2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
36 #include <sys/types.h>
38 #include <urcu/uatomic.h>
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
52 #include <common/dynamic-buffer.h>
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
62 #include "kernel-consumer.h"
66 #include "ust-consumer.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "ust-thread.h"
72 #include "agent-thread.h"
74 #include "load-session-thread.h"
75 #include "notification-thread.h"
76 #include "notification-thread-commands.h"
77 #include "rotation-thread.h"
78 #include "lttng-syscall.h"
80 #include "ht-cleanup.h"
81 #include "sessiond-config.h"
82 #include "sessiond-timer.h"
84 static const char *help_msg
=
85 #ifdef LTTNG_EMBED_HELP
86 #include <lttng-sessiond.8.h>
93 static pid_t ppid
; /* Parent PID for --sig-parent option */
94 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
95 static int lockfile_fd
= -1;
97 /* Set to 1 when a SIGUSR1 signal is received. */
98 static int recv_child_signal
;
100 static struct lttng_kernel_tracer_version kernel_tracer_version
;
101 static struct lttng_kernel_tracer_abi_version kernel_tracer_abi_version
;
104 * Consumer daemon specific control data. Every value not initialized here is
105 * set to 0 by the static definition.
107 static struct consumer_data kconsumer_data
= {
108 .type
= LTTNG_CONSUMER_KERNEL
,
111 .channel_monitor_pipe
= -1,
112 .channel_rotate_pipe
= -1,
113 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
114 .lock
= PTHREAD_MUTEX_INITIALIZER
,
115 .cond
= PTHREAD_COND_INITIALIZER
,
116 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
118 static struct consumer_data ustconsumer64_data
= {
119 .type
= LTTNG_CONSUMER64_UST
,
122 .channel_monitor_pipe
= -1,
123 .channel_rotate_pipe
= -1,
124 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
125 .lock
= PTHREAD_MUTEX_INITIALIZER
,
126 .cond
= PTHREAD_COND_INITIALIZER
,
127 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
129 static struct consumer_data ustconsumer32_data
= {
130 .type
= LTTNG_CONSUMER32_UST
,
133 .channel_monitor_pipe
= -1,
134 .channel_rotate_pipe
= -1,
135 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
136 .lock
= PTHREAD_MUTEX_INITIALIZER
,
137 .cond
= PTHREAD_COND_INITIALIZER
,
138 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
141 /* Command line options */
142 static const struct option long_options
[] = {
143 { "client-sock", required_argument
, 0, 'c' },
144 { "apps-sock", required_argument
, 0, 'a' },
145 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
146 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
147 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
148 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
149 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
150 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
151 { "consumerd32-path", required_argument
, 0, '\0' },
152 { "consumerd32-libdir", required_argument
, 0, '\0' },
153 { "consumerd64-path", required_argument
, 0, '\0' },
154 { "consumerd64-libdir", required_argument
, 0, '\0' },
155 { "daemonize", no_argument
, 0, 'd' },
156 { "background", no_argument
, 0, 'b' },
157 { "sig-parent", no_argument
, 0, 'S' },
158 { "help", no_argument
, 0, 'h' },
159 { "group", required_argument
, 0, 'g' },
160 { "version", no_argument
, 0, 'V' },
161 { "quiet", no_argument
, 0, 'q' },
162 { "verbose", no_argument
, 0, 'v' },
163 { "verbose-consumer", no_argument
, 0, '\0' },
164 { "no-kernel", no_argument
, 0, '\0' },
165 { "pidfile", required_argument
, 0, 'p' },
166 { "agent-tcp-port", required_argument
, 0, '\0' },
167 { "config", required_argument
, 0, 'f' },
168 { "load", required_argument
, 0, 'l' },
169 { "kmod-probes", required_argument
, 0, '\0' },
170 { "extra-kmod-probes", required_argument
, 0, '\0' },
174 struct sessiond_config config
;
176 /* Command line options to ignore from configuration file */
177 static const char *config_ignore_options
[] = { "help", "version", "config" };
179 /* Shared between threads */
180 static int dispatch_thread_exit
;
182 /* Sockets and FDs */
183 static int client_sock
= -1;
184 static int apps_sock
= -1;
185 int kernel_tracer_fd
= -1;
186 static int kernel_poll_pipe
[2] = { -1, -1 };
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
192 static int thread_quit_pipe
[2] = { -1, -1 };
195 * This pipe is used to inform the thread managing application communication
196 * that a command is queued and ready to be processed.
198 static int apps_cmd_pipe
[2] = { -1, -1 };
200 int apps_cmd_notify_pipe
[2] = { -1, -1 };
202 /* Pthread, Mutexes and Semaphores */
203 static pthread_t apps_thread
;
204 static pthread_t apps_notify_thread
;
205 static pthread_t reg_apps_thread
;
206 static pthread_t client_thread
;
207 static pthread_t kernel_thread
;
208 static pthread_t dispatch_thread
;
209 static pthread_t health_thread
;
210 static pthread_t ht_cleanup_thread
;
211 static pthread_t agent_reg_thread
;
212 static pthread_t load_session_thread
;
213 static pthread_t notification_thread
;
214 static pthread_t rotation_thread
;
215 static pthread_t timer_thread
;
218 * UST registration command queue. This queue is tied with a futex and uses a N
219 * wakers / 1 waiter implemented and detailed in futex.c/.h
221 * The thread_registration_apps and thread_dispatch_ust_registration uses this
222 * queue along with the wait/wake scheme. The thread_manage_apps receives down
223 * the line new application socket and monitors it for any I/O error or clean
224 * close that triggers an unregistration of the application.
226 static struct ust_cmd_queue ust_cmd_queue
;
229 * Pointer initialized before thread creation.
231 * This points to the tracing session list containing the session count and a
232 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
233 * MUST NOT be taken if you call a public function in session.c.
235 * The lock is nested inside the structure: session_list_ptr->lock. Please use
236 * session_lock_list and session_unlock_list for lock acquisition.
238 static struct ltt_session_list
*session_list_ptr
;
240 int ust_consumerd64_fd
= -1;
241 int ust_consumerd32_fd
= -1;
243 static const char *module_proc_lttng
= "/proc/lttng";
246 * Consumer daemon state which is changed when spawning it, killing it or in
247 * case of a fatal error.
249 enum consumerd_state
{
250 CONSUMER_STARTED
= 1,
251 CONSUMER_STOPPED
= 2,
256 * This consumer daemon state is used to validate if a client command will be
257 * able to reach the consumer. If not, the client is informed. For instance,
258 * doing a "lttng start" when the consumer state is set to ERROR will return an
259 * error to the client.
261 * The following example shows a possible race condition of this scheme:
263 * consumer thread error happens
265 * client cmd checks state -> still OK
266 * consumer thread exit, sets error
267 * client cmd try to talk to consumer
270 * However, since the consumer is a different daemon, we have no way of making
271 * sure the command will reach it safely even with this state flag. This is why
272 * we consider that up to the state validation during command processing, the
273 * command is safe. After that, we can not guarantee the correctness of the
274 * client request vis-a-vis the consumer.
276 static enum consumerd_state ust_consumerd_state
;
277 static enum consumerd_state kernel_consumerd_state
;
279 /* Set in main() with the current page size. */
282 /* Application health monitoring */
283 struct health_app
*health_sessiond
;
285 /* Am I root or not. */
286 int is_root
; /* Set to 1 if the daemon is running as root */
288 const char * const config_section_name
= "sessiond";
290 /* Load session thread information to operate. */
291 struct load_session_thread_data
*load_info
;
293 /* Notification thread handle. */
294 struct notification_thread_handle
*notification_thread_handle
;
296 /* Rotation thread handle. */
297 struct rotation_thread_handle
*rotation_thread_handle
;
299 /* Global hash tables */
300 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
303 * The initialization of the session daemon is done in multiple phases.
305 * While all threads are launched near-simultaneously, only some of them
306 * are needed to ensure the session daemon can start to respond to client
309 * There are two important guarantees that we wish to offer with respect
310 * to the initialisation of the session daemon:
311 * - When the daemonize/background launcher process exits, the sessiond
312 * is fully able to respond to client requests,
313 * - Auto-loaded sessions are visible to clients.
315 * In order to achieve this, a number of support threads have to be launched
316 * to allow the "client" thread to function properly. Moreover, since the
317 * "load session" thread needs the client thread, we must provide a way
318 * for the "load session" thread to know that the "client" thread is up
321 * Hence, the support threads decrement the lttng_sessiond_ready counter
322 * while the "client" threads waits for it to reach 0. Once the "client" thread
323 * unblocks, it posts the message_thread_ready semaphore which allows the
324 * "load session" thread to progress.
326 * This implies that the "load session" thread is the last to be initialized
327 * and will explicitly call sessiond_signal_parents(), which signals the parents
328 * that the session daemon is fully initialized.
330 * The four (4) support threads are:
332 * - notification_thread
336 #define NR_LTTNG_SESSIOND_SUPPORT_THREADS 4
337 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_SUPPORT_THREADS
;
339 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
341 return (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) ? 1 : 0;
344 /* Notify parents that we are ready for cmd and health check */
346 void sessiond_signal_parents(void)
349 * Notify parent pid that we are ready to accept command
350 * for client side. This ppid is the one from the
351 * external process that spawned us.
353 if (config
.sig_parent
) {
358 * Notify the parent of the fork() process that we are
361 if (config
.daemonize
|| config
.background
) {
362 kill(child_ppid
, SIGUSR1
);
367 void sessiond_notify_ready(void)
370 * This memory barrier is paired with the one performed by
371 * the client thread after it has seen that 'lttng_sessiond_ready' is 0.
373 * The purpose of these memory barriers is to ensure that all
374 * initialization operations of the various threads that call this
375 * function to signal that they are ready are commited/published
376 * before the client thread can see the 'lttng_sessiond_ready' counter
379 * Note that this could be a 'write' memory barrier, but a full barrier
380 * is used in case the code using this utility changes. The performance
381 * implications of this choice are minimal since this is a slow path.
384 uatomic_sub(<tng_sessiond_ready
, 1);
388 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
395 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
401 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
413 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
415 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
417 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
421 * Init thread quit pipe.
423 * Return -1 on error or 0 if all pipes are created.
425 static int __init_thread_quit_pipe(int *a_pipe
)
431 PERROR("thread quit pipe");
435 for (i
= 0; i
< 2; i
++) {
436 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
447 static int init_thread_quit_pipe(void)
449 return __init_thread_quit_pipe(thread_quit_pipe
);
453 * Stop all threads by closing the thread quit pipe.
455 static void stop_threads(void)
459 /* Stopping all threads */
460 DBG("Terminating all threads");
461 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
463 ERR("write error on thread quit pipe");
466 /* Dispatch thread */
467 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
468 futex_nto1_wake(&ust_cmd_queue
.futex
);
472 * Close every consumer sockets.
474 static void close_consumer_sockets(void)
478 if (kconsumer_data
.err_sock
>= 0) {
479 ret
= close(kconsumer_data
.err_sock
);
481 PERROR("kernel consumer err_sock close");
484 if (ustconsumer32_data
.err_sock
>= 0) {
485 ret
= close(ustconsumer32_data
.err_sock
);
487 PERROR("UST consumerd32 err_sock close");
490 if (ustconsumer64_data
.err_sock
>= 0) {
491 ret
= close(ustconsumer64_data
.err_sock
);
493 PERROR("UST consumerd64 err_sock close");
496 if (kconsumer_data
.cmd_sock
>= 0) {
497 ret
= close(kconsumer_data
.cmd_sock
);
499 PERROR("kernel consumer cmd_sock close");
502 if (ustconsumer32_data
.cmd_sock
>= 0) {
503 ret
= close(ustconsumer32_data
.cmd_sock
);
505 PERROR("UST consumerd32 cmd_sock close");
508 if (ustconsumer64_data
.cmd_sock
>= 0) {
509 ret
= close(ustconsumer64_data
.cmd_sock
);
511 PERROR("UST consumerd64 cmd_sock close");
514 if (kconsumer_data
.channel_monitor_pipe
>= 0) {
515 ret
= close(kconsumer_data
.channel_monitor_pipe
);
517 PERROR("kernel consumer channel monitor pipe close");
520 if (ustconsumer32_data
.channel_monitor_pipe
>= 0) {
521 ret
= close(ustconsumer32_data
.channel_monitor_pipe
);
523 PERROR("UST consumerd32 channel monitor pipe close");
526 if (ustconsumer64_data
.channel_monitor_pipe
>= 0) {
527 ret
= close(ustconsumer64_data
.channel_monitor_pipe
);
529 PERROR("UST consumerd64 channel monitor pipe close");
532 if (kconsumer_data
.channel_rotate_pipe
>= 0) {
533 ret
= close(kconsumer_data
.channel_rotate_pipe
);
535 PERROR("kernel consumer channel rotate pipe close");
538 if (ustconsumer32_data
.channel_rotate_pipe
>= 0) {
539 ret
= close(ustconsumer32_data
.channel_rotate_pipe
);
541 PERROR("UST consumerd32 channel rotate pipe close");
544 if (ustconsumer64_data
.channel_rotate_pipe
>= 0) {
545 ret
= close(ustconsumer64_data
.channel_rotate_pipe
);
547 PERROR("UST consumerd64 channel rotate pipe close");
553 * Wait on consumer process termination.
555 * Need to be called with the consumer data lock held or from a context
556 * ensuring no concurrent access to data (e.g: cleanup).
558 static void wait_consumer(struct consumer_data
*consumer_data
)
563 if (consumer_data
->pid
<= 0) {
567 DBG("Waiting for complete teardown of consumerd (PID: %d)",
569 ret
= waitpid(consumer_data
->pid
, &status
, 0);
571 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
572 } else if (!WIFEXITED(status
)) {
573 ERR("consumerd termination with error: %d",
576 consumer_data
->pid
= 0;
580 * Cleanup the session daemon's data structures.
582 static void sessiond_cleanup(void)
585 struct ltt_session
*sess
, *stmp
;
587 DBG("Cleanup sessiond");
590 * Close the thread quit pipe. It has already done its job,
591 * since we are now called.
593 utils_close_pipe(thread_quit_pipe
);
596 * If config.pid_file_path.value is undefined, the default file will be
597 * wiped when removing the rundir.
599 if (config
.pid_file_path
.value
) {
600 ret
= remove(config
.pid_file_path
.value
);
602 PERROR("remove pidfile %s", config
.pid_file_path
.value
);
606 DBG("Removing sessiond and consumerd content of directory %s",
607 config
.rundir
.value
);
610 DBG("Removing %s", config
.pid_file_path
.value
);
611 (void) unlink(config
.pid_file_path
.value
);
613 DBG("Removing %s", config
.agent_port_file_path
.value
);
614 (void) unlink(config
.agent_port_file_path
.value
);
617 DBG("Removing %s", kconsumer_data
.err_unix_sock_path
);
618 (void) unlink(kconsumer_data
.err_unix_sock_path
);
620 DBG("Removing directory %s", config
.kconsumerd_path
.value
);
621 (void) rmdir(config
.kconsumerd_path
.value
);
623 /* ust consumerd 32 */
624 DBG("Removing %s", config
.consumerd32_err_unix_sock_path
.value
);
625 (void) unlink(config
.consumerd32_err_unix_sock_path
.value
);
627 DBG("Removing directory %s", config
.consumerd32_path
.value
);
628 (void) rmdir(config
.consumerd32_path
.value
);
630 /* ust consumerd 64 */
631 DBG("Removing %s", config
.consumerd64_err_unix_sock_path
.value
);
632 (void) unlink(config
.consumerd64_err_unix_sock_path
.value
);
634 DBG("Removing directory %s", config
.consumerd64_path
.value
);
635 (void) rmdir(config
.consumerd64_path
.value
);
637 DBG("Cleaning up all sessions");
639 /* Destroy session list mutex */
640 if (session_list_ptr
!= NULL
) {
641 pthread_mutex_destroy(&session_list_ptr
->lock
);
643 /* Cleanup ALL session */
644 cds_list_for_each_entry_safe(sess
, stmp
,
645 &session_list_ptr
->head
, list
) {
646 cmd_destroy_session(sess
, kernel_poll_pipe
[1],
647 notification_thread_handle
);
651 wait_consumer(&kconsumer_data
);
652 wait_consumer(&ustconsumer64_data
);
653 wait_consumer(&ustconsumer32_data
);
655 DBG("Cleaning up all agent apps");
656 agent_app_ht_clean();
658 DBG("Closing all UST sockets");
659 ust_app_clean_list();
660 buffer_reg_destroy_registries();
662 if (is_root
&& !config
.no_kernel
) {
663 DBG2("Closing kernel fd");
664 if (kernel_tracer_fd
>= 0) {
665 ret
= close(kernel_tracer_fd
);
670 DBG("Unloading kernel modules");
671 modprobe_remove_lttng_all();
675 close_consumer_sockets();
678 load_session_destroy_data(load_info
);
683 * We do NOT rmdir rundir because there are other processes
684 * using it, for instance lttng-relayd, which can start in
685 * parallel with this teardown.
690 * Cleanup the daemon's option data structures.
692 static void sessiond_cleanup_options(void)
694 DBG("Cleaning up options");
696 sessiond_config_fini(&config
);
698 run_as_destroy_worker();
702 * Send data on a unix socket using the liblttsessiondcomm API.
704 * Return lttcomm error code.
706 static int send_unix_sock(int sock
, void *buf
, size_t len
)
708 /* Check valid length */
713 return lttcomm_send_unix_sock(sock
, buf
, len
);
717 * Free memory of a command context structure.
719 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
721 DBG("Clean command context structure");
723 if ((*cmd_ctx
)->llm
) {
724 free((*cmd_ctx
)->llm
);
726 if ((*cmd_ctx
)->lsm
) {
727 free((*cmd_ctx
)->lsm
);
735 * Notify UST applications using the shm mmap futex.
737 static int notify_ust_apps(int active
)
741 DBG("Notifying applications of session daemon state: %d", active
);
743 /* See shm.c for this call implying mmap, shm and futex calls */
744 wait_shm_mmap
= shm_ust_get_mmap(config
.wait_shm_path
.value
, is_root
);
745 if (wait_shm_mmap
== NULL
) {
749 /* Wake waiting process */
750 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
752 /* Apps notified successfully */
760 * Setup the outgoing data buffer for the response (llm) by allocating the
761 * right amount of memory and copying the original information from the lsm
764 * Return 0 on success, negative value on error.
766 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
767 const void *payload_buf
, size_t payload_len
,
768 const void *cmd_header_buf
, size_t cmd_header_len
)
771 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
772 const size_t cmd_header_offset
= header_len
;
773 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
774 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
776 cmd_ctx
->llm
= zmalloc(total_msg_size
);
778 if (cmd_ctx
->llm
== NULL
) {
784 /* Copy common data */
785 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
786 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
787 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
788 cmd_ctx
->llm
->data_size
= payload_len
;
789 cmd_ctx
->lttng_msg_size
= total_msg_size
;
791 /* Copy command header */
792 if (cmd_header_len
) {
793 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
799 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
808 * Version of setup_lttng_msg() without command header.
810 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
811 void *payload_buf
, size_t payload_len
)
813 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
816 * Update the kernel poll set of all channel fd available over all tracing
817 * session. Add the wakeup pipe at the end of the set.
819 static int update_kernel_poll(struct lttng_poll_event
*events
)
822 struct ltt_session
*session
;
823 struct ltt_kernel_channel
*channel
;
825 DBG("Updating kernel poll set");
828 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
829 session_lock(session
);
830 if (session
->kernel_session
== NULL
) {
831 session_unlock(session
);
835 cds_list_for_each_entry(channel
,
836 &session
->kernel_session
->channel_list
.head
, list
) {
837 /* Add channel fd to the kernel poll set */
838 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
840 session_unlock(session
);
843 DBG("Channel fd %d added to kernel set", channel
->fd
);
845 session_unlock(session
);
847 session_unlock_list();
852 session_unlock_list();
857 * Find the channel fd from 'fd' over all tracing session. When found, check
858 * for new channel stream and send those stream fds to the kernel consumer.
860 * Useful for CPU hotplug feature.
862 static int update_kernel_stream(int fd
)
865 struct ltt_session
*session
;
866 struct ltt_kernel_session
*ksess
;
867 struct ltt_kernel_channel
*channel
;
869 DBG("Updating kernel streams for channel fd %d", fd
);
872 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
873 session_lock(session
);
874 if (session
->kernel_session
== NULL
) {
875 session_unlock(session
);
878 ksess
= session
->kernel_session
;
880 cds_list_for_each_entry(channel
,
881 &ksess
->channel_list
.head
, list
) {
882 struct lttng_ht_iter iter
;
883 struct consumer_socket
*socket
;
885 if (channel
->fd
!= fd
) {
888 DBG("Channel found, updating kernel streams");
889 ret
= kernel_open_channel_stream(channel
);
893 /* Update the stream global counter */
894 ksess
->stream_count_global
+= ret
;
897 * Have we already sent fds to the consumer? If yes, it
898 * means that tracing is started so it is safe to send
899 * our updated stream fds.
901 if (ksess
->consumer_fds_sent
!= 1
902 || ksess
->consumer
== NULL
) {
908 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
909 &iter
.iter
, socket
, node
.node
) {
910 pthread_mutex_lock(socket
->lock
);
911 ret
= kernel_consumer_send_channel_streams(socket
,
913 session
->output_traces
? 1 : 0);
914 pthread_mutex_unlock(socket
->lock
);
922 session_unlock(session
);
924 session_unlock_list();
928 session_unlock(session
);
929 session_unlock_list();
934 * For each tracing session, update newly registered apps. The session list
935 * lock MUST be acquired before calling this.
937 static void update_ust_app(int app_sock
)
939 struct ltt_session
*sess
, *stmp
;
941 /* Consumer is in an ERROR state. Stop any application update. */
942 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
943 /* Stop the update process since the consumer is dead. */
947 /* For all tracing session(s) */
948 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
952 if (!sess
->ust_session
) {
957 assert(app_sock
>= 0);
958 app
= ust_app_find_by_sock(app_sock
);
961 * Application can be unregistered before so
962 * this is possible hence simply stopping the
965 DBG3("UST app update failed to find app sock %d",
969 ust_app_global_update(sess
->ust_session
, app
);
973 session_unlock(sess
);
978 * This thread manage event coming from the kernel.
980 * Features supported in this thread:
983 static void *thread_manage_kernel(void *data
)
985 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
986 uint32_t revents
, nb_fd
;
988 struct lttng_poll_event events
;
990 DBG("[thread] Thread manage kernel started");
992 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
995 * This first step of the while is to clean this structure which could free
996 * non NULL pointers so initialize it before the loop.
998 lttng_poll_init(&events
);
1000 if (testpoint(sessiond_thread_manage_kernel
)) {
1001 goto error_testpoint
;
1004 health_code_update();
1006 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1007 goto error_testpoint
;
1011 health_code_update();
1013 if (update_poll_flag
== 1) {
1014 /* Clean events object. We are about to populate it again. */
1015 lttng_poll_clean(&events
);
1017 ret
= sessiond_set_thread_pollset(&events
, 2);
1019 goto error_poll_create
;
1022 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1027 /* This will add the available kernel channel if any. */
1028 ret
= update_kernel_poll(&events
);
1032 update_poll_flag
= 0;
1035 DBG("Thread kernel polling");
1037 /* Poll infinite value of time */
1039 health_poll_entry();
1040 ret
= lttng_poll_wait(&events
, -1);
1041 DBG("Thread kernel return from poll on %d fds",
1042 LTTNG_POLL_GETNB(&events
));
1046 * Restart interrupted system call.
1048 if (errno
== EINTR
) {
1052 } else if (ret
== 0) {
1053 /* Should not happen since timeout is infinite */
1054 ERR("Return value of poll is 0 with an infinite timeout.\n"
1055 "This should not have happened! Continuing...");
1061 for (i
= 0; i
< nb_fd
; i
++) {
1062 /* Fetch once the poll data */
1063 revents
= LTTNG_POLL_GETEV(&events
, i
);
1064 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1066 health_code_update();
1069 /* No activity for this FD (poll implementation). */
1073 /* Thread quit pipe has been closed. Killing thread. */
1074 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1080 /* Check for data on kernel pipe */
1081 if (revents
& LPOLLIN
) {
1082 if (pollfd
== kernel_poll_pipe
[0]) {
1083 (void) lttng_read(kernel_poll_pipe
[0],
1086 * Ret value is useless here, if this pipe gets any actions an
1087 * update is required anyway.
1089 update_poll_flag
= 1;
1093 * New CPU detected by the kernel. Adding kernel stream to
1094 * kernel session and updating the kernel consumer
1096 ret
= update_kernel_stream(pollfd
);
1102 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1103 update_poll_flag
= 1;
1106 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1114 lttng_poll_clean(&events
);
1117 utils_close_pipe(kernel_poll_pipe
);
1118 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1121 ERR("Health error occurred in %s", __func__
);
1122 WARN("Kernel thread died unexpectedly. "
1123 "Kernel tracing can continue but CPU hotplug is disabled.");
1125 health_unregister(health_sessiond
);
1126 DBG("Kernel thread dying");
1131 * Signal pthread condition of the consumer data that the thread.
1133 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1135 pthread_mutex_lock(&data
->cond_mutex
);
1138 * The state is set before signaling. It can be any value, it's the waiter
1139 * job to correctly interpret this condition variable associated to the
1140 * consumer pthread_cond.
1142 * A value of 0 means that the corresponding thread of the consumer data
1143 * was not started. 1 indicates that the thread has started and is ready
1144 * for action. A negative value means that there was an error during the
1147 data
->consumer_thread_is_ready
= state
;
1148 (void) pthread_cond_signal(&data
->cond
);
1150 pthread_mutex_unlock(&data
->cond_mutex
);
1154 * This thread manage the consumer error sent back to the session daemon.
1156 static void *thread_manage_consumer(void *data
)
1158 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1159 uint32_t revents
, nb_fd
;
1160 enum lttcomm_return_code code
;
1161 struct lttng_poll_event events
;
1162 struct consumer_data
*consumer_data
= data
;
1163 struct consumer_socket
*cmd_socket_wrapper
= NULL
;
1165 DBG("[thread] Manage consumer started");
1167 rcu_register_thread();
1168 rcu_thread_online();
1170 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1172 health_code_update();
1175 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1176 * metadata_sock. Nothing more will be added to this poll set.
1178 ret
= sessiond_set_thread_pollset(&events
, 3);
1184 * The error socket here is already in a listening state which was done
1185 * just before spawning this thread to avoid a race between the consumer
1186 * daemon exec trying to connect and the listen() call.
1188 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1193 health_code_update();
1195 /* Infinite blocking call, waiting for transmission */
1197 health_poll_entry();
1199 if (testpoint(sessiond_thread_manage_consumer
)) {
1203 ret
= lttng_poll_wait(&events
, -1);
1207 * Restart interrupted system call.
1209 if (errno
== EINTR
) {
1217 for (i
= 0; i
< nb_fd
; i
++) {
1218 /* Fetch once the poll data */
1219 revents
= LTTNG_POLL_GETEV(&events
, i
);
1220 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1222 health_code_update();
1225 /* No activity for this FD (poll implementation). */
1229 /* Thread quit pipe has been closed. Killing thread. */
1230 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1236 /* Event on the registration socket */
1237 if (pollfd
== consumer_data
->err_sock
) {
1238 if (revents
& LPOLLIN
) {
1240 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1241 ERR("consumer err socket poll error");
1244 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1250 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1256 * Set the CLOEXEC flag. Return code is useless because either way, the
1259 (void) utils_set_fd_cloexec(sock
);
1261 health_code_update();
1263 DBG2("Receiving code from consumer err_sock");
1265 /* Getting status code from kconsumerd */
1266 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1267 sizeof(enum lttcomm_return_code
));
1272 health_code_update();
1273 if (code
!= LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1274 ERR("consumer error when waiting for SOCK_READY : %s",
1275 lttcomm_get_readable_code(-code
));
1279 /* Connect both command and metadata sockets. */
1280 consumer_data
->cmd_sock
=
1281 lttcomm_connect_unix_sock(
1282 consumer_data
->cmd_unix_sock_path
);
1283 consumer_data
->metadata_fd
=
1284 lttcomm_connect_unix_sock(
1285 consumer_data
->cmd_unix_sock_path
);
1286 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
1287 PERROR("consumer connect cmd socket");
1288 /* On error, signal condition and quit. */
1289 signal_consumer_condition(consumer_data
, -1);
1293 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1295 /* Create metadata socket lock. */
1296 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1297 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1298 PERROR("zmalloc pthread mutex");
1301 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1303 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1304 DBG("Consumer metadata socket ready (fd: %d)",
1305 consumer_data
->metadata_fd
);
1308 * Remove the consumerd error sock since we've established a connection.
1310 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1315 /* Add new accepted error socket. */
1316 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1321 /* Add metadata socket that is successfully connected. */
1322 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1323 LPOLLIN
| LPOLLRDHUP
);
1328 health_code_update();
1331 * Transfer the write-end of the channel monitoring and rotate pipe
1332 * to the consumer by issuing a SET_CHANNEL_MONITOR_PIPE and
1333 * SET_CHANNEL_ROTATE_PIPE commands.
1335 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
1336 if (!cmd_socket_wrapper
) {
1339 cmd_socket_wrapper
->lock
= &consumer_data
->lock
;
1341 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
1342 consumer_data
->channel_monitor_pipe
);
1347 ret
= consumer_send_channel_rotate_pipe(cmd_socket_wrapper
,
1348 consumer_data
->channel_rotate_pipe
);
1353 /* Discard the socket wrapper as it is no longer needed. */
1354 consumer_destroy_socket(cmd_socket_wrapper
);
1355 cmd_socket_wrapper
= NULL
;
1357 /* The thread is completely initialized, signal that it is ready. */
1358 signal_consumer_condition(consumer_data
, 1);
1360 /* Infinite blocking call, waiting for transmission */
1363 health_code_update();
1365 /* Exit the thread because the thread quit pipe has been triggered. */
1367 /* Not a health error. */
1372 health_poll_entry();
1373 ret
= lttng_poll_wait(&events
, -1);
1377 * Restart interrupted system call.
1379 if (errno
== EINTR
) {
1387 for (i
= 0; i
< nb_fd
; i
++) {
1388 /* Fetch once the poll data */
1389 revents
= LTTNG_POLL_GETEV(&events
, i
);
1390 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1392 health_code_update();
1395 /* No activity for this FD (poll implementation). */
1400 * Thread quit pipe has been triggered, flag that we should stop
1401 * but continue the current loop to handle potential data from
1404 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1406 if (pollfd
== sock
) {
1407 /* Event on the consumerd socket */
1408 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1409 && !(revents
& LPOLLIN
)) {
1410 ERR("consumer err socket second poll error");
1413 health_code_update();
1414 /* Wait for any kconsumerd error */
1415 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1416 sizeof(enum lttcomm_return_code
));
1418 ERR("consumer closed the command socket");
1422 ERR("consumer return code : %s",
1423 lttcomm_get_readable_code(-code
));
1426 } else if (pollfd
== consumer_data
->metadata_fd
) {
1427 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1428 && !(revents
& LPOLLIN
)) {
1429 ERR("consumer err metadata socket second poll error");
1432 /* UST metadata requests */
1433 ret
= ust_consumer_metadata_request(
1434 &consumer_data
->metadata_sock
);
1436 ERR("Handling metadata request");
1440 /* No need for an else branch all FDs are tested prior. */
1442 health_code_update();
1448 * We lock here because we are about to close the sockets and some other
1449 * thread might be using them so get exclusive access which will abort all
1450 * other consumer command by other threads.
1452 pthread_mutex_lock(&consumer_data
->lock
);
1454 /* Immediately set the consumerd state to stopped */
1455 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1456 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1457 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1458 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1459 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1461 /* Code flow error... */
1465 if (consumer_data
->err_sock
>= 0) {
1466 ret
= close(consumer_data
->err_sock
);
1470 consumer_data
->err_sock
= -1;
1472 if (consumer_data
->cmd_sock
>= 0) {
1473 ret
= close(consumer_data
->cmd_sock
);
1477 consumer_data
->cmd_sock
= -1;
1479 if (consumer_data
->metadata_sock
.fd_ptr
&&
1480 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1481 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1493 unlink(consumer_data
->err_unix_sock_path
);
1494 unlink(consumer_data
->cmd_unix_sock_path
);
1495 pthread_mutex_unlock(&consumer_data
->lock
);
1497 /* Cleanup metadata socket mutex. */
1498 if (consumer_data
->metadata_sock
.lock
) {
1499 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1500 free(consumer_data
->metadata_sock
.lock
);
1502 lttng_poll_clean(&events
);
1504 if (cmd_socket_wrapper
) {
1505 consumer_destroy_socket(cmd_socket_wrapper
);
1510 ERR("Health error occurred in %s", __func__
);
1512 health_unregister(health_sessiond
);
1513 DBG("consumer thread cleanup completed");
1515 rcu_thread_offline();
1516 rcu_unregister_thread();
1522 * This thread receives application command sockets (FDs) on the
1523 * apps_cmd_pipe and waits (polls) on them until they are closed
1524 * or an error occurs.
1526 * At that point, it flushes the data (tracing and metadata) associated
1527 * with this application and tears down ust app sessions and other
1528 * associated data structures through ust_app_unregister().
1530 * Note that this thread never sends commands to the applications
1531 * through the command sockets; it merely listens for hang-ups
1532 * and errors on those sockets and cleans-up as they occur.
1534 static void *thread_manage_apps(void *data
)
1536 int i
, ret
, pollfd
, err
= -1;
1538 uint32_t revents
, nb_fd
;
1539 struct lttng_poll_event events
;
1541 DBG("[thread] Manage application started");
1543 rcu_register_thread();
1544 rcu_thread_online();
1546 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1548 if (testpoint(sessiond_thread_manage_apps
)) {
1549 goto error_testpoint
;
1552 health_code_update();
1554 ret
= sessiond_set_thread_pollset(&events
, 2);
1556 goto error_poll_create
;
1559 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1564 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1568 health_code_update();
1571 DBG("Apps thread polling");
1573 /* Inifinite blocking call, waiting for transmission */
1575 health_poll_entry();
1576 ret
= lttng_poll_wait(&events
, -1);
1577 DBG("Apps thread return from poll on %d fds",
1578 LTTNG_POLL_GETNB(&events
));
1582 * Restart interrupted system call.
1584 if (errno
== EINTR
) {
1592 for (i
= 0; i
< nb_fd
; i
++) {
1593 /* Fetch once the poll data */
1594 revents
= LTTNG_POLL_GETEV(&events
, i
);
1595 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1597 health_code_update();
1600 /* No activity for this FD (poll implementation). */
1604 /* Thread quit pipe has been closed. Killing thread. */
1605 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1611 /* Inspect the apps cmd pipe */
1612 if (pollfd
== apps_cmd_pipe
[0]) {
1613 if (revents
& LPOLLIN
) {
1617 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1618 if (size_ret
< sizeof(sock
)) {
1619 PERROR("read apps cmd pipe");
1623 health_code_update();
1626 * Since this is a command socket (write then read),
1627 * we only monitor the error events of the socket.
1629 ret
= lttng_poll_add(&events
, sock
,
1630 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1635 DBG("Apps with sock %d added to poll set", sock
);
1636 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1637 ERR("Apps command pipe error");
1640 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1645 * At this point, we know that a registered application made
1646 * the event at poll_wait.
1648 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1649 /* Removing from the poll set */
1650 ret
= lttng_poll_del(&events
, pollfd
);
1655 /* Socket closed on remote end. */
1656 ust_app_unregister(pollfd
);
1658 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1663 health_code_update();
1669 lttng_poll_clean(&events
);
1672 utils_close_pipe(apps_cmd_pipe
);
1673 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1676 * We don't clean the UST app hash table here since already registered
1677 * applications can still be controlled so let them be until the session
1678 * daemon dies or the applications stop.
1683 ERR("Health error occurred in %s", __func__
);
1685 health_unregister(health_sessiond
);
1686 DBG("Application communication apps thread cleanup complete");
1687 rcu_thread_offline();
1688 rcu_unregister_thread();
1693 * Send a socket to a thread This is called from the dispatch UST registration
1694 * thread once all sockets are set for the application.
1696 * The sock value can be invalid, we don't really care, the thread will handle
1697 * it and make the necessary cleanup if so.
1699 * On success, return 0 else a negative value being the errno message of the
1702 static int send_socket_to_thread(int fd
, int sock
)
1707 * It's possible that the FD is set as invalid with -1 concurrently just
1708 * before calling this function being a shutdown state of the thread.
1715 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1716 if (ret
< sizeof(sock
)) {
1717 PERROR("write apps pipe %d", fd
);
1724 /* All good. Don't send back the write positive ret value. */
1731 * Sanitize the wait queue of the dispatch registration thread meaning removing
1732 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1733 * notify socket is never received.
1735 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1737 int ret
, nb_fd
= 0, i
;
1738 unsigned int fd_added
= 0;
1739 struct lttng_poll_event events
;
1740 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1744 lttng_poll_init(&events
);
1746 /* Just skip everything for an empty queue. */
1747 if (!wait_queue
->count
) {
1751 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1756 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1757 &wait_queue
->head
, head
) {
1758 assert(wait_node
->app
);
1759 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1760 LPOLLHUP
| LPOLLERR
);
1773 * Poll but don't block so we can quickly identify the faulty events and
1774 * clean them afterwards from the wait queue.
1776 ret
= lttng_poll_wait(&events
, 0);
1782 for (i
= 0; i
< nb_fd
; i
++) {
1783 /* Get faulty FD. */
1784 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1785 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1788 /* No activity for this FD (poll implementation). */
1792 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1793 &wait_queue
->head
, head
) {
1794 if (pollfd
== wait_node
->app
->sock
&&
1795 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1796 cds_list_del(&wait_node
->head
);
1797 wait_queue
->count
--;
1798 ust_app_destroy(wait_node
->app
);
1801 * Silence warning of use-after-free in
1802 * cds_list_for_each_entry_safe which uses
1803 * __typeof__(*wait_node).
1808 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1815 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1819 lttng_poll_clean(&events
);
1823 lttng_poll_clean(&events
);
1825 ERR("Unable to sanitize wait queue");
1830 * Dispatch request from the registration threads to the application
1831 * communication thread.
1833 static void *thread_dispatch_ust_registration(void *data
)
1836 struct cds_wfcq_node
*node
;
1837 struct ust_command
*ust_cmd
= NULL
;
1838 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1839 struct ust_reg_wait_queue wait_queue
= {
1843 rcu_register_thread();
1845 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1847 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1848 goto error_testpoint
;
1851 health_code_update();
1853 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1855 DBG("[thread] Dispatch UST command started");
1858 health_code_update();
1860 /* Atomically prepare the queue futex */
1861 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1863 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1868 struct ust_app
*app
= NULL
;
1872 * Make sure we don't have node(s) that have hung up before receiving
1873 * the notify socket. This is to clean the list in order to avoid
1874 * memory leaks from notify socket that are never seen.
1876 sanitize_wait_queue(&wait_queue
);
1878 health_code_update();
1879 /* Dequeue command for registration */
1880 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1882 DBG("Woken up but nothing in the UST command queue");
1883 /* Continue thread execution */
1887 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1889 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1890 " gid:%d sock:%d name:%s (version %d.%d)",
1891 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1892 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1893 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1894 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1896 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1897 wait_node
= zmalloc(sizeof(*wait_node
));
1899 PERROR("zmalloc wait_node dispatch");
1900 ret
= close(ust_cmd
->sock
);
1902 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1904 lttng_fd_put(LTTNG_FD_APPS
, 1);
1908 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1910 /* Create application object if socket is CMD. */
1911 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1913 if (!wait_node
->app
) {
1914 ret
= close(ust_cmd
->sock
);
1916 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1918 lttng_fd_put(LTTNG_FD_APPS
, 1);
1924 * Add application to the wait queue so we can set the notify
1925 * socket before putting this object in the global ht.
1927 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1932 * We have to continue here since we don't have the notify
1933 * socket and the application MUST be added to the hash table
1934 * only at that moment.
1939 * Look for the application in the local wait queue and set the
1940 * notify socket if found.
1942 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1943 &wait_queue
.head
, head
) {
1944 health_code_update();
1945 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1946 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1947 cds_list_del(&wait_node
->head
);
1949 app
= wait_node
->app
;
1951 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1957 * With no application at this stage the received socket is
1958 * basically useless so close it before we free the cmd data
1959 * structure for good.
1962 ret
= close(ust_cmd
->sock
);
1964 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1966 lttng_fd_put(LTTNG_FD_APPS
, 1);
1973 * @session_lock_list
1975 * Lock the global session list so from the register up to the
1976 * registration done message, no thread can see the application
1977 * and change its state.
1979 session_lock_list();
1983 * Add application to the global hash table. This needs to be
1984 * done before the update to the UST registry can locate the
1989 /* Set app version. This call will print an error if needed. */
1990 (void) ust_app_version(app
);
1992 /* Send notify socket through the notify pipe. */
1993 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1997 session_unlock_list();
1999 * No notify thread, stop the UST tracing. However, this is
2000 * not an internal error of the this thread thus setting
2001 * the health error code to a normal exit.
2008 * Update newly registered application with the tracing
2009 * registry info already enabled information.
2011 update_ust_app(app
->sock
);
2014 * Don't care about return value. Let the manage apps threads
2015 * handle app unregistration upon socket close.
2017 (void) ust_app_register_done(app
);
2020 * Even if the application socket has been closed, send the app
2021 * to the thread and unregistration will take place at that
2024 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2027 session_unlock_list();
2029 * No apps. thread, stop the UST tracing. However, this is
2030 * not an internal error of the this thread thus setting
2031 * the health error code to a normal exit.
2038 session_unlock_list();
2040 } while (node
!= NULL
);
2042 health_poll_entry();
2043 /* Futex wait on queue. Blocking call on futex() */
2044 futex_nto1_wait(&ust_cmd_queue
.futex
);
2047 /* Normal exit, no error */
2051 /* Clean up wait queue. */
2052 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2053 &wait_queue
.head
, head
) {
2054 cds_list_del(&wait_node
->head
);
2059 /* Empty command queue. */
2061 /* Dequeue command for registration */
2062 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2066 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2067 ret
= close(ust_cmd
->sock
);
2069 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2071 lttng_fd_put(LTTNG_FD_APPS
, 1);
2076 DBG("Dispatch thread dying");
2079 ERR("Health error occurred in %s", __func__
);
2081 health_unregister(health_sessiond
);
2082 rcu_unregister_thread();
2087 * This thread manage application registration.
2089 static void *thread_registration_apps(void *data
)
2091 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2092 uint32_t revents
, nb_fd
;
2093 struct lttng_poll_event events
;
2095 * Get allocated in this thread, enqueued to a global queue, dequeued and
2096 * freed in the manage apps thread.
2098 struct ust_command
*ust_cmd
= NULL
;
2100 DBG("[thread] Manage application registration started");
2102 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2104 if (testpoint(sessiond_thread_registration_apps
)) {
2105 goto error_testpoint
;
2108 ret
= lttcomm_listen_unix_sock(apps_sock
);
2114 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2115 * more will be added to this poll set.
2117 ret
= sessiond_set_thread_pollset(&events
, 2);
2119 goto error_create_poll
;
2122 /* Add the application registration socket */
2123 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2125 goto error_poll_add
;
2128 /* Notify all applications to register */
2129 ret
= notify_ust_apps(1);
2131 ERR("Failed to notify applications or create the wait shared memory.\n"
2132 "Execution continues but there might be problem for already\n"
2133 "running applications that wishes to register.");
2137 DBG("Accepting application registration");
2139 /* Inifinite blocking call, waiting for transmission */
2141 health_poll_entry();
2142 ret
= lttng_poll_wait(&events
, -1);
2146 * Restart interrupted system call.
2148 if (errno
== EINTR
) {
2156 for (i
= 0; i
< nb_fd
; i
++) {
2157 health_code_update();
2159 /* Fetch once the poll data */
2160 revents
= LTTNG_POLL_GETEV(&events
, i
);
2161 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2164 /* No activity for this FD (poll implementation). */
2168 /* Thread quit pipe has been closed. Killing thread. */
2169 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2175 /* Event on the registration socket */
2176 if (pollfd
== apps_sock
) {
2177 if (revents
& LPOLLIN
) {
2178 sock
= lttcomm_accept_unix_sock(apps_sock
);
2184 * Set socket timeout for both receiving and ending.
2185 * app_socket_timeout is in seconds, whereas
2186 * lttcomm_setsockopt_rcv_timeout and
2187 * lttcomm_setsockopt_snd_timeout expect msec as
2190 if (config
.app_socket_timeout
>= 0) {
2191 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2192 config
.app_socket_timeout
* 1000);
2193 (void) lttcomm_setsockopt_snd_timeout(sock
,
2194 config
.app_socket_timeout
* 1000);
2198 * Set the CLOEXEC flag. Return code is useless because
2199 * either way, the show must go on.
2201 (void) utils_set_fd_cloexec(sock
);
2203 /* Create UST registration command for enqueuing */
2204 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2205 if (ust_cmd
== NULL
) {
2206 PERROR("ust command zmalloc");
2215 * Using message-based transmissions to ensure we don't
2216 * have to deal with partially received messages.
2218 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2220 ERR("Exhausted file descriptors allowed for applications.");
2230 health_code_update();
2231 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2234 /* Close socket of the application. */
2239 lttng_fd_put(LTTNG_FD_APPS
, 1);
2243 health_code_update();
2245 ust_cmd
->sock
= sock
;
2248 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2249 " gid:%d sock:%d name:%s (version %d.%d)",
2250 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2251 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2252 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2253 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2256 * Lock free enqueue the registration request. The red pill
2257 * has been taken! This apps will be part of the *system*.
2259 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2262 * Wake the registration queue futex. Implicit memory
2263 * barrier with the exchange in cds_wfcq_enqueue.
2265 futex_nto1_wake(&ust_cmd_queue
.futex
);
2266 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2267 ERR("Register apps socket poll error");
2270 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2279 /* Notify that the registration thread is gone */
2282 if (apps_sock
>= 0) {
2283 ret
= close(apps_sock
);
2293 lttng_fd_put(LTTNG_FD_APPS
, 1);
2295 unlink(config
.apps_unix_sock_path
.value
);
2298 lttng_poll_clean(&events
);
2302 DBG("UST Registration thread cleanup complete");
2305 ERR("Health error occurred in %s", __func__
);
2307 health_unregister(health_sessiond
);
2313 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2314 * exec or it will fails.
2316 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2319 struct timespec timeout
;
2322 * Make sure we set the readiness flag to 0 because we are NOT ready.
2323 * This access to consumer_thread_is_ready does not need to be
2324 * protected by consumer_data.cond_mutex (yet) since the consumer
2325 * management thread has not been started at this point.
2327 consumer_data
->consumer_thread_is_ready
= 0;
2329 /* Setup pthread condition */
2330 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2333 PERROR("pthread_condattr_init consumer data");
2338 * Set the monotonic clock in order to make sure we DO NOT jump in time
2339 * between the clock_gettime() call and the timedwait call. See bug #324
2340 * for a more details and how we noticed it.
2342 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2345 PERROR("pthread_condattr_setclock consumer data");
2349 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2352 PERROR("pthread_cond_init consumer data");
2356 ret
= pthread_create(&consumer_data
->thread
, default_pthread_attr(),
2357 thread_manage_consumer
, consumer_data
);
2360 PERROR("pthread_create consumer");
2365 /* We are about to wait on a pthread condition */
2366 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2368 /* Get time for sem_timedwait absolute timeout */
2369 clock_ret
= lttng_clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2371 * Set the timeout for the condition timed wait even if the clock gettime
2372 * call fails since we might loop on that call and we want to avoid to
2373 * increment the timeout too many times.
2375 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2378 * The following loop COULD be skipped in some conditions so this is why we
2379 * set ret to 0 in order to make sure at least one round of the loop is
2385 * Loop until the condition is reached or when a timeout is reached. Note
2386 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2387 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2388 * possible. This loop does not take any chances and works with both of
2391 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2392 if (clock_ret
< 0) {
2393 PERROR("clock_gettime spawn consumer");
2394 /* Infinite wait for the consumerd thread to be ready */
2395 ret
= pthread_cond_wait(&consumer_data
->cond
,
2396 &consumer_data
->cond_mutex
);
2398 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2399 &consumer_data
->cond_mutex
, &timeout
);
2403 /* Release the pthread condition */
2404 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2408 if (ret
== ETIMEDOUT
) {
2412 * Call has timed out so we kill the kconsumerd_thread and return
2415 ERR("Condition timed out. The consumer thread was never ready."
2417 pth_ret
= pthread_cancel(consumer_data
->thread
);
2419 PERROR("pthread_cancel consumer thread");
2422 PERROR("pthread_cond_wait failed consumer thread");
2424 /* Caller is expecting a negative value on failure. */
2429 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2430 if (consumer_data
->pid
== 0) {
2431 ERR("Consumerd did not start");
2432 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2435 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2444 * Join consumer thread
2446 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2450 /* Consumer pid must be a real one. */
2451 if (consumer_data
->pid
> 0) {
2453 ret
= kill(consumer_data
->pid
, SIGTERM
);
2455 PERROR("Error killing consumer daemon");
2458 return pthread_join(consumer_data
->thread
, &status
);
2465 * Fork and exec a consumer daemon (consumerd).
2467 * Return pid if successful else -1.
2469 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2473 const char *consumer_to_use
;
2474 const char *verbosity
;
2477 DBG("Spawning consumerd");
2484 if (config
.verbose_consumer
) {
2485 verbosity
= "--verbose";
2486 } else if (lttng_opt_quiet
) {
2487 verbosity
= "--quiet";
2492 switch (consumer_data
->type
) {
2493 case LTTNG_CONSUMER_KERNEL
:
2495 * Find out which consumerd to execute. We will first try the
2496 * 64-bit path, then the sessiond's installation directory, and
2497 * fallback on the 32-bit one,
2499 DBG3("Looking for a kernel consumer at these locations:");
2500 DBG3(" 1) %s", config
.consumerd64_bin_path
.value
? : "NULL");
2501 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, DEFAULT_CONSUMERD_FILE
);
2502 DBG3(" 3) %s", config
.consumerd32_bin_path
.value
? : "NULL");
2503 if (stat(config
.consumerd64_bin_path
.value
, &st
) == 0) {
2504 DBG3("Found location #1");
2505 consumer_to_use
= config
.consumerd64_bin_path
.value
;
2506 } else if (stat(INSTALL_BIN_PATH
"/" DEFAULT_CONSUMERD_FILE
, &st
) == 0) {
2507 DBG3("Found location #2");
2508 consumer_to_use
= INSTALL_BIN_PATH
"/" DEFAULT_CONSUMERD_FILE
;
2509 } else if (stat(config
.consumerd32_bin_path
.value
, &st
) == 0) {
2510 DBG3("Found location #3");
2511 consumer_to_use
= config
.consumerd32_bin_path
.value
;
2513 DBG("Could not find any valid consumerd executable");
2517 DBG("Using kernel consumer at: %s", consumer_to_use
);
2518 (void) execl(consumer_to_use
,
2519 "lttng-consumerd", verbosity
, "-k",
2520 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2521 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2522 "--group", config
.tracing_group_name
.value
,
2525 case LTTNG_CONSUMER64_UST
:
2527 if (config
.consumerd64_lib_dir
.value
) {
2532 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2536 tmplen
= strlen(config
.consumerd64_lib_dir
.value
) + 1 /* : */ + strlen(tmp
);
2537 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2542 strcat(tmpnew
, config
.consumerd64_lib_dir
.value
);
2543 if (tmp
[0] != '\0') {
2544 strcat(tmpnew
, ":");
2545 strcat(tmpnew
, tmp
);
2547 ret
= setenv("LD_LIBRARY_PATH", tmpnew
, 1);
2554 DBG("Using 64-bit UST consumer at: %s", config
.consumerd64_bin_path
.value
);
2555 (void) execl(config
.consumerd64_bin_path
.value
, "lttng-consumerd", verbosity
, "-u",
2556 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2557 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2558 "--group", config
.tracing_group_name
.value
,
2562 case LTTNG_CONSUMER32_UST
:
2564 if (config
.consumerd32_lib_dir
.value
) {
2569 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2573 tmplen
= strlen(config
.consumerd32_lib_dir
.value
) + 1 /* : */ + strlen(tmp
);
2574 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2579 strcat(tmpnew
, config
.consumerd32_lib_dir
.value
);
2580 if (tmp
[0] != '\0') {
2581 strcat(tmpnew
, ":");
2582 strcat(tmpnew
, tmp
);
2584 ret
= setenv("LD_LIBRARY_PATH", tmpnew
, 1);
2591 DBG("Using 32-bit UST consumer at: %s", config
.consumerd32_bin_path
.value
);
2592 (void) execl(config
.consumerd32_bin_path
.value
, "lttng-consumerd", verbosity
, "-u",
2593 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2594 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2595 "--group", config
.tracing_group_name
.value
,
2600 ERR("unknown consumer type");
2604 PERROR("Consumer execl()");
2606 /* Reaching this point, we got a failure on our execl(). */
2608 } else if (pid
> 0) {
2611 PERROR("start consumer fork");
2619 * Spawn the consumerd daemon and session daemon thread.
2621 static int start_consumerd(struct consumer_data
*consumer_data
)
2626 * Set the listen() state on the socket since there is a possible race
2627 * between the exec() of the consumer daemon and this call if place in the
2628 * consumer thread. See bug #366 for more details.
2630 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2635 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2636 if (consumer_data
->pid
!= 0) {
2637 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2641 ret
= spawn_consumerd(consumer_data
);
2643 ERR("Spawning consumerd failed");
2644 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2648 /* Setting up the consumer_data pid */
2649 consumer_data
->pid
= ret
;
2650 DBG2("Consumer pid %d", consumer_data
->pid
);
2651 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2653 DBG2("Spawning consumer control thread");
2654 ret
= spawn_consumer_thread(consumer_data
);
2656 ERR("Fatal error spawning consumer control thread");
2664 /* Cleanup already created sockets on error. */
2665 if (consumer_data
->err_sock
>= 0) {
2668 err
= close(consumer_data
->err_sock
);
2670 PERROR("close consumer data error socket");
2677 * Setup necessary data for kernel tracer action.
2679 static int init_kernel_tracer(void)
2683 /* Modprobe lttng kernel modules */
2684 ret
= modprobe_lttng_control();
2689 /* Open debugfs lttng */
2690 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2691 if (kernel_tracer_fd
< 0) {
2692 DBG("Failed to open %s", module_proc_lttng
);
2696 /* Validate kernel version */
2697 ret
= kernel_validate_version(kernel_tracer_fd
, &kernel_tracer_version
,
2698 &kernel_tracer_abi_version
);
2703 ret
= modprobe_lttng_data();
2708 ret
= kernel_supports_ring_buffer_snapshot_sample_positions(
2715 WARN("Kernel tracer does not support buffer monitoring. "
2716 "The monitoring timer of channels in the kernel domain "
2717 "will be set to 0 (disabled).");
2720 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2724 modprobe_remove_lttng_control();
2725 ret
= close(kernel_tracer_fd
);
2729 kernel_tracer_fd
= -1;
2730 return LTTNG_ERR_KERN_VERSION
;
2733 ret
= close(kernel_tracer_fd
);
2739 modprobe_remove_lttng_control();
2742 WARN("No kernel tracer available");
2743 kernel_tracer_fd
= -1;
2745 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2747 return LTTNG_ERR_KERN_NA
;
2753 * Copy consumer output from the tracing session to the domain session. The
2754 * function also applies the right modification on a per domain basis for the
2755 * trace files destination directory.
2757 * Should *NOT* be called with RCU read-side lock held.
2759 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2762 const char *dir_name
;
2763 struct consumer_output
*consumer
;
2766 assert(session
->consumer
);
2769 case LTTNG_DOMAIN_KERNEL
:
2770 DBG3("Copying tracing session consumer output in kernel session");
2772 * XXX: We should audit the session creation and what this function
2773 * does "extra" in order to avoid a destroy since this function is used
2774 * in the domain session creation (kernel and ust) only. Same for UST
2777 if (session
->kernel_session
->consumer
) {
2778 consumer_output_put(session
->kernel_session
->consumer
);
2780 session
->kernel_session
->consumer
=
2781 consumer_copy_output(session
->consumer
);
2782 /* Ease our life a bit for the next part */
2783 consumer
= session
->kernel_session
->consumer
;
2784 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2786 case LTTNG_DOMAIN_JUL
:
2787 case LTTNG_DOMAIN_LOG4J
:
2788 case LTTNG_DOMAIN_PYTHON
:
2789 case LTTNG_DOMAIN_UST
:
2790 DBG3("Copying tracing session consumer output in UST session");
2791 if (session
->ust_session
->consumer
) {
2792 consumer_output_put(session
->ust_session
->consumer
);
2794 session
->ust_session
->consumer
=
2795 consumer_copy_output(session
->consumer
);
2796 /* Ease our life a bit for the next part */
2797 consumer
= session
->ust_session
->consumer
;
2798 dir_name
= DEFAULT_UST_TRACE_DIR
;
2801 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2805 /* Append correct directory to subdir */
2806 strncat(consumer
->subdir
, dir_name
,
2807 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2808 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2817 * Create an UST session and add it to the session ust list.
2819 * Should *NOT* be called with RCU read-side lock held.
2821 static int create_ust_session(struct ltt_session
*session
,
2822 struct lttng_domain
*domain
)
2825 struct ltt_ust_session
*lus
= NULL
;
2829 assert(session
->consumer
);
2831 switch (domain
->type
) {
2832 case LTTNG_DOMAIN_JUL
:
2833 case LTTNG_DOMAIN_LOG4J
:
2834 case LTTNG_DOMAIN_PYTHON
:
2835 case LTTNG_DOMAIN_UST
:
2838 ERR("Unknown UST domain on create session %d", domain
->type
);
2839 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2843 DBG("Creating UST session");
2845 lus
= trace_ust_create_session(session
->id
);
2847 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2851 lus
->uid
= session
->uid
;
2852 lus
->gid
= session
->gid
;
2853 lus
->output_traces
= session
->output_traces
;
2854 lus
->snapshot_mode
= session
->snapshot_mode
;
2855 lus
->live_timer_interval
= session
->live_timer
;
2856 session
->ust_session
= lus
;
2857 if (session
->shm_path
[0]) {
2858 strncpy(lus
->root_shm_path
, session
->shm_path
,
2859 sizeof(lus
->root_shm_path
));
2860 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2861 strncpy(lus
->shm_path
, session
->shm_path
,
2862 sizeof(lus
->shm_path
));
2863 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2864 strncat(lus
->shm_path
, "/ust",
2865 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2867 /* Copy session output to the newly created UST session */
2868 ret
= copy_session_consumer(domain
->type
, session
);
2869 if (ret
!= LTTNG_OK
) {
2877 session
->ust_session
= NULL
;
2882 * Create a kernel tracer session then create the default channel.
2884 static int create_kernel_session(struct ltt_session
*session
)
2888 DBG("Creating kernel session");
2890 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2892 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2896 /* Code flow safety */
2897 assert(session
->kernel_session
);
2899 /* Copy session output to the newly created Kernel session */
2900 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2901 if (ret
!= LTTNG_OK
) {
2905 session
->kernel_session
->uid
= session
->uid
;
2906 session
->kernel_session
->gid
= session
->gid
;
2907 session
->kernel_session
->output_traces
= session
->output_traces
;
2908 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2913 trace_kernel_destroy_session(session
->kernel_session
);
2914 session
->kernel_session
= NULL
;
2919 * Count number of session permitted by uid/gid.
2921 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2924 struct ltt_session
*session
;
2926 DBG("Counting number of available session for UID %d GID %d",
2928 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2930 * Only list the sessions the user can control.
2932 if (!session_access_ok(session
, uid
, gid
)) {
2941 * Check if the current kernel tracer supports the session rotation feature.
2942 * Return 1 if it does, 0 otherwise.
2944 static int check_rotate_compatible(void)
2948 if (kernel_tracer_version
.major
!= 2 || kernel_tracer_version
.minor
< 11) {
2949 DBG("Kernel tracer version is not compatible with the rotation feature");
2957 * Process the command requested by the lttng client within the command
2958 * context structure. This function make sure that the return structure (llm)
2959 * is set and ready for transmission before returning.
2961 * Return any error encountered or 0 for success.
2963 * "sock" is only used for special-case var. len data.
2965 * Should *NOT* be called with RCU read-side lock held.
2967 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2971 int need_tracing_session
= 1;
2974 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2976 assert(!rcu_read_ongoing());
2980 switch (cmd_ctx
->lsm
->cmd_type
) {
2981 case LTTNG_CREATE_SESSION
:
2982 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2983 case LTTNG_CREATE_SESSION_LIVE
:
2984 case LTTNG_DESTROY_SESSION
:
2985 case LTTNG_LIST_SESSIONS
:
2986 case LTTNG_LIST_DOMAINS
:
2987 case LTTNG_START_TRACE
:
2988 case LTTNG_STOP_TRACE
:
2989 case LTTNG_DATA_PENDING
:
2990 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2991 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2992 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2993 case LTTNG_SNAPSHOT_RECORD
:
2994 case LTTNG_SAVE_SESSION
:
2995 case LTTNG_SET_SESSION_SHM_PATH
:
2996 case LTTNG_REGENERATE_METADATA
:
2997 case LTTNG_REGENERATE_STATEDUMP
:
2998 case LTTNG_REGISTER_TRIGGER
:
2999 case LTTNG_UNREGISTER_TRIGGER
:
3000 case LTTNG_ROTATE_SESSION
:
3001 case LTTNG_ROTATION_GET_INFO
:
3002 case LTTNG_SESSION_GET_CURRENT_OUTPUT
:
3003 case LTTNG_ROTATION_SET_SCHEDULE
:
3004 case LTTNG_ROTATION_SCHEDULE_GET_TIMER_PERIOD
:
3005 case LTTNG_ROTATION_SCHEDULE_GET_SIZE
:
3012 if (config
.no_kernel
&& need_domain
3013 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3015 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3017 ret
= LTTNG_ERR_KERN_NA
;
3022 /* Deny register consumer if we already have a spawned consumer. */
3023 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3024 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3025 if (kconsumer_data
.pid
> 0) {
3026 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3027 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3030 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3034 * Check for command that don't needs to allocate a returned payload. We do
3035 * this here so we don't have to make the call for no payload at each
3038 switch(cmd_ctx
->lsm
->cmd_type
) {
3039 case LTTNG_LIST_SESSIONS
:
3040 case LTTNG_LIST_TRACEPOINTS
:
3041 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3042 case LTTNG_LIST_DOMAINS
:
3043 case LTTNG_LIST_CHANNELS
:
3044 case LTTNG_LIST_EVENTS
:
3045 case LTTNG_LIST_SYSCALLS
:
3046 case LTTNG_LIST_TRACKER_IDS
:
3047 case LTTNG_DATA_PENDING
:
3048 case LTTNG_ROTATE_SESSION
:
3049 case LTTNG_ROTATION_GET_INFO
:
3050 case LTTNG_ROTATION_SCHEDULE_GET_TIMER_PERIOD
:
3051 case LTTNG_ROTATION_SCHEDULE_GET_SIZE
:
3054 /* Setup lttng message with no payload */
3055 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3057 /* This label does not try to unlock the session */
3058 goto init_setup_error
;
3062 /* Commands that DO NOT need a session. */
3063 switch (cmd_ctx
->lsm
->cmd_type
) {
3064 case LTTNG_CREATE_SESSION
:
3065 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3066 case LTTNG_CREATE_SESSION_LIVE
:
3067 case LTTNG_LIST_SESSIONS
:
3068 case LTTNG_LIST_TRACEPOINTS
:
3069 case LTTNG_LIST_SYSCALLS
:
3070 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3071 case LTTNG_SAVE_SESSION
:
3072 case LTTNG_REGISTER_TRIGGER
:
3073 case LTTNG_UNREGISTER_TRIGGER
:
3074 need_tracing_session
= 0;
3077 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3079 * We keep the session list lock across _all_ commands
3080 * for now, because the per-session lock does not
3081 * handle teardown properly.
3083 session_lock_list();
3084 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3085 if (cmd_ctx
->session
== NULL
) {
3086 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3089 /* Acquire lock for the session */
3090 session_lock(cmd_ctx
->session
);
3096 * Commands that need a valid session but should NOT create one if none
3097 * exists. Instead of creating one and destroying it when the command is
3098 * handled, process that right before so we save some round trip in useless
3101 switch (cmd_ctx
->lsm
->cmd_type
) {
3102 case LTTNG_DISABLE_CHANNEL
:
3103 case LTTNG_DISABLE_EVENT
:
3104 switch (cmd_ctx
->lsm
->domain
.type
) {
3105 case LTTNG_DOMAIN_KERNEL
:
3106 if (!cmd_ctx
->session
->kernel_session
) {
3107 ret
= LTTNG_ERR_NO_CHANNEL
;
3111 case LTTNG_DOMAIN_JUL
:
3112 case LTTNG_DOMAIN_LOG4J
:
3113 case LTTNG_DOMAIN_PYTHON
:
3114 case LTTNG_DOMAIN_UST
:
3115 if (!cmd_ctx
->session
->ust_session
) {
3116 ret
= LTTNG_ERR_NO_CHANNEL
;
3121 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3133 * Check domain type for specific "pre-action".
3135 switch (cmd_ctx
->lsm
->domain
.type
) {
3136 case LTTNG_DOMAIN_KERNEL
:
3138 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3142 /* Kernel tracer check */
3143 if (kernel_tracer_fd
== -1) {
3144 /* Basically, load kernel tracer modules */
3145 ret
= init_kernel_tracer();
3151 /* Consumer is in an ERROR state. Report back to client */
3152 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3153 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3157 /* Need a session for kernel command */
3158 if (need_tracing_session
) {
3159 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3160 ret
= create_kernel_session(cmd_ctx
->session
);
3162 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3167 /* Start the kernel consumer daemon */
3168 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3169 if (kconsumer_data
.pid
== 0 &&
3170 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3171 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3172 ret
= start_consumerd(&kconsumer_data
);
3174 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3177 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3179 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3183 * The consumer was just spawned so we need to add the socket to
3184 * the consumer output of the session if exist.
3186 ret
= consumer_create_socket(&kconsumer_data
,
3187 cmd_ctx
->session
->kernel_session
->consumer
);
3194 case LTTNG_DOMAIN_JUL
:
3195 case LTTNG_DOMAIN_LOG4J
:
3196 case LTTNG_DOMAIN_PYTHON
:
3197 case LTTNG_DOMAIN_UST
:
3199 if (!ust_app_supported()) {
3200 ret
= LTTNG_ERR_NO_UST
;
3203 /* Consumer is in an ERROR state. Report back to client */
3204 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3205 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3209 if (need_tracing_session
) {
3210 /* Create UST session if none exist. */
3211 if (cmd_ctx
->session
->ust_session
== NULL
) {
3212 ret
= create_ust_session(cmd_ctx
->session
,
3213 &cmd_ctx
->lsm
->domain
);
3214 if (ret
!= LTTNG_OK
) {
3219 /* Start the UST consumer daemons */
3221 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3222 if (config
.consumerd64_bin_path
.value
&&
3223 ustconsumer64_data
.pid
== 0 &&
3224 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3225 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3226 ret
= start_consumerd(&ustconsumer64_data
);
3228 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3229 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3233 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3234 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3236 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3240 * Setup socket for consumer 64 bit. No need for atomic access
3241 * since it was set above and can ONLY be set in this thread.
3243 ret
= consumer_create_socket(&ustconsumer64_data
,
3244 cmd_ctx
->session
->ust_session
->consumer
);
3250 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3251 if (config
.consumerd32_bin_path
.value
&&
3252 ustconsumer32_data
.pid
== 0 &&
3253 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3254 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3255 ret
= start_consumerd(&ustconsumer32_data
);
3257 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3258 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3262 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3263 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3265 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3269 * Setup socket for consumer 64 bit. No need for atomic access
3270 * since it was set above and can ONLY be set in this thread.
3272 ret
= consumer_create_socket(&ustconsumer32_data
,
3273 cmd_ctx
->session
->ust_session
->consumer
);
3285 /* Validate consumer daemon state when start/stop trace command */
3286 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3287 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3288 switch (cmd_ctx
->lsm
->domain
.type
) {
3289 case LTTNG_DOMAIN_NONE
:
3291 case LTTNG_DOMAIN_JUL
:
3292 case LTTNG_DOMAIN_LOG4J
:
3293 case LTTNG_DOMAIN_PYTHON
:
3294 case LTTNG_DOMAIN_UST
:
3295 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3296 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3300 case LTTNG_DOMAIN_KERNEL
:
3301 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3302 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3307 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3313 * Check that the UID or GID match that of the tracing session.
3314 * The root user can interact with all sessions.
3316 if (need_tracing_session
) {
3317 if (!session_access_ok(cmd_ctx
->session
,
3318 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3319 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3320 ret
= LTTNG_ERR_EPERM
;
3326 * Send relayd information to consumer as soon as we have a domain and a
3329 if (cmd_ctx
->session
&& need_domain
) {
3331 * Setup relayd if not done yet. If the relayd information was already
3332 * sent to the consumer, this call will gracefully return.
3334 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3335 if (ret
!= LTTNG_OK
) {
3340 /* Process by command type */
3341 switch (cmd_ctx
->lsm
->cmd_type
) {
3342 case LTTNG_ADD_CONTEXT
:
3345 * An LTTNG_ADD_CONTEXT command might have a supplementary
3346 * payload if the context being added is an application context.
3348 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3349 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3350 char *provider_name
= NULL
, *context_name
= NULL
;
3351 size_t provider_name_len
=
3352 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3353 size_t context_name_len
=
3354 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3356 if (provider_name_len
== 0 || context_name_len
== 0) {
3358 * Application provider and context names MUST
3361 ret
= -LTTNG_ERR_INVALID
;
3365 provider_name
= zmalloc(provider_name_len
+ 1);
3366 if (!provider_name
) {
3367 ret
= -LTTNG_ERR_NOMEM
;
3370 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3373 context_name
= zmalloc(context_name_len
+ 1);
3374 if (!context_name
) {
3375 ret
= -LTTNG_ERR_NOMEM
;
3376 goto error_add_context
;
3378 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3381 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3384 goto error_add_context
;
3387 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3390 goto error_add_context
;
3395 * cmd_add_context assumes ownership of the provider and context
3398 ret
= cmd_add_context(cmd_ctx
->session
,
3399 cmd_ctx
->lsm
->domain
.type
,
3400 cmd_ctx
->lsm
->u
.context
.channel_name
,
3401 &cmd_ctx
->lsm
->u
.context
.ctx
,
3402 kernel_poll_pipe
[1]);
3404 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3405 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3407 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3408 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3414 case LTTNG_DISABLE_CHANNEL
:
3416 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3417 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3420 case LTTNG_DISABLE_EVENT
:
3424 * FIXME: handle filter; for now we just receive the filter's
3425 * bytecode along with the filter expression which are sent by
3426 * liblttng-ctl and discard them.
3428 * This fixes an issue where the client may block while sending
3429 * the filter payload and encounter an error because the session
3430 * daemon closes the socket without ever handling this data.
3432 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3433 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3436 char data
[LTTNG_FILTER_MAX_LEN
];
3438 DBG("Discarding disable event command payload of size %zu", count
);
3440 ret
= lttcomm_recv_unix_sock(sock
, data
,
3441 count
> sizeof(data
) ? sizeof(data
) : count
);
3446 count
-= (size_t) ret
;
3449 /* FIXME: passing packed structure to non-packed pointer */
3450 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3451 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3452 &cmd_ctx
->lsm
->u
.disable
.event
);
3455 case LTTNG_ENABLE_CHANNEL
:
3457 cmd_ctx
->lsm
->u
.channel
.chan
.attr
.extended
.ptr
=
3458 (struct lttng_channel_extended
*) &cmd_ctx
->lsm
->u
.channel
.extended
;
3459 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3460 &cmd_ctx
->lsm
->u
.channel
.chan
,
3461 kernel_poll_pipe
[1]);
3464 case LTTNG_TRACK_ID
:
3466 struct lttng_tracker_id id
;
3468 memset(&id
, 0, sizeof(id
));
3469 id
.type
= cmd_ctx
->lsm
->u
.id_tracker
.id_type
;
3473 case LTTNG_ID_VALUE
:
3474 id
.value
= cmd_ctx
->lsm
->u
.id_tracker
.u
.value
;
3476 case LTTNG_ID_STRING
:
3478 size_t var_len
= cmd_ctx
->lsm
->u
.id_tracker
.u
.var_len
;
3480 id
.string
= zmalloc(var_len
);
3482 ret
= LTTNG_ERR_NOMEM
;
3485 DBG("Receiving var len tracker id string from client.");
3486 ret
= lttcomm_recv_unix_sock(sock
, id
.string
, var_len
);
3488 DBG("Nothing received.");
3491 ret
= LTTNG_ERR_INVALID
;
3494 if (strnlen(id
.string
, var_len
) != var_len
- 1) {
3495 DBG("Corrupted string.");
3497 ret
= LTTNG_ERR_INVALID
;
3503 ret
= LTTNG_ERR_INVALID
;
3506 ret
= cmd_track_id(cmd_ctx
->session
,
3507 cmd_ctx
->lsm
->u
.id_tracker
.tracker_type
,
3508 cmd_ctx
->lsm
->domain
.type
,
3513 case LTTNG_UNTRACK_ID
:
3515 struct lttng_tracker_id id
;
3517 memset(&id
, 0, sizeof(id
));
3518 id
.type
= cmd_ctx
->lsm
->u
.id_tracker
.id_type
;
3522 case LTTNG_ID_VALUE
:
3523 id
.value
= cmd_ctx
->lsm
->u
.id_tracker
.u
.value
;
3525 case LTTNG_ID_STRING
:
3527 size_t var_len
= cmd_ctx
->lsm
->u
.id_tracker
.u
.var_len
;
3529 id
.string
= zmalloc(var_len
);
3531 ret
= LTTNG_ERR_NOMEM
;
3534 DBG("Receiving var len tracker id string from client.");
3535 ret
= lttcomm_recv_unix_sock(sock
, id
.string
, var_len
);
3537 DBG("Nothing received.");
3540 ret
= LTTNG_ERR_INVALID
;
3543 if (strnlen(id
.string
, var_len
) != var_len
- 1) {
3544 DBG("Corrupted string.");
3546 ret
= LTTNG_ERR_INVALID
;
3552 ret
= LTTNG_ERR_INVALID
;
3555 ret
= cmd_untrack_id(cmd_ctx
->session
,
3556 cmd_ctx
->lsm
->u
.id_tracker
.tracker_type
,
3557 cmd_ctx
->lsm
->domain
.type
,
3562 case LTTNG_ENABLE_EVENT
:
3564 struct lttng_event_exclusion
*exclusion
= NULL
;
3565 struct lttng_filter_bytecode
*bytecode
= NULL
;
3566 char *filter_expression
= NULL
;
3568 /* Handle exclusion events and receive it from the client. */
3569 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3570 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3572 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3573 (count
* LTTNG_SYMBOL_NAME_LEN
));
3575 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3579 DBG("Receiving var len exclusion event list from client ...");
3580 exclusion
->count
= count
;
3581 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3582 count
* LTTNG_SYMBOL_NAME_LEN
);
3584 DBG("Nothing recv() from client var len data... continuing");
3587 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3592 /* Get filter expression from client. */
3593 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3594 size_t expression_len
=
3595 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3597 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3598 ret
= LTTNG_ERR_FILTER_INVAL
;
3603 filter_expression
= zmalloc(expression_len
);
3604 if (!filter_expression
) {
3606 ret
= LTTNG_ERR_FILTER_NOMEM
;
3610 /* Receive var. len. data */
3611 DBG("Receiving var len filter's expression from client ...");
3612 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3615 DBG("Nothing recv() from client car len data... continuing");
3617 free(filter_expression
);
3619 ret
= LTTNG_ERR_FILTER_INVAL
;
3624 /* Handle filter and get bytecode from client. */
3625 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3626 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3628 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3629 ret
= LTTNG_ERR_FILTER_INVAL
;
3630 free(filter_expression
);
3635 bytecode
= zmalloc(bytecode_len
);
3637 free(filter_expression
);
3639 ret
= LTTNG_ERR_FILTER_NOMEM
;
3643 /* Receive var. len. data */
3644 DBG("Receiving var len filter's bytecode from client ...");
3645 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3647 DBG("Nothing recv() from client car len data... continuing");
3649 free(filter_expression
);
3652 ret
= LTTNG_ERR_FILTER_INVAL
;
3656 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3657 free(filter_expression
);
3660 ret
= LTTNG_ERR_FILTER_INVAL
;
3665 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3666 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3667 &cmd_ctx
->lsm
->u
.enable
.event
,
3668 filter_expression
, bytecode
, exclusion
,
3669 kernel_poll_pipe
[1]);
3672 case LTTNG_LIST_TRACEPOINTS
:
3674 struct lttng_event
*events
;
3677 session_lock_list();
3678 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3679 session_unlock_list();
3680 if (nb_events
< 0) {
3681 /* Return value is a negative lttng_error_code. */
3687 * Setup lttng message with payload size set to the event list size in
3688 * bytes and then copy list into the llm payload.
3690 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3691 sizeof(struct lttng_event
) * nb_events
);
3701 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3703 struct lttng_event_field
*fields
;
3706 session_lock_list();
3707 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3709 session_unlock_list();
3710 if (nb_fields
< 0) {
3711 /* Return value is a negative lttng_error_code. */
3717 * Setup lttng message with payload size set to the event list size in
3718 * bytes and then copy list into the llm payload.
3720 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3721 sizeof(struct lttng_event_field
) * nb_fields
);
3731 case LTTNG_LIST_SYSCALLS
:
3733 struct lttng_event
*events
;
3736 nb_events
= cmd_list_syscalls(&events
);
3737 if (nb_events
< 0) {
3738 /* Return value is a negative lttng_error_code. */
3744 * Setup lttng message with payload size set to the event list size in
3745 * bytes and then copy list into the llm payload.
3747 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3748 sizeof(struct lttng_event
) * nb_events
);
3758 case LTTNG_LIST_TRACKER_IDS
:
3760 struct lttcomm_tracker_command_header cmd_header
;
3761 struct lttng_tracker_id
*ids
= NULL
;
3763 struct lttng_dynamic_buffer buf
;
3765 nr_ids
= cmd_list_tracker_ids(cmd_ctx
->lsm
->u
.id_tracker
.tracker_type
,
3767 cmd_ctx
->lsm
->domain
.type
, &ids
);
3769 /* Return value is a negative lttng_error_code. */
3774 lttng_dynamic_buffer_init(&buf
);
3775 for (i
= 0; i
< nr_ids
; i
++) {
3776 struct lttng_tracker_id
*id
= &ids
[i
];
3777 struct lttcomm_tracker_id_header id_hdr
;
3778 size_t var_data_len
= 0;
3780 memset(&id_hdr
, 0, sizeof(id_hdr
));
3781 id_hdr
.type
= id
->type
;
3785 case LTTNG_ID_VALUE
:
3786 id_hdr
.u
.value
= id
->value
;
3788 case LTTNG_ID_STRING
:
3789 id_hdr
.u
.var_data_len
= var_data_len
= strlen(id
->string
) + 1;
3792 ret
= LTTNG_ERR_INVALID
;
3795 ret
= lttng_dynamic_buffer_append(&buf
, &id_hdr
, sizeof(id_hdr
));
3797 ret
= LTTNG_ERR_NOMEM
;
3800 ret
= lttng_dynamic_buffer_append(&buf
, id
->string
, var_data_len
);
3802 ret
= LTTNG_ERR_NOMEM
;
3808 cmd_header
.nb_tracker_id
= nr_ids
;
3809 ret
= setup_lttng_msg(cmd_ctx
, buf
.data
, buf
.size
, &cmd_header
,
3810 sizeof(cmd_header
));
3812 lttng_dynamic_buffer_reset(&buf
);
3820 case LTTNG_SET_CONSUMER_URI
:
3823 struct lttng_uri
*uris
;
3825 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3826 len
= nb_uri
* sizeof(struct lttng_uri
);
3829 ret
= LTTNG_ERR_INVALID
;
3833 uris
= zmalloc(len
);
3835 ret
= LTTNG_ERR_FATAL
;
3839 /* Receive variable len data */
3840 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3841 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3843 DBG("No URIs received from client... continuing");
3845 ret
= LTTNG_ERR_SESSION_FAIL
;
3850 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3852 if (ret
!= LTTNG_OK
) {
3859 case LTTNG_START_TRACE
:
3862 * On the first start, if we have a kernel session and we have
3863 * enabled time or size-based rotations, we have to make sure
3864 * the kernel tracer supports it.
3866 if (!cmd_ctx
->session
->has_been_started
&& \
3867 cmd_ctx
->session
->kernel_session
&& \
3868 (cmd_ctx
->session
->rotate_timer_period
|| \
3869 cmd_ctx
->session
->rotate_size
) && \
3870 !check_rotate_compatible()) {
3871 DBG("Kernel tracer version is not compatible with the rotation feature");
3872 ret
= LTTNG_ERR_ROTATION_WRONG_VERSION
;
3875 ret
= cmd_start_trace(cmd_ctx
->session
);
3878 case LTTNG_STOP_TRACE
:
3880 ret
= cmd_stop_trace(cmd_ctx
->session
);
3883 case LTTNG_CREATE_SESSION
:
3886 struct lttng_uri
*uris
= NULL
;
3888 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3889 len
= nb_uri
* sizeof(struct lttng_uri
);
3892 uris
= zmalloc(len
);
3894 ret
= LTTNG_ERR_FATAL
;
3898 /* Receive variable len data */
3899 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3900 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3902 DBG("No URIs received from client... continuing");
3904 ret
= LTTNG_ERR_SESSION_FAIL
;
3909 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3910 DBG("Creating session with ONE network URI is a bad call");
3911 ret
= LTTNG_ERR_SESSION_FAIL
;
3917 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3918 &cmd_ctx
->creds
, 0);
3924 case LTTNG_DESTROY_SESSION
:
3926 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1],
3927 notification_thread_handle
);
3929 /* Set session to NULL so we do not unlock it after free. */
3930 cmd_ctx
->session
= NULL
;
3933 case LTTNG_LIST_DOMAINS
:
3936 struct lttng_domain
*domains
= NULL
;
3938 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3940 /* Return value is a negative lttng_error_code. */
3945 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3946 nb_dom
* sizeof(struct lttng_domain
));
3956 case LTTNG_LIST_CHANNELS
:
3958 ssize_t payload_size
;
3959 struct lttng_channel
*channels
= NULL
;
3961 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3962 cmd_ctx
->session
, &channels
);
3963 if (payload_size
< 0) {
3964 /* Return value is a negative lttng_error_code. */
3965 ret
= -payload_size
;
3969 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3980 case LTTNG_LIST_EVENTS
:
3983 struct lttng_event
*events
= NULL
;
3984 struct lttcomm_event_command_header cmd_header
;
3987 memset(&cmd_header
, 0, sizeof(cmd_header
));
3988 /* Extended infos are included at the end of events */
3989 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3990 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3991 &events
, &total_size
);
3994 /* Return value is a negative lttng_error_code. */
3999 cmd_header
.nb_events
= nb_event
;
4000 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
4001 &cmd_header
, sizeof(cmd_header
));
4011 case LTTNG_LIST_SESSIONS
:
4013 unsigned int nr_sessions
;
4014 void *sessions_payload
;
4017 session_lock_list();
4018 nr_sessions
= lttng_sessions_count(
4019 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
4020 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
4021 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
4022 sessions_payload
= zmalloc(payload_len
);
4024 if (!sessions_payload
) {
4025 session_unlock_list();
4030 cmd_list_lttng_sessions(sessions_payload
,
4031 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
4032 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
4033 session_unlock_list();
4035 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
4037 free(sessions_payload
);
4046 case LTTNG_REGISTER_CONSUMER
:
4048 struct consumer_data
*cdata
;
4050 switch (cmd_ctx
->lsm
->domain
.type
) {
4051 case LTTNG_DOMAIN_KERNEL
:
4052 cdata
= &kconsumer_data
;
4055 ret
= LTTNG_ERR_UND
;
4059 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
4060 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
4063 case LTTNG_DATA_PENDING
:
4066 uint8_t pending_ret_byte
;
4068 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
4073 * This function may returns 0 or 1 to indicate whether or not
4074 * there is data pending. In case of error, it should return an
4075 * LTTNG_ERR code. However, some code paths may still return
4076 * a nondescript error code, which we handle by returning an
4079 if (pending_ret
== 0 || pending_ret
== 1) {
4081 * ret will be set to LTTNG_OK at the end of
4084 } else if (pending_ret
< 0) {
4085 ret
= LTTNG_ERR_UNK
;
4092 pending_ret_byte
= (uint8_t) pending_ret
;
4094 /* 1 byte to return whether or not data is pending */
4095 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
4096 &pending_ret_byte
, 1);
4105 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
4107 struct lttcomm_lttng_output_id reply
;
4109 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
4110 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
4111 if (ret
!= LTTNG_OK
) {
4115 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
4121 /* Copy output list into message payload */
4125 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
4127 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
4128 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
4131 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4134 struct lttng_snapshot_output
*outputs
= NULL
;
4136 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4137 if (nb_output
< 0) {
4142 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4143 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4144 nb_output
* sizeof(struct lttng_snapshot_output
));
4154 case LTTNG_SNAPSHOT_RECORD
:
4156 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4157 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4158 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4161 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4164 struct lttng_uri
*uris
= NULL
;
4166 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4167 len
= nb_uri
* sizeof(struct lttng_uri
);
4170 uris
= zmalloc(len
);
4172 ret
= LTTNG_ERR_FATAL
;
4176 /* Receive variable len data */
4177 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4178 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4180 DBG("No URIs received from client... continuing");
4182 ret
= LTTNG_ERR_SESSION_FAIL
;
4187 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4188 DBG("Creating session with ONE network URI is a bad call");
4189 ret
= LTTNG_ERR_SESSION_FAIL
;
4195 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4196 nb_uri
, &cmd_ctx
->creds
);
4200 case LTTNG_CREATE_SESSION_LIVE
:
4203 struct lttng_uri
*uris
= NULL
;
4205 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4206 len
= nb_uri
* sizeof(struct lttng_uri
);
4209 uris
= zmalloc(len
);
4211 ret
= LTTNG_ERR_FATAL
;
4215 /* Receive variable len data */
4216 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4217 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4219 DBG("No URIs received from client... continuing");
4221 ret
= LTTNG_ERR_SESSION_FAIL
;
4226 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4227 DBG("Creating session with ONE network URI is a bad call");
4228 ret
= LTTNG_ERR_SESSION_FAIL
;
4234 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4235 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4239 case LTTNG_SAVE_SESSION
:
4241 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4245 case LTTNG_SET_SESSION_SHM_PATH
:
4247 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4248 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4251 case LTTNG_REGENERATE_METADATA
:
4253 ret
= cmd_regenerate_metadata(cmd_ctx
->session
);
4256 case LTTNG_REGENERATE_STATEDUMP
:
4258 ret
= cmd_regenerate_statedump(cmd_ctx
->session
);
4261 case LTTNG_REGISTER_TRIGGER
:
4263 ret
= cmd_register_trigger(cmd_ctx
, sock
,
4264 notification_thread_handle
);
4267 case LTTNG_UNREGISTER_TRIGGER
:
4269 ret
= cmd_unregister_trigger(cmd_ctx
, sock
,
4270 notification_thread_handle
);
4273 case LTTNG_ROTATE_SESSION
:
4275 struct lttng_rotate_session_return rotate_return
;
4277 DBG("Client rotate session \"%s\"", cmd_ctx
->session
->name
);
4279 memset(&rotate_return
, 0, sizeof(rotate_return
));
4280 if (cmd_ctx
->session
->kernel_session
&& !check_rotate_compatible()) {
4281 DBG("Kernel tracer version is not compatible with the rotation feature");
4282 ret
= LTTNG_ERR_ROTATION_WRONG_VERSION
;
4286 ret
= cmd_rotate_session(cmd_ctx
->session
, &rotate_return
);
4292 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &rotate_return
,
4293 sizeof(rotate_return
));
4302 case LTTNG_ROTATION_GET_INFO
:
4304 struct lttng_rotation_get_info_return get_info_return
;
4306 memset(&get_info_return
, 0, sizeof(get_info_return
));
4307 ret
= cmd_rotate_get_info(cmd_ctx
->session
, &get_info_return
,
4308 cmd_ctx
->lsm
->u
.get_rotation_info
.rotation_id
);
4314 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &get_info_return
,
4315 sizeof(get_info_return
));
4324 case LTTNG_SESSION_GET_CURRENT_OUTPUT
:
4326 struct lttng_session_get_current_output_return output_return
;
4328 memset(&output_return
, 0, sizeof(output_return
));
4329 ret
= cmd_session_get_current_output(cmd_ctx
->session
,
4336 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &output_return
,
4337 sizeof(output_return
));
4346 case LTTNG_ROTATION_SET_SCHEDULE
:
4348 if (cmd_ctx
->session
->kernel_session
&& !check_rotate_compatible()) {
4349 DBG("Kernel tracer version does not support session rotations");
4350 ret
= LTTNG_ERR_ROTATION_WRONG_VERSION
;
4354 ret
= cmd_rotation_set_schedule(cmd_ctx
->session
,
4355 cmd_ctx
->lsm
->u
.rotate_setup
.timer_us
,
4356 cmd_ctx
->lsm
->u
.rotate_setup
.size
,
4357 notification_thread_handle
);
4358 if (ret
!= LTTNG_OK
) {
4364 case LTTNG_ROTATION_SCHEDULE_GET_TIMER_PERIOD
:
4366 struct lttng_rotation_schedule_get_timer_period
*get_timer
;
4368 get_timer
= zmalloc(sizeof(struct lttng_rotation_schedule_get_timer_period
));
4373 get_timer
->rotate_timer
= cmd_ctx
->session
->rotate_timer_period
;
4375 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, get_timer
,
4376 sizeof(struct lttng_rotation_schedule_get_timer_period
));
4386 case LTTNG_ROTATION_SCHEDULE_GET_SIZE
:
4388 struct lttng_rotation_schedule_get_size
*get_size
;
4390 get_size
= zmalloc(sizeof(struct lttng_rotation_schedule_get_size
));
4395 get_size
->rotate_size
= cmd_ctx
->session
->rotate_size
;
4397 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, get_size
,
4398 sizeof(struct lttng_rotation_schedule_get_size
));
4409 ret
= LTTNG_ERR_UND
;
4414 if (cmd_ctx
->llm
== NULL
) {
4415 DBG("Missing llm structure. Allocating one.");
4416 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4420 /* Set return code */
4421 cmd_ctx
->llm
->ret_code
= ret
;
4423 if (cmd_ctx
->session
) {
4424 session_unlock(cmd_ctx
->session
);
4426 if (need_tracing_session
) {
4427 session_unlock_list();
4430 assert(!rcu_read_ongoing());
4435 * Thread managing health check socket.
4437 static void *thread_manage_health(void *data
)
4439 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4440 uint32_t revents
, nb_fd
;
4441 struct lttng_poll_event events
;
4442 struct health_comm_msg msg
;
4443 struct health_comm_reply reply
;
4445 DBG("[thread] Manage health check started");
4447 rcu_register_thread();
4449 /* We might hit an error path before this is created. */
4450 lttng_poll_init(&events
);
4452 /* Create unix socket */
4453 sock
= lttcomm_create_unix_sock(config
.health_unix_sock_path
.value
);
4455 ERR("Unable to create health check Unix socket");
4460 /* lttng health client socket path permissions */
4461 ret
= chown(config
.health_unix_sock_path
.value
, 0,
4462 utils_get_group_id(config
.tracing_group_name
.value
));
4464 ERR("Unable to set group on %s", config
.health_unix_sock_path
.value
);
4469 ret
= chmod(config
.health_unix_sock_path
.value
,
4470 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4472 ERR("Unable to set permissions on %s", config
.health_unix_sock_path
.value
);
4479 * Set the CLOEXEC flag. Return code is useless because either way, the
4482 (void) utils_set_fd_cloexec(sock
);
4484 ret
= lttcomm_listen_unix_sock(sock
);
4490 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4491 * more will be added to this poll set.
4493 ret
= sessiond_set_thread_pollset(&events
, 2);
4498 /* Add the application registration socket */
4499 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4504 sessiond_notify_ready();
4507 DBG("Health check ready");
4509 /* Inifinite blocking call, waiting for transmission */
4511 ret
= lttng_poll_wait(&events
, -1);
4514 * Restart interrupted system call.
4516 if (errno
== EINTR
) {
4524 for (i
= 0; i
< nb_fd
; i
++) {
4525 /* Fetch once the poll data */
4526 revents
= LTTNG_POLL_GETEV(&events
, i
);
4527 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4530 /* No activity for this FD (poll implementation). */
4534 /* Thread quit pipe has been closed. Killing thread. */
4535 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4541 /* Event on the registration socket */
4542 if (pollfd
== sock
) {
4543 if (revents
& LPOLLIN
) {
4545 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4546 ERR("Health socket poll error");
4549 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4555 new_sock
= lttcomm_accept_unix_sock(sock
);
4561 * Set the CLOEXEC flag. Return code is useless because either way, the
4564 (void) utils_set_fd_cloexec(new_sock
);
4566 DBG("Receiving data from client for health...");
4567 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4569 DBG("Nothing recv() from client... continuing");
4570 ret
= close(new_sock
);
4577 rcu_thread_online();
4579 memset(&reply
, 0, sizeof(reply
));
4580 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4582 * health_check_state returns 0 if health is
4585 if (!health_check_state(health_sessiond
, i
)) {
4586 reply
.ret_code
|= 1ULL << i
;
4590 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4592 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4594 ERR("Failed to send health data back to client");
4597 /* End of transmission */
4598 ret
= close(new_sock
);
4607 ERR("Health error occurred in %s", __func__
);
4609 DBG("Health check thread dying");
4610 unlink(config
.health_unix_sock_path
.value
);
4618 lttng_poll_clean(&events
);
4620 rcu_unregister_thread();
4625 * This thread manage all clients request using the unix client socket for
4628 static void *thread_manage_clients(void *data
)
4630 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4632 uint32_t revents
, nb_fd
;
4633 struct command_ctx
*cmd_ctx
= NULL
;
4634 struct lttng_poll_event events
;
4636 DBG("[thread] Manage client started");
4638 rcu_register_thread();
4640 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4642 health_code_update();
4644 ret
= lttcomm_listen_unix_sock(client_sock
);
4650 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4651 * more will be added to this poll set.
4653 ret
= sessiond_set_thread_pollset(&events
, 2);
4655 goto error_create_poll
;
4658 /* Add the application registration socket */
4659 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4664 ret
= sem_post(&load_info
->message_thread_ready
);
4666 PERROR("sem_post message_thread_ready");
4671 * Wait until all support threads are initialized before accepting
4674 while (uatomic_read(<tng_sessiond_ready
) != 0) {
4676 struct timeval timeout
;
4679 FD_SET(thread_quit_pipe
[0], &read_fds
);
4680 memset(&timeout
, 0, sizeof(timeout
));
4681 timeout
.tv_usec
= 1000;
4684 * If a support thread failed to launch, it may signal that
4685 * we must exit and the sessiond would never be marked as
4688 * The timeout is set to 1ms, which serves as a way to
4689 * pace down this check.
4691 ret
= select(thread_quit_pipe
[0] + 1, &read_fds
, NULL
, NULL
,
4693 if (ret
> 0 || (ret
< 0 && errno
!= EINTR
)) {
4698 * This barrier is paired with the one in sessiond_notify_ready() to
4699 * ensure that loads accessing data initialized by the other threads,
4700 * on which this thread was waiting, are not performed before this point.
4702 * Note that this could be a 'read' memory barrier, but a full barrier
4703 * is used in case the code changes. The performance implications of
4704 * this choice are minimal since this is a slow path.
4708 /* This testpoint is after we signal readiness to the parent. */
4709 if (testpoint(sessiond_thread_manage_clients
)) {
4713 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4717 health_code_update();
4720 DBG("Accepting client command ...");
4722 /* Inifinite blocking call, waiting for transmission */
4724 health_poll_entry();
4725 ret
= lttng_poll_wait(&events
, -1);
4729 * Restart interrupted system call.
4731 if (errno
== EINTR
) {
4739 for (i
= 0; i
< nb_fd
; i
++) {
4740 /* Fetch once the poll data */
4741 revents
= LTTNG_POLL_GETEV(&events
, i
);
4742 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4744 health_code_update();
4747 /* No activity for this FD (poll implementation). */
4751 /* Thread quit pipe has been closed. Killing thread. */
4752 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4758 /* Event on the registration socket */
4759 if (pollfd
== client_sock
) {
4760 if (revents
& LPOLLIN
) {
4762 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4763 ERR("Client socket poll error");
4766 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4772 DBG("Wait for client response");
4774 health_code_update();
4776 sock
= lttcomm_accept_unix_sock(client_sock
);
4782 * Set the CLOEXEC flag. Return code is useless because either way, the
4785 (void) utils_set_fd_cloexec(sock
);
4787 /* Set socket option for credentials retrieval */
4788 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4793 /* Allocate context command to process the client request */
4794 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4795 if (cmd_ctx
== NULL
) {
4796 PERROR("zmalloc cmd_ctx");
4800 /* Allocate data buffer for reception */
4801 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4802 if (cmd_ctx
->lsm
== NULL
) {
4803 PERROR("zmalloc cmd_ctx->lsm");
4807 cmd_ctx
->llm
= NULL
;
4808 cmd_ctx
->session
= NULL
;
4810 health_code_update();
4813 * Data is received from the lttng client. The struct
4814 * lttcomm_session_msg (lsm) contains the command and data request of
4817 DBG("Receiving data from client ...");
4818 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4819 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4821 DBG("Nothing recv() from client... continuing");
4827 clean_command_ctx(&cmd_ctx
);
4831 health_code_update();
4833 // TODO: Validate cmd_ctx including sanity check for
4834 // security purpose.
4836 rcu_thread_online();
4838 * This function dispatch the work to the kernel or userspace tracer
4839 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4840 * informations for the client. The command context struct contains
4841 * everything this function may needs.
4843 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4844 rcu_thread_offline();
4852 * TODO: Inform client somehow of the fatal error. At
4853 * this point, ret < 0 means that a zmalloc failed
4854 * (ENOMEM). Error detected but still accept
4855 * command, unless a socket error has been
4858 clean_command_ctx(&cmd_ctx
);
4862 health_code_update();
4864 DBG("Sending response (size: %d, retcode: %s (%d))",
4865 cmd_ctx
->lttng_msg_size
,
4866 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4867 cmd_ctx
->llm
->ret_code
);
4868 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4870 ERR("Failed to send data back to client");
4873 /* End of transmission */
4880 clean_command_ctx(&cmd_ctx
);
4882 health_code_update();
4894 lttng_poll_clean(&events
);
4895 clean_command_ctx(&cmd_ctx
);
4899 unlink(config
.client_unix_sock_path
.value
);
4900 if (client_sock
>= 0) {
4901 ret
= close(client_sock
);
4909 ERR("Health error occurred in %s", __func__
);
4912 health_unregister(health_sessiond
);
4914 DBG("Client thread dying");
4916 rcu_unregister_thread();
4919 * Since we are creating the consumer threads, we own them, so we need
4920 * to join them before our thread exits.
4922 ret
= join_consumer_thread(&kconsumer_data
);
4925 PERROR("join_consumer");
4928 ret
= join_consumer_thread(&ustconsumer32_data
);
4931 PERROR("join_consumer ust32");
4934 ret
= join_consumer_thread(&ustconsumer64_data
);
4937 PERROR("join_consumer ust64");
4942 static int string_match(const char *str1
, const char *str2
)
4944 return (str1
&& str2
) && !strcmp(str1
, str2
);
4948 * Take an option from the getopt output and set it in the right variable to be
4951 * Return 0 on success else a negative value.
4953 static int set_option(int opt
, const char *arg
, const char *optname
)
4957 if (string_match(optname
, "client-sock") || opt
== 'c') {
4958 if (!arg
|| *arg
== '\0') {
4962 if (lttng_is_setuid_setgid()) {
4963 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4964 "-c, --client-sock");
4966 config_string_set(&config
.client_unix_sock_path
,
4968 if (!config
.client_unix_sock_path
.value
) {
4973 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4974 if (!arg
|| *arg
== '\0') {
4978 if (lttng_is_setuid_setgid()) {
4979 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4982 config_string_set(&config
.apps_unix_sock_path
,
4984 if (!config
.apps_unix_sock_path
.value
) {
4989 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4990 config
.daemonize
= true;
4991 } else if (string_match(optname
, "background") || opt
== 'b') {
4992 config
.background
= true;
4993 } else if (string_match(optname
, "group") || opt
== 'g') {
4994 if (!arg
|| *arg
== '\0') {
4998 if (lttng_is_setuid_setgid()) {
4999 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5002 config_string_set(&config
.tracing_group_name
,
5004 if (!config
.tracing_group_name
.value
) {
5009 } else if (string_match(optname
, "help") || opt
== 'h') {
5010 ret
= utils_show_help(8, "lttng-sessiond", help_msg
);
5012 ERR("Cannot show --help for `lttng-sessiond`");
5015 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
5016 } else if (string_match(optname
, "version") || opt
== 'V') {
5017 fprintf(stdout
, "%s\n", VERSION
);
5019 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
5020 config
.sig_parent
= true;
5021 } else if (string_match(optname
, "kconsumerd-err-sock")) {
5022 if (!arg
|| *arg
== '\0') {
5026 if (lttng_is_setuid_setgid()) {
5027 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5028 "--kconsumerd-err-sock");
5030 config_string_set(&config
.kconsumerd_err_unix_sock_path
,
5032 if (!config
.kconsumerd_err_unix_sock_path
.value
) {
5037 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
5038 if (!arg
|| *arg
== '\0') {
5042 if (lttng_is_setuid_setgid()) {
5043 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5044 "--kconsumerd-cmd-sock");
5046 config_string_set(&config
.kconsumerd_cmd_unix_sock_path
,
5048 if (!config
.kconsumerd_cmd_unix_sock_path
.value
) {
5053 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
5054 if (!arg
|| *arg
== '\0') {
5058 if (lttng_is_setuid_setgid()) {
5059 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5060 "--ustconsumerd64-err-sock");
5062 config_string_set(&config
.consumerd64_err_unix_sock_path
,
5064 if (!config
.consumerd64_err_unix_sock_path
.value
) {
5069 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
5070 if (!arg
|| *arg
== '\0') {
5074 if (lttng_is_setuid_setgid()) {
5075 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5076 "--ustconsumerd64-cmd-sock");
5078 config_string_set(&config
.consumerd64_cmd_unix_sock_path
,
5080 if (!config
.consumerd64_cmd_unix_sock_path
.value
) {
5085 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
5086 if (!arg
|| *arg
== '\0') {
5090 if (lttng_is_setuid_setgid()) {
5091 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5092 "--ustconsumerd32-err-sock");
5094 config_string_set(&config
.consumerd32_err_unix_sock_path
,
5096 if (!config
.consumerd32_err_unix_sock_path
.value
) {
5101 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
5102 if (!arg
|| *arg
== '\0') {
5106 if (lttng_is_setuid_setgid()) {
5107 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5108 "--ustconsumerd32-cmd-sock");
5110 config_string_set(&config
.consumerd32_cmd_unix_sock_path
,
5112 if (!config
.consumerd32_cmd_unix_sock_path
.value
) {
5117 } else if (string_match(optname
, "no-kernel")) {
5118 config
.no_kernel
= true;
5119 } else if (string_match(optname
, "quiet") || opt
== 'q') {
5120 config
.quiet
= true;
5121 } else if (string_match(optname
, "verbose") || opt
== 'v') {
5122 /* Verbose level can increase using multiple -v */
5124 /* Value obtained from config file */
5125 config
.verbose
= config_parse_value(arg
);
5127 /* -v used on command line */
5130 /* Clamp value to [0, 3] */
5131 config
.verbose
= config
.verbose
< 0 ? 0 :
5132 (config
.verbose
<= 3 ? config
.verbose
: 3);
5133 } else if (string_match(optname
, "verbose-consumer")) {
5135 config
.verbose_consumer
= config_parse_value(arg
);
5137 config
.verbose_consumer
++;
5139 } else if (string_match(optname
, "consumerd32-path")) {
5140 if (!arg
|| *arg
== '\0') {
5144 if (lttng_is_setuid_setgid()) {
5145 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5146 "--consumerd32-path");
5148 config_string_set(&config
.consumerd32_bin_path
,
5150 if (!config
.consumerd32_bin_path
.value
) {
5155 } else if (string_match(optname
, "consumerd32-libdir")) {
5156 if (!arg
|| *arg
== '\0') {
5160 if (lttng_is_setuid_setgid()) {
5161 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5162 "--consumerd32-libdir");
5164 config_string_set(&config
.consumerd32_lib_dir
,
5166 if (!config
.consumerd32_lib_dir
.value
) {
5171 } else if (string_match(optname
, "consumerd64-path")) {
5172 if (!arg
|| *arg
== '\0') {
5176 if (lttng_is_setuid_setgid()) {
5177 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5178 "--consumerd64-path");
5180 config_string_set(&config
.consumerd64_bin_path
,
5182 if (!config
.consumerd64_bin_path
.value
) {
5187 } else if (string_match(optname
, "consumerd64-libdir")) {
5188 if (!arg
|| *arg
== '\0') {
5192 if (lttng_is_setuid_setgid()) {
5193 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5194 "--consumerd64-libdir");
5196 config_string_set(&config
.consumerd64_lib_dir
,
5198 if (!config
.consumerd64_lib_dir
.value
) {
5203 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
5204 if (!arg
|| *arg
== '\0') {
5208 if (lttng_is_setuid_setgid()) {
5209 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5212 config_string_set(&config
.pid_file_path
, strdup(arg
));
5213 if (!config
.pid_file_path
.value
) {
5218 } else if (string_match(optname
, "agent-tcp-port")) {
5219 if (!arg
|| *arg
== '\0') {
5223 if (lttng_is_setuid_setgid()) {
5224 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5225 "--agent-tcp-port");
5230 v
= strtoul(arg
, NULL
, 0);
5231 if (errno
!= 0 || !isdigit(arg
[0])) {
5232 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
5235 if (v
== 0 || v
>= 65535) {
5236 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
5239 config
.agent_tcp_port
.begin
= config
.agent_tcp_port
.end
= (int) v
;
5240 DBG3("Agent TCP port set to non default: %i", (int) v
);
5242 } else if (string_match(optname
, "load") || opt
== 'l') {
5243 if (!arg
|| *arg
== '\0') {
5247 if (lttng_is_setuid_setgid()) {
5248 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5251 config_string_set(&config
.load_session_path
, strdup(arg
));
5252 if (!config
.load_session_path
.value
) {
5257 } else if (string_match(optname
, "kmod-probes")) {
5258 if (!arg
|| *arg
== '\0') {
5262 if (lttng_is_setuid_setgid()) {
5263 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5266 config_string_set(&config
.kmod_probes_list
, strdup(arg
));
5267 if (!config
.kmod_probes_list
.value
) {
5272 } else if (string_match(optname
, "extra-kmod-probes")) {
5273 if (!arg
|| *arg
== '\0') {
5277 if (lttng_is_setuid_setgid()) {
5278 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5279 "--extra-kmod-probes");
5281 config_string_set(&config
.kmod_extra_probes_list
,
5283 if (!config
.kmod_extra_probes_list
.value
) {
5288 } else if (string_match(optname
, "config") || opt
== 'f') {
5289 /* This is handled in set_options() thus silent skip. */
5292 /* Unknown option or other error.
5293 * Error is printed by getopt, just return */
5298 if (ret
== -EINVAL
) {
5299 const char *opt_name
= "unknown";
5302 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
5304 if (opt
== long_options
[i
].val
) {
5305 opt_name
= long_options
[i
].name
;
5310 WARN("Invalid argument provided for option \"%s\", using default value.",
5318 * config_entry_handler_cb used to handle options read from a config file.
5319 * See config_entry_handler_cb comment in common/config/session-config.h for the
5320 * return value conventions.
5322 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
5326 if (!entry
|| !entry
->name
|| !entry
->value
) {
5331 /* Check if the option is to be ignored */
5332 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5333 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5338 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5341 /* Ignore if not fully matched. */
5342 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5347 * If the option takes no argument on the command line, we have to
5348 * check if the value is "true". We support non-zero numeric values,
5351 if (!long_options
[i
].has_arg
) {
5352 ret
= config_parse_value(entry
->value
);
5355 WARN("Invalid configuration value \"%s\" for option %s",
5356 entry
->value
, entry
->name
);
5358 /* False, skip boolean config option. */
5363 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5367 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5374 * daemon configuration loading and argument parsing
5376 static int set_options(int argc
, char **argv
)
5378 int ret
= 0, c
= 0, option_index
= 0;
5379 int orig_optopt
= optopt
, orig_optind
= optind
;
5381 const char *config_path
= NULL
;
5383 optstring
= utils_generate_optstring(long_options
,
5384 sizeof(long_options
) / sizeof(struct option
));
5390 /* Check for the --config option */
5391 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5392 &option_index
)) != -1) {
5396 } else if (c
!= 'f') {
5397 /* if not equal to --config option. */
5401 if (lttng_is_setuid_setgid()) {
5402 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5405 config_path
= utils_expand_path(optarg
);
5407 ERR("Failed to resolve path: %s", optarg
);
5412 ret
= config_get_section_entries(config_path
, config_section_name
,
5413 config_entry_handler
, NULL
);
5416 ERR("Invalid configuration option at line %i", ret
);
5422 /* Reset getopt's global state */
5423 optopt
= orig_optopt
;
5424 optind
= orig_optind
;
5428 * getopt_long() will not set option_index if it encounters a
5431 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5438 * Pass NULL as the long option name if popt left the index
5441 ret
= set_option(c
, optarg
,
5442 option_index
< 0 ? NULL
:
5443 long_options
[option_index
].name
);
5455 * Creates the two needed socket by the daemon.
5456 * apps_sock - The communication socket for all UST apps.
5457 * client_sock - The communication of the cli tool (lttng).
5459 static int init_daemon_socket(void)
5464 old_umask
= umask(0);
5466 /* Create client tool unix socket */
5467 client_sock
= lttcomm_create_unix_sock(config
.client_unix_sock_path
.value
);
5468 if (client_sock
< 0) {
5469 ERR("Create unix sock failed: %s", config
.client_unix_sock_path
.value
);
5474 /* Set the cloexec flag */
5475 ret
= utils_set_fd_cloexec(client_sock
);
5477 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5478 "Continuing but note that the consumer daemon will have a "
5479 "reference to this socket on exec()", client_sock
);
5482 /* File permission MUST be 660 */
5483 ret
= chmod(config
.client_unix_sock_path
.value
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5485 ERR("Set file permissions failed: %s", config
.client_unix_sock_path
.value
);
5490 /* Create the application unix socket */
5491 apps_sock
= lttcomm_create_unix_sock(config
.apps_unix_sock_path
.value
);
5492 if (apps_sock
< 0) {
5493 ERR("Create unix sock failed: %s", config
.apps_unix_sock_path
.value
);
5498 /* Set the cloexec flag */
5499 ret
= utils_set_fd_cloexec(apps_sock
);
5501 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5502 "Continuing but note that the consumer daemon will have a "
5503 "reference to this socket on exec()", apps_sock
);
5506 /* File permission MUST be 666 */
5507 ret
= chmod(config
.apps_unix_sock_path
.value
,
5508 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5510 ERR("Set file permissions failed: %s", config
.apps_unix_sock_path
.value
);
5515 DBG3("Session daemon client socket %d and application socket %d created",
5516 client_sock
, apps_sock
);
5524 * Create lockfile using the rundir and return its fd.
5526 static int create_lockfile(void)
5528 return utils_create_lock_file(config
.lock_file_path
.value
);
5532 * Check if the global socket is available, and if a daemon is answering at the
5533 * other side. If yes, error is returned.
5535 * Also attempts to create and hold the lock file.
5537 static int check_existing_daemon(void)
5541 /* Is there anybody out there ? */
5542 if (lttng_session_daemon_alive()) {
5547 lockfile_fd
= create_lockfile();
5548 if (lockfile_fd
< 0) {
5556 static void sessiond_cleanup_lock_file(void)
5561 * Cleanup lock file by deleting it and finaly closing it which will
5562 * release the file system lock.
5564 if (lockfile_fd
>= 0) {
5565 ret
= remove(config
.lock_file_path
.value
);
5567 PERROR("remove lock file");
5569 ret
= close(lockfile_fd
);
5571 PERROR("close lock file");
5577 * Set the tracing group gid onto the client socket.
5579 * Race window between mkdir and chown is OK because we are going from more
5580 * permissive (root.root) to less permissive (root.tracing).
5582 static int set_permissions(char *rundir
)
5587 gid
= utils_get_group_id(config
.tracing_group_name
.value
);
5589 /* Set lttng run dir */
5590 ret
= chown(rundir
, 0, gid
);
5592 ERR("Unable to set group on %s", rundir
);
5597 * Ensure all applications and tracing group can search the run
5598 * dir. Allow everyone to read the directory, since it does not
5599 * buy us anything to hide its content.
5601 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5603 ERR("Unable to set permissions on %s", rundir
);
5607 /* lttng client socket path */
5608 ret
= chown(config
.client_unix_sock_path
.value
, 0, gid
);
5610 ERR("Unable to set group on %s", config
.client_unix_sock_path
.value
);
5614 /* kconsumer error socket path */
5615 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5617 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5621 /* 64-bit ustconsumer error socket path */
5622 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5624 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5628 /* 32-bit ustconsumer compat32 error socket path */
5629 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5631 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5635 DBG("All permissions are set");
5641 * Create the lttng run directory needed for all global sockets and pipe.
5643 static int create_lttng_rundir(void)
5647 DBG3("Creating LTTng run directory: %s", config
.rundir
.value
);
5649 ret
= mkdir(config
.rundir
.value
, S_IRWXU
);
5651 if (errno
!= EEXIST
) {
5652 ERR("Unable to create %s", config
.rundir
.value
);
5664 * Setup sockets and directory needed by the consumerds' communication with the
5667 static int set_consumer_sockets(struct consumer_data
*consumer_data
)
5672 switch (consumer_data
->type
) {
5673 case LTTNG_CONSUMER_KERNEL
:
5674 path
= config
.kconsumerd_path
.value
;
5676 case LTTNG_CONSUMER64_UST
:
5677 path
= config
.consumerd64_path
.value
;
5679 case LTTNG_CONSUMER32_UST
:
5680 path
= config
.consumerd32_path
.value
;
5683 ERR("Consumer type unknown");
5689 DBG2("Creating consumer directory: %s", path
);
5691 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5692 if (ret
< 0 && errno
!= EEXIST
) {
5694 ERR("Failed to create %s", path
);
5698 ret
= chown(path
, 0, utils_get_group_id(config
.tracing_group_name
.value
));
5700 ERR("Unable to set group on %s", path
);
5706 /* Create the consumerd error unix socket */
5707 consumer_data
->err_sock
=
5708 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5709 if (consumer_data
->err_sock
< 0) {
5710 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5716 * Set the CLOEXEC flag. Return code is useless because either way, the
5719 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5721 PERROR("utils_set_fd_cloexec");
5722 /* continue anyway */
5725 /* File permission MUST be 660 */
5726 ret
= chmod(consumer_data
->err_unix_sock_path
,
5727 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5729 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5739 * Signal handler for the daemon
5741 * Simply stop all worker threads, leaving main() return gracefully after
5742 * joining all threads and calling cleanup().
5744 static void sighandler(int sig
)
5748 DBG("SIGINT caught");
5752 DBG("SIGTERM caught");
5756 CMM_STORE_SHARED(recv_child_signal
, 1);
5764 * Setup signal handler for :
5765 * SIGINT, SIGTERM, SIGPIPE
5767 static int set_signal_handler(void)
5770 struct sigaction sa
;
5773 if ((ret
= sigemptyset(&sigset
)) < 0) {
5774 PERROR("sigemptyset");
5778 sa
.sa_mask
= sigset
;
5781 sa
.sa_handler
= sighandler
;
5782 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5783 PERROR("sigaction");
5787 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5788 PERROR("sigaction");
5792 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5793 PERROR("sigaction");
5797 sa
.sa_handler
= SIG_IGN
;
5798 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5799 PERROR("sigaction");
5803 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5809 * Set open files limit to unlimited. This daemon can open a large number of
5810 * file descriptors in order to consume multiple kernel traces.
5812 static void set_ulimit(void)
5817 /* The kernel does not allow an infinite limit for open files */
5818 lim
.rlim_cur
= 65535;
5819 lim
.rlim_max
= 65535;
5821 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5823 PERROR("failed to set open files limit");
5827 static int write_pidfile(void)
5829 return utils_create_pid_file(getpid(), config
.pid_file_path
.value
);
5832 static int set_clock_plugin_env(void)
5835 char *env_value
= NULL
;
5837 if (!config
.lttng_ust_clock_plugin
.value
) {
5841 ret
= asprintf(&env_value
, "LTTNG_UST_CLOCK_PLUGIN=%s",
5842 config
.lttng_ust_clock_plugin
.value
);
5848 ret
= putenv(env_value
);
5851 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5855 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5856 config
.lttng_ust_clock_plugin
.value
);
5862 struct rotation_thread_timer_queue
*create_rotate_timer_queue(void)
5864 struct rotation_thread_timer_queue
*queue
= NULL
;
5866 queue
= zmalloc(sizeof(struct rotation_thread_timer_queue
));
5868 PERROR("Failed to allocate timer rotate queue");
5872 queue
->event_pipe
= lttng_pipe_open(FD_CLOEXEC
| O_NONBLOCK
);
5873 CDS_INIT_LIST_HEAD(&queue
->list
);
5874 pthread_mutex_init(&queue
->lock
, NULL
);
5881 void destroy_rotate_timer_queue(struct rotation_thread_timer_queue
*queue
)
5883 struct sessiond_rotation_timer
*node
, *tmp_node
;
5889 lttng_pipe_destroy(queue
->event_pipe
);
5891 pthread_mutex_lock(&queue
->lock
);
5892 /* Empty wait queue. */
5893 cds_list_for_each_entry_safe(node
, tmp_node
, &queue
->list
, head
) {
5894 cds_list_del(&node
->head
);
5897 pthread_mutex_unlock(&queue
->lock
);
5899 pthread_mutex_destroy(&queue
->lock
);
5906 int main(int argc
, char **argv
)
5908 int ret
= 0, retval
= 0;
5910 const char *env_app_timeout
;
5911 struct lttng_pipe
*ust32_channel_monitor_pipe
= NULL
,
5912 *ust64_channel_monitor_pipe
= NULL
,
5913 *kernel_channel_monitor_pipe
= NULL
;
5914 bool notification_thread_launched
= false;
5915 bool rotation_thread_launched
= false;
5916 bool timer_thread_launched
= false;
5917 struct lttng_pipe
*ust32_channel_rotate_pipe
= NULL
,
5918 *ust64_channel_rotate_pipe
= NULL
,
5919 *kernel_channel_rotate_pipe
= NULL
;
5920 struct timer_thread_parameters timer_thread_ctx
;
5921 /* Queue of rotation jobs populated by the sessiond-timer. */
5922 struct rotation_thread_timer_queue
*rotation_timer_queue
= NULL
;
5923 sem_t notification_thread_ready
;
5925 init_kernel_workarounds();
5927 rcu_register_thread();
5929 if (set_signal_handler()) {
5931 goto exit_set_signal_handler
;
5934 if (sessiond_timer_signal_init()) {
5936 goto exit_set_signal_handler
;
5939 page_size
= sysconf(_SC_PAGESIZE
);
5940 if (page_size
< 0) {
5941 PERROR("sysconf _SC_PAGESIZE");
5942 page_size
= LONG_MAX
;
5943 WARN("Fallback page size to %ld", page_size
);
5946 ret
= sessiond_config_init(&config
);
5949 goto exit_set_signal_handler
;
5953 * Init config from environment variables.
5954 * Command line option override env configuration per-doc. Do env first.
5956 sessiond_config_apply_env_config(&config
);
5959 * Parse arguments and load the daemon configuration file.
5961 * We have an exit_options exit path to free memory reserved by
5962 * set_options. This is needed because the rest of sessiond_cleanup()
5963 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5964 * depends on set_options.
5967 if (set_options(argc
, argv
)) {
5973 * Resolve all paths received as arguments, configuration option, or
5974 * through environment variable as absolute paths. This is necessary
5975 * since daemonizing causes the sessiond's current working directory
5978 ret
= sessiond_config_resolve_paths(&config
);
5984 lttng_opt_verbose
= config
.verbose
;
5985 lttng_opt_quiet
= config
.quiet
;
5986 kconsumer_data
.err_unix_sock_path
=
5987 config
.kconsumerd_err_unix_sock_path
.value
;
5988 kconsumer_data
.cmd_unix_sock_path
=
5989 config
.kconsumerd_cmd_unix_sock_path
.value
;
5990 ustconsumer32_data
.err_unix_sock_path
=
5991 config
.consumerd32_err_unix_sock_path
.value
;
5992 ustconsumer32_data
.cmd_unix_sock_path
=
5993 config
.consumerd32_cmd_unix_sock_path
.value
;
5994 ustconsumer64_data
.err_unix_sock_path
=
5995 config
.consumerd64_err_unix_sock_path
.value
;
5996 ustconsumer64_data
.cmd_unix_sock_path
=
5997 config
.consumerd64_cmd_unix_sock_path
.value
;
5998 set_clock_plugin_env();
6000 sessiond_config_log(&config
);
6002 if (create_lttng_rundir()) {
6007 /* Abort launch if a session daemon is already running. */
6008 if (check_existing_daemon()) {
6009 ERR("A session daemon is already running.");
6015 if (config
.daemonize
|| config
.background
) {
6018 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
6019 !config
.background
);
6026 * We are in the child. Make sure all other file descriptors are
6027 * closed, in case we are called with more opened file
6028 * descriptors than the standard ones and the lock file.
6030 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
6031 if (i
== lockfile_fd
) {
6038 if (run_as_create_worker(argv
[0]) < 0) {
6039 goto exit_create_run_as_worker_cleanup
;
6043 * Starting from here, we can create threads. This needs to be after
6044 * lttng_daemonize due to RCU.
6048 * Initialize the health check subsystem. This call should set the
6049 * appropriate time values.
6051 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
6052 if (!health_sessiond
) {
6053 PERROR("health_app_create error");
6055 goto exit_health_sessiond_cleanup
;
6058 /* Create thread to clean up RCU hash tables */
6059 if (init_ht_cleanup_thread(&ht_cleanup_thread
)) {
6061 goto exit_ht_cleanup
;
6064 /* Create thread quit pipe */
6065 if (init_thread_quit_pipe()) {
6067 goto exit_init_data
;
6070 /* Check if daemon is UID = 0 */
6071 is_root
= !getuid();
6073 /* Create global run dir with root access */
6075 kernel_channel_monitor_pipe
= lttng_pipe_open(0);
6076 if (!kernel_channel_monitor_pipe
) {
6077 ERR("Failed to create kernel consumer channel monitor pipe");
6079 goto exit_init_data
;
6081 kconsumer_data
.channel_monitor_pipe
=
6082 lttng_pipe_release_writefd(
6083 kernel_channel_monitor_pipe
);
6084 if (kconsumer_data
.channel_monitor_pipe
< 0) {
6086 goto exit_init_data
;
6088 kernel_channel_rotate_pipe
= lttng_pipe_open(0);
6089 if (!kernel_channel_rotate_pipe
) {
6090 ERR("Failed to create kernel consumer channel rotate pipe");
6092 goto exit_init_data
;
6094 kconsumer_data
.channel_rotate_pipe
=
6095 lttng_pipe_release_writefd(
6096 kernel_channel_rotate_pipe
);
6097 if (kconsumer_data
.channel_rotate_pipe
< 0) {
6099 goto exit_init_data
;
6103 /* Set consumer initial state */
6104 kernel_consumerd_state
= CONSUMER_STOPPED
;
6105 ust_consumerd_state
= CONSUMER_STOPPED
;
6107 ust32_channel_monitor_pipe
= lttng_pipe_open(0);
6108 if (!ust32_channel_monitor_pipe
) {
6109 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
6111 goto exit_init_data
;
6113 ustconsumer32_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
6114 ust32_channel_monitor_pipe
);
6115 if (ustconsumer32_data
.channel_monitor_pipe
< 0) {
6117 goto exit_init_data
;
6119 ust32_channel_rotate_pipe
= lttng_pipe_open(0);
6120 if (!ust32_channel_rotate_pipe
) {
6121 ERR("Failed to create 32-bit user space consumer channel rotate pipe");
6123 goto exit_init_data
;
6125 ustconsumer32_data
.channel_rotate_pipe
= lttng_pipe_release_writefd(
6126 ust32_channel_rotate_pipe
);
6127 if (ustconsumer32_data
.channel_rotate_pipe
< 0) {
6129 goto exit_init_data
;
6133 * The rotation_timer_queue structure is shared between the sessiond timer
6134 * thread and the rotation thread. The main() keeps the ownership and
6135 * destroys it when both threads have quit.
6137 rotation_timer_queue
= create_rotate_timer_queue();
6138 if (!rotation_timer_queue
) {
6140 goto exit_init_data
;
6142 timer_thread_ctx
.rotation_timer_queue
= rotation_timer_queue
;
6144 ust64_channel_monitor_pipe
= lttng_pipe_open(0);
6145 if (!ust64_channel_monitor_pipe
) {
6146 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
6148 goto exit_init_data
;
6150 ustconsumer64_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
6151 ust64_channel_monitor_pipe
);
6152 if (ustconsumer64_data
.channel_monitor_pipe
< 0) {
6154 goto exit_init_data
;
6156 ust64_channel_rotate_pipe
= lttng_pipe_open(0);
6157 if (!ust64_channel_rotate_pipe
) {
6158 ERR("Failed to create 64-bit user space consumer channel rotate pipe");
6160 goto exit_init_data
;
6162 ustconsumer64_data
.channel_rotate_pipe
= lttng_pipe_release_writefd(
6163 ust64_channel_rotate_pipe
);
6164 if (ustconsumer64_data
.channel_rotate_pipe
< 0) {
6166 goto exit_init_data
;
6170 * Init UST app hash table. Alloc hash table before this point since
6171 * cleanup() can get called after that point.
6173 if (ust_app_ht_alloc()) {
6174 ERR("Failed to allocate UST app hash table");
6176 goto exit_init_data
;
6180 * Initialize agent app hash table. We allocate the hash table here
6181 * since cleanup() can get called after this point.
6183 if (agent_app_ht_alloc()) {
6184 ERR("Failed to allocate Agent app hash table");
6186 goto exit_init_data
;
6190 * These actions must be executed as root. We do that *after* setting up
6191 * the sockets path because we MUST make the check for another daemon using
6192 * those paths *before* trying to set the kernel consumer sockets and init
6196 if (set_consumer_sockets(&kconsumer_data
)) {
6198 goto exit_init_data
;
6201 /* Setup kernel tracer */
6202 if (!config
.no_kernel
) {
6203 init_kernel_tracer();
6204 if (kernel_tracer_fd
>= 0) {
6205 ret
= syscall_init_table();
6207 ERR("Unable to populate syscall table. "
6208 "Syscall tracing won't work "
6209 "for this session daemon.");
6214 /* Set ulimit for open files */
6217 /* init lttng_fd tracking must be done after set_ulimit. */
6220 if (set_consumer_sockets(&ustconsumer64_data
)) {
6222 goto exit_init_data
;
6225 if (set_consumer_sockets(&ustconsumer32_data
)) {
6227 goto exit_init_data
;
6230 /* Setup the needed unix socket */
6231 if (init_daemon_socket()) {
6233 goto exit_init_data
;
6236 /* Set credentials to socket */
6237 if (is_root
&& set_permissions(config
.rundir
.value
)) {
6239 goto exit_init_data
;
6242 /* Get parent pid if -S, --sig-parent is specified. */
6243 if (config
.sig_parent
) {
6247 /* Setup the kernel pipe for waking up the kernel thread */
6248 if (is_root
&& !config
.no_kernel
) {
6249 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
6251 goto exit_init_data
;
6255 /* Setup the thread apps communication pipe. */
6256 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
6258 goto exit_init_data
;
6261 /* Setup the thread apps notify communication pipe. */
6262 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
6264 goto exit_init_data
;
6267 /* Initialize global buffer per UID and PID registry. */
6268 buffer_reg_init_uid_registry();
6269 buffer_reg_init_pid_registry();
6271 /* Init UST command queue. */
6272 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
6275 * Get session list pointer. This pointer MUST NOT be free'd. This list
6276 * is statically declared in session.c
6278 session_list_ptr
= session_get_list();
6282 /* Check for the application socket timeout env variable. */
6283 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
6284 if (env_app_timeout
) {
6285 config
.app_socket_timeout
= atoi(env_app_timeout
);
6287 config
.app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
6290 ret
= write_pidfile();
6292 ERR("Error in write_pidfile");
6294 goto exit_init_data
;
6297 /* Initialize communication library */
6299 /* Initialize TCP timeout values */
6300 lttcomm_inet_init();
6302 if (load_session_init_data(&load_info
) < 0) {
6304 goto exit_init_data
;
6306 load_info
->path
= config
.load_session_path
.value
;
6308 /* Create health-check thread. */
6309 ret
= pthread_create(&health_thread
, default_pthread_attr(),
6310 thread_manage_health
, (void *) NULL
);
6313 PERROR("pthread_create health");
6319 * The rotation thread needs the notification thread to be ready before
6320 * creating the rotate_notification_channel, so we use this semaphore as
6321 * a rendez-vous point.
6323 sem_init(¬ification_thread_ready
, 0, 0);
6325 /* notification_thread_data acquires the pipes' read side. */
6326 notification_thread_handle
= notification_thread_handle_create(
6327 ust32_channel_monitor_pipe
,
6328 ust64_channel_monitor_pipe
,
6329 kernel_channel_monitor_pipe
,
6330 ¬ification_thread_ready
);
6331 if (!notification_thread_handle
) {
6333 ERR("Failed to create notification thread shared data");
6335 goto exit_notification
;
6338 /* Create notification thread. */
6339 ret
= pthread_create(¬ification_thread
, default_pthread_attr(),
6340 thread_notification
, notification_thread_handle
);
6343 PERROR("pthread_create notification");
6346 goto exit_notification
;
6348 notification_thread_launched
= true;
6350 /* Create timer thread. */
6351 ret
= pthread_create(&timer_thread
, default_pthread_attr(),
6352 sessiond_timer_thread
, &timer_thread_ctx
);
6355 PERROR("pthread_create timer");
6358 goto exit_notification
;
6360 timer_thread_launched
= true;
6362 /* rotation_thread_data acquires the pipes' read side. */
6363 rotation_thread_handle
= rotation_thread_handle_create(
6364 ust32_channel_rotate_pipe
,
6365 ust64_channel_rotate_pipe
,
6366 kernel_channel_rotate_pipe
,
6367 thread_quit_pipe
[0],
6368 rotation_timer_queue
,
6369 notification_thread_handle
,
6370 ¬ification_thread_ready
);
6371 if (!rotation_thread_handle
) {
6373 ERR("Failed to create rotation thread shared data");
6378 /* Create rotation thread. */
6379 ret
= pthread_create(&rotation_thread
, default_pthread_attr(),
6380 thread_rotation
, rotation_thread_handle
);
6383 PERROR("pthread_create rotation");
6388 rotation_thread_launched
= true;
6390 /* Create thread to manage the client socket */
6391 ret
= pthread_create(&client_thread
, default_pthread_attr(),
6392 thread_manage_clients
, (void *) NULL
);
6395 PERROR("pthread_create clients");
6401 /* Create thread to dispatch registration */
6402 ret
= pthread_create(&dispatch_thread
, default_pthread_attr(),
6403 thread_dispatch_ust_registration
, (void *) NULL
);
6406 PERROR("pthread_create dispatch");
6412 /* Create thread to manage application registration. */
6413 ret
= pthread_create(®_apps_thread
, default_pthread_attr(),
6414 thread_registration_apps
, (void *) NULL
);
6417 PERROR("pthread_create registration");
6423 /* Create thread to manage application socket */
6424 ret
= pthread_create(&apps_thread
, default_pthread_attr(),
6425 thread_manage_apps
, (void *) NULL
);
6428 PERROR("pthread_create apps");
6434 /* Create thread to manage application notify socket */
6435 ret
= pthread_create(&apps_notify_thread
, default_pthread_attr(),
6436 ust_thread_manage_notify
, (void *) NULL
);
6439 PERROR("pthread_create notify");
6442 goto exit_apps_notify
;
6445 /* Create agent registration thread. */
6446 ret
= pthread_create(&agent_reg_thread
, default_pthread_attr(),
6447 agent_thread_manage_registration
, (void *) NULL
);
6450 PERROR("pthread_create agent");
6453 goto exit_agent_reg
;
6456 /* Don't start this thread if kernel tracing is not requested nor root */
6457 if (is_root
&& !config
.no_kernel
) {
6458 /* Create kernel thread to manage kernel event */
6459 ret
= pthread_create(&kernel_thread
, default_pthread_attr(),
6460 thread_manage_kernel
, (void *) NULL
);
6463 PERROR("pthread_create kernel");
6470 /* Create session loading thread. */
6471 ret
= pthread_create(&load_session_thread
, default_pthread_attr(),
6472 thread_load_session
, load_info
);
6475 PERROR("pthread_create load_session_thread");
6478 goto exit_load_session
;
6482 * This is where we start awaiting program completion (e.g. through
6483 * signal that asks threads to teardown).
6486 ret
= pthread_join(load_session_thread
, &status
);
6489 PERROR("pthread_join load_session_thread");
6494 if (is_root
&& !config
.no_kernel
) {
6495 ret
= pthread_join(kernel_thread
, &status
);
6498 PERROR("pthread_join");
6504 ret
= pthread_join(agent_reg_thread
, &status
);
6507 PERROR("pthread_join agent");
6512 ret
= pthread_join(apps_notify_thread
, &status
);
6515 PERROR("pthread_join apps notify");
6520 ret
= pthread_join(apps_thread
, &status
);
6523 PERROR("pthread_join apps");
6528 ret
= pthread_join(reg_apps_thread
, &status
);
6531 PERROR("pthread_join");
6537 * Join dispatch thread after joining reg_apps_thread to ensure
6538 * we don't leak applications in the queue.
6540 ret
= pthread_join(dispatch_thread
, &status
);
6543 PERROR("pthread_join");
6548 ret
= pthread_join(client_thread
, &status
);
6551 PERROR("pthread_join");
6558 sem_destroy(¬ification_thread_ready
);
6559 ret
= pthread_join(health_thread
, &status
);
6562 PERROR("pthread_join health thread");
6569 * Wait for all pending call_rcu work to complete before tearing
6570 * down data structures. call_rcu worker may be trying to
6571 * perform lookups in those structures.
6575 * sessiond_cleanup() is called when no other thread is running, except
6576 * the ht_cleanup thread, which is needed to destroy the hash tables.
6578 rcu_thread_online();
6582 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6583 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6584 * the queue is empty before shutting down the clean-up thread.
6589 * The teardown of the notification system is performed after the
6590 * session daemon's teardown in order to allow it to be notified
6591 * of the active session and channels at the moment of the teardown.
6593 if (notification_thread_handle
) {
6594 if (notification_thread_launched
) {
6595 notification_thread_command_quit(
6596 notification_thread_handle
);
6597 ret
= pthread_join(notification_thread
, &status
);
6600 PERROR("pthread_join notification thread");
6604 notification_thread_handle_destroy(notification_thread_handle
);
6607 if (rotation_thread_handle
) {
6608 if (rotation_thread_launched
) {
6609 ret
= pthread_join(rotation_thread
, &status
);
6612 PERROR("pthread_join rotation thread");
6616 rotation_thread_handle_destroy(rotation_thread_handle
);
6619 if (timer_thread_launched
) {
6620 kill(getpid(), LTTNG_SESSIOND_SIG_EXIT
);
6621 ret
= pthread_join(timer_thread
, &status
);
6624 PERROR("pthread_join timer thread");
6630 * After the rotation and timer thread have quit, we can safely destroy
6631 * the rotation_timer_queue.
6633 destroy_rotate_timer_queue(rotation_timer_queue
);
6635 rcu_thread_offline();
6636 rcu_unregister_thread();
6638 ret
= fini_ht_cleanup_thread(&ht_cleanup_thread
);
6642 lttng_pipe_destroy(ust32_channel_monitor_pipe
);
6643 lttng_pipe_destroy(ust64_channel_monitor_pipe
);
6644 lttng_pipe_destroy(kernel_channel_monitor_pipe
);
6645 lttng_pipe_destroy(ust32_channel_rotate_pipe
);
6646 lttng_pipe_destroy(ust64_channel_rotate_pipe
);
6647 lttng_pipe_destroy(kernel_channel_rotate_pipe
);
6650 health_app_destroy(health_sessiond
);
6651 exit_health_sessiond_cleanup
:
6652 exit_create_run_as_worker_cleanup
:
6655 sessiond_cleanup_lock_file();
6656 sessiond_cleanup_options();
6658 exit_set_signal_handler
: