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.
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
35 #include <sys/types.h>
37 #include <urcu/uatomic.h>
41 #include <common/common.h>
42 #include <common/compat/socket.h>
43 #include <common/defaults.h>
44 #include <common/kernel-consumer/kernel-consumer.h>
45 #include <common/futex.h>
46 #include <common/relayd/relayd.h>
47 #include <common/utils.h>
48 #include <common/daemonize.h>
50 #include "lttng-sessiond.h"
51 #include "buffer-registry.h"
58 #include "kernel-consumer.h"
62 #include "ust-consumer.h"
65 #include "health-sessiond.h"
66 #include "testpoint.h"
67 #include "ust-thread.h"
68 #include "jul-thread.h"
70 #define CONSUMERD_FILE "lttng-consumerd"
73 static const char *tracing_group_name
= DEFAULT_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
, opt_background
;
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 */
81 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
84 /* Set to 1 when a SIGUSR1 signal is received. */
85 static int recv_child_signal
;
88 * Consumer daemon specific control data. Every value not initialized here is
89 * set to 0 by the static definition.
91 static struct consumer_data kconsumer_data
= {
92 .type
= LTTNG_CONSUMER_KERNEL
,
93 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
94 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
97 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
98 .lock
= PTHREAD_MUTEX_INITIALIZER
,
99 .cond
= PTHREAD_COND_INITIALIZER
,
100 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
102 static struct consumer_data ustconsumer64_data
= {
103 .type
= LTTNG_CONSUMER64_UST
,
104 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
105 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
108 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
109 .lock
= PTHREAD_MUTEX_INITIALIZER
,
110 .cond
= PTHREAD_COND_INITIALIZER
,
111 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
113 static struct consumer_data ustconsumer32_data
= {
114 .type
= LTTNG_CONSUMER32_UST
,
115 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
116 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
119 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
120 .lock
= PTHREAD_MUTEX_INITIALIZER
,
121 .cond
= PTHREAD_COND_INITIALIZER
,
122 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
125 /* Shared between threads */
126 static int dispatch_thread_exit
;
128 /* Global application Unix socket path */
129 static char apps_unix_sock_path
[PATH_MAX
];
130 /* Global client Unix socket path */
131 static char client_unix_sock_path
[PATH_MAX
];
132 /* global wait shm path for UST */
133 static char wait_shm_path
[PATH_MAX
];
134 /* Global health check unix path */
135 static char health_unix_sock_path
[PATH_MAX
];
137 /* Sockets and FDs */
138 static int client_sock
= -1;
139 static int apps_sock
= -1;
140 int kernel_tracer_fd
= -1;
141 static int kernel_poll_pipe
[2] = { -1, -1 };
144 * Quit pipe for all threads. This permits a single cancellation point
145 * for all threads when receiving an event on the pipe.
147 static int thread_quit_pipe
[2] = { -1, -1 };
150 * This pipe is used to inform the thread managing application communication
151 * that a command is queued and ready to be processed.
153 static int apps_cmd_pipe
[2] = { -1, -1 };
155 int apps_cmd_notify_pipe
[2] = { -1, -1 };
157 /* Pthread, Mutexes and Semaphores */
158 static pthread_t apps_thread
;
159 static pthread_t apps_notify_thread
;
160 static pthread_t reg_apps_thread
;
161 static pthread_t client_thread
;
162 static pthread_t kernel_thread
;
163 static pthread_t dispatch_thread
;
164 static pthread_t health_thread
;
165 static pthread_t ht_cleanup_thread
;
166 static pthread_t jul_reg_thread
;
169 * UST registration command queue. This queue is tied with a futex and uses a N
170 * wakers / 1 waiter implemented and detailed in futex.c/.h
172 * The thread_manage_apps and thread_dispatch_ust_registration interact with
173 * this queue and the wait/wake scheme.
175 static struct ust_cmd_queue ust_cmd_queue
;
178 * Pointer initialized before thread creation.
180 * This points to the tracing session list containing the session count and a
181 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
182 * MUST NOT be taken if you call a public function in session.c.
184 * The lock is nested inside the structure: session_list_ptr->lock. Please use
185 * session_lock_list and session_unlock_list for lock acquisition.
187 static struct ltt_session_list
*session_list_ptr
;
189 int ust_consumerd64_fd
= -1;
190 int ust_consumerd32_fd
= -1;
192 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
193 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
194 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
195 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
197 static const char *module_proc_lttng
= "/proc/lttng";
200 * Consumer daemon state which is changed when spawning it, killing it or in
201 * case of a fatal error.
203 enum consumerd_state
{
204 CONSUMER_STARTED
= 1,
205 CONSUMER_STOPPED
= 2,
210 * This consumer daemon state is used to validate if a client command will be
211 * able to reach the consumer. If not, the client is informed. For instance,
212 * doing a "lttng start" when the consumer state is set to ERROR will return an
213 * error to the client.
215 * The following example shows a possible race condition of this scheme:
217 * consumer thread error happens
219 * client cmd checks state -> still OK
220 * consumer thread exit, sets error
221 * client cmd try to talk to consumer
224 * However, since the consumer is a different daemon, we have no way of making
225 * sure the command will reach it safely even with this state flag. This is why
226 * we consider that up to the state validation during command processing, the
227 * command is safe. After that, we can not guarantee the correctness of the
228 * client request vis-a-vis the consumer.
230 static enum consumerd_state ust_consumerd_state
;
231 static enum consumerd_state kernel_consumerd_state
;
234 * Socket timeout for receiving and sending in seconds.
236 static int app_socket_timeout
;
238 /* Set in main() with the current page size. */
241 /* Application health monitoring */
242 struct health_app
*health_sessiond
;
244 /* JUL TCP port for registration. Used by the JUL thread. */
245 unsigned int jul_tcp_port
= DEFAULT_JUL_TCP_PORT
;
248 * Whether sessiond is ready for commands/health check requests.
249 * NR_LTTNG_SESSIOND_READY must match the number of calls to
250 * lttng_sessiond_notify_ready().
252 #define NR_LTTNG_SESSIOND_READY 2
253 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
255 /* Notify parents that we are ready for cmd and health check */
257 void lttng_sessiond_notify_ready(void)
259 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
261 * Notify parent pid that we are ready to accept command
262 * for client side. This ppid is the one from the
263 * external process that spawned us.
265 if (opt_sig_parent
) {
270 * Notify the parent of the fork() process that we are
273 if (opt_daemon
|| opt_background
) {
274 kill(child_ppid
, SIGUSR1
);
280 void setup_consumerd_path(void)
282 const char *bin
, *libdir
;
285 * Allow INSTALL_BIN_PATH to be used as a target path for the
286 * native architecture size consumer if CONFIG_CONSUMER*_PATH
287 * has not been defined.
289 #if (CAA_BITS_PER_LONG == 32)
290 if (!consumerd32_bin
[0]) {
291 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
293 if (!consumerd32_libdir
[0]) {
294 consumerd32_libdir
= INSTALL_LIB_PATH
;
296 #elif (CAA_BITS_PER_LONG == 64)
297 if (!consumerd64_bin
[0]) {
298 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
300 if (!consumerd64_libdir
[0]) {
301 consumerd64_libdir
= INSTALL_LIB_PATH
;
304 #error "Unknown bitness"
308 * runtime env. var. overrides the build default.
310 bin
= getenv("LTTNG_CONSUMERD32_BIN");
312 consumerd32_bin
= bin
;
314 bin
= getenv("LTTNG_CONSUMERD64_BIN");
316 consumerd64_bin
= bin
;
318 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
320 consumerd32_libdir
= libdir
;
322 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
324 consumerd64_libdir
= libdir
;
329 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
331 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
337 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
343 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
355 * Check if the thread quit pipe was triggered.
357 * Return 1 if it was triggered else 0;
359 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
361 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
369 * Init thread quit pipe.
371 * Return -1 on error or 0 if all pipes are created.
373 static int init_thread_quit_pipe(void)
377 ret
= pipe(thread_quit_pipe
);
379 PERROR("thread quit pipe");
383 for (i
= 0; i
< 2; i
++) {
384 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
396 * Stop all threads by closing the thread quit pipe.
398 static void stop_threads(void)
402 /* Stopping all threads */
403 DBG("Terminating all threads");
404 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
406 ERR("write error on thread quit pipe");
409 /* Dispatch thread */
410 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
411 futex_nto1_wake(&ust_cmd_queue
.futex
);
415 * Close every consumer sockets.
417 static void close_consumer_sockets(void)
421 if (kconsumer_data
.err_sock
>= 0) {
422 ret
= close(kconsumer_data
.err_sock
);
424 PERROR("kernel consumer err_sock close");
427 if (ustconsumer32_data
.err_sock
>= 0) {
428 ret
= close(ustconsumer32_data
.err_sock
);
430 PERROR("UST consumerd32 err_sock close");
433 if (ustconsumer64_data
.err_sock
>= 0) {
434 ret
= close(ustconsumer64_data
.err_sock
);
436 PERROR("UST consumerd64 err_sock close");
439 if (kconsumer_data
.cmd_sock
>= 0) {
440 ret
= close(kconsumer_data
.cmd_sock
);
442 PERROR("kernel consumer cmd_sock close");
445 if (ustconsumer32_data
.cmd_sock
>= 0) {
446 ret
= close(ustconsumer32_data
.cmd_sock
);
448 PERROR("UST consumerd32 cmd_sock close");
451 if (ustconsumer64_data
.cmd_sock
>= 0) {
452 ret
= close(ustconsumer64_data
.cmd_sock
);
454 PERROR("UST consumerd64 cmd_sock close");
462 static void cleanup(void)
465 struct ltt_session
*sess
, *stmp
;
471 * Close the thread quit pipe. It has already done its job,
472 * since we are now called.
474 utils_close_pipe(thread_quit_pipe
);
477 * If opt_pidfile is undefined, the default file will be wiped when
478 * removing the rundir.
481 ret
= remove(opt_pidfile
);
483 PERROR("remove pidfile %s", opt_pidfile
);
487 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
490 snprintf(path
, PATH_MAX
,
492 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
493 DBG("Removing %s", path
);
496 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
497 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
);
498 DBG("Removing %s", path
);
502 snprintf(path
, PATH_MAX
,
503 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
505 DBG("Removing %s", path
);
508 snprintf(path
, PATH_MAX
,
509 DEFAULT_KCONSUMERD_PATH
,
511 DBG("Removing directory %s", path
);
514 /* ust consumerd 32 */
515 snprintf(path
, PATH_MAX
,
516 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
518 DBG("Removing %s", path
);
521 snprintf(path
, PATH_MAX
,
522 DEFAULT_USTCONSUMERD32_PATH
,
524 DBG("Removing directory %s", path
);
527 /* ust consumerd 64 */
528 snprintf(path
, PATH_MAX
,
529 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
531 DBG("Removing %s", path
);
534 snprintf(path
, PATH_MAX
,
535 DEFAULT_USTCONSUMERD64_PATH
,
537 DBG("Removing directory %s", path
);
541 * We do NOT rmdir rundir because there are other processes
542 * using it, for instance lttng-relayd, which can start in
543 * parallel with this teardown.
548 DBG("Cleaning up all sessions");
550 /* Destroy session list mutex */
551 if (session_list_ptr
!= NULL
) {
552 pthread_mutex_destroy(&session_list_ptr
->lock
);
554 /* Cleanup ALL session */
555 cds_list_for_each_entry_safe(sess
, stmp
,
556 &session_list_ptr
->head
, list
) {
557 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
561 DBG("Closing all UST sockets");
562 ust_app_clean_list();
563 buffer_reg_destroy_registries();
565 if (is_root
&& !opt_no_kernel
) {
566 DBG2("Closing kernel fd");
567 if (kernel_tracer_fd
>= 0) {
568 ret
= close(kernel_tracer_fd
);
573 DBG("Unloading kernel modules");
574 modprobe_remove_lttng_all();
577 close_consumer_sockets();
580 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
581 "Matthew, BEET driven development works!%c[%dm",
582 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
587 * Send data on a unix socket using the liblttsessiondcomm API.
589 * Return lttcomm error code.
591 static int send_unix_sock(int sock
, void *buf
, size_t len
)
593 /* Check valid length */
598 return lttcomm_send_unix_sock(sock
, buf
, len
);
602 * Free memory of a command context structure.
604 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
606 DBG("Clean command context structure");
608 if ((*cmd_ctx
)->llm
) {
609 free((*cmd_ctx
)->llm
);
611 if ((*cmd_ctx
)->lsm
) {
612 free((*cmd_ctx
)->lsm
);
620 * Notify UST applications using the shm mmap futex.
622 static int notify_ust_apps(int active
)
626 DBG("Notifying applications of session daemon state: %d", active
);
628 /* See shm.c for this call implying mmap, shm and futex calls */
629 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
630 if (wait_shm_mmap
== NULL
) {
634 /* Wake waiting process */
635 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
637 /* Apps notified successfully */
645 * Setup the outgoing data buffer for the response (llm) by allocating the
646 * right amount of memory and copying the original information from the lsm
649 * Return total size of the buffer pointed by buf.
651 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
657 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
658 if (cmd_ctx
->llm
== NULL
) {
664 /* Copy common data */
665 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
666 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
668 cmd_ctx
->llm
->data_size
= size
;
669 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
678 * Update the kernel poll set of all channel fd available over all tracing
679 * session. Add the wakeup pipe at the end of the set.
681 static int update_kernel_poll(struct lttng_poll_event
*events
)
684 struct ltt_session
*session
;
685 struct ltt_kernel_channel
*channel
;
687 DBG("Updating kernel poll set");
690 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
691 session_lock(session
);
692 if (session
->kernel_session
== NULL
) {
693 session_unlock(session
);
697 cds_list_for_each_entry(channel
,
698 &session
->kernel_session
->channel_list
.head
, list
) {
699 /* Add channel fd to the kernel poll set */
700 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
702 session_unlock(session
);
705 DBG("Channel fd %d added to kernel set", channel
->fd
);
707 session_unlock(session
);
709 session_unlock_list();
714 session_unlock_list();
719 * Find the channel fd from 'fd' over all tracing session. When found, check
720 * for new channel stream and send those stream fds to the kernel consumer.
722 * Useful for CPU hotplug feature.
724 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
727 struct ltt_session
*session
;
728 struct ltt_kernel_session
*ksess
;
729 struct ltt_kernel_channel
*channel
;
731 DBG("Updating kernel streams for channel fd %d", fd
);
734 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
735 session_lock(session
);
736 if (session
->kernel_session
== NULL
) {
737 session_unlock(session
);
740 ksess
= session
->kernel_session
;
742 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
743 if (channel
->fd
== fd
) {
744 DBG("Channel found, updating kernel streams");
745 ret
= kernel_open_channel_stream(channel
);
749 /* Update the stream global counter */
750 ksess
->stream_count_global
+= ret
;
753 * Have we already sent fds to the consumer? If yes, it means
754 * that tracing is started so it is safe to send our updated
757 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
758 struct lttng_ht_iter iter
;
759 struct consumer_socket
*socket
;
762 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
763 &iter
.iter
, socket
, node
.node
) {
764 pthread_mutex_lock(socket
->lock
);
765 ret
= kernel_consumer_send_channel_stream(socket
,
767 session
->output_traces
? 1 : 0);
768 pthread_mutex_unlock(socket
->lock
);
779 session_unlock(session
);
781 session_unlock_list();
785 session_unlock(session
);
786 session_unlock_list();
791 * For each tracing session, update newly registered apps. The session list
792 * lock MUST be acquired before calling this.
794 static void update_ust_app(int app_sock
)
796 struct ltt_session
*sess
, *stmp
;
798 /* Consumer is in an ERROR state. Stop any application update. */
799 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
800 /* Stop the update process since the consumer is dead. */
804 /* For all tracing session(s) */
805 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
807 if (sess
->ust_session
) {
808 ust_app_global_update(sess
->ust_session
, app_sock
);
810 session_unlock(sess
);
815 * This thread manage event coming from the kernel.
817 * Features supported in this thread:
820 static void *thread_manage_kernel(void *data
)
822 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
823 uint32_t revents
, nb_fd
;
825 struct lttng_poll_event events
;
827 DBG("[thread] Thread manage kernel started");
829 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
832 * This first step of the while is to clean this structure which could free
833 * non NULL pointers so initialize it before the loop.
835 lttng_poll_init(&events
);
837 if (testpoint(sessiond_thread_manage_kernel
)) {
838 goto error_testpoint
;
841 health_code_update();
843 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
844 goto error_testpoint
;
848 health_code_update();
850 if (update_poll_flag
== 1) {
851 /* Clean events object. We are about to populate it again. */
852 lttng_poll_clean(&events
);
854 ret
= sessiond_set_thread_pollset(&events
, 2);
856 goto error_poll_create
;
859 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
864 /* This will add the available kernel channel if any. */
865 ret
= update_kernel_poll(&events
);
869 update_poll_flag
= 0;
872 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
874 /* Poll infinite value of time */
877 ret
= lttng_poll_wait(&events
, -1);
881 * Restart interrupted system call.
883 if (errno
== EINTR
) {
887 } else if (ret
== 0) {
888 /* Should not happen since timeout is infinite */
889 ERR("Return value of poll is 0 with an infinite timeout.\n"
890 "This should not have happened! Continuing...");
896 for (i
= 0; i
< nb_fd
; i
++) {
897 /* Fetch once the poll data */
898 revents
= LTTNG_POLL_GETEV(&events
, i
);
899 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
901 health_code_update();
903 /* Thread quit pipe has been closed. Killing thread. */
904 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
910 /* Check for data on kernel pipe */
911 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
912 (void) lttng_read(kernel_poll_pipe
[0],
915 * Ret value is useless here, if this pipe gets any actions an
916 * update is required anyway.
918 update_poll_flag
= 1;
922 * New CPU detected by the kernel. Adding kernel stream to
923 * kernel session and updating the kernel consumer
925 if (revents
& LPOLLIN
) {
926 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
932 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
933 * and unregister kernel stream at this point.
942 lttng_poll_clean(&events
);
945 utils_close_pipe(kernel_poll_pipe
);
946 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
949 ERR("Health error occurred in %s", __func__
);
950 WARN("Kernel thread died unexpectedly. "
951 "Kernel tracing can continue but CPU hotplug is disabled.");
953 health_unregister(health_sessiond
);
954 DBG("Kernel thread dying");
959 * Signal pthread condition of the consumer data that the thread.
961 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
963 pthread_mutex_lock(&data
->cond_mutex
);
966 * The state is set before signaling. It can be any value, it's the waiter
967 * job to correctly interpret this condition variable associated to the
968 * consumer pthread_cond.
970 * A value of 0 means that the corresponding thread of the consumer data
971 * was not started. 1 indicates that the thread has started and is ready
972 * for action. A negative value means that there was an error during the
975 data
->consumer_thread_is_ready
= state
;
976 (void) pthread_cond_signal(&data
->cond
);
978 pthread_mutex_unlock(&data
->cond_mutex
);
982 * This thread manage the consumer error sent back to the session daemon.
984 static void *thread_manage_consumer(void *data
)
986 int sock
= -1, i
, ret
, pollfd
, err
= -1;
987 uint32_t revents
, nb_fd
;
988 enum lttcomm_return_code code
;
989 struct lttng_poll_event events
;
990 struct consumer_data
*consumer_data
= data
;
992 DBG("[thread] Manage consumer started");
994 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
996 health_code_update();
999 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1000 * metadata_sock. Nothing more will be added to this poll set.
1002 ret
= sessiond_set_thread_pollset(&events
, 3);
1008 * The error socket here is already in a listening state which was done
1009 * just before spawning this thread to avoid a race between the consumer
1010 * daemon exec trying to connect and the listen() call.
1012 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1017 health_code_update();
1019 /* Infinite blocking call, waiting for transmission */
1021 health_poll_entry();
1023 if (testpoint(sessiond_thread_manage_consumer
)) {
1027 ret
= lttng_poll_wait(&events
, -1);
1031 * Restart interrupted system call.
1033 if (errno
== EINTR
) {
1041 for (i
= 0; i
< nb_fd
; i
++) {
1042 /* Fetch once the poll data */
1043 revents
= LTTNG_POLL_GETEV(&events
, i
);
1044 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1046 health_code_update();
1048 /* Thread quit pipe has been closed. Killing thread. */
1049 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1055 /* Event on the registration socket */
1056 if (pollfd
== consumer_data
->err_sock
) {
1057 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1058 ERR("consumer err socket poll error");
1064 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1070 * Set the CLOEXEC flag. Return code is useless because either way, the
1073 (void) utils_set_fd_cloexec(sock
);
1075 health_code_update();
1077 DBG2("Receiving code from consumer err_sock");
1079 /* Getting status code from kconsumerd */
1080 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1081 sizeof(enum lttcomm_return_code
));
1086 health_code_update();
1087 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1088 /* Connect both socket, command and metadata. */
1089 consumer_data
->cmd_sock
=
1090 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1091 consumer_data
->metadata_fd
=
1092 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1093 if (consumer_data
->cmd_sock
< 0
1094 || consumer_data
->metadata_fd
< 0) {
1095 PERROR("consumer connect cmd socket");
1096 /* On error, signal condition and quit. */
1097 signal_consumer_condition(consumer_data
, -1);
1100 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1101 /* Create metadata socket lock. */
1102 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1103 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1104 PERROR("zmalloc pthread mutex");
1108 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1110 signal_consumer_condition(consumer_data
, 1);
1111 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1112 DBG("Consumer metadata socket ready (fd: %d)",
1113 consumer_data
->metadata_fd
);
1115 ERR("consumer error when waiting for SOCK_READY : %s",
1116 lttcomm_get_readable_code(-code
));
1120 /* Remove the consumerd error sock since we've established a connexion */
1121 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1126 /* Add new accepted error socket. */
1127 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1132 /* Add metadata socket that is successfully connected. */
1133 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1134 LPOLLIN
| LPOLLRDHUP
);
1139 health_code_update();
1141 /* Infinite blocking call, waiting for transmission */
1144 health_poll_entry();
1145 ret
= lttng_poll_wait(&events
, -1);
1149 * Restart interrupted system call.
1151 if (errno
== EINTR
) {
1159 for (i
= 0; i
< nb_fd
; i
++) {
1160 /* Fetch once the poll data */
1161 revents
= LTTNG_POLL_GETEV(&events
, i
);
1162 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1164 health_code_update();
1166 /* Thread quit pipe has been closed. Killing thread. */
1167 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1173 if (pollfd
== sock
) {
1174 /* Event on the consumerd socket */
1175 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1176 ERR("consumer err socket second poll error");
1179 health_code_update();
1180 /* Wait for any kconsumerd error */
1181 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1182 sizeof(enum lttcomm_return_code
));
1184 ERR("consumer closed the command socket");
1188 ERR("consumer return code : %s",
1189 lttcomm_get_readable_code(-code
));
1192 } else if (pollfd
== consumer_data
->metadata_fd
) {
1193 /* UST metadata requests */
1194 ret
= ust_consumer_metadata_request(
1195 &consumer_data
->metadata_sock
);
1197 ERR("Handling metadata request");
1202 ERR("Unknown pollfd");
1206 health_code_update();
1212 * We lock here because we are about to close the sockets and some other
1213 * thread might be using them so get exclusive access which will abort all
1214 * other consumer command by other threads.
1216 pthread_mutex_lock(&consumer_data
->lock
);
1218 /* Immediately set the consumerd state to stopped */
1219 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1220 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1221 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1222 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1223 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1225 /* Code flow error... */
1229 if (consumer_data
->err_sock
>= 0) {
1230 ret
= close(consumer_data
->err_sock
);
1234 consumer_data
->err_sock
= -1;
1236 if (consumer_data
->cmd_sock
>= 0) {
1237 ret
= close(consumer_data
->cmd_sock
);
1241 consumer_data
->cmd_sock
= -1;
1243 if (consumer_data
->metadata_sock
.fd_ptr
&&
1244 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1245 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1257 unlink(consumer_data
->err_unix_sock_path
);
1258 unlink(consumer_data
->cmd_unix_sock_path
);
1259 consumer_data
->pid
= 0;
1260 pthread_mutex_unlock(&consumer_data
->lock
);
1262 /* Cleanup metadata socket mutex. */
1263 if (consumer_data
->metadata_sock
.lock
) {
1264 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1265 free(consumer_data
->metadata_sock
.lock
);
1267 lttng_poll_clean(&events
);
1271 ERR("Health error occurred in %s", __func__
);
1273 health_unregister(health_sessiond
);
1274 DBG("consumer thread cleanup completed");
1280 * This thread manage application communication.
1282 static void *thread_manage_apps(void *data
)
1284 int i
, ret
, pollfd
, err
= -1;
1286 uint32_t revents
, nb_fd
;
1287 struct lttng_poll_event events
;
1289 DBG("[thread] Manage application started");
1291 rcu_register_thread();
1292 rcu_thread_online();
1294 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1296 if (testpoint(sessiond_thread_manage_apps
)) {
1297 goto error_testpoint
;
1300 health_code_update();
1302 ret
= sessiond_set_thread_pollset(&events
, 2);
1304 goto error_poll_create
;
1307 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1312 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1316 health_code_update();
1319 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1321 /* Inifinite blocking call, waiting for transmission */
1323 health_poll_entry();
1324 ret
= lttng_poll_wait(&events
, -1);
1328 * Restart interrupted system call.
1330 if (errno
== EINTR
) {
1338 for (i
= 0; i
< nb_fd
; i
++) {
1339 /* Fetch once the poll data */
1340 revents
= LTTNG_POLL_GETEV(&events
, i
);
1341 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1343 health_code_update();
1345 /* Thread quit pipe has been closed. Killing thread. */
1346 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1352 /* Inspect the apps cmd pipe */
1353 if (pollfd
== apps_cmd_pipe
[0]) {
1354 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1355 ERR("Apps command pipe error");
1357 } else if (revents
& LPOLLIN
) {
1361 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1362 if (size_ret
< sizeof(sock
)) {
1363 PERROR("read apps cmd pipe");
1367 health_code_update();
1370 * We only monitor the error events of the socket. This
1371 * thread does not handle any incoming data from UST
1374 ret
= lttng_poll_add(&events
, sock
,
1375 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1380 DBG("Apps with sock %d added to poll set", sock
);
1384 * At this point, we know that a registered application made
1385 * the event at poll_wait.
1387 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1388 /* Removing from the poll set */
1389 ret
= lttng_poll_del(&events
, pollfd
);
1394 /* Socket closed on remote end. */
1395 ust_app_unregister(pollfd
);
1399 health_code_update();
1405 lttng_poll_clean(&events
);
1408 utils_close_pipe(apps_cmd_pipe
);
1409 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1412 * We don't clean the UST app hash table here since already registered
1413 * applications can still be controlled so let them be until the session
1414 * daemon dies or the applications stop.
1419 ERR("Health error occurred in %s", __func__
);
1421 health_unregister(health_sessiond
);
1422 DBG("Application communication apps thread cleanup complete");
1423 rcu_thread_offline();
1424 rcu_unregister_thread();
1429 * Send a socket to a thread This is called from the dispatch UST registration
1430 * thread once all sockets are set for the application.
1432 * The sock value can be invalid, we don't really care, the thread will handle
1433 * it and make the necessary cleanup if so.
1435 * On success, return 0 else a negative value being the errno message of the
1438 static int send_socket_to_thread(int fd
, int sock
)
1443 * It's possible that the FD is set as invalid with -1 concurrently just
1444 * before calling this function being a shutdown state of the thread.
1451 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1452 if (ret
< sizeof(sock
)) {
1453 PERROR("write apps pipe %d", fd
);
1460 /* All good. Don't send back the write positive ret value. */
1467 * Sanitize the wait queue of the dispatch registration thread meaning removing
1468 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1469 * notify socket is never received.
1471 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1473 int ret
, nb_fd
= 0, i
;
1474 unsigned int fd_added
= 0;
1475 struct lttng_poll_event events
;
1476 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1480 lttng_poll_init(&events
);
1482 /* Just skip everything for an empty queue. */
1483 if (!wait_queue
->count
) {
1487 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1492 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1493 &wait_queue
->head
, head
) {
1494 assert(wait_node
->app
);
1495 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1496 LPOLLHUP
| LPOLLERR
);
1509 * Poll but don't block so we can quickly identify the faulty events and
1510 * clean them afterwards from the wait queue.
1512 ret
= lttng_poll_wait(&events
, 0);
1518 for (i
= 0; i
< nb_fd
; i
++) {
1519 /* Get faulty FD. */
1520 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1521 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1523 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1524 &wait_queue
->head
, head
) {
1525 if (pollfd
== wait_node
->app
->sock
&&
1526 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1527 cds_list_del(&wait_node
->head
);
1528 wait_queue
->count
--;
1529 ust_app_destroy(wait_node
->app
);
1537 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1541 lttng_poll_clean(&events
);
1545 lttng_poll_clean(&events
);
1547 ERR("Unable to sanitize wait queue");
1552 * Dispatch request from the registration threads to the application
1553 * communication thread.
1555 static void *thread_dispatch_ust_registration(void *data
)
1558 struct cds_wfq_node
*node
;
1559 struct ust_command
*ust_cmd
= NULL
;
1560 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1561 struct ust_reg_wait_queue wait_queue
= {
1565 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1567 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1568 goto error_testpoint
;
1571 health_code_update();
1573 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1575 DBG("[thread] Dispatch UST command started");
1577 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1578 health_code_update();
1580 /* Atomically prepare the queue futex */
1581 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1584 struct ust_app
*app
= NULL
;
1588 * Make sure we don't have node(s) that have hung up before receiving
1589 * the notify socket. This is to clean the list in order to avoid
1590 * memory leaks from notify socket that are never seen.
1592 sanitize_wait_queue(&wait_queue
);
1594 health_code_update();
1595 /* Dequeue command for registration */
1596 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1598 DBG("Woken up but nothing in the UST command queue");
1599 /* Continue thread execution */
1603 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1605 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1606 " gid:%d sock:%d name:%s (version %d.%d)",
1607 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1608 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1609 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1610 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1612 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1613 wait_node
= zmalloc(sizeof(*wait_node
));
1615 PERROR("zmalloc wait_node dispatch");
1616 ret
= close(ust_cmd
->sock
);
1618 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1620 lttng_fd_put(LTTNG_FD_APPS
, 1);
1624 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1626 /* Create application object if socket is CMD. */
1627 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1629 if (!wait_node
->app
) {
1630 ret
= close(ust_cmd
->sock
);
1632 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1634 lttng_fd_put(LTTNG_FD_APPS
, 1);
1640 * Add application to the wait queue so we can set the notify
1641 * socket before putting this object in the global ht.
1643 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1648 * We have to continue here since we don't have the notify
1649 * socket and the application MUST be added to the hash table
1650 * only at that moment.
1655 * Look for the application in the local wait queue and set the
1656 * notify socket if found.
1658 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1659 &wait_queue
.head
, head
) {
1660 health_code_update();
1661 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1662 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1663 cds_list_del(&wait_node
->head
);
1665 app
= wait_node
->app
;
1667 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1673 * With no application at this stage the received socket is
1674 * basically useless so close it before we free the cmd data
1675 * structure for good.
1678 ret
= close(ust_cmd
->sock
);
1680 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1682 lttng_fd_put(LTTNG_FD_APPS
, 1);
1689 * @session_lock_list
1691 * Lock the global session list so from the register up to the
1692 * registration done message, no thread can see the application
1693 * and change its state.
1695 session_lock_list();
1699 * Add application to the global hash table. This needs to be
1700 * done before the update to the UST registry can locate the
1705 /* Set app version. This call will print an error if needed. */
1706 (void) ust_app_version(app
);
1708 /* Send notify socket through the notify pipe. */
1709 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1713 session_unlock_list();
1715 * No notify thread, stop the UST tracing. However, this is
1716 * not an internal error of the this thread thus setting
1717 * the health error code to a normal exit.
1724 * Update newly registered application with the tracing
1725 * registry info already enabled information.
1727 update_ust_app(app
->sock
);
1730 * Don't care about return value. Let the manage apps threads
1731 * handle app unregistration upon socket close.
1733 (void) ust_app_register_done(app
->sock
);
1736 * Even if the application socket has been closed, send the app
1737 * to the thread and unregistration will take place at that
1740 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1743 session_unlock_list();
1745 * No apps. thread, stop the UST tracing. However, this is
1746 * not an internal error of the this thread thus setting
1747 * the health error code to a normal exit.
1754 session_unlock_list();
1756 } while (node
!= NULL
);
1758 health_poll_entry();
1759 /* Futex wait on queue. Blocking call on futex() */
1760 futex_nto1_wait(&ust_cmd_queue
.futex
);
1763 /* Normal exit, no error */
1767 /* Clean up wait queue. */
1768 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1769 &wait_queue
.head
, head
) {
1770 cds_list_del(&wait_node
->head
);
1776 DBG("Dispatch thread dying");
1779 ERR("Health error occurred in %s", __func__
);
1781 health_unregister(health_sessiond
);
1786 * This thread manage application registration.
1788 static void *thread_registration_apps(void *data
)
1790 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1791 uint32_t revents
, nb_fd
;
1792 struct lttng_poll_event events
;
1794 * Get allocated in this thread, enqueued to a global queue, dequeued and
1795 * freed in the manage apps thread.
1797 struct ust_command
*ust_cmd
= NULL
;
1799 DBG("[thread] Manage application registration started");
1801 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1803 if (testpoint(sessiond_thread_registration_apps
)) {
1804 goto error_testpoint
;
1807 ret
= lttcomm_listen_unix_sock(apps_sock
);
1813 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1814 * more will be added to this poll set.
1816 ret
= sessiond_set_thread_pollset(&events
, 2);
1818 goto error_create_poll
;
1821 /* Add the application registration socket */
1822 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1824 goto error_poll_add
;
1827 /* Notify all applications to register */
1828 ret
= notify_ust_apps(1);
1830 ERR("Failed to notify applications or create the wait shared memory.\n"
1831 "Execution continues but there might be problem for already\n"
1832 "running applications that wishes to register.");
1836 DBG("Accepting application registration");
1838 /* Inifinite blocking call, waiting for transmission */
1840 health_poll_entry();
1841 ret
= lttng_poll_wait(&events
, -1);
1845 * Restart interrupted system call.
1847 if (errno
== EINTR
) {
1855 for (i
= 0; i
< nb_fd
; i
++) {
1856 health_code_update();
1858 /* Fetch once the poll data */
1859 revents
= LTTNG_POLL_GETEV(&events
, i
);
1860 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1862 /* Thread quit pipe has been closed. Killing thread. */
1863 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1869 /* Event on the registration socket */
1870 if (pollfd
== apps_sock
) {
1871 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1872 ERR("Register apps socket poll error");
1874 } else if (revents
& LPOLLIN
) {
1875 sock
= lttcomm_accept_unix_sock(apps_sock
);
1881 * Set socket timeout for both receiving and ending.
1882 * app_socket_timeout is in seconds, whereas
1883 * lttcomm_setsockopt_rcv_timeout and
1884 * lttcomm_setsockopt_snd_timeout expect msec as
1887 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1888 app_socket_timeout
* 1000);
1889 (void) lttcomm_setsockopt_snd_timeout(sock
,
1890 app_socket_timeout
* 1000);
1893 * Set the CLOEXEC flag. Return code is useless because
1894 * either way, the show must go on.
1896 (void) utils_set_fd_cloexec(sock
);
1898 /* Create UST registration command for enqueuing */
1899 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1900 if (ust_cmd
== NULL
) {
1901 PERROR("ust command zmalloc");
1906 * Using message-based transmissions to ensure we don't
1907 * have to deal with partially received messages.
1909 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1911 ERR("Exhausted file descriptors allowed for applications.");
1921 health_code_update();
1922 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1925 /* Close socket of the application. */
1930 lttng_fd_put(LTTNG_FD_APPS
, 1);
1934 health_code_update();
1936 ust_cmd
->sock
= sock
;
1939 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1940 " gid:%d sock:%d name:%s (version %d.%d)",
1941 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1942 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1943 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1944 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1947 * Lock free enqueue the registration request. The red pill
1948 * has been taken! This apps will be part of the *system*.
1950 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1953 * Wake the registration queue futex. Implicit memory
1954 * barrier with the exchange in cds_wfq_enqueue.
1956 futex_nto1_wake(&ust_cmd_queue
.futex
);
1964 /* Notify that the registration thread is gone */
1967 if (apps_sock
>= 0) {
1968 ret
= close(apps_sock
);
1978 lttng_fd_put(LTTNG_FD_APPS
, 1);
1980 unlink(apps_unix_sock_path
);
1983 lttng_poll_clean(&events
);
1987 DBG("UST Registration thread cleanup complete");
1990 ERR("Health error occurred in %s", __func__
);
1992 health_unregister(health_sessiond
);
1998 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1999 * exec or it will fails.
2001 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2004 struct timespec timeout
;
2006 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2007 consumer_data
->consumer_thread_is_ready
= 0;
2009 /* Setup pthread condition */
2010 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2013 PERROR("pthread_condattr_init consumer data");
2018 * Set the monotonic clock in order to make sure we DO NOT jump in time
2019 * between the clock_gettime() call and the timedwait call. See bug #324
2020 * for a more details and how we noticed it.
2022 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2025 PERROR("pthread_condattr_setclock consumer data");
2029 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2032 PERROR("pthread_cond_init consumer data");
2036 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2039 PERROR("pthread_create consumer");
2044 /* We are about to wait on a pthread condition */
2045 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2047 /* Get time for sem_timedwait absolute timeout */
2048 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2050 * Set the timeout for the condition timed wait even if the clock gettime
2051 * call fails since we might loop on that call and we want to avoid to
2052 * increment the timeout too many times.
2054 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2057 * The following loop COULD be skipped in some conditions so this is why we
2058 * set ret to 0 in order to make sure at least one round of the loop is
2064 * Loop until the condition is reached or when a timeout is reached. Note
2065 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2066 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2067 * possible. This loop does not take any chances and works with both of
2070 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2071 if (clock_ret
< 0) {
2072 PERROR("clock_gettime spawn consumer");
2073 /* Infinite wait for the consumerd thread to be ready */
2074 ret
= pthread_cond_wait(&consumer_data
->cond
,
2075 &consumer_data
->cond_mutex
);
2077 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2078 &consumer_data
->cond_mutex
, &timeout
);
2082 /* Release the pthread condition */
2083 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2087 if (ret
== ETIMEDOUT
) {
2091 * Call has timed out so we kill the kconsumerd_thread and return
2094 ERR("Condition timed out. The consumer thread was never ready."
2096 pth_ret
= pthread_cancel(consumer_data
->thread
);
2098 PERROR("pthread_cancel consumer thread");
2101 PERROR("pthread_cond_wait failed consumer thread");
2103 /* Caller is expecting a negative value on failure. */
2108 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2109 if (consumer_data
->pid
== 0) {
2110 ERR("Consumerd did not start");
2111 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2114 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2123 * Join consumer thread
2125 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2129 /* Consumer pid must be a real one. */
2130 if (consumer_data
->pid
> 0) {
2132 ret
= kill(consumer_data
->pid
, SIGTERM
);
2134 ERR("Error killing consumer daemon");
2137 return pthread_join(consumer_data
->thread
, &status
);
2144 * Fork and exec a consumer daemon (consumerd).
2146 * Return pid if successful else -1.
2148 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2152 const char *consumer_to_use
;
2153 const char *verbosity
;
2156 DBG("Spawning consumerd");
2163 if (opt_verbose_consumer
) {
2164 verbosity
= "--verbose";
2166 verbosity
= "--quiet";
2168 switch (consumer_data
->type
) {
2169 case LTTNG_CONSUMER_KERNEL
:
2171 * Find out which consumerd to execute. We will first try the
2172 * 64-bit path, then the sessiond's installation directory, and
2173 * fallback on the 32-bit one,
2175 DBG3("Looking for a kernel consumer at these locations:");
2176 DBG3(" 1) %s", consumerd64_bin
);
2177 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2178 DBG3(" 3) %s", consumerd32_bin
);
2179 if (stat(consumerd64_bin
, &st
) == 0) {
2180 DBG3("Found location #1");
2181 consumer_to_use
= consumerd64_bin
;
2182 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2183 DBG3("Found location #2");
2184 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2185 } else if (stat(consumerd32_bin
, &st
) == 0) {
2186 DBG3("Found location #3");
2187 consumer_to_use
= consumerd32_bin
;
2189 DBG("Could not find any valid consumerd executable");
2193 DBG("Using kernel consumer at: %s", consumer_to_use
);
2194 ret
= execl(consumer_to_use
,
2195 "lttng-consumerd", verbosity
, "-k",
2196 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2197 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2198 "--group", tracing_group_name
,
2201 case LTTNG_CONSUMER64_UST
:
2203 char *tmpnew
= NULL
;
2205 if (consumerd64_libdir
[0] != '\0') {
2209 tmp
= getenv("LD_LIBRARY_PATH");
2213 tmplen
= strlen("LD_LIBRARY_PATH=")
2214 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2215 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2220 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2221 strcat(tmpnew
, consumerd64_libdir
);
2222 if (tmp
[0] != '\0') {
2223 strcat(tmpnew
, ":");
2224 strcat(tmpnew
, tmp
);
2226 ret
= putenv(tmpnew
);
2233 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2234 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2235 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2236 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2237 "--group", tracing_group_name
,
2239 if (consumerd64_libdir
[0] != '\0') {
2244 case LTTNG_CONSUMER32_UST
:
2246 char *tmpnew
= NULL
;
2248 if (consumerd32_libdir
[0] != '\0') {
2252 tmp
= getenv("LD_LIBRARY_PATH");
2256 tmplen
= strlen("LD_LIBRARY_PATH=")
2257 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2258 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2263 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2264 strcat(tmpnew
, consumerd32_libdir
);
2265 if (tmp
[0] != '\0') {
2266 strcat(tmpnew
, ":");
2267 strcat(tmpnew
, tmp
);
2269 ret
= putenv(tmpnew
);
2276 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2277 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2278 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2279 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2280 "--group", tracing_group_name
,
2282 if (consumerd32_libdir
[0] != '\0') {
2288 PERROR("unknown consumer type");
2292 PERROR("Consumer execl()");
2294 /* Reaching this point, we got a failure on our execl(). */
2296 } else if (pid
> 0) {
2299 PERROR("start consumer fork");
2307 * Spawn the consumerd daemon and session daemon thread.
2309 static int start_consumerd(struct consumer_data
*consumer_data
)
2314 * Set the listen() state on the socket since there is a possible race
2315 * between the exec() of the consumer daemon and this call if place in the
2316 * consumer thread. See bug #366 for more details.
2318 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2323 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2324 if (consumer_data
->pid
!= 0) {
2325 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2329 ret
= spawn_consumerd(consumer_data
);
2331 ERR("Spawning consumerd failed");
2332 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2336 /* Setting up the consumer_data pid */
2337 consumer_data
->pid
= ret
;
2338 DBG2("Consumer pid %d", consumer_data
->pid
);
2339 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2341 DBG2("Spawning consumer control thread");
2342 ret
= spawn_consumer_thread(consumer_data
);
2344 ERR("Fatal error spawning consumer control thread");
2352 /* Cleanup already created sockets on error. */
2353 if (consumer_data
->err_sock
>= 0) {
2356 err
= close(consumer_data
->err_sock
);
2358 PERROR("close consumer data error socket");
2365 * Setup necessary data for kernel tracer action.
2367 static int init_kernel_tracer(void)
2371 /* Modprobe lttng kernel modules */
2372 ret
= modprobe_lttng_control();
2377 /* Open debugfs lttng */
2378 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2379 if (kernel_tracer_fd
< 0) {
2380 DBG("Failed to open %s", module_proc_lttng
);
2385 /* Validate kernel version */
2386 ret
= kernel_validate_version(kernel_tracer_fd
);
2391 ret
= modprobe_lttng_data();
2396 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2400 modprobe_remove_lttng_control();
2401 ret
= close(kernel_tracer_fd
);
2405 kernel_tracer_fd
= -1;
2406 return LTTNG_ERR_KERN_VERSION
;
2409 ret
= close(kernel_tracer_fd
);
2415 modprobe_remove_lttng_control();
2418 WARN("No kernel tracer available");
2419 kernel_tracer_fd
= -1;
2421 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2423 return LTTNG_ERR_KERN_NA
;
2429 * Copy consumer output from the tracing session to the domain session. The
2430 * function also applies the right modification on a per domain basis for the
2431 * trace files destination directory.
2433 * Should *NOT* be called with RCU read-side lock held.
2435 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2438 const char *dir_name
;
2439 struct consumer_output
*consumer
;
2442 assert(session
->consumer
);
2445 case LTTNG_DOMAIN_KERNEL
:
2446 DBG3("Copying tracing session consumer output in kernel session");
2448 * XXX: We should audit the session creation and what this function
2449 * does "extra" in order to avoid a destroy since this function is used
2450 * in the domain session creation (kernel and ust) only. Same for UST
2453 if (session
->kernel_session
->consumer
) {
2454 consumer_destroy_output(session
->kernel_session
->consumer
);
2456 session
->kernel_session
->consumer
=
2457 consumer_copy_output(session
->consumer
);
2458 /* Ease our life a bit for the next part */
2459 consumer
= session
->kernel_session
->consumer
;
2460 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2462 case LTTNG_DOMAIN_JUL
:
2463 case LTTNG_DOMAIN_UST
:
2464 DBG3("Copying tracing session consumer output in UST session");
2465 if (session
->ust_session
->consumer
) {
2466 consumer_destroy_output(session
->ust_session
->consumer
);
2468 session
->ust_session
->consumer
=
2469 consumer_copy_output(session
->consumer
);
2470 /* Ease our life a bit for the next part */
2471 consumer
= session
->ust_session
->consumer
;
2472 dir_name
= DEFAULT_UST_TRACE_DIR
;
2475 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2479 /* Append correct directory to subdir */
2480 strncat(consumer
->subdir
, dir_name
,
2481 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2482 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2491 * Create an UST session and add it to the session ust list.
2493 * Should *NOT* be called with RCU read-side lock held.
2495 static int create_ust_session(struct ltt_session
*session
,
2496 struct lttng_domain
*domain
)
2499 struct ltt_ust_session
*lus
= NULL
;
2503 assert(session
->consumer
);
2505 switch (domain
->type
) {
2506 case LTTNG_DOMAIN_JUL
:
2507 case LTTNG_DOMAIN_UST
:
2510 ERR("Unknown UST domain on create session %d", domain
->type
);
2511 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2515 DBG("Creating UST session");
2517 lus
= trace_ust_create_session(session
->id
);
2519 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2523 lus
->uid
= session
->uid
;
2524 lus
->gid
= session
->gid
;
2525 lus
->output_traces
= session
->output_traces
;
2526 lus
->snapshot_mode
= session
->snapshot_mode
;
2527 lus
->live_timer_interval
= session
->live_timer
;
2528 session
->ust_session
= lus
;
2530 /* Copy session output to the newly created UST session */
2531 ret
= copy_session_consumer(domain
->type
, session
);
2532 if (ret
!= LTTNG_OK
) {
2540 session
->ust_session
= NULL
;
2545 * Create a kernel tracer session then create the default channel.
2547 static int create_kernel_session(struct ltt_session
*session
)
2551 DBG("Creating kernel session");
2553 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2555 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2559 /* Code flow safety */
2560 assert(session
->kernel_session
);
2562 /* Copy session output to the newly created Kernel session */
2563 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2564 if (ret
!= LTTNG_OK
) {
2568 /* Create directory(ies) on local filesystem. */
2569 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2570 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2571 ret
= run_as_mkdir_recursive(
2572 session
->kernel_session
->consumer
->dst
.trace_path
,
2573 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2575 if (ret
!= -EEXIST
) {
2576 ERR("Trace directory creation error");
2582 session
->kernel_session
->uid
= session
->uid
;
2583 session
->kernel_session
->gid
= session
->gid
;
2584 session
->kernel_session
->output_traces
= session
->output_traces
;
2585 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2590 trace_kernel_destroy_session(session
->kernel_session
);
2591 session
->kernel_session
= NULL
;
2596 * Count number of session permitted by uid/gid.
2598 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2601 struct ltt_session
*session
;
2603 DBG("Counting number of available session for UID %d GID %d",
2605 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2607 * Only list the sessions the user can control.
2609 if (!session_access_ok(session
, uid
, gid
)) {
2618 * Process the command requested by the lttng client within the command
2619 * context structure. This function make sure that the return structure (llm)
2620 * is set and ready for transmission before returning.
2622 * Return any error encountered or 0 for success.
2624 * "sock" is only used for special-case var. len data.
2626 * Should *NOT* be called with RCU read-side lock held.
2628 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2632 int need_tracing_session
= 1;
2635 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2639 switch (cmd_ctx
->lsm
->cmd_type
) {
2640 case LTTNG_CREATE_SESSION
:
2641 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2642 case LTTNG_CREATE_SESSION_LIVE
:
2643 case LTTNG_DESTROY_SESSION
:
2644 case LTTNG_LIST_SESSIONS
:
2645 case LTTNG_LIST_DOMAINS
:
2646 case LTTNG_START_TRACE
:
2647 case LTTNG_STOP_TRACE
:
2648 case LTTNG_DATA_PENDING
:
2649 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2650 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2651 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2652 case LTTNG_SNAPSHOT_RECORD
:
2659 if (opt_no_kernel
&& need_domain
2660 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2662 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2664 ret
= LTTNG_ERR_KERN_NA
;
2669 /* Deny register consumer if we already have a spawned consumer. */
2670 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2671 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2672 if (kconsumer_data
.pid
> 0) {
2673 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2674 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2677 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2681 * Check for command that don't needs to allocate a returned payload. We do
2682 * this here so we don't have to make the call for no payload at each
2685 switch(cmd_ctx
->lsm
->cmd_type
) {
2686 case LTTNG_LIST_SESSIONS
:
2687 case LTTNG_LIST_TRACEPOINTS
:
2688 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2689 case LTTNG_LIST_DOMAINS
:
2690 case LTTNG_LIST_CHANNELS
:
2691 case LTTNG_LIST_EVENTS
:
2694 /* Setup lttng message with no payload */
2695 ret
= setup_lttng_msg(cmd_ctx
, 0);
2697 /* This label does not try to unlock the session */
2698 goto init_setup_error
;
2702 /* Commands that DO NOT need a session. */
2703 switch (cmd_ctx
->lsm
->cmd_type
) {
2704 case LTTNG_CREATE_SESSION
:
2705 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2706 case LTTNG_CREATE_SESSION_LIVE
:
2707 case LTTNG_CALIBRATE
:
2708 case LTTNG_LIST_SESSIONS
:
2709 case LTTNG_LIST_TRACEPOINTS
:
2710 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2711 need_tracing_session
= 0;
2714 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2716 * We keep the session list lock across _all_ commands
2717 * for now, because the per-session lock does not
2718 * handle teardown properly.
2720 session_lock_list();
2721 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2722 if (cmd_ctx
->session
== NULL
) {
2723 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2726 /* Acquire lock for the session */
2727 session_lock(cmd_ctx
->session
);
2737 * Check domain type for specific "pre-action".
2739 switch (cmd_ctx
->lsm
->domain
.type
) {
2740 case LTTNG_DOMAIN_KERNEL
:
2742 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2746 /* Kernel tracer check */
2747 if (kernel_tracer_fd
== -1) {
2748 /* Basically, load kernel tracer modules */
2749 ret
= init_kernel_tracer();
2755 /* Consumer is in an ERROR state. Report back to client */
2756 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2757 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2761 /* Need a session for kernel command */
2762 if (need_tracing_session
) {
2763 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2764 ret
= create_kernel_session(cmd_ctx
->session
);
2766 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2771 /* Start the kernel consumer daemon */
2772 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2773 if (kconsumer_data
.pid
== 0 &&
2774 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2775 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2776 ret
= start_consumerd(&kconsumer_data
);
2778 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2781 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2783 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2787 * The consumer was just spawned so we need to add the socket to
2788 * the consumer output of the session if exist.
2790 ret
= consumer_create_socket(&kconsumer_data
,
2791 cmd_ctx
->session
->kernel_session
->consumer
);
2798 case LTTNG_DOMAIN_JUL
:
2799 case LTTNG_DOMAIN_UST
:
2801 if (!ust_app_supported()) {
2802 ret
= LTTNG_ERR_NO_UST
;
2805 /* Consumer is in an ERROR state. Report back to client */
2806 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2807 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2811 if (need_tracing_session
) {
2812 /* Create UST session if none exist. */
2813 if (cmd_ctx
->session
->ust_session
== NULL
) {
2814 ret
= create_ust_session(cmd_ctx
->session
,
2815 &cmd_ctx
->lsm
->domain
);
2816 if (ret
!= LTTNG_OK
) {
2821 /* Start the UST consumer daemons */
2823 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2824 if (consumerd64_bin
[0] != '\0' &&
2825 ustconsumer64_data
.pid
== 0 &&
2826 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2827 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2828 ret
= start_consumerd(&ustconsumer64_data
);
2830 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2831 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2835 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2836 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2838 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2842 * Setup socket for consumer 64 bit. No need for atomic access
2843 * since it was set above and can ONLY be set in this thread.
2845 ret
= consumer_create_socket(&ustconsumer64_data
,
2846 cmd_ctx
->session
->ust_session
->consumer
);
2852 if (consumerd32_bin
[0] != '\0' &&
2853 ustconsumer32_data
.pid
== 0 &&
2854 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2855 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2856 ret
= start_consumerd(&ustconsumer32_data
);
2858 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2859 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2863 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2864 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2866 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2870 * Setup socket for consumer 64 bit. No need for atomic access
2871 * since it was set above and can ONLY be set in this thread.
2873 ret
= consumer_create_socket(&ustconsumer32_data
,
2874 cmd_ctx
->session
->ust_session
->consumer
);
2886 /* Validate consumer daemon state when start/stop trace command */
2887 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2888 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2889 switch (cmd_ctx
->lsm
->domain
.type
) {
2890 case LTTNG_DOMAIN_JUL
:
2891 case LTTNG_DOMAIN_UST
:
2892 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2893 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2897 case LTTNG_DOMAIN_KERNEL
:
2898 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2899 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2907 * Check that the UID or GID match that of the tracing session.
2908 * The root user can interact with all sessions.
2910 if (need_tracing_session
) {
2911 if (!session_access_ok(cmd_ctx
->session
,
2912 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2913 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2914 ret
= LTTNG_ERR_EPERM
;
2920 * Send relayd information to consumer as soon as we have a domain and a
2923 if (cmd_ctx
->session
&& need_domain
) {
2925 * Setup relayd if not done yet. If the relayd information was already
2926 * sent to the consumer, this call will gracefully return.
2928 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2929 if (ret
!= LTTNG_OK
) {
2934 /* Process by command type */
2935 switch (cmd_ctx
->lsm
->cmd_type
) {
2936 case LTTNG_ADD_CONTEXT
:
2938 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2939 cmd_ctx
->lsm
->u
.context
.channel_name
,
2940 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2943 case LTTNG_DISABLE_CHANNEL
:
2945 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2946 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2949 case LTTNG_DISABLE_EVENT
:
2951 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2952 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2953 cmd_ctx
->lsm
->u
.disable
.name
);
2956 case LTTNG_DISABLE_ALL_EVENT
:
2958 DBG("Disabling all events");
2960 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2961 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2964 case LTTNG_ENABLE_CHANNEL
:
2966 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2967 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2970 case LTTNG_ENABLE_EVENT
:
2972 struct lttng_event_exclusion
*exclusion
= NULL
;
2973 struct lttng_filter_bytecode
*bytecode
= NULL
;
2975 /* Handle exclusion events and receive it from the client. */
2976 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
2977 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
2979 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
2980 (count
* LTTNG_SYMBOL_NAME_LEN
));
2982 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
2986 DBG("Receiving var len exclusion event list from client ...");
2987 exclusion
->count
= count
;
2988 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
2989 count
* LTTNG_SYMBOL_NAME_LEN
);
2991 DBG("Nothing recv() from client var len data... continuing");
2994 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
2999 /* Handle filter and get bytecode from client. */
3000 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3001 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3003 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3004 ret
= LTTNG_ERR_FILTER_INVAL
;
3009 bytecode
= zmalloc(bytecode_len
);
3012 ret
= LTTNG_ERR_FILTER_NOMEM
;
3016 /* Receive var. len. data */
3017 DBG("Receiving var len filter's bytecode from client ...");
3018 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3020 DBG("Nothing recv() from client car len data... continuing");
3024 ret
= LTTNG_ERR_FILTER_INVAL
;
3028 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3031 ret
= LTTNG_ERR_FILTER_INVAL
;
3036 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3037 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3038 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, exclusion
,
3039 kernel_poll_pipe
[1]);
3042 case LTTNG_ENABLE_ALL_EVENT
:
3044 DBG("Enabling all events");
3046 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3047 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3048 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
3051 case LTTNG_LIST_TRACEPOINTS
:
3053 struct lttng_event
*events
;
3056 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3057 if (nb_events
< 0) {
3058 /* Return value is a negative lttng_error_code. */
3064 * Setup lttng message with payload size set to the event list size in
3065 * bytes and then copy list into the llm payload.
3067 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3073 /* Copy event list into message payload */
3074 memcpy(cmd_ctx
->llm
->payload
, events
,
3075 sizeof(struct lttng_event
) * nb_events
);
3082 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3084 struct lttng_event_field
*fields
;
3087 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3089 if (nb_fields
< 0) {
3090 /* Return value is a negative lttng_error_code. */
3096 * Setup lttng message with payload size set to the event list size in
3097 * bytes and then copy list into the llm payload.
3099 ret
= setup_lttng_msg(cmd_ctx
,
3100 sizeof(struct lttng_event_field
) * nb_fields
);
3106 /* Copy event list into message payload */
3107 memcpy(cmd_ctx
->llm
->payload
, fields
,
3108 sizeof(struct lttng_event_field
) * nb_fields
);
3115 case LTTNG_SET_CONSUMER_URI
:
3118 struct lttng_uri
*uris
;
3120 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3121 len
= nb_uri
* sizeof(struct lttng_uri
);
3124 ret
= LTTNG_ERR_INVALID
;
3128 uris
= zmalloc(len
);
3130 ret
= LTTNG_ERR_FATAL
;
3134 /* Receive variable len data */
3135 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3136 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3138 DBG("No URIs received from client... continuing");
3140 ret
= LTTNG_ERR_SESSION_FAIL
;
3145 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3147 if (ret
!= LTTNG_OK
) {
3153 * XXX: 0 means that this URI should be applied on the session. Should
3154 * be a DOMAIN enuam.
3156 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3157 /* Add the URI for the UST session if a consumer is present. */
3158 if (cmd_ctx
->session
->ust_session
&&
3159 cmd_ctx
->session
->ust_session
->consumer
) {
3160 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3162 } else if (cmd_ctx
->session
->kernel_session
&&
3163 cmd_ctx
->session
->kernel_session
->consumer
) {
3164 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3165 cmd_ctx
->session
, nb_uri
, uris
);
3173 case LTTNG_START_TRACE
:
3175 ret
= cmd_start_trace(cmd_ctx
->session
);
3178 case LTTNG_STOP_TRACE
:
3180 ret
= cmd_stop_trace(cmd_ctx
->session
);
3183 case LTTNG_CREATE_SESSION
:
3186 struct lttng_uri
*uris
= NULL
;
3188 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3189 len
= nb_uri
* sizeof(struct lttng_uri
);
3192 uris
= zmalloc(len
);
3194 ret
= LTTNG_ERR_FATAL
;
3198 /* Receive variable len data */
3199 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3200 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3202 DBG("No URIs received from client... continuing");
3204 ret
= LTTNG_ERR_SESSION_FAIL
;
3209 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3210 DBG("Creating session with ONE network URI is a bad call");
3211 ret
= LTTNG_ERR_SESSION_FAIL
;
3217 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3218 &cmd_ctx
->creds
, 0);
3224 case LTTNG_DESTROY_SESSION
:
3226 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3228 /* Set session to NULL so we do not unlock it after free. */
3229 cmd_ctx
->session
= NULL
;
3232 case LTTNG_LIST_DOMAINS
:
3235 struct lttng_domain
*domains
;
3237 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3239 /* Return value is a negative lttng_error_code. */
3244 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3250 /* Copy event list into message payload */
3251 memcpy(cmd_ctx
->llm
->payload
, domains
,
3252 nb_dom
* sizeof(struct lttng_domain
));
3259 case LTTNG_LIST_CHANNELS
:
3262 struct lttng_channel
*channels
;
3264 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3265 cmd_ctx
->session
, &channels
);
3267 /* Return value is a negative lttng_error_code. */
3272 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3278 /* Copy event list into message payload */
3279 memcpy(cmd_ctx
->llm
->payload
, channels
,
3280 nb_chan
* sizeof(struct lttng_channel
));
3287 case LTTNG_LIST_EVENTS
:
3290 struct lttng_event
*events
= NULL
;
3292 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3293 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3295 /* Return value is a negative lttng_error_code. */
3300 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3306 /* Copy event list into message payload */
3307 memcpy(cmd_ctx
->llm
->payload
, events
,
3308 nb_event
* sizeof(struct lttng_event
));
3315 case LTTNG_LIST_SESSIONS
:
3317 unsigned int nr_sessions
;
3319 session_lock_list();
3320 nr_sessions
= lttng_sessions_count(
3321 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3322 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3324 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3326 session_unlock_list();
3330 /* Filled the session array */
3331 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3332 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3333 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3335 session_unlock_list();
3340 case LTTNG_CALIBRATE
:
3342 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3343 &cmd_ctx
->lsm
->u
.calibrate
);
3346 case LTTNG_REGISTER_CONSUMER
:
3348 struct consumer_data
*cdata
;
3350 switch (cmd_ctx
->lsm
->domain
.type
) {
3351 case LTTNG_DOMAIN_KERNEL
:
3352 cdata
= &kconsumer_data
;
3355 ret
= LTTNG_ERR_UND
;
3359 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3360 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3363 case LTTNG_DATA_PENDING
:
3365 ret
= cmd_data_pending(cmd_ctx
->session
);
3368 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3370 struct lttcomm_lttng_output_id reply
;
3372 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3373 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3374 if (ret
!= LTTNG_OK
) {
3378 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3383 /* Copy output list into message payload */
3384 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3388 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3390 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3391 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3394 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3397 struct lttng_snapshot_output
*outputs
= NULL
;
3399 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3400 if (nb_output
< 0) {
3405 ret
= setup_lttng_msg(cmd_ctx
,
3406 nb_output
* sizeof(struct lttng_snapshot_output
));
3413 /* Copy output list into message payload */
3414 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3415 nb_output
* sizeof(struct lttng_snapshot_output
));
3422 case LTTNG_SNAPSHOT_RECORD
:
3424 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3425 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3426 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3429 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3432 struct lttng_uri
*uris
= NULL
;
3434 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3435 len
= nb_uri
* sizeof(struct lttng_uri
);
3438 uris
= zmalloc(len
);
3440 ret
= LTTNG_ERR_FATAL
;
3444 /* Receive variable len data */
3445 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3446 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3448 DBG("No URIs received from client... continuing");
3450 ret
= LTTNG_ERR_SESSION_FAIL
;
3455 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3456 DBG("Creating session with ONE network URI is a bad call");
3457 ret
= LTTNG_ERR_SESSION_FAIL
;
3463 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3464 nb_uri
, &cmd_ctx
->creds
);
3468 case LTTNG_CREATE_SESSION_LIVE
:
3471 struct lttng_uri
*uris
= NULL
;
3473 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3474 len
= nb_uri
* sizeof(struct lttng_uri
);
3477 uris
= zmalloc(len
);
3479 ret
= LTTNG_ERR_FATAL
;
3483 /* Receive variable len data */
3484 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3485 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3487 DBG("No URIs received from client... continuing");
3489 ret
= LTTNG_ERR_SESSION_FAIL
;
3494 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3495 DBG("Creating session with ONE network URI is a bad call");
3496 ret
= LTTNG_ERR_SESSION_FAIL
;
3502 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3503 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3508 ret
= LTTNG_ERR_UND
;
3513 if (cmd_ctx
->llm
== NULL
) {
3514 DBG("Missing llm structure. Allocating one.");
3515 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3519 /* Set return code */
3520 cmd_ctx
->llm
->ret_code
= ret
;
3522 if (cmd_ctx
->session
) {
3523 session_unlock(cmd_ctx
->session
);
3525 if (need_tracing_session
) {
3526 session_unlock_list();
3533 * Thread managing health check socket.
3535 static void *thread_manage_health(void *data
)
3537 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3538 uint32_t revents
, nb_fd
;
3539 struct lttng_poll_event events
;
3540 struct health_comm_msg msg
;
3541 struct health_comm_reply reply
;
3543 DBG("[thread] Manage health check started");
3545 rcu_register_thread();
3547 /* We might hit an error path before this is created. */
3548 lttng_poll_init(&events
);
3550 /* Create unix socket */
3551 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3553 ERR("Unable to create health check Unix socket");
3559 /* lttng health client socket path permissions */
3560 ret
= chown(health_unix_sock_path
, 0,
3561 utils_get_group_id(tracing_group_name
));
3563 ERR("Unable to set group on %s", health_unix_sock_path
);
3569 ret
= chmod(health_unix_sock_path
,
3570 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3572 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3580 * Set the CLOEXEC flag. Return code is useless because either way, the
3583 (void) utils_set_fd_cloexec(sock
);
3585 ret
= lttcomm_listen_unix_sock(sock
);
3591 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3592 * more will be added to this poll set.
3594 ret
= sessiond_set_thread_pollset(&events
, 2);
3599 /* Add the application registration socket */
3600 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3605 lttng_sessiond_notify_ready();
3608 DBG("Health check ready");
3610 /* Inifinite blocking call, waiting for transmission */
3612 ret
= lttng_poll_wait(&events
, -1);
3615 * Restart interrupted system call.
3617 if (errno
== EINTR
) {
3625 for (i
= 0; i
< nb_fd
; i
++) {
3626 /* Fetch once the poll data */
3627 revents
= LTTNG_POLL_GETEV(&events
, i
);
3628 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3630 /* Thread quit pipe has been closed. Killing thread. */
3631 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3637 /* Event on the registration socket */
3638 if (pollfd
== sock
) {
3639 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3640 ERR("Health socket poll error");
3646 new_sock
= lttcomm_accept_unix_sock(sock
);
3652 * Set the CLOEXEC flag. Return code is useless because either way, the
3655 (void) utils_set_fd_cloexec(new_sock
);
3657 DBG("Receiving data from client for health...");
3658 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3660 DBG("Nothing recv() from client... continuing");
3661 ret
= close(new_sock
);
3669 rcu_thread_online();
3671 memset(&reply
, 0, sizeof(reply
));
3672 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3674 * health_check_state returns 0 if health is
3677 if (!health_check_state(health_sessiond
, i
)) {
3678 reply
.ret_code
|= 1ULL << i
;
3682 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3684 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3686 ERR("Failed to send health data back to client");
3689 /* End of transmission */
3690 ret
= close(new_sock
);
3700 ERR("Health error occurred in %s", __func__
);
3702 DBG("Health check thread dying");
3703 unlink(health_unix_sock_path
);
3711 lttng_poll_clean(&events
);
3713 rcu_unregister_thread();
3718 * This thread manage all clients request using the unix client socket for
3721 static void *thread_manage_clients(void *data
)
3723 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3725 uint32_t revents
, nb_fd
;
3726 struct command_ctx
*cmd_ctx
= NULL
;
3727 struct lttng_poll_event events
;
3729 DBG("[thread] Manage client started");
3731 rcu_register_thread();
3733 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3735 health_code_update();
3737 ret
= lttcomm_listen_unix_sock(client_sock
);
3743 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3744 * more will be added to this poll set.
3746 ret
= sessiond_set_thread_pollset(&events
, 2);
3748 goto error_create_poll
;
3751 /* Add the application registration socket */
3752 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3757 lttng_sessiond_notify_ready();
3759 /* This testpoint is after we signal readiness to the parent. */
3760 if (testpoint(sessiond_thread_manage_clients
)) {
3764 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
3768 health_code_update();
3771 DBG("Accepting client command ...");
3773 /* Inifinite blocking call, waiting for transmission */
3775 health_poll_entry();
3776 ret
= lttng_poll_wait(&events
, -1);
3780 * Restart interrupted system call.
3782 if (errno
== EINTR
) {
3790 for (i
= 0; i
< nb_fd
; i
++) {
3791 /* Fetch once the poll data */
3792 revents
= LTTNG_POLL_GETEV(&events
, i
);
3793 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3795 health_code_update();
3797 /* Thread quit pipe has been closed. Killing thread. */
3798 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3804 /* Event on the registration socket */
3805 if (pollfd
== client_sock
) {
3806 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3807 ERR("Client socket poll error");
3813 DBG("Wait for client response");
3815 health_code_update();
3817 sock
= lttcomm_accept_unix_sock(client_sock
);
3823 * Set the CLOEXEC flag. Return code is useless because either way, the
3826 (void) utils_set_fd_cloexec(sock
);
3828 /* Set socket option for credentials retrieval */
3829 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3834 /* Allocate context command to process the client request */
3835 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3836 if (cmd_ctx
== NULL
) {
3837 PERROR("zmalloc cmd_ctx");
3841 /* Allocate data buffer for reception */
3842 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3843 if (cmd_ctx
->lsm
== NULL
) {
3844 PERROR("zmalloc cmd_ctx->lsm");
3848 cmd_ctx
->llm
= NULL
;
3849 cmd_ctx
->session
= NULL
;
3851 health_code_update();
3854 * Data is received from the lttng client. The struct
3855 * lttcomm_session_msg (lsm) contains the command and data request of
3858 DBG("Receiving data from client ...");
3859 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3860 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3862 DBG("Nothing recv() from client... continuing");
3868 clean_command_ctx(&cmd_ctx
);
3872 health_code_update();
3874 // TODO: Validate cmd_ctx including sanity check for
3875 // security purpose.
3877 rcu_thread_online();
3879 * This function dispatch the work to the kernel or userspace tracer
3880 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3881 * informations for the client. The command context struct contains
3882 * everything this function may needs.
3884 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3885 rcu_thread_offline();
3893 * TODO: Inform client somehow of the fatal error. At
3894 * this point, ret < 0 means that a zmalloc failed
3895 * (ENOMEM). Error detected but still accept
3896 * command, unless a socket error has been
3899 clean_command_ctx(&cmd_ctx
);
3903 health_code_update();
3905 DBG("Sending response (size: %d, retcode: %s)",
3906 cmd_ctx
->lttng_msg_size
,
3907 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3908 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3910 ERR("Failed to send data back to client");
3913 /* End of transmission */
3920 clean_command_ctx(&cmd_ctx
);
3922 health_code_update();
3934 lttng_poll_clean(&events
);
3935 clean_command_ctx(&cmd_ctx
);
3939 unlink(client_unix_sock_path
);
3940 if (client_sock
>= 0) {
3941 ret
= close(client_sock
);
3949 ERR("Health error occurred in %s", __func__
);
3952 health_unregister(health_sessiond
);
3954 DBG("Client thread dying");
3956 rcu_unregister_thread();
3962 * usage function on stderr
3964 static void usage(void)
3966 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3967 fprintf(stderr
, " -h, --help Display this usage.\n");
3968 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3969 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3970 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3971 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3972 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3973 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3974 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3975 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3976 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3977 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3978 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3979 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3980 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3981 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
3982 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3983 fprintf(stderr
, " -V, --version Show version number.\n");
3984 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
3985 fprintf(stderr
, " -q, --quiet No output at all.\n");
3986 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3987 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3988 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3989 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3990 fprintf(stderr
, " --jul-tcp-port JUL application registration TCP port\n");
3994 * daemon argument parsing
3996 static int parse_args(int argc
, char **argv
)
4000 static struct option long_options
[] = {
4001 { "client-sock", 1, 0, 'c' },
4002 { "apps-sock", 1, 0, 'a' },
4003 { "kconsumerd-cmd-sock", 1, 0, 'C' },
4004 { "kconsumerd-err-sock", 1, 0, 'E' },
4005 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
4006 { "ustconsumerd32-err-sock", 1, 0, 'H' },
4007 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
4008 { "ustconsumerd64-err-sock", 1, 0, 'F' },
4009 { "consumerd32-path", 1, 0, 'u' },
4010 { "consumerd32-libdir", 1, 0, 'U' },
4011 { "consumerd64-path", 1, 0, 't' },
4012 { "consumerd64-libdir", 1, 0, 'T' },
4013 { "daemonize", 0, 0, 'd' },
4014 { "sig-parent", 0, 0, 'S' },
4015 { "help", 0, 0, 'h' },
4016 { "group", 1, 0, 'g' },
4017 { "version", 0, 0, 'V' },
4018 { "quiet", 0, 0, 'q' },
4019 { "verbose", 0, 0, 'v' },
4020 { "verbose-consumer", 0, 0, 'Z' },
4021 { "no-kernel", 0, 0, 'N' },
4022 { "pidfile", 1, 0, 'p' },
4023 { "jul-tcp-port", 1, 0, 'J' },
4024 { "background", 0, 0, 'b' },
4029 int option_index
= 0;
4030 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:b",
4031 long_options
, &option_index
);
4038 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
4040 fprintf(stderr
, " with arg %s\n", optarg
);
4044 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4047 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4056 tracing_group_name
= optarg
;
4062 fprintf(stdout
, "%s\n", VERSION
);
4068 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4071 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4074 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4077 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4080 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4083 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4089 lttng_opt_quiet
= 1;
4092 /* Verbose level can increase using multiple -v */
4093 lttng_opt_verbose
+= 1;
4096 opt_verbose_consumer
+= 1;
4099 consumerd32_bin
= optarg
;
4102 consumerd32_libdir
= optarg
;
4105 consumerd64_bin
= optarg
;
4108 consumerd64_libdir
= optarg
;
4111 opt_pidfile
= optarg
;
4113 case 'J': /* JUL TCP port. */
4118 v
= strtoul(optarg
, NULL
, 0);
4119 if (errno
!= 0 || !isdigit(optarg
[0])) {
4120 ERR("Wrong value in --jul-tcp-port parameter: %s", optarg
);
4123 if (v
== 0 || v
>= 65535) {
4124 ERR("Port overflow in --jul-tcp-port parameter: %s", optarg
);
4127 jul_tcp_port
= (uint32_t) v
;
4128 DBG3("JUL TCP port set to non default: %u", jul_tcp_port
);
4132 /* Unknown option or other error.
4133 * Error is printed by getopt, just return */
4142 * Creates the two needed socket by the daemon.
4143 * apps_sock - The communication socket for all UST apps.
4144 * client_sock - The communication of the cli tool (lttng).
4146 static int init_daemon_socket(void)
4151 old_umask
= umask(0);
4153 /* Create client tool unix socket */
4154 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4155 if (client_sock
< 0) {
4156 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4161 /* Set the cloexec flag */
4162 ret
= utils_set_fd_cloexec(client_sock
);
4164 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4165 "Continuing but note that the consumer daemon will have a "
4166 "reference to this socket on exec()", client_sock
);
4169 /* File permission MUST be 660 */
4170 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4172 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4177 /* Create the application unix socket */
4178 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4179 if (apps_sock
< 0) {
4180 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4185 /* Set the cloexec flag */
4186 ret
= utils_set_fd_cloexec(apps_sock
);
4188 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4189 "Continuing but note that the consumer daemon will have a "
4190 "reference to this socket on exec()", apps_sock
);
4193 /* File permission MUST be 666 */
4194 ret
= chmod(apps_unix_sock_path
,
4195 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4197 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4202 DBG3("Session daemon client socket %d and application socket %d created",
4203 client_sock
, apps_sock
);
4211 * Check if the global socket is available, and if a daemon is answering at the
4212 * other side. If yes, error is returned.
4214 static int check_existing_daemon(void)
4216 /* Is there anybody out there ? */
4217 if (lttng_session_daemon_alive()) {
4225 * Set the tracing group gid onto the client socket.
4227 * Race window between mkdir and chown is OK because we are going from more
4228 * permissive (root.root) to less permissive (root.tracing).
4230 static int set_permissions(char *rundir
)
4235 gid
= utils_get_group_id(tracing_group_name
);
4237 /* Set lttng run dir */
4238 ret
= chown(rundir
, 0, gid
);
4240 ERR("Unable to set group on %s", rundir
);
4245 * Ensure all applications and tracing group can search the run
4246 * dir. Allow everyone to read the directory, since it does not
4247 * buy us anything to hide its content.
4249 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4251 ERR("Unable to set permissions on %s", rundir
);
4255 /* lttng client socket path */
4256 ret
= chown(client_unix_sock_path
, 0, gid
);
4258 ERR("Unable to set group on %s", client_unix_sock_path
);
4262 /* kconsumer error socket path */
4263 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4265 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4269 /* 64-bit ustconsumer error socket path */
4270 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4272 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4276 /* 32-bit ustconsumer compat32 error socket path */
4277 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4279 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4283 DBG("All permissions are set");
4289 * Create the lttng run directory needed for all global sockets and pipe.
4291 static int create_lttng_rundir(const char *rundir
)
4295 DBG3("Creating LTTng run directory: %s", rundir
);
4297 ret
= mkdir(rundir
, S_IRWXU
);
4299 if (errno
!= EEXIST
) {
4300 ERR("Unable to create %s", rundir
);
4312 * Setup sockets and directory needed by the kconsumerd communication with the
4315 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4319 char path
[PATH_MAX
];
4321 switch (consumer_data
->type
) {
4322 case LTTNG_CONSUMER_KERNEL
:
4323 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4325 case LTTNG_CONSUMER64_UST
:
4326 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4328 case LTTNG_CONSUMER32_UST
:
4329 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4332 ERR("Consumer type unknown");
4337 DBG2("Creating consumer directory: %s", path
);
4339 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4341 if (errno
!= EEXIST
) {
4343 ERR("Failed to create %s", path
);
4349 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4351 ERR("Unable to set group on %s", path
);
4357 /* Create the kconsumerd error unix socket */
4358 consumer_data
->err_sock
=
4359 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4360 if (consumer_data
->err_sock
< 0) {
4361 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4367 * Set the CLOEXEC flag. Return code is useless because either way, the
4370 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4372 PERROR("utils_set_fd_cloexec");
4373 /* continue anyway */
4376 /* File permission MUST be 660 */
4377 ret
= chmod(consumer_data
->err_unix_sock_path
,
4378 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4380 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4390 * Signal handler for the daemon
4392 * Simply stop all worker threads, leaving main() return gracefully after
4393 * joining all threads and calling cleanup().
4395 static void sighandler(int sig
)
4399 DBG("SIGPIPE caught");
4402 DBG("SIGINT caught");
4406 DBG("SIGTERM caught");
4410 CMM_STORE_SHARED(recv_child_signal
, 1);
4418 * Setup signal handler for :
4419 * SIGINT, SIGTERM, SIGPIPE
4421 static int set_signal_handler(void)
4424 struct sigaction sa
;
4427 if ((ret
= sigemptyset(&sigset
)) < 0) {
4428 PERROR("sigemptyset");
4432 sa
.sa_handler
= sighandler
;
4433 sa
.sa_mask
= sigset
;
4435 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4436 PERROR("sigaction");
4440 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4441 PERROR("sigaction");
4445 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4446 PERROR("sigaction");
4450 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4451 PERROR("sigaction");
4455 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4461 * Set open files limit to unlimited. This daemon can open a large number of
4462 * file descriptors in order to consumer multiple kernel traces.
4464 static void set_ulimit(void)
4469 /* The kernel does not allowed an infinite limit for open files */
4470 lim
.rlim_cur
= 65535;
4471 lim
.rlim_max
= 65535;
4473 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4475 PERROR("failed to set open files limit");
4480 * Write pidfile using the rundir and opt_pidfile.
4482 static void write_pidfile(void)
4485 char pidfile_path
[PATH_MAX
];
4490 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4492 /* Build pidfile path from rundir and opt_pidfile. */
4493 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4494 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4496 PERROR("snprintf pidfile path");
4502 * Create pid file in rundir. Return value is of no importance. The
4503 * execution will continue even though we are not able to write the file.
4505 (void) utils_create_pid_file(getpid(), pidfile_path
);
4512 * Write JUL TCP port using the rundir.
4514 static void write_julport(void)
4517 char path
[PATH_MAX
];
4521 ret
= snprintf(path
, sizeof(path
), "%s/"
4522 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
, rundir
);
4524 PERROR("snprintf julport path");
4529 * Create TCP JUL port file in rundir. Return value is of no importance.
4530 * The execution will continue even though we are not able to write the
4533 (void) utils_create_pid_file(jul_tcp_port
, path
);
4542 int main(int argc
, char **argv
)
4546 const char *home_path
, *env_app_timeout
;
4548 init_kernel_workarounds();
4550 rcu_register_thread();
4552 if ((ret
= set_signal_handler()) < 0) {
4556 setup_consumerd_path();
4558 page_size
= sysconf(_SC_PAGESIZE
);
4559 if (page_size
< 0) {
4560 PERROR("sysconf _SC_PAGESIZE");
4561 page_size
= LONG_MAX
;
4562 WARN("Fallback page size to %ld", page_size
);
4565 /* Parse arguments */
4567 if ((ret
= parse_args(argc
, argv
)) < 0) {
4572 if (opt_daemon
|| opt_background
) {
4575 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4582 * We are in the child. Make sure all other file descriptors are
4583 * closed, in case we are called with more opened file descriptors than
4584 * the standard ones.
4586 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4591 /* Create thread quit pipe */
4592 if ((ret
= init_thread_quit_pipe()) < 0) {
4596 /* Check if daemon is UID = 0 */
4597 is_root
= !getuid();
4600 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4602 /* Create global run dir with root access */
4603 ret
= create_lttng_rundir(rundir
);
4608 if (strlen(apps_unix_sock_path
) == 0) {
4609 snprintf(apps_unix_sock_path
, PATH_MAX
,
4610 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4613 if (strlen(client_unix_sock_path
) == 0) {
4614 snprintf(client_unix_sock_path
, PATH_MAX
,
4615 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4618 /* Set global SHM for ust */
4619 if (strlen(wait_shm_path
) == 0) {
4620 snprintf(wait_shm_path
, PATH_MAX
,
4621 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4624 if (strlen(health_unix_sock_path
) == 0) {
4625 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4626 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4629 /* Setup kernel consumerd path */
4630 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4631 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4632 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4633 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4635 DBG2("Kernel consumer err path: %s",
4636 kconsumer_data
.err_unix_sock_path
);
4637 DBG2("Kernel consumer cmd path: %s",
4638 kconsumer_data
.cmd_unix_sock_path
);
4640 home_path
= utils_get_home_dir();
4641 if (home_path
== NULL
) {
4642 /* TODO: Add --socket PATH option */
4643 ERR("Can't get HOME directory for sockets creation.");
4649 * Create rundir from home path. This will create something like
4652 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4658 ret
= create_lttng_rundir(rundir
);
4663 if (strlen(apps_unix_sock_path
) == 0) {
4664 snprintf(apps_unix_sock_path
, PATH_MAX
,
4665 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4668 /* Set the cli tool unix socket path */
4669 if (strlen(client_unix_sock_path
) == 0) {
4670 snprintf(client_unix_sock_path
, PATH_MAX
,
4671 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4674 /* Set global SHM for ust */
4675 if (strlen(wait_shm_path
) == 0) {
4676 snprintf(wait_shm_path
, PATH_MAX
,
4677 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4680 /* Set health check Unix path */
4681 if (strlen(health_unix_sock_path
) == 0) {
4682 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4683 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4687 /* Set consumer initial state */
4688 kernel_consumerd_state
= CONSUMER_STOPPED
;
4689 ust_consumerd_state
= CONSUMER_STOPPED
;
4691 DBG("Client socket path %s", client_unix_sock_path
);
4692 DBG("Application socket path %s", apps_unix_sock_path
);
4693 DBG("Application wait path %s", wait_shm_path
);
4694 DBG("LTTng run directory path: %s", rundir
);
4696 /* 32 bits consumerd path setup */
4697 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4698 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4699 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4700 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4702 DBG2("UST consumer 32 bits err path: %s",
4703 ustconsumer32_data
.err_unix_sock_path
);
4704 DBG2("UST consumer 32 bits cmd path: %s",
4705 ustconsumer32_data
.cmd_unix_sock_path
);
4707 /* 64 bits consumerd path setup */
4708 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4709 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4710 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4711 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4713 DBG2("UST consumer 64 bits err path: %s",
4714 ustconsumer64_data
.err_unix_sock_path
);
4715 DBG2("UST consumer 64 bits cmd path: %s",
4716 ustconsumer64_data
.cmd_unix_sock_path
);
4719 * See if daemon already exist.
4721 if ((ret
= check_existing_daemon()) < 0) {
4722 ERR("Already running daemon.\n");
4724 * We do not goto exit because we must not cleanup()
4725 * because a daemon is already running.
4731 * Init UST app hash table. Alloc hash table before this point since
4732 * cleanup() can get called after that point.
4736 /* Initialize JUL domain subsystem. */
4737 if ((ret
= jul_init()) < 0) {
4738 /* ENOMEM at this point. */
4742 /* After this point, we can safely call cleanup() with "goto exit" */
4745 * These actions must be executed as root. We do that *after* setting up
4746 * the sockets path because we MUST make the check for another daemon using
4747 * those paths *before* trying to set the kernel consumer sockets and init
4751 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4756 /* Setup kernel tracer */
4757 if (!opt_no_kernel
) {
4758 init_kernel_tracer();
4761 /* Set ulimit for open files */
4764 /* init lttng_fd tracking must be done after set_ulimit. */
4767 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4772 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4777 /* Setup the needed unix socket */
4778 if ((ret
= init_daemon_socket()) < 0) {
4782 /* Set credentials to socket */
4783 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4787 /* Get parent pid if -S, --sig-parent is specified. */
4788 if (opt_sig_parent
) {
4792 /* Setup the kernel pipe for waking up the kernel thread */
4793 if (is_root
&& !opt_no_kernel
) {
4794 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4799 /* Setup the thread ht_cleanup communication pipe. */
4800 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4804 /* Setup the thread apps communication pipe. */
4805 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4809 /* Setup the thread apps notify communication pipe. */
4810 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4814 /* Initialize global buffer per UID and PID registry. */
4815 buffer_reg_init_uid_registry();
4816 buffer_reg_init_pid_registry();
4818 /* Init UST command queue. */
4819 cds_wfq_init(&ust_cmd_queue
.queue
);
4822 * Get session list pointer. This pointer MUST NOT be free(). This list is
4823 * statically declared in session.c
4825 session_list_ptr
= session_get_list();
4827 /* Set up max poll set size */
4828 lttng_poll_set_max_size();
4832 /* Check for the application socket timeout env variable. */
4833 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4834 if (env_app_timeout
) {
4835 app_socket_timeout
= atoi(env_app_timeout
);
4837 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4843 /* Initialize communication library */
4845 /* This is to get the TCP timeout value. */
4846 lttcomm_inet_init();
4849 * Initialize the health check subsystem. This call should set the
4850 * appropriate time values.
4852 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
4853 if (!health_sessiond
) {
4854 PERROR("health_app_create error");
4855 goto exit_health_sessiond_cleanup
;
4858 /* Create thread to clean up RCU hash tables */
4859 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4860 thread_ht_cleanup
, (void *) NULL
);
4862 PERROR("pthread_create ht_cleanup");
4863 goto exit_ht_cleanup
;
4866 /* Create health-check thread */
4867 ret
= pthread_create(&health_thread
, NULL
,
4868 thread_manage_health
, (void *) NULL
);
4870 PERROR("pthread_create health");
4874 /* Create thread to manage the client socket */
4875 ret
= pthread_create(&client_thread
, NULL
,
4876 thread_manage_clients
, (void *) NULL
);
4878 PERROR("pthread_create clients");
4882 /* Create thread to dispatch registration */
4883 ret
= pthread_create(&dispatch_thread
, NULL
,
4884 thread_dispatch_ust_registration
, (void *) NULL
);
4886 PERROR("pthread_create dispatch");
4890 /* Create thread to manage application registration. */
4891 ret
= pthread_create(®_apps_thread
, NULL
,
4892 thread_registration_apps
, (void *) NULL
);
4894 PERROR("pthread_create registration");
4898 /* Create thread to manage application socket */
4899 ret
= pthread_create(&apps_thread
, NULL
,
4900 thread_manage_apps
, (void *) NULL
);
4902 PERROR("pthread_create apps");
4906 /* Create thread to manage application notify socket */
4907 ret
= pthread_create(&apps_notify_thread
, NULL
,
4908 ust_thread_manage_notify
, (void *) NULL
);
4910 PERROR("pthread_create apps");
4911 goto exit_apps_notify
;
4914 /* Create JUL registration thread. */
4915 ret
= pthread_create(&jul_reg_thread
, NULL
,
4916 jul_thread_manage_registration
, (void *) NULL
);
4918 PERROR("pthread_create apps");
4922 /* Don't start this thread if kernel tracing is not requested nor root */
4923 if (is_root
&& !opt_no_kernel
) {
4924 /* Create kernel thread to manage kernel event */
4925 ret
= pthread_create(&kernel_thread
, NULL
,
4926 thread_manage_kernel
, (void *) NULL
);
4928 PERROR("pthread_create kernel");
4932 ret
= pthread_join(kernel_thread
, &status
);
4934 PERROR("pthread_join");
4935 goto error
; /* join error, exit without cleanup */
4940 ret
= pthread_join(jul_reg_thread
, &status
);
4942 PERROR("pthread_join JUL");
4943 goto error
; /* join error, exit without cleanup */
4947 ret
= pthread_join(apps_notify_thread
, &status
);
4949 PERROR("pthread_join apps notify");
4950 goto error
; /* join error, exit without cleanup */
4954 ret
= pthread_join(apps_thread
, &status
);
4956 PERROR("pthread_join apps");
4957 goto error
; /* join error, exit without cleanup */
4962 ret
= pthread_join(reg_apps_thread
, &status
);
4964 PERROR("pthread_join");
4965 goto error
; /* join error, exit without cleanup */
4969 ret
= pthread_join(dispatch_thread
, &status
);
4971 PERROR("pthread_join");
4972 goto error
; /* join error, exit without cleanup */
4976 ret
= pthread_join(client_thread
, &status
);
4978 PERROR("pthread_join");
4979 goto error
; /* join error, exit without cleanup */
4982 ret
= join_consumer_thread(&kconsumer_data
);
4984 PERROR("join_consumer");
4985 goto error
; /* join error, exit without cleanup */
4988 ret
= join_consumer_thread(&ustconsumer32_data
);
4990 PERROR("join_consumer ust32");
4991 goto error
; /* join error, exit without cleanup */
4994 ret
= join_consumer_thread(&ustconsumer64_data
);
4996 PERROR("join_consumer ust64");
4997 goto error
; /* join error, exit without cleanup */
5001 ret
= pthread_join(health_thread
, &status
);
5003 PERROR("pthread_join health thread");
5004 goto error
; /* join error, exit without cleanup */
5008 ret
= pthread_join(ht_cleanup_thread
, &status
);
5010 PERROR("pthread_join ht cleanup thread");
5011 goto error
; /* join error, exit without cleanup */
5014 health_app_destroy(health_sessiond
);
5015 exit_health_sessiond_cleanup
:
5018 * cleanup() is called when no other thread is running.
5020 rcu_thread_online();
5022 rcu_thread_offline();
5023 rcu_unregister_thread();