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
);
680 * Have we already sent fds to the consumer? If yes, it means
681 * that tracing is started so it is safe to send our updated
684 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
685 struct lttng_ht_iter iter
;
686 struct consumer_socket
*socket
;
689 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
690 &iter
.iter
, socket
, node
.node
) {
691 /* Code flow error */
692 assert(socket
->fd
>= 0);
694 pthread_mutex_lock(socket
->lock
);
695 ret
= kernel_consumer_send_channel_stream(socket
,
697 session
->output_traces
? 1 : 0);
698 pthread_mutex_unlock(socket
->lock
);
709 session_unlock(session
);
711 session_unlock_list();
715 session_unlock(session
);
716 session_unlock_list();
721 * For each tracing session, update newly registered apps. The session list
722 * lock MUST be acquired before calling this.
724 static void update_ust_app(int app_sock
)
726 struct ltt_session
*sess
, *stmp
;
728 /* For all tracing session(s) */
729 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
731 if (sess
->ust_session
) {
732 ust_app_global_update(sess
->ust_session
, app_sock
);
734 session_unlock(sess
);
739 * This thread manage event coming from the kernel.
741 * Features supported in this thread:
744 static void *thread_manage_kernel(void *data
)
746 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
747 uint32_t revents
, nb_fd
;
749 struct lttng_poll_event events
;
751 DBG("[thread] Thread manage kernel started");
753 health_register(HEALTH_TYPE_KERNEL
);
756 * This first step of the while is to clean this structure which could free
757 * non NULL pointers so initialize it before the loop.
759 lttng_poll_init(&events
);
761 if (testpoint(thread_manage_kernel
)) {
762 goto error_testpoint
;
765 health_code_update();
767 if (testpoint(thread_manage_kernel_before_loop
)) {
768 goto error_testpoint
;
772 health_code_update();
774 if (update_poll_flag
== 1) {
775 /* Clean events object. We are about to populate it again. */
776 lttng_poll_clean(&events
);
778 ret
= sessiond_set_thread_pollset(&events
, 2);
780 goto error_poll_create
;
783 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
788 /* This will add the available kernel channel if any. */
789 ret
= update_kernel_poll(&events
);
793 update_poll_flag
= 0;
796 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
798 /* Poll infinite value of time */
801 ret
= lttng_poll_wait(&events
, -1);
805 * Restart interrupted system call.
807 if (errno
== EINTR
) {
811 } else if (ret
== 0) {
812 /* Should not happen since timeout is infinite */
813 ERR("Return value of poll is 0 with an infinite timeout.\n"
814 "This should not have happened! Continuing...");
820 for (i
= 0; i
< nb_fd
; i
++) {
821 /* Fetch once the poll data */
822 revents
= LTTNG_POLL_GETEV(&events
, i
);
823 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
825 health_code_update();
827 /* Thread quit pipe has been closed. Killing thread. */
828 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
834 /* Check for data on kernel pipe */
835 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
837 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
838 } while (ret
< 0 && errno
== EINTR
);
840 * Ret value is useless here, if this pipe gets any actions an
841 * update is required anyway.
843 update_poll_flag
= 1;
847 * New CPU detected by the kernel. Adding kernel stream to
848 * kernel session and updating the kernel consumer
850 if (revents
& LPOLLIN
) {
851 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
857 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
858 * and unregister kernel stream at this point.
867 lttng_poll_clean(&events
);
870 utils_close_pipe(kernel_poll_pipe
);
871 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
874 ERR("Health error occurred in %s", __func__
);
875 WARN("Kernel thread died unexpectedly. "
876 "Kernel tracing can continue but CPU hotplug is disabled.");
879 DBG("Kernel thread dying");
884 * Signal pthread condition of the consumer data that the thread.
886 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
888 pthread_mutex_lock(&data
->cond_mutex
);
891 * The state is set before signaling. It can be any value, it's the waiter
892 * job to correctly interpret this condition variable associated to the
893 * consumer pthread_cond.
895 * A value of 0 means that the corresponding thread of the consumer data
896 * was not started. 1 indicates that the thread has started and is ready
897 * for action. A negative value means that there was an error during the
900 data
->consumer_thread_is_ready
= state
;
901 (void) pthread_cond_signal(&data
->cond
);
903 pthread_mutex_unlock(&data
->cond_mutex
);
907 * This thread manage the consumer error sent back to the session daemon.
909 static void *thread_manage_consumer(void *data
)
911 int sock
= -1, i
, ret
, pollfd
, err
= -1;
912 uint32_t revents
, nb_fd
;
913 enum lttcomm_return_code code
;
914 struct lttng_poll_event events
;
915 struct consumer_data
*consumer_data
= data
;
917 DBG("[thread] Manage consumer started");
919 health_register(HEALTH_TYPE_CONSUMER
);
921 health_code_update();
924 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
925 * metadata_sock. Nothing more will be added to this poll set.
927 ret
= sessiond_set_thread_pollset(&events
, 3);
933 * The error socket here is already in a listening state which was done
934 * just before spawning this thread to avoid a race between the consumer
935 * daemon exec trying to connect and the listen() call.
937 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
942 health_code_update();
944 /* Infinite blocking call, waiting for transmission */
948 if (testpoint(thread_manage_consumer
)) {
952 ret
= lttng_poll_wait(&events
, -1);
956 * Restart interrupted system call.
958 if (errno
== EINTR
) {
966 for (i
= 0; i
< nb_fd
; i
++) {
967 /* Fetch once the poll data */
968 revents
= LTTNG_POLL_GETEV(&events
, i
);
969 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
971 health_code_update();
973 /* Thread quit pipe has been closed. Killing thread. */
974 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
980 /* Event on the registration socket */
981 if (pollfd
== consumer_data
->err_sock
) {
982 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
983 ERR("consumer err socket poll error");
989 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
995 * Set the CLOEXEC flag. Return code is useless because either way, the
998 (void) utils_set_fd_cloexec(sock
);
1000 health_code_update();
1002 DBG2("Receiving code from consumer err_sock");
1004 /* Getting status code from kconsumerd */
1005 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1006 sizeof(enum lttcomm_return_code
));
1011 health_code_update();
1013 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1014 /* Connect both socket, command and metadata. */
1015 consumer_data
->cmd_sock
=
1016 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1017 consumer_data
->metadata_sock
.fd
=
1018 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1019 if (consumer_data
->cmd_sock
< 0 ||
1020 consumer_data
->metadata_sock
.fd
< 0) {
1021 PERROR("consumer connect cmd socket");
1022 /* On error, signal condition and quit. */
1023 signal_consumer_condition(consumer_data
, -1);
1026 /* Create metadata socket lock. */
1027 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1028 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1029 PERROR("zmalloc pthread mutex");
1033 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1035 signal_consumer_condition(consumer_data
, 1);
1036 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1037 DBG("Consumer metadata socket ready (fd: %d)",
1038 consumer_data
->metadata_sock
.fd
);
1040 ERR("consumer error when waiting for SOCK_READY : %s",
1041 lttcomm_get_readable_code(-code
));
1045 /* Remove the consumerd error sock since we've established a connexion */
1046 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1051 /* Add new accepted error socket. */
1052 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1057 /* Add metadata socket that is successfully connected. */
1058 ret
= lttng_poll_add(&events
, consumer_data
->metadata_sock
.fd
,
1059 LPOLLIN
| LPOLLRDHUP
);
1064 health_code_update();
1066 /* Infinite blocking call, waiting for transmission */
1069 health_poll_entry();
1070 ret
= lttng_poll_wait(&events
, -1);
1074 * Restart interrupted system call.
1076 if (errno
== EINTR
) {
1084 for (i
= 0; i
< nb_fd
; i
++) {
1085 /* Fetch once the poll data */
1086 revents
= LTTNG_POLL_GETEV(&events
, i
);
1087 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1089 health_code_update();
1091 /* Thread quit pipe has been closed. Killing thread. */
1092 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1098 if (pollfd
== sock
) {
1099 /* Event on the consumerd socket */
1100 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1101 ERR("consumer err socket second poll error");
1104 health_code_update();
1105 /* Wait for any kconsumerd error */
1106 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1107 sizeof(enum lttcomm_return_code
));
1109 ERR("consumer closed the command socket");
1113 ERR("consumer return code : %s",
1114 lttcomm_get_readable_code(-code
));
1117 } else if (pollfd
== consumer_data
->metadata_sock
.fd
) {
1118 /* UST metadata requests */
1119 ret
= ust_consumer_metadata_request(
1120 &consumer_data
->metadata_sock
);
1122 ERR("Handling metadata request");
1127 ERR("Unknown pollfd");
1131 health_code_update();
1136 /* Immediately set the consumerd state to stopped */
1137 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1138 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1139 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1140 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1141 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1143 /* Code flow error... */
1147 if (consumer_data
->err_sock
>= 0) {
1148 ret
= close(consumer_data
->err_sock
);
1152 consumer_data
->err_sock
= -1;
1154 if (consumer_data
->cmd_sock
>= 0) {
1155 ret
= close(consumer_data
->cmd_sock
);
1159 consumer_data
->cmd_sock
= -1;
1161 if (consumer_data
->metadata_sock
.fd
>= 0) {
1162 ret
= close(consumer_data
->metadata_sock
.fd
);
1167 /* Cleanup metadata socket mutex. */
1168 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1169 free(consumer_data
->metadata_sock
.lock
);
1178 unlink(consumer_data
->err_unix_sock_path
);
1179 unlink(consumer_data
->cmd_unix_sock_path
);
1180 consumer_data
->pid
= 0;
1182 lttng_poll_clean(&events
);
1186 ERR("Health error occurred in %s", __func__
);
1188 health_unregister();
1189 DBG("consumer thread cleanup completed");
1195 * This thread manage application communication.
1197 static void *thread_manage_apps(void *data
)
1199 int i
, ret
, pollfd
, err
= -1;
1200 uint32_t revents
, nb_fd
;
1201 struct lttng_poll_event events
;
1203 DBG("[thread] Manage application started");
1205 rcu_register_thread();
1206 rcu_thread_online();
1208 health_register(HEALTH_TYPE_APP_MANAGE
);
1210 if (testpoint(thread_manage_apps
)) {
1211 goto error_testpoint
;
1214 health_code_update();
1216 ret
= sessiond_set_thread_pollset(&events
, 2);
1218 goto error_poll_create
;
1221 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1226 if (testpoint(thread_manage_apps_before_loop
)) {
1230 health_code_update();
1233 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1235 /* Inifinite blocking call, waiting for transmission */
1237 health_poll_entry();
1238 ret
= lttng_poll_wait(&events
, -1);
1242 * Restart interrupted system call.
1244 if (errno
== EINTR
) {
1252 for (i
= 0; i
< nb_fd
; i
++) {
1253 /* Fetch once the poll data */
1254 revents
= LTTNG_POLL_GETEV(&events
, i
);
1255 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1257 health_code_update();
1259 /* Thread quit pipe has been closed. Killing thread. */
1260 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1266 /* Inspect the apps cmd pipe */
1267 if (pollfd
== apps_cmd_pipe
[0]) {
1268 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1269 ERR("Apps command pipe error");
1271 } else if (revents
& LPOLLIN
) {
1276 ret
= read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1277 } while (ret
< 0 && errno
== EINTR
);
1278 if (ret
< 0 || ret
< sizeof(sock
)) {
1279 PERROR("read apps cmd pipe");
1283 health_code_update();
1286 * We only monitor the error events of the socket. This
1287 * thread does not handle any incoming data from UST
1290 ret
= lttng_poll_add(&events
, sock
,
1291 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1296 /* Set socket timeout for both receiving and ending */
1297 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1298 app_socket_timeout
);
1299 (void) lttcomm_setsockopt_snd_timeout(sock
,
1300 app_socket_timeout
);
1302 DBG("Apps with sock %d added to poll set", sock
);
1304 health_code_update();
1310 * At this point, we know that a registered application made
1311 * the event at poll_wait.
1313 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1314 /* Removing from the poll set */
1315 ret
= lttng_poll_del(&events
, pollfd
);
1320 /* Socket closed on remote end. */
1321 ust_app_unregister(pollfd
);
1326 health_code_update();
1332 lttng_poll_clean(&events
);
1335 utils_close_pipe(apps_cmd_pipe
);
1336 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1339 * We don't clean the UST app hash table here since already registered
1340 * applications can still be controlled so let them be until the session
1341 * daemon dies or the applications stop.
1346 ERR("Health error occurred in %s", __func__
);
1348 health_unregister();
1349 DBG("Application communication apps thread cleanup complete");
1350 rcu_thread_offline();
1351 rcu_unregister_thread();
1356 * Send a socket to a thread This is called from the dispatch UST registration
1357 * thread once all sockets are set for the application.
1359 * On success, return 0 else a negative value being the errno message of the
1362 static int send_socket_to_thread(int fd
, int sock
)
1366 /* Sockets MUST be set or else this should not have been called. */
1371 ret
= write(fd
, &sock
, sizeof(sock
));
1372 } while (ret
< 0 && errno
== EINTR
);
1373 if (ret
< 0 || ret
!= sizeof(sock
)) {
1374 PERROR("write apps pipe %d", fd
);
1381 /* All good. Don't send back the write positive ret value. */
1388 * Sanitize the wait queue of the dispatch registration thread meaning removing
1389 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1390 * notify socket is never received.
1392 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1394 int ret
, nb_fd
= 0, i
;
1395 unsigned int fd_added
= 0;
1396 struct lttng_poll_event events
;
1397 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1401 lttng_poll_init(&events
);
1403 /* Just skip everything for an empty queue. */
1404 if (!wait_queue
->count
) {
1408 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1413 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1414 &wait_queue
->head
, head
) {
1415 assert(wait_node
->app
);
1416 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1417 LPOLLHUP
| LPOLLERR
);
1430 * Poll but don't block so we can quickly identify the faulty events and
1431 * clean them afterwards from the wait queue.
1433 ret
= lttng_poll_wait(&events
, 0);
1439 for (i
= 0; i
< nb_fd
; i
++) {
1440 /* Get faulty FD. */
1441 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1442 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1444 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1445 &wait_queue
->head
, head
) {
1446 if (pollfd
== wait_node
->app
->sock
&&
1447 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1448 cds_list_del(&wait_node
->head
);
1449 wait_queue
->count
--;
1450 ust_app_destroy(wait_node
->app
);
1458 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1462 lttng_poll_clean(&events
);
1466 lttng_poll_clean(&events
);
1468 ERR("Unable to sanitize wait queue");
1473 * Dispatch request from the registration threads to the application
1474 * communication thread.
1476 static void *thread_dispatch_ust_registration(void *data
)
1479 struct cds_wfq_node
*node
;
1480 struct ust_command
*ust_cmd
= NULL
;
1481 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1482 struct ust_reg_wait_queue wait_queue
= {
1486 health_register(HEALTH_TYPE_APP_REG_DISPATCH
);
1488 health_code_update();
1490 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1492 DBG("[thread] Dispatch UST command started");
1494 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1495 health_code_update();
1497 /* Atomically prepare the queue futex */
1498 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1501 struct ust_app
*app
= NULL
;
1505 * Make sure we don't have node(s) that have hung up before receiving
1506 * the notify socket. This is to clean the list in order to avoid
1507 * memory leaks from notify socket that are never seen.
1509 sanitize_wait_queue(&wait_queue
);
1511 health_code_update();
1512 /* Dequeue command for registration */
1513 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1515 DBG("Woken up but nothing in the UST command queue");
1516 /* Continue thread execution */
1520 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1522 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1523 " gid:%d sock:%d name:%s (version %d.%d)",
1524 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1525 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1526 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1527 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1529 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1530 wait_node
= zmalloc(sizeof(*wait_node
));
1532 PERROR("zmalloc wait_node dispatch");
1533 ret
= close(ust_cmd
->sock
);
1535 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1537 lttng_fd_put(1, LTTNG_FD_APPS
);
1541 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1543 /* Create application object if socket is CMD. */
1544 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1546 if (!wait_node
->app
) {
1547 ret
= close(ust_cmd
->sock
);
1549 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1551 lttng_fd_put(1, LTTNG_FD_APPS
);
1557 * Add application to the wait queue so we can set the notify
1558 * socket before putting this object in the global ht.
1560 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1565 * We have to continue here since we don't have the notify
1566 * socket and the application MUST be added to the hash table
1567 * only at that moment.
1572 * Look for the application in the local wait queue and set the
1573 * notify socket if found.
1575 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1576 &wait_queue
.head
, head
) {
1577 health_code_update();
1578 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1579 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1580 cds_list_del(&wait_node
->head
);
1582 app
= wait_node
->app
;
1584 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1590 * With no application at this stage the received socket is
1591 * basically useless so close it before we free the cmd data
1592 * structure for good.
1595 ret
= close(ust_cmd
->sock
);
1597 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1599 lttng_fd_put(1, LTTNG_FD_APPS
);
1606 * @session_lock_list
1608 * Lock the global session list so from the register up to the
1609 * registration done message, no thread can see the application
1610 * and change its state.
1612 session_lock_list();
1616 * Add application to the global hash table. This needs to be
1617 * done before the update to the UST registry can locate the
1622 /* Set app version. This call will print an error if needed. */
1623 (void) ust_app_version(app
);
1625 /* Send notify socket through the notify pipe. */
1626 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1630 session_unlock_list();
1631 /* No notify thread, stop the UST tracing. */
1636 * Update newly registered application with the tracing
1637 * registry info already enabled information.
1639 update_ust_app(app
->sock
);
1642 * Don't care about return value. Let the manage apps threads
1643 * handle app unregistration upon socket close.
1645 (void) ust_app_register_done(app
->sock
);
1648 * Even if the application socket has been closed, send the app
1649 * to the thread and unregistration will take place at that
1652 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1655 session_unlock_list();
1656 /* No apps. thread, stop the UST tracing. */
1661 session_unlock_list();
1663 } while (node
!= NULL
);
1665 health_poll_entry();
1666 /* Futex wait on queue. Blocking call on futex() */
1667 futex_nto1_wait(&ust_cmd_queue
.futex
);
1670 /* Normal exit, no error */
1674 /* Clean up wait queue. */
1675 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1676 &wait_queue
.head
, head
) {
1677 cds_list_del(&wait_node
->head
);
1682 DBG("Dispatch thread dying");
1685 ERR("Health error occurred in %s", __func__
);
1687 health_unregister();
1692 * This thread manage application registration.
1694 static void *thread_registration_apps(void *data
)
1696 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1697 uint32_t revents
, nb_fd
;
1698 struct lttng_poll_event events
;
1700 * Get allocated in this thread, enqueued to a global queue, dequeued and
1701 * freed in the manage apps thread.
1703 struct ust_command
*ust_cmd
= NULL
;
1705 DBG("[thread] Manage application registration started");
1707 health_register(HEALTH_TYPE_APP_REG
);
1709 if (testpoint(thread_registration_apps
)) {
1710 goto error_testpoint
;
1713 ret
= lttcomm_listen_unix_sock(apps_sock
);
1719 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1720 * more will be added to this poll set.
1722 ret
= sessiond_set_thread_pollset(&events
, 2);
1724 goto error_create_poll
;
1727 /* Add the application registration socket */
1728 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1730 goto error_poll_add
;
1733 /* Notify all applications to register */
1734 ret
= notify_ust_apps(1);
1736 ERR("Failed to notify applications or create the wait shared memory.\n"
1737 "Execution continues but there might be problem for already\n"
1738 "running applications that wishes to register.");
1742 DBG("Accepting application registration");
1744 /* Inifinite blocking call, waiting for transmission */
1746 health_poll_entry();
1747 ret
= lttng_poll_wait(&events
, -1);
1751 * Restart interrupted system call.
1753 if (errno
== EINTR
) {
1761 for (i
= 0; i
< nb_fd
; i
++) {
1762 health_code_update();
1764 /* Fetch once the poll data */
1765 revents
= LTTNG_POLL_GETEV(&events
, i
);
1766 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1768 /* Thread quit pipe has been closed. Killing thread. */
1769 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1775 /* Event on the registration socket */
1776 if (pollfd
== apps_sock
) {
1777 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1778 ERR("Register apps socket poll error");
1780 } else if (revents
& LPOLLIN
) {
1781 sock
= lttcomm_accept_unix_sock(apps_sock
);
1787 * Set the CLOEXEC flag. Return code is useless because
1788 * either way, the show must go on.
1790 (void) utils_set_fd_cloexec(sock
);
1792 /* Create UST registration command for enqueuing */
1793 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1794 if (ust_cmd
== NULL
) {
1795 PERROR("ust command zmalloc");
1800 * Using message-based transmissions to ensure we don't
1801 * have to deal with partially received messages.
1803 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1805 ERR("Exhausted file descriptors allowed for applications.");
1815 health_code_update();
1816 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1819 /* Close socket of the application. */
1824 lttng_fd_put(LTTNG_FD_APPS
, 1);
1828 health_code_update();
1830 ust_cmd
->sock
= sock
;
1833 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1834 " gid:%d sock:%d name:%s (version %d.%d)",
1835 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1836 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1837 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1838 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1841 * Lock free enqueue the registration request. The red pill
1842 * has been taken! This apps will be part of the *system*.
1844 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1847 * Wake the registration queue futex. Implicit memory
1848 * barrier with the exchange in cds_wfq_enqueue.
1850 futex_nto1_wake(&ust_cmd_queue
.futex
);
1860 ERR("Health error occurred in %s", __func__
);
1863 /* Notify that the registration thread is gone */
1866 if (apps_sock
>= 0) {
1867 ret
= close(apps_sock
);
1877 lttng_fd_put(LTTNG_FD_APPS
, 1);
1879 unlink(apps_unix_sock_path
);
1882 lttng_poll_clean(&events
);
1886 DBG("UST Registration thread cleanup complete");
1887 health_unregister();
1893 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1894 * exec or it will fails.
1896 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1899 struct timespec timeout
;
1901 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1902 consumer_data
->consumer_thread_is_ready
= 0;
1904 /* Setup pthread condition */
1905 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1908 PERROR("pthread_condattr_init consumer data");
1913 * Set the monotonic clock in order to make sure we DO NOT jump in time
1914 * between the clock_gettime() call and the timedwait call. See bug #324
1915 * for a more details and how we noticed it.
1917 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1920 PERROR("pthread_condattr_setclock consumer data");
1924 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1927 PERROR("pthread_cond_init consumer data");
1931 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1934 PERROR("pthread_create consumer");
1939 /* We are about to wait on a pthread condition */
1940 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1942 /* Get time for sem_timedwait absolute timeout */
1943 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1945 * Set the timeout for the condition timed wait even if the clock gettime
1946 * call fails since we might loop on that call and we want to avoid to
1947 * increment the timeout too many times.
1949 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1952 * The following loop COULD be skipped in some conditions so this is why we
1953 * set ret to 0 in order to make sure at least one round of the loop is
1959 * Loop until the condition is reached or when a timeout is reached. Note
1960 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1961 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1962 * possible. This loop does not take any chances and works with both of
1965 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1966 if (clock_ret
< 0) {
1967 PERROR("clock_gettime spawn consumer");
1968 /* Infinite wait for the consumerd thread to be ready */
1969 ret
= pthread_cond_wait(&consumer_data
->cond
,
1970 &consumer_data
->cond_mutex
);
1972 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1973 &consumer_data
->cond_mutex
, &timeout
);
1977 /* Release the pthread condition */
1978 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1982 if (ret
== ETIMEDOUT
) {
1984 * Call has timed out so we kill the kconsumerd_thread and return
1987 ERR("Condition timed out. The consumer thread was never ready."
1989 ret
= pthread_cancel(consumer_data
->thread
);
1991 PERROR("pthread_cancel consumer thread");
1994 PERROR("pthread_cond_wait failed consumer thread");
1999 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2000 if (consumer_data
->pid
== 0) {
2001 ERR("Consumerd did not start");
2002 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2005 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2014 * Join consumer thread
2016 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2020 /* Consumer pid must be a real one. */
2021 if (consumer_data
->pid
> 0) {
2023 ret
= kill(consumer_data
->pid
, SIGTERM
);
2025 ERR("Error killing consumer daemon");
2028 return pthread_join(consumer_data
->thread
, &status
);
2035 * Fork and exec a consumer daemon (consumerd).
2037 * Return pid if successful else -1.
2039 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2043 const char *consumer_to_use
;
2044 const char *verbosity
;
2047 DBG("Spawning consumerd");
2054 if (opt_verbose_consumer
) {
2055 verbosity
= "--verbose";
2057 verbosity
= "--quiet";
2059 switch (consumer_data
->type
) {
2060 case LTTNG_CONSUMER_KERNEL
:
2062 * Find out which consumerd to execute. We will first try the
2063 * 64-bit path, then the sessiond's installation directory, and
2064 * fallback on the 32-bit one,
2066 DBG3("Looking for a kernel consumer at these locations:");
2067 DBG3(" 1) %s", consumerd64_bin
);
2068 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2069 DBG3(" 3) %s", consumerd32_bin
);
2070 if (stat(consumerd64_bin
, &st
) == 0) {
2071 DBG3("Found location #1");
2072 consumer_to_use
= consumerd64_bin
;
2073 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2074 DBG3("Found location #2");
2075 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2076 } else if (stat(consumerd32_bin
, &st
) == 0) {
2077 DBG3("Found location #3");
2078 consumer_to_use
= consumerd32_bin
;
2080 DBG("Could not find any valid consumerd executable");
2083 DBG("Using kernel consumer at: %s", consumer_to_use
);
2084 execl(consumer_to_use
,
2085 "lttng-consumerd", verbosity
, "-k",
2086 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2087 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2090 case LTTNG_CONSUMER64_UST
:
2092 char *tmpnew
= NULL
;
2094 if (consumerd64_libdir
[0] != '\0') {
2098 tmp
= getenv("LD_LIBRARY_PATH");
2102 tmplen
= strlen("LD_LIBRARY_PATH=")
2103 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2104 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2109 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2110 strcat(tmpnew
, consumerd64_libdir
);
2111 if (tmp
[0] != '\0') {
2112 strcat(tmpnew
, ":");
2113 strcat(tmpnew
, tmp
);
2115 ret
= putenv(tmpnew
);
2122 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2123 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2124 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2125 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2127 if (consumerd64_libdir
[0] != '\0') {
2135 case LTTNG_CONSUMER32_UST
:
2137 char *tmpnew
= NULL
;
2139 if (consumerd32_libdir
[0] != '\0') {
2143 tmp
= getenv("LD_LIBRARY_PATH");
2147 tmplen
= strlen("LD_LIBRARY_PATH=")
2148 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2149 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2154 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2155 strcat(tmpnew
, consumerd32_libdir
);
2156 if (tmp
[0] != '\0') {
2157 strcat(tmpnew
, ":");
2158 strcat(tmpnew
, tmp
);
2160 ret
= putenv(tmpnew
);
2167 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2168 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2169 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2170 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2172 if (consumerd32_libdir
[0] != '\0') {
2181 PERROR("unknown consumer type");
2185 PERROR("kernel start consumer exec");
2188 } else if (pid
> 0) {
2191 PERROR("start consumer fork");
2199 * Spawn the consumerd daemon and session daemon thread.
2201 static int start_consumerd(struct consumer_data
*consumer_data
)
2206 * Set the listen() state on the socket since there is a possible race
2207 * between the exec() of the consumer daemon and this call if place in the
2208 * consumer thread. See bug #366 for more details.
2210 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2215 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2216 if (consumer_data
->pid
!= 0) {
2217 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2221 ret
= spawn_consumerd(consumer_data
);
2223 ERR("Spawning consumerd failed");
2224 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2228 /* Setting up the consumer_data pid */
2229 consumer_data
->pid
= ret
;
2230 DBG2("Consumer pid %d", consumer_data
->pid
);
2231 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2233 DBG2("Spawning consumer control thread");
2234 ret
= spawn_consumer_thread(consumer_data
);
2236 ERR("Fatal error spawning consumer control thread");
2244 /* Cleanup already created sockets on error. */
2245 if (consumer_data
->err_sock
>= 0) {
2248 err
= close(consumer_data
->err_sock
);
2250 PERROR("close consumer data error socket");
2257 * Compute health status of each consumer. If one of them is zero (bad
2258 * state), we return 0.
2260 static int check_consumer_health(void)
2264 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2266 DBG3("Health consumer check %d", ret
);
2272 * Setup necessary data for kernel tracer action.
2274 static int init_kernel_tracer(void)
2278 /* Modprobe lttng kernel modules */
2279 ret
= modprobe_lttng_control();
2284 /* Open debugfs lttng */
2285 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2286 if (kernel_tracer_fd
< 0) {
2287 DBG("Failed to open %s", module_proc_lttng
);
2292 /* Validate kernel version */
2293 ret
= kernel_validate_version(kernel_tracer_fd
);
2298 ret
= modprobe_lttng_data();
2303 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2307 modprobe_remove_lttng_control();
2308 ret
= close(kernel_tracer_fd
);
2312 kernel_tracer_fd
= -1;
2313 return LTTNG_ERR_KERN_VERSION
;
2316 ret
= close(kernel_tracer_fd
);
2322 modprobe_remove_lttng_control();
2325 WARN("No kernel tracer available");
2326 kernel_tracer_fd
= -1;
2328 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2330 return LTTNG_ERR_KERN_NA
;
2336 * Copy consumer output from the tracing session to the domain session. The
2337 * function also applies the right modification on a per domain basis for the
2338 * trace files destination directory.
2340 * Should *NOT* be called with RCU read-side lock held.
2342 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2345 const char *dir_name
;
2346 struct consumer_output
*consumer
;
2349 assert(session
->consumer
);
2352 case LTTNG_DOMAIN_KERNEL
:
2353 DBG3("Copying tracing session consumer output in kernel session");
2355 * XXX: We should audit the session creation and what this function
2356 * does "extra" in order to avoid a destroy since this function is used
2357 * in the domain session creation (kernel and ust) only. Same for UST
2360 if (session
->kernel_session
->consumer
) {
2361 consumer_destroy_output(session
->kernel_session
->consumer
);
2363 session
->kernel_session
->consumer
=
2364 consumer_copy_output(session
->consumer
);
2365 /* Ease our life a bit for the next part */
2366 consumer
= session
->kernel_session
->consumer
;
2367 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2369 case LTTNG_DOMAIN_UST
:
2370 DBG3("Copying tracing session consumer output in UST session");
2371 if (session
->ust_session
->consumer
) {
2372 consumer_destroy_output(session
->ust_session
->consumer
);
2374 session
->ust_session
->consumer
=
2375 consumer_copy_output(session
->consumer
);
2376 /* Ease our life a bit for the next part */
2377 consumer
= session
->ust_session
->consumer
;
2378 dir_name
= DEFAULT_UST_TRACE_DIR
;
2381 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2385 /* Append correct directory to subdir */
2386 strncat(consumer
->subdir
, dir_name
,
2387 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2388 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2397 * Create an UST session and add it to the session ust list.
2399 * Should *NOT* be called with RCU read-side lock held.
2401 static int create_ust_session(struct ltt_session
*session
,
2402 struct lttng_domain
*domain
)
2405 struct ltt_ust_session
*lus
= NULL
;
2409 assert(session
->consumer
);
2411 switch (domain
->type
) {
2412 case LTTNG_DOMAIN_UST
:
2415 ERR("Unknown UST domain on create session %d", domain
->type
);
2416 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2420 DBG("Creating UST session");
2422 lus
= trace_ust_create_session(session
->id
);
2424 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2428 lus
->uid
= session
->uid
;
2429 lus
->gid
= session
->gid
;
2430 lus
->output_traces
= session
->output_traces
;
2431 lus
->snapshot_mode
= session
->snapshot_mode
;
2432 session
->ust_session
= lus
;
2434 /* Copy session output to the newly created UST session */
2435 ret
= copy_session_consumer(domain
->type
, session
);
2436 if (ret
!= LTTNG_OK
) {
2444 session
->ust_session
= NULL
;
2449 * Create a kernel tracer session then create the default channel.
2451 static int create_kernel_session(struct ltt_session
*session
)
2455 DBG("Creating kernel session");
2457 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2459 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2463 /* Code flow safety */
2464 assert(session
->kernel_session
);
2466 /* Copy session output to the newly created Kernel session */
2467 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2468 if (ret
!= LTTNG_OK
) {
2472 /* Create directory(ies) on local filesystem. */
2473 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2474 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2475 ret
= run_as_mkdir_recursive(
2476 session
->kernel_session
->consumer
->dst
.trace_path
,
2477 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2479 if (ret
!= -EEXIST
) {
2480 ERR("Trace directory creation error");
2486 session
->kernel_session
->uid
= session
->uid
;
2487 session
->kernel_session
->gid
= session
->gid
;
2488 session
->kernel_session
->output_traces
= session
->output_traces
;
2489 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2494 trace_kernel_destroy_session(session
->kernel_session
);
2495 session
->kernel_session
= NULL
;
2500 * Count number of session permitted by uid/gid.
2502 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2505 struct ltt_session
*session
;
2507 DBG("Counting number of available session for UID %d GID %d",
2509 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2511 * Only list the sessions the user can control.
2513 if (!session_access_ok(session
, uid
, gid
)) {
2522 * Process the command requested by the lttng client within the command
2523 * context structure. This function make sure that the return structure (llm)
2524 * is set and ready for transmission before returning.
2526 * Return any error encountered or 0 for success.
2528 * "sock" is only used for special-case var. len data.
2530 * Should *NOT* be called with RCU read-side lock held.
2532 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2536 int need_tracing_session
= 1;
2539 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2543 switch (cmd_ctx
->lsm
->cmd_type
) {
2544 case LTTNG_CREATE_SESSION
:
2545 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2546 case LTTNG_DESTROY_SESSION
:
2547 case LTTNG_LIST_SESSIONS
:
2548 case LTTNG_LIST_DOMAINS
:
2549 case LTTNG_START_TRACE
:
2550 case LTTNG_STOP_TRACE
:
2551 case LTTNG_DATA_PENDING
:
2552 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2553 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2554 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2555 case LTTNG_SNAPSHOT_RECORD
:
2562 if (opt_no_kernel
&& need_domain
2563 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2565 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2567 ret
= LTTNG_ERR_KERN_NA
;
2572 /* Deny register consumer if we already have a spawned consumer. */
2573 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2574 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2575 if (kconsumer_data
.pid
> 0) {
2576 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2577 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2580 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2584 * Check for command that don't needs to allocate a returned payload. We do
2585 * this here so we don't have to make the call for no payload at each
2588 switch(cmd_ctx
->lsm
->cmd_type
) {
2589 case LTTNG_LIST_SESSIONS
:
2590 case LTTNG_LIST_TRACEPOINTS
:
2591 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2592 case LTTNG_LIST_DOMAINS
:
2593 case LTTNG_LIST_CHANNELS
:
2594 case LTTNG_LIST_EVENTS
:
2597 /* Setup lttng message with no payload */
2598 ret
= setup_lttng_msg(cmd_ctx
, 0);
2600 /* This label does not try to unlock the session */
2601 goto init_setup_error
;
2605 /* Commands that DO NOT need a session. */
2606 switch (cmd_ctx
->lsm
->cmd_type
) {
2607 case LTTNG_CREATE_SESSION
:
2608 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2609 case LTTNG_CALIBRATE
:
2610 case LTTNG_LIST_SESSIONS
:
2611 case LTTNG_LIST_TRACEPOINTS
:
2612 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2613 need_tracing_session
= 0;
2616 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2618 * We keep the session list lock across _all_ commands
2619 * for now, because the per-session lock does not
2620 * handle teardown properly.
2622 session_lock_list();
2623 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2624 if (cmd_ctx
->session
== NULL
) {
2625 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2628 /* Acquire lock for the session */
2629 session_lock(cmd_ctx
->session
);
2639 * Check domain type for specific "pre-action".
2641 switch (cmd_ctx
->lsm
->domain
.type
) {
2642 case LTTNG_DOMAIN_KERNEL
:
2644 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2648 /* Kernel tracer check */
2649 if (kernel_tracer_fd
== -1) {
2650 /* Basically, load kernel tracer modules */
2651 ret
= init_kernel_tracer();
2657 /* Consumer is in an ERROR state. Report back to client */
2658 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2659 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2663 /* Need a session for kernel command */
2664 if (need_tracing_session
) {
2665 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2666 ret
= create_kernel_session(cmd_ctx
->session
);
2668 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2673 /* Start the kernel consumer daemon */
2674 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2675 if (kconsumer_data
.pid
== 0 &&
2676 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2677 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2678 ret
= start_consumerd(&kconsumer_data
);
2680 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2683 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2685 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2689 * The consumer was just spawned so we need to add the socket to
2690 * the consumer output of the session if exist.
2692 ret
= consumer_create_socket(&kconsumer_data
,
2693 cmd_ctx
->session
->kernel_session
->consumer
);
2700 case LTTNG_DOMAIN_UST
:
2702 /* Consumer is in an ERROR state. Report back to client */
2703 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2704 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2708 if (need_tracing_session
) {
2709 /* Create UST session if none exist. */
2710 if (cmd_ctx
->session
->ust_session
== NULL
) {
2711 ret
= create_ust_session(cmd_ctx
->session
,
2712 &cmd_ctx
->lsm
->domain
);
2713 if (ret
!= LTTNG_OK
) {
2718 /* Start the UST consumer daemons */
2720 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2721 if (consumerd64_bin
[0] != '\0' &&
2722 ustconsumer64_data
.pid
== 0 &&
2723 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2724 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2725 ret
= start_consumerd(&ustconsumer64_data
);
2727 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2728 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2732 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2733 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2735 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2739 * Setup socket for consumer 64 bit. No need for atomic access
2740 * since it was set above and can ONLY be set in this thread.
2742 ret
= consumer_create_socket(&ustconsumer64_data
,
2743 cmd_ctx
->session
->ust_session
->consumer
);
2749 if (consumerd32_bin
[0] != '\0' &&
2750 ustconsumer32_data
.pid
== 0 &&
2751 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2752 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2753 ret
= start_consumerd(&ustconsumer32_data
);
2755 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2756 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2760 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2761 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2763 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2767 * Setup socket for consumer 64 bit. No need for atomic access
2768 * since it was set above and can ONLY be set in this thread.
2770 ret
= consumer_create_socket(&ustconsumer32_data
,
2771 cmd_ctx
->session
->ust_session
->consumer
);
2783 /* Validate consumer daemon state when start/stop trace command */
2784 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2785 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2786 switch (cmd_ctx
->lsm
->domain
.type
) {
2787 case LTTNG_DOMAIN_UST
:
2788 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2789 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2793 case LTTNG_DOMAIN_KERNEL
:
2794 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2795 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2803 * Check that the UID or GID match that of the tracing session.
2804 * The root user can interact with all sessions.
2806 if (need_tracing_session
) {
2807 if (!session_access_ok(cmd_ctx
->session
,
2808 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2809 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2810 ret
= LTTNG_ERR_EPERM
;
2816 * Send relayd information to consumer as soon as we have a domain and a
2819 if (cmd_ctx
->session
&& need_domain
) {
2821 * Setup relayd if not done yet. If the relayd information was already
2822 * sent to the consumer, this call will gracefully return.
2824 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2825 if (ret
!= LTTNG_OK
) {
2830 /* Process by command type */
2831 switch (cmd_ctx
->lsm
->cmd_type
) {
2832 case LTTNG_ADD_CONTEXT
:
2834 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2835 cmd_ctx
->lsm
->u
.context
.channel_name
,
2836 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2839 case LTTNG_DISABLE_CHANNEL
:
2841 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2842 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2845 case LTTNG_DISABLE_EVENT
:
2847 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2848 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2849 cmd_ctx
->lsm
->u
.disable
.name
);
2852 case LTTNG_DISABLE_ALL_EVENT
:
2854 DBG("Disabling all events");
2856 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2857 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2860 case LTTNG_ENABLE_CHANNEL
:
2862 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2863 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2866 case LTTNG_ENABLE_EVENT
:
2868 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2869 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2870 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2873 case LTTNG_ENABLE_ALL_EVENT
:
2875 DBG("Enabling all events");
2877 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2878 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2879 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2882 case LTTNG_LIST_TRACEPOINTS
:
2884 struct lttng_event
*events
;
2887 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2888 if (nb_events
< 0) {
2889 /* Return value is a negative lttng_error_code. */
2895 * Setup lttng message with payload size set to the event list size in
2896 * bytes and then copy list into the llm payload.
2898 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2904 /* Copy event list into message payload */
2905 memcpy(cmd_ctx
->llm
->payload
, events
,
2906 sizeof(struct lttng_event
) * nb_events
);
2913 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2915 struct lttng_event_field
*fields
;
2918 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2920 if (nb_fields
< 0) {
2921 /* Return value is a negative lttng_error_code. */
2927 * Setup lttng message with payload size set to the event list size in
2928 * bytes and then copy list into the llm payload.
2930 ret
= setup_lttng_msg(cmd_ctx
,
2931 sizeof(struct lttng_event_field
) * nb_fields
);
2937 /* Copy event list into message payload */
2938 memcpy(cmd_ctx
->llm
->payload
, fields
,
2939 sizeof(struct lttng_event_field
) * nb_fields
);
2946 case LTTNG_SET_CONSUMER_URI
:
2949 struct lttng_uri
*uris
;
2951 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2952 len
= nb_uri
* sizeof(struct lttng_uri
);
2955 ret
= LTTNG_ERR_INVALID
;
2959 uris
= zmalloc(len
);
2961 ret
= LTTNG_ERR_FATAL
;
2965 /* Receive variable len data */
2966 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2967 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2969 DBG("No URIs received from client... continuing");
2971 ret
= LTTNG_ERR_SESSION_FAIL
;
2976 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2978 if (ret
!= LTTNG_OK
) {
2984 * XXX: 0 means that this URI should be applied on the session. Should
2985 * be a DOMAIN enuam.
2987 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2988 /* Add the URI for the UST session if a consumer is present. */
2989 if (cmd_ctx
->session
->ust_session
&&
2990 cmd_ctx
->session
->ust_session
->consumer
) {
2991 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2993 } else if (cmd_ctx
->session
->kernel_session
&&
2994 cmd_ctx
->session
->kernel_session
->consumer
) {
2995 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2996 cmd_ctx
->session
, nb_uri
, uris
);
3004 case LTTNG_START_TRACE
:
3006 ret
= cmd_start_trace(cmd_ctx
->session
);
3009 case LTTNG_STOP_TRACE
:
3011 ret
= cmd_stop_trace(cmd_ctx
->session
);
3014 case LTTNG_CREATE_SESSION
:
3017 struct lttng_uri
*uris
= NULL
;
3019 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3020 len
= nb_uri
* sizeof(struct lttng_uri
);
3023 uris
= zmalloc(len
);
3025 ret
= LTTNG_ERR_FATAL
;
3029 /* Receive variable len data */
3030 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3031 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3033 DBG("No URIs received from client... continuing");
3035 ret
= LTTNG_ERR_SESSION_FAIL
;
3040 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3041 DBG("Creating session with ONE network URI is a bad call");
3042 ret
= LTTNG_ERR_SESSION_FAIL
;
3048 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3055 case LTTNG_DESTROY_SESSION
:
3057 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3059 /* Set session to NULL so we do not unlock it after free. */
3060 cmd_ctx
->session
= NULL
;
3063 case LTTNG_LIST_DOMAINS
:
3066 struct lttng_domain
*domains
;
3068 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3070 /* Return value is a negative lttng_error_code. */
3075 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3081 /* Copy event list into message payload */
3082 memcpy(cmd_ctx
->llm
->payload
, domains
,
3083 nb_dom
* sizeof(struct lttng_domain
));
3090 case LTTNG_LIST_CHANNELS
:
3093 struct lttng_channel
*channels
;
3095 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3096 cmd_ctx
->session
, &channels
);
3098 /* Return value is a negative lttng_error_code. */
3103 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3109 /* Copy event list into message payload */
3110 memcpy(cmd_ctx
->llm
->payload
, channels
,
3111 nb_chan
* sizeof(struct lttng_channel
));
3118 case LTTNG_LIST_EVENTS
:
3121 struct lttng_event
*events
= NULL
;
3123 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3124 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3126 /* Return value is a negative lttng_error_code. */
3131 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3137 /* Copy event list into message payload */
3138 memcpy(cmd_ctx
->llm
->payload
, events
,
3139 nb_event
* sizeof(struct lttng_event
));
3146 case LTTNG_LIST_SESSIONS
:
3148 unsigned int nr_sessions
;
3150 session_lock_list();
3151 nr_sessions
= lttng_sessions_count(
3152 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3153 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3155 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3157 session_unlock_list();
3161 /* Filled the session array */
3162 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3163 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3164 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3166 session_unlock_list();
3171 case LTTNG_CALIBRATE
:
3173 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3174 &cmd_ctx
->lsm
->u
.calibrate
);
3177 case LTTNG_REGISTER_CONSUMER
:
3179 struct consumer_data
*cdata
;
3181 switch (cmd_ctx
->lsm
->domain
.type
) {
3182 case LTTNG_DOMAIN_KERNEL
:
3183 cdata
= &kconsumer_data
;
3186 ret
= LTTNG_ERR_UND
;
3190 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3191 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3194 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3196 struct lttng_filter_bytecode
*bytecode
;
3198 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3199 ret
= LTTNG_ERR_FILTER_INVAL
;
3202 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3203 ret
= LTTNG_ERR_FILTER_INVAL
;
3206 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3208 ret
= LTTNG_ERR_FILTER_NOMEM
;
3211 /* Receive var. len. data */
3212 DBG("Receiving var len data from client ...");
3213 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3214 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3216 DBG("Nothing recv() from client var len data... continuing");
3218 ret
= LTTNG_ERR_FILTER_INVAL
;
3222 if (bytecode
->len
+ sizeof(*bytecode
)
3223 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3225 ret
= LTTNG_ERR_FILTER_INVAL
;
3229 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3230 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3231 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3234 case LTTNG_DATA_PENDING
:
3236 ret
= cmd_data_pending(cmd_ctx
->session
);
3239 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3241 struct lttcomm_lttng_output_id reply
;
3243 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3244 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3245 if (ret
!= LTTNG_OK
) {
3249 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3254 /* Copy output list into message payload */
3255 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3259 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3261 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3262 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3265 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3268 struct lttng_snapshot_output
*outputs
= NULL
;
3270 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3271 if (nb_output
< 0) {
3276 ret
= setup_lttng_msg(cmd_ctx
,
3277 nb_output
* sizeof(struct lttng_snapshot_output
));
3284 /* Copy output list into message payload */
3285 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3286 nb_output
* sizeof(struct lttng_snapshot_output
));
3293 case LTTNG_SNAPSHOT_RECORD
:
3295 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3296 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3297 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3300 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3303 struct lttng_uri
*uris
= NULL
;
3305 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3306 len
= nb_uri
* sizeof(struct lttng_uri
);
3309 uris
= zmalloc(len
);
3311 ret
= LTTNG_ERR_FATAL
;
3315 /* Receive variable len data */
3316 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3317 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3319 DBG("No URIs received from client... continuing");
3321 ret
= LTTNG_ERR_SESSION_FAIL
;
3326 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3327 DBG("Creating session with ONE network URI is a bad call");
3328 ret
= LTTNG_ERR_SESSION_FAIL
;
3334 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3335 nb_uri
, &cmd_ctx
->creds
);
3340 ret
= LTTNG_ERR_UND
;
3345 if (cmd_ctx
->llm
== NULL
) {
3346 DBG("Missing llm structure. Allocating one.");
3347 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3351 /* Set return code */
3352 cmd_ctx
->llm
->ret_code
= ret
;
3354 if (cmd_ctx
->session
) {
3355 session_unlock(cmd_ctx
->session
);
3357 if (need_tracing_session
) {
3358 session_unlock_list();
3365 * Thread managing health check socket.
3367 static void *thread_manage_health(void *data
)
3369 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3370 uint32_t revents
, nb_fd
;
3371 struct lttng_poll_event events
;
3372 struct lttcomm_health_msg msg
;
3373 struct lttcomm_health_data reply
;
3375 DBG("[thread] Manage health check started");
3377 rcu_register_thread();
3379 /* We might hit an error path before this is created. */
3380 lttng_poll_init(&events
);
3382 /* Create unix socket */
3383 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3385 ERR("Unable to create health check Unix socket");
3391 * Set the CLOEXEC flag. Return code is useless because either way, the
3394 (void) utils_set_fd_cloexec(sock
);
3396 ret
= lttcomm_listen_unix_sock(sock
);
3402 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3403 * more will be added to this poll set.
3405 ret
= sessiond_set_thread_pollset(&events
, 2);
3410 /* Add the application registration socket */
3411 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3417 DBG("Health check ready");
3419 /* Inifinite blocking call, waiting for transmission */
3421 ret
= lttng_poll_wait(&events
, -1);
3424 * Restart interrupted system call.
3426 if (errno
== EINTR
) {
3434 for (i
= 0; i
< nb_fd
; i
++) {
3435 /* Fetch once the poll data */
3436 revents
= LTTNG_POLL_GETEV(&events
, i
);
3437 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3439 /* Thread quit pipe has been closed. Killing thread. */
3440 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3446 /* Event on the registration socket */
3447 if (pollfd
== sock
) {
3448 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3449 ERR("Health socket poll error");
3455 new_sock
= lttcomm_accept_unix_sock(sock
);
3461 * Set the CLOEXEC flag. Return code is useless because either way, the
3464 (void) utils_set_fd_cloexec(new_sock
);
3466 DBG("Receiving data from client for health...");
3467 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3469 DBG("Nothing recv() from client... continuing");
3470 ret
= close(new_sock
);
3478 rcu_thread_online();
3480 switch (msg
.component
) {
3481 case LTTNG_HEALTH_CMD
:
3482 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3484 case LTTNG_HEALTH_APP_MANAGE
:
3485 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3487 case LTTNG_HEALTH_APP_REG
:
3488 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3490 case LTTNG_HEALTH_KERNEL
:
3491 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3493 case LTTNG_HEALTH_CONSUMER
:
3494 reply
.ret_code
= check_consumer_health();
3496 case LTTNG_HEALTH_HT_CLEANUP
:
3497 reply
.ret_code
= health_check_state(HEALTH_TYPE_HT_CLEANUP
);
3499 case LTTNG_HEALTH_APP_MANAGE_NOTIFY
:
3500 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3502 case LTTNG_HEALTH_APP_REG_DISPATCH
:
3503 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3505 case LTTNG_HEALTH_ALL
:
3507 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3508 health_check_state(HEALTH_TYPE_APP_REG
) &&
3509 health_check_state(HEALTH_TYPE_CMD
) &&
3510 health_check_state(HEALTH_TYPE_KERNEL
) &&
3511 check_consumer_health() &&
3512 health_check_state(HEALTH_TYPE_HT_CLEANUP
) &&
3513 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
) &&
3514 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3517 reply
.ret_code
= LTTNG_ERR_UND
;
3522 * Flip ret value since 0 is a success and 1 indicates a bad health for
3523 * the client where in the sessiond it is the opposite. Again, this is
3524 * just to make things easier for us poor developer which enjoy a lot
3527 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3528 reply
.ret_code
= !reply
.ret_code
;
3531 DBG2("Health check return value %d", reply
.ret_code
);
3533 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3535 ERR("Failed to send health data back to client");
3538 /* End of transmission */
3539 ret
= close(new_sock
);
3549 ERR("Health error occurred in %s", __func__
);
3551 DBG("Health check thread dying");
3552 unlink(health_unix_sock_path
);
3560 lttng_poll_clean(&events
);
3562 rcu_unregister_thread();
3567 * This thread manage all clients request using the unix client socket for
3570 static void *thread_manage_clients(void *data
)
3572 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3574 uint32_t revents
, nb_fd
;
3575 struct command_ctx
*cmd_ctx
= NULL
;
3576 struct lttng_poll_event events
;
3578 DBG("[thread] Manage client started");
3580 rcu_register_thread();
3582 health_register(HEALTH_TYPE_CMD
);
3584 if (testpoint(thread_manage_clients
)) {
3585 goto error_testpoint
;
3588 health_code_update();
3590 ret
= lttcomm_listen_unix_sock(client_sock
);
3596 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3597 * more will be added to this poll set.
3599 ret
= sessiond_set_thread_pollset(&events
, 2);
3601 goto error_create_poll
;
3604 /* Add the application registration socket */
3605 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3611 * Notify parent pid that we are ready to accept command for client side.
3613 if (opt_sig_parent
) {
3614 kill(ppid
, SIGUSR1
);
3617 if (testpoint(thread_manage_clients_before_loop
)) {
3621 health_code_update();
3624 DBG("Accepting client command ...");
3626 /* Inifinite blocking call, waiting for transmission */
3628 health_poll_entry();
3629 ret
= lttng_poll_wait(&events
, -1);
3633 * Restart interrupted system call.
3635 if (errno
== EINTR
) {
3643 for (i
= 0; i
< nb_fd
; i
++) {
3644 /* Fetch once the poll data */
3645 revents
= LTTNG_POLL_GETEV(&events
, i
);
3646 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3648 health_code_update();
3650 /* Thread quit pipe has been closed. Killing thread. */
3651 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3657 /* Event on the registration socket */
3658 if (pollfd
== client_sock
) {
3659 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3660 ERR("Client socket poll error");
3666 DBG("Wait for client response");
3668 health_code_update();
3670 sock
= lttcomm_accept_unix_sock(client_sock
);
3676 * Set the CLOEXEC flag. Return code is useless because either way, the
3679 (void) utils_set_fd_cloexec(sock
);
3681 /* Set socket option for credentials retrieval */
3682 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3687 /* Allocate context command to process the client request */
3688 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3689 if (cmd_ctx
== NULL
) {
3690 PERROR("zmalloc cmd_ctx");
3694 /* Allocate data buffer for reception */
3695 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3696 if (cmd_ctx
->lsm
== NULL
) {
3697 PERROR("zmalloc cmd_ctx->lsm");
3701 cmd_ctx
->llm
= NULL
;
3702 cmd_ctx
->session
= NULL
;
3704 health_code_update();
3707 * Data is received from the lttng client. The struct
3708 * lttcomm_session_msg (lsm) contains the command and data request of
3711 DBG("Receiving data from client ...");
3712 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3713 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3715 DBG("Nothing recv() from client... continuing");
3721 clean_command_ctx(&cmd_ctx
);
3725 health_code_update();
3727 // TODO: Validate cmd_ctx including sanity check for
3728 // security purpose.
3730 rcu_thread_online();
3732 * This function dispatch the work to the kernel or userspace tracer
3733 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3734 * informations for the client. The command context struct contains
3735 * everything this function may needs.
3737 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3738 rcu_thread_offline();
3746 * TODO: Inform client somehow of the fatal error. At
3747 * this point, ret < 0 means that a zmalloc failed
3748 * (ENOMEM). Error detected but still accept
3749 * command, unless a socket error has been
3752 clean_command_ctx(&cmd_ctx
);
3756 health_code_update();
3758 DBG("Sending response (size: %d, retcode: %s)",
3759 cmd_ctx
->lttng_msg_size
,
3760 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3761 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3763 ERR("Failed to send data back to client");
3766 /* End of transmission */
3773 clean_command_ctx(&cmd_ctx
);
3775 health_code_update();
3787 lttng_poll_clean(&events
);
3788 clean_command_ctx(&cmd_ctx
);
3793 unlink(client_unix_sock_path
);
3794 if (client_sock
>= 0) {
3795 ret
= close(client_sock
);
3803 ERR("Health error occurred in %s", __func__
);
3806 health_unregister();
3808 DBG("Client thread dying");
3810 rcu_unregister_thread();
3816 * usage function on stderr
3818 static void usage(void)
3820 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3821 fprintf(stderr
, " -h, --help Display this usage.\n");
3822 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3823 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3824 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3825 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3826 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3827 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3828 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3829 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3830 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3831 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3832 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3833 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3834 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3835 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3836 fprintf(stderr
, " -V, --version Show version number.\n");
3837 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3838 fprintf(stderr
, " -q, --quiet No output at all.\n");
3839 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3840 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3841 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3842 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3846 * daemon argument parsing
3848 static int parse_args(int argc
, char **argv
)
3852 static struct option long_options
[] = {
3853 { "client-sock", 1, 0, 'c' },
3854 { "apps-sock", 1, 0, 'a' },
3855 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3856 { "kconsumerd-err-sock", 1, 0, 'E' },
3857 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3858 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3859 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3860 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3861 { "consumerd32-path", 1, 0, 'u' },
3862 { "consumerd32-libdir", 1, 0, 'U' },
3863 { "consumerd64-path", 1, 0, 't' },
3864 { "consumerd64-libdir", 1, 0, 'T' },
3865 { "daemonize", 0, 0, 'd' },
3866 { "sig-parent", 0, 0, 'S' },
3867 { "help", 0, 0, 'h' },
3868 { "group", 1, 0, 'g' },
3869 { "version", 0, 0, 'V' },
3870 { "quiet", 0, 0, 'q' },
3871 { "verbose", 0, 0, 'v' },
3872 { "verbose-consumer", 0, 0, 'Z' },
3873 { "no-kernel", 0, 0, 'N' },
3874 { "pidfile", 1, 0, 'p' },
3879 int option_index
= 0;
3880 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3881 long_options
, &option_index
);
3888 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3890 fprintf(stderr
, " with arg %s\n", optarg
);
3894 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3897 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3903 opt_tracing_group
= optarg
;
3909 fprintf(stdout
, "%s\n", VERSION
);
3915 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3918 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3921 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3924 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3927 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3930 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3936 lttng_opt_quiet
= 1;
3939 /* Verbose level can increase using multiple -v */
3940 lttng_opt_verbose
+= 1;
3943 opt_verbose_consumer
+= 1;
3946 consumerd32_bin
= optarg
;
3949 consumerd32_libdir
= optarg
;
3952 consumerd64_bin
= optarg
;
3955 consumerd64_libdir
= optarg
;
3958 opt_pidfile
= optarg
;
3961 /* Unknown option or other error.
3962 * Error is printed by getopt, just return */
3971 * Creates the two needed socket by the daemon.
3972 * apps_sock - The communication socket for all UST apps.
3973 * client_sock - The communication of the cli tool (lttng).
3975 static int init_daemon_socket(void)
3980 old_umask
= umask(0);
3982 /* Create client tool unix socket */
3983 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3984 if (client_sock
< 0) {
3985 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3990 /* Set the cloexec flag */
3991 ret
= utils_set_fd_cloexec(client_sock
);
3993 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3994 "Continuing but note that the consumer daemon will have a "
3995 "reference to this socket on exec()", client_sock
);
3998 /* File permission MUST be 660 */
3999 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4001 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4006 /* Create the application unix socket */
4007 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4008 if (apps_sock
< 0) {
4009 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4014 /* Set the cloexec flag */
4015 ret
= utils_set_fd_cloexec(apps_sock
);
4017 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4018 "Continuing but note that the consumer daemon will have a "
4019 "reference to this socket on exec()", apps_sock
);
4022 /* File permission MUST be 666 */
4023 ret
= chmod(apps_unix_sock_path
,
4024 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4026 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4031 DBG3("Session daemon client socket %d and application socket %d created",
4032 client_sock
, apps_sock
);
4040 * Check if the global socket is available, and if a daemon is answering at the
4041 * other side. If yes, error is returned.
4043 static int check_existing_daemon(void)
4045 /* Is there anybody out there ? */
4046 if (lttng_session_daemon_alive()) {
4054 * Set the tracing group gid onto the client socket.
4056 * Race window between mkdir and chown is OK because we are going from more
4057 * permissive (root.root) to less permissive (root.tracing).
4059 static int set_permissions(char *rundir
)
4064 ret
= allowed_group();
4066 WARN("No tracing group detected");
4073 /* Set lttng run dir */
4074 ret
= chown(rundir
, 0, gid
);
4076 ERR("Unable to set group on %s", rundir
);
4080 /* Ensure tracing group can search the run dir */
4081 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
4083 ERR("Unable to set permissions on %s", rundir
);
4087 /* lttng client socket path */
4088 ret
= chown(client_unix_sock_path
, 0, gid
);
4090 ERR("Unable to set group on %s", client_unix_sock_path
);
4094 /* kconsumer error socket path */
4095 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
4097 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4101 /* 64-bit ustconsumer error socket path */
4102 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
4104 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4108 /* 32-bit ustconsumer compat32 error socket path */
4109 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
4111 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4115 DBG("All permissions are set");
4122 * Create the lttng run directory needed for all global sockets and pipe.
4124 static int create_lttng_rundir(const char *rundir
)
4128 DBG3("Creating LTTng run directory: %s", rundir
);
4130 ret
= mkdir(rundir
, S_IRWXU
);
4132 if (errno
!= EEXIST
) {
4133 ERR("Unable to create %s", rundir
);
4145 * Setup sockets and directory needed by the kconsumerd communication with the
4148 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4152 char path
[PATH_MAX
];
4154 switch (consumer_data
->type
) {
4155 case LTTNG_CONSUMER_KERNEL
:
4156 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4158 case LTTNG_CONSUMER64_UST
:
4159 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4161 case LTTNG_CONSUMER32_UST
:
4162 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4165 ERR("Consumer type unknown");
4170 DBG2("Creating consumer directory: %s", path
);
4172 ret
= mkdir(path
, S_IRWXU
);
4174 if (errno
!= EEXIST
) {
4176 ERR("Failed to create %s", path
);
4182 /* Create the kconsumerd error unix socket */
4183 consumer_data
->err_sock
=
4184 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4185 if (consumer_data
->err_sock
< 0) {
4186 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4192 * Set the CLOEXEC flag. Return code is useless because either way, the
4195 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4197 PERROR("utils_set_fd_cloexec");
4198 /* continue anyway */
4201 /* File permission MUST be 660 */
4202 ret
= chmod(consumer_data
->err_unix_sock_path
,
4203 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4205 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4215 * Signal handler for the daemon
4217 * Simply stop all worker threads, leaving main() return gracefully after
4218 * joining all threads and calling cleanup().
4220 static void sighandler(int sig
)
4224 DBG("SIGPIPE caught");
4227 DBG("SIGINT caught");
4231 DBG("SIGTERM caught");
4240 * Setup signal handler for :
4241 * SIGINT, SIGTERM, SIGPIPE
4243 static int set_signal_handler(void)
4246 struct sigaction sa
;
4249 if ((ret
= sigemptyset(&sigset
)) < 0) {
4250 PERROR("sigemptyset");
4254 sa
.sa_handler
= sighandler
;
4255 sa
.sa_mask
= sigset
;
4257 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4258 PERROR("sigaction");
4262 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4263 PERROR("sigaction");
4267 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4268 PERROR("sigaction");
4272 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4278 * Set open files limit to unlimited. This daemon can open a large number of
4279 * file descriptors in order to consumer multiple kernel traces.
4281 static void set_ulimit(void)
4286 /* The kernel does not allowed an infinite limit for open files */
4287 lim
.rlim_cur
= 65535;
4288 lim
.rlim_max
= 65535;
4290 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4292 PERROR("failed to set open files limit");
4297 * Write pidfile using the rundir and opt_pidfile.
4299 static void write_pidfile(void)
4302 char pidfile_path
[PATH_MAX
];
4307 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4309 /* Build pidfile path from rundir and opt_pidfile. */
4310 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4311 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4313 PERROR("snprintf pidfile path");
4319 * Create pid file in rundir. Return value is of no importance. The
4320 * execution will continue even though we are not able to write the file.
4322 (void) utils_create_pid_file(getpid(), pidfile_path
);
4331 int main(int argc
, char **argv
)
4335 const char *home_path
, *env_app_timeout
;
4337 init_kernel_workarounds();
4339 rcu_register_thread();
4341 setup_consumerd_path();
4343 page_size
= sysconf(_SC_PAGESIZE
);
4344 if (page_size
< 0) {
4345 PERROR("sysconf _SC_PAGESIZE");
4346 page_size
= LONG_MAX
;
4347 WARN("Fallback page size to %ld", page_size
);
4350 /* Parse arguments */
4352 if ((ret
= parse_args(argc
, argv
)) < 0) {
4362 * child: setsid, close FD 0, 1, 2, chdir /
4363 * parent: exit (if fork is successful)
4371 * We are in the child. Make sure all other file
4372 * descriptors are closed, in case we are called with
4373 * more opened file descriptors than the standard ones.
4375 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4380 /* Create thread quit pipe */
4381 if ((ret
= init_thread_quit_pipe()) < 0) {
4385 /* Check if daemon is UID = 0 */
4386 is_root
= !getuid();
4389 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4391 /* Create global run dir with root access */
4392 ret
= create_lttng_rundir(rundir
);
4397 if (strlen(apps_unix_sock_path
) == 0) {
4398 snprintf(apps_unix_sock_path
, PATH_MAX
,
4399 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4402 if (strlen(client_unix_sock_path
) == 0) {
4403 snprintf(client_unix_sock_path
, PATH_MAX
,
4404 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4407 /* Set global SHM for ust */
4408 if (strlen(wait_shm_path
) == 0) {
4409 snprintf(wait_shm_path
, PATH_MAX
,
4410 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4413 if (strlen(health_unix_sock_path
) == 0) {
4414 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4415 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4418 /* Setup kernel consumerd path */
4419 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4420 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4421 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4422 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4424 DBG2("Kernel consumer err path: %s",
4425 kconsumer_data
.err_unix_sock_path
);
4426 DBG2("Kernel consumer cmd path: %s",
4427 kconsumer_data
.cmd_unix_sock_path
);
4429 home_path
= utils_get_home_dir();
4430 if (home_path
== NULL
) {
4431 /* TODO: Add --socket PATH option */
4432 ERR("Can't get HOME directory for sockets creation.");
4438 * Create rundir from home path. This will create something like
4441 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4447 ret
= create_lttng_rundir(rundir
);
4452 if (strlen(apps_unix_sock_path
) == 0) {
4453 snprintf(apps_unix_sock_path
, PATH_MAX
,
4454 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4457 /* Set the cli tool unix socket path */
4458 if (strlen(client_unix_sock_path
) == 0) {
4459 snprintf(client_unix_sock_path
, PATH_MAX
,
4460 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4463 /* Set global SHM for ust */
4464 if (strlen(wait_shm_path
) == 0) {
4465 snprintf(wait_shm_path
, PATH_MAX
,
4466 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4469 /* Set health check Unix path */
4470 if (strlen(health_unix_sock_path
) == 0) {
4471 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4472 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4476 /* Set consumer initial state */
4477 kernel_consumerd_state
= CONSUMER_STOPPED
;
4478 ust_consumerd_state
= CONSUMER_STOPPED
;
4480 DBG("Client socket path %s", client_unix_sock_path
);
4481 DBG("Application socket path %s", apps_unix_sock_path
);
4482 DBG("Application wait path %s", wait_shm_path
);
4483 DBG("LTTng run directory path: %s", rundir
);
4485 /* 32 bits consumerd path setup */
4486 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4487 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4488 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4489 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4491 DBG2("UST consumer 32 bits err path: %s",
4492 ustconsumer32_data
.err_unix_sock_path
);
4493 DBG2("UST consumer 32 bits cmd path: %s",
4494 ustconsumer32_data
.cmd_unix_sock_path
);
4496 /* 64 bits consumerd path setup */
4497 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4498 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4499 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4500 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4502 DBG2("UST consumer 64 bits err path: %s",
4503 ustconsumer64_data
.err_unix_sock_path
);
4504 DBG2("UST consumer 64 bits cmd path: %s",
4505 ustconsumer64_data
.cmd_unix_sock_path
);
4508 * See if daemon already exist.
4510 if ((ret
= check_existing_daemon()) < 0) {
4511 ERR("Already running daemon.\n");
4513 * We do not goto exit because we must not cleanup()
4514 * because a daemon is already running.
4520 * Init UST app hash table. Alloc hash table before this point since
4521 * cleanup() can get called after that point.
4525 /* After this point, we can safely call cleanup() with "goto exit" */
4528 * These actions must be executed as root. We do that *after* setting up
4529 * the sockets path because we MUST make the check for another daemon using
4530 * those paths *before* trying to set the kernel consumer sockets and init
4534 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4539 /* Setup kernel tracer */
4540 if (!opt_no_kernel
) {
4541 init_kernel_tracer();
4544 /* Set ulimit for open files */
4547 /* init lttng_fd tracking must be done after set_ulimit. */
4550 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4555 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4560 if ((ret
= set_signal_handler()) < 0) {
4564 /* Setup the needed unix socket */
4565 if ((ret
= init_daemon_socket()) < 0) {
4569 /* Set credentials to socket */
4570 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4574 /* Get parent pid if -S, --sig-parent is specified. */
4575 if (opt_sig_parent
) {
4579 /* Setup the kernel pipe for waking up the kernel thread */
4580 if (is_root
&& !opt_no_kernel
) {
4581 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4586 /* Setup the thread ht_cleanup communication pipe. */
4587 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4591 /* Setup the thread apps communication pipe. */
4592 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4596 /* Setup the thread apps notify communication pipe. */
4597 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4601 /* Initialize global buffer per UID and PID registry. */
4602 buffer_reg_init_uid_registry();
4603 buffer_reg_init_pid_registry();
4605 /* Init UST command queue. */
4606 cds_wfq_init(&ust_cmd_queue
.queue
);
4609 * Get session list pointer. This pointer MUST NOT be free(). This list is
4610 * statically declared in session.c
4612 session_list_ptr
= session_get_list();
4614 /* Set up max poll set size */
4615 lttng_poll_set_max_size();
4619 /* Check for the application socket timeout env variable. */
4620 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4621 if (env_app_timeout
) {
4622 app_socket_timeout
= atoi(env_app_timeout
);
4624 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4629 /* Create thread to manage the client socket */
4630 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4631 thread_ht_cleanup
, (void *) NULL
);
4633 PERROR("pthread_create ht_cleanup");
4634 goto exit_ht_cleanup
;
4637 /* Create thread to manage the client socket */
4638 ret
= pthread_create(&health_thread
, NULL
,
4639 thread_manage_health
, (void *) NULL
);
4641 PERROR("pthread_create health");
4645 /* Create thread to manage the client socket */
4646 ret
= pthread_create(&client_thread
, NULL
,
4647 thread_manage_clients
, (void *) NULL
);
4649 PERROR("pthread_create clients");
4653 /* Create thread to dispatch registration */
4654 ret
= pthread_create(&dispatch_thread
, NULL
,
4655 thread_dispatch_ust_registration
, (void *) NULL
);
4657 PERROR("pthread_create dispatch");
4661 /* Create thread to manage application registration. */
4662 ret
= pthread_create(®_apps_thread
, NULL
,
4663 thread_registration_apps
, (void *) NULL
);
4665 PERROR("pthread_create registration");
4669 /* Create thread to manage application socket */
4670 ret
= pthread_create(&apps_thread
, NULL
,
4671 thread_manage_apps
, (void *) NULL
);
4673 PERROR("pthread_create apps");
4677 /* Create thread to manage application notify socket */
4678 ret
= pthread_create(&apps_notify_thread
, NULL
,
4679 ust_thread_manage_notify
, (void *) NULL
);
4681 PERROR("pthread_create apps");
4685 /* Don't start this thread if kernel tracing is not requested nor root */
4686 if (is_root
&& !opt_no_kernel
) {
4687 /* Create kernel thread to manage kernel event */
4688 ret
= pthread_create(&kernel_thread
, NULL
,
4689 thread_manage_kernel
, (void *) NULL
);
4691 PERROR("pthread_create kernel");
4695 ret
= pthread_join(kernel_thread
, &status
);
4697 PERROR("pthread_join");
4698 goto error
; /* join error, exit without cleanup */
4703 ret
= pthread_join(apps_thread
, &status
);
4705 PERROR("pthread_join");
4706 goto error
; /* join error, exit without cleanup */
4710 ret
= pthread_join(reg_apps_thread
, &status
);
4712 PERROR("pthread_join");
4713 goto error
; /* join error, exit without cleanup */
4717 ret
= pthread_join(dispatch_thread
, &status
);
4719 PERROR("pthread_join");
4720 goto error
; /* join error, exit without cleanup */
4724 ret
= pthread_join(client_thread
, &status
);
4726 PERROR("pthread_join");
4727 goto error
; /* join error, exit without cleanup */
4730 ret
= join_consumer_thread(&kconsumer_data
);
4732 PERROR("join_consumer");
4733 goto error
; /* join error, exit without cleanup */
4736 ret
= join_consumer_thread(&ustconsumer32_data
);
4738 PERROR("join_consumer ust32");
4739 goto error
; /* join error, exit without cleanup */
4742 ret
= join_consumer_thread(&ustconsumer64_data
);
4744 PERROR("join_consumer ust64");
4745 goto error
; /* join error, exit without cleanup */
4749 ret
= pthread_join(health_thread
, &status
);
4751 PERROR("pthread_join health thread");
4752 goto error
; /* join error, exit without cleanup */
4756 ret
= pthread_join(ht_cleanup_thread
, &status
);
4758 PERROR("pthread_join ht cleanup thread");
4759 goto error
; /* join error, exit without cleanup */
4764 * cleanup() is called when no other thread is running.
4766 rcu_thread_online();
4768 rcu_thread_offline();
4769 rcu_unregister_thread();