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 .metadata_sock
.fd
= -1,
94 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
95 .lock
= PTHREAD_MUTEX_INITIALIZER
,
96 .cond
= PTHREAD_COND_INITIALIZER
,
97 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
99 static struct consumer_data ustconsumer64_data
= {
100 .type
= LTTNG_CONSUMER64_UST
,
101 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
102 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
105 .metadata_sock
.fd
= -1,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer32_data
= {
112 .type
= LTTNG_CONSUMER32_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
117 .metadata_sock
.fd
= -1,
118 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 .lock
= PTHREAD_MUTEX_INITIALIZER
,
120 .cond
= PTHREAD_COND_INITIALIZER
,
121 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
124 /* Shared between threads */
125 static int dispatch_thread_exit
;
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path
[PATH_MAX
];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path
[PATH_MAX
];
131 /* global wait shm path for UST */
132 static char wait_shm_path
[PATH_MAX
];
133 /* Global health check unix path */
134 static char health_unix_sock_path
[PATH_MAX
];
136 /* Sockets and FDs */
137 static int client_sock
= -1;
138 static int apps_sock
= -1;
139 int kernel_tracer_fd
= -1;
140 static int kernel_poll_pipe
[2] = { -1, -1 };
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
146 static int thread_quit_pipe
[2] = { -1, -1 };
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
152 static int apps_cmd_pipe
[2] = { -1, -1 };
154 int apps_cmd_notify_pipe
[2] = { -1, -1 };
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread
;
158 static pthread_t apps_notify_thread
;
159 static pthread_t reg_apps_thread
;
160 static pthread_t client_thread
;
161 static pthread_t kernel_thread
;
162 static pthread_t dispatch_thread
;
163 static pthread_t health_thread
;
164 static pthread_t ht_cleanup_thread
;
167 * UST registration command queue. This queue is tied with a futex and uses a N
168 * wakers / 1 waiter implemented and detailed in futex.c/.h
170 * The thread_manage_apps and thread_dispatch_ust_registration interact with
171 * this queue and the wait/wake scheme.
173 static struct ust_cmd_queue ust_cmd_queue
;
176 * Pointer initialized before thread creation.
178 * This points to the tracing session list containing the session count and a
179 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
180 * MUST NOT be taken if you call a public function in session.c.
182 * The lock is nested inside the structure: session_list_ptr->lock. Please use
183 * session_lock_list and session_unlock_list for lock acquisition.
185 static struct ltt_session_list
*session_list_ptr
;
187 int ust_consumerd64_fd
= -1;
188 int ust_consumerd32_fd
= -1;
190 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
191 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
192 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
193 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
195 static const char *module_proc_lttng
= "/proc/lttng";
198 * Consumer daemon state which is changed when spawning it, killing it or in
199 * case of a fatal error.
201 enum consumerd_state
{
202 CONSUMER_STARTED
= 1,
203 CONSUMER_STOPPED
= 2,
208 * This consumer daemon state is used to validate if a client command will be
209 * able to reach the consumer. If not, the client is informed. For instance,
210 * doing a "lttng start" when the consumer state is set to ERROR will return an
211 * error to the client.
213 * The following example shows a possible race condition of this scheme:
215 * consumer thread error happens
217 * client cmd checks state -> still OK
218 * consumer thread exit, sets error
219 * client cmd try to talk to consumer
222 * However, since the consumer is a different daemon, we have no way of making
223 * sure the command will reach it safely even with this state flag. This is why
224 * we consider that up to the state validation during command processing, the
225 * command is safe. After that, we can not guarantee the correctness of the
226 * client request vis-a-vis the consumer.
228 static enum consumerd_state ust_consumerd_state
;
229 static enum consumerd_state kernel_consumerd_state
;
232 * Socket timeout for receiving and sending in seconds.
234 static int app_socket_timeout
;
236 /* Set in main() with the current page size. */
240 void setup_consumerd_path(void)
242 const char *bin
, *libdir
;
245 * Allow INSTALL_BIN_PATH to be used as a target path for the
246 * native architecture size consumer if CONFIG_CONSUMER*_PATH
247 * has not been defined.
249 #if (CAA_BITS_PER_LONG == 32)
250 if (!consumerd32_bin
[0]) {
251 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
253 if (!consumerd32_libdir
[0]) {
254 consumerd32_libdir
= INSTALL_LIB_PATH
;
256 #elif (CAA_BITS_PER_LONG == 64)
257 if (!consumerd64_bin
[0]) {
258 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
260 if (!consumerd64_libdir
[0]) {
261 consumerd64_libdir
= INSTALL_LIB_PATH
;
264 #error "Unknown bitness"
268 * runtime env. var. overrides the build default.
270 bin
= getenv("LTTNG_CONSUMERD32_BIN");
272 consumerd32_bin
= bin
;
274 bin
= getenv("LTTNG_CONSUMERD64_BIN");
276 consumerd64_bin
= bin
;
278 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
280 consumerd32_libdir
= libdir
;
282 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
284 consumerd64_libdir
= libdir
;
289 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
291 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
297 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
303 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
315 * Check if the thread quit pipe was triggered.
317 * Return 1 if it was triggered else 0;
319 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
321 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
329 * Return group ID of the tracing group or -1 if not found.
331 static gid_t
allowed_group(void)
335 if (opt_tracing_group
) {
336 grp
= getgrnam(opt_tracing_group
);
338 grp
= getgrnam(default_tracing_group
);
348 * Init thread quit pipe.
350 * Return -1 on error or 0 if all pipes are created.
352 static int init_thread_quit_pipe(void)
356 ret
= pipe(thread_quit_pipe
);
358 PERROR("thread quit pipe");
362 for (i
= 0; i
< 2; i
++) {
363 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
375 * Stop all threads by closing the thread quit pipe.
377 static void stop_threads(void)
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
385 ERR("write error on thread quit pipe");
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
390 futex_nto1_wake(&ust_cmd_queue
.futex
);
394 * Close every consumer sockets.
396 static void close_consumer_sockets(void)
400 if (kconsumer_data
.err_sock
>= 0) {
401 ret
= close(kconsumer_data
.err_sock
);
403 PERROR("kernel consumer err_sock close");
406 if (ustconsumer32_data
.err_sock
>= 0) {
407 ret
= close(ustconsumer32_data
.err_sock
);
409 PERROR("UST consumerd32 err_sock close");
412 if (ustconsumer64_data
.err_sock
>= 0) {
413 ret
= close(ustconsumer64_data
.err_sock
);
415 PERROR("UST consumerd64 err_sock close");
418 if (kconsumer_data
.cmd_sock
>= 0) {
419 ret
= close(kconsumer_data
.cmd_sock
);
421 PERROR("kernel consumer cmd_sock close");
424 if (ustconsumer32_data
.cmd_sock
>= 0) {
425 ret
= close(ustconsumer32_data
.cmd_sock
);
427 PERROR("UST consumerd32 cmd_sock close");
430 if (ustconsumer64_data
.cmd_sock
>= 0) {
431 ret
= close(ustconsumer64_data
.cmd_sock
);
433 PERROR("UST consumerd64 cmd_sock close");
441 static void cleanup(void)
445 struct ltt_session
*sess
, *stmp
;
449 /* First thing first, stop all threads */
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 /* Code flow error */
694 assert(socket
->fd
>= 0);
696 pthread_mutex_lock(socket
->lock
);
697 ret
= kernel_consumer_send_channel_stream(socket
,
699 session
->output_traces
? 1 : 0);
700 pthread_mutex_unlock(socket
->lock
);
711 session_unlock(session
);
713 session_unlock_list();
717 session_unlock(session
);
718 session_unlock_list();
723 * For each tracing session, update newly registered apps. The session list
724 * lock MUST be acquired before calling this.
726 static void update_ust_app(int app_sock
)
728 struct ltt_session
*sess
, *stmp
;
730 /* For all tracing session(s) */
731 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
733 if (sess
->ust_session
) {
734 ust_app_global_update(sess
->ust_session
, app_sock
);
736 session_unlock(sess
);
741 * This thread manage event coming from the kernel.
743 * Features supported in this thread:
746 static void *thread_manage_kernel(void *data
)
748 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
749 uint32_t revents
, nb_fd
;
751 struct lttng_poll_event events
;
753 DBG("[thread] Thread manage kernel started");
755 health_register(HEALTH_TYPE_KERNEL
);
758 * This first step of the while is to clean this structure which could free
759 * non NULL pointers so initialize it before the loop.
761 lttng_poll_init(&events
);
763 if (testpoint(thread_manage_kernel
)) {
764 goto error_testpoint
;
767 health_code_update();
769 if (testpoint(thread_manage_kernel_before_loop
)) {
770 goto error_testpoint
;
774 health_code_update();
776 if (update_poll_flag
== 1) {
777 /* Clean events object. We are about to populate it again. */
778 lttng_poll_clean(&events
);
780 ret
= sessiond_set_thread_pollset(&events
, 2);
782 goto error_poll_create
;
785 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
790 /* This will add the available kernel channel if any. */
791 ret
= update_kernel_poll(&events
);
795 update_poll_flag
= 0;
798 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
800 /* Poll infinite value of time */
803 ret
= lttng_poll_wait(&events
, -1);
807 * Restart interrupted system call.
809 if (errno
== EINTR
) {
813 } else if (ret
== 0) {
814 /* Should not happen since timeout is infinite */
815 ERR("Return value of poll is 0 with an infinite timeout.\n"
816 "This should not have happened! Continuing...");
822 for (i
= 0; i
< nb_fd
; i
++) {
823 /* Fetch once the poll data */
824 revents
= LTTNG_POLL_GETEV(&events
, i
);
825 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
827 health_code_update();
829 /* Thread quit pipe has been closed. Killing thread. */
830 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
836 /* Check for data on kernel pipe */
837 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
839 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
840 } while (ret
< 0 && errno
== EINTR
);
842 * Ret value is useless here, if this pipe gets any actions an
843 * update is required anyway.
845 update_poll_flag
= 1;
849 * New CPU detected by the kernel. Adding kernel stream to
850 * kernel session and updating the kernel consumer
852 if (revents
& LPOLLIN
) {
853 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
859 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
860 * and unregister kernel stream at this point.
869 lttng_poll_clean(&events
);
872 utils_close_pipe(kernel_poll_pipe
);
873 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
876 ERR("Health error occurred in %s", __func__
);
877 WARN("Kernel thread died unexpectedly. "
878 "Kernel tracing can continue but CPU hotplug is disabled.");
881 DBG("Kernel thread dying");
886 * Signal pthread condition of the consumer data that the thread.
888 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
890 pthread_mutex_lock(&data
->cond_mutex
);
893 * The state is set before signaling. It can be any value, it's the waiter
894 * job to correctly interpret this condition variable associated to the
895 * consumer pthread_cond.
897 * A value of 0 means that the corresponding thread of the consumer data
898 * was not started. 1 indicates that the thread has started and is ready
899 * for action. A negative value means that there was an error during the
902 data
->consumer_thread_is_ready
= state
;
903 (void) pthread_cond_signal(&data
->cond
);
905 pthread_mutex_unlock(&data
->cond_mutex
);
909 * This thread manage the consumer error sent back to the session daemon.
911 static void *thread_manage_consumer(void *data
)
913 int sock
= -1, i
, ret
, pollfd
, err
= -1;
914 uint32_t revents
, nb_fd
;
915 enum lttcomm_return_code code
;
916 struct lttng_poll_event events
;
917 struct consumer_data
*consumer_data
= data
;
919 DBG("[thread] Manage consumer started");
921 health_register(HEALTH_TYPE_CONSUMER
);
923 health_code_update();
926 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
927 * metadata_sock. Nothing more will be added to this poll set.
929 ret
= sessiond_set_thread_pollset(&events
, 3);
935 * The error socket here is already in a listening state which was done
936 * just before spawning this thread to avoid a race between the consumer
937 * daemon exec trying to connect and the listen() call.
939 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
944 health_code_update();
946 /* Infinite blocking call, waiting for transmission */
950 if (testpoint(thread_manage_consumer
)) {
954 ret
= lttng_poll_wait(&events
, -1);
958 * Restart interrupted system call.
960 if (errno
== EINTR
) {
968 for (i
= 0; i
< nb_fd
; i
++) {
969 /* Fetch once the poll data */
970 revents
= LTTNG_POLL_GETEV(&events
, i
);
971 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
973 health_code_update();
975 /* Thread quit pipe has been closed. Killing thread. */
976 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
982 /* Event on the registration socket */
983 if (pollfd
== consumer_data
->err_sock
) {
984 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
985 ERR("consumer err socket poll error");
991 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
997 * Set the CLOEXEC flag. Return code is useless because either way, the
1000 (void) utils_set_fd_cloexec(sock
);
1002 health_code_update();
1004 DBG2("Receiving code from consumer err_sock");
1006 /* Getting status code from kconsumerd */
1007 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1008 sizeof(enum lttcomm_return_code
));
1013 health_code_update();
1015 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1016 /* Connect both socket, command and metadata. */
1017 consumer_data
->cmd_sock
=
1018 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1019 consumer_data
->metadata_sock
.fd
=
1020 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1021 if (consumer_data
->cmd_sock
< 0 ||
1022 consumer_data
->metadata_sock
.fd
< 0) {
1023 PERROR("consumer connect cmd socket");
1024 /* On error, signal condition and quit. */
1025 signal_consumer_condition(consumer_data
, -1);
1028 /* Create metadata socket lock. */
1029 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1030 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1031 PERROR("zmalloc pthread mutex");
1035 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1037 signal_consumer_condition(consumer_data
, 1);
1038 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1039 DBG("Consumer metadata socket ready (fd: %d)",
1040 consumer_data
->metadata_sock
.fd
);
1042 ERR("consumer error when waiting for SOCK_READY : %s",
1043 lttcomm_get_readable_code(-code
));
1047 /* Remove the consumerd error sock since we've established a connexion */
1048 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1053 /* Add new accepted error socket. */
1054 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1059 /* Add metadata socket that is successfully connected. */
1060 ret
= lttng_poll_add(&events
, consumer_data
->metadata_sock
.fd
,
1061 LPOLLIN
| LPOLLRDHUP
);
1066 health_code_update();
1068 /* Infinite blocking call, waiting for transmission */
1071 health_poll_entry();
1072 ret
= lttng_poll_wait(&events
, -1);
1076 * Restart interrupted system call.
1078 if (errno
== EINTR
) {
1086 for (i
= 0; i
< nb_fd
; i
++) {
1087 /* Fetch once the poll data */
1088 revents
= LTTNG_POLL_GETEV(&events
, i
);
1089 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1091 health_code_update();
1093 /* Thread quit pipe has been closed. Killing thread. */
1094 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1100 if (pollfd
== sock
) {
1101 /* Event on the consumerd socket */
1102 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1103 ERR("consumer err socket second poll error");
1106 health_code_update();
1107 /* Wait for any kconsumerd error */
1108 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1109 sizeof(enum lttcomm_return_code
));
1111 ERR("consumer closed the command socket");
1115 ERR("consumer return code : %s",
1116 lttcomm_get_readable_code(-code
));
1119 } else if (pollfd
== consumer_data
->metadata_sock
.fd
) {
1120 /* UST metadata requests */
1121 ret
= ust_consumer_metadata_request(
1122 &consumer_data
->metadata_sock
);
1124 ERR("Handling metadata request");
1129 ERR("Unknown pollfd");
1133 health_code_update();
1138 /* Immediately set the consumerd state to stopped */
1139 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1140 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1141 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1142 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1143 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1145 /* Code flow error... */
1149 if (consumer_data
->err_sock
>= 0) {
1150 ret
= close(consumer_data
->err_sock
);
1154 consumer_data
->err_sock
= -1;
1156 if (consumer_data
->cmd_sock
>= 0) {
1157 ret
= close(consumer_data
->cmd_sock
);
1161 consumer_data
->cmd_sock
= -1;
1163 if (consumer_data
->metadata_sock
.fd
>= 0) {
1164 ret
= close(consumer_data
->metadata_sock
.fd
);
1169 /* Cleanup metadata socket mutex. */
1170 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1171 free(consumer_data
->metadata_sock
.lock
);
1180 unlink(consumer_data
->err_unix_sock_path
);
1181 unlink(consumer_data
->cmd_unix_sock_path
);
1182 consumer_data
->pid
= 0;
1184 lttng_poll_clean(&events
);
1188 ERR("Health error occurred in %s", __func__
);
1190 health_unregister();
1191 DBG("consumer thread cleanup completed");
1197 * This thread manage application communication.
1199 static void *thread_manage_apps(void *data
)
1201 int i
, ret
, pollfd
, err
= -1;
1202 uint32_t revents
, nb_fd
;
1203 struct lttng_poll_event events
;
1205 DBG("[thread] Manage application started");
1207 rcu_register_thread();
1208 rcu_thread_online();
1210 health_register(HEALTH_TYPE_APP_MANAGE
);
1212 if (testpoint(thread_manage_apps
)) {
1213 goto error_testpoint
;
1216 health_code_update();
1218 ret
= sessiond_set_thread_pollset(&events
, 2);
1220 goto error_poll_create
;
1223 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1228 if (testpoint(thread_manage_apps_before_loop
)) {
1232 health_code_update();
1235 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1237 /* Inifinite blocking call, waiting for transmission */
1239 health_poll_entry();
1240 ret
= lttng_poll_wait(&events
, -1);
1244 * Restart interrupted system call.
1246 if (errno
== EINTR
) {
1254 for (i
= 0; i
< nb_fd
; i
++) {
1255 /* Fetch once the poll data */
1256 revents
= LTTNG_POLL_GETEV(&events
, i
);
1257 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1259 health_code_update();
1261 /* Thread quit pipe has been closed. Killing thread. */
1262 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1268 /* Inspect the apps cmd pipe */
1269 if (pollfd
== apps_cmd_pipe
[0]) {
1270 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1271 ERR("Apps command pipe error");
1273 } else if (revents
& LPOLLIN
) {
1278 ret
= read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1279 } while (ret
< 0 && errno
== EINTR
);
1280 if (ret
< 0 || ret
< sizeof(sock
)) {
1281 PERROR("read apps cmd pipe");
1285 health_code_update();
1288 * We only monitor the error events of the socket. This
1289 * thread does not handle any incoming data from UST
1292 ret
= lttng_poll_add(&events
, sock
,
1293 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1298 /* Set socket timeout for both receiving and ending */
1299 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1300 app_socket_timeout
);
1301 (void) lttcomm_setsockopt_snd_timeout(sock
,
1302 app_socket_timeout
);
1304 DBG("Apps with sock %d added to poll set", sock
);
1306 health_code_update();
1312 * At this point, we know that a registered application made
1313 * the event at poll_wait.
1315 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1316 /* Removing from the poll set */
1317 ret
= lttng_poll_del(&events
, pollfd
);
1322 /* Socket closed on remote end. */
1323 ust_app_unregister(pollfd
);
1328 health_code_update();
1334 lttng_poll_clean(&events
);
1337 utils_close_pipe(apps_cmd_pipe
);
1338 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1341 * We don't clean the UST app hash table here since already registered
1342 * applications can still be controlled so let them be until the session
1343 * daemon dies or the applications stop.
1348 ERR("Health error occurred in %s", __func__
);
1350 health_unregister();
1351 DBG("Application communication apps thread cleanup complete");
1352 rcu_thread_offline();
1353 rcu_unregister_thread();
1358 * Send a socket to a thread This is called from the dispatch UST registration
1359 * thread once all sockets are set for the application.
1361 * On success, return 0 else a negative value being the errno message of the
1364 static int send_socket_to_thread(int fd
, int sock
)
1368 /* Sockets MUST be set or else this should not have been called. */
1373 ret
= write(fd
, &sock
, sizeof(sock
));
1374 } while (ret
< 0 && errno
== EINTR
);
1375 if (ret
< 0 || ret
!= sizeof(sock
)) {
1376 PERROR("write apps pipe %d", fd
);
1383 /* All good. Don't send back the write positive ret value. */
1390 * Sanitize the wait queue of the dispatch registration thread meaning removing
1391 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1392 * notify socket is never received.
1394 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1396 int ret
, nb_fd
= 0, i
;
1397 unsigned int fd_added
= 0;
1398 struct lttng_poll_event events
;
1399 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1403 lttng_poll_init(&events
);
1405 /* Just skip everything for an empty queue. */
1406 if (!wait_queue
->count
) {
1410 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1415 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1416 &wait_queue
->head
, head
) {
1417 assert(wait_node
->app
);
1418 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1419 LPOLLHUP
| LPOLLERR
);
1432 * Poll but don't block so we can quickly identify the faulty events and
1433 * clean them afterwards from the wait queue.
1435 ret
= lttng_poll_wait(&events
, 0);
1441 for (i
= 0; i
< nb_fd
; i
++) {
1442 /* Get faulty FD. */
1443 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1444 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1446 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1447 &wait_queue
->head
, head
) {
1448 if (pollfd
== wait_node
->app
->sock
&&
1449 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1450 cds_list_del(&wait_node
->head
);
1451 wait_queue
->count
--;
1452 ust_app_destroy(wait_node
->app
);
1460 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1464 lttng_poll_clean(&events
);
1468 lttng_poll_clean(&events
);
1470 ERR("Unable to sanitize wait queue");
1475 * Dispatch request from the registration threads to the application
1476 * communication thread.
1478 static void *thread_dispatch_ust_registration(void *data
)
1481 struct cds_wfq_node
*node
;
1482 struct ust_command
*ust_cmd
= NULL
;
1483 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1484 struct ust_reg_wait_queue wait_queue
= {
1488 health_register(HEALTH_TYPE_APP_REG_DISPATCH
);
1490 health_code_update();
1492 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1494 DBG("[thread] Dispatch UST command started");
1496 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1497 health_code_update();
1499 /* Atomically prepare the queue futex */
1500 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1503 struct ust_app
*app
= NULL
;
1507 * Make sure we don't have node(s) that have hung up before receiving
1508 * the notify socket. This is to clean the list in order to avoid
1509 * memory leaks from notify socket that are never seen.
1511 sanitize_wait_queue(&wait_queue
);
1513 health_code_update();
1514 /* Dequeue command for registration */
1515 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1517 DBG("Woken up but nothing in the UST command queue");
1518 /* Continue thread execution */
1522 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1524 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1525 " gid:%d sock:%d name:%s (version %d.%d)",
1526 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1527 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1528 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1529 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1531 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1532 wait_node
= zmalloc(sizeof(*wait_node
));
1534 PERROR("zmalloc wait_node dispatch");
1535 ret
= close(ust_cmd
->sock
);
1537 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1539 lttng_fd_put(1, LTTNG_FD_APPS
);
1543 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1545 /* Create application object if socket is CMD. */
1546 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1548 if (!wait_node
->app
) {
1549 ret
= close(ust_cmd
->sock
);
1551 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1553 lttng_fd_put(1, LTTNG_FD_APPS
);
1559 * Add application to the wait queue so we can set the notify
1560 * socket before putting this object in the global ht.
1562 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1567 * We have to continue here since we don't have the notify
1568 * socket and the application MUST be added to the hash table
1569 * only at that moment.
1574 * Look for the application in the local wait queue and set the
1575 * notify socket if found.
1577 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1578 &wait_queue
.head
, head
) {
1579 health_code_update();
1580 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1581 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1582 cds_list_del(&wait_node
->head
);
1584 app
= wait_node
->app
;
1586 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1592 * With no application at this stage the received socket is
1593 * basically useless so close it before we free the cmd data
1594 * structure for good.
1597 ret
= close(ust_cmd
->sock
);
1599 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1601 lttng_fd_put(1, LTTNG_FD_APPS
);
1608 * @session_lock_list
1610 * Lock the global session list so from the register up to the
1611 * registration done message, no thread can see the application
1612 * and change its state.
1614 session_lock_list();
1618 * Add application to the global hash table. This needs to be
1619 * done before the update to the UST registry can locate the
1624 /* Set app version. This call will print an error if needed. */
1625 (void) ust_app_version(app
);
1627 /* Send notify socket through the notify pipe. */
1628 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1632 session_unlock_list();
1633 /* No notify thread, stop the UST tracing. */
1638 * Update newly registered application with the tracing
1639 * registry info already enabled information.
1641 update_ust_app(app
->sock
);
1644 * Don't care about return value. Let the manage apps threads
1645 * handle app unregistration upon socket close.
1647 (void) ust_app_register_done(app
->sock
);
1650 * Even if the application socket has been closed, send the app
1651 * to the thread and unregistration will take place at that
1654 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1657 session_unlock_list();
1658 /* No apps. thread, stop the UST tracing. */
1663 session_unlock_list();
1665 } while (node
!= NULL
);
1667 health_poll_entry();
1668 /* Futex wait on queue. Blocking call on futex() */
1669 futex_nto1_wait(&ust_cmd_queue
.futex
);
1672 /* Normal exit, no error */
1676 /* Clean up wait queue. */
1677 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1678 &wait_queue
.head
, head
) {
1679 cds_list_del(&wait_node
->head
);
1684 DBG("Dispatch thread dying");
1687 ERR("Health error occurred in %s", __func__
);
1689 health_unregister();
1694 * This thread manage application registration.
1696 static void *thread_registration_apps(void *data
)
1698 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1699 uint32_t revents
, nb_fd
;
1700 struct lttng_poll_event events
;
1702 * Get allocated in this thread, enqueued to a global queue, dequeued and
1703 * freed in the manage apps thread.
1705 struct ust_command
*ust_cmd
= NULL
;
1707 DBG("[thread] Manage application registration started");
1709 health_register(HEALTH_TYPE_APP_REG
);
1711 if (testpoint(thread_registration_apps
)) {
1712 goto error_testpoint
;
1715 ret
= lttcomm_listen_unix_sock(apps_sock
);
1721 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1722 * more will be added to this poll set.
1724 ret
= sessiond_set_thread_pollset(&events
, 2);
1726 goto error_create_poll
;
1729 /* Add the application registration socket */
1730 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1732 goto error_poll_add
;
1735 /* Notify all applications to register */
1736 ret
= notify_ust_apps(1);
1738 ERR("Failed to notify applications or create the wait shared memory.\n"
1739 "Execution continues but there might be problem for already\n"
1740 "running applications that wishes to register.");
1744 DBG("Accepting application registration");
1746 /* Inifinite blocking call, waiting for transmission */
1748 health_poll_entry();
1749 ret
= lttng_poll_wait(&events
, -1);
1753 * Restart interrupted system call.
1755 if (errno
== EINTR
) {
1763 for (i
= 0; i
< nb_fd
; i
++) {
1764 health_code_update();
1766 /* Fetch once the poll data */
1767 revents
= LTTNG_POLL_GETEV(&events
, i
);
1768 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1770 /* Thread quit pipe has been closed. Killing thread. */
1771 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1777 /* Event on the registration socket */
1778 if (pollfd
== apps_sock
) {
1779 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1780 ERR("Register apps socket poll error");
1782 } else if (revents
& LPOLLIN
) {
1783 sock
= lttcomm_accept_unix_sock(apps_sock
);
1789 * Set the CLOEXEC flag. Return code is useless because
1790 * either way, the show must go on.
1792 (void) utils_set_fd_cloexec(sock
);
1794 /* Create UST registration command for enqueuing */
1795 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1796 if (ust_cmd
== NULL
) {
1797 PERROR("ust command zmalloc");
1802 * Using message-based transmissions to ensure we don't
1803 * have to deal with partially received messages.
1805 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1807 ERR("Exhausted file descriptors allowed for applications.");
1817 health_code_update();
1818 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1821 /* Close socket of the application. */
1826 lttng_fd_put(LTTNG_FD_APPS
, 1);
1830 health_code_update();
1832 ust_cmd
->sock
= sock
;
1835 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1836 " gid:%d sock:%d name:%s (version %d.%d)",
1837 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1838 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1839 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1840 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1843 * Lock free enqueue the registration request. The red pill
1844 * has been taken! This apps will be part of the *system*.
1846 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1849 * Wake the registration queue futex. Implicit memory
1850 * barrier with the exchange in cds_wfq_enqueue.
1852 futex_nto1_wake(&ust_cmd_queue
.futex
);
1862 ERR("Health error occurred in %s", __func__
);
1865 /* Notify that the registration thread is gone */
1868 if (apps_sock
>= 0) {
1869 ret
= close(apps_sock
);
1879 lttng_fd_put(LTTNG_FD_APPS
, 1);
1881 unlink(apps_unix_sock_path
);
1884 lttng_poll_clean(&events
);
1888 DBG("UST Registration thread cleanup complete");
1889 health_unregister();
1895 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1896 * exec or it will fails.
1898 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1901 struct timespec timeout
;
1903 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1904 consumer_data
->consumer_thread_is_ready
= 0;
1906 /* Setup pthread condition */
1907 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1910 PERROR("pthread_condattr_init consumer data");
1915 * Set the monotonic clock in order to make sure we DO NOT jump in time
1916 * between the clock_gettime() call and the timedwait call. See bug #324
1917 * for a more details and how we noticed it.
1919 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1922 PERROR("pthread_condattr_setclock consumer data");
1926 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1929 PERROR("pthread_cond_init consumer data");
1933 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1936 PERROR("pthread_create consumer");
1941 /* We are about to wait on a pthread condition */
1942 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1944 /* Get time for sem_timedwait absolute timeout */
1945 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1947 * Set the timeout for the condition timed wait even if the clock gettime
1948 * call fails since we might loop on that call and we want to avoid to
1949 * increment the timeout too many times.
1951 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1954 * The following loop COULD be skipped in some conditions so this is why we
1955 * set ret to 0 in order to make sure at least one round of the loop is
1961 * Loop until the condition is reached or when a timeout is reached. Note
1962 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1963 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1964 * possible. This loop does not take any chances and works with both of
1967 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1968 if (clock_ret
< 0) {
1969 PERROR("clock_gettime spawn consumer");
1970 /* Infinite wait for the consumerd thread to be ready */
1971 ret
= pthread_cond_wait(&consumer_data
->cond
,
1972 &consumer_data
->cond_mutex
);
1974 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1975 &consumer_data
->cond_mutex
, &timeout
);
1979 /* Release the pthread condition */
1980 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1984 if (ret
== ETIMEDOUT
) {
1986 * Call has timed out so we kill the kconsumerd_thread and return
1989 ERR("Condition timed out. The consumer thread was never ready."
1991 ret
= pthread_cancel(consumer_data
->thread
);
1993 PERROR("pthread_cancel consumer thread");
1996 PERROR("pthread_cond_wait failed consumer thread");
2001 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2002 if (consumer_data
->pid
== 0) {
2003 ERR("Consumerd did not start");
2004 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2007 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2016 * Join consumer thread
2018 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2022 /* Consumer pid must be a real one. */
2023 if (consumer_data
->pid
> 0) {
2025 ret
= kill(consumer_data
->pid
, SIGTERM
);
2027 ERR("Error killing consumer daemon");
2030 return pthread_join(consumer_data
->thread
, &status
);
2037 * Fork and exec a consumer daemon (consumerd).
2039 * Return pid if successful else -1.
2041 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2045 const char *consumer_to_use
;
2046 const char *verbosity
;
2049 DBG("Spawning consumerd");
2056 if (opt_verbose_consumer
) {
2057 verbosity
= "--verbose";
2059 verbosity
= "--quiet";
2061 switch (consumer_data
->type
) {
2062 case LTTNG_CONSUMER_KERNEL
:
2064 * Find out which consumerd to execute. We will first try the
2065 * 64-bit path, then the sessiond's installation directory, and
2066 * fallback on the 32-bit one,
2068 DBG3("Looking for a kernel consumer at these locations:");
2069 DBG3(" 1) %s", consumerd64_bin
);
2070 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2071 DBG3(" 3) %s", consumerd32_bin
);
2072 if (stat(consumerd64_bin
, &st
) == 0) {
2073 DBG3("Found location #1");
2074 consumer_to_use
= consumerd64_bin
;
2075 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2076 DBG3("Found location #2");
2077 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2078 } else if (stat(consumerd32_bin
, &st
) == 0) {
2079 DBG3("Found location #3");
2080 consumer_to_use
= consumerd32_bin
;
2082 DBG("Could not find any valid consumerd executable");
2085 DBG("Using kernel consumer at: %s", consumer_to_use
);
2086 execl(consumer_to_use
,
2087 "lttng-consumerd", verbosity
, "-k",
2088 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2089 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2092 case LTTNG_CONSUMER64_UST
:
2094 char *tmpnew
= NULL
;
2096 if (consumerd64_libdir
[0] != '\0') {
2100 tmp
= getenv("LD_LIBRARY_PATH");
2104 tmplen
= strlen("LD_LIBRARY_PATH=")
2105 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2106 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2111 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2112 strcat(tmpnew
, consumerd64_libdir
);
2113 if (tmp
[0] != '\0') {
2114 strcat(tmpnew
, ":");
2115 strcat(tmpnew
, tmp
);
2117 ret
= putenv(tmpnew
);
2124 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2125 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2126 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2127 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2129 if (consumerd64_libdir
[0] != '\0') {
2137 case LTTNG_CONSUMER32_UST
:
2139 char *tmpnew
= NULL
;
2141 if (consumerd32_libdir
[0] != '\0') {
2145 tmp
= getenv("LD_LIBRARY_PATH");
2149 tmplen
= strlen("LD_LIBRARY_PATH=")
2150 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2151 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2156 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2157 strcat(tmpnew
, consumerd32_libdir
);
2158 if (tmp
[0] != '\0') {
2159 strcat(tmpnew
, ":");
2160 strcat(tmpnew
, tmp
);
2162 ret
= putenv(tmpnew
);
2169 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2170 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2171 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2172 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2174 if (consumerd32_libdir
[0] != '\0') {
2183 PERROR("unknown consumer type");
2187 PERROR("kernel start consumer exec");
2190 } else if (pid
> 0) {
2193 PERROR("start consumer fork");
2201 * Spawn the consumerd daemon and session daemon thread.
2203 static int start_consumerd(struct consumer_data
*consumer_data
)
2208 * Set the listen() state on the socket since there is a possible race
2209 * between the exec() of the consumer daemon and this call if place in the
2210 * consumer thread. See bug #366 for more details.
2212 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2217 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2218 if (consumer_data
->pid
!= 0) {
2219 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2223 ret
= spawn_consumerd(consumer_data
);
2225 ERR("Spawning consumerd failed");
2226 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2230 /* Setting up the consumer_data pid */
2231 consumer_data
->pid
= ret
;
2232 DBG2("Consumer pid %d", consumer_data
->pid
);
2233 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2235 DBG2("Spawning consumer control thread");
2236 ret
= spawn_consumer_thread(consumer_data
);
2238 ERR("Fatal error spawning consumer control thread");
2246 /* Cleanup already created sockets on error. */
2247 if (consumer_data
->err_sock
>= 0) {
2250 err
= close(consumer_data
->err_sock
);
2252 PERROR("close consumer data error socket");
2259 * Compute health status of each consumer. If one of them is zero (bad
2260 * state), we return 0.
2262 static int check_consumer_health(void)
2266 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2268 DBG3("Health consumer check %d", ret
);
2274 * Setup necessary data for kernel tracer action.
2276 static int init_kernel_tracer(void)
2280 /* Modprobe lttng kernel modules */
2281 ret
= modprobe_lttng_control();
2286 /* Open debugfs lttng */
2287 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2288 if (kernel_tracer_fd
< 0) {
2289 DBG("Failed to open %s", module_proc_lttng
);
2294 /* Validate kernel version */
2295 ret
= kernel_validate_version(kernel_tracer_fd
);
2300 ret
= modprobe_lttng_data();
2305 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2309 modprobe_remove_lttng_control();
2310 ret
= close(kernel_tracer_fd
);
2314 kernel_tracer_fd
= -1;
2315 return LTTNG_ERR_KERN_VERSION
;
2318 ret
= close(kernel_tracer_fd
);
2324 modprobe_remove_lttng_control();
2327 WARN("No kernel tracer available");
2328 kernel_tracer_fd
= -1;
2330 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2332 return LTTNG_ERR_KERN_NA
;
2338 * Copy consumer output from the tracing session to the domain session. The
2339 * function also applies the right modification on a per domain basis for the
2340 * trace files destination directory.
2342 * Should *NOT* be called with RCU read-side lock held.
2344 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2347 const char *dir_name
;
2348 struct consumer_output
*consumer
;
2351 assert(session
->consumer
);
2354 case LTTNG_DOMAIN_KERNEL
:
2355 DBG3("Copying tracing session consumer output in kernel session");
2357 * XXX: We should audit the session creation and what this function
2358 * does "extra" in order to avoid a destroy since this function is used
2359 * in the domain session creation (kernel and ust) only. Same for UST
2362 if (session
->kernel_session
->consumer
) {
2363 consumer_destroy_output(session
->kernel_session
->consumer
);
2365 session
->kernel_session
->consumer
=
2366 consumer_copy_output(session
->consumer
);
2367 /* Ease our life a bit for the next part */
2368 consumer
= session
->kernel_session
->consumer
;
2369 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2371 case LTTNG_DOMAIN_UST
:
2372 DBG3("Copying tracing session consumer output in UST session");
2373 if (session
->ust_session
->consumer
) {
2374 consumer_destroy_output(session
->ust_session
->consumer
);
2376 session
->ust_session
->consumer
=
2377 consumer_copy_output(session
->consumer
);
2378 /* Ease our life a bit for the next part */
2379 consumer
= session
->ust_session
->consumer
;
2380 dir_name
= DEFAULT_UST_TRACE_DIR
;
2383 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2387 /* Append correct directory to subdir */
2388 strncat(consumer
->subdir
, dir_name
,
2389 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2390 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2399 * Create an UST session and add it to the session ust list.
2401 * Should *NOT* be called with RCU read-side lock held.
2403 static int create_ust_session(struct ltt_session
*session
,
2404 struct lttng_domain
*domain
)
2407 struct ltt_ust_session
*lus
= NULL
;
2411 assert(session
->consumer
);
2413 switch (domain
->type
) {
2414 case LTTNG_DOMAIN_UST
:
2417 ERR("Unknown UST domain on create session %d", domain
->type
);
2418 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2422 DBG("Creating UST session");
2424 lus
= trace_ust_create_session(session
->id
);
2426 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2430 lus
->uid
= session
->uid
;
2431 lus
->gid
= session
->gid
;
2432 lus
->output_traces
= session
->output_traces
;
2433 lus
->snapshot_mode
= session
->snapshot_mode
;
2434 session
->ust_session
= lus
;
2436 /* Copy session output to the newly created UST session */
2437 ret
= copy_session_consumer(domain
->type
, session
);
2438 if (ret
!= LTTNG_OK
) {
2446 session
->ust_session
= NULL
;
2451 * Create a kernel tracer session then create the default channel.
2453 static int create_kernel_session(struct ltt_session
*session
)
2457 DBG("Creating kernel session");
2459 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2461 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2465 /* Code flow safety */
2466 assert(session
->kernel_session
);
2468 /* Copy session output to the newly created Kernel session */
2469 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2470 if (ret
!= LTTNG_OK
) {
2474 /* Create directory(ies) on local filesystem. */
2475 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2476 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2477 ret
= run_as_mkdir_recursive(
2478 session
->kernel_session
->consumer
->dst
.trace_path
,
2479 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2481 if (ret
!= -EEXIST
) {
2482 ERR("Trace directory creation error");
2488 session
->kernel_session
->uid
= session
->uid
;
2489 session
->kernel_session
->gid
= session
->gid
;
2490 session
->kernel_session
->output_traces
= session
->output_traces
;
2491 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2496 trace_kernel_destroy_session(session
->kernel_session
);
2497 session
->kernel_session
= NULL
;
2502 * Count number of session permitted by uid/gid.
2504 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2507 struct ltt_session
*session
;
2509 DBG("Counting number of available session for UID %d GID %d",
2511 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2513 * Only list the sessions the user can control.
2515 if (!session_access_ok(session
, uid
, gid
)) {
2524 * Process the command requested by the lttng client within the command
2525 * context structure. This function make sure that the return structure (llm)
2526 * is set and ready for transmission before returning.
2528 * Return any error encountered or 0 for success.
2530 * "sock" is only used for special-case var. len data.
2532 * Should *NOT* be called with RCU read-side lock held.
2534 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2538 int need_tracing_session
= 1;
2541 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2545 switch (cmd_ctx
->lsm
->cmd_type
) {
2546 case LTTNG_CREATE_SESSION
:
2547 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2548 case LTTNG_DESTROY_SESSION
:
2549 case LTTNG_LIST_SESSIONS
:
2550 case LTTNG_LIST_DOMAINS
:
2551 case LTTNG_START_TRACE
:
2552 case LTTNG_STOP_TRACE
:
2553 case LTTNG_DATA_PENDING
:
2554 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2555 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2556 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2557 case LTTNG_SNAPSHOT_RECORD
:
2564 if (opt_no_kernel
&& need_domain
2565 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2567 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2569 ret
= LTTNG_ERR_KERN_NA
;
2574 /* Deny register consumer if we already have a spawned consumer. */
2575 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2576 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2577 if (kconsumer_data
.pid
> 0) {
2578 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2579 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2582 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2586 * Check for command that don't needs to allocate a returned payload. We do
2587 * this here so we don't have to make the call for no payload at each
2590 switch(cmd_ctx
->lsm
->cmd_type
) {
2591 case LTTNG_LIST_SESSIONS
:
2592 case LTTNG_LIST_TRACEPOINTS
:
2593 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2594 case LTTNG_LIST_DOMAINS
:
2595 case LTTNG_LIST_CHANNELS
:
2596 case LTTNG_LIST_EVENTS
:
2599 /* Setup lttng message with no payload */
2600 ret
= setup_lttng_msg(cmd_ctx
, 0);
2602 /* This label does not try to unlock the session */
2603 goto init_setup_error
;
2607 /* Commands that DO NOT need a session. */
2608 switch (cmd_ctx
->lsm
->cmd_type
) {
2609 case LTTNG_CREATE_SESSION
:
2610 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2611 case LTTNG_CALIBRATE
:
2612 case LTTNG_LIST_SESSIONS
:
2613 case LTTNG_LIST_TRACEPOINTS
:
2614 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2615 need_tracing_session
= 0;
2618 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2620 * We keep the session list lock across _all_ commands
2621 * for now, because the per-session lock does not
2622 * handle teardown properly.
2624 session_lock_list();
2625 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2626 if (cmd_ctx
->session
== NULL
) {
2627 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2630 /* Acquire lock for the session */
2631 session_lock(cmd_ctx
->session
);
2641 * Check domain type for specific "pre-action".
2643 switch (cmd_ctx
->lsm
->domain
.type
) {
2644 case LTTNG_DOMAIN_KERNEL
:
2646 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2650 /* Kernel tracer check */
2651 if (kernel_tracer_fd
== -1) {
2652 /* Basically, load kernel tracer modules */
2653 ret
= init_kernel_tracer();
2659 /* Consumer is in an ERROR state. Report back to client */
2660 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2661 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2665 /* Need a session for kernel command */
2666 if (need_tracing_session
) {
2667 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2668 ret
= create_kernel_session(cmd_ctx
->session
);
2670 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2675 /* Start the kernel consumer daemon */
2676 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2677 if (kconsumer_data
.pid
== 0 &&
2678 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2679 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2680 ret
= start_consumerd(&kconsumer_data
);
2682 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2685 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2687 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2691 * The consumer was just spawned so we need to add the socket to
2692 * the consumer output of the session if exist.
2694 ret
= consumer_create_socket(&kconsumer_data
,
2695 cmd_ctx
->session
->kernel_session
->consumer
);
2702 case LTTNG_DOMAIN_UST
:
2704 /* Consumer is in an ERROR state. Report back to client */
2705 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2706 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2710 if (need_tracing_session
) {
2711 /* Create UST session if none exist. */
2712 if (cmd_ctx
->session
->ust_session
== NULL
) {
2713 ret
= create_ust_session(cmd_ctx
->session
,
2714 &cmd_ctx
->lsm
->domain
);
2715 if (ret
!= LTTNG_OK
) {
2720 /* Start the UST consumer daemons */
2722 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2723 if (consumerd64_bin
[0] != '\0' &&
2724 ustconsumer64_data
.pid
== 0 &&
2725 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2726 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2727 ret
= start_consumerd(&ustconsumer64_data
);
2729 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2730 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2734 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2735 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2737 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2741 * Setup socket for consumer 64 bit. No need for atomic access
2742 * since it was set above and can ONLY be set in this thread.
2744 ret
= consumer_create_socket(&ustconsumer64_data
,
2745 cmd_ctx
->session
->ust_session
->consumer
);
2751 if (consumerd32_bin
[0] != '\0' &&
2752 ustconsumer32_data
.pid
== 0 &&
2753 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2754 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2755 ret
= start_consumerd(&ustconsumer32_data
);
2757 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2758 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2762 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2763 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2765 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2769 * Setup socket for consumer 64 bit. No need for atomic access
2770 * since it was set above and can ONLY be set in this thread.
2772 ret
= consumer_create_socket(&ustconsumer32_data
,
2773 cmd_ctx
->session
->ust_session
->consumer
);
2785 /* Validate consumer daemon state when start/stop trace command */
2786 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2787 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2788 switch (cmd_ctx
->lsm
->domain
.type
) {
2789 case LTTNG_DOMAIN_UST
:
2790 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2791 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2795 case LTTNG_DOMAIN_KERNEL
:
2796 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2797 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2805 * Check that the UID or GID match that of the tracing session.
2806 * The root user can interact with all sessions.
2808 if (need_tracing_session
) {
2809 if (!session_access_ok(cmd_ctx
->session
,
2810 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2811 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2812 ret
= LTTNG_ERR_EPERM
;
2818 * Send relayd information to consumer as soon as we have a domain and a
2821 if (cmd_ctx
->session
&& need_domain
) {
2823 * Setup relayd if not done yet. If the relayd information was already
2824 * sent to the consumer, this call will gracefully return.
2826 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2827 if (ret
!= LTTNG_OK
) {
2832 /* Process by command type */
2833 switch (cmd_ctx
->lsm
->cmd_type
) {
2834 case LTTNG_ADD_CONTEXT
:
2836 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2837 cmd_ctx
->lsm
->u
.context
.channel_name
,
2838 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2841 case LTTNG_DISABLE_CHANNEL
:
2843 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2844 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2847 case LTTNG_DISABLE_EVENT
:
2849 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2850 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2851 cmd_ctx
->lsm
->u
.disable
.name
);
2854 case LTTNG_DISABLE_ALL_EVENT
:
2856 DBG("Disabling all events");
2858 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2859 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2862 case LTTNG_ENABLE_CHANNEL
:
2864 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2865 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2868 case LTTNG_ENABLE_EVENT
:
2870 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2871 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2872 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2875 case LTTNG_ENABLE_ALL_EVENT
:
2877 DBG("Enabling all events");
2879 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2880 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2881 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2884 case LTTNG_LIST_TRACEPOINTS
:
2886 struct lttng_event
*events
;
2889 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2890 if (nb_events
< 0) {
2891 /* Return value is a negative lttng_error_code. */
2897 * Setup lttng message with payload size set to the event list size in
2898 * bytes and then copy list into the llm payload.
2900 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2906 /* Copy event list into message payload */
2907 memcpy(cmd_ctx
->llm
->payload
, events
,
2908 sizeof(struct lttng_event
) * nb_events
);
2915 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2917 struct lttng_event_field
*fields
;
2920 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2922 if (nb_fields
< 0) {
2923 /* Return value is a negative lttng_error_code. */
2929 * Setup lttng message with payload size set to the event list size in
2930 * bytes and then copy list into the llm payload.
2932 ret
= setup_lttng_msg(cmd_ctx
,
2933 sizeof(struct lttng_event_field
) * nb_fields
);
2939 /* Copy event list into message payload */
2940 memcpy(cmd_ctx
->llm
->payload
, fields
,
2941 sizeof(struct lttng_event_field
) * nb_fields
);
2948 case LTTNG_SET_CONSUMER_URI
:
2951 struct lttng_uri
*uris
;
2953 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2954 len
= nb_uri
* sizeof(struct lttng_uri
);
2957 ret
= LTTNG_ERR_INVALID
;
2961 uris
= zmalloc(len
);
2963 ret
= LTTNG_ERR_FATAL
;
2967 /* Receive variable len data */
2968 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2969 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2971 DBG("No URIs received from client... continuing");
2973 ret
= LTTNG_ERR_SESSION_FAIL
;
2978 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2980 if (ret
!= LTTNG_OK
) {
2986 * XXX: 0 means that this URI should be applied on the session. Should
2987 * be a DOMAIN enuam.
2989 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2990 /* Add the URI for the UST session if a consumer is present. */
2991 if (cmd_ctx
->session
->ust_session
&&
2992 cmd_ctx
->session
->ust_session
->consumer
) {
2993 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2995 } else if (cmd_ctx
->session
->kernel_session
&&
2996 cmd_ctx
->session
->kernel_session
->consumer
) {
2997 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2998 cmd_ctx
->session
, nb_uri
, uris
);
3006 case LTTNG_START_TRACE
:
3008 ret
= cmd_start_trace(cmd_ctx
->session
);
3011 case LTTNG_STOP_TRACE
:
3013 ret
= cmd_stop_trace(cmd_ctx
->session
);
3016 case LTTNG_CREATE_SESSION
:
3019 struct lttng_uri
*uris
= NULL
;
3021 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3022 len
= nb_uri
* sizeof(struct lttng_uri
);
3025 uris
= zmalloc(len
);
3027 ret
= LTTNG_ERR_FATAL
;
3031 /* Receive variable len data */
3032 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3033 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3035 DBG("No URIs received from client... continuing");
3037 ret
= LTTNG_ERR_SESSION_FAIL
;
3042 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3043 DBG("Creating session with ONE network URI is a bad call");
3044 ret
= LTTNG_ERR_SESSION_FAIL
;
3050 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3057 case LTTNG_DESTROY_SESSION
:
3059 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3061 /* Set session to NULL so we do not unlock it after free. */
3062 cmd_ctx
->session
= NULL
;
3065 case LTTNG_LIST_DOMAINS
:
3068 struct lttng_domain
*domains
;
3070 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3072 /* Return value is a negative lttng_error_code. */
3077 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3083 /* Copy event list into message payload */
3084 memcpy(cmd_ctx
->llm
->payload
, domains
,
3085 nb_dom
* sizeof(struct lttng_domain
));
3092 case LTTNG_LIST_CHANNELS
:
3095 struct lttng_channel
*channels
;
3097 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3098 cmd_ctx
->session
, &channels
);
3100 /* Return value is a negative lttng_error_code. */
3105 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3111 /* Copy event list into message payload */
3112 memcpy(cmd_ctx
->llm
->payload
, channels
,
3113 nb_chan
* sizeof(struct lttng_channel
));
3120 case LTTNG_LIST_EVENTS
:
3123 struct lttng_event
*events
= NULL
;
3125 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3126 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3128 /* Return value is a negative lttng_error_code. */
3133 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3139 /* Copy event list into message payload */
3140 memcpy(cmd_ctx
->llm
->payload
, events
,
3141 nb_event
* sizeof(struct lttng_event
));
3148 case LTTNG_LIST_SESSIONS
:
3150 unsigned int nr_sessions
;
3152 session_lock_list();
3153 nr_sessions
= lttng_sessions_count(
3154 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3155 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3157 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3159 session_unlock_list();
3163 /* Filled the session array */
3164 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3165 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3166 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3168 session_unlock_list();
3173 case LTTNG_CALIBRATE
:
3175 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3176 &cmd_ctx
->lsm
->u
.calibrate
);
3179 case LTTNG_REGISTER_CONSUMER
:
3181 struct consumer_data
*cdata
;
3183 switch (cmd_ctx
->lsm
->domain
.type
) {
3184 case LTTNG_DOMAIN_KERNEL
:
3185 cdata
= &kconsumer_data
;
3188 ret
= LTTNG_ERR_UND
;
3192 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3193 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3196 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3198 struct lttng_filter_bytecode
*bytecode
;
3200 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3201 ret
= LTTNG_ERR_FILTER_INVAL
;
3204 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3205 ret
= LTTNG_ERR_FILTER_INVAL
;
3208 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3210 ret
= LTTNG_ERR_FILTER_NOMEM
;
3213 /* Receive var. len. data */
3214 DBG("Receiving var len data from client ...");
3215 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3216 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3218 DBG("Nothing recv() from client var len data... continuing");
3220 ret
= LTTNG_ERR_FILTER_INVAL
;
3224 if (bytecode
->len
+ sizeof(*bytecode
)
3225 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3227 ret
= LTTNG_ERR_FILTER_INVAL
;
3231 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3232 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3233 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3236 case LTTNG_DATA_PENDING
:
3238 ret
= cmd_data_pending(cmd_ctx
->session
);
3241 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3243 struct lttcomm_lttng_output_id reply
;
3245 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3246 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3247 if (ret
!= LTTNG_OK
) {
3251 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3256 /* Copy output list into message payload */
3257 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3261 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3263 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3264 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3267 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3270 struct lttng_snapshot_output
*outputs
= NULL
;
3272 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3273 if (nb_output
< 0) {
3278 ret
= setup_lttng_msg(cmd_ctx
,
3279 nb_output
* sizeof(struct lttng_snapshot_output
));
3286 /* Copy output list into message payload */
3287 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3288 nb_output
* sizeof(struct lttng_snapshot_output
));
3295 case LTTNG_SNAPSHOT_RECORD
:
3297 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3298 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3299 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3302 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3305 struct lttng_uri
*uris
= NULL
;
3307 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3308 len
= nb_uri
* sizeof(struct lttng_uri
);
3311 uris
= zmalloc(len
);
3313 ret
= LTTNG_ERR_FATAL
;
3317 /* Receive variable len data */
3318 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3319 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3321 DBG("No URIs received from client... continuing");
3323 ret
= LTTNG_ERR_SESSION_FAIL
;
3328 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3329 DBG("Creating session with ONE network URI is a bad call");
3330 ret
= LTTNG_ERR_SESSION_FAIL
;
3336 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3337 nb_uri
, &cmd_ctx
->creds
);
3342 ret
= LTTNG_ERR_UND
;
3347 if (cmd_ctx
->llm
== NULL
) {
3348 DBG("Missing llm structure. Allocating one.");
3349 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3353 /* Set return code */
3354 cmd_ctx
->llm
->ret_code
= ret
;
3356 if (cmd_ctx
->session
) {
3357 session_unlock(cmd_ctx
->session
);
3359 if (need_tracing_session
) {
3360 session_unlock_list();
3367 * Thread managing health check socket.
3369 static void *thread_manage_health(void *data
)
3371 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3372 uint32_t revents
, nb_fd
;
3373 struct lttng_poll_event events
;
3374 struct lttcomm_health_msg msg
;
3375 struct lttcomm_health_data reply
;
3377 DBG("[thread] Manage health check started");
3379 rcu_register_thread();
3381 /* We might hit an error path before this is created. */
3382 lttng_poll_init(&events
);
3384 /* Create unix socket */
3385 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3387 ERR("Unable to create health check Unix socket");
3393 * Set the CLOEXEC flag. Return code is useless because either way, the
3396 (void) utils_set_fd_cloexec(sock
);
3398 ret
= lttcomm_listen_unix_sock(sock
);
3404 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3405 * more will be added to this poll set.
3407 ret
= sessiond_set_thread_pollset(&events
, 2);
3412 /* Add the application registration socket */
3413 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3419 DBG("Health check ready");
3421 /* Inifinite blocking call, waiting for transmission */
3423 ret
= lttng_poll_wait(&events
, -1);
3426 * Restart interrupted system call.
3428 if (errno
== EINTR
) {
3436 for (i
= 0; i
< nb_fd
; i
++) {
3437 /* Fetch once the poll data */
3438 revents
= LTTNG_POLL_GETEV(&events
, i
);
3439 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3441 /* Thread quit pipe has been closed. Killing thread. */
3442 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3448 /* Event on the registration socket */
3449 if (pollfd
== sock
) {
3450 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3451 ERR("Health socket poll error");
3457 new_sock
= lttcomm_accept_unix_sock(sock
);
3463 * Set the CLOEXEC flag. Return code is useless because either way, the
3466 (void) utils_set_fd_cloexec(new_sock
);
3468 DBG("Receiving data from client for health...");
3469 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3471 DBG("Nothing recv() from client... continuing");
3472 ret
= close(new_sock
);
3480 rcu_thread_online();
3482 switch (msg
.component
) {
3483 case LTTNG_HEALTH_CMD
:
3484 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3486 case LTTNG_HEALTH_APP_MANAGE
:
3487 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3489 case LTTNG_HEALTH_APP_REG
:
3490 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3492 case LTTNG_HEALTH_KERNEL
:
3493 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3495 case LTTNG_HEALTH_CONSUMER
:
3496 reply
.ret_code
= check_consumer_health();
3498 case LTTNG_HEALTH_HT_CLEANUP
:
3499 reply
.ret_code
= health_check_state(HEALTH_TYPE_HT_CLEANUP
);
3501 case LTTNG_HEALTH_APP_MANAGE_NOTIFY
:
3502 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3504 case LTTNG_HEALTH_APP_REG_DISPATCH
:
3505 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3507 case LTTNG_HEALTH_ALL
:
3509 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3510 health_check_state(HEALTH_TYPE_APP_REG
) &&
3511 health_check_state(HEALTH_TYPE_CMD
) &&
3512 health_check_state(HEALTH_TYPE_KERNEL
) &&
3513 check_consumer_health() &&
3514 health_check_state(HEALTH_TYPE_HT_CLEANUP
) &&
3515 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
) &&
3516 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3519 reply
.ret_code
= LTTNG_ERR_UND
;
3524 * Flip ret value since 0 is a success and 1 indicates a bad health for
3525 * the client where in the sessiond it is the opposite. Again, this is
3526 * just to make things easier for us poor developer which enjoy a lot
3529 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3530 reply
.ret_code
= !reply
.ret_code
;
3533 DBG2("Health check return value %d", reply
.ret_code
);
3535 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3537 ERR("Failed to send health data back to client");
3540 /* End of transmission */
3541 ret
= close(new_sock
);
3551 ERR("Health error occurred in %s", __func__
);
3553 DBG("Health check thread dying");
3554 unlink(health_unix_sock_path
);
3562 lttng_poll_clean(&events
);
3564 rcu_unregister_thread();
3569 * This thread manage all clients request using the unix client socket for
3572 static void *thread_manage_clients(void *data
)
3574 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3576 uint32_t revents
, nb_fd
;
3577 struct command_ctx
*cmd_ctx
= NULL
;
3578 struct lttng_poll_event events
;
3580 DBG("[thread] Manage client started");
3582 rcu_register_thread();
3584 health_register(HEALTH_TYPE_CMD
);
3586 if (testpoint(thread_manage_clients
)) {
3587 goto error_testpoint
;
3590 health_code_update();
3592 ret
= lttcomm_listen_unix_sock(client_sock
);
3598 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3599 * more will be added to this poll set.
3601 ret
= sessiond_set_thread_pollset(&events
, 2);
3603 goto error_create_poll
;
3606 /* Add the application registration socket */
3607 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3613 * Notify parent pid that we are ready to accept command for client side.
3615 if (opt_sig_parent
) {
3616 kill(ppid
, SIGUSR1
);
3619 if (testpoint(thread_manage_clients_before_loop
)) {
3623 health_code_update();
3626 DBG("Accepting client command ...");
3628 /* Inifinite blocking call, waiting for transmission */
3630 health_poll_entry();
3631 ret
= lttng_poll_wait(&events
, -1);
3635 * Restart interrupted system call.
3637 if (errno
== EINTR
) {
3645 for (i
= 0; i
< nb_fd
; i
++) {
3646 /* Fetch once the poll data */
3647 revents
= LTTNG_POLL_GETEV(&events
, i
);
3648 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3650 health_code_update();
3652 /* Thread quit pipe has been closed. Killing thread. */
3653 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3659 /* Event on the registration socket */
3660 if (pollfd
== client_sock
) {
3661 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3662 ERR("Client socket poll error");
3668 DBG("Wait for client response");
3670 health_code_update();
3672 sock
= lttcomm_accept_unix_sock(client_sock
);
3678 * Set the CLOEXEC flag. Return code is useless because either way, the
3681 (void) utils_set_fd_cloexec(sock
);
3683 /* Set socket option for credentials retrieval */
3684 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3689 /* Allocate context command to process the client request */
3690 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3691 if (cmd_ctx
== NULL
) {
3692 PERROR("zmalloc cmd_ctx");
3696 /* Allocate data buffer for reception */
3697 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3698 if (cmd_ctx
->lsm
== NULL
) {
3699 PERROR("zmalloc cmd_ctx->lsm");
3703 cmd_ctx
->llm
= NULL
;
3704 cmd_ctx
->session
= NULL
;
3706 health_code_update();
3709 * Data is received from the lttng client. The struct
3710 * lttcomm_session_msg (lsm) contains the command and data request of
3713 DBG("Receiving data from client ...");
3714 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3715 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3717 DBG("Nothing recv() from client... continuing");
3723 clean_command_ctx(&cmd_ctx
);
3727 health_code_update();
3729 // TODO: Validate cmd_ctx including sanity check for
3730 // security purpose.
3732 rcu_thread_online();
3734 * This function dispatch the work to the kernel or userspace tracer
3735 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3736 * informations for the client. The command context struct contains
3737 * everything this function may needs.
3739 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3740 rcu_thread_offline();
3748 * TODO: Inform client somehow of the fatal error. At
3749 * this point, ret < 0 means that a zmalloc failed
3750 * (ENOMEM). Error detected but still accept
3751 * command, unless a socket error has been
3754 clean_command_ctx(&cmd_ctx
);
3758 health_code_update();
3760 DBG("Sending response (size: %d, retcode: %s)",
3761 cmd_ctx
->lttng_msg_size
,
3762 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3763 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3765 ERR("Failed to send data back to client");
3768 /* End of transmission */
3775 clean_command_ctx(&cmd_ctx
);
3777 health_code_update();
3789 lttng_poll_clean(&events
);
3790 clean_command_ctx(&cmd_ctx
);
3795 unlink(client_unix_sock_path
);
3796 if (client_sock
>= 0) {
3797 ret
= close(client_sock
);
3805 ERR("Health error occurred in %s", __func__
);
3808 health_unregister();
3810 DBG("Client thread dying");
3812 rcu_unregister_thread();
3818 * usage function on stderr
3820 static void usage(void)
3822 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3823 fprintf(stderr
, " -h, --help Display this usage.\n");
3824 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3825 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3826 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3827 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3828 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3829 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3830 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3831 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3832 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3833 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3834 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3835 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3836 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3837 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3838 fprintf(stderr
, " -V, --version Show version number.\n");
3839 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3840 fprintf(stderr
, " -q, --quiet No output at all.\n");
3841 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3842 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3843 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3844 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3848 * daemon argument parsing
3850 static int parse_args(int argc
, char **argv
)
3854 static struct option long_options
[] = {
3855 { "client-sock", 1, 0, 'c' },
3856 { "apps-sock", 1, 0, 'a' },
3857 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3858 { "kconsumerd-err-sock", 1, 0, 'E' },
3859 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3860 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3861 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3862 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3863 { "consumerd32-path", 1, 0, 'u' },
3864 { "consumerd32-libdir", 1, 0, 'U' },
3865 { "consumerd64-path", 1, 0, 't' },
3866 { "consumerd64-libdir", 1, 0, 'T' },
3867 { "daemonize", 0, 0, 'd' },
3868 { "sig-parent", 0, 0, 'S' },
3869 { "help", 0, 0, 'h' },
3870 { "group", 1, 0, 'g' },
3871 { "version", 0, 0, 'V' },
3872 { "quiet", 0, 0, 'q' },
3873 { "verbose", 0, 0, 'v' },
3874 { "verbose-consumer", 0, 0, 'Z' },
3875 { "no-kernel", 0, 0, 'N' },
3876 { "pidfile", 1, 0, 'p' },
3881 int option_index
= 0;
3882 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3883 long_options
, &option_index
);
3890 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3892 fprintf(stderr
, " with arg %s\n", optarg
);
3896 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3899 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3905 opt_tracing_group
= optarg
;
3911 fprintf(stdout
, "%s\n", VERSION
);
3917 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3920 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3923 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3926 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3929 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3932 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3938 lttng_opt_quiet
= 1;
3941 /* Verbose level can increase using multiple -v */
3942 lttng_opt_verbose
+= 1;
3945 opt_verbose_consumer
+= 1;
3948 consumerd32_bin
= optarg
;
3951 consumerd32_libdir
= optarg
;
3954 consumerd64_bin
= optarg
;
3957 consumerd64_libdir
= optarg
;
3960 opt_pidfile
= optarg
;
3963 /* Unknown option or other error.
3964 * Error is printed by getopt, just return */
3973 * Creates the two needed socket by the daemon.
3974 * apps_sock - The communication socket for all UST apps.
3975 * client_sock - The communication of the cli tool (lttng).
3977 static int init_daemon_socket(void)
3982 old_umask
= umask(0);
3984 /* Create client tool unix socket */
3985 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3986 if (client_sock
< 0) {
3987 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3992 /* Set the cloexec flag */
3993 ret
= utils_set_fd_cloexec(client_sock
);
3995 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3996 "Continuing but note that the consumer daemon will have a "
3997 "reference to this socket on exec()", client_sock
);
4000 /* File permission MUST be 660 */
4001 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4003 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4008 /* Create the application unix socket */
4009 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4010 if (apps_sock
< 0) {
4011 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4016 /* Set the cloexec flag */
4017 ret
= utils_set_fd_cloexec(apps_sock
);
4019 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4020 "Continuing but note that the consumer daemon will have a "
4021 "reference to this socket on exec()", apps_sock
);
4024 /* File permission MUST be 666 */
4025 ret
= chmod(apps_unix_sock_path
,
4026 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4028 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4033 DBG3("Session daemon client socket %d and application socket %d created",
4034 client_sock
, apps_sock
);
4042 * Check if the global socket is available, and if a daemon is answering at the
4043 * other side. If yes, error is returned.
4045 static int check_existing_daemon(void)
4047 /* Is there anybody out there ? */
4048 if (lttng_session_daemon_alive()) {
4056 * Set the tracing group gid onto the client socket.
4058 * Race window between mkdir and chown is OK because we are going from more
4059 * permissive (root.root) to less permissive (root.tracing).
4061 static int set_permissions(char *rundir
)
4066 ret
= allowed_group();
4068 WARN("No tracing group detected");
4075 /* Set lttng run dir */
4076 ret
= chown(rundir
, 0, gid
);
4078 ERR("Unable to set group on %s", rundir
);
4082 /* Ensure tracing group can search the run dir */
4083 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
4085 ERR("Unable to set permissions on %s", rundir
);
4089 /* lttng client socket path */
4090 ret
= chown(client_unix_sock_path
, 0, gid
);
4092 ERR("Unable to set group on %s", client_unix_sock_path
);
4096 /* kconsumer error socket path */
4097 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
4099 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4103 /* 64-bit ustconsumer error socket path */
4104 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
4106 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4110 /* 32-bit ustconsumer compat32 error socket path */
4111 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
4113 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4117 DBG("All permissions are set");
4124 * Create the lttng run directory needed for all global sockets and pipe.
4126 static int create_lttng_rundir(const char *rundir
)
4130 DBG3("Creating LTTng run directory: %s", rundir
);
4132 ret
= mkdir(rundir
, S_IRWXU
);
4134 if (errno
!= EEXIST
) {
4135 ERR("Unable to create %s", rundir
);
4147 * Setup sockets and directory needed by the kconsumerd communication with the
4150 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4154 char path
[PATH_MAX
];
4156 switch (consumer_data
->type
) {
4157 case LTTNG_CONSUMER_KERNEL
:
4158 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4160 case LTTNG_CONSUMER64_UST
:
4161 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4163 case LTTNG_CONSUMER32_UST
:
4164 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4167 ERR("Consumer type unknown");
4172 DBG2("Creating consumer directory: %s", path
);
4174 ret
= mkdir(path
, S_IRWXU
);
4176 if (errno
!= EEXIST
) {
4178 ERR("Failed to create %s", path
);
4184 /* Create the kconsumerd error unix socket */
4185 consumer_data
->err_sock
=
4186 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4187 if (consumer_data
->err_sock
< 0) {
4188 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4194 * Set the CLOEXEC flag. Return code is useless because either way, the
4197 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4199 PERROR("utils_set_fd_cloexec");
4200 /* continue anyway */
4203 /* File permission MUST be 660 */
4204 ret
= chmod(consumer_data
->err_unix_sock_path
,
4205 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4207 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4217 * Signal handler for the daemon
4219 * Simply stop all worker threads, leaving main() return gracefully after
4220 * joining all threads and calling cleanup().
4222 static void sighandler(int sig
)
4226 DBG("SIGPIPE caught");
4229 DBG("SIGINT caught");
4233 DBG("SIGTERM caught");
4242 * Setup signal handler for :
4243 * SIGINT, SIGTERM, SIGPIPE
4245 static int set_signal_handler(void)
4248 struct sigaction sa
;
4251 if ((ret
= sigemptyset(&sigset
)) < 0) {
4252 PERROR("sigemptyset");
4256 sa
.sa_handler
= sighandler
;
4257 sa
.sa_mask
= sigset
;
4259 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4260 PERROR("sigaction");
4264 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4265 PERROR("sigaction");
4269 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4270 PERROR("sigaction");
4274 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4280 * Set open files limit to unlimited. This daemon can open a large number of
4281 * file descriptors in order to consumer multiple kernel traces.
4283 static void set_ulimit(void)
4288 /* The kernel does not allowed an infinite limit for open files */
4289 lim
.rlim_cur
= 65535;
4290 lim
.rlim_max
= 65535;
4292 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4294 PERROR("failed to set open files limit");
4299 * Write pidfile using the rundir and opt_pidfile.
4301 static void write_pidfile(void)
4304 char pidfile_path
[PATH_MAX
];
4309 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4311 /* Build pidfile path from rundir and opt_pidfile. */
4312 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4313 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4315 PERROR("snprintf pidfile path");
4321 * Create pid file in rundir. Return value is of no importance. The
4322 * execution will continue even though we are not able to write the file.
4324 (void) utils_create_pid_file(getpid(), pidfile_path
);
4333 int main(int argc
, char **argv
)
4337 const char *home_path
, *env_app_timeout
;
4339 init_kernel_workarounds();
4341 rcu_register_thread();
4343 setup_consumerd_path();
4345 page_size
= sysconf(_SC_PAGESIZE
);
4346 if (page_size
< 0) {
4347 PERROR("sysconf _SC_PAGESIZE");
4348 page_size
= LONG_MAX
;
4349 WARN("Fallback page size to %ld", page_size
);
4352 /* Parse arguments */
4354 if ((ret
= parse_args(argc
, argv
)) < 0) {
4364 * child: setsid, close FD 0, 1, 2, chdir /
4365 * parent: exit (if fork is successful)
4373 * We are in the child. Make sure all other file
4374 * descriptors are closed, in case we are called with
4375 * more opened file descriptors than the standard ones.
4377 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4382 /* Create thread quit pipe */
4383 if ((ret
= init_thread_quit_pipe()) < 0) {
4387 /* Check if daemon is UID = 0 */
4388 is_root
= !getuid();
4391 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4393 /* Create global run dir with root access */
4394 ret
= create_lttng_rundir(rundir
);
4399 if (strlen(apps_unix_sock_path
) == 0) {
4400 snprintf(apps_unix_sock_path
, PATH_MAX
,
4401 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4404 if (strlen(client_unix_sock_path
) == 0) {
4405 snprintf(client_unix_sock_path
, PATH_MAX
,
4406 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4409 /* Set global SHM for ust */
4410 if (strlen(wait_shm_path
) == 0) {
4411 snprintf(wait_shm_path
, PATH_MAX
,
4412 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4415 if (strlen(health_unix_sock_path
) == 0) {
4416 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4417 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4420 /* Setup kernel consumerd path */
4421 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4422 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4423 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4424 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4426 DBG2("Kernel consumer err path: %s",
4427 kconsumer_data
.err_unix_sock_path
);
4428 DBG2("Kernel consumer cmd path: %s",
4429 kconsumer_data
.cmd_unix_sock_path
);
4431 home_path
= utils_get_home_dir();
4432 if (home_path
== NULL
) {
4433 /* TODO: Add --socket PATH option */
4434 ERR("Can't get HOME directory for sockets creation.");
4440 * Create rundir from home path. This will create something like
4443 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4449 ret
= create_lttng_rundir(rundir
);
4454 if (strlen(apps_unix_sock_path
) == 0) {
4455 snprintf(apps_unix_sock_path
, PATH_MAX
,
4456 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4459 /* Set the cli tool unix socket path */
4460 if (strlen(client_unix_sock_path
) == 0) {
4461 snprintf(client_unix_sock_path
, PATH_MAX
,
4462 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4465 /* Set global SHM for ust */
4466 if (strlen(wait_shm_path
) == 0) {
4467 snprintf(wait_shm_path
, PATH_MAX
,
4468 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4471 /* Set health check Unix path */
4472 if (strlen(health_unix_sock_path
) == 0) {
4473 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4474 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4478 /* Set consumer initial state */
4479 kernel_consumerd_state
= CONSUMER_STOPPED
;
4480 ust_consumerd_state
= CONSUMER_STOPPED
;
4482 DBG("Client socket path %s", client_unix_sock_path
);
4483 DBG("Application socket path %s", apps_unix_sock_path
);
4484 DBG("Application wait path %s", wait_shm_path
);
4485 DBG("LTTng run directory path: %s", rundir
);
4487 /* 32 bits consumerd path setup */
4488 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4489 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4490 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4491 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4493 DBG2("UST consumer 32 bits err path: %s",
4494 ustconsumer32_data
.err_unix_sock_path
);
4495 DBG2("UST consumer 32 bits cmd path: %s",
4496 ustconsumer32_data
.cmd_unix_sock_path
);
4498 /* 64 bits consumerd path setup */
4499 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4500 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4501 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4502 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4504 DBG2("UST consumer 64 bits err path: %s",
4505 ustconsumer64_data
.err_unix_sock_path
);
4506 DBG2("UST consumer 64 bits cmd path: %s",
4507 ustconsumer64_data
.cmd_unix_sock_path
);
4510 * See if daemon already exist.
4512 if ((ret
= check_existing_daemon()) < 0) {
4513 ERR("Already running daemon.\n");
4515 * We do not goto exit because we must not cleanup()
4516 * because a daemon is already running.
4522 * Init UST app hash table. Alloc hash table before this point since
4523 * cleanup() can get called after that point.
4527 /* After this point, we can safely call cleanup() with "goto exit" */
4530 * These actions must be executed as root. We do that *after* setting up
4531 * the sockets path because we MUST make the check for another daemon using
4532 * those paths *before* trying to set the kernel consumer sockets and init
4536 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4541 /* Setup kernel tracer */
4542 if (!opt_no_kernel
) {
4543 init_kernel_tracer();
4546 /* Set ulimit for open files */
4549 /* init lttng_fd tracking must be done after set_ulimit. */
4552 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4557 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4562 if ((ret
= set_signal_handler()) < 0) {
4566 /* Setup the needed unix socket */
4567 if ((ret
= init_daemon_socket()) < 0) {
4571 /* Set credentials to socket */
4572 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4576 /* Get parent pid if -S, --sig-parent is specified. */
4577 if (opt_sig_parent
) {
4581 /* Setup the kernel pipe for waking up the kernel thread */
4582 if (is_root
&& !opt_no_kernel
) {
4583 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4588 /* Setup the thread ht_cleanup communication pipe. */
4589 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4593 /* Setup the thread apps communication pipe. */
4594 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4598 /* Setup the thread apps notify communication pipe. */
4599 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4603 /* Initialize global buffer per UID and PID registry. */
4604 buffer_reg_init_uid_registry();
4605 buffer_reg_init_pid_registry();
4607 /* Init UST command queue. */
4608 cds_wfq_init(&ust_cmd_queue
.queue
);
4611 * Get session list pointer. This pointer MUST NOT be free(). This list is
4612 * statically declared in session.c
4614 session_list_ptr
= session_get_list();
4616 /* Set up max poll set size */
4617 lttng_poll_set_max_size();
4621 /* Check for the application socket timeout env variable. */
4622 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4623 if (env_app_timeout
) {
4624 app_socket_timeout
= atoi(env_app_timeout
);
4626 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4631 /* Create thread to manage the client socket */
4632 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4633 thread_ht_cleanup
, (void *) NULL
);
4635 PERROR("pthread_create ht_cleanup");
4636 goto exit_ht_cleanup
;
4639 /* Create thread to manage the client socket */
4640 ret
= pthread_create(&health_thread
, NULL
,
4641 thread_manage_health
, (void *) NULL
);
4643 PERROR("pthread_create health");
4647 /* Create thread to manage the client socket */
4648 ret
= pthread_create(&client_thread
, NULL
,
4649 thread_manage_clients
, (void *) NULL
);
4651 PERROR("pthread_create clients");
4655 /* Create thread to dispatch registration */
4656 ret
= pthread_create(&dispatch_thread
, NULL
,
4657 thread_dispatch_ust_registration
, (void *) NULL
);
4659 PERROR("pthread_create dispatch");
4663 /* Create thread to manage application registration. */
4664 ret
= pthread_create(®_apps_thread
, NULL
,
4665 thread_registration_apps
, (void *) NULL
);
4667 PERROR("pthread_create registration");
4671 /* Create thread to manage application socket */
4672 ret
= pthread_create(&apps_thread
, NULL
,
4673 thread_manage_apps
, (void *) NULL
);
4675 PERROR("pthread_create apps");
4679 /* Create thread to manage application notify socket */
4680 ret
= pthread_create(&apps_notify_thread
, NULL
,
4681 ust_thread_manage_notify
, (void *) NULL
);
4683 PERROR("pthread_create apps");
4687 /* Don't start this thread if kernel tracing is not requested nor root */
4688 if (is_root
&& !opt_no_kernel
) {
4689 /* Create kernel thread to manage kernel event */
4690 ret
= pthread_create(&kernel_thread
, NULL
,
4691 thread_manage_kernel
, (void *) NULL
);
4693 PERROR("pthread_create kernel");
4697 ret
= pthread_join(kernel_thread
, &status
);
4699 PERROR("pthread_join");
4700 goto error
; /* join error, exit without cleanup */
4705 ret
= pthread_join(apps_thread
, &status
);
4707 PERROR("pthread_join");
4708 goto error
; /* join error, exit without cleanup */
4712 ret
= pthread_join(reg_apps_thread
, &status
);
4714 PERROR("pthread_join");
4715 goto error
; /* join error, exit without cleanup */
4719 ret
= pthread_join(dispatch_thread
, &status
);
4721 PERROR("pthread_join");
4722 goto error
; /* join error, exit without cleanup */
4726 ret
= pthread_join(client_thread
, &status
);
4728 PERROR("pthread_join");
4729 goto error
; /* join error, exit without cleanup */
4732 ret
= join_consumer_thread(&kconsumer_data
);
4734 PERROR("join_consumer");
4735 goto error
; /* join error, exit without cleanup */
4738 ret
= join_consumer_thread(&ustconsumer32_data
);
4740 PERROR("join_consumer ust32");
4741 goto error
; /* join error, exit without cleanup */
4744 ret
= join_consumer_thread(&ustconsumer64_data
);
4746 PERROR("join_consumer ust64");
4747 goto error
; /* join error, exit without cleanup */
4751 ret
= pthread_join(health_thread
, &status
);
4753 PERROR("pthread_join health thread");
4754 goto error
; /* join error, exit without cleanup */
4758 ret
= pthread_join(ht_cleanup_thread
, &status
);
4760 PERROR("pthread_join ht cleanup thread");
4761 goto error
; /* join error, exit without cleanup */
4766 * cleanup() is called when no other thread is running.
4768 rcu_thread_online();
4770 rcu_thread_offline();
4771 rcu_unregister_thread();