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 * On success, return 0 else a negative value being the errno message of the
1383 static int send_socket_to_thread(int fd
, int sock
)
1387 /* Sockets MUST be set or else this should not have been called. */
1392 ret
= write(fd
, &sock
, sizeof(sock
));
1393 } while (ret
< 0 && errno
== EINTR
);
1394 if (ret
< 0 || ret
!= sizeof(sock
)) {
1395 PERROR("write apps pipe %d", fd
);
1402 /* All good. Don't send back the write positive ret value. */
1409 * Sanitize the wait queue of the dispatch registration thread meaning removing
1410 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1411 * notify socket is never received.
1413 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1415 int ret
, nb_fd
= 0, i
;
1416 unsigned int fd_added
= 0;
1417 struct lttng_poll_event events
;
1418 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1422 lttng_poll_init(&events
);
1424 /* Just skip everything for an empty queue. */
1425 if (!wait_queue
->count
) {
1429 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1434 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1435 &wait_queue
->head
, head
) {
1436 assert(wait_node
->app
);
1437 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1438 LPOLLHUP
| LPOLLERR
);
1451 * Poll but don't block so we can quickly identify the faulty events and
1452 * clean them afterwards from the wait queue.
1454 ret
= lttng_poll_wait(&events
, 0);
1460 for (i
= 0; i
< nb_fd
; i
++) {
1461 /* Get faulty FD. */
1462 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1463 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1465 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1466 &wait_queue
->head
, head
) {
1467 if (pollfd
== wait_node
->app
->sock
&&
1468 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1469 cds_list_del(&wait_node
->head
);
1470 wait_queue
->count
--;
1471 ust_app_destroy(wait_node
->app
);
1479 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1483 lttng_poll_clean(&events
);
1487 lttng_poll_clean(&events
);
1489 ERR("Unable to sanitize wait queue");
1494 * Dispatch request from the registration threads to the application
1495 * communication thread.
1497 static void *thread_dispatch_ust_registration(void *data
)
1500 struct cds_wfq_node
*node
;
1501 struct ust_command
*ust_cmd
= NULL
;
1502 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1503 struct ust_reg_wait_queue wait_queue
= {
1507 health_register(HEALTH_TYPE_APP_REG_DISPATCH
);
1509 health_code_update();
1511 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1513 DBG("[thread] Dispatch UST command started");
1515 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1516 health_code_update();
1518 /* Atomically prepare the queue futex */
1519 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1522 struct ust_app
*app
= NULL
;
1526 * Make sure we don't have node(s) that have hung up before receiving
1527 * the notify socket. This is to clean the list in order to avoid
1528 * memory leaks from notify socket that are never seen.
1530 sanitize_wait_queue(&wait_queue
);
1532 health_code_update();
1533 /* Dequeue command for registration */
1534 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1536 DBG("Woken up but nothing in the UST command queue");
1537 /* Continue thread execution */
1541 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1543 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1544 " gid:%d sock:%d name:%s (version %d.%d)",
1545 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1546 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1547 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1548 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1550 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1551 wait_node
= zmalloc(sizeof(*wait_node
));
1553 PERROR("zmalloc wait_node dispatch");
1554 ret
= close(ust_cmd
->sock
);
1556 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1558 lttng_fd_put(1, LTTNG_FD_APPS
);
1562 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1564 /* Create application object if socket is CMD. */
1565 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1567 if (!wait_node
->app
) {
1568 ret
= close(ust_cmd
->sock
);
1570 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1572 lttng_fd_put(1, LTTNG_FD_APPS
);
1578 * Add application to the wait queue so we can set the notify
1579 * socket before putting this object in the global ht.
1581 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1586 * We have to continue here since we don't have the notify
1587 * socket and the application MUST be added to the hash table
1588 * only at that moment.
1593 * Look for the application in the local wait queue and set the
1594 * notify socket if found.
1596 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1597 &wait_queue
.head
, head
) {
1598 health_code_update();
1599 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1600 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1601 cds_list_del(&wait_node
->head
);
1603 app
= wait_node
->app
;
1605 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1611 * With no application at this stage the received socket is
1612 * basically useless so close it before we free the cmd data
1613 * structure for good.
1616 ret
= close(ust_cmd
->sock
);
1618 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1620 lttng_fd_put(1, LTTNG_FD_APPS
);
1627 * @session_lock_list
1629 * Lock the global session list so from the register up to the
1630 * registration done message, no thread can see the application
1631 * and change its state.
1633 session_lock_list();
1637 * Add application to the global hash table. This needs to be
1638 * done before the update to the UST registry can locate the
1643 /* Set app version. This call will print an error if needed. */
1644 (void) ust_app_version(app
);
1646 /* Send notify socket through the notify pipe. */
1647 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1651 session_unlock_list();
1652 /* No notify thread, stop the UST tracing. */
1657 * Update newly registered application with the tracing
1658 * registry info already enabled information.
1660 update_ust_app(app
->sock
);
1663 * Don't care about return value. Let the manage apps threads
1664 * handle app unregistration upon socket close.
1666 (void) ust_app_register_done(app
->sock
);
1669 * Even if the application socket has been closed, send the app
1670 * to the thread and unregistration will take place at that
1673 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1676 session_unlock_list();
1677 /* No apps. thread, stop the UST tracing. */
1682 session_unlock_list();
1684 } while (node
!= NULL
);
1686 health_poll_entry();
1687 /* Futex wait on queue. Blocking call on futex() */
1688 futex_nto1_wait(&ust_cmd_queue
.futex
);
1691 /* Normal exit, no error */
1695 /* Clean up wait queue. */
1696 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1697 &wait_queue
.head
, head
) {
1698 cds_list_del(&wait_node
->head
);
1703 DBG("Dispatch thread dying");
1706 ERR("Health error occurred in %s", __func__
);
1708 health_unregister();
1713 * This thread manage application registration.
1715 static void *thread_registration_apps(void *data
)
1717 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1718 uint32_t revents
, nb_fd
;
1719 struct lttng_poll_event events
;
1721 * Get allocated in this thread, enqueued to a global queue, dequeued and
1722 * freed in the manage apps thread.
1724 struct ust_command
*ust_cmd
= NULL
;
1726 DBG("[thread] Manage application registration started");
1728 health_register(HEALTH_TYPE_APP_REG
);
1730 if (testpoint(thread_registration_apps
)) {
1731 goto error_testpoint
;
1734 ret
= lttcomm_listen_unix_sock(apps_sock
);
1740 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1741 * more will be added to this poll set.
1743 ret
= sessiond_set_thread_pollset(&events
, 2);
1745 goto error_create_poll
;
1748 /* Add the application registration socket */
1749 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1751 goto error_poll_add
;
1754 /* Notify all applications to register */
1755 ret
= notify_ust_apps(1);
1757 ERR("Failed to notify applications or create the wait shared memory.\n"
1758 "Execution continues but there might be problem for already\n"
1759 "running applications that wishes to register.");
1763 DBG("Accepting application registration");
1765 /* Inifinite blocking call, waiting for transmission */
1767 health_poll_entry();
1768 ret
= lttng_poll_wait(&events
, -1);
1772 * Restart interrupted system call.
1774 if (errno
== EINTR
) {
1782 for (i
= 0; i
< nb_fd
; i
++) {
1783 health_code_update();
1785 /* Fetch once the poll data */
1786 revents
= LTTNG_POLL_GETEV(&events
, i
);
1787 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1789 /* Thread quit pipe has been closed. Killing thread. */
1790 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1796 /* Event on the registration socket */
1797 if (pollfd
== apps_sock
) {
1798 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1799 ERR("Register apps socket poll error");
1801 } else if (revents
& LPOLLIN
) {
1802 sock
= lttcomm_accept_unix_sock(apps_sock
);
1808 * Set the CLOEXEC flag. Return code is useless because
1809 * either way, the show must go on.
1811 (void) utils_set_fd_cloexec(sock
);
1813 /* Create UST registration command for enqueuing */
1814 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1815 if (ust_cmd
== NULL
) {
1816 PERROR("ust command zmalloc");
1821 * Using message-based transmissions to ensure we don't
1822 * have to deal with partially received messages.
1824 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1826 ERR("Exhausted file descriptors allowed for applications.");
1836 health_code_update();
1837 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1840 /* Close socket of the application. */
1845 lttng_fd_put(LTTNG_FD_APPS
, 1);
1849 health_code_update();
1851 ust_cmd
->sock
= sock
;
1854 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1855 " gid:%d sock:%d name:%s (version %d.%d)",
1856 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1857 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1858 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1859 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1862 * Lock free enqueue the registration request. The red pill
1863 * has been taken! This apps will be part of the *system*.
1865 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1868 * Wake the registration queue futex. Implicit memory
1869 * barrier with the exchange in cds_wfq_enqueue.
1871 futex_nto1_wake(&ust_cmd_queue
.futex
);
1881 ERR("Health error occurred in %s", __func__
);
1884 /* Notify that the registration thread is gone */
1887 if (apps_sock
>= 0) {
1888 ret
= close(apps_sock
);
1898 lttng_fd_put(LTTNG_FD_APPS
, 1);
1900 unlink(apps_unix_sock_path
);
1903 lttng_poll_clean(&events
);
1907 DBG("UST Registration thread cleanup complete");
1908 health_unregister();
1914 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1915 * exec or it will fails.
1917 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1920 struct timespec timeout
;
1922 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1923 consumer_data
->consumer_thread_is_ready
= 0;
1925 /* Setup pthread condition */
1926 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1929 PERROR("pthread_condattr_init consumer data");
1934 * Set the monotonic clock in order to make sure we DO NOT jump in time
1935 * between the clock_gettime() call and the timedwait call. See bug #324
1936 * for a more details and how we noticed it.
1938 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1941 PERROR("pthread_condattr_setclock consumer data");
1945 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1948 PERROR("pthread_cond_init consumer data");
1952 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1955 PERROR("pthread_create consumer");
1960 /* We are about to wait on a pthread condition */
1961 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1963 /* Get time for sem_timedwait absolute timeout */
1964 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1966 * Set the timeout for the condition timed wait even if the clock gettime
1967 * call fails since we might loop on that call and we want to avoid to
1968 * increment the timeout too many times.
1970 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1973 * The following loop COULD be skipped in some conditions so this is why we
1974 * set ret to 0 in order to make sure at least one round of the loop is
1980 * Loop until the condition is reached or when a timeout is reached. Note
1981 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1982 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1983 * possible. This loop does not take any chances and works with both of
1986 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1987 if (clock_ret
< 0) {
1988 PERROR("clock_gettime spawn consumer");
1989 /* Infinite wait for the consumerd thread to be ready */
1990 ret
= pthread_cond_wait(&consumer_data
->cond
,
1991 &consumer_data
->cond_mutex
);
1993 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1994 &consumer_data
->cond_mutex
, &timeout
);
1998 /* Release the pthread condition */
1999 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2003 if (ret
== ETIMEDOUT
) {
2005 * Call has timed out so we kill the kconsumerd_thread and return
2008 ERR("Condition timed out. The consumer thread was never ready."
2010 ret
= pthread_cancel(consumer_data
->thread
);
2012 PERROR("pthread_cancel consumer thread");
2015 PERROR("pthread_cond_wait failed consumer thread");
2020 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2021 if (consumer_data
->pid
== 0) {
2022 ERR("Consumerd did not start");
2023 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2026 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2035 * Join consumer thread
2037 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2041 /* Consumer pid must be a real one. */
2042 if (consumer_data
->pid
> 0) {
2044 ret
= kill(consumer_data
->pid
, SIGTERM
);
2046 ERR("Error killing consumer daemon");
2049 return pthread_join(consumer_data
->thread
, &status
);
2056 * Fork and exec a consumer daemon (consumerd).
2058 * Return pid if successful else -1.
2060 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2064 const char *consumer_to_use
;
2065 const char *verbosity
;
2068 DBG("Spawning consumerd");
2075 if (opt_verbose_consumer
) {
2076 verbosity
= "--verbose";
2078 verbosity
= "--quiet";
2080 switch (consumer_data
->type
) {
2081 case LTTNG_CONSUMER_KERNEL
:
2083 * Find out which consumerd to execute. We will first try the
2084 * 64-bit path, then the sessiond's installation directory, and
2085 * fallback on the 32-bit one,
2087 DBG3("Looking for a kernel consumer at these locations:");
2088 DBG3(" 1) %s", consumerd64_bin
);
2089 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2090 DBG3(" 3) %s", consumerd32_bin
);
2091 if (stat(consumerd64_bin
, &st
) == 0) {
2092 DBG3("Found location #1");
2093 consumer_to_use
= consumerd64_bin
;
2094 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2095 DBG3("Found location #2");
2096 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2097 } else if (stat(consumerd32_bin
, &st
) == 0) {
2098 DBG3("Found location #3");
2099 consumer_to_use
= consumerd32_bin
;
2101 DBG("Could not find any valid consumerd executable");
2104 DBG("Using kernel consumer at: %s", consumer_to_use
);
2105 execl(consumer_to_use
,
2106 "lttng-consumerd", verbosity
, "-k",
2107 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2108 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2111 case LTTNG_CONSUMER64_UST
:
2113 char *tmpnew
= NULL
;
2115 if (consumerd64_libdir
[0] != '\0') {
2119 tmp
= getenv("LD_LIBRARY_PATH");
2123 tmplen
= strlen("LD_LIBRARY_PATH=")
2124 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2125 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2130 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2131 strcat(tmpnew
, consumerd64_libdir
);
2132 if (tmp
[0] != '\0') {
2133 strcat(tmpnew
, ":");
2134 strcat(tmpnew
, tmp
);
2136 ret
= putenv(tmpnew
);
2143 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2144 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2145 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2146 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2148 if (consumerd64_libdir
[0] != '\0') {
2156 case LTTNG_CONSUMER32_UST
:
2158 char *tmpnew
= NULL
;
2160 if (consumerd32_libdir
[0] != '\0') {
2164 tmp
= getenv("LD_LIBRARY_PATH");
2168 tmplen
= strlen("LD_LIBRARY_PATH=")
2169 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2170 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2175 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2176 strcat(tmpnew
, consumerd32_libdir
);
2177 if (tmp
[0] != '\0') {
2178 strcat(tmpnew
, ":");
2179 strcat(tmpnew
, tmp
);
2181 ret
= putenv(tmpnew
);
2188 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2189 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2190 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2191 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2193 if (consumerd32_libdir
[0] != '\0') {
2202 PERROR("unknown consumer type");
2206 PERROR("kernel start consumer exec");
2209 } else if (pid
> 0) {
2212 PERROR("start consumer fork");
2220 * Spawn the consumerd daemon and session daemon thread.
2222 static int start_consumerd(struct consumer_data
*consumer_data
)
2227 * Set the listen() state on the socket since there is a possible race
2228 * between the exec() of the consumer daemon and this call if place in the
2229 * consumer thread. See bug #366 for more details.
2231 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2236 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2237 if (consumer_data
->pid
!= 0) {
2238 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2242 ret
= spawn_consumerd(consumer_data
);
2244 ERR("Spawning consumerd failed");
2245 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2249 /* Setting up the consumer_data pid */
2250 consumer_data
->pid
= ret
;
2251 DBG2("Consumer pid %d", consumer_data
->pid
);
2252 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2254 DBG2("Spawning consumer control thread");
2255 ret
= spawn_consumer_thread(consumer_data
);
2257 ERR("Fatal error spawning consumer control thread");
2265 /* Cleanup already created sockets on error. */
2266 if (consumer_data
->err_sock
>= 0) {
2269 err
= close(consumer_data
->err_sock
);
2271 PERROR("close consumer data error socket");
2278 * Compute health status of each consumer. If one of them is zero (bad
2279 * state), we return 0.
2281 static int check_consumer_health(void)
2285 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2287 DBG3("Health consumer check %d", ret
);
2293 * Setup necessary data for kernel tracer action.
2295 static int init_kernel_tracer(void)
2299 /* Modprobe lttng kernel modules */
2300 ret
= modprobe_lttng_control();
2305 /* Open debugfs lttng */
2306 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2307 if (kernel_tracer_fd
< 0) {
2308 DBG("Failed to open %s", module_proc_lttng
);
2313 /* Validate kernel version */
2314 ret
= kernel_validate_version(kernel_tracer_fd
);
2319 ret
= modprobe_lttng_data();
2324 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2328 modprobe_remove_lttng_control();
2329 ret
= close(kernel_tracer_fd
);
2333 kernel_tracer_fd
= -1;
2334 return LTTNG_ERR_KERN_VERSION
;
2337 ret
= close(kernel_tracer_fd
);
2343 modprobe_remove_lttng_control();
2346 WARN("No kernel tracer available");
2347 kernel_tracer_fd
= -1;
2349 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2351 return LTTNG_ERR_KERN_NA
;
2357 * Copy consumer output from the tracing session to the domain session. The
2358 * function also applies the right modification on a per domain basis for the
2359 * trace files destination directory.
2361 * Should *NOT* be called with RCU read-side lock held.
2363 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2366 const char *dir_name
;
2367 struct consumer_output
*consumer
;
2370 assert(session
->consumer
);
2373 case LTTNG_DOMAIN_KERNEL
:
2374 DBG3("Copying tracing session consumer output in kernel session");
2376 * XXX: We should audit the session creation and what this function
2377 * does "extra" in order to avoid a destroy since this function is used
2378 * in the domain session creation (kernel and ust) only. Same for UST
2381 if (session
->kernel_session
->consumer
) {
2382 consumer_destroy_output(session
->kernel_session
->consumer
);
2384 session
->kernel_session
->consumer
=
2385 consumer_copy_output(session
->consumer
);
2386 /* Ease our life a bit for the next part */
2387 consumer
= session
->kernel_session
->consumer
;
2388 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2390 case LTTNG_DOMAIN_UST
:
2391 DBG3("Copying tracing session consumer output in UST session");
2392 if (session
->ust_session
->consumer
) {
2393 consumer_destroy_output(session
->ust_session
->consumer
);
2395 session
->ust_session
->consumer
=
2396 consumer_copy_output(session
->consumer
);
2397 /* Ease our life a bit for the next part */
2398 consumer
= session
->ust_session
->consumer
;
2399 dir_name
= DEFAULT_UST_TRACE_DIR
;
2402 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2406 /* Append correct directory to subdir */
2407 strncat(consumer
->subdir
, dir_name
,
2408 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2409 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2418 * Create an UST session and add it to the session ust list.
2420 * Should *NOT* be called with RCU read-side lock held.
2422 static int create_ust_session(struct ltt_session
*session
,
2423 struct lttng_domain
*domain
)
2426 struct ltt_ust_session
*lus
= NULL
;
2430 assert(session
->consumer
);
2432 switch (domain
->type
) {
2433 case LTTNG_DOMAIN_UST
:
2436 ERR("Unknown UST domain on create session %d", domain
->type
);
2437 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2441 DBG("Creating UST session");
2443 lus
= trace_ust_create_session(session
->id
);
2445 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2449 lus
->uid
= session
->uid
;
2450 lus
->gid
= session
->gid
;
2451 lus
->output_traces
= session
->output_traces
;
2452 lus
->snapshot_mode
= session
->snapshot_mode
;
2453 session
->ust_session
= lus
;
2455 /* Copy session output to the newly created UST session */
2456 ret
= copy_session_consumer(domain
->type
, session
);
2457 if (ret
!= LTTNG_OK
) {
2465 session
->ust_session
= NULL
;
2470 * Create a kernel tracer session then create the default channel.
2472 static int create_kernel_session(struct ltt_session
*session
)
2476 DBG("Creating kernel session");
2478 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2480 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2484 /* Code flow safety */
2485 assert(session
->kernel_session
);
2487 /* Copy session output to the newly created Kernel session */
2488 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2489 if (ret
!= LTTNG_OK
) {
2493 /* Create directory(ies) on local filesystem. */
2494 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2495 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2496 ret
= run_as_mkdir_recursive(
2497 session
->kernel_session
->consumer
->dst
.trace_path
,
2498 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2500 if (ret
!= -EEXIST
) {
2501 ERR("Trace directory creation error");
2507 session
->kernel_session
->uid
= session
->uid
;
2508 session
->kernel_session
->gid
= session
->gid
;
2509 session
->kernel_session
->output_traces
= session
->output_traces
;
2510 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2515 trace_kernel_destroy_session(session
->kernel_session
);
2516 session
->kernel_session
= NULL
;
2521 * Count number of session permitted by uid/gid.
2523 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2526 struct ltt_session
*session
;
2528 DBG("Counting number of available session for UID %d GID %d",
2530 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2532 * Only list the sessions the user can control.
2534 if (!session_access_ok(session
, uid
, gid
)) {
2543 * Process the command requested by the lttng client within the command
2544 * context structure. This function make sure that the return structure (llm)
2545 * is set and ready for transmission before returning.
2547 * Return any error encountered or 0 for success.
2549 * "sock" is only used for special-case var. len data.
2551 * Should *NOT* be called with RCU read-side lock held.
2553 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2557 int need_tracing_session
= 1;
2560 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2564 switch (cmd_ctx
->lsm
->cmd_type
) {
2565 case LTTNG_CREATE_SESSION
:
2566 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2567 case LTTNG_DESTROY_SESSION
:
2568 case LTTNG_LIST_SESSIONS
:
2569 case LTTNG_LIST_DOMAINS
:
2570 case LTTNG_START_TRACE
:
2571 case LTTNG_STOP_TRACE
:
2572 case LTTNG_DATA_PENDING
:
2573 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2574 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2575 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2576 case LTTNG_SNAPSHOT_RECORD
:
2583 if (opt_no_kernel
&& need_domain
2584 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2586 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2588 ret
= LTTNG_ERR_KERN_NA
;
2593 /* Deny register consumer if we already have a spawned consumer. */
2594 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2595 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2596 if (kconsumer_data
.pid
> 0) {
2597 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2598 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2601 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2605 * Check for command that don't needs to allocate a returned payload. We do
2606 * this here so we don't have to make the call for no payload at each
2609 switch(cmd_ctx
->lsm
->cmd_type
) {
2610 case LTTNG_LIST_SESSIONS
:
2611 case LTTNG_LIST_TRACEPOINTS
:
2612 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2613 case LTTNG_LIST_DOMAINS
:
2614 case LTTNG_LIST_CHANNELS
:
2615 case LTTNG_LIST_EVENTS
:
2618 /* Setup lttng message with no payload */
2619 ret
= setup_lttng_msg(cmd_ctx
, 0);
2621 /* This label does not try to unlock the session */
2622 goto init_setup_error
;
2626 /* Commands that DO NOT need a session. */
2627 switch (cmd_ctx
->lsm
->cmd_type
) {
2628 case LTTNG_CREATE_SESSION
:
2629 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2630 case LTTNG_CALIBRATE
:
2631 case LTTNG_LIST_SESSIONS
:
2632 case LTTNG_LIST_TRACEPOINTS
:
2633 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2634 need_tracing_session
= 0;
2637 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2639 * We keep the session list lock across _all_ commands
2640 * for now, because the per-session lock does not
2641 * handle teardown properly.
2643 session_lock_list();
2644 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2645 if (cmd_ctx
->session
== NULL
) {
2646 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2649 /* Acquire lock for the session */
2650 session_lock(cmd_ctx
->session
);
2660 * Check domain type for specific "pre-action".
2662 switch (cmd_ctx
->lsm
->domain
.type
) {
2663 case LTTNG_DOMAIN_KERNEL
:
2665 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2669 /* Kernel tracer check */
2670 if (kernel_tracer_fd
== -1) {
2671 /* Basically, load kernel tracer modules */
2672 ret
= init_kernel_tracer();
2678 /* Consumer is in an ERROR state. Report back to client */
2679 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2680 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2684 /* Need a session for kernel command */
2685 if (need_tracing_session
) {
2686 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2687 ret
= create_kernel_session(cmd_ctx
->session
);
2689 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2694 /* Start the kernel consumer daemon */
2695 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2696 if (kconsumer_data
.pid
== 0 &&
2697 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2698 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2699 ret
= start_consumerd(&kconsumer_data
);
2701 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2704 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2706 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2710 * The consumer was just spawned so we need to add the socket to
2711 * the consumer output of the session if exist.
2713 ret
= consumer_create_socket(&kconsumer_data
,
2714 cmd_ctx
->session
->kernel_session
->consumer
);
2721 case LTTNG_DOMAIN_UST
:
2723 if (!ust_app_supported()) {
2724 ret
= LTTNG_ERR_NO_UST
;
2727 /* Consumer is in an ERROR state. Report back to client */
2728 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2729 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2733 if (need_tracing_session
) {
2734 /* Create UST session if none exist. */
2735 if (cmd_ctx
->session
->ust_session
== NULL
) {
2736 ret
= create_ust_session(cmd_ctx
->session
,
2737 &cmd_ctx
->lsm
->domain
);
2738 if (ret
!= LTTNG_OK
) {
2743 /* Start the UST consumer daemons */
2745 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2746 if (consumerd64_bin
[0] != '\0' &&
2747 ustconsumer64_data
.pid
== 0 &&
2748 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2749 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2750 ret
= start_consumerd(&ustconsumer64_data
);
2752 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2753 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2757 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2758 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2760 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2764 * Setup socket for consumer 64 bit. No need for atomic access
2765 * since it was set above and can ONLY be set in this thread.
2767 ret
= consumer_create_socket(&ustconsumer64_data
,
2768 cmd_ctx
->session
->ust_session
->consumer
);
2774 if (consumerd32_bin
[0] != '\0' &&
2775 ustconsumer32_data
.pid
== 0 &&
2776 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2777 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2778 ret
= start_consumerd(&ustconsumer32_data
);
2780 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2781 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2785 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2786 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2788 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2792 * Setup socket for consumer 64 bit. No need for atomic access
2793 * since it was set above and can ONLY be set in this thread.
2795 ret
= consumer_create_socket(&ustconsumer32_data
,
2796 cmd_ctx
->session
->ust_session
->consumer
);
2808 /* Validate consumer daemon state when start/stop trace command */
2809 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2810 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2811 switch (cmd_ctx
->lsm
->domain
.type
) {
2812 case LTTNG_DOMAIN_UST
:
2813 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2814 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2818 case LTTNG_DOMAIN_KERNEL
:
2819 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2820 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2828 * Check that the UID or GID match that of the tracing session.
2829 * The root user can interact with all sessions.
2831 if (need_tracing_session
) {
2832 if (!session_access_ok(cmd_ctx
->session
,
2833 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2834 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2835 ret
= LTTNG_ERR_EPERM
;
2841 * Send relayd information to consumer as soon as we have a domain and a
2844 if (cmd_ctx
->session
&& need_domain
) {
2846 * Setup relayd if not done yet. If the relayd information was already
2847 * sent to the consumer, this call will gracefully return.
2849 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2850 if (ret
!= LTTNG_OK
) {
2855 /* Process by command type */
2856 switch (cmd_ctx
->lsm
->cmd_type
) {
2857 case LTTNG_ADD_CONTEXT
:
2859 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2860 cmd_ctx
->lsm
->u
.context
.channel_name
,
2861 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2864 case LTTNG_DISABLE_CHANNEL
:
2866 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2867 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2870 case LTTNG_DISABLE_EVENT
:
2872 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2873 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2874 cmd_ctx
->lsm
->u
.disable
.name
);
2877 case LTTNG_DISABLE_ALL_EVENT
:
2879 DBG("Disabling all events");
2881 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2882 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2885 case LTTNG_ENABLE_CHANNEL
:
2887 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2888 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2891 case LTTNG_ENABLE_EVENT
:
2893 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2894 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2895 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2898 case LTTNG_ENABLE_ALL_EVENT
:
2900 DBG("Enabling all events");
2902 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2903 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2904 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2907 case LTTNG_LIST_TRACEPOINTS
:
2909 struct lttng_event
*events
;
2912 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2913 if (nb_events
< 0) {
2914 /* Return value is a negative lttng_error_code. */
2920 * Setup lttng message with payload size set to the event list size in
2921 * bytes and then copy list into the llm payload.
2923 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2929 /* Copy event list into message payload */
2930 memcpy(cmd_ctx
->llm
->payload
, events
,
2931 sizeof(struct lttng_event
) * nb_events
);
2938 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2940 struct lttng_event_field
*fields
;
2943 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2945 if (nb_fields
< 0) {
2946 /* Return value is a negative lttng_error_code. */
2952 * Setup lttng message with payload size set to the event list size in
2953 * bytes and then copy list into the llm payload.
2955 ret
= setup_lttng_msg(cmd_ctx
,
2956 sizeof(struct lttng_event_field
) * nb_fields
);
2962 /* Copy event list into message payload */
2963 memcpy(cmd_ctx
->llm
->payload
, fields
,
2964 sizeof(struct lttng_event_field
) * nb_fields
);
2971 case LTTNG_SET_CONSUMER_URI
:
2974 struct lttng_uri
*uris
;
2976 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2977 len
= nb_uri
* sizeof(struct lttng_uri
);
2980 ret
= LTTNG_ERR_INVALID
;
2984 uris
= zmalloc(len
);
2986 ret
= LTTNG_ERR_FATAL
;
2990 /* Receive variable len data */
2991 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2992 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2994 DBG("No URIs received from client... continuing");
2996 ret
= LTTNG_ERR_SESSION_FAIL
;
3001 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3003 if (ret
!= LTTNG_OK
) {
3009 * XXX: 0 means that this URI should be applied on the session. Should
3010 * be a DOMAIN enuam.
3012 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3013 /* Add the URI for the UST session if a consumer is present. */
3014 if (cmd_ctx
->session
->ust_session
&&
3015 cmd_ctx
->session
->ust_session
->consumer
) {
3016 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3018 } else if (cmd_ctx
->session
->kernel_session
&&
3019 cmd_ctx
->session
->kernel_session
->consumer
) {
3020 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3021 cmd_ctx
->session
, nb_uri
, uris
);
3029 case LTTNG_START_TRACE
:
3031 ret
= cmd_start_trace(cmd_ctx
->session
);
3034 case LTTNG_STOP_TRACE
:
3036 ret
= cmd_stop_trace(cmd_ctx
->session
);
3039 case LTTNG_CREATE_SESSION
:
3042 struct lttng_uri
*uris
= NULL
;
3044 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3045 len
= nb_uri
* sizeof(struct lttng_uri
);
3048 uris
= zmalloc(len
);
3050 ret
= LTTNG_ERR_FATAL
;
3054 /* Receive variable len data */
3055 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3056 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3058 DBG("No URIs received from client... continuing");
3060 ret
= LTTNG_ERR_SESSION_FAIL
;
3065 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3066 DBG("Creating session with ONE network URI is a bad call");
3067 ret
= LTTNG_ERR_SESSION_FAIL
;
3073 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3080 case LTTNG_DESTROY_SESSION
:
3082 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3084 /* Set session to NULL so we do not unlock it after free. */
3085 cmd_ctx
->session
= NULL
;
3088 case LTTNG_LIST_DOMAINS
:
3091 struct lttng_domain
*domains
;
3093 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3095 /* Return value is a negative lttng_error_code. */
3100 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3106 /* Copy event list into message payload */
3107 memcpy(cmd_ctx
->llm
->payload
, domains
,
3108 nb_dom
* sizeof(struct lttng_domain
));
3115 case LTTNG_LIST_CHANNELS
:
3118 struct lttng_channel
*channels
;
3120 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3121 cmd_ctx
->session
, &channels
);
3123 /* Return value is a negative lttng_error_code. */
3128 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3134 /* Copy event list into message payload */
3135 memcpy(cmd_ctx
->llm
->payload
, channels
,
3136 nb_chan
* sizeof(struct lttng_channel
));
3143 case LTTNG_LIST_EVENTS
:
3146 struct lttng_event
*events
= NULL
;
3148 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3149 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3151 /* Return value is a negative lttng_error_code. */
3156 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3162 /* Copy event list into message payload */
3163 memcpy(cmd_ctx
->llm
->payload
, events
,
3164 nb_event
* sizeof(struct lttng_event
));
3171 case LTTNG_LIST_SESSIONS
:
3173 unsigned int nr_sessions
;
3175 session_lock_list();
3176 nr_sessions
= lttng_sessions_count(
3177 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3178 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3180 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3182 session_unlock_list();
3186 /* Filled the session array */
3187 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3188 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3189 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3191 session_unlock_list();
3196 case LTTNG_CALIBRATE
:
3198 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3199 &cmd_ctx
->lsm
->u
.calibrate
);
3202 case LTTNG_REGISTER_CONSUMER
:
3204 struct consumer_data
*cdata
;
3206 switch (cmd_ctx
->lsm
->domain
.type
) {
3207 case LTTNG_DOMAIN_KERNEL
:
3208 cdata
= &kconsumer_data
;
3211 ret
= LTTNG_ERR_UND
;
3215 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3216 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3219 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3221 struct lttng_filter_bytecode
*bytecode
;
3223 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3224 ret
= LTTNG_ERR_FILTER_INVAL
;
3227 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3228 ret
= LTTNG_ERR_FILTER_INVAL
;
3231 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3233 ret
= LTTNG_ERR_FILTER_NOMEM
;
3236 /* Receive var. len. data */
3237 DBG("Receiving var len data from client ...");
3238 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3239 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3241 DBG("Nothing recv() from client var len data... continuing");
3243 ret
= LTTNG_ERR_FILTER_INVAL
;
3247 if (bytecode
->len
+ sizeof(*bytecode
)
3248 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3250 ret
= LTTNG_ERR_FILTER_INVAL
;
3254 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3255 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3256 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3259 case LTTNG_DATA_PENDING
:
3261 ret
= cmd_data_pending(cmd_ctx
->session
);
3264 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3266 struct lttcomm_lttng_output_id reply
;
3268 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3269 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3270 if (ret
!= LTTNG_OK
) {
3274 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3279 /* Copy output list into message payload */
3280 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3284 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3286 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3287 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3290 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3293 struct lttng_snapshot_output
*outputs
= NULL
;
3295 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3296 if (nb_output
< 0) {
3301 ret
= setup_lttng_msg(cmd_ctx
,
3302 nb_output
* sizeof(struct lttng_snapshot_output
));
3309 /* Copy output list into message payload */
3310 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3311 nb_output
* sizeof(struct lttng_snapshot_output
));
3318 case LTTNG_SNAPSHOT_RECORD
:
3320 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3321 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3322 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3325 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3328 struct lttng_uri
*uris
= NULL
;
3330 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3331 len
= nb_uri
* sizeof(struct lttng_uri
);
3334 uris
= zmalloc(len
);
3336 ret
= LTTNG_ERR_FATAL
;
3340 /* Receive variable len data */
3341 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3342 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3344 DBG("No URIs received from client... continuing");
3346 ret
= LTTNG_ERR_SESSION_FAIL
;
3351 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3352 DBG("Creating session with ONE network URI is a bad call");
3353 ret
= LTTNG_ERR_SESSION_FAIL
;
3359 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3360 nb_uri
, &cmd_ctx
->creds
);
3365 ret
= LTTNG_ERR_UND
;
3370 if (cmd_ctx
->llm
== NULL
) {
3371 DBG("Missing llm structure. Allocating one.");
3372 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3376 /* Set return code */
3377 cmd_ctx
->llm
->ret_code
= ret
;
3379 if (cmd_ctx
->session
) {
3380 session_unlock(cmd_ctx
->session
);
3382 if (need_tracing_session
) {
3383 session_unlock_list();
3390 * Thread managing health check socket.
3392 static void *thread_manage_health(void *data
)
3394 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3395 uint32_t revents
, nb_fd
;
3396 struct lttng_poll_event events
;
3397 struct lttcomm_health_msg msg
;
3398 struct lttcomm_health_data reply
;
3400 DBG("[thread] Manage health check started");
3402 rcu_register_thread();
3404 /* We might hit an error path before this is created. */
3405 lttng_poll_init(&events
);
3407 /* Create unix socket */
3408 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3410 ERR("Unable to create health check Unix socket");
3416 * Set the CLOEXEC flag. Return code is useless because either way, the
3419 (void) utils_set_fd_cloexec(sock
);
3421 ret
= lttcomm_listen_unix_sock(sock
);
3427 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3428 * more will be added to this poll set.
3430 ret
= sessiond_set_thread_pollset(&events
, 2);
3435 /* Add the application registration socket */
3436 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3442 DBG("Health check ready");
3444 /* Inifinite blocking call, waiting for transmission */
3446 ret
= lttng_poll_wait(&events
, -1);
3449 * Restart interrupted system call.
3451 if (errno
== EINTR
) {
3459 for (i
= 0; i
< nb_fd
; i
++) {
3460 /* Fetch once the poll data */
3461 revents
= LTTNG_POLL_GETEV(&events
, i
);
3462 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3464 /* Thread quit pipe has been closed. Killing thread. */
3465 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3471 /* Event on the registration socket */
3472 if (pollfd
== sock
) {
3473 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3474 ERR("Health socket poll error");
3480 new_sock
= lttcomm_accept_unix_sock(sock
);
3486 * Set the CLOEXEC flag. Return code is useless because either way, the
3489 (void) utils_set_fd_cloexec(new_sock
);
3491 DBG("Receiving data from client for health...");
3492 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3494 DBG("Nothing recv() from client... continuing");
3495 ret
= close(new_sock
);
3503 rcu_thread_online();
3505 switch (msg
.component
) {
3506 case LTTNG_HEALTH_CMD
:
3507 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3509 case LTTNG_HEALTH_APP_MANAGE
:
3510 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3512 case LTTNG_HEALTH_APP_REG
:
3513 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3515 case LTTNG_HEALTH_KERNEL
:
3516 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3518 case LTTNG_HEALTH_CONSUMER
:
3519 reply
.ret_code
= check_consumer_health();
3521 case LTTNG_HEALTH_HT_CLEANUP
:
3522 reply
.ret_code
= health_check_state(HEALTH_TYPE_HT_CLEANUP
);
3524 case LTTNG_HEALTH_APP_MANAGE_NOTIFY
:
3525 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3527 case LTTNG_HEALTH_APP_REG_DISPATCH
:
3528 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3530 case LTTNG_HEALTH_ALL
:
3532 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3533 health_check_state(HEALTH_TYPE_APP_REG
) &&
3534 health_check_state(HEALTH_TYPE_CMD
) &&
3535 health_check_state(HEALTH_TYPE_KERNEL
) &&
3536 check_consumer_health() &&
3537 health_check_state(HEALTH_TYPE_HT_CLEANUP
) &&
3538 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
) &&
3539 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3542 reply
.ret_code
= LTTNG_ERR_UND
;
3547 * Flip ret value since 0 is a success and 1 indicates a bad health for
3548 * the client where in the sessiond it is the opposite. Again, this is
3549 * just to make things easier for us poor developer which enjoy a lot
3552 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3553 reply
.ret_code
= !reply
.ret_code
;
3556 DBG2("Health check return value %d", reply
.ret_code
);
3558 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3560 ERR("Failed to send health data back to client");
3563 /* End of transmission */
3564 ret
= close(new_sock
);
3574 ERR("Health error occurred in %s", __func__
);
3576 DBG("Health check thread dying");
3577 unlink(health_unix_sock_path
);
3585 lttng_poll_clean(&events
);
3587 rcu_unregister_thread();
3592 * This thread manage all clients request using the unix client socket for
3595 static void *thread_manage_clients(void *data
)
3597 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3599 uint32_t revents
, nb_fd
;
3600 struct command_ctx
*cmd_ctx
= NULL
;
3601 struct lttng_poll_event events
;
3603 DBG("[thread] Manage client started");
3605 rcu_register_thread();
3607 health_register(HEALTH_TYPE_CMD
);
3609 if (testpoint(thread_manage_clients
)) {
3610 goto error_testpoint
;
3613 health_code_update();
3615 ret
= lttcomm_listen_unix_sock(client_sock
);
3621 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3622 * more will be added to this poll set.
3624 ret
= sessiond_set_thread_pollset(&events
, 2);
3626 goto error_create_poll
;
3629 /* Add the application registration socket */
3630 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3636 * Notify parent pid that we are ready to accept command for client side.
3638 if (opt_sig_parent
) {
3639 kill(ppid
, SIGUSR1
);
3642 if (testpoint(thread_manage_clients_before_loop
)) {
3646 health_code_update();
3649 DBG("Accepting client command ...");
3651 /* Inifinite blocking call, waiting for transmission */
3653 health_poll_entry();
3654 ret
= lttng_poll_wait(&events
, -1);
3658 * Restart interrupted system call.
3660 if (errno
== EINTR
) {
3668 for (i
= 0; i
< nb_fd
; i
++) {
3669 /* Fetch once the poll data */
3670 revents
= LTTNG_POLL_GETEV(&events
, i
);
3671 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3673 health_code_update();
3675 /* Thread quit pipe has been closed. Killing thread. */
3676 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3682 /* Event on the registration socket */
3683 if (pollfd
== client_sock
) {
3684 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3685 ERR("Client socket poll error");
3691 DBG("Wait for client response");
3693 health_code_update();
3695 sock
= lttcomm_accept_unix_sock(client_sock
);
3701 * Set the CLOEXEC flag. Return code is useless because either way, the
3704 (void) utils_set_fd_cloexec(sock
);
3706 /* Set socket option for credentials retrieval */
3707 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3712 /* Allocate context command to process the client request */
3713 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3714 if (cmd_ctx
== NULL
) {
3715 PERROR("zmalloc cmd_ctx");
3719 /* Allocate data buffer for reception */
3720 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3721 if (cmd_ctx
->lsm
== NULL
) {
3722 PERROR("zmalloc cmd_ctx->lsm");
3726 cmd_ctx
->llm
= NULL
;
3727 cmd_ctx
->session
= NULL
;
3729 health_code_update();
3732 * Data is received from the lttng client. The struct
3733 * lttcomm_session_msg (lsm) contains the command and data request of
3736 DBG("Receiving data from client ...");
3737 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3738 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3740 DBG("Nothing recv() from client... continuing");
3746 clean_command_ctx(&cmd_ctx
);
3750 health_code_update();
3752 // TODO: Validate cmd_ctx including sanity check for
3753 // security purpose.
3755 rcu_thread_online();
3757 * This function dispatch the work to the kernel or userspace tracer
3758 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3759 * informations for the client. The command context struct contains
3760 * everything this function may needs.
3762 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3763 rcu_thread_offline();
3771 * TODO: Inform client somehow of the fatal error. At
3772 * this point, ret < 0 means that a zmalloc failed
3773 * (ENOMEM). Error detected but still accept
3774 * command, unless a socket error has been
3777 clean_command_ctx(&cmd_ctx
);
3781 health_code_update();
3783 DBG("Sending response (size: %d, retcode: %s)",
3784 cmd_ctx
->lttng_msg_size
,
3785 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3786 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3788 ERR("Failed to send data back to client");
3791 /* End of transmission */
3798 clean_command_ctx(&cmd_ctx
);
3800 health_code_update();
3812 lttng_poll_clean(&events
);
3813 clean_command_ctx(&cmd_ctx
);
3818 unlink(client_unix_sock_path
);
3819 if (client_sock
>= 0) {
3820 ret
= close(client_sock
);
3828 ERR("Health error occurred in %s", __func__
);
3831 health_unregister();
3833 DBG("Client thread dying");
3835 rcu_unregister_thread();
3841 * usage function on stderr
3843 static void usage(void)
3845 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3846 fprintf(stderr
, " -h, --help Display this usage.\n");
3847 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3848 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3849 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3850 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3851 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3852 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3853 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3854 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3855 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3856 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3857 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3858 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3859 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3860 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3861 fprintf(stderr
, " -V, --version Show version number.\n");
3862 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3863 fprintf(stderr
, " -q, --quiet No output at all.\n");
3864 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3865 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3866 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3867 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3871 * daemon argument parsing
3873 static int parse_args(int argc
, char **argv
)
3877 static struct option long_options
[] = {
3878 { "client-sock", 1, 0, 'c' },
3879 { "apps-sock", 1, 0, 'a' },
3880 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3881 { "kconsumerd-err-sock", 1, 0, 'E' },
3882 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3883 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3884 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3885 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3886 { "consumerd32-path", 1, 0, 'u' },
3887 { "consumerd32-libdir", 1, 0, 'U' },
3888 { "consumerd64-path", 1, 0, 't' },
3889 { "consumerd64-libdir", 1, 0, 'T' },
3890 { "daemonize", 0, 0, 'd' },
3891 { "sig-parent", 0, 0, 'S' },
3892 { "help", 0, 0, 'h' },
3893 { "group", 1, 0, 'g' },
3894 { "version", 0, 0, 'V' },
3895 { "quiet", 0, 0, 'q' },
3896 { "verbose", 0, 0, 'v' },
3897 { "verbose-consumer", 0, 0, 'Z' },
3898 { "no-kernel", 0, 0, 'N' },
3899 { "pidfile", 1, 0, 'p' },
3904 int option_index
= 0;
3905 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3906 long_options
, &option_index
);
3913 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3915 fprintf(stderr
, " with arg %s\n", optarg
);
3919 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3922 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3928 opt_tracing_group
= optarg
;
3934 fprintf(stdout
, "%s\n", VERSION
);
3940 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3943 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3946 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3949 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3952 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3955 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3961 lttng_opt_quiet
= 1;
3964 /* Verbose level can increase using multiple -v */
3965 lttng_opt_verbose
+= 1;
3968 opt_verbose_consumer
+= 1;
3971 consumerd32_bin
= optarg
;
3974 consumerd32_libdir
= optarg
;
3977 consumerd64_bin
= optarg
;
3980 consumerd64_libdir
= optarg
;
3983 opt_pidfile
= optarg
;
3986 /* Unknown option or other error.
3987 * Error is printed by getopt, just return */
3996 * Creates the two needed socket by the daemon.
3997 * apps_sock - The communication socket for all UST apps.
3998 * client_sock - The communication of the cli tool (lttng).
4000 static int init_daemon_socket(void)
4005 old_umask
= umask(0);
4007 /* Create client tool unix socket */
4008 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4009 if (client_sock
< 0) {
4010 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4015 /* Set the cloexec flag */
4016 ret
= utils_set_fd_cloexec(client_sock
);
4018 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4019 "Continuing but note that the consumer daemon will have a "
4020 "reference to this socket on exec()", client_sock
);
4023 /* File permission MUST be 660 */
4024 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4026 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4031 /* Create the application unix socket */
4032 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4033 if (apps_sock
< 0) {
4034 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4039 /* Set the cloexec flag */
4040 ret
= utils_set_fd_cloexec(apps_sock
);
4042 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4043 "Continuing but note that the consumer daemon will have a "
4044 "reference to this socket on exec()", apps_sock
);
4047 /* File permission MUST be 666 */
4048 ret
= chmod(apps_unix_sock_path
,
4049 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4051 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4056 DBG3("Session daemon client socket %d and application socket %d created",
4057 client_sock
, apps_sock
);
4065 * Check if the global socket is available, and if a daemon is answering at the
4066 * other side. If yes, error is returned.
4068 static int check_existing_daemon(void)
4070 /* Is there anybody out there ? */
4071 if (lttng_session_daemon_alive()) {
4079 * Set the tracing group gid onto the client socket.
4081 * Race window between mkdir and chown is OK because we are going from more
4082 * permissive (root.root) to less permissive (root.tracing).
4084 static int set_permissions(char *rundir
)
4089 ret
= allowed_group();
4091 WARN("No tracing group detected");
4092 /* Setting gid to 0 if no tracing group is found */
4098 /* Set lttng run dir */
4099 ret
= chown(rundir
, 0, gid
);
4101 ERR("Unable to set group on %s", rundir
);
4105 /* Ensure all applications and tracing group can search the run dir */
4106 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
4108 ERR("Unable to set permissions on %s", rundir
);
4112 /* lttng client socket path */
4113 ret
= chown(client_unix_sock_path
, 0, gid
);
4115 ERR("Unable to set group on %s", client_unix_sock_path
);
4119 /* kconsumer error socket path */
4120 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
4122 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4126 /* 64-bit ustconsumer error socket path */
4127 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
4129 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4133 /* 32-bit ustconsumer compat32 error socket path */
4134 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
4136 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4140 DBG("All permissions are set");
4146 * Create the lttng run directory needed for all global sockets and pipe.
4148 static int create_lttng_rundir(const char *rundir
)
4152 DBG3("Creating LTTng run directory: %s", rundir
);
4154 ret
= mkdir(rundir
, S_IRWXU
);
4156 if (errno
!= EEXIST
) {
4157 ERR("Unable to create %s", rundir
);
4169 * Setup sockets and directory needed by the kconsumerd communication with the
4172 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4176 char path
[PATH_MAX
];
4178 switch (consumer_data
->type
) {
4179 case LTTNG_CONSUMER_KERNEL
:
4180 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4182 case LTTNG_CONSUMER64_UST
:
4183 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4185 case LTTNG_CONSUMER32_UST
:
4186 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4189 ERR("Consumer type unknown");
4194 DBG2("Creating consumer directory: %s", path
);
4196 ret
= mkdir(path
, S_IRWXU
);
4198 if (errno
!= EEXIST
) {
4200 ERR("Failed to create %s", path
);
4206 /* Create the kconsumerd error unix socket */
4207 consumer_data
->err_sock
=
4208 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4209 if (consumer_data
->err_sock
< 0) {
4210 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4216 * Set the CLOEXEC flag. Return code is useless because either way, the
4219 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4221 PERROR("utils_set_fd_cloexec");
4222 /* continue anyway */
4225 /* File permission MUST be 660 */
4226 ret
= chmod(consumer_data
->err_unix_sock_path
,
4227 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4229 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4239 * Signal handler for the daemon
4241 * Simply stop all worker threads, leaving main() return gracefully after
4242 * joining all threads and calling cleanup().
4244 static void sighandler(int sig
)
4248 DBG("SIGPIPE caught");
4251 DBG("SIGINT caught");
4255 DBG("SIGTERM caught");
4264 * Setup signal handler for :
4265 * SIGINT, SIGTERM, SIGPIPE
4267 static int set_signal_handler(void)
4270 struct sigaction sa
;
4273 if ((ret
= sigemptyset(&sigset
)) < 0) {
4274 PERROR("sigemptyset");
4278 sa
.sa_handler
= sighandler
;
4279 sa
.sa_mask
= sigset
;
4281 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4282 PERROR("sigaction");
4286 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4287 PERROR("sigaction");
4291 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4292 PERROR("sigaction");
4296 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4302 * Set open files limit to unlimited. This daemon can open a large number of
4303 * file descriptors in order to consumer multiple kernel traces.
4305 static void set_ulimit(void)
4310 /* The kernel does not allowed an infinite limit for open files */
4311 lim
.rlim_cur
= 65535;
4312 lim
.rlim_max
= 65535;
4314 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4316 PERROR("failed to set open files limit");
4321 * Write pidfile using the rundir and opt_pidfile.
4323 static void write_pidfile(void)
4326 char pidfile_path
[PATH_MAX
];
4331 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4333 /* Build pidfile path from rundir and opt_pidfile. */
4334 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4335 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4337 PERROR("snprintf pidfile path");
4343 * Create pid file in rundir. Return value is of no importance. The
4344 * execution will continue even though we are not able to write the file.
4346 (void) utils_create_pid_file(getpid(), pidfile_path
);
4355 int main(int argc
, char **argv
)
4359 const char *home_path
, *env_app_timeout
;
4361 init_kernel_workarounds();
4363 rcu_register_thread();
4365 setup_consumerd_path();
4367 page_size
= sysconf(_SC_PAGESIZE
);
4368 if (page_size
< 0) {
4369 PERROR("sysconf _SC_PAGESIZE");
4370 page_size
= LONG_MAX
;
4371 WARN("Fallback page size to %ld", page_size
);
4374 /* Parse arguments */
4376 if ((ret
= parse_args(argc
, argv
)) < 0) {
4386 * child: setsid, close FD 0, 1, 2, chdir /
4387 * parent: exit (if fork is successful)
4395 * We are in the child. Make sure all other file
4396 * descriptors are closed, in case we are called with
4397 * more opened file descriptors than the standard ones.
4399 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4404 /* Create thread quit pipe */
4405 if ((ret
= init_thread_quit_pipe()) < 0) {
4409 /* Check if daemon is UID = 0 */
4410 is_root
= !getuid();
4413 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4415 /* Create global run dir with root access */
4416 ret
= create_lttng_rundir(rundir
);
4421 if (strlen(apps_unix_sock_path
) == 0) {
4422 snprintf(apps_unix_sock_path
, PATH_MAX
,
4423 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4426 if (strlen(client_unix_sock_path
) == 0) {
4427 snprintf(client_unix_sock_path
, PATH_MAX
,
4428 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4431 /* Set global SHM for ust */
4432 if (strlen(wait_shm_path
) == 0) {
4433 snprintf(wait_shm_path
, PATH_MAX
,
4434 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4437 if (strlen(health_unix_sock_path
) == 0) {
4438 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4439 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4442 /* Setup kernel consumerd path */
4443 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4444 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4445 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4446 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4448 DBG2("Kernel consumer err path: %s",
4449 kconsumer_data
.err_unix_sock_path
);
4450 DBG2("Kernel consumer cmd path: %s",
4451 kconsumer_data
.cmd_unix_sock_path
);
4453 home_path
= utils_get_home_dir();
4454 if (home_path
== NULL
) {
4455 /* TODO: Add --socket PATH option */
4456 ERR("Can't get HOME directory for sockets creation.");
4462 * Create rundir from home path. This will create something like
4465 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4471 ret
= create_lttng_rundir(rundir
);
4476 if (strlen(apps_unix_sock_path
) == 0) {
4477 snprintf(apps_unix_sock_path
, PATH_MAX
,
4478 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4481 /* Set the cli tool unix socket path */
4482 if (strlen(client_unix_sock_path
) == 0) {
4483 snprintf(client_unix_sock_path
, PATH_MAX
,
4484 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4487 /* Set global SHM for ust */
4488 if (strlen(wait_shm_path
) == 0) {
4489 snprintf(wait_shm_path
, PATH_MAX
,
4490 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4493 /* Set health check Unix path */
4494 if (strlen(health_unix_sock_path
) == 0) {
4495 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4496 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4500 /* Set consumer initial state */
4501 kernel_consumerd_state
= CONSUMER_STOPPED
;
4502 ust_consumerd_state
= CONSUMER_STOPPED
;
4504 DBG("Client socket path %s", client_unix_sock_path
);
4505 DBG("Application socket path %s", apps_unix_sock_path
);
4506 DBG("Application wait path %s", wait_shm_path
);
4507 DBG("LTTng run directory path: %s", rundir
);
4509 /* 32 bits consumerd path setup */
4510 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4511 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4512 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4513 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4515 DBG2("UST consumer 32 bits err path: %s",
4516 ustconsumer32_data
.err_unix_sock_path
);
4517 DBG2("UST consumer 32 bits cmd path: %s",
4518 ustconsumer32_data
.cmd_unix_sock_path
);
4520 /* 64 bits consumerd path setup */
4521 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4522 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4523 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4524 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4526 DBG2("UST consumer 64 bits err path: %s",
4527 ustconsumer64_data
.err_unix_sock_path
);
4528 DBG2("UST consumer 64 bits cmd path: %s",
4529 ustconsumer64_data
.cmd_unix_sock_path
);
4532 * See if daemon already exist.
4534 if ((ret
= check_existing_daemon()) < 0) {
4535 ERR("Already running daemon.\n");
4537 * We do not goto exit because we must not cleanup()
4538 * because a daemon is already running.
4544 * Init UST app hash table. Alloc hash table before this point since
4545 * cleanup() can get called after that point.
4549 /* After this point, we can safely call cleanup() with "goto exit" */
4552 * These actions must be executed as root. We do that *after* setting up
4553 * the sockets path because we MUST make the check for another daemon using
4554 * those paths *before* trying to set the kernel consumer sockets and init
4558 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4563 /* Setup kernel tracer */
4564 if (!opt_no_kernel
) {
4565 init_kernel_tracer();
4568 /* Set ulimit for open files */
4571 /* init lttng_fd tracking must be done after set_ulimit. */
4574 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4579 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4584 if ((ret
= set_signal_handler()) < 0) {
4588 /* Setup the needed unix socket */
4589 if ((ret
= init_daemon_socket()) < 0) {
4593 /* Set credentials to socket */
4594 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4598 /* Get parent pid if -S, --sig-parent is specified. */
4599 if (opt_sig_parent
) {
4603 /* Setup the kernel pipe for waking up the kernel thread */
4604 if (is_root
&& !opt_no_kernel
) {
4605 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4610 /* Setup the thread ht_cleanup communication pipe. */
4611 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4615 /* Setup the thread apps communication pipe. */
4616 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4620 /* Setup the thread apps notify communication pipe. */
4621 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4625 /* Initialize global buffer per UID and PID registry. */
4626 buffer_reg_init_uid_registry();
4627 buffer_reg_init_pid_registry();
4629 /* Init UST command queue. */
4630 cds_wfq_init(&ust_cmd_queue
.queue
);
4633 * Get session list pointer. This pointer MUST NOT be free(). This list is
4634 * statically declared in session.c
4636 session_list_ptr
= session_get_list();
4638 /* Set up max poll set size */
4639 lttng_poll_set_max_size();
4643 /* Check for the application socket timeout env variable. */
4644 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4645 if (env_app_timeout
) {
4646 app_socket_timeout
= atoi(env_app_timeout
);
4648 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4653 /* Initialize communication library */
4655 /* This is to get the TCP timeout value. */
4656 lttcomm_inet_init();
4659 * Initialize the health check subsystem. This call should set the
4660 * appropriate time values.
4664 /* Create thread to manage the client socket */
4665 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4666 thread_ht_cleanup
, (void *) NULL
);
4668 PERROR("pthread_create ht_cleanup");
4669 goto exit_ht_cleanup
;
4672 /* Create thread to manage the client socket */
4673 ret
= pthread_create(&health_thread
, NULL
,
4674 thread_manage_health
, (void *) NULL
);
4676 PERROR("pthread_create health");
4680 /* Create thread to manage the client socket */
4681 ret
= pthread_create(&client_thread
, NULL
,
4682 thread_manage_clients
, (void *) NULL
);
4684 PERROR("pthread_create clients");
4688 /* Create thread to dispatch registration */
4689 ret
= pthread_create(&dispatch_thread
, NULL
,
4690 thread_dispatch_ust_registration
, (void *) NULL
);
4692 PERROR("pthread_create dispatch");
4696 /* Create thread to manage application registration. */
4697 ret
= pthread_create(®_apps_thread
, NULL
,
4698 thread_registration_apps
, (void *) NULL
);
4700 PERROR("pthread_create registration");
4704 /* Create thread to manage application socket */
4705 ret
= pthread_create(&apps_thread
, NULL
,
4706 thread_manage_apps
, (void *) NULL
);
4708 PERROR("pthread_create apps");
4712 /* Create thread to manage application notify socket */
4713 ret
= pthread_create(&apps_notify_thread
, NULL
,
4714 ust_thread_manage_notify
, (void *) NULL
);
4716 PERROR("pthread_create apps");
4720 /* Don't start this thread if kernel tracing is not requested nor root */
4721 if (is_root
&& !opt_no_kernel
) {
4722 /* Create kernel thread to manage kernel event */
4723 ret
= pthread_create(&kernel_thread
, NULL
,
4724 thread_manage_kernel
, (void *) NULL
);
4726 PERROR("pthread_create kernel");
4730 ret
= pthread_join(kernel_thread
, &status
);
4732 PERROR("pthread_join");
4733 goto error
; /* join error, exit without cleanup */
4738 ret
= pthread_join(apps_thread
, &status
);
4740 PERROR("pthread_join");
4741 goto error
; /* join error, exit without cleanup */
4745 ret
= pthread_join(reg_apps_thread
, &status
);
4747 PERROR("pthread_join");
4748 goto error
; /* join error, exit without cleanup */
4752 ret
= pthread_join(dispatch_thread
, &status
);
4754 PERROR("pthread_join");
4755 goto error
; /* join error, exit without cleanup */
4759 ret
= pthread_join(client_thread
, &status
);
4761 PERROR("pthread_join");
4762 goto error
; /* join error, exit without cleanup */
4765 ret
= join_consumer_thread(&kconsumer_data
);
4767 PERROR("join_consumer");
4768 goto error
; /* join error, exit without cleanup */
4771 ret
= join_consumer_thread(&ustconsumer32_data
);
4773 PERROR("join_consumer ust32");
4774 goto error
; /* join error, exit without cleanup */
4777 ret
= join_consumer_thread(&ustconsumer64_data
);
4779 PERROR("join_consumer ust64");
4780 goto error
; /* join error, exit without cleanup */
4784 ret
= pthread_join(health_thread
, &status
);
4786 PERROR("pthread_join health thread");
4787 goto error
; /* join error, exit without cleanup */
4791 ret
= pthread_join(ht_cleanup_thread
, &status
);
4793 PERROR("pthread_join ht cleanup thread");
4794 goto error
; /* join error, exit without cleanup */
4799 * cleanup() is called when no other thread is running.
4801 rcu_thread_online();
4803 rcu_thread_offline();
4804 rcu_unregister_thread();