_ctx
->session
);
3057 case LTTNG_CREATE_SESSION
:
3060 struct lttng_uri
*uris
= NULL
;
3062 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3063 len
= nb_uri
* sizeof(struct lttng_uri
);
3066 uris
= zmalloc(len
);
3068 ret
= LTTNG_ERR_FATAL
;
3072 /* Receive variable len data */
3073 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3074 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3076 DBG("No URIs received from client... continuing");
3078 ret
= LTTNG_ERR_SESSION_FAIL
;
3083 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3084 DBG("Creating session with ONE network URI is a bad call");
3085 ret
= LTTNG_ERR_SESSION_FAIL
;
3091 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3098 case LTTNG_DESTROY_SESSION
:
3100 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3102 /* Set session to NULL so we do not unlock it after free. */
3103 cmd_ctx
->session
= NULL
;
3106 case LTTNG_LIST_DOMAINS
:
3109 struct lttng_domain
*domains
;
3111 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3113 /* Return value is a negative lttng_error_code. */
3118 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3124 /* Copy event list into message payload */
3125 memcpy(cmd_ctx
->llm
->payload
, domains
,
3126 nb_dom
* sizeof(struct lttng_domain
));
3133 case LTTNG_LIST_CHANNELS
:
3136 struct lttng_channel
*channels
;
3138 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3139 cmd_ctx
->session
, &channels
);
3141 /* Return value is a negative lttng_error_code. */
3146 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3152 /* Copy event list into message payload */
3153 memcpy(cmd_ctx
->llm
->payload
, channels
,
3154 nb_chan
* sizeof(struct lttng_channel
));
3161 case LTTNG_LIST_EVENTS
:
3164 struct lttng_event
*events
= NULL
;
3166 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3167 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3169 /* Return value is a negative lttng_error_code. */
3174 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3180 /* Copy event list into message payload */
3181 memcpy(cmd_ctx
->llm
->payload
, events
,
3182 nb_event
* sizeof(struct lttng_event
));
3189 case LTTNG_LIST_SESSIONS
:
3191 unsigned int nr_sessions
;
3193 session_lock_list();
3194 nr_sessions
= lttng_sessions_count(
3195 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3196 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3198 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3200 session_unlock_list();
3204 /* Filled the session array */
3205 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3206 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3207 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3209 session_unlock_list();
3214 case LTTNG_CALIBRATE
:
3216 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3217 &cmd_ctx
->lsm
->u
.calibrate
);
3220 case LTTNG_REGISTER_CONSUMER
:
3222 struct consumer_data
*cdata
;
3224 switch (cmd_ctx
->lsm
->domain
.type
) {
3225 case LTTNG_DOMAIN_KERNEL
:
3226 cdata
= &kconsumer_data
;
3229 ret
= LTTNG_ERR_UND
;
3233 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3234 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3237 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3239 struct lttng_filter_bytecode
*bytecode
;
3241 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3242 ret
= LTTNG_ERR_FILTER_INVAL
;
3245 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3246 ret
= LTTNG_ERR_FILTER_INVAL
;
3249 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3251 ret
= LTTNG_ERR_FILTER_NOMEM
;
3254 /* Receive var. len. data */
3255 DBG("Receiving var len data from client ...");
3256 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3257 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3259 DBG("Nothing recv() from client var len data... continuing");
3261 ret
= LTTNG_ERR_FILTER_INVAL
;
3265 if (bytecode
->len
+ sizeof(*bytecode
)
3266 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3268 ret
= LTTNG_ERR_FILTER_INVAL
;
3272 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3273 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3274 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3277 case LTTNG_DATA_PENDING
:
3279 ret
= cmd_data_pending(cmd_ctx
->session
);
3282 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3284 struct lttcomm_lttng_output_id reply
;
3286 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3287 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3288 if (ret
!= LTTNG_OK
) {
3292 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3297 /* Copy output list into message payload */
3298 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3302 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3304 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3305 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3308 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3311 struct lttng_snapshot_output
*outputs
= NULL
;
3313 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3314 if (nb_output
< 0) {
3319 ret
= setup_lttng_msg(cmd_ctx
,
3320 nb_output
* sizeof(struct lttng_snapshot_output
));
3327 /* Copy output list into message payload */
3328 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3329 nb_output
* sizeof(struct lttng_snapshot_output
));
3336 case LTTNG_SNAPSHOT_RECORD
:
3338 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3339 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3340 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3343 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3346 struct lttng_uri
*uris
= NULL
;
3348 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3349 len
= nb_uri
* sizeof(struct lttng_uri
);
3352 uris
= zmalloc(len
);
3354 ret
= LTTNG_ERR_FATAL
;
3358 /* Receive variable len data */
3359 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3360 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3362 DBG("No URIs received from client... continuing");
3364 ret
= LTTNG_ERR_SESSION_FAIL
;
3369 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3370 DBG("Creating session with ONE network URI is a bad call");
3371 ret
= LTTNG_ERR_SESSION_FAIL
;
3377 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3378 nb_uri
, &cmd_ctx
->creds
);
3383 ret
= LTTNG_ERR_UND
;
3388 if (cmd_ctx
->llm
== NULL
) {
3389 DBG("Missing llm structure. Allocating one.");
3390 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3394 /* Set return code */
3395 cmd_ctx
->llm
->ret_code
= ret
;
3397 if (cmd_ctx
->session
) {
3398 session_unlock(cmd_ctx
->session
);
3400 if (need_tracing_session
) {
3401 session_unlock_list();
3408 * Thread managing health check socket.
3410 static void *thread_manage_health(void *data
)
3412 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3413 uint32_t revents
, nb_fd
;
3414 struct lttng_poll_event events
;
3415 struct lttcomm_health_msg msg
;
3416 struct lttcomm_health_data reply
;
3418 DBG("[thread] Manage health check started");
3420 rcu_register_thread();
3422 /* We might hit an error path before this is created. */
3423 lttng_poll_init(&events
);
3425 /* Create unix socket */
3426 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3428 ERR("Unable to create health check Unix socket");
3434 * Set the CLOEXEC flag. Return code is useless because either way, the
3437 (void) utils_set_fd_cloexec(sock
);
3439 ret
= lttcomm_listen_unix_sock(sock
);
3445 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3446 * more will be added to this poll set.
3448 ret
= sessiond_set_thread_pollset(&events
, 2);
3453 /* Add the application registration socket */
3454 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3460 DBG("Health check ready");
3462 /* Inifinite blocking call, waiting for transmission */
3464 ret
= lttng_poll_wait(&events
, -1);
3467 * Restart interrupted system call.
3469 if (errno
== EINTR
) {
3477 for (i
= 0; i
< nb_fd
; i
++) {
3478 /* Fetch once the poll data */
3479 revents
= LTTNG_POLL_GETEV(&events
, i
);
3480 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3482 /* Thread quit pipe has been closed. Killing thread. */
3483 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3489 /* Event on the registration socket */
3490 if (pollfd
== sock
) {
3491 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3492 ERR("Health socket poll error");
3498 new_sock
= lttcomm_accept_unix_sock(sock
);
3504 * Set the CLOEXEC flag. Return code is useless because either way, the
3507 (void) utils_set_fd_cloexec(new_sock
);
3509 DBG("Receiving data from client for health...");
3510 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3512 DBG("Nothing recv() from client... continuing");
3513 ret
= close(new_sock
);
3521 rcu_thread_online();
3523 switch (msg
.component
) {
3524 case LTTNG_HEALTH_CMD
:
3525 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3527 case LTTNG_HEALTH_APP_MANAGE
:
3528 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3530 case LTTNG_HEALTH_APP_REG
:
3531 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3533 case LTTNG_HEALTH_KERNEL
:
3534 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3536 case LTTNG_HEALTH_CONSUMER
:
3537 reply
.ret_code
= check_consumer_health();
3539 case LTTNG_HEALTH_HT_CLEANUP
:
3540 reply
.ret_code
= health_check_state(HEALTH_TYPE_HT_CLEANUP
);
3542 case LTTNG_HEALTH_APP_MANAGE_NOTIFY
:
3543 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3545 case LTTNG_HEALTH_APP_REG_DISPATCH
:
3546 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3548 case LTTNG_HEALTH_ALL
:
3550 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3551 health_check_state(HEALTH_TYPE_APP_REG
) &&
3552 health_check_state(HEALTH_TYPE_CMD
) &&
3553 health_check_state(HEALTH_TYPE_KERNEL
) &&
3554 check_consumer_health() &&
3555 health_check_state(HEALTH_TYPE_HT_CLEANUP
) &&
3556 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
) &&
3557 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3560 reply
.ret_code
= LTTNG_ERR_UND
;
3565 * Flip ret value since 0 is a success and 1 indicates a bad health for
3566 * the client where in the sessiond it is the opposite. Again, this is
3567 * just to make things easier for us poor developer which enjoy a lot
3570 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3571 reply
.ret_code
= !reply
.ret_code
;
3574 DBG2("Health check return value %d", reply
.ret_code
);
3576 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3578 ERR("Failed to send health data back to client");
3581 /* End of transmission */
3582 ret
= close(new_sock
);
3592 ERR("Health error occurred in %s", __func__
);
3594 DBG("Health check thread dying");
3595 unlink(health_unix_sock_path
);
3603 lttng_poll_clean(&events
);
3605 rcu_unregister_thread();
3610 * This thread manage all clients request using the unix client socket for
3613 static void *thread_manage_clients(void *data
)
3615 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3617 uint32_t revents
, nb_fd
;
3618 struct command_ctx
*cmd_ctx
= NULL
;
3619 struct lttng_poll_event events
;
3621 DBG("[thread] Manage client started");
3623 rcu_register_thread();
3625 health_register(HEALTH_TYPE_CMD
);
3627 if (testpoint(thread_manage_clients
)) {
3628 goto error_testpoint
;
3631 health_code_update();
3633 ret
= lttcomm_listen_unix_sock(client_sock
);
3639 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3640 * more will be added to this poll set.
3642 ret
= sessiond_set_thread_pollset(&events
, 2);
3644 goto error_create_poll
;
3647 /* Add the application registration socket */
3648 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3654 * Notify parent pid that we are ready to accept command for client side.
3656 if (opt_sig_parent
) {
3657 kill(ppid
, SIGUSR1
);
3660 if (testpoint(thread_manage_clients_before_loop
)) {
3664 health_code_update();
3667 DBG("Accepting client command ...");
3669 /* Inifinite blocking call, waiting for transmission */
3671 health_poll_entry();
3672 ret
= lttng_poll_wait(&events
, -1);
3676 * Restart interrupted system call.
3678 if (errno
== EINTR
) {
3686 for (i
= 0; i
< nb_fd
; i
++) {
3687 /* Fetch once the poll data */
3688 revents
= LTTNG_POLL_GETEV(&events
, i
);
3689 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3691 health_code_update();
3693 /* Thread quit pipe has been closed. Killing thread. */
3694 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3700 /* Event on the registration socket */
3701 if (pollfd
== client_sock
) {
3702 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3703 ERR("Client socket poll error");
3709 DBG("Wait for client response");
3711 health_code_update();
3713 sock
= lttcomm_accept_unix_sock(client_sock
);
3719 * Set the CLOEXEC flag. Return code is useless because either way, the
3722 (void) utils_set_fd_cloexec(sock
);
3724 /* Set socket option for credentials retrieval */
3725 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3730 /* Allocate context command to process the client request */
3731 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3732 if (cmd_ctx
== NULL
) {
3733 PERROR("zmalloc cmd_ctx");
3737 /* Allocate data buffer for reception */
3738 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3739 if (cmd_ctx
->lsm
== NULL
) {
3740 PERROR("zmalloc cmd_ctx->lsm");
3744 cmd_ctx
->llm
= NULL
;
3745 cmd_ctx
->session
= NULL
;
3747 health_code_update();
3750 * Data is received from the lttng client. The struct
3751 * lttcomm_session_msg (lsm) contains the command and data request of
3754 DBG("Receiving data from client ...");
3755 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3756 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3758 DBG("Nothing recv() from client... continuing");
3764 clean_command_ctx(&cmd_ctx
);
3768 health_code_update();
3770 // TODO: Validate cmd_ctx including sanity check for
3771 // security purpose.
3773 rcu_thread_online();
3775 * This function dispatch the work to the kernel or userspace tracer
3776 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3777 * informations for the client. The command context struct contains
3778 * everything this function may needs.
3780 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3781 rcu_thread_offline();
3789 * TODO: Inform client somehow of the fatal error. At
3790 * this point, ret < 0 means that a zmalloc failed
3791 * (ENOMEM). Error detected but still accept
3792 * command, unless a socket error has been
3795 clean_command_ctx(&cmd_ctx
);
3799 health_code_update();
3801 DBG("Sending response (size: %d, retcode: %s)",
3802 cmd_ctx
->lttng_msg_size
,
3803 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3804 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3806 ERR("Failed to send data back to client");
3809 /* End of transmission */
3816 clean_command_ctx(&cmd_ctx
);
3818 health_code_update();
3830 lttng_poll_clean(&events
);
3831 clean_command_ctx(&cmd_ctx
);
3836 unlink(client_unix_sock_path
);
3837 if (client_sock
>= 0) {
3838 ret
= close(client_sock
);
3846 ERR("Health error occurred in %s", __func__
);
3849 health_unregister();
3851 DBG("Client thread dying");
3853 rcu_unregister_thread();
3859 * usage function on stderr
3861 static void usage(void)
3863 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3864 fprintf(stderr
, " -h, --help Display this usage.\n");
3865 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3866 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3867 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3868 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3869 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3870 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3871 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3872 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3873 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3874 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3875 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3876 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3877 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3878 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3879 fprintf(stderr
, " -V, --version Show version number.\n");
3880 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3881 fprintf(stderr
, " -q, --quiet No output at all.\n");
3882 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3883 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3884 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3885 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3889 * daemon argument parsing
3891 static int parse_args(int argc
, char **argv
)
3895 static struct option long_options
[] = {
3896 { "client-sock", 1, 0, 'c' },
3897 { "apps-sock", 1, 0, 'a' },
3898 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3899 { "kconsumerd-err-sock", 1, 0, 'E' },
3900 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3901 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3902 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3903 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3904 { "consumerd32-path", 1, 0, 'u' },
3905 { "consumerd32-libdir", 1, 0, 'U' },
3906 { "consumerd64-path", 1, 0, 't' },
3907 { "consumerd64-libdir", 1, 0, 'T' },
3908 { "daemonize", 0, 0, 'd' },
3909 { "sig-parent", 0, 0, 'S' },
3910 { "help", 0, 0, 'h' },
3911 { "group", 1, 0, 'g' },
3912 { "version", 0, 0, 'V' },
3913 { "quiet", 0, 0, 'q' },
3914 { "verbose", 0, 0, 'v' },
3915 { "verbose-consumer", 0, 0, 'Z' },
3916 { "no-kernel", 0, 0, 'N' },
3917 { "pidfile", 1, 0, 'p' },
3922 int option_index
= 0;
3923 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3924 long_options
, &option_index
);
3931 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3933 fprintf(stderr
, " with arg %s\n", optarg
);
3937 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3940 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3946 opt_tracing_group
= optarg
;
3952 fprintf(stdout
, "%s\n", VERSION
);
3958 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3961 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3964 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3967 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3970 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3973 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3979 lttng_opt_quiet
= 1;
3982 /* Verbose level can increase using multiple -v */
3983 lttng_opt_verbose
+= 1;
3986 opt_verbose_consumer
+= 1;
3989 consumerd32_bin
= optarg
;
3992 consumerd32_libdir
= optarg
;
3995 consumerd64_bin
= optarg
;
3998 consumerd64_libdir
= optarg
;
4001 opt_pidfile
= optarg
;
4004 /* Unknown option or other error.
4005 * Error is printed by getopt, just return */
4014 * Creates the two needed socket by the daemon.
4015 * apps_sock - The communication socket for all UST apps.
4016 * client_sock - The communication of the cli tool (lttng).
4018 static int init_daemon_socket(void)
4023 old_umask
= umask(0);
4025 /* Create client tool unix socket */
4026 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4027 if (client_sock
< 0) {
4028 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4033 /* Set the cloexec flag */
4034 ret
= utils_set_fd_cloexec(client_sock
);
4036 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4037 "Continuing but note that the consumer daemon will have a "
4038 "reference to this socket on exec()", client_sock
);
4041 /* File permission MUST be 660 */
4042 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4044 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4049 /* Create the application unix socket */
4050 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4051 if (apps_sock
< 0) {
4052 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4057 /* Set the cloexec flag */
4058 ret
= utils_set_fd_cloexec(apps_sock
);
4060 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4061 "Continuing but note that the consumer daemon will have a "
4062 "reference to this socket on exec()", apps_sock
);
4065 /* File permission MUST be 666 */
4066 ret
= chmod(apps_unix_sock_path
,
4067 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4069 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4074 DBG3("Session daemon client socket %d and application socket %d created",
4075 client_sock
, apps_sock
);
4083 * Check if the global socket is available, and if a daemon is answering at the
4084 * other side. If yes, error is returned.
4086 static int check_existing_daemon(void)
4088 /* Is there anybody out there ? */
4089 if (lttng_session_daemon_alive()) {
4097 * Set the tracing group gid onto the client socket.
4099 * Race window between mkdir and chown is OK because we are going from more
4100 * permissive (root.root) to less permissive (root.tracing).
4102 static int set_permissions(char *rundir
)
4107 ret
= allowed_group();
4109 WARN("No tracing group detected");
4110 /* Setting gid to 0 if no tracing group is found */
4116 /* Set lttng run dir */
4117 ret
= chown(rundir
, 0, gid
);
4119 ERR("Unable to set group on %s", rundir
);
4123 /* Ensure all applications and tracing group can search the run dir */
4124 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
4126 ERR("Unable to set permissions on %s", rundir
);
4130 /* lttng client socket path */
4131 ret
= chown(client_unix_sock_path
, 0, gid
);
4133 ERR("Unable to set group on %s", client_unix_sock_path
);
4137 /* kconsumer error socket path */
4138 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
4140 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4144 /* 64-bit ustconsumer error socket path */
4145 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
4147 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4151 /* 32-bit ustconsumer compat32 error socket path */
4152 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
4154 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4158 DBG("All permissions are set");
4164 * Create the lttng run directory needed for all global sockets and pipe.
4166 static int create_lttng_rundir(const char *rundir
)
4170 DBG3("Creating LTTng run directory: %s", rundir
);
4172 ret
= mkdir(rundir
, S_IRWXU
);
4174 if (errno
!= EEXIST
) {
4175 ERR("Unable to create %s", rundir
);
4187 * Setup sockets and directory needed by the kconsumerd communication with the
4190 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4194 char path
[PATH_MAX
];
4196 switch (consumer_data
->type
) {
4197 case LTTNG_CONSUMER_KERNEL
:
4198 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4200 case LTTNG_CONSUMER64_UST
:
4201 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4203 case LTTNG_CONSUMER32_UST
:
4204 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4207 ERR("Consumer type unknown");
4212 DBG2("Creating consumer directory: %s", path
);
4214 ret
= mkdir(path
, S_IRWXU
);
4216 if (errno
!= EEXIST
) {
4218 ERR("Failed to create %s", path
);
4224 /* Create the kconsumerd error unix socket */
4225 consumer_data
->err_sock
=
4226 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4227 if (consumer_data
->err_sock
< 0) {
4228 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4234 * Set the CLOEXEC flag. Return code is useless because either way, the
4237 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4239 PERROR("utils_set_fd_cloexec");
4240 /* continue anyway */
4243 /* File permission MUST be 660 */
4244 ret
= chmod(consumer_data
->err_unix_sock_path
,
4245 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4247 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4257 * Signal handler for the daemon
4259 * Simply stop all worker threads, leaving main() return gracefully after
4260 * joining all threads and calling cleanup().
4262 static void sighandler(int sig
)
4266 DBG("SIGPIPE caught");
4269 DBG("SIGINT caught");
4273 DBG("SIGTERM caught");
4282 * Setup signal handler for :
4283 * SIGINT, SIGTERM, SIGPIPE
4285 static int set_signal_handler(void)
4288 struct sigaction sa
;
4291 if ((ret
= sigemptyset(&sigset
)) < 0) {
4292 PERROR("sigemptyset");
4296 sa
.sa_handler
= sighandler
;
4297 sa
.sa_mask
= sigset
;
4299 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4300 PERROR("sigaction");
4304 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4305 PERROR("sigaction");
4309 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4310 PERROR("sigaction");
4314 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4320 * Set open files limit to unlimited. This daemon can open a large number of
4321 * file descriptors in order to consumer multiple kernel traces.
4323 static void set_ulimit(void)
4328 /* The kernel does not allowed an infinite limit for open files */
4329 lim
.rlim_cur
= 65535;
4330 lim
.rlim_max
= 65535;
4332 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4334 PERROR("failed to set open files limit");
4339 * Write pidfile using the rundir and opt_pidfile.
4341 static void write_pidfile(void)
4344 char pidfile_path
[PATH_MAX
];
4349 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4351 /* Build pidfile path from rundir and opt_pidfile. */
4352 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4353 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4355 PERROR("snprintf pidfile path");
4361 * Create pid file in rundir. Return value is of no importance. The
4362 * execution will continue even though we are not able to write the file.
4364 (void) utils_create_pid_file(getpid(), pidfile_path
);
4373 int main(int argc
, char **argv
)
4377 const char *home_path
, *env_app_timeout
;
4379 init_kernel_workarounds();
4381 rcu_register_thread();
4383 setup_consumerd_path();
4385 page_size
= sysconf(_SC_PAGESIZE
);
4386 if (page_size
< 0) {
4387 PERROR("sysconf _SC_PAGESIZE");
4388 page_size
= LONG_MAX
;
4389 WARN("Fallback page size to %ld", page_size
);
4392 /* Parse arguments */
4394 if ((ret
= parse_args(argc
, argv
)) < 0) {
4404 * child: setsid, close FD 0, 1, 2, chdir /
4405 * parent: exit (if fork is successful)
4413 * We are in the child. Make sure all other file
4414 * descriptors are closed, in case we are called with
4415 * more opened file descriptors than the standard ones.
4417 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4422 /* Create thread quit pipe */
4423 if ((ret
= init_thread_quit_pipe()) < 0) {
4427 /* Check if daemon is UID = 0 */
4428 is_root
= !getuid();
4431 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4433 /* Create global run dir with root access */
4434 ret
= create_lttng_rundir(rundir
);
4439 if (strlen(apps_unix_sock_path
) == 0) {
4440 snprintf(apps_unix_sock_path
, PATH_MAX
,
4441 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4444 if (strlen(client_unix_sock_path
) == 0) {
4445 snprintf(client_unix_sock_path
, PATH_MAX
,
4446 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4449 /* Set global SHM for ust */
4450 if (strlen(wait_shm_path
) == 0) {
4451 snprintf(wait_shm_path
, PATH_MAX
,
4452 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4455 if (strlen(health_unix_sock_path
) == 0) {
4456 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4457 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4460 /* Setup kernel consumerd path */
4461 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4462 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4463 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4464 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4466 DBG2("Kernel consumer err path: %s",
4467 kconsumer_data
.err_unix_sock_path
);
4468 DBG2("Kernel consumer cmd path: %s",
4469 kconsumer_data
.cmd_unix_sock_path
);
4471 home_path
= utils_get_home_dir();
4472 if (home_path
== NULL
) {
4473 /* TODO: Add --socket PATH option */
4474 ERR("Can't get HOME directory for sockets creation.");
4480 * Create rundir from home path. This will create something like
4483 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4489 ret
= create_lttng_rundir(rundir
);
4494 if (strlen(apps_unix_sock_path
) == 0) {
4495 snprintf(apps_unix_sock_path
, PATH_MAX
,
4496 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4499 /* Set the cli tool unix socket path */
4500 if (strlen(client_unix_sock_path
) == 0) {
4501 snprintf(client_unix_sock_path
, PATH_MAX
,
4502 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4505 /* Set global SHM for ust */
4506 if (strlen(wait_shm_path
) == 0) {
4507 snprintf(wait_shm_path
, PATH_MAX
,
4508 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4511 /* Set health check Unix path */
4512 if (strlen(health_unix_sock_path
) == 0) {
4513 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4514 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4518 /* Set consumer initial state */
4519 kernel_consumerd_state
= CONSUMER_STOPPED
;
4520 ust_consumerd_state
= CONSUMER_STOPPED
;
4522 DBG("Client socket path %s", client_unix_sock_path
);
4523 DBG("Application socket path %s", apps_unix_sock_path
);
4524 DBG("Application wait path %s", wait_shm_path
);
4525 DBG("LTTng run directory path: %s", rundir
);
4527 /* 32 bits consumerd path setup */
4528 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4529 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4530 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4531 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4533 DBG2("UST consumer 32 bits err path: %s",
4534 ustconsumer32_data
.err_unix_sock_path
);
4535 DBG2("UST consumer 32 bits cmd path: %s",
4536 ustconsumer32_data
.cmd_unix_sock_path
);
4538 /* 64 bits consumerd path setup */
4539 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4540 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4541 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4542 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4544 DBG2("UST consumer 64 bits err path: %s",
4545 ustconsumer64_data
.err_unix_sock_path
);
4546 DBG2("UST consumer 64 bits cmd path: %s",
4547 ustconsumer64_data
.cmd_unix_sock_path
);
4550 * See if daemon already exist.
4552 if ((ret
= check_existing_daemon()) < 0) {
4553 ERR("Already running daemon.\n");
4555 * We do not goto exit because we must not cleanup()
4556 * because a daemon is already running.
4562 * Init UST app hash table. Alloc hash table before this point since
4563 * cleanup() can get called after that point.
4567 /* After this point, we can safely call cleanup() with "goto exit" */
4570 * These actions must be executed as root. We do that *after* setting up
4571 * the sockets path because we MUST make the check for another daemon using
4572 * those paths *before* trying to set the kernel consumer sockets and init
4576 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4581 /* Setup kernel tracer */
4582 if (!opt_no_kernel
) {
4583 init_kernel_tracer();
4586 /* Set ulimit for open files */
4589 /* init lttng_fd tracking must be done after set_ulimit. */
4592 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4597 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4602 if ((ret
= set_signal_handler()) < 0) {
4606 /* Setup the needed unix socket */
4607 if ((ret
= init_daemon_socket()) < 0) {
4611 /* Set credentials to socket */
4612 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4616 /* Get parent pid if -S, --sig-parent is specified. */
4617 if (opt_sig_parent
) {
4621 /* Setup the kernel pipe for waking up the kernel thread */
4622 if (is_root
&& !opt_no_kernel
) {
4623 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4628 /* Setup the thread ht_cleanup communication pipe. */
4629 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4633 /* Setup the thread apps communication pipe. */
4634 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4638 /* Setup the thread apps notify communication pipe. */
4639 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4643 /* Initialize global buffer per UID and PID registry. */
4644 buffer_reg_init_uid_registry();
4645 buffer_reg_init_pid_registry();
4647 /* Init UST command queue. */
4648 cds_wfq_init(&ust_cmd_queue
.queue
);
4651 * Get session list pointer. This pointer MUST NOT be free(). This list is
4652 * statically declared in session.c
4654 session_list_ptr
= session_get_list();
4656 /* Set up max poll set size */
4657 lttng_poll_set_max_size();
4661 /* Check for the application socket timeout env variable. */
4662 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4663 if (env_app_timeout
) {
4664 app_socket_timeout
= atoi(env_app_timeout
);
4666 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4671 /* Initialize communication library */
4673 /* This is to get the TCP timeout value. */
4674 lttcomm_inet_init();
4677 * Initialize the health check subsystem. This call should set the
4678 * appropriate time values.
4682 /* Create thread to manage the client socket */
4683 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4684 thread_ht_cleanup
, (void *) NULL
);
4686 PERROR("pthread_create ht_cleanup");
4687 goto exit_ht_cleanup
;
4690 /* Create thread to manage the client socket */
4691 ret
= pthread_create(&health_thread
, NULL
,
4692 thread_manage_health
, (void *) NULL
);
4694 PERROR("pthread_create health");
4698 /* Create thread to manage the client socket */
4699 ret
= pthread_create(&client_thread
, NULL
,
4700 thread_manage_clients
, (void *) NULL
);
4702 PERROR("pthread_create clients");
4706 /* Create thread to dispatch registration */
4707 ret
= pthread_create(&dispatch_thread
, NULL
,
4708 thread_dispatch_ust_registration
, (void *) NULL
);
4710 PERROR("pthread_create dispatch");
4714 /* Create thread to manage application registration. */
4715 ret
= pthread_create(®_apps_thread
, NULL
,
4716 thread_registration_apps
, (void *) NULL
);
4718 PERROR("pthread_create registration");
4722 /* Create thread to manage application socket */
4723 ret
= pthread_create(&apps_thread
, NULL
,
4724 thread_manage_apps
, (void *) NULL
);
4726 PERROR("pthread_create apps");
4730 /* Create thread to manage application notify socket */
4731 ret
= pthread_create(&apps_notify_thread
, NULL
,
4732 ust_thread_manage_notify
, (void *) NULL
);
4734 PERROR("pthread_create apps");
4738 /* Don't start this thread if kernel tracing is not requested nor root */
4739 if (is_root
&& !opt_no_kernel
) {
4740 /* Create kernel thread to manage kernel event */
4741 ret
= pthread_create(&kernel_thread
, NULL
,
4742 thread_manage_kernel
, (void *) NULL
);
4744 PERROR("pthread_create kernel");
4748 ret
= pthread_join(kernel_thread
, &status
);
4750 PERROR("pthread_join");
4751 goto error
; /* join error, exit without cleanup */
4756 ret
= pthread_join(apps_thread
, &status
);
4758 PERROR("pthread_join");
4759 goto error
; /* join error, exit without cleanup */
4763 ret
= pthread_join(reg_apps_thread
, &status
);
4765 PERROR("pthread_join");
4766 goto error
; /* join error, exit without cleanup */
4770 ret
= pthread_join(dispatch_thread
, &status
);
4772 PERROR("pthread_join");
4773 goto error
; /* join error, exit without cleanup */
4777 ret
= pthread_join(client_thread
, &status
);
4779 PERROR("pthread_join");
4780 goto error
; /* join error, exit without cleanup */
4783 ret
= join_consumer_thread(&kconsumer_data
);
4785 PERROR("join_consumer");
4786 goto error
; /* join error, exit without cleanup */
4789 ret
= join_consumer_thread(&ustconsumer32_data
);
4791 PERROR("join_consumer ust32");
4792 goto error
; /* join error, exit without cleanup */
4795 ret
= join_consumer_thread(&ustconsumer64_data
);
4797 PERROR("join_consumer ust64");
4798 goto error
; /* join error, exit without cleanup */
4802 ret
= pthread_join(health_thread
, &status
);
4804 PERROR("pthread_join health thread");
4805 goto error
; /* join error, exit without cleanup */
4809 ret
= pthread_join(ht_cleanup_thread
, &status
);
4811 PERROR("pthread_join ht cleanup thread");
4812 goto error
; /* join error, exit without cleanup */
4817 * cleanup() is called when no other thread is running.
4819 rcu_thread_online();
4821 rcu_thread_offline();
4822 rcu_unregister_thread();