2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
34 #include <sys/types.h>
36 #include <urcu/uatomic.h>
40 #include <common/common.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
48 #include "lttng-sessiond.h"
49 #include "buffer-registry.h"
56 #include "kernel-consumer.h"
60 #include "ust-consumer.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
67 #define CONSUMERD_FILE "lttng-consumerd"
70 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
73 const char *opt_tracing_group
;
74 static const char *opt_pidfile
;
75 static int opt_sig_parent
;
76 static int opt_verbose_consumer
;
77 static int opt_daemon
;
78 static int opt_no_kernel
;
79 static int is_root
; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid
; /* Parent PID for --sig-parent option */
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
87 static struct consumer_data kconsumer_data
= {
88 .type
= LTTNG_CONSUMER_KERNEL
,
89 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
90 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
93 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
94 .lock
= PTHREAD_MUTEX_INITIALIZER
,
95 .cond
= PTHREAD_COND_INITIALIZER
,
96 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
98 static struct consumer_data ustconsumer64_data
= {
99 .type
= LTTNG_CONSUMER64_UST
,
100 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
101 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
104 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
105 .lock
= PTHREAD_MUTEX_INITIALIZER
,
106 .cond
= PTHREAD_COND_INITIALIZER
,
107 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
109 static struct consumer_data ustconsumer32_data
= {
110 .type
= LTTNG_CONSUMER32_UST
,
111 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
112 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
115 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
116 .lock
= PTHREAD_MUTEX_INITIALIZER
,
117 .cond
= PTHREAD_COND_INITIALIZER
,
118 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
121 /* Shared between threads */
122 static int dispatch_thread_exit
;
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path
[PATH_MAX
];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path
[PATH_MAX
];
128 /* global wait shm path for UST */
129 static char wait_shm_path
[PATH_MAX
];
130 /* Global health check unix path */
131 static char health_unix_sock_path
[PATH_MAX
];
133 /* Sockets and FDs */
134 static int client_sock
= -1;
135 static int apps_sock
= -1;
136 int kernel_tracer_fd
= -1;
137 static int kernel_poll_pipe
[2] = { -1, -1 };
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
143 static int thread_quit_pipe
[2] = { -1, -1 };
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
149 static int apps_cmd_pipe
[2] = { -1, -1 };
151 int apps_cmd_notify_pipe
[2] = { -1, -1 };
153 /* Pthread, Mutexes and Semaphores */
154 static pthread_t apps_thread
;
155 static pthread_t apps_notify_thread
;
156 static pthread_t reg_apps_thread
;
157 static pthread_t client_thread
;
158 static pthread_t kernel_thread
;
159 static pthread_t dispatch_thread
;
160 static pthread_t health_thread
;
161 static pthread_t ht_cleanup_thread
;
164 * UST registration command queue. This queue is tied with a futex and uses a N
165 * wakers / 1 waiter implemented and detailed in futex.c/.h
167 * The thread_manage_apps and thread_dispatch_ust_registration interact with
168 * this queue and the wait/wake scheme.
170 static struct ust_cmd_queue ust_cmd_queue
;
173 * Pointer initialized before thread creation.
175 * This points to the tracing session list containing the session count and a
176 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
177 * MUST NOT be taken if you call a public function in session.c.
179 * The lock is nested inside the structure: session_list_ptr->lock. Please use
180 * session_lock_list and session_unlock_list for lock acquisition.
182 static struct ltt_session_list
*session_list_ptr
;
184 int ust_consumerd64_fd
= -1;
185 int ust_consumerd32_fd
= -1;
187 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
188 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
189 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
190 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
192 static const char *module_proc_lttng
= "/proc/lttng";
195 * Consumer daemon state which is changed when spawning it, killing it or in
196 * case of a fatal error.
198 enum consumerd_state
{
199 CONSUMER_STARTED
= 1,
200 CONSUMER_STOPPED
= 2,
205 * This consumer daemon state is used to validate if a client command will be
206 * able to reach the consumer. If not, the client is informed. For instance,
207 * doing a "lttng start" when the consumer state is set to ERROR will return an
208 * error to the client.
210 * The following example shows a possible race condition of this scheme:
212 * consumer thread error happens
214 * client cmd checks state -> still OK
215 * consumer thread exit, sets error
216 * client cmd try to talk to consumer
219 * However, since the consumer is a different daemon, we have no way of making
220 * sure the command will reach it safely even with this state flag. This is why
221 * we consider that up to the state validation during command processing, the
222 * command is safe. After that, we can not guarantee the correctness of the
223 * client request vis-a-vis the consumer.
225 static enum consumerd_state ust_consumerd_state
;
226 static enum consumerd_state kernel_consumerd_state
;
229 * Socket timeout for receiving and sending in seconds.
231 static int app_socket_timeout
;
233 /* Set in main() with the current page size. */
237 void setup_consumerd_path(void)
239 const char *bin
, *libdir
;
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin
[0]) {
248 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
250 if (!consumerd32_libdir
[0]) {
251 consumerd32_libdir
= INSTALL_LIB_PATH
;
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin
[0]) {
255 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
257 if (!consumerd64_libdir
[0]) {
258 consumerd64_libdir
= INSTALL_LIB_PATH
;
261 #error "Unknown bitness"
265 * runtime env. var. overrides the build default.
267 bin
= getenv("LTTNG_CONSUMERD32_BIN");
269 consumerd32_bin
= bin
;
271 bin
= getenv("LTTNG_CONSUMERD64_BIN");
273 consumerd64_bin
= bin
;
275 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
277 consumerd32_libdir
= libdir
;
279 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
281 consumerd64_libdir
= libdir
;
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
288 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
294 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
300 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
312 * Check if the thread quit pipe was triggered.
314 * Return 1 if it was triggered else 0;
316 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
318 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
326 * Return group ID of the tracing group or -1 if not found.
328 static gid_t
allowed_group(void)
332 if (opt_tracing_group
) {
333 grp
= getgrnam(opt_tracing_group
);
335 grp
= getgrnam(default_tracing_group
);
345 * Init thread quit pipe.
347 * Return -1 on error or 0 if all pipes are created.
349 static int init_thread_quit_pipe(void)
353 ret
= pipe(thread_quit_pipe
);
355 PERROR("thread quit pipe");
359 for (i
= 0; i
< 2; i
++) {
360 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
372 * Stop all threads by closing the thread quit pipe.
374 static void stop_threads(void)
378 /* Stopping all threads */
379 DBG("Terminating all threads");
380 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
382 ERR("write error on thread quit pipe");
385 /* Dispatch thread */
386 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
387 futex_nto1_wake(&ust_cmd_queue
.futex
);
391 * Close every consumer sockets.
393 static void close_consumer_sockets(void)
397 if (kconsumer_data
.err_sock
>= 0) {
398 ret
= close(kconsumer_data
.err_sock
);
400 PERROR("kernel consumer err_sock close");
403 if (ustconsumer32_data
.err_sock
>= 0) {
404 ret
= close(ustconsumer32_data
.err_sock
);
406 PERROR("UST consumerd32 err_sock close");
409 if (ustconsumer64_data
.err_sock
>= 0) {
410 ret
= close(ustconsumer64_data
.err_sock
);
412 PERROR("UST consumerd64 err_sock close");
415 if (kconsumer_data
.cmd_sock
>= 0) {
416 ret
= close(kconsumer_data
.cmd_sock
);
418 PERROR("kernel consumer cmd_sock close");
421 if (ustconsumer32_data
.cmd_sock
>= 0) {
422 ret
= close(ustconsumer32_data
.cmd_sock
);
424 PERROR("UST consumerd32 cmd_sock close");
427 if (ustconsumer64_data
.cmd_sock
>= 0) {
428 ret
= close(ustconsumer64_data
.cmd_sock
);
430 PERROR("UST consumerd64 cmd_sock close");
438 static void cleanup(void)
442 struct ltt_session
*sess
, *stmp
;
447 * Close the thread quit pipe. It has already done its job,
448 * since we are now called.
450 utils_close_pipe(thread_quit_pipe
);
453 * If opt_pidfile is undefined, the default file will be wiped when
454 * removing the rundir.
457 ret
= remove(opt_pidfile
);
459 PERROR("remove pidfile %s", opt_pidfile
);
463 DBG("Removing %s directory", rundir
);
464 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
466 ERR("asprintf failed. Something is really wrong!");
469 /* Remove lttng run directory */
472 ERR("Unable to clean %s", rundir
);
477 DBG("Cleaning up all sessions");
479 /* Destroy session list mutex */
480 if (session_list_ptr
!= NULL
) {
481 pthread_mutex_destroy(&session_list_ptr
->lock
);
483 /* Cleanup ALL session */
484 cds_list_for_each_entry_safe(sess
, stmp
,
485 &session_list_ptr
->head
, list
) {
486 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
490 DBG("Closing all UST sockets");
491 ust_app_clean_list();
492 buffer_reg_destroy_registries();
494 if (is_root
&& !opt_no_kernel
) {
495 DBG2("Closing kernel fd");
496 if (kernel_tracer_fd
>= 0) {
497 ret
= close(kernel_tracer_fd
);
502 DBG("Unloading kernel modules");
503 modprobe_remove_lttng_all();
506 close_consumer_sockets();
509 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
510 "Matthew, BEET driven development works!%c[%dm",
511 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
516 * Send data on a unix socket using the liblttsessiondcomm API.
518 * Return lttcomm error code.
520 static int send_unix_sock(int sock
, void *buf
, size_t len
)
522 /* Check valid length */
527 return lttcomm_send_unix_sock(sock
, buf
, len
);
531 * Free memory of a command context structure.
533 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
535 DBG("Clean command context structure");
537 if ((*cmd_ctx
)->llm
) {
538 free((*cmd_ctx
)->llm
);
540 if ((*cmd_ctx
)->lsm
) {
541 free((*cmd_ctx
)->lsm
);
549 * Notify UST applications using the shm mmap futex.
551 static int notify_ust_apps(int active
)
555 DBG("Notifying applications of session daemon state: %d", active
);
557 /* See shm.c for this call implying mmap, shm and futex calls */
558 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
559 if (wait_shm_mmap
== NULL
) {
563 /* Wake waiting process */
564 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
566 /* Apps notified successfully */
574 * Setup the outgoing data buffer for the response (llm) by allocating the
575 * right amount of memory and copying the original information from the lsm
578 * Return total size of the buffer pointed by buf.
580 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
586 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
587 if (cmd_ctx
->llm
== NULL
) {
593 /* Copy common data */
594 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
595 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
597 cmd_ctx
->llm
->data_size
= size
;
598 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
607 * Update the kernel poll set of all channel fd available over all tracing
608 * session. Add the wakeup pipe at the end of the set.
610 static int update_kernel_poll(struct lttng_poll_event
*events
)
613 struct ltt_session
*session
;
614 struct ltt_kernel_channel
*channel
;
616 DBG("Updating kernel poll set");
619 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
620 session_lock(session
);
621 if (session
->kernel_session
== NULL
) {
622 session_unlock(session
);
626 cds_list_for_each_entry(channel
,
627 &session
->kernel_session
->channel_list
.head
, list
) {
628 /* Add channel fd to the kernel poll set */
629 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
631 session_unlock(session
);
634 DBG("Channel fd %d added to kernel set", channel
->fd
);
636 session_unlock(session
);
638 session_unlock_list();
643 session_unlock_list();
648 * Find the channel fd from 'fd' over all tracing session. When found, check
649 * for new channel stream and send those stream fds to the kernel consumer.
651 * Useful for CPU hotplug feature.
653 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
656 struct ltt_session
*session
;
657 struct ltt_kernel_session
*ksess
;
658 struct ltt_kernel_channel
*channel
;
660 DBG("Updating kernel streams for channel fd %d", fd
);
663 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
664 session_lock(session
);
665 if (session
->kernel_session
== NULL
) {
666 session_unlock(session
);
669 ksess
= session
->kernel_session
;
671 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
672 if (channel
->fd
== fd
) {
673 DBG("Channel found, updating kernel streams");
674 ret
= kernel_open_channel_stream(channel
);
678 /* Update the stream global counter */
679 ksess
->stream_count_global
+= ret
;
682 * Have we already sent fds to the consumer? If yes, it means
683 * that tracing is started so it is safe to send our updated
686 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
687 struct lttng_ht_iter iter
;
688 struct consumer_socket
*socket
;
691 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
692 &iter
.iter
, socket
, node
.node
) {
693 /* Code flow error */
696 pthread_mutex_lock(socket
->lock
);
697 ret
= kernel_consumer_send_channel_stream(socket
,
699 session
->output_traces
? 1 : 0);
700 pthread_mutex_unlock(socket
->lock
);
711 session_unlock(session
);
713 session_unlock_list();
717 session_unlock(session
);
718 session_unlock_list();
723 * For each tracing session, update newly registered apps. The session list
724 * lock MUST be acquired before calling this.
726 static void update_ust_app(int app_sock
)
728 struct ltt_session
*sess
, *stmp
;
730 /* Consumer is in an ERROR state. Stop any application update. */
731 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
732 /* Stop the update process since the consumer is dead. */
736 /* For all tracing session(s) */
737 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
739 if (sess
->ust_session
) {
740 ust_app_global_update(sess
->ust_session
, app_sock
);
742 session_unlock(sess
);
747 * This thread manage event coming from the kernel.
749 * Features supported in this thread:
752 static void *thread_manage_kernel(void *data
)
754 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
755 uint32_t revents
, nb_fd
;
757 struct lttng_poll_event events
;
759 DBG("[thread] Thread manage kernel started");
761 health_register(HEALTH_TYPE_KERNEL
);
764 * This first step of the while is to clean this structure which could free
765 * non NULL pointers so initialize it before the loop.
767 lttng_poll_init(&events
);
769 if (testpoint(thread_manage_kernel
)) {
770 goto error_testpoint
;
773 health_code_update();
775 if (testpoint(thread_manage_kernel_before_loop
)) {
776 goto error_testpoint
;
780 health_code_update();
782 if (update_poll_flag
== 1) {
783 /* Clean events object. We are about to populate it again. */
784 lttng_poll_clean(&events
);
786 ret
= sessiond_set_thread_pollset(&events
, 2);
788 goto error_poll_create
;
791 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
796 /* This will add the available kernel channel if any. */
797 ret
= update_kernel_poll(&events
);
801 update_poll_flag
= 0;
804 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
806 /* Poll infinite value of time */
809 ret
= lttng_poll_wait(&events
, -1);
813 * Restart interrupted system call.
815 if (errno
== EINTR
) {
819 } else if (ret
== 0) {
820 /* Should not happen since timeout is infinite */
821 ERR("Return value of poll is 0 with an infinite timeout.\n"
822 "This should not have happened! Continuing...");
828 for (i
= 0; i
< nb_fd
; i
++) {
829 /* Fetch once the poll data */
830 revents
= LTTNG_POLL_GETEV(&events
, i
);
831 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
833 health_code_update();
835 /* Thread quit pipe has been closed. Killing thread. */
836 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
842 /* Check for data on kernel pipe */
843 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
845 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
846 } while (ret
< 0 && errno
== EINTR
);
848 * Ret value is useless here, if this pipe gets any actions an
849 * update is required anyway.
851 update_poll_flag
= 1;
855 * New CPU detected by the kernel. Adding kernel stream to
856 * kernel session and updating the kernel consumer
858 if (revents
& LPOLLIN
) {
859 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
865 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
866 * and unregister kernel stream at this point.
875 lttng_poll_clean(&events
);
878 utils_close_pipe(kernel_poll_pipe
);
879 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
882 ERR("Health error occurred in %s", __func__
);
883 WARN("Kernel thread died unexpectedly. "
884 "Kernel tracing can continue but CPU hotplug is disabled.");
887 DBG("Kernel thread dying");
892 * Signal pthread condition of the consumer data that the thread.
894 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
896 pthread_mutex_lock(&data
->cond_mutex
);
899 * The state is set before signaling. It can be any value, it's the waiter
900 * job to correctly interpret this condition variable associated to the
901 * consumer pthread_cond.
903 * A value of 0 means that the corresponding thread of the consumer data
904 * was not started. 1 indicates that the thread has started and is ready
905 * for action. A negative value means that there was an error during the
908 data
->consumer_thread_is_ready
= state
;
909 (void) pthread_cond_signal(&data
->cond
);
911 pthread_mutex_unlock(&data
->cond_mutex
);
915 * This thread manage the consumer error sent back to the session daemon.
917 static void *thread_manage_consumer(void *data
)
919 int sock
= -1, i
, ret
, pollfd
, err
= -1;
920 uint32_t revents
, nb_fd
;
921 enum lttcomm_return_code code
;
922 struct lttng_poll_event events
;
923 struct consumer_data
*consumer_data
= data
;
925 DBG("[thread] Manage consumer started");
927 health_register(HEALTH_TYPE_CONSUMER
);
929 health_code_update();
932 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
933 * metadata_sock. Nothing more will be added to this poll set.
935 ret
= sessiond_set_thread_pollset(&events
, 3);
941 * The error socket here is already in a listening state which was done
942 * just before spawning this thread to avoid a race between the consumer
943 * daemon exec trying to connect and the listen() call.
945 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
950 health_code_update();
952 /* Infinite blocking call, waiting for transmission */
956 if (testpoint(thread_manage_consumer
)) {
960 ret
= lttng_poll_wait(&events
, -1);
964 * Restart interrupted system call.
966 if (errno
== EINTR
) {
974 for (i
= 0; i
< nb_fd
; i
++) {
975 /* Fetch once the poll data */
976 revents
= LTTNG_POLL_GETEV(&events
, i
);
977 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
979 health_code_update();
981 /* Thread quit pipe has been closed. Killing thread. */
982 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
988 /* Event on the registration socket */
989 if (pollfd
== consumer_data
->err_sock
) {
990 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
991 ERR("consumer err socket poll error");
997 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1003 * Set the CLOEXEC flag. Return code is useless because either way, the
1006 (void) utils_set_fd_cloexec(sock
);
1008 health_code_update();
1010 DBG2("Receiving code from consumer err_sock");
1012 /* Getting status code from kconsumerd */
1013 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1014 sizeof(enum lttcomm_return_code
));
1019 health_code_update();
1021 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1022 /* Connect both socket, command and metadata. */
1023 consumer_data
->cmd_sock
=
1024 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1025 consumer_data
->metadata_fd
=
1026 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1027 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
1028 PERROR("consumer connect cmd socket");
1029 /* On error, signal condition and quit. */
1030 signal_consumer_condition(consumer_data
, -1);
1033 consumer_data
->metadata_sock
.fd
= &consumer_data
->metadata_fd
;
1034 /* Create metadata socket lock. */
1035 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1036 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1037 PERROR("zmalloc pthread mutex");
1041 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1043 signal_consumer_condition(consumer_data
, 1);
1044 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1045 DBG("Consumer metadata socket ready (fd: %d)",
1046 consumer_data
->metadata_fd
);
1048 ERR("consumer error when waiting for SOCK_READY : %s",
1049 lttcomm_get_readable_code(-code
));
1053 /* Remove the consumerd error sock since we've established a connexion */
1054 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1059 /* Add new accepted error socket. */
1060 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1065 /* Add metadata socket that is successfully connected. */
1066 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1067 LPOLLIN
| LPOLLRDHUP
);
1072 health_code_update();
1074 /* Infinite blocking call, waiting for transmission */
1077 health_poll_entry();
1078 ret
= lttng_poll_wait(&events
, -1);
1082 * Restart interrupted system call.
1084 if (errno
== EINTR
) {
1092 for (i
= 0; i
< nb_fd
; i
++) {
1093 /* Fetch once the poll data */
1094 revents
= LTTNG_POLL_GETEV(&events
, i
);
1095 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1097 health_code_update();
1099 /* Thread quit pipe has been closed. Killing thread. */
1100 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1106 if (pollfd
== sock
) {
1107 /* Event on the consumerd socket */
1108 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1109 ERR("consumer err socket second poll error");
1112 health_code_update();
1113 /* Wait for any kconsumerd error */
1114 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1115 sizeof(enum lttcomm_return_code
));
1117 ERR("consumer closed the command socket");
1121 ERR("consumer return code : %s",
1122 lttcomm_get_readable_code(-code
));
1125 } else if (pollfd
== consumer_data
->metadata_fd
) {
1126 /* UST metadata requests */
1127 ret
= ust_consumer_metadata_request(
1128 &consumer_data
->metadata_sock
);
1130 ERR("Handling metadata request");
1135 ERR("Unknown pollfd");
1139 health_code_update();
1145 * We lock here because we are about to close the sockets and some other
1146 * thread might be using them so wait before we are exclusive which will
1147 * abort all other consumer command by other threads.
1149 pthread_mutex_lock(&consumer_data
->lock
);
1151 /* Immediately set the consumerd state to stopped */
1152 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1153 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1154 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1155 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1156 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1158 /* Code flow error... */
1162 if (consumer_data
->err_sock
>= 0) {
1163 ret
= close(consumer_data
->err_sock
);
1167 consumer_data
->err_sock
= -1;
1169 if (consumer_data
->cmd_sock
>= 0) {
1170 ret
= close(consumer_data
->cmd_sock
);
1174 consumer_data
->cmd_sock
= -1;
1176 if (*consumer_data
->metadata_sock
.fd
>= 0) {
1177 ret
= close(*consumer_data
->metadata_sock
.fd
);
1190 unlink(consumer_data
->err_unix_sock_path
);
1191 unlink(consumer_data
->cmd_unix_sock_path
);
1192 consumer_data
->pid
= 0;
1193 pthread_mutex_unlock(&consumer_data
->lock
);
1194 /* Cleanup metadata socket mutex. */
1195 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1196 free(consumer_data
->metadata_sock
.lock
);
1198 lttng_poll_clean(&events
);
1202 ERR("Health error occurred in %s", __func__
);
1204 health_unregister();
1205 DBG("consumer thread cleanup completed");
1211 * This thread manage application communication.
1213 static void *thread_manage_apps(void *data
)
1215 int i
, ret
, pollfd
, err
= -1;
1216 uint32_t revents
, nb_fd
;
1217 struct lttng_poll_event events
;
1219 DBG("[thread] Manage application started");
1221 rcu_register_thread();
1222 rcu_thread_online();
1224 health_register(HEALTH_TYPE_APP_MANAGE
);
1226 if (testpoint(thread_manage_apps
)) {
1227 goto error_testpoint
;
1230 health_code_update();
1232 ret
= sessiond_set_thread_pollset(&events
, 2);
1234 goto error_poll_create
;
1237 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1242 if (testpoint(thread_manage_apps_before_loop
)) {
1246 health_code_update();
1249 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1251 /* Inifinite blocking call, waiting for transmission */
1253 health_poll_entry();
1254 ret
= lttng_poll_wait(&events
, -1);
1258 * Restart interrupted system call.
1260 if (errno
== EINTR
) {
1268 for (i
= 0; i
< nb_fd
; i
++) {
1269 /* Fetch once the poll data */
1270 revents
= LTTNG_POLL_GETEV(&events
, i
);
1271 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1273 health_code_update();
1275 /* Thread quit pipe has been closed. Killing thread. */
1276 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1282 /* Inspect the apps cmd pipe */
1283 if (pollfd
== apps_cmd_pipe
[0]) {
1284 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1285 ERR("Apps command pipe error");
1287 } else if (revents
& LPOLLIN
) {
1292 ret
= read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1293 } while (ret
< 0 && errno
== EINTR
);
1294 if (ret
< 0 || ret
< sizeof(sock
)) {
1295 PERROR("read apps cmd pipe");
1299 health_code_update();
1302 * We only monitor the error events of the socket. This
1303 * thread does not handle any incoming data from UST
1306 ret
= lttng_poll_add(&events
, sock
,
1307 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1313 * Set socket timeout for both receiving and ending.
1314 * app_socket_timeout is in seconds, whereas
1315 * lttcomm_setsockopt_rcv_timeout and
1316 * lttcomm_setsockopt_snd_timeout expect msec as
1319 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1320 app_socket_timeout
* 1000);
1321 (void) lttcomm_setsockopt_snd_timeout(sock
,
1322 app_socket_timeout
* 1000);
1324 DBG("Apps with sock %d added to poll set", sock
);
1326 health_code_update();
1332 * At this point, we know that a registered application made
1333 * the event at poll_wait.
1335 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1336 /* Removing from the poll set */
1337 ret
= lttng_poll_del(&events
, pollfd
);
1342 /* Socket closed on remote end. */
1343 ust_app_unregister(pollfd
);
1348 health_code_update();
1354 lttng_poll_clean(&events
);
1357 utils_close_pipe(apps_cmd_pipe
);
1358 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1361 * We don't clean the UST app hash table here since already registered
1362 * applications can still be controlled so let them be until the session
1363 * daemon dies or the applications stop.
1368 ERR("Health error occurred in %s", __func__
);
1370 health_unregister();
1371 DBG("Application communication apps thread cleanup complete");
1372 rcu_thread_offline();
1373 rcu_unregister_thread();
1378 * Send a socket to a thread This is called from the dispatch UST registration
1379 * thread once all sockets are set for the application.
1381 * On success, return 0 else a negative value being the errno message of the
1384 static int send_socket_to_thread(int fd
, int sock
)
1388 /* Sockets MUST be set or else this should not have been called. */
1393 ret
= write(fd
, &sock
, sizeof(sock
));
1394 } while (ret
< 0 && errno
== EINTR
);
1395 if (ret
< 0 || ret
!= sizeof(sock
)) {
1396 PERROR("write apps pipe %d", fd
);
1403 /* All good. Don't send back the write positive ret value. */
1410 * Sanitize the wait queue of the dispatch registration thread meaning removing
1411 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1412 * notify socket is never received.
1414 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1416 int ret
, nb_fd
= 0, i
;
1417 unsigned int fd_added
= 0;
1418 struct lttng_poll_event events
;
1419 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1423 lttng_poll_init(&events
);
1425 /* Just skip everything for an empty queue. */
1426 if (!wait_queue
->count
) {
1430 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1435 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1436 &wait_queue
->head
, head
) {
1437 assert(wait_node
->app
);
1438 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1439 LPOLLHUP
| LPOLLERR
);
1452 * Poll but don't block so we can quickly identify the faulty events and
1453 * clean them afterwards from the wait queue.
1455 ret
= lttng_poll_wait(&events
, 0);
1461 for (i
= 0; i
< nb_fd
; i
++) {
1462 /* Get faulty FD. */
1463 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1464 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1466 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1467 &wait_queue
->head
, head
) {
1468 if (pollfd
== wait_node
->app
->sock
&&
1469 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1470 cds_list_del(&wait_node
->head
);
1471 wait_queue
->count
--;
1472 ust_app_destroy(wait_node
->app
);
1480 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1484 lttng_poll_clean(&events
);
1488 lttng_poll_clean(&events
);
1490 ERR("Unable to sanitize wait queue");
1495 * Dispatch request from the registration threads to the application
1496 * communication thread.
1498 static void *thread_dispatch_ust_registration(void *data
)
1501 struct cds_wfq_node
*node
;
1502 struct ust_command
*ust_cmd
= NULL
;
1503 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1504 struct ust_reg_wait_queue wait_queue
= {
1508 health_register(HEALTH_TYPE_APP_REG_DISPATCH
);
1510 health_code_update();
1512 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1514 DBG("[thread] Dispatch UST command started");
1516 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1517 health_code_update();
1519 /* Atomically prepare the queue futex */
1520 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1523 struct ust_app
*app
= NULL
;
1527 * Make sure we don't have node(s) that have hung up before receiving
1528 * the notify socket. This is to clean the list in order to avoid
1529 * memory leaks from notify socket that are never seen.
1531 sanitize_wait_queue(&wait_queue
);
1533 health_code_update();
1534 /* Dequeue command for registration */
1535 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1537 DBG("Woken up but nothing in the UST command queue");
1538 /* Continue thread execution */
1542 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1544 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1545 " gid:%d sock:%d name:%s (version %d.%d)",
1546 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1547 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1548 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1549 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1551 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1552 wait_node
= zmalloc(sizeof(*wait_node
));
1554 PERROR("zmalloc wait_node dispatch");
1555 ret
= close(ust_cmd
->sock
);
1557 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1559 lttng_fd_put(1, LTTNG_FD_APPS
);
1563 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1565 /* Create application object if socket is CMD. */
1566 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1568 if (!wait_node
->app
) {
1569 ret
= close(ust_cmd
->sock
);
1571 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1573 lttng_fd_put(1, LTTNG_FD_APPS
);
1579 * Add application to the wait queue so we can set the notify
1580 * socket before putting this object in the global ht.
1582 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1587 * We have to continue here since we don't have the notify
1588 * socket and the application MUST be added to the hash table
1589 * only at that moment.
1594 * Look for the application in the local wait queue and set the
1595 * notify socket if found.
1597 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1598 &wait_queue
.head
, head
) {
1599 health_code_update();
1600 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1601 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1602 cds_list_del(&wait_node
->head
);
1604 app
= wait_node
->app
;
1606 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1612 * With no application at this stage the received socket is
1613 * basically useless so close it before we free the cmd data
1614 * structure for good.
1617 ret
= close(ust_cmd
->sock
);
1619 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1621 lttng_fd_put(1, LTTNG_FD_APPS
);
1628 * @session_lock_list
1630 * Lock the global session list so from the register up to the
1631 * registration done message, no thread can see the application
1632 * and change its state.
1634 session_lock_list();
1638 * Add application to the global hash table. This needs to be
1639 * done before the update to the UST registry can locate the
1644 /* Set app version. This call will print an error if needed. */
1645 (void) ust_app_version(app
);
1647 /* Send notify socket through the notify pipe. */
1648 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1652 session_unlock_list();
1653 /* No notify thread, stop the UST tracing. */
1658 * Update newly registered application with the tracing
1659 * registry info already enabled information.
1661 update_ust_app(app
->sock
);
1664 * Don't care about return value. Let the manage apps threads
1665 * handle app unregistration upon socket close.
1667 (void) ust_app_register_done(app
->sock
);
1670 * Even if the application socket has been closed, send the app
1671 * to the thread and unregistration will take place at that
1674 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1677 session_unlock_list();
1678 /* No apps. thread, stop the UST tracing. */
1683 session_unlock_list();
1685 } while (node
!= NULL
);
1687 health_poll_entry();
1688 /* Futex wait on queue. Blocking call on futex() */
1689 futex_nto1_wait(&ust_cmd_queue
.futex
);
1692 /* Normal exit, no error */
1696 /* Clean up wait queue. */
1697 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1698 &wait_queue
.head
, head
) {
1699 cds_list_del(&wait_node
->head
);
1704 DBG("Dispatch thread dying");
1707 ERR("Health error occurred in %s", __func__
);
1709 health_unregister();
1714 * This thread manage application registration.
1716 static void *thread_registration_apps(void *data
)
1718 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1719 uint32_t revents
, nb_fd
;
1720 struct lttng_poll_event events
;
1722 * Get allocated in this thread, enqueued to a global queue, dequeued and
1723 * freed in the manage apps thread.
1725 struct ust_command
*ust_cmd
= NULL
;
1727 DBG("[thread] Manage application registration started");
1729 health_register(HEALTH_TYPE_APP_REG
);
1731 if (testpoint(thread_registration_apps
)) {
1732 goto error_testpoint
;
1735 ret
= lttcomm_listen_unix_sock(apps_sock
);
1741 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1742 * more will be added to this poll set.
1744 ret
= sessiond_set_thread_pollset(&events
, 2);
1746 goto error_create_poll
;
1749 /* Add the application registration socket */
1750 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1752 goto error_poll_add
;
1755 /* Notify all applications to register */
1756 ret
= notify_ust_apps(1);
1758 ERR("Failed to notify applications or create the wait shared memory.\n"
1759 "Execution continues but there might be problem for already\n"
1760 "running applications that wishes to register.");
1764 DBG("Accepting application registration");
1766 /* Inifinite blocking call, waiting for transmission */
1768 health_poll_entry();
1769 ret
= lttng_poll_wait(&events
, -1);
1773 * Restart interrupted system call.
1775 if (errno
== EINTR
) {
1783 for (i
= 0; i
< nb_fd
; i
++) {
1784 health_code_update();
1786 /* Fetch once the poll data */
1787 revents
= LTTNG_POLL_GETEV(&events
, i
);
1788 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1790 /* Thread quit pipe has been closed. Killing thread. */
1791 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1797 /* Event on the registration socket */
1798 if (pollfd
== apps_sock
) {
1799 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1800 ERR("Register apps socket poll error");
1802 } else if (revents
& LPOLLIN
) {
1803 sock
= lttcomm_accept_unix_sock(apps_sock
);
1809 * Set the CLOEXEC flag. Return code is useless because
1810 * either way, the show must go on.
1812 (void) utils_set_fd_cloexec(sock
);
1814 /* Create UST registration command for enqueuing */
1815 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1816 if (ust_cmd
== NULL
) {
1817 PERROR("ust command zmalloc");
1822 * Using message-based transmissions to ensure we don't
1823 * have to deal with partially received messages.
1825 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1827 ERR("Exhausted file descriptors allowed for applications.");
1837 health_code_update();
1838 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1841 /* Close socket of the application. */
1846 lttng_fd_put(LTTNG_FD_APPS
, 1);
1850 health_code_update();
1852 ust_cmd
->sock
= sock
;
1855 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1856 " gid:%d sock:%d name:%s (version %d.%d)",
1857 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1858 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1859 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1860 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1863 * Lock free enqueue the registration request. The red pill
1864 * has been taken! This apps will be part of the *system*.
1866 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1869 * Wake the registration queue futex. Implicit memory
1870 * barrier with the exchange in cds_wfq_enqueue.
1872 futex_nto1_wake(&ust_cmd_queue
.futex
);
1882 ERR("Health error occurred in %s", __func__
);
1885 /* Notify that the registration thread is gone */
1888 if (apps_sock
>= 0) {
1889 ret
= close(apps_sock
);
1899 lttng_fd_put(LTTNG_FD_APPS
, 1);
1901 unlink(apps_unix_sock_path
);
1904 lttng_poll_clean(&events
);
1908 DBG("UST Registration thread cleanup complete");
1909 health_unregister();
1915 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1916 * exec or it will fails.
1918 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1921 struct timespec timeout
;
1923 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1924 consumer_data
->consumer_thread_is_ready
= 0;
1926 /* Setup pthread condition */
1927 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1930 PERROR("pthread_condattr_init consumer data");
1935 * Set the monotonic clock in order to make sure we DO NOT jump in time
1936 * between the clock_gettime() call and the timedwait call. See bug #324
1937 * for a more details and how we noticed it.
1939 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1942 PERROR("pthread_condattr_setclock consumer data");
1946 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1949 PERROR("pthread_cond_init consumer data");
1953 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1956 PERROR("pthread_create consumer");
1961 /* We are about to wait on a pthread condition */
1962 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1964 /* Get time for sem_timedwait absolute timeout */
1965 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1967 * Set the timeout for the condition timed wait even if the clock gettime
1968 * call fails since we might loop on that call and we want to avoid to
1969 * increment the timeout too many times.
1971 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1974 * The following loop COULD be skipped in some conditions so this is why we
1975 * set ret to 0 in order to make sure at least one round of the loop is
1981 * Loop until the condition is reached or when a timeout is reached. Note
1982 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1983 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1984 * possible. This loop does not take any chances and works with both of
1987 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1988 if (clock_ret
< 0) {
1989 PERROR("clock_gettime spawn consumer");
1990 /* Infinite wait for the consumerd thread to be ready */
1991 ret
= pthread_cond_wait(&consumer_data
->cond
,
1992 &consumer_data
->cond_mutex
);
1994 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1995 &consumer_data
->cond_mutex
, &timeout
);
1999 /* Release the pthread condition */
2000 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2004 if (ret
== ETIMEDOUT
) {
2006 * Call has timed out so we kill the kconsumerd_thread and return
2009 ERR("Condition timed out. The consumer thread was never ready."
2011 ret
= pthread_cancel(consumer_data
->thread
);
2013 PERROR("pthread_cancel consumer thread");
2016 PERROR("pthread_cond_wait failed consumer thread");
2021 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2022 if (consumer_data
->pid
== 0) {
2023 ERR("Consumerd did not start");
2024 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2027 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2036 * Join consumer thread
2038 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2042 /* Consumer pid must be a real one. */
2043 if (consumer_data
->pid
> 0) {
2045 ret
= kill(consumer_data
->pid
, SIGTERM
);
2047 ERR("Error killing consumer daemon");
2050 return pthread_join(consumer_data
->thread
, &status
);
2057 * Fork and exec a consumer daemon (consumerd).
2059 * Return pid if successful else -1.
2061 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2065 const char *consumer_to_use
;
2066 const char *verbosity
;
2069 DBG("Spawning consumerd");
2076 if (opt_verbose_consumer
) {
2077 verbosity
= "--verbose";
2079 verbosity
= "--quiet";
2081 switch (consumer_data
->type
) {
2082 case LTTNG_CONSUMER_KERNEL
:
2084 * Find out which consumerd to execute. We will first try the
2085 * 64-bit path, then the sessiond's installation directory, and
2086 * fallback on the 32-bit one,
2088 DBG3("Looking for a kernel consumer at these locations:");
2089 DBG3(" 1) %s", consumerd64_bin
);
2090 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2091 DBG3(" 3) %s", consumerd32_bin
);
2092 if (stat(consumerd64_bin
, &st
) == 0) {
2093 DBG3("Found location #1");
2094 consumer_to_use
= consumerd64_bin
;
2095 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2096 DBG3("Found location #2");
2097 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2098 } else if (stat(consumerd32_bin
, &st
) == 0) {
2099 DBG3("Found location #3");
2100 consumer_to_use
= consumerd32_bin
;
2102 DBG("Could not find any valid consumerd executable");
2105 DBG("Using kernel consumer at: %s", consumer_to_use
);
2106 execl(consumer_to_use
,
2107 "lttng-consumerd", verbosity
, "-k",
2108 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2109 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2112 case LTTNG_CONSUMER64_UST
:
2114 char *tmpnew
= NULL
;
2116 if (consumerd64_libdir
[0] != '\0') {
2120 tmp
= getenv("LD_LIBRARY_PATH");
2124 tmplen
= strlen("LD_LIBRARY_PATH=")
2125 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2126 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2131 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2132 strcat(tmpnew
, consumerd64_libdir
);
2133 if (tmp
[0] != '\0') {
2134 strcat(tmpnew
, ":");
2135 strcat(tmpnew
, tmp
);
2137 ret
= putenv(tmpnew
);
2144 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2145 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2146 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2147 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2149 if (consumerd64_libdir
[0] != '\0') {
2157 case LTTNG_CONSUMER32_UST
:
2159 char *tmpnew
= NULL
;
2161 if (consumerd32_libdir
[0] != '\0') {
2165 tmp
= getenv("LD_LIBRARY_PATH");
2169 tmplen
= strlen("LD_LIBRARY_PATH=")
2170 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2171 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2176 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2177 strcat(tmpnew
, consumerd32_libdir
);
2178 if (tmp
[0] != '\0') {
2179 strcat(tmpnew
, ":");
2180 strcat(tmpnew
, tmp
);
2182 ret
= putenv(tmpnew
);
2189 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2190 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2191 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2192 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2194 if (consumerd32_libdir
[0] != '\0') {
2203 PERROR("unknown consumer type");
2207 PERROR("kernel start consumer exec");
2210 } else if (pid
> 0) {
2213 PERROR("start consumer fork");
2221 * Spawn the consumerd daemon and session daemon thread.
2223 static int start_consumerd(struct consumer_data
*consumer_data
)
2228 * Set the listen() state on the socket since there is a possible race
2229 * between the exec() of the consumer daemon and this call if place in the
2230 * consumer thread. See bug #366 for more details.
2232 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2237 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2238 if (consumer_data
->pid
!= 0) {
2239 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2243 ret
= spawn_consumerd(consumer_data
);
2245 ERR("Spawning consumerd failed");
2246 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2250 /* Setting up the consumer_data pid */
2251 consumer_data
->pid
= ret
;
2252 DBG2("Consumer pid %d", consumer_data
->pid
);
2253 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2255 DBG2("Spawning consumer control thread");
2256 ret
= spawn_consumer_thread(consumer_data
);
2258 ERR("Fatal error spawning consumer control thread");
2266 /* Cleanup already created sockets on error. */
2267 if (consumer_data
->err_sock
>= 0) {
2270 err
= close(consumer_data
->err_sock
);
2272 PERROR("close consumer data error socket");
2279 * Compute health status of each consumer. If one of them is zero (bad
2280 * state), we return 0.
2282 static int check_consumer_health(void)
2286 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2288 DBG3("Health consumer check %d", ret
);
2294 * Setup necessary data for kernel tracer action.
2296 static int init_kernel_tracer(void)
2300 /* Modprobe lttng kernel modules */
2301 ret
= modprobe_lttng_control();
2306 /* Open debugfs lttng */
2307 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2308 if (kernel_tracer_fd
< 0) {
2309 DBG("Failed to open %s", module_proc_lttng
);
2314 /* Validate kernel version */
2315 ret
= kernel_validate_version(kernel_tracer_fd
);
2320 ret
= modprobe_lttng_data();
2325 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2329 modprobe_remove_lttng_control();
2330 ret
= close(kernel_tracer_fd
);
2334 kernel_tracer_fd
= -1;
2335 return LTTNG_ERR_KERN_VERSION
;
2338 ret
= close(kernel_tracer_fd
);
2344 modprobe_remove_lttng_control();
2347 WARN("No kernel tracer available");
2348 kernel_tracer_fd
= -1;
2350 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2352 return LTTNG_ERR_KERN_NA
;
2358 * Copy consumer output from the tracing session to the domain session. The
2359 * function also applies the right modification on a per domain basis for the
2360 * trace files destination directory.
2362 * Should *NOT* be called with RCU read-side lock held.
2364 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2367 const char *dir_name
;
2368 struct consumer_output
*consumer
;
2371 assert(session
->consumer
);
2374 case LTTNG_DOMAIN_KERNEL
:
2375 DBG3("Copying tracing session consumer output in kernel session");
2377 * XXX: We should audit the session creation and what this function
2378 * does "extra" in order to avoid a destroy since this function is used
2379 * in the domain session creation (kernel and ust) only. Same for UST
2382 if (session
->kernel_session
->consumer
) {
2383 consumer_destroy_output(session
->kernel_session
->consumer
);
2385 session
->kernel_session
->consumer
=
2386 consumer_copy_output(session
->consumer
);
2387 /* Ease our life a bit for the next part */
2388 consumer
= session
->kernel_session
->consumer
;
2389 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2391 case LTTNG_DOMAIN_UST
:
2392 DBG3("Copying tracing session consumer output in UST session");
2393 if (session
->ust_session
->consumer
) {
2394 consumer_destroy_output(session
->ust_session
->consumer
);
2396 session
->ust_session
->consumer
=
2397 consumer_copy_output(session
->consumer
);
2398 /* Ease our life a bit for the next part */
2399 consumer
= session
->ust_session
->consumer
;
2400 dir_name
= DEFAULT_UST_TRACE_DIR
;
2403 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2407 /* Append correct directory to subdir */
2408 strncat(consumer
->subdir
, dir_name
,
2409 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2410 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2419 * Create an UST session and add it to the session ust list.
2421 * Should *NOT* be called with RCU read-side lock held.
2423 static int create_ust_session(struct ltt_session
*session
,
2424 struct lttng_domain
*domain
)
2427 struct ltt_ust_session
*lus
= NULL
;
2431 assert(session
->consumer
);
2433 switch (domain
->type
) {
2434 case LTTNG_DOMAIN_UST
:
2437 ERR("Unknown UST domain on create session %d", domain
->type
);
2438 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2442 DBG("Creating UST session");
2444 lus
= trace_ust_create_session(session
->id
);
2446 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2450 lus
->uid
= session
->uid
;
2451 lus
->gid
= session
->gid
;
2452 lus
->output_traces
= session
->output_traces
;
2453 lus
->snapshot_mode
= session
->snapshot_mode
;
2454 session
->ust_session
= lus
;
2456 /* Copy session output to the newly created UST session */
2457 ret
= copy_session_consumer(domain
->type
, session
);
2458 if (ret
!= LTTNG_OK
) {
2466 session
->ust_session
= NULL
;
2471 * Create a kernel tracer session then create the default channel.
2473 static int create_kernel_session(struct ltt_session
*session
)
2477 DBG("Creating kernel session");
2479 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2481 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2485 /* Code flow safety */
2486 assert(session
->kernel_session
);
2488 /* Copy session output to the newly created Kernel session */
2489 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2490 if (ret
!= LTTNG_OK
) {
2494 /* Create directory(ies) on local filesystem. */
2495 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2496 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2497 ret
= run_as_mkdir_recursive(
2498 session
->kernel_session
->consumer
->dst
.trace_path
,
2499 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2501 if (ret
!= -EEXIST
) {
2502 ERR("Trace directory creation error");
2508 session
->kernel_session
->uid
= session
->uid
;
2509 session
->kernel_session
->gid
= session
->gid
;
2510 session
->kernel_session
->output_traces
= session
->output_traces
;
2511 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2516 trace_kernel_destroy_session(session
->kernel_session
);
2517 session
->kernel_session
= NULL
;
2522 * Count number of session permitted by uid/gid.
2524 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2527 struct ltt_session
*session
;
2529 DBG("Counting number of available session for UID %d GID %d",
2531 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2533 * Only list the sessions the user can control.
2535 if (!session_access_ok(session
, uid
, gid
)) {
2544 * Process the command requested by the lttng client within the command
2545 * context structure. This function make sure that the return structure (llm)
2546 * is set and ready for transmission before returning.
2548 * Return any error encountered or 0 for success.
2550 * "sock" is only used for special-case var. len data.
2552 * Should *NOT* be called with RCU read-side lock held.
2554 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2558 int need_tracing_session
= 1;
2561 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2565 switch (cmd_ctx
->lsm
->cmd_type
) {
2566 case LTTNG_CREATE_SESSION
:
2567 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2568 case LTTNG_DESTROY_SESSION
:
2569 case LTTNG_LIST_SESSIONS
:
2570 case LTTNG_LIST_DOMAINS
:
2571 case LTTNG_START_TRACE
:
2572 case LTTNG_STOP_TRACE
:
2573 case LTTNG_DATA_PENDING
:
2574 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2575 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2576 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2577 case LTTNG_SNAPSHOT_RECORD
:
2584 if (opt_no_kernel
&& need_domain
2585 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2587 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2589 ret
= LTTNG_ERR_KERN_NA
;
2594 /* Deny register consumer if we already have a spawned consumer. */
2595 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2596 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2597 if (kconsumer_data
.pid
> 0) {
2598 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2599 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2602 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2606 * Check for command that don't needs to allocate a returned payload. We do
2607 * this here so we don't have to make the call for no payload at each
2610 switch(cmd_ctx
->lsm
->cmd_type
) {
2611 case LTTNG_LIST_SESSIONS
:
2612 case LTTNG_LIST_TRACEPOINTS
:
2613 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2614 case LTTNG_LIST_DOMAINS
:
2615 case LTTNG_LIST_CHANNELS
:
2616 case LTTNG_LIST_EVENTS
:
2619 /* Setup lttng message with no payload */
2620 ret
= setup_lttng_msg(cmd_ctx
, 0);
2622 /* This label does not try to unlock the session */
2623 goto init_setup_error
;
2627 /* Commands that DO NOT need a session. */
2628 switch (cmd_ctx
->lsm
->cmd_type
) {
2629 case LTTNG_CREATE_SESSION
:
2630 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2631 case LTTNG_CALIBRATE
:
2632 case LTTNG_LIST_SESSIONS
:
2633 case LTTNG_LIST_TRACEPOINTS
:
2634 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2635 need_tracing_session
= 0;
2638 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2640 * We keep the session list lock across _all_ commands
2641 * for now, because the per-session lock does not
2642 * handle teardown properly.
2644 session_lock_list();
2645 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2646 if (cmd_ctx
->session
== NULL
) {
2647 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2650 /* Acquire lock for the session */
2651 session_lock(cmd_ctx
->session
);
2661 * Check domain type for specific "pre-action".
2663 switch (cmd_ctx
->lsm
->domain
.type
) {
2664 case LTTNG_DOMAIN_KERNEL
:
2666 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2670 /* Kernel tracer check */
2671 if (kernel_tracer_fd
== -1) {
2672 /* Basically, load kernel tracer modules */
2673 ret
= init_kernel_tracer();
2679 /* Consumer is in an ERROR state. Report back to client */
2680 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2681 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2685 /* Need a session for kernel command */
2686 if (need_tracing_session
) {
2687 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2688 ret
= create_kernel_session(cmd_ctx
->session
);
2690 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2695 /* Start the kernel consumer daemon */
2696 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2697 if (kconsumer_data
.pid
== 0 &&
2698 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2699 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2700 ret
= start_consumerd(&kconsumer_data
);
2702 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2705 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2707 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2711 * The consumer was just spawned so we need to add the socket to
2712 * the consumer output of the session if exist.
2714 ret
= consumer_create_socket(&kconsumer_data
,
2715 cmd_ctx
->session
->kernel_session
->consumer
);
2722 case LTTNG_DOMAIN_UST
:
2724 if (!ust_app_supported()) {
2725 ret
= LTTNG_ERR_NO_UST
;
2728 /* Consumer is in an ERROR state. Report back to client */
2729 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2730 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2734 if (need_tracing_session
) {
2735 /* Create UST session if none exist. */
2736 if (cmd_ctx
->session
->ust_session
== NULL
) {
2737 ret
= create_ust_session(cmd_ctx
->session
,
2738 &cmd_ctx
->lsm
->domain
);
2739 if (ret
!= LTTNG_OK
) {
2744 /* Start the UST consumer daemons */
2746 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2747 if (consumerd64_bin
[0] != '\0' &&
2748 ustconsumer64_data
.pid
== 0 &&
2749 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2750 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2751 ret
= start_consumerd(&ustconsumer64_data
);
2753 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2754 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2758 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2759 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2761 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2765 * Setup socket for consumer 64 bit. No need for atomic access
2766 * since it was set above and can ONLY be set in this thread.
2768 ret
= consumer_create_socket(&ustconsumer64_data
,
2769 cmd_ctx
->session
->ust_session
->consumer
);
2775 if (consumerd32_bin
[0] != '\0' &&
2776 ustconsumer32_data
.pid
== 0 &&
2777 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2778 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2779 ret
= start_consumerd(&ustconsumer32_data
);
2781 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2782 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2786 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2787 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2789 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2793 * Setup socket for consumer 64 bit. No need for atomic access
2794 * since it was set above and can ONLY be set in this thread.
2796 ret
= consumer_create_socket(&ustconsumer32_data
,
2797 cmd_ctx
->session
->ust_session
->consumer
);
2809 /* Validate consumer daemon state when start/stop trace command */
2810 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2811 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2812 switch (cmd_ctx
->lsm
->domain
.type
) {
2813 case LTTNG_DOMAIN_UST
:
2814 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2815 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2819 case LTTNG_DOMAIN_KERNEL
:
2820 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2821 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2829 * Check that the UID or GID match that of the tracing session.
2830 * The root user can interact with all sessions.
2832 if (need_tracing_session
) {
2833 if (!session_access_ok(cmd_ctx
->session
,
2834 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2835 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2836 ret
= LTTNG_ERR_EPERM
;
2842 * Send relayd information to consumer as soon as we have a domain and a
2845 if (cmd_ctx
->session
&& need_domain
) {
2847 * Setup relayd if not done yet. If the relayd information was already
2848 * sent to the consumer, this call will gracefully return.
2850 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2851 if (ret
!= LTTNG_OK
) {
2856 /* Process by command type */
2857 switch (cmd_ctx
->lsm
->cmd_type
) {
2858 case LTTNG_ADD_CONTEXT
:
2860 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2861 cmd_ctx
->lsm
->u
.context
.channel_name
,
2862 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2865 case LTTNG_DISABLE_CHANNEL
:
2867 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2868 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2871 case LTTNG_DISABLE_EVENT
:
2873 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2874 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2875 cmd_ctx
->lsm
->u
.disable
.name
);
2878 case LTTNG_DISABLE_ALL_EVENT
:
2880 DBG("Disabling all events");
2882 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2883 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2886 case LTTNG_ENABLE_CHANNEL
:
2888 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2889 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2892 case LTTNG_ENABLE_EVENT
:
2894 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2895 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2896 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2899 case LTTNG_ENABLE_ALL_EVENT
:
2901 DBG("Enabling all events");
2903 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2904 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2905 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2908 case LTTNG_LIST_TRACEPOINTS
:
2910 struct lttng_event
*events
;
2913 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2914 if (nb_events
< 0) {
2915 /* Return value is a negative lttng_error_code. */
2921 * Setup lttng message with payload size set to the event list size in
2922 * bytes and then copy list into the llm payload.
2924 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2930 /* Copy event list into message payload */
2931 memcpy(cmd_ctx
->llm
->payload
, events
,
2932 sizeof(struct lttng_event
) * nb_events
);
2939 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2941 struct lttng_event_field
*fields
;
2944 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2946 if (nb_fields
< 0) {
2947 /* Return value is a negative lttng_error_code. */
2953 * Setup lttng message with payload size set to the event list size in
2954 * bytes and then copy list into the llm payload.
2956 ret
= setup_lttng_msg(cmd_ctx
,
2957 sizeof(struct lttng_event_field
) * nb_fields
);
2963 /* Copy event list into message payload */
2964 memcpy(cmd_ctx
->llm
->payload
, fields
,
2965 sizeof(struct lttng_event_field
) * nb_fields
);
2972 case LTTNG_SET_CONSUMER_URI
:
2975 struct lttng_uri
*uris
;
2977 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2978 len
= nb_uri
* sizeof(struct lttng_uri
);
2981 ret
= LTTNG_ERR_INVALID
;
2985 uris
= zmalloc(len
);
2987 ret
= LTTNG_ERR_FATAL
;
2991 /* Receive variable len data */
2992 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2993 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2995 DBG("No URIs received from client... continuing");
2997 ret
= LTTNG_ERR_SESSION_FAIL
;
3002 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3004 if (ret
!= LTTNG_OK
) {
3010 * XXX: 0 means that this URI should be applied on the session. Should
3011 * be a DOMAIN enuam.
3013 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3014 /* Add the URI for the UST session if a consumer is present. */
3015 if (cmd_ctx
->session
->ust_session
&&
3016 cmd_ctx
->session
->ust_session
->consumer
) {
3017 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3019 } else if (cmd_ctx
->session
->kernel_session
&&
3020 cmd_ctx
->session
->kernel_session
->consumer
) {
3021 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3022 cmd_ctx
->session
, nb_uri
, uris
);
3030 case LTTNG_START_TRACE
:
3032 ret
= cmd_start_trace(cmd_ctx
->session
);
3035 case LTTNG_STOP_TRACE
:
3037 ret
= cmd_stop_trace(cmd_ctx
->session
);
3040 case LTTNG_CREATE_SESSION
:
3043 struct lttng_uri
*uris
= NULL
;
3045 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3046 len
= nb_uri
* sizeof(struct lttng_uri
);
3049 uris
= zmalloc(len
);
3051 ret
= LTTNG_ERR_FATAL
;
3055 /* Receive variable len data */
3056 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3057 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3059 DBG("No URIs received from client... continuing");
3061 ret
= LTTNG_ERR_SESSION_FAIL
;
3066 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3067 DBG("Creating session with ONE network URI is a bad call");
3068 ret
= LTTNG_ERR_SESSION_FAIL
;
3074 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3081 case LTTNG_DESTROY_SESSION
:
3083 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3085 /* Set session to NULL so we do not unlock it after free. */
3086 cmd_ctx
->session
= NULL
;
3089 case LTTNG_LIST_DOMAINS
:
3092 struct lttng_domain
*domains
;
3094 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3096 /* Return value is a negative lttng_error_code. */
3101 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3107 /* Copy event list into message payload */
3108 memcpy(cmd_ctx
->llm
->payload
, domains
,
3109 nb_dom
* sizeof(struct lttng_domain
));
3116 case LTTNG_LIST_CHANNELS
:
3119 struct lttng_channel
*channels
;
3121 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3122 cmd_ctx
->session
, &channels
);
3124 /* Return value is a negative lttng_error_code. */
3129 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3135 /* Copy event list into message payload */
3136 memcpy(cmd_ctx
->llm
->payload
, channels
,
3137 nb_chan
* sizeof(struct lttng_channel
));
3144 case LTTNG_LIST_EVENTS
:
3147 struct lttng_event
*events
= NULL
;
3149 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3150 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3152 /* Return value is a negative lttng_error_code. */
3157 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3163 /* Copy event list into message payload */
3164 memcpy(cmd_ctx
->llm
->payload
, events
,
3165 nb_event
* sizeof(struct lttng_event
));
3172 case LTTNG_LIST_SESSIONS
:
3174 unsigned int nr_sessions
;
3176 session_lock_list();
3177 nr_sessions
= lttng_sessions_count(
3178 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3179 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3181 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3183 session_unlock_list();
3187 /* Filled the session array */
3188 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3189 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3190 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3192 session_unlock_list();
3197 case LTTNG_CALIBRATE
:
3199 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3200 &cmd_ctx
->lsm
->u
.calibrate
);
3203 case LTTNG_REGISTER_CONSUMER
:
3205 struct consumer_data
*cdata
;
3207 switch (cmd_ctx
->lsm
->domain
.type
) {
3208 case LTTNG_DOMAIN_KERNEL
:
3209 cdata
= &kconsumer_data
;
3212 ret
= LTTNG_ERR_UND
;
3216 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3217 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3220 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3222 struct lttng_filter_bytecode
*bytecode
;
3224 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3225 ret
= LTTNG_ERR_FILTER_INVAL
;
3228 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3229 ret
= LTTNG_ERR_FILTER_INVAL
;
3232 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3234 ret
= LTTNG_ERR_FILTER_NOMEM
;
3237 /* Receive var. len. data */
3238 DBG("Receiving var len data from client ...");
3239 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3240 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3242 DBG("Nothing recv() from client var len data... continuing");
3244 ret
= LTTNG_ERR_FILTER_INVAL
;
3248 if (bytecode
->len
+ sizeof(*bytecode
)
3249 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3251 ret
= LTTNG_ERR_FILTER_INVAL
;
3255 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3256 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3257 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3260 case LTTNG_DATA_PENDING
:
3262 ret
= cmd_data_pending(cmd_ctx
->session
);
3265 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3267 struct lttcomm_lttng_output_id reply
;
3269 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3270 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3271 if (ret
!= LTTNG_OK
) {
3275 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3280 /* Copy output list into message payload */
3281 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3285 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3287 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3288 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3291 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3294 struct lttng_snapshot_output
*outputs
= NULL
;
3296 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3297 if (nb_output
< 0) {
3302 ret
= setup_lttng_msg(cmd_ctx
,
3303 nb_output
* sizeof(struct lttng_snapshot_output
));
3310 /* Copy output list into message payload */
3311 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3312 nb_output
* sizeof(struct lttng_snapshot_output
));
3319 case LTTNG_SNAPSHOT_RECORD
:
3321 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3322 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3323 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3326 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3329 struct lttng_uri
*uris
= NULL
;
3331 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3332 len
= nb_uri
* sizeof(struct lttng_uri
);
3335 uris
= zmalloc(len
);
3337 ret
= LTTNG_ERR_FATAL
;
3341 /* Receive variable len data */
3342 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3343 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3345 DBG("No URIs received from client... continuing");
3347 ret
= LTTNG_ERR_SESSION_FAIL
;
3352 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3353 DBG("Creating session with ONE network URI is a bad call");
3354 ret
= LTTNG_ERR_SESSION_FAIL
;
3360 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3361 nb_uri
, &cmd_ctx
->creds
);
3366 ret
= LTTNG_ERR_UND
;
3371 if (cmd_ctx
->llm
== NULL
) {
3372 DBG("Missing llm structure. Allocating one.");
3373 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3377 /* Set return code */
3378 cmd_ctx
->llm
->ret_code
= ret
;
3380 if (cmd_ctx
->session
) {
3381 session_unlock(cmd_ctx
->session
);
3383 if (need_tracing_session
) {
3384 session_unlock_list();
3391 * Thread managing health check socket.
3393 static void *thread_manage_health(void *data
)
3395 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3396 uint32_t revents
, nb_fd
;
3397 struct lttng_poll_event events
;
3398 struct lttcomm_health_msg msg
;
3399 struct lttcomm_health_data reply
;
3401 DBG("[thread] Manage health check started");
3403 rcu_register_thread();
3405 /* We might hit an error path before this is created. */
3406 lttng_poll_init(&events
);
3408 /* Create unix socket */
3409 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3411 ERR("Unable to create health check Unix socket");
3417 * Set the CLOEXEC flag. Return code is useless because either way, the
3420 (void) utils_set_fd_cloexec(sock
);
3422 ret
= lttcomm_listen_unix_sock(sock
);
3428 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3429 * more will be added to this poll set.
3431 ret
= sessiond_set_thread_pollset(&events
, 2);
3436 /* Add the application registration socket */
3437 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3443 DBG("Health check ready");
3445 /* Inifinite blocking call, waiting for transmission */
3447 ret
= lttng_poll_wait(&events
, -1);
3450 * Restart interrupted system call.
3452 if (errno
== EINTR
) {
3460 for (i
= 0; i
< nb_fd
; i
++) {
3461 /* Fetch once the poll data */
3462 revents
= LTTNG_POLL_GETEV(&events
, i
);
3463 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3465 /* Thread quit pipe has been closed. Killing thread. */
3466 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3472 /* Event on the registration socket */
3473 if (pollfd
== sock
) {
3474 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3475 ERR("Health socket poll error");
3481 new_sock
= lttcomm_accept_unix_sock(sock
);
3487 * Set the CLOEXEC flag. Return code is useless because either way, the
3490 (void) utils_set_fd_cloexec(new_sock
);
3492 DBG("Receiving data from client for health...");
3493 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3495 DBG("Nothing recv() from client... continuing");
3496 ret
= close(new_sock
);
3504 rcu_thread_online();
3506 switch (msg
.component
) {
3507 case LTTNG_HEALTH_CMD
:
3508 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3510 case LTTNG_HEALTH_APP_MANAGE
:
3511 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3513 case LTTNG_HEALTH_APP_REG
:
3514 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3516 case LTTNG_HEALTH_KERNEL
:
3517 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3519 case LTTNG_HEALTH_CONSUMER
:
3520 reply
.ret_code
= check_consumer_health();
3522 case LTTNG_HEALTH_HT_CLEANUP
:
3523 reply
.ret_code
= health_check_state(HEALTH_TYPE_HT_CLEANUP
);
3525 case LTTNG_HEALTH_APP_MANAGE_NOTIFY
:
3526 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3528 case LTTNG_HEALTH_APP_REG_DISPATCH
:
3529 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3531 case LTTNG_HEALTH_ALL
:
3533 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3534 health_check_state(HEALTH_TYPE_APP_REG
) &&
3535 health_check_state(HEALTH_TYPE_CMD
) &&
3536 health_check_state(HEALTH_TYPE_KERNEL
) &&
3537 check_consumer_health() &&
3538 health_check_state(HEALTH_TYPE_HT_CLEANUP
) &&
3539 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
) &&
3540 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3543 reply
.ret_code
= LTTNG_ERR_UND
;
3548 * Flip ret value since 0 is a success and 1 indicates a bad health for
3549 * the client where in the sessiond it is the opposite. Again, this is
3550 * just to make things easier for us poor developer which enjoy a lot
3553 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3554 reply
.ret_code
= !reply
.ret_code
;
3557 DBG2("Health check return value %d", reply
.ret_code
);
3559 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3561 ERR("Failed to send health data back to client");
3564 /* End of transmission */
3565 ret
= close(new_sock
);
3575 ERR("Health error occurred in %s", __func__
);
3577 DBG("Health check thread dying");
3578 unlink(health_unix_sock_path
);
3586 lttng_poll_clean(&events
);
3588 rcu_unregister_thread();
3593 * This thread manage all clients request using the unix client socket for
3596 static void *thread_manage_clients(void *data
)
3598 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3600 uint32_t revents
, nb_fd
;
3601 struct command_ctx
*cmd_ctx
= NULL
;
3602 struct lttng_poll_event events
;
3604 DBG("[thread] Manage client started");
3606 rcu_register_thread();
3608 health_register(HEALTH_TYPE_CMD
);
3610 if (testpoint(thread_manage_clients
)) {
3611 goto error_testpoint
;
3614 health_code_update();
3616 ret
= lttcomm_listen_unix_sock(client_sock
);
3622 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3623 * more will be added to this poll set.
3625 ret
= sessiond_set_thread_pollset(&events
, 2);
3627 goto error_create_poll
;
3630 /* Add the application registration socket */
3631 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3637 * Notify parent pid that we are ready to accept command for client side.
3639 if (opt_sig_parent
) {
3640 kill(ppid
, SIGUSR1
);
3643 if (testpoint(thread_manage_clients_before_loop
)) {
3647 health_code_update();
3650 DBG("Accepting client command ...");
3652 /* Inifinite blocking call, waiting for transmission */
3654 health_poll_entry();
3655 ret
= lttng_poll_wait(&events
, -1);
3659 * Restart interrupted system call.
3661 if (errno
== EINTR
) {
3669 for (i
= 0; i
< nb_fd
; i
++) {
3670 /* Fetch once the poll data */
3671 revents
= LTTNG_POLL_GETEV(&events
, i
);
3672 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3674 health_code_update();
3676 /* Thread quit pipe has been closed. Killing thread. */
3677 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3683 /* Event on the registration socket */
3684 if (pollfd
== client_sock
) {
3685 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3686 ERR("Client socket poll error");
3692 DBG("Wait for client response");
3694 health_code_update();
3696 sock
= lttcomm_accept_unix_sock(client_sock
);
3702 * Set the CLOEXEC flag. Return code is useless because either way, the
3705 (void) utils_set_fd_cloexec(sock
);
3707 /* Set socket option for credentials retrieval */
3708 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3713 /* Allocate context command to process the client request */
3714 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3715 if (cmd_ctx
== NULL
) {
3716 PERROR("zmalloc cmd_ctx");
3720 /* Allocate data buffer for reception */
3721 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3722 if (cmd_ctx
->lsm
== NULL
) {
3723 PERROR("zmalloc cmd_ctx->lsm");
3727 cmd_ctx
->llm
= NULL
;
3728 cmd_ctx
->session
= NULL
;
3730 health_code_update();
3733 * Data is received from the lttng client. The struct
3734 * lttcomm_session_msg (lsm) contains the command and data request of
3737 DBG("Receiving data from client ...");
3738 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3739 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3741 DBG("Nothing recv() from client... continuing");
3747 clean_command_ctx(&cmd_ctx
);
3751 health_code_update();
3753 // TODO: Validate cmd_ctx including sanity check for
3754 // security purpose.
3756 rcu_thread_online();
3758 * This function dispatch the work to the kernel or userspace tracer
3759 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3760 * informations for the client. The command context struct contains
3761 * everything this function may needs.
3763 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3764 rcu_thread_offline();
3772 * TODO: Inform client somehow of the fatal error. At
3773 * this point, ret < 0 means that a zmalloc failed
3774 * (ENOMEM). Error detected but still accept
3775 * command, unless a socket error has been
3778 clean_command_ctx(&cmd_ctx
);
3782 health_code_update();
3784 DBG("Sending response (size: %d, retcode: %s)",
3785 cmd_ctx
->lttng_msg_size
,
3786 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3787 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3789 ERR("Failed to send data back to client");
3792 /* End of transmission */
3799 clean_command_ctx(&cmd_ctx
);
3801 health_code_update();
3813 lttng_poll_clean(&events
);
3814 clean_command_ctx(&cmd_ctx
);
3819 unlink(client_unix_sock_path
);
3820 if (client_sock
>= 0) {
3821 ret
= close(client_sock
);
3829 ERR("Health error occurred in %s", __func__
);
3832 health_unregister();
3834 DBG("Client thread dying");
3836 rcu_unregister_thread();
3842 * usage function on stderr
3844 static void usage(void)
3846 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3847 fprintf(stderr
, " -h, --help Display this usage.\n");
3848 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3849 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3850 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3851 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3852 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3853 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3854 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3855 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3856 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3857 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3858 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3859 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3860 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3861 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3862 fprintf(stderr
, " -V, --version Show version number.\n");
3863 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3864 fprintf(stderr
, " -q, --quiet No output at all.\n");
3865 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3866 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3867 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3868 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3872 * daemon argument parsing
3874 static int parse_args(int argc
, char **argv
)
3878 static struct option long_options
[] = {
3879 { "client-sock", 1, 0, 'c' },
3880 { "apps-sock", 1, 0, 'a' },
3881 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3882 { "kconsumerd-err-sock", 1, 0, 'E' },
3883 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3884 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3885 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3886 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3887 { "consumerd32-path", 1, 0, 'u' },
3888 { "consumerd32-libdir", 1, 0, 'U' },
3889 { "consumerd64-path", 1, 0, 't' },
3890 { "consumerd64-libdir", 1, 0, 'T' },
3891 { "daemonize", 0, 0, 'd' },
3892 { "sig-parent", 0, 0, 'S' },
3893 { "help", 0, 0, 'h' },
3894 { "group", 1, 0, 'g' },
3895 { "version", 0, 0, 'V' },
3896 { "quiet", 0, 0, 'q' },
3897 { "verbose", 0, 0, 'v' },
3898 { "verbose-consumer", 0, 0, 'Z' },
3899 { "no-kernel", 0, 0, 'N' },
3900 { "pidfile", 1, 0, 'p' },
3905 int option_index
= 0;
3906 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3907 long_options
, &option_index
);
3914 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3916 fprintf(stderr
, " with arg %s\n", optarg
);
3920 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3923 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3929 opt_tracing_group
= optarg
;
3935 fprintf(stdout
, "%s\n", VERSION
);
3941 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3944 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3947 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3950 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3953 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3956 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3962 lttng_opt_quiet
= 1;
3965 /* Verbose level can increase using multiple -v */
3966 lttng_opt_verbose
+= 1;
3969 opt_verbose_consumer
+= 1;
3972 consumerd32_bin
= optarg
;
3975 consumerd32_libdir
= optarg
;
3978 consumerd64_bin
= optarg
;
3981 consumerd64_libdir
= optarg
;
3984 opt_pidfile
= optarg
;
3987 /* Unknown option or other error.
3988 * Error is printed by getopt, just return */
3997 * Creates the two needed socket by the daemon.
3998 * apps_sock - The communication socket for all UST apps.
3999 * client_sock - The communication of the cli tool (lttng).
4001 static int init_daemon_socket(void)
4006 old_umask
= umask(0);
4008 /* Create client tool unix socket */
4009 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4010 if (client_sock
< 0) {
4011 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4016 /* Set the cloexec flag */
4017 ret
= utils_set_fd_cloexec(client_sock
);
4019 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4020 "Continuing but note that the consumer daemon will have a "
4021 "reference to this socket on exec()", client_sock
);
4024 /* File permission MUST be 660 */
4025 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4027 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4032 /* Create the application unix socket */
4033 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4034 if (apps_sock
< 0) {
4035 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4040 /* Set the cloexec flag */
4041 ret
= utils_set_fd_cloexec(apps_sock
);
4043 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4044 "Continuing but note that the consumer daemon will have a "
4045 "reference to this socket on exec()", apps_sock
);
4048 /* File permission MUST be 666 */
4049 ret
= chmod(apps_unix_sock_path
,
4050 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4052 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4057 DBG3("Session daemon client socket %d and application socket %d created",
4058 client_sock
, apps_sock
);
4066 * Check if the global socket is available, and if a daemon is answering at the
4067 * other side. If yes, error is returned.
4069 static int check_existing_daemon(void)
4071 /* Is there anybody out there ? */
4072 if (lttng_session_daemon_alive()) {
4080 * Set the tracing group gid onto the client socket.
4082 * Race window between mkdir and chown is OK because we are going from more
4083 * permissive (root.root) to less permissive (root.tracing).
4085 static int set_permissions(char *rundir
)
4090 ret
= allowed_group();
4092 WARN("No tracing group detected");
4093 /* Setting gid to 0 if no tracing group is found */
4099 /* Set lttng run dir */
4100 ret
= chown(rundir
, 0, gid
);
4102 ERR("Unable to set group on %s", rundir
);
4106 /* Ensure all applications and tracing group can search the run dir */
4107 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
4109 ERR("Unable to set permissions on %s", rundir
);
4113 /* lttng client socket path */
4114 ret
= chown(client_unix_sock_path
, 0, gid
);
4116 ERR("Unable to set group on %s", client_unix_sock_path
);
4120 /* kconsumer error socket path */
4121 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
4123 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4127 /* 64-bit ustconsumer error socket path */
4128 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
4130 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4134 /* 32-bit ustconsumer compat32 error socket path */
4135 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
4137 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4141 DBG("All permissions are set");
4147 * Create the lttng run directory needed for all global sockets and pipe.
4149 static int create_lttng_rundir(const char *rundir
)
4153 DBG3("Creating LTTng run directory: %s", rundir
);
4155 ret
= mkdir(rundir
, S_IRWXU
);
4157 if (errno
!= EEXIST
) {
4158 ERR("Unable to create %s", rundir
);
4170 * Setup sockets and directory needed by the kconsumerd communication with the
4173 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4177 char path
[PATH_MAX
];
4179 switch (consumer_data
->type
) {
4180 case LTTNG_CONSUMER_KERNEL
:
4181 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4183 case LTTNG_CONSUMER64_UST
:
4184 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4186 case LTTNG_CONSUMER32_UST
:
4187 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4190 ERR("Consumer type unknown");
4195 DBG2("Creating consumer directory: %s", path
);
4197 ret
= mkdir(path
, S_IRWXU
);
4199 if (errno
!= EEXIST
) {
4201 ERR("Failed to create %s", path
);
4207 /* Create the kconsumerd error unix socket */
4208 consumer_data
->err_sock
=
4209 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4210 if (consumer_data
->err_sock
< 0) {
4211 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4217 * Set the CLOEXEC flag. Return code is useless because either way, the
4220 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4222 PERROR("utils_set_fd_cloexec");
4223 /* continue anyway */
4226 /* File permission MUST be 660 */
4227 ret
= chmod(consumer_data
->err_unix_sock_path
,
4228 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4230 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4240 * Signal handler for the daemon
4242 * Simply stop all worker threads, leaving main() return gracefully after
4243 * joining all threads and calling cleanup().
4245 static void sighandler(int sig
)
4249 DBG("SIGPIPE caught");
4252 DBG("SIGINT caught");
4256 DBG("SIGTERM caught");
4265 * Setup signal handler for :
4266 * SIGINT, SIGTERM, SIGPIPE
4268 static int set_signal_handler(void)
4271 struct sigaction sa
;
4274 if ((ret
= sigemptyset(&sigset
)) < 0) {
4275 PERROR("sigemptyset");
4279 sa
.sa_handler
= sighandler
;
4280 sa
.sa_mask
= sigset
;
4282 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4283 PERROR("sigaction");
4287 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4288 PERROR("sigaction");
4292 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4293 PERROR("sigaction");
4297 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4303 * Set open files limit to unlimited. This daemon can open a large number of
4304 * file descriptors in order to consumer multiple kernel traces.
4306 static void set_ulimit(void)
4311 /* The kernel does not allowed an infinite limit for open files */
4312 lim
.rlim_cur
= 65535;
4313 lim
.rlim_max
= 65535;
4315 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4317 PERROR("failed to set open files limit");
4322 * Write pidfile using the rundir and opt_pidfile.
4324 static void write_pidfile(void)
4327 char pidfile_path
[PATH_MAX
];
4332 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4334 /* Build pidfile path from rundir and opt_pidfile. */
4335 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4336 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4338 PERROR("snprintf pidfile path");
4344 * Create pid file in rundir. Return value is of no importance. The
4345 * execution will continue even though we are not able to write the file.
4347 (void) utils_create_pid_file(getpid(), pidfile_path
);
4356 int main(int argc
, char **argv
)
4360 const char *home_path
, *env_app_timeout
;
4362 init_kernel_workarounds();
4364 rcu_register_thread();
4366 setup_consumerd_path();
4368 page_size
= sysconf(_SC_PAGESIZE
);
4369 if (page_size
< 0) {
4370 PERROR("sysconf _SC_PAGESIZE");
4371 page_size
= LONG_MAX
;
4372 WARN("Fallback page size to %ld", page_size
);
4375 /* Parse arguments */
4377 if ((ret
= parse_args(argc
, argv
)) < 0) {
4387 * child: setsid, close FD 0, 1, 2, chdir /
4388 * parent: exit (if fork is successful)
4396 * We are in the child. Make sure all other file
4397 * descriptors are closed, in case we are called with
4398 * more opened file descriptors than the standard ones.
4400 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4405 /* Create thread quit pipe */
4406 if ((ret
= init_thread_quit_pipe()) < 0) {
4410 /* Check if daemon is UID = 0 */
4411 is_root
= !getuid();
4414 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4416 /* Create global run dir with root access */
4417 ret
= create_lttng_rundir(rundir
);
4422 if (strlen(apps_unix_sock_path
) == 0) {
4423 snprintf(apps_unix_sock_path
, PATH_MAX
,
4424 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4427 if (strlen(client_unix_sock_path
) == 0) {
4428 snprintf(client_unix_sock_path
, PATH_MAX
,
4429 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4432 /* Set global SHM for ust */
4433 if (strlen(wait_shm_path
) == 0) {
4434 snprintf(wait_shm_path
, PATH_MAX
,
4435 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4438 if (strlen(health_unix_sock_path
) == 0) {
4439 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4440 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4443 /* Setup kernel consumerd path */
4444 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4445 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4446 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4447 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4449 DBG2("Kernel consumer err path: %s",
4450 kconsumer_data
.err_unix_sock_path
);
4451 DBG2("Kernel consumer cmd path: %s",
4452 kconsumer_data
.cmd_unix_sock_path
);
4454 home_path
= utils_get_home_dir();
4455 if (home_path
== NULL
) {
4456 /* TODO: Add --socket PATH option */
4457 ERR("Can't get HOME directory for sockets creation.");
4463 * Create rundir from home path. This will create something like
4466 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4472 ret
= create_lttng_rundir(rundir
);
4477 if (strlen(apps_unix_sock_path
) == 0) {
4478 snprintf(apps_unix_sock_path
, PATH_MAX
,
4479 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4482 /* Set the cli tool unix socket path */
4483 if (strlen(client_unix_sock_path
) == 0) {
4484 snprintf(client_unix_sock_path
, PATH_MAX
,
4485 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4488 /* Set global SHM for ust */
4489 if (strlen(wait_shm_path
) == 0) {
4490 snprintf(wait_shm_path
, PATH_MAX
,
4491 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4494 /* Set health check Unix path */
4495 if (strlen(health_unix_sock_path
) == 0) {
4496 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4497 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4501 /* Set consumer initial state */
4502 kernel_consumerd_state
= CONSUMER_STOPPED
;
4503 ust_consumerd_state
= CONSUMER_STOPPED
;
4505 DBG("Client socket path %s", client_unix_sock_path
);
4506 DBG("Application socket path %s", apps_unix_sock_path
);
4507 DBG("Application wait path %s", wait_shm_path
);
4508 DBG("LTTng run directory path: %s", rundir
);
4510 /* 32 bits consumerd path setup */
4511 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4512 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4513 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4514 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4516 DBG2("UST consumer 32 bits err path: %s",
4517 ustconsumer32_data
.err_unix_sock_path
);
4518 DBG2("UST consumer 32 bits cmd path: %s",
4519 ustconsumer32_data
.cmd_unix_sock_path
);
4521 /* 64 bits consumerd path setup */
4522 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4523 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4524 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4525 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4527 DBG2("UST consumer 64 bits err path: %s",
4528 ustconsumer64_data
.err_unix_sock_path
);
4529 DBG2("UST consumer 64 bits cmd path: %s",
4530 ustconsumer64_data
.cmd_unix_sock_path
);
4533 * See if daemon already exist.
4535 if ((ret
= check_existing_daemon()) < 0) {
4536 ERR("Already running daemon.\n");
4538 * We do not goto exit because we must not cleanup()
4539 * because a daemon is already running.
4545 * Init UST app hash table. Alloc hash table before this point since
4546 * cleanup() can get called after that point.
4550 /* After this point, we can safely call cleanup() with "goto exit" */
4553 * These actions must be executed as root. We do that *after* setting up
4554 * the sockets path because we MUST make the check for another daemon using
4555 * those paths *before* trying to set the kernel consumer sockets and init
4559 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4564 /* Setup kernel tracer */
4565 if (!opt_no_kernel
) {
4566 init_kernel_tracer();
4569 /* Set ulimit for open files */
4572 /* init lttng_fd tracking must be done after set_ulimit. */
4575 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4580 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4585 if ((ret
= set_signal_handler()) < 0) {
4589 /* Setup the needed unix socket */
4590 if ((ret
= init_daemon_socket()) < 0) {
4594 /* Set credentials to socket */
4595 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4599 /* Get parent pid if -S, --sig-parent is specified. */
4600 if (opt_sig_parent
) {
4604 /* Setup the kernel pipe for waking up the kernel thread */
4605 if (is_root
&& !opt_no_kernel
) {
4606 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4611 /* Setup the thread ht_cleanup communication pipe. */
4612 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4616 /* Setup the thread apps communication pipe. */
4617 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4621 /* Setup the thread apps notify communication pipe. */
4622 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4626 /* Initialize global buffer per UID and PID registry. */
4627 buffer_reg_init_uid_registry();
4628 buffer_reg_init_pid_registry();
4630 /* Init UST command queue. */
4631 cds_wfq_init(&ust_cmd_queue
.queue
);
4634 * Get session list pointer. This pointer MUST NOT be free(). This list is
4635 * statically declared in session.c
4637 session_list_ptr
= session_get_list();
4639 /* Set up max poll set size */
4640 lttng_poll_set_max_size();
4644 /* Check for the application socket timeout env variable. */
4645 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4646 if (env_app_timeout
) {
4647 app_socket_timeout
= atoi(env_app_timeout
);
4649 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4654 /* Initialize communication library */
4657 /* Create thread to manage the client socket */
4658 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4659 thread_ht_cleanup
, (void *) NULL
);
4661 PERROR("pthread_create ht_cleanup");
4662 goto exit_ht_cleanup
;
4665 /* Create thread to manage the client socket */
4666 ret
= pthread_create(&health_thread
, NULL
,
4667 thread_manage_health
, (void *) NULL
);
4669 PERROR("pthread_create health");
4673 /* Create thread to manage the client socket */
4674 ret
= pthread_create(&client_thread
, NULL
,
4675 thread_manage_clients
, (void *) NULL
);
4677 PERROR("pthread_create clients");
4681 /* Create thread to dispatch registration */
4682 ret
= pthread_create(&dispatch_thread
, NULL
,
4683 thread_dispatch_ust_registration
, (void *) NULL
);
4685 PERROR("pthread_create dispatch");
4689 /* Create thread to manage application registration. */
4690 ret
= pthread_create(®_apps_thread
, NULL
,
4691 thread_registration_apps
, (void *) NULL
);
4693 PERROR("pthread_create registration");
4697 /* Create thread to manage application socket */
4698 ret
= pthread_create(&apps_thread
, NULL
,
4699 thread_manage_apps
, (void *) NULL
);
4701 PERROR("pthread_create apps");
4705 /* Create thread to manage application notify socket */
4706 ret
= pthread_create(&apps_notify_thread
, NULL
,
4707 ust_thread_manage_notify
, (void *) NULL
);
4709 PERROR("pthread_create apps");
4713 /* Don't start this thread if kernel tracing is not requested nor root */
4714 if (is_root
&& !opt_no_kernel
) {
4715 /* Create kernel thread to manage kernel event */
4716 ret
= pthread_create(&kernel_thread
, NULL
,
4717 thread_manage_kernel
, (void *) NULL
);
4719 PERROR("pthread_create kernel");
4723 ret
= pthread_join(kernel_thread
, &status
);
4725 PERROR("pthread_join");
4726 goto error
; /* join error, exit without cleanup */
4731 ret
= pthread_join(apps_thread
, &status
);
4733 PERROR("pthread_join");
4734 goto error
; /* join error, exit without cleanup */
4738 ret
= pthread_join(reg_apps_thread
, &status
);
4740 PERROR("pthread_join");
4741 goto error
; /* join error, exit without cleanup */
4745 ret
= pthread_join(dispatch_thread
, &status
);
4747 PERROR("pthread_join");
4748 goto error
; /* join error, exit without cleanup */
4752 ret
= pthread_join(client_thread
, &status
);
4754 PERROR("pthread_join");
4755 goto error
; /* join error, exit without cleanup */
4758 ret
= join_consumer_thread(&kconsumer_data
);
4760 PERROR("join_consumer");
4761 goto error
; /* join error, exit without cleanup */
4764 ret
= join_consumer_thread(&ustconsumer32_data
);
4766 PERROR("join_consumer ust32");
4767 goto error
; /* join error, exit without cleanup */
4770 ret
= join_consumer_thread(&ustconsumer64_data
);
4772 PERROR("join_consumer ust64");
4773 goto error
; /* join error, exit without cleanup */
4777 ret
= pthread_join(health_thread
, &status
);
4779 PERROR("pthread_join health thread");
4780 goto error
; /* join error, exit without cleanup */
4784 ret
= pthread_join(ht_cleanup_thread
, &status
);
4786 PERROR("pthread_join ht cleanup thread");
4787 goto error
; /* join error, exit without cleanup */
4792 * cleanup() is called when no other thread is running.
4794 rcu_thread_online();
4796 rcu_thread_offline();
4797 rcu_unregister_thread();