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>
53 #include <lttng/userspace-probe-internal.h>
54 #include <lttng/event-internal.h>
56 #include "lttng-sessiond.h"
57 #include "buffer-registry.h"
64 #include "kernel-consumer.h"
68 #include "ust-consumer.h"
71 #include "health-sessiond.h"
72 #include "testpoint.h"
73 #include "ust-thread.h"
74 #include "agent-thread.h"
76 #include "load-session-thread.h"
77 #include "notification-thread.h"
78 #include "notification-thread-commands.h"
79 #include "rotation-thread.h"
80 #include "lttng-syscall.h"
82 #include "ht-cleanup.h"
83 #include "sessiond-config.h"
84 #include "sessiond-timer.h"
86 static const char *help_msg
=
87 #ifdef LTTNG_EMBED_HELP
88 #include <lttng-sessiond.8.h>
95 static pid_t ppid
; /* Parent PID for --sig-parent option */
96 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
97 static int lockfile_fd
= -1;
99 /* Set to 1 when a SIGUSR1 signal is received. */
100 static int recv_child_signal
;
102 static struct lttng_kernel_tracer_version kernel_tracer_version
;
103 static struct lttng_kernel_tracer_abi_version kernel_tracer_abi_version
;
106 * Consumer daemon specific control data. Every value not initialized here is
107 * set to 0 by the static definition.
109 static struct consumer_data kconsumer_data
= {
110 .type
= LTTNG_CONSUMER_KERNEL
,
113 .channel_monitor_pipe
= -1,
114 .channel_rotate_pipe
= -1,
115 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
116 .lock
= PTHREAD_MUTEX_INITIALIZER
,
117 .cond
= PTHREAD_COND_INITIALIZER
,
118 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
120 static struct consumer_data ustconsumer64_data
= {
121 .type
= LTTNG_CONSUMER64_UST
,
124 .channel_monitor_pipe
= -1,
125 .channel_rotate_pipe
= -1,
126 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
127 .lock
= PTHREAD_MUTEX_INITIALIZER
,
128 .cond
= PTHREAD_COND_INITIALIZER
,
129 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
131 static struct consumer_data ustconsumer32_data
= {
132 .type
= LTTNG_CONSUMER32_UST
,
135 .channel_monitor_pipe
= -1,
136 .channel_rotate_pipe
= -1,
137 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
138 .lock
= PTHREAD_MUTEX_INITIALIZER
,
139 .cond
= PTHREAD_COND_INITIALIZER
,
140 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
143 /* Command line options */
144 static const struct option long_options
[] = {
145 { "client-sock", required_argument
, 0, 'c' },
146 { "apps-sock", required_argument
, 0, 'a' },
147 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
148 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
149 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
150 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
151 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
152 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
153 { "consumerd32-path", required_argument
, 0, '\0' },
154 { "consumerd32-libdir", required_argument
, 0, '\0' },
155 { "consumerd64-path", required_argument
, 0, '\0' },
156 { "consumerd64-libdir", required_argument
, 0, '\0' },
157 { "daemonize", no_argument
, 0, 'd' },
158 { "background", no_argument
, 0, 'b' },
159 { "sig-parent", no_argument
, 0, 'S' },
160 { "help", no_argument
, 0, 'h' },
161 { "group", required_argument
, 0, 'g' },
162 { "version", no_argument
, 0, 'V' },
163 { "quiet", no_argument
, 0, 'q' },
164 { "verbose", no_argument
, 0, 'v' },
165 { "verbose-consumer", no_argument
, 0, '\0' },
166 { "no-kernel", no_argument
, 0, '\0' },
167 { "pidfile", required_argument
, 0, 'p' },
168 { "agent-tcp-port", required_argument
, 0, '\0' },
169 { "config", required_argument
, 0, 'f' },
170 { "load", required_argument
, 0, 'l' },
171 { "kmod-probes", required_argument
, 0, '\0' },
172 { "extra-kmod-probes", required_argument
, 0, '\0' },
176 struct sessiond_config config
;
178 /* Command line options to ignore from configuration file */
179 static const char *config_ignore_options
[] = { "help", "version", "config" };
181 /* Shared between threads */
182 static int dispatch_thread_exit
;
184 /* Sockets and FDs */
185 static int client_sock
= -1;
186 static int apps_sock
= -1;
187 int kernel_tracer_fd
= -1;
188 static int kernel_poll_pipe
[2] = { -1, -1 };
191 * Quit pipe for all threads. This permits a single cancellation point
192 * for all threads when receiving an event on the pipe.
194 static int thread_quit_pipe
[2] = { -1, -1 };
197 * This pipe is used to inform the thread managing application communication
198 * that a command is queued and ready to be processed.
200 static int apps_cmd_pipe
[2] = { -1, -1 };
202 int apps_cmd_notify_pipe
[2] = { -1, -1 };
204 /* Pthread, Mutexes and Semaphores */
205 static pthread_t apps_thread
;
206 static pthread_t apps_notify_thread
;
207 static pthread_t reg_apps_thread
;
208 static pthread_t client_thread
;
209 static pthread_t kernel_thread
;
210 static pthread_t dispatch_thread
;
211 static pthread_t health_thread
;
212 static pthread_t ht_cleanup_thread
;
213 static pthread_t agent_reg_thread
;
214 static pthread_t load_session_thread
;
215 static pthread_t notification_thread
;
216 static pthread_t rotation_thread
;
217 static pthread_t timer_thread
;
220 * UST registration command queue. This queue is tied with a futex and uses a N
221 * wakers / 1 waiter implemented and detailed in futex.c/.h
223 * The thread_registration_apps and thread_dispatch_ust_registration uses this
224 * queue along with the wait/wake scheme. The thread_manage_apps receives down
225 * the line new application socket and monitors it for any I/O error or clean
226 * close that triggers an unregistration of the application.
228 static struct ust_cmd_queue ust_cmd_queue
;
231 * Pointer initialized before thread creation.
233 * This points to the tracing session list containing the session count and a
234 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
235 * MUST NOT be taken if you call a public function in session.c.
237 * The lock is nested inside the structure: session_list_ptr->lock. Please use
238 * session_lock_list and session_unlock_list for lock acquisition.
240 static struct ltt_session_list
*session_list_ptr
;
242 int ust_consumerd64_fd
= -1;
243 int ust_consumerd32_fd
= -1;
245 static const char *module_proc_lttng
= "/proc/lttng";
248 * Consumer daemon state which is changed when spawning it, killing it or in
249 * case of a fatal error.
251 enum consumerd_state
{
252 CONSUMER_STARTED
= 1,
253 CONSUMER_STOPPED
= 2,
258 * This consumer daemon state is used to validate if a client command will be
259 * able to reach the consumer. If not, the client is informed. For instance,
260 * doing a "lttng start" when the consumer state is set to ERROR will return an
261 * error to the client.
263 * The following example shows a possible race condition of this scheme:
265 * consumer thread error happens
267 * client cmd checks state -> still OK
268 * consumer thread exit, sets error
269 * client cmd try to talk to consumer
272 * However, since the consumer is a different daemon, we have no way of making
273 * sure the command will reach it safely even with this state flag. This is why
274 * we consider that up to the state validation during command processing, the
275 * command is safe. After that, we can not guarantee the correctness of the
276 * client request vis-a-vis the consumer.
278 static enum consumerd_state ust_consumerd_state
;
279 static enum consumerd_state kernel_consumerd_state
;
281 /* Set in main() with the current page size. */
284 /* Application health monitoring */
285 struct health_app
*health_sessiond
;
287 /* Am I root or not. */
288 int is_root
; /* Set to 1 if the daemon is running as root */
290 const char * const config_section_name
= "sessiond";
292 /* Load session thread information to operate. */
293 struct load_session_thread_data
*load_info
;
295 /* Notification thread handle. */
296 struct notification_thread_handle
*notification_thread_handle
;
298 /* Rotation thread handle. */
299 struct rotation_thread_handle
*rotation_thread_handle
;
301 /* Global hash tables */
302 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
305 * The initialization of the session daemon is done in multiple phases.
307 * While all threads are launched near-simultaneously, only some of them
308 * are needed to ensure the session daemon can start to respond to client
311 * There are two important guarantees that we wish to offer with respect
312 * to the initialisation of the session daemon:
313 * - When the daemonize/background launcher process exits, the sessiond
314 * is fully able to respond to client requests,
315 * - Auto-loaded sessions are visible to clients.
317 * In order to achieve this, a number of support threads have to be launched
318 * to allow the "client" thread to function properly. Moreover, since the
319 * "load session" thread needs the client thread, we must provide a way
320 * for the "load session" thread to know that the "client" thread is up
323 * Hence, the support threads decrement the lttng_sessiond_ready counter
324 * while the "client" threads waits for it to reach 0. Once the "client" thread
325 * unblocks, it posts the message_thread_ready semaphore which allows the
326 * "load session" thread to progress.
328 * This implies that the "load session" thread is the last to be initialized
329 * and will explicitly call sessiond_signal_parents(), which signals the parents
330 * that the session daemon is fully initialized.
332 * The four (4) support threads are:
334 * - notification_thread
338 #define NR_LTTNG_SESSIOND_SUPPORT_THREADS 4
339 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_SUPPORT_THREADS
;
341 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
343 return (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) ? 1 : 0;
346 /* Notify parents that we are ready for cmd and health check */
348 void sessiond_signal_parents(void)
351 * Notify parent pid that we are ready to accept command
352 * for client side. This ppid is the one from the
353 * external process that spawned us.
355 if (config
.sig_parent
) {
360 * Notify the parent of the fork() process that we are
363 if (config
.daemonize
|| config
.background
) {
364 kill(child_ppid
, SIGUSR1
);
369 void sessiond_notify_ready(void)
372 * This memory barrier is paired with the one performed by
373 * the client thread after it has seen that 'lttng_sessiond_ready' is 0.
375 * The purpose of these memory barriers is to ensure that all
376 * initialization operations of the various threads that call this
377 * function to signal that they are ready are commited/published
378 * before the client thread can see the 'lttng_sessiond_ready' counter
381 * Note that this could be a 'write' memory barrier, but a full barrier
382 * is used in case the code using this utility changes. The performance
383 * implications of this choice are minimal since this is a slow path.
386 uatomic_sub(<tng_sessiond_ready
, 1);
390 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
397 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
403 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
415 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
417 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
419 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
423 * Init thread quit pipe.
425 * Return -1 on error or 0 if all pipes are created.
427 static int __init_thread_quit_pipe(int *a_pipe
)
433 PERROR("thread quit pipe");
437 for (i
= 0; i
< 2; i
++) {
438 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
449 static int init_thread_quit_pipe(void)
451 return __init_thread_quit_pipe(thread_quit_pipe
);
455 * Stop all threads by closing the thread quit pipe.
457 static void stop_threads(void)
461 /* Stopping all threads */
462 DBG("Terminating all threads");
463 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
465 ERR("write error on thread quit pipe");
468 /* Dispatch thread */
469 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
470 futex_nto1_wake(&ust_cmd_queue
.futex
);
474 * Close every consumer sockets.
476 static void close_consumer_sockets(void)
480 if (kconsumer_data
.err_sock
>= 0) {
481 ret
= close(kconsumer_data
.err_sock
);
483 PERROR("kernel consumer err_sock close");
486 if (ustconsumer32_data
.err_sock
>= 0) {
487 ret
= close(ustconsumer32_data
.err_sock
);
489 PERROR("UST consumerd32 err_sock close");
492 if (ustconsumer64_data
.err_sock
>= 0) {
493 ret
= close(ustconsumer64_data
.err_sock
);
495 PERROR("UST consumerd64 err_sock close");
498 if (kconsumer_data
.cmd_sock
>= 0) {
499 ret
= close(kconsumer_data
.cmd_sock
);
501 PERROR("kernel consumer cmd_sock close");
504 if (ustconsumer32_data
.cmd_sock
>= 0) {
505 ret
= close(ustconsumer32_data
.cmd_sock
);
507 PERROR("UST consumerd32 cmd_sock close");
510 if (ustconsumer64_data
.cmd_sock
>= 0) {
511 ret
= close(ustconsumer64_data
.cmd_sock
);
513 PERROR("UST consumerd64 cmd_sock close");
516 if (kconsumer_data
.channel_monitor_pipe
>= 0) {
517 ret
= close(kconsumer_data
.channel_monitor_pipe
);
519 PERROR("kernel consumer channel monitor pipe close");
522 if (ustconsumer32_data
.channel_monitor_pipe
>= 0) {
523 ret
= close(ustconsumer32_data
.channel_monitor_pipe
);
525 PERROR("UST consumerd32 channel monitor pipe close");
528 if (ustconsumer64_data
.channel_monitor_pipe
>= 0) {
529 ret
= close(ustconsumer64_data
.channel_monitor_pipe
);
531 PERROR("UST consumerd64 channel monitor pipe close");
534 if (kconsumer_data
.channel_rotate_pipe
>= 0) {
535 ret
= close(kconsumer_data
.channel_rotate_pipe
);
537 PERROR("kernel consumer channel rotate pipe close");
540 if (ustconsumer32_data
.channel_rotate_pipe
>= 0) {
541 ret
= close(ustconsumer32_data
.channel_rotate_pipe
);
543 PERROR("UST consumerd32 channel rotate pipe close");
546 if (ustconsumer64_data
.channel_rotate_pipe
>= 0) {
547 ret
= close(ustconsumer64_data
.channel_rotate_pipe
);
549 PERROR("UST consumerd64 channel rotate pipe close");
555 * Wait on consumer process termination.
557 * Need to be called with the consumer data lock held or from a context
558 * ensuring no concurrent access to data (e.g: cleanup).
560 static void wait_consumer(struct consumer_data
*consumer_data
)
565 if (consumer_data
->pid
<= 0) {
569 DBG("Waiting for complete teardown of consumerd (PID: %d)",
571 ret
= waitpid(consumer_data
->pid
, &status
, 0);
573 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
574 } else if (!WIFEXITED(status
)) {
575 ERR("consumerd termination with error: %d",
578 consumer_data
->pid
= 0;
582 * Cleanup the session daemon's data structures.
584 static void sessiond_cleanup(void)
587 struct ltt_session
*sess
, *stmp
;
589 DBG("Cleanup sessiond");
592 * Close the thread quit pipe. It has already done its job,
593 * since we are now called.
595 utils_close_pipe(thread_quit_pipe
);
597 ret
= remove(config
.pid_file_path
.value
);
599 PERROR("remove pidfile %s", config
.pid_file_path
.value
);
602 DBG("Removing sessiond and consumerd content of directory %s",
603 config
.rundir
.value
);
606 DBG("Removing %s", config
.pid_file_path
.value
);
607 (void) unlink(config
.pid_file_path
.value
);
609 DBG("Removing %s", config
.agent_port_file_path
.value
);
610 (void) unlink(config
.agent_port_file_path
.value
);
613 DBG("Removing %s", kconsumer_data
.err_unix_sock_path
);
614 (void) unlink(kconsumer_data
.err_unix_sock_path
);
616 DBG("Removing directory %s", config
.kconsumerd_path
.value
);
617 (void) rmdir(config
.kconsumerd_path
.value
);
619 /* ust consumerd 32 */
620 DBG("Removing %s", config
.consumerd32_err_unix_sock_path
.value
);
621 (void) unlink(config
.consumerd32_err_unix_sock_path
.value
);
623 DBG("Removing directory %s", config
.consumerd32_path
.value
);
624 (void) rmdir(config
.consumerd32_path
.value
);
626 /* ust consumerd 64 */
627 DBG("Removing %s", config
.consumerd64_err_unix_sock_path
.value
);
628 (void) unlink(config
.consumerd64_err_unix_sock_path
.value
);
630 DBG("Removing directory %s", config
.consumerd64_path
.value
);
631 (void) rmdir(config
.consumerd64_path
.value
);
633 DBG("Cleaning up all sessions");
635 /* Destroy session list mutex */
636 if (session_list_ptr
!= NULL
) {
637 pthread_mutex_destroy(&session_list_ptr
->lock
);
639 /* Cleanup ALL session */
640 cds_list_for_each_entry_safe(sess
, stmp
,
641 &session_list_ptr
->head
, list
) {
642 cmd_destroy_session(sess
, kernel_poll_pipe
[1],
643 notification_thread_handle
);
647 wait_consumer(&kconsumer_data
);
648 wait_consumer(&ustconsumer64_data
);
649 wait_consumer(&ustconsumer32_data
);
651 DBG("Cleaning up all agent apps");
652 agent_app_ht_clean();
654 DBG("Closing all UST sockets");
655 ust_app_clean_list();
656 buffer_reg_destroy_registries();
658 if (is_root
&& !config
.no_kernel
) {
659 DBG2("Closing kernel fd");
660 if (kernel_tracer_fd
>= 0) {
661 ret
= close(kernel_tracer_fd
);
666 DBG("Unloading kernel modules");
667 modprobe_remove_lttng_all();
671 close_consumer_sockets();
674 load_session_destroy_data(load_info
);
679 * We do NOT rmdir rundir because there are other processes
680 * using it, for instance lttng-relayd, which can start in
681 * parallel with this teardown.
686 * Cleanup the daemon's option data structures.
688 static void sessiond_cleanup_options(void)
690 DBG("Cleaning up options");
692 sessiond_config_fini(&config
);
694 run_as_destroy_worker();
698 * Send data on a unix socket using the liblttsessiondcomm API.
700 * Return lttcomm error code.
702 static int send_unix_sock(int sock
, void *buf
, size_t len
)
704 /* Check valid length */
709 return lttcomm_send_unix_sock(sock
, buf
, len
);
713 * Free memory of a command context structure.
715 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
717 DBG("Clean command context structure");
719 if ((*cmd_ctx
)->llm
) {
720 free((*cmd_ctx
)->llm
);
722 if ((*cmd_ctx
)->lsm
) {
723 free((*cmd_ctx
)->lsm
);
731 * Notify UST applications using the shm mmap futex.
733 static int notify_ust_apps(int active
)
737 DBG("Notifying applications of session daemon state: %d", active
);
739 /* See shm.c for this call implying mmap, shm and futex calls */
740 wait_shm_mmap
= shm_ust_get_mmap(config
.wait_shm_path
.value
, is_root
);
741 if (wait_shm_mmap
== NULL
) {
745 /* Wake waiting process */
746 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
748 /* Apps notified successfully */
756 * Setup the outgoing data buffer for the response (llm) by allocating the
757 * right amount of memory and copying the original information from the lsm
760 * Return 0 on success, negative value on error.
762 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
763 const void *payload_buf
, size_t payload_len
,
764 const void *cmd_header_buf
, size_t cmd_header_len
)
767 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
768 const size_t cmd_header_offset
= header_len
;
769 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
770 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
772 cmd_ctx
->llm
= zmalloc(total_msg_size
);
774 if (cmd_ctx
->llm
== NULL
) {
780 /* Copy common data */
781 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
782 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
783 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
784 cmd_ctx
->llm
->data_size
= payload_len
;
785 cmd_ctx
->lttng_msg_size
= total_msg_size
;
787 /* Copy command header */
788 if (cmd_header_len
) {
789 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
795 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
804 * Version of setup_lttng_msg() without command header.
806 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
807 void *payload_buf
, size_t payload_len
)
809 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
812 * Update the kernel poll set of all channel fd available over all tracing
813 * session. Add the wakeup pipe at the end of the set.
815 static int update_kernel_poll(struct lttng_poll_event
*events
)
818 struct ltt_session
*session
;
819 struct ltt_kernel_channel
*channel
;
821 DBG("Updating kernel poll set");
824 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
825 session_lock(session
);
826 if (session
->kernel_session
== NULL
) {
827 session_unlock(session
);
831 cds_list_for_each_entry(channel
,
832 &session
->kernel_session
->channel_list
.head
, list
) {
833 /* Add channel fd to the kernel poll set */
834 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
836 session_unlock(session
);
839 DBG("Channel fd %d added to kernel set", channel
->fd
);
841 session_unlock(session
);
843 session_unlock_list();
848 session_unlock_list();
853 * Find the channel fd from 'fd' over all tracing session. When found, check
854 * for new channel stream and send those stream fds to the kernel consumer.
856 * Useful for CPU hotplug feature.
858 static int update_kernel_stream(int fd
)
861 struct ltt_session
*session
;
862 struct ltt_kernel_session
*ksess
;
863 struct ltt_kernel_channel
*channel
;
865 DBG("Updating kernel streams for channel fd %d", fd
);
868 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
869 session_lock(session
);
870 if (session
->kernel_session
== NULL
) {
871 session_unlock(session
);
874 ksess
= session
->kernel_session
;
876 cds_list_for_each_entry(channel
,
877 &ksess
->channel_list
.head
, list
) {
878 struct lttng_ht_iter iter
;
879 struct consumer_socket
*socket
;
881 if (channel
->fd
!= fd
) {
884 DBG("Channel found, updating kernel streams");
885 ret
= kernel_open_channel_stream(channel
);
889 /* Update the stream global counter */
890 ksess
->stream_count_global
+= ret
;
893 * Have we already sent fds to the consumer? If yes, it
894 * means that tracing is started so it is safe to send
895 * our updated stream fds.
897 if (ksess
->consumer_fds_sent
!= 1
898 || ksess
->consumer
== NULL
) {
904 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
905 &iter
.iter
, socket
, node
.node
) {
906 pthread_mutex_lock(socket
->lock
);
907 ret
= kernel_consumer_send_channel_streams(socket
,
909 session
->output_traces
? 1 : 0);
910 pthread_mutex_unlock(socket
->lock
);
918 session_unlock(session
);
920 session_unlock_list();
924 session_unlock(session
);
925 session_unlock_list();
930 * For each tracing session, update newly registered apps. The session list
931 * lock MUST be acquired before calling this.
933 static void update_ust_app(int app_sock
)
935 struct ltt_session
*sess
, *stmp
;
937 /* Consumer is in an ERROR state. Stop any application update. */
938 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
939 /* Stop the update process since the consumer is dead. */
943 /* For all tracing session(s) */
944 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
948 if (!sess
->ust_session
) {
953 assert(app_sock
>= 0);
954 app
= ust_app_find_by_sock(app_sock
);
957 * Application can be unregistered before so
958 * this is possible hence simply stopping the
961 DBG3("UST app update failed to find app sock %d",
965 ust_app_global_update(sess
->ust_session
, app
);
969 session_unlock(sess
);
974 * This thread manage event coming from the kernel.
976 * Features supported in this thread:
979 static void *thread_manage_kernel(void *data
)
981 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
982 uint32_t revents
, nb_fd
;
984 struct lttng_poll_event events
;
986 DBG("[thread] Thread manage kernel started");
988 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
991 * This first step of the while is to clean this structure which could free
992 * non NULL pointers so initialize it before the loop.
994 lttng_poll_init(&events
);
996 if (testpoint(sessiond_thread_manage_kernel
)) {
997 goto error_testpoint
;
1000 health_code_update();
1002 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1003 goto error_testpoint
;
1007 health_code_update();
1009 if (update_poll_flag
== 1) {
1010 /* Clean events object. We are about to populate it again. */
1011 lttng_poll_clean(&events
);
1013 ret
= sessiond_set_thread_pollset(&events
, 2);
1015 goto error_poll_create
;
1018 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1023 /* This will add the available kernel channel if any. */
1024 ret
= update_kernel_poll(&events
);
1028 update_poll_flag
= 0;
1031 DBG("Thread kernel polling");
1033 /* Poll infinite value of time */
1035 health_poll_entry();
1036 ret
= lttng_poll_wait(&events
, -1);
1037 DBG("Thread kernel return from poll on %d fds",
1038 LTTNG_POLL_GETNB(&events
));
1042 * Restart interrupted system call.
1044 if (errno
== EINTR
) {
1048 } else if (ret
== 0) {
1049 /* Should not happen since timeout is infinite */
1050 ERR("Return value of poll is 0 with an infinite timeout.\n"
1051 "This should not have happened! Continuing...");
1057 for (i
= 0; i
< nb_fd
; i
++) {
1058 /* Fetch once the poll data */
1059 revents
= LTTNG_POLL_GETEV(&events
, i
);
1060 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1062 health_code_update();
1065 /* No activity for this FD (poll implementation). */
1069 /* Thread quit pipe has been closed. Killing thread. */
1070 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1076 /* Check for data on kernel pipe */
1077 if (revents
& LPOLLIN
) {
1078 if (pollfd
== kernel_poll_pipe
[0]) {
1079 (void) lttng_read(kernel_poll_pipe
[0],
1082 * Ret value is useless here, if this pipe gets any actions an
1083 * update is required anyway.
1085 update_poll_flag
= 1;
1089 * New CPU detected by the kernel. Adding kernel stream to
1090 * kernel session and updating the kernel consumer
1092 ret
= update_kernel_stream(pollfd
);
1098 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1099 update_poll_flag
= 1;
1102 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1110 lttng_poll_clean(&events
);
1113 utils_close_pipe(kernel_poll_pipe
);
1114 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1117 ERR("Health error occurred in %s", __func__
);
1118 WARN("Kernel thread died unexpectedly. "
1119 "Kernel tracing can continue but CPU hotplug is disabled.");
1121 health_unregister(health_sessiond
);
1122 DBG("Kernel thread dying");
1127 * Signal pthread condition of the consumer data that the thread.
1129 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1131 pthread_mutex_lock(&data
->cond_mutex
);
1134 * The state is set before signaling. It can be any value, it's the waiter
1135 * job to correctly interpret this condition variable associated to the
1136 * consumer pthread_cond.
1138 * A value of 0 means that the corresponding thread of the consumer data
1139 * was not started. 1 indicates that the thread has started and is ready
1140 * for action. A negative value means that there was an error during the
1143 data
->consumer_thread_is_ready
= state
;
1144 (void) pthread_cond_signal(&data
->cond
);
1146 pthread_mutex_unlock(&data
->cond_mutex
);
1150 * This thread manage the consumer error sent back to the session daemon.
1152 static void *thread_manage_consumer(void *data
)
1154 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1155 uint32_t revents
, nb_fd
;
1156 enum lttcomm_return_code code
;
1157 struct lttng_poll_event events
;
1158 struct consumer_data
*consumer_data
= data
;
1159 struct consumer_socket
*cmd_socket_wrapper
= NULL
;
1161 DBG("[thread] Manage consumer started");
1163 rcu_register_thread();
1164 rcu_thread_online();
1166 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1168 health_code_update();
1171 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1172 * metadata_sock. Nothing more will be added to this poll set.
1174 ret
= sessiond_set_thread_pollset(&events
, 3);
1180 * The error socket here is already in a listening state which was done
1181 * just before spawning this thread to avoid a race between the consumer
1182 * daemon exec trying to connect and the listen() call.
1184 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1189 health_code_update();
1191 /* Infinite blocking call, waiting for transmission */
1193 health_poll_entry();
1195 if (testpoint(sessiond_thread_manage_consumer
)) {
1199 ret
= lttng_poll_wait(&events
, -1);
1203 * Restart interrupted system call.
1205 if (errno
== EINTR
) {
1213 for (i
= 0; i
< nb_fd
; i
++) {
1214 /* Fetch once the poll data */
1215 revents
= LTTNG_POLL_GETEV(&events
, i
);
1216 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1218 health_code_update();
1221 /* No activity for this FD (poll implementation). */
1225 /* Thread quit pipe has been closed. Killing thread. */
1226 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1232 /* Event on the registration socket */
1233 if (pollfd
== consumer_data
->err_sock
) {
1234 if (revents
& LPOLLIN
) {
1236 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1237 ERR("consumer err socket poll error");
1240 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1246 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1252 * Set the CLOEXEC flag. Return code is useless because either way, the
1255 (void) utils_set_fd_cloexec(sock
);
1257 health_code_update();
1259 DBG2("Receiving code from consumer err_sock");
1261 /* Getting status code from kconsumerd */
1262 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1263 sizeof(enum lttcomm_return_code
));
1268 health_code_update();
1269 if (code
!= LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1270 ERR("consumer error when waiting for SOCK_READY : %s",
1271 lttcomm_get_readable_code(-code
));
1275 /* Connect both command and metadata sockets. */
1276 consumer_data
->cmd_sock
=
1277 lttcomm_connect_unix_sock(
1278 consumer_data
->cmd_unix_sock_path
);
1279 consumer_data
->metadata_fd
=
1280 lttcomm_connect_unix_sock(
1281 consumer_data
->cmd_unix_sock_path
);
1282 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
1283 PERROR("consumer connect cmd socket");
1284 /* On error, signal condition and quit. */
1285 signal_consumer_condition(consumer_data
, -1);
1289 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1291 /* Create metadata socket lock. */
1292 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1293 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1294 PERROR("zmalloc pthread mutex");
1297 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1299 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1300 DBG("Consumer metadata socket ready (fd: %d)",
1301 consumer_data
->metadata_fd
);
1304 * Remove the consumerd error sock since we've established a connection.
1306 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1311 /* Add new accepted error socket. */
1312 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1317 /* Add metadata socket that is successfully connected. */
1318 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1319 LPOLLIN
| LPOLLRDHUP
);
1324 health_code_update();
1327 * Transfer the write-end of the channel monitoring and rotate pipe
1328 * to the consumer by issuing a SET_CHANNEL_MONITOR_PIPE and
1329 * SET_CHANNEL_ROTATE_PIPE commands.
1331 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
1332 if (!cmd_socket_wrapper
) {
1335 cmd_socket_wrapper
->lock
= &consumer_data
->lock
;
1337 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
1338 consumer_data
->channel_monitor_pipe
);
1343 ret
= consumer_send_channel_rotate_pipe(cmd_socket_wrapper
,
1344 consumer_data
->channel_rotate_pipe
);
1349 /* Discard the socket wrapper as it is no longer needed. */
1350 consumer_destroy_socket(cmd_socket_wrapper
);
1351 cmd_socket_wrapper
= NULL
;
1353 /* The thread is completely initialized, signal that it is ready. */
1354 signal_consumer_condition(consumer_data
, 1);
1356 /* Infinite blocking call, waiting for transmission */
1359 health_code_update();
1361 /* Exit the thread because the thread quit pipe has been triggered. */
1363 /* Not a health error. */
1368 health_poll_entry();
1369 ret
= lttng_poll_wait(&events
, -1);
1373 * Restart interrupted system call.
1375 if (errno
== EINTR
) {
1383 for (i
= 0; i
< nb_fd
; i
++) {
1384 /* Fetch once the poll data */
1385 revents
= LTTNG_POLL_GETEV(&events
, i
);
1386 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1388 health_code_update();
1391 /* No activity for this FD (poll implementation). */
1396 * Thread quit pipe has been triggered, flag that we should stop
1397 * but continue the current loop to handle potential data from
1400 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1402 if (pollfd
== sock
) {
1403 /* Event on the consumerd socket */
1404 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1405 && !(revents
& LPOLLIN
)) {
1406 ERR("consumer err socket second poll error");
1409 health_code_update();
1410 /* Wait for any kconsumerd error */
1411 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1412 sizeof(enum lttcomm_return_code
));
1414 ERR("consumer closed the command socket");
1418 ERR("consumer return code : %s",
1419 lttcomm_get_readable_code(-code
));
1422 } else if (pollfd
== consumer_data
->metadata_fd
) {
1423 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1424 && !(revents
& LPOLLIN
)) {
1425 ERR("consumer err metadata socket second poll error");
1428 /* UST metadata requests */
1429 ret
= ust_consumer_metadata_request(
1430 &consumer_data
->metadata_sock
);
1432 ERR("Handling metadata request");
1436 /* No need for an else branch all FDs are tested prior. */
1438 health_code_update();
1444 * We lock here because we are about to close the sockets and some other
1445 * thread might be using them so get exclusive access which will abort all
1446 * other consumer command by other threads.
1448 pthread_mutex_lock(&consumer_data
->lock
);
1450 /* Immediately set the consumerd state to stopped */
1451 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1452 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1453 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1454 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1455 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1457 /* Code flow error... */
1461 if (consumer_data
->err_sock
>= 0) {
1462 ret
= close(consumer_data
->err_sock
);
1466 consumer_data
->err_sock
= -1;
1468 if (consumer_data
->cmd_sock
>= 0) {
1469 ret
= close(consumer_data
->cmd_sock
);
1473 consumer_data
->cmd_sock
= -1;
1475 if (consumer_data
->metadata_sock
.fd_ptr
&&
1476 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1477 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1489 unlink(consumer_data
->err_unix_sock_path
);
1490 unlink(consumer_data
->cmd_unix_sock_path
);
1491 pthread_mutex_unlock(&consumer_data
->lock
);
1493 /* Cleanup metadata socket mutex. */
1494 if (consumer_data
->metadata_sock
.lock
) {
1495 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1496 free(consumer_data
->metadata_sock
.lock
);
1498 lttng_poll_clean(&events
);
1500 if (cmd_socket_wrapper
) {
1501 consumer_destroy_socket(cmd_socket_wrapper
);
1506 ERR("Health error occurred in %s", __func__
);
1508 health_unregister(health_sessiond
);
1509 DBG("consumer thread cleanup completed");
1511 rcu_thread_offline();
1512 rcu_unregister_thread();
1518 * This thread receives application command sockets (FDs) on the
1519 * apps_cmd_pipe and waits (polls) on them until they are closed
1520 * or an error occurs.
1522 * At that point, it flushes the data (tracing and metadata) associated
1523 * with this application and tears down ust app sessions and other
1524 * associated data structures through ust_app_unregister().
1526 * Note that this thread never sends commands to the applications
1527 * through the command sockets; it merely listens for hang-ups
1528 * and errors on those sockets and cleans-up as they occur.
1530 static void *thread_manage_apps(void *data
)
1532 int i
, ret
, pollfd
, err
= -1;
1534 uint32_t revents
, nb_fd
;
1535 struct lttng_poll_event events
;
1537 DBG("[thread] Manage application started");
1539 rcu_register_thread();
1540 rcu_thread_online();
1542 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1544 if (testpoint(sessiond_thread_manage_apps
)) {
1545 goto error_testpoint
;
1548 health_code_update();
1550 ret
= sessiond_set_thread_pollset(&events
, 2);
1552 goto error_poll_create
;
1555 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1560 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1564 health_code_update();
1567 DBG("Apps thread polling");
1569 /* Inifinite blocking call, waiting for transmission */
1571 health_poll_entry();
1572 ret
= lttng_poll_wait(&events
, -1);
1573 DBG("Apps thread return from poll on %d fds",
1574 LTTNG_POLL_GETNB(&events
));
1578 * Restart interrupted system call.
1580 if (errno
== EINTR
) {
1588 for (i
= 0; i
< nb_fd
; i
++) {
1589 /* Fetch once the poll data */
1590 revents
= LTTNG_POLL_GETEV(&events
, i
);
1591 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1593 health_code_update();
1596 /* No activity for this FD (poll implementation). */
1600 /* Thread quit pipe has been closed. Killing thread. */
1601 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1607 /* Inspect the apps cmd pipe */
1608 if (pollfd
== apps_cmd_pipe
[0]) {
1609 if (revents
& LPOLLIN
) {
1613 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1614 if (size_ret
< sizeof(sock
)) {
1615 PERROR("read apps cmd pipe");
1619 health_code_update();
1622 * Since this is a command socket (write then read),
1623 * we only monitor the error events of the socket.
1625 ret
= lttng_poll_add(&events
, sock
,
1626 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1631 DBG("Apps with sock %d added to poll set", sock
);
1632 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1633 ERR("Apps command pipe error");
1636 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1641 * At this point, we know that a registered application made
1642 * the event at poll_wait.
1644 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1645 /* Removing from the poll set */
1646 ret
= lttng_poll_del(&events
, pollfd
);
1651 /* Socket closed on remote end. */
1652 ust_app_unregister(pollfd
);
1654 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1659 health_code_update();
1665 lttng_poll_clean(&events
);
1668 utils_close_pipe(apps_cmd_pipe
);
1669 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1672 * We don't clean the UST app hash table here since already registered
1673 * applications can still be controlled so let them be until the session
1674 * daemon dies or the applications stop.
1679 ERR("Health error occurred in %s", __func__
);
1681 health_unregister(health_sessiond
);
1682 DBG("Application communication apps thread cleanup complete");
1683 rcu_thread_offline();
1684 rcu_unregister_thread();
1689 * Send a socket to a thread This is called from the dispatch UST registration
1690 * thread once all sockets are set for the application.
1692 * The sock value can be invalid, we don't really care, the thread will handle
1693 * it and make the necessary cleanup if so.
1695 * On success, return 0 else a negative value being the errno message of the
1698 static int send_socket_to_thread(int fd
, int sock
)
1703 * It's possible that the FD is set as invalid with -1 concurrently just
1704 * before calling this function being a shutdown state of the thread.
1711 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1712 if (ret
< sizeof(sock
)) {
1713 PERROR("write apps pipe %d", fd
);
1720 /* All good. Don't send back the write positive ret value. */
1727 * Sanitize the wait queue of the dispatch registration thread meaning removing
1728 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1729 * notify socket is never received.
1731 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1733 int ret
, nb_fd
= 0, i
;
1734 unsigned int fd_added
= 0;
1735 struct lttng_poll_event events
;
1736 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1740 lttng_poll_init(&events
);
1742 /* Just skip everything for an empty queue. */
1743 if (!wait_queue
->count
) {
1747 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1752 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1753 &wait_queue
->head
, head
) {
1754 assert(wait_node
->app
);
1755 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1756 LPOLLHUP
| LPOLLERR
);
1769 * Poll but don't block so we can quickly identify the faulty events and
1770 * clean them afterwards from the wait queue.
1772 ret
= lttng_poll_wait(&events
, 0);
1778 for (i
= 0; i
< nb_fd
; i
++) {
1779 /* Get faulty FD. */
1780 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1781 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1784 /* No activity for this FD (poll implementation). */
1788 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1789 &wait_queue
->head
, head
) {
1790 if (pollfd
== wait_node
->app
->sock
&&
1791 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1792 cds_list_del(&wait_node
->head
);
1793 wait_queue
->count
--;
1794 ust_app_destroy(wait_node
->app
);
1797 * Silence warning of use-after-free in
1798 * cds_list_for_each_entry_safe which uses
1799 * __typeof__(*wait_node).
1804 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1811 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1815 lttng_poll_clean(&events
);
1819 lttng_poll_clean(&events
);
1821 ERR("Unable to sanitize wait queue");
1826 * Dispatch request from the registration threads to the application
1827 * communication thread.
1829 static void *thread_dispatch_ust_registration(void *data
)
1832 struct cds_wfcq_node
*node
;
1833 struct ust_command
*ust_cmd
= NULL
;
1834 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1835 struct ust_reg_wait_queue wait_queue
= {
1839 rcu_register_thread();
1841 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1843 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1844 goto error_testpoint
;
1847 health_code_update();
1849 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1851 DBG("[thread] Dispatch UST command started");
1854 health_code_update();
1856 /* Atomically prepare the queue futex */
1857 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1859 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1864 struct ust_app
*app
= NULL
;
1868 * Make sure we don't have node(s) that have hung up before receiving
1869 * the notify socket. This is to clean the list in order to avoid
1870 * memory leaks from notify socket that are never seen.
1872 sanitize_wait_queue(&wait_queue
);
1874 health_code_update();
1875 /* Dequeue command for registration */
1876 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1878 DBG("Woken up but nothing in the UST command queue");
1879 /* Continue thread execution */
1883 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1885 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1886 " gid:%d sock:%d name:%s (version %d.%d)",
1887 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1888 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1889 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1890 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1892 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1893 wait_node
= zmalloc(sizeof(*wait_node
));
1895 PERROR("zmalloc wait_node dispatch");
1896 ret
= close(ust_cmd
->sock
);
1898 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1900 lttng_fd_put(LTTNG_FD_APPS
, 1);
1904 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1906 /* Create application object if socket is CMD. */
1907 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1909 if (!wait_node
->app
) {
1910 ret
= close(ust_cmd
->sock
);
1912 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1914 lttng_fd_put(LTTNG_FD_APPS
, 1);
1920 * Add application to the wait queue so we can set the notify
1921 * socket before putting this object in the global ht.
1923 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1928 * We have to continue here since we don't have the notify
1929 * socket and the application MUST be added to the hash table
1930 * only at that moment.
1935 * Look for the application in the local wait queue and set the
1936 * notify socket if found.
1938 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1939 &wait_queue
.head
, head
) {
1940 health_code_update();
1941 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1942 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1943 cds_list_del(&wait_node
->head
);
1945 app
= wait_node
->app
;
1947 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1953 * With no application at this stage the received socket is
1954 * basically useless so close it before we free the cmd data
1955 * structure for good.
1958 ret
= close(ust_cmd
->sock
);
1960 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1962 lttng_fd_put(LTTNG_FD_APPS
, 1);
1969 * @session_lock_list
1971 * Lock the global session list so from the register up to the
1972 * registration done message, no thread can see the application
1973 * and change its state.
1975 session_lock_list();
1979 * Add application to the global hash table. This needs to be
1980 * done before the update to the UST registry can locate the
1985 /* Set app version. This call will print an error if needed. */
1986 (void) ust_app_version(app
);
1988 /* Send notify socket through the notify pipe. */
1989 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1993 session_unlock_list();
1995 * No notify thread, stop the UST tracing. However, this is
1996 * not an internal error of the this thread thus setting
1997 * the health error code to a normal exit.
2004 * Update newly registered application with the tracing
2005 * registry info already enabled information.
2007 update_ust_app(app
->sock
);
2010 * Don't care about return value. Let the manage apps threads
2011 * handle app unregistration upon socket close.
2013 (void) ust_app_register_done(app
);
2016 * Even if the application socket has been closed, send the app
2017 * to the thread and unregistration will take place at that
2020 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2023 session_unlock_list();
2025 * No apps. thread, stop the UST tracing. However, this is
2026 * not an internal error of the this thread thus setting
2027 * the health error code to a normal exit.
2034 session_unlock_list();
2036 } while (node
!= NULL
);
2038 health_poll_entry();
2039 /* Futex wait on queue. Blocking call on futex() */
2040 futex_nto1_wait(&ust_cmd_queue
.futex
);
2043 /* Normal exit, no error */
2047 /* Clean up wait queue. */
2048 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2049 &wait_queue
.head
, head
) {
2050 cds_list_del(&wait_node
->head
);
2055 /* Empty command queue. */
2057 /* Dequeue command for registration */
2058 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2062 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2063 ret
= close(ust_cmd
->sock
);
2065 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2067 lttng_fd_put(LTTNG_FD_APPS
, 1);
2072 DBG("Dispatch thread dying");
2075 ERR("Health error occurred in %s", __func__
);
2077 health_unregister(health_sessiond
);
2078 rcu_unregister_thread();
2083 * This thread manage application registration.
2085 static void *thread_registration_apps(void *data
)
2087 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2088 uint32_t revents
, nb_fd
;
2089 struct lttng_poll_event events
;
2091 * Get allocated in this thread, enqueued to a global queue, dequeued and
2092 * freed in the manage apps thread.
2094 struct ust_command
*ust_cmd
= NULL
;
2096 DBG("[thread] Manage application registration started");
2098 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2100 if (testpoint(sessiond_thread_registration_apps
)) {
2101 goto error_testpoint
;
2104 ret
= lttcomm_listen_unix_sock(apps_sock
);
2110 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2111 * more will be added to this poll set.
2113 ret
= sessiond_set_thread_pollset(&events
, 2);
2115 goto error_create_poll
;
2118 /* Add the application registration socket */
2119 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2121 goto error_poll_add
;
2124 /* Notify all applications to register */
2125 ret
= notify_ust_apps(1);
2127 ERR("Failed to notify applications or create the wait shared memory.\n"
2128 "Execution continues but there might be problem for already\n"
2129 "running applications that wishes to register.");
2133 DBG("Accepting application registration");
2135 /* Inifinite blocking call, waiting for transmission */
2137 health_poll_entry();
2138 ret
= lttng_poll_wait(&events
, -1);
2142 * Restart interrupted system call.
2144 if (errno
== EINTR
) {
2152 for (i
= 0; i
< nb_fd
; i
++) {
2153 health_code_update();
2155 /* Fetch once the poll data */
2156 revents
= LTTNG_POLL_GETEV(&events
, i
);
2157 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2160 /* No activity for this FD (poll implementation). */
2164 /* Thread quit pipe has been closed. Killing thread. */
2165 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2171 /* Event on the registration socket */
2172 if (pollfd
== apps_sock
) {
2173 if (revents
& LPOLLIN
) {
2174 sock
= lttcomm_accept_unix_sock(apps_sock
);
2180 * Set socket timeout for both receiving and ending.
2181 * app_socket_timeout is in seconds, whereas
2182 * lttcomm_setsockopt_rcv_timeout and
2183 * lttcomm_setsockopt_snd_timeout expect msec as
2186 if (config
.app_socket_timeout
>= 0) {
2187 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2188 config
.app_socket_timeout
* 1000);
2189 (void) lttcomm_setsockopt_snd_timeout(sock
,
2190 config
.app_socket_timeout
* 1000);
2194 * Set the CLOEXEC flag. Return code is useless because
2195 * either way, the show must go on.
2197 (void) utils_set_fd_cloexec(sock
);
2199 /* Create UST registration command for enqueuing */
2200 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2201 if (ust_cmd
== NULL
) {
2202 PERROR("ust command zmalloc");
2211 * Using message-based transmissions to ensure we don't
2212 * have to deal with partially received messages.
2214 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2216 ERR("Exhausted file descriptors allowed for applications.");
2226 health_code_update();
2227 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2230 /* Close socket of the application. */
2235 lttng_fd_put(LTTNG_FD_APPS
, 1);
2239 health_code_update();
2241 ust_cmd
->sock
= sock
;
2244 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2245 " gid:%d sock:%d name:%s (version %d.%d)",
2246 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2247 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2248 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2249 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2252 * Lock free enqueue the registration request. The red pill
2253 * has been taken! This apps will be part of the *system*.
2255 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2258 * Wake the registration queue futex. Implicit memory
2259 * barrier with the exchange in cds_wfcq_enqueue.
2261 futex_nto1_wake(&ust_cmd_queue
.futex
);
2262 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2263 ERR("Register apps socket poll error");
2266 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2275 /* Notify that the registration thread is gone */
2278 if (apps_sock
>= 0) {
2279 ret
= close(apps_sock
);
2289 lttng_fd_put(LTTNG_FD_APPS
, 1);
2291 unlink(config
.apps_unix_sock_path
.value
);
2294 lttng_poll_clean(&events
);
2298 DBG("UST Registration thread cleanup complete");
2301 ERR("Health error occurred in %s", __func__
);
2303 health_unregister(health_sessiond
);
2309 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2310 * exec or it will fails.
2312 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2315 struct timespec timeout
;
2318 * Make sure we set the readiness flag to 0 because we are NOT ready.
2319 * This access to consumer_thread_is_ready does not need to be
2320 * protected by consumer_data.cond_mutex (yet) since the consumer
2321 * management thread has not been started at this point.
2323 consumer_data
->consumer_thread_is_ready
= 0;
2325 /* Setup pthread condition */
2326 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2329 PERROR("pthread_condattr_init consumer data");
2334 * Set the monotonic clock in order to make sure we DO NOT jump in time
2335 * between the clock_gettime() call and the timedwait call. See bug #324
2336 * for a more details and how we noticed it.
2338 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2341 PERROR("pthread_condattr_setclock consumer data");
2345 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2348 PERROR("pthread_cond_init consumer data");
2352 ret
= pthread_create(&consumer_data
->thread
, default_pthread_attr(),
2353 thread_manage_consumer
, consumer_data
);
2356 PERROR("pthread_create consumer");
2361 /* We are about to wait on a pthread condition */
2362 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2364 /* Get time for sem_timedwait absolute timeout */
2365 clock_ret
= lttng_clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2367 * Set the timeout for the condition timed wait even if the clock gettime
2368 * call fails since we might loop on that call and we want to avoid to
2369 * increment the timeout too many times.
2371 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2374 * The following loop COULD be skipped in some conditions so this is why we
2375 * set ret to 0 in order to make sure at least one round of the loop is
2381 * Loop until the condition is reached or when a timeout is reached. Note
2382 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2383 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2384 * possible. This loop does not take any chances and works with both of
2387 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2388 if (clock_ret
< 0) {
2389 PERROR("clock_gettime spawn consumer");
2390 /* Infinite wait for the consumerd thread to be ready */
2391 ret
= pthread_cond_wait(&consumer_data
->cond
,
2392 &consumer_data
->cond_mutex
);
2394 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2395 &consumer_data
->cond_mutex
, &timeout
);
2399 /* Release the pthread condition */
2400 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2404 if (ret
== ETIMEDOUT
) {
2408 * Call has timed out so we kill the kconsumerd_thread and return
2411 ERR("Condition timed out. The consumer thread was never ready."
2413 pth_ret
= pthread_cancel(consumer_data
->thread
);
2415 PERROR("pthread_cancel consumer thread");
2418 PERROR("pthread_cond_wait failed consumer thread");
2420 /* Caller is expecting a negative value on failure. */
2425 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2426 if (consumer_data
->pid
== 0) {
2427 ERR("Consumerd did not start");
2428 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2431 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2440 * Join consumer thread
2442 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2446 /* Consumer pid must be a real one. */
2447 if (consumer_data
->pid
> 0) {
2449 ret
= kill(consumer_data
->pid
, SIGTERM
);
2451 PERROR("Error killing consumer daemon");
2454 return pthread_join(consumer_data
->thread
, &status
);
2461 * Fork and exec a consumer daemon (consumerd).
2463 * Return pid if successful else -1.
2465 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2469 const char *consumer_to_use
;
2470 const char *verbosity
;
2473 DBG("Spawning consumerd");
2480 if (config
.verbose_consumer
) {
2481 verbosity
= "--verbose";
2482 } else if (lttng_opt_quiet
) {
2483 verbosity
= "--quiet";
2488 switch (consumer_data
->type
) {
2489 case LTTNG_CONSUMER_KERNEL
:
2491 * Find out which consumerd to execute. We will first try the
2492 * 64-bit path, then the sessiond's installation directory, and
2493 * fallback on the 32-bit one,
2495 DBG3("Looking for a kernel consumer at these locations:");
2496 DBG3(" 1) %s", config
.consumerd64_bin_path
.value
? : "NULL");
2497 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, DEFAULT_CONSUMERD_FILE
);
2498 DBG3(" 3) %s", config
.consumerd32_bin_path
.value
? : "NULL");
2499 if (stat(config
.consumerd64_bin_path
.value
, &st
) == 0) {
2500 DBG3("Found location #1");
2501 consumer_to_use
= config
.consumerd64_bin_path
.value
;
2502 } else if (stat(INSTALL_BIN_PATH
"/" DEFAULT_CONSUMERD_FILE
, &st
) == 0) {
2503 DBG3("Found location #2");
2504 consumer_to_use
= INSTALL_BIN_PATH
"/" DEFAULT_CONSUMERD_FILE
;
2505 } else if (config
.consumerd32_bin_path
.value
&&
2506 stat(config
.consumerd32_bin_path
.value
, &st
) == 0) {
2507 DBG3("Found location #3");
2508 consumer_to_use
= config
.consumerd32_bin_path
.value
;
2510 DBG("Could not find any valid consumerd executable");
2514 DBG("Using kernel consumer at: %s", consumer_to_use
);
2515 (void) execl(consumer_to_use
,
2516 "lttng-consumerd", verbosity
, "-k",
2517 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2518 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2519 "--group", config
.tracing_group_name
.value
,
2522 case LTTNG_CONSUMER64_UST
:
2524 if (config
.consumerd64_lib_dir
.value
) {
2529 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2533 tmplen
= strlen(config
.consumerd64_lib_dir
.value
) + 1 /* : */ + strlen(tmp
);
2534 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2539 strcat(tmpnew
, config
.consumerd64_lib_dir
.value
);
2540 if (tmp
[0] != '\0') {
2541 strcat(tmpnew
, ":");
2542 strcat(tmpnew
, tmp
);
2544 ret
= setenv("LD_LIBRARY_PATH", tmpnew
, 1);
2551 DBG("Using 64-bit UST consumer at: %s", config
.consumerd64_bin_path
.value
);
2552 (void) execl(config
.consumerd64_bin_path
.value
, "lttng-consumerd", verbosity
, "-u",
2553 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2554 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2555 "--group", config
.tracing_group_name
.value
,
2559 case LTTNG_CONSUMER32_UST
:
2561 if (config
.consumerd32_lib_dir
.value
) {
2566 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2570 tmplen
= strlen(config
.consumerd32_lib_dir
.value
) + 1 /* : */ + strlen(tmp
);
2571 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2576 strcat(tmpnew
, config
.consumerd32_lib_dir
.value
);
2577 if (tmp
[0] != '\0') {
2578 strcat(tmpnew
, ":");
2579 strcat(tmpnew
, tmp
);
2581 ret
= setenv("LD_LIBRARY_PATH", tmpnew
, 1);
2588 DBG("Using 32-bit UST consumer at: %s", config
.consumerd32_bin_path
.value
);
2589 (void) execl(config
.consumerd32_bin_path
.value
, "lttng-consumerd", verbosity
, "-u",
2590 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2591 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2592 "--group", config
.tracing_group_name
.value
,
2597 ERR("unknown consumer type");
2601 PERROR("Consumer execl()");
2603 /* Reaching this point, we got a failure on our execl(). */
2605 } else if (pid
> 0) {
2608 PERROR("start consumer fork");
2616 * Spawn the consumerd daemon and session daemon thread.
2618 static int start_consumerd(struct consumer_data
*consumer_data
)
2623 * Set the listen() state on the socket since there is a possible race
2624 * between the exec() of the consumer daemon and this call if place in the
2625 * consumer thread. See bug #366 for more details.
2627 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2632 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2633 if (consumer_data
->pid
!= 0) {
2634 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2638 ret
= spawn_consumerd(consumer_data
);
2640 ERR("Spawning consumerd failed");
2641 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2645 /* Setting up the consumer_data pid */
2646 consumer_data
->pid
= ret
;
2647 DBG2("Consumer pid %d", consumer_data
->pid
);
2648 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2650 DBG2("Spawning consumer control thread");
2651 ret
= spawn_consumer_thread(consumer_data
);
2653 ERR("Fatal error spawning consumer control thread");
2661 /* Cleanup already created sockets on error. */
2662 if (consumer_data
->err_sock
>= 0) {
2665 err
= close(consumer_data
->err_sock
);
2667 PERROR("close consumer data error socket");
2674 * Setup necessary data for kernel tracer action.
2676 static int init_kernel_tracer(void)
2680 /* Modprobe lttng kernel modules */
2681 ret
= modprobe_lttng_control();
2686 /* Open debugfs lttng */
2687 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2688 if (kernel_tracer_fd
< 0) {
2689 DBG("Failed to open %s", module_proc_lttng
);
2693 /* Validate kernel version */
2694 ret
= kernel_validate_version(kernel_tracer_fd
, &kernel_tracer_version
,
2695 &kernel_tracer_abi_version
);
2700 ret
= modprobe_lttng_data();
2705 ret
= kernel_supports_ring_buffer_snapshot_sample_positions(
2712 WARN("Kernel tracer does not support buffer monitoring. "
2713 "The monitoring timer of channels in the kernel domain "
2714 "will be set to 0 (disabled).");
2717 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2721 modprobe_remove_lttng_control();
2722 ret
= close(kernel_tracer_fd
);
2726 kernel_tracer_fd
= -1;
2727 return LTTNG_ERR_KERN_VERSION
;
2730 ret
= close(kernel_tracer_fd
);
2736 modprobe_remove_lttng_control();
2739 WARN("No kernel tracer available");
2740 kernel_tracer_fd
= -1;
2742 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2744 return LTTNG_ERR_KERN_NA
;
2750 * Copy consumer output from the tracing session to the domain session. The
2751 * function also applies the right modification on a per domain basis for the
2752 * trace files destination directory.
2754 * Should *NOT* be called with RCU read-side lock held.
2756 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2759 const char *dir_name
;
2760 struct consumer_output
*consumer
;
2763 assert(session
->consumer
);
2766 case LTTNG_DOMAIN_KERNEL
:
2767 DBG3("Copying tracing session consumer output in kernel session");
2769 * XXX: We should audit the session creation and what this function
2770 * does "extra" in order to avoid a destroy since this function is used
2771 * in the domain session creation (kernel and ust) only. Same for UST
2774 if (session
->kernel_session
->consumer
) {
2775 consumer_output_put(session
->kernel_session
->consumer
);
2777 session
->kernel_session
->consumer
=
2778 consumer_copy_output(session
->consumer
);
2779 /* Ease our life a bit for the next part */
2780 consumer
= session
->kernel_session
->consumer
;
2781 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2783 case LTTNG_DOMAIN_JUL
:
2784 case LTTNG_DOMAIN_LOG4J
:
2785 case LTTNG_DOMAIN_PYTHON
:
2786 case LTTNG_DOMAIN_UST
:
2787 DBG3("Copying tracing session consumer output in UST session");
2788 if (session
->ust_session
->consumer
) {
2789 consumer_output_put(session
->ust_session
->consumer
);
2791 session
->ust_session
->consumer
=
2792 consumer_copy_output(session
->consumer
);
2793 /* Ease our life a bit for the next part */
2794 consumer
= session
->ust_session
->consumer
;
2795 dir_name
= DEFAULT_UST_TRACE_DIR
;
2798 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2802 /* Append correct directory to subdir */
2803 strncat(consumer
->subdir
, dir_name
,
2804 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2805 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2814 * Create an UST session and add it to the session ust list.
2816 * Should *NOT* be called with RCU read-side lock held.
2818 static int create_ust_session(struct ltt_session
*session
,
2819 struct lttng_domain
*domain
)
2822 struct ltt_ust_session
*lus
= NULL
;
2826 assert(session
->consumer
);
2828 switch (domain
->type
) {
2829 case LTTNG_DOMAIN_JUL
:
2830 case LTTNG_DOMAIN_LOG4J
:
2831 case LTTNG_DOMAIN_PYTHON
:
2832 case LTTNG_DOMAIN_UST
:
2835 ERR("Unknown UST domain on create session %d", domain
->type
);
2836 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2840 DBG("Creating UST session");
2842 lus
= trace_ust_create_session(session
->id
);
2844 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2848 lus
->uid
= session
->uid
;
2849 lus
->gid
= session
->gid
;
2850 lus
->output_traces
= session
->output_traces
;
2851 lus
->snapshot_mode
= session
->snapshot_mode
;
2852 lus
->live_timer_interval
= session
->live_timer
;
2853 session
->ust_session
= lus
;
2854 if (session
->shm_path
[0]) {
2855 strncpy(lus
->root_shm_path
, session
->shm_path
,
2856 sizeof(lus
->root_shm_path
));
2857 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2858 strncpy(lus
->shm_path
, session
->shm_path
,
2859 sizeof(lus
->shm_path
));
2860 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2861 strncat(lus
->shm_path
, "/ust",
2862 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2864 /* Copy session output to the newly created UST session */
2865 ret
= copy_session_consumer(domain
->type
, session
);
2866 if (ret
!= LTTNG_OK
) {
2874 session
->ust_session
= NULL
;
2879 * Create a kernel tracer session then create the default channel.
2881 static int create_kernel_session(struct ltt_session
*session
)
2885 DBG("Creating kernel session");
2887 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2889 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2893 /* Code flow safety */
2894 assert(session
->kernel_session
);
2896 /* Copy session output to the newly created Kernel session */
2897 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2898 if (ret
!= LTTNG_OK
) {
2902 session
->kernel_session
->uid
= session
->uid
;
2903 session
->kernel_session
->gid
= session
->gid
;
2904 session
->kernel_session
->output_traces
= session
->output_traces
;
2905 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2910 trace_kernel_destroy_session(session
->kernel_session
);
2911 session
->kernel_session
= NULL
;
2916 * Count number of session permitted by uid/gid.
2918 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2921 struct ltt_session
*session
;
2923 DBG("Counting number of available session for UID %d GID %d",
2925 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2927 * Only list the sessions the user can control.
2929 if (!session_access_ok(session
, uid
, gid
)) {
2937 static int receive_userspace_probe(struct command_ctx
*cmd_ctx
, int sock
,
2938 int *sock_error
, struct lttng_event
*event
)
2941 struct lttng_userspace_probe_location
*probe_location
;
2942 const struct lttng_userspace_probe_location_lookup_method
*lookup
= NULL
;
2943 struct lttng_dynamic_buffer probe_location_buffer
;
2944 struct lttng_buffer_view buffer_view
;
2947 * Create a buffer to store the serialized version of the probe
2950 lttng_dynamic_buffer_init(&probe_location_buffer
);
2951 ret
= lttng_dynamic_buffer_set_size(&probe_location_buffer
,
2952 cmd_ctx
->lsm
->u
.enable
.userspace_probe_location_len
);
2954 ret
= LTTNG_ERR_NOMEM
;
2959 * Receive the probe location.
2961 ret
= lttcomm_recv_unix_sock(sock
, probe_location_buffer
.data
,
2962 probe_location_buffer
.size
);
2964 DBG("Nothing recv() from client var len data... continuing");
2966 lttng_dynamic_buffer_reset(&probe_location_buffer
);
2967 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
2971 buffer_view
= lttng_buffer_view_from_dynamic_buffer(
2972 &probe_location_buffer
, 0, probe_location_buffer
.size
);
2975 * Extract the probe location from the serialized version.
2977 ret
= lttng_userspace_probe_location_create_from_buffer(
2978 &buffer_view
, &probe_location
);
2980 WARN("Failed to create a userspace probe location from the received buffer");
2981 lttng_dynamic_buffer_reset( &probe_location_buffer
);
2982 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
2987 * Receive the file descriptor to the target binary from the client.
2989 DBG("Receiving userspace probe target FD from client ...");
2990 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
2992 DBG("Nothing recv() from client userspace probe fd... continuing");
2994 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
2999 * Set the file descriptor received from the client through the unix
3000 * socket in the probe location.
3002 lookup
= lttng_userspace_probe_location_get_lookup_method(probe_location
);
3004 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
3009 * From the kernel tracer's perspective, all userspace probe event types
3010 * are all the same: a file and an offset.
3012 switch (lttng_userspace_probe_location_lookup_method_get_type(lookup
)) {
3013 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF
:
3014 ret
= lttng_userspace_probe_location_function_set_binary_fd(
3015 probe_location
, fd
);
3017 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT
:
3018 ret
= lttng_userspace_probe_location_tracepoint_set_binary_fd(
3019 probe_location
, fd
);
3022 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
3027 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
3031 /* Attach the probe location to the event. */
3032 ret
= lttng_event_set_userspace_probe_location(event
, probe_location
);
3034 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
3038 lttng_dynamic_buffer_reset(&probe_location_buffer
);
3044 * Check if the current kernel tracer supports the session rotation feature.
3045 * Return 1 if it does, 0 otherwise.
3047 static int check_rotate_compatible(void)
3051 if (kernel_tracer_version
.major
!= 2 || kernel_tracer_version
.minor
< 11) {
3052 DBG("Kernel tracer version is not compatible with the rotation feature");
3060 * Process the command requested by the lttng client within the command
3061 * context structure. This function make sure that the return structure (llm)
3062 * is set and ready for transmission before returning.
3064 * Return any error encountered or 0 for success.
3066 * "sock" is only used for special-case var. len data.
3068 * Should *NOT* be called with RCU read-side lock held.
3070 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
3074 int need_tracing_session
= 1;
3077 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
3079 assert(!rcu_read_ongoing());
3083 switch (cmd_ctx
->lsm
->cmd_type
) {
3084 case LTTNG_CREATE_SESSION
:
3085 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3086 case LTTNG_CREATE_SESSION_LIVE
:
3087 case LTTNG_DESTROY_SESSION
:
3088 case LTTNG_LIST_SESSIONS
:
3089 case LTTNG_LIST_DOMAINS
:
3090 case LTTNG_START_TRACE
:
3091 case LTTNG_STOP_TRACE
:
3092 case LTTNG_DATA_PENDING
:
3093 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3094 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3095 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3096 case LTTNG_SNAPSHOT_RECORD
:
3097 case LTTNG_SAVE_SESSION
:
3098 case LTTNG_SET_SESSION_SHM_PATH
:
3099 case LTTNG_REGENERATE_METADATA
:
3100 case LTTNG_REGENERATE_STATEDUMP
:
3101 case LTTNG_REGISTER_TRIGGER
:
3102 case LTTNG_UNREGISTER_TRIGGER
:
3103 case LTTNG_ROTATE_SESSION
:
3104 case LTTNG_ROTATION_GET_INFO
:
3105 case LTTNG_SESSION_GET_CURRENT_OUTPUT
:
3106 case LTTNG_ROTATION_SET_SCHEDULE
:
3107 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES
:
3114 if (config
.no_kernel
&& need_domain
3115 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3117 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3119 ret
= LTTNG_ERR_KERN_NA
;
3124 /* Deny register consumer if we already have a spawned consumer. */
3125 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3126 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3127 if (kconsumer_data
.pid
> 0) {
3128 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3129 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3132 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3136 * Check for command that don't needs to allocate a returned payload. We do
3137 * this here so we don't have to make the call for no payload at each
3140 switch(cmd_ctx
->lsm
->cmd_type
) {
3141 case LTTNG_LIST_SESSIONS
:
3142 case LTTNG_LIST_TRACEPOINTS
:
3143 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3144 case LTTNG_LIST_DOMAINS
:
3145 case LTTNG_LIST_CHANNELS
:
3146 case LTTNG_LIST_EVENTS
:
3147 case LTTNG_LIST_SYSCALLS
:
3148 case LTTNG_LIST_TRACKER_PIDS
:
3149 case LTTNG_DATA_PENDING
:
3150 case LTTNG_ROTATE_SESSION
:
3151 case LTTNG_ROTATION_GET_INFO
:
3152 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES
:
3155 /* Setup lttng message with no payload */
3156 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3158 /* This label does not try to unlock the session */
3159 goto init_setup_error
;
3163 /* Commands that DO NOT need a session. */
3164 switch (cmd_ctx
->lsm
->cmd_type
) {
3165 case LTTNG_CREATE_SESSION
:
3166 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3167 case LTTNG_CREATE_SESSION_LIVE
:
3168 case LTTNG_LIST_SESSIONS
:
3169 case LTTNG_LIST_TRACEPOINTS
:
3170 case LTTNG_LIST_SYSCALLS
:
3171 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3172 case LTTNG_SAVE_SESSION
:
3173 case LTTNG_REGISTER_TRIGGER
:
3174 case LTTNG_UNREGISTER_TRIGGER
:
3175 need_tracing_session
= 0;
3178 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3180 * We keep the session list lock across _all_ commands
3181 * for now, because the per-session lock does not
3182 * handle teardown properly.
3184 session_lock_list();
3185 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3186 if (cmd_ctx
->session
== NULL
) {
3187 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3190 /* Acquire lock for the session */
3191 session_lock(cmd_ctx
->session
);
3197 * Commands that need a valid session but should NOT create one if none
3198 * exists. Instead of creating one and destroying it when the command is
3199 * handled, process that right before so we save some round trip in useless
3202 switch (cmd_ctx
->lsm
->cmd_type
) {
3203 case LTTNG_DISABLE_CHANNEL
:
3204 case LTTNG_DISABLE_EVENT
:
3205 switch (cmd_ctx
->lsm
->domain
.type
) {
3206 case LTTNG_DOMAIN_KERNEL
:
3207 if (!cmd_ctx
->session
->kernel_session
) {
3208 ret
= LTTNG_ERR_NO_CHANNEL
;
3212 case LTTNG_DOMAIN_JUL
:
3213 case LTTNG_DOMAIN_LOG4J
:
3214 case LTTNG_DOMAIN_PYTHON
:
3215 case LTTNG_DOMAIN_UST
:
3216 if (!cmd_ctx
->session
->ust_session
) {
3217 ret
= LTTNG_ERR_NO_CHANNEL
;
3222 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3234 * Check domain type for specific "pre-action".
3236 switch (cmd_ctx
->lsm
->domain
.type
) {
3237 case LTTNG_DOMAIN_KERNEL
:
3239 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3243 /* Kernel tracer check */
3244 if (kernel_tracer_fd
== -1) {
3245 /* Basically, load kernel tracer modules */
3246 ret
= init_kernel_tracer();
3252 /* Consumer is in an ERROR state. Report back to client */
3253 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3254 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3258 /* Need a session for kernel command */
3259 if (need_tracing_session
) {
3260 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3261 ret
= create_kernel_session(cmd_ctx
->session
);
3263 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3268 /* Start the kernel consumer daemon */
3269 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3270 if (kconsumer_data
.pid
== 0 &&
3271 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3272 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3273 ret
= start_consumerd(&kconsumer_data
);
3275 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3278 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3280 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3284 * The consumer was just spawned so we need to add the socket to
3285 * the consumer output of the session if exist.
3287 ret
= consumer_create_socket(&kconsumer_data
,
3288 cmd_ctx
->session
->kernel_session
->consumer
);
3295 case LTTNG_DOMAIN_JUL
:
3296 case LTTNG_DOMAIN_LOG4J
:
3297 case LTTNG_DOMAIN_PYTHON
:
3298 case LTTNG_DOMAIN_UST
:
3300 if (!ust_app_supported()) {
3301 ret
= LTTNG_ERR_NO_UST
;
3304 /* Consumer is in an ERROR state. Report back to client */
3305 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3306 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3310 if (need_tracing_session
) {
3311 /* Create UST session if none exist. */
3312 if (cmd_ctx
->session
->ust_session
== NULL
) {
3313 ret
= create_ust_session(cmd_ctx
->session
,
3314 &cmd_ctx
->lsm
->domain
);
3315 if (ret
!= LTTNG_OK
) {
3320 /* Start the UST consumer daemons */
3322 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3323 if (config
.consumerd64_bin_path
.value
&&
3324 ustconsumer64_data
.pid
== 0 &&
3325 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3326 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3327 ret
= start_consumerd(&ustconsumer64_data
);
3329 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3330 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3334 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3335 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3337 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3341 * Setup socket for consumer 64 bit. No need for atomic access
3342 * since it was set above and can ONLY be set in this thread.
3344 ret
= consumer_create_socket(&ustconsumer64_data
,
3345 cmd_ctx
->session
->ust_session
->consumer
);
3351 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3352 if (config
.consumerd32_bin_path
.value
&&
3353 ustconsumer32_data
.pid
== 0 &&
3354 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3355 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3356 ret
= start_consumerd(&ustconsumer32_data
);
3358 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3359 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3363 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3364 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3366 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3370 * Setup socket for consumer 64 bit. No need for atomic access
3371 * since it was set above and can ONLY be set in this thread.
3373 ret
= consumer_create_socket(&ustconsumer32_data
,
3374 cmd_ctx
->session
->ust_session
->consumer
);
3386 /* Validate consumer daemon state when start/stop trace command */
3387 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3388 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3389 switch (cmd_ctx
->lsm
->domain
.type
) {
3390 case LTTNG_DOMAIN_NONE
:
3392 case LTTNG_DOMAIN_JUL
:
3393 case LTTNG_DOMAIN_LOG4J
:
3394 case LTTNG_DOMAIN_PYTHON
:
3395 case LTTNG_DOMAIN_UST
:
3396 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3397 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3401 case LTTNG_DOMAIN_KERNEL
:
3402 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3403 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3408 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3414 * Check that the UID or GID match that of the tracing session.
3415 * The root user can interact with all sessions.
3417 if (need_tracing_session
) {
3418 if (!session_access_ok(cmd_ctx
->session
,
3419 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3420 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3421 ret
= LTTNG_ERR_EPERM
;
3427 * Send relayd information to consumer as soon as we have a domain and a
3430 if (cmd_ctx
->session
&& need_domain
) {
3432 * Setup relayd if not done yet. If the relayd information was already
3433 * sent to the consumer, this call will gracefully return.
3435 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3436 if (ret
!= LTTNG_OK
) {
3441 /* Process by command type */
3442 switch (cmd_ctx
->lsm
->cmd_type
) {
3443 case LTTNG_ADD_CONTEXT
:
3446 * An LTTNG_ADD_CONTEXT command might have a supplementary
3447 * payload if the context being added is an application context.
3449 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3450 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3451 char *provider_name
= NULL
, *context_name
= NULL
;
3452 size_t provider_name_len
=
3453 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3454 size_t context_name_len
=
3455 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3457 if (provider_name_len
== 0 || context_name_len
== 0) {
3459 * Application provider and context names MUST
3462 ret
= -LTTNG_ERR_INVALID
;
3466 provider_name
= zmalloc(provider_name_len
+ 1);
3467 if (!provider_name
) {
3468 ret
= -LTTNG_ERR_NOMEM
;
3471 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3474 context_name
= zmalloc(context_name_len
+ 1);
3475 if (!context_name
) {
3476 ret
= -LTTNG_ERR_NOMEM
;
3477 goto error_add_context
;
3479 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3482 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3485 goto error_add_context
;
3488 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3491 goto error_add_context
;
3496 * cmd_add_context assumes ownership of the provider and context
3499 ret
= cmd_add_context(cmd_ctx
->session
,
3500 cmd_ctx
->lsm
->domain
.type
,
3501 cmd_ctx
->lsm
->u
.context
.channel_name
,
3502 &cmd_ctx
->lsm
->u
.context
.ctx
,
3503 kernel_poll_pipe
[1]);
3505 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3506 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3508 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3509 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3515 case LTTNG_DISABLE_CHANNEL
:
3517 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3518 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3521 case LTTNG_DISABLE_EVENT
:
3525 * FIXME: handle filter; for now we just receive the filter's
3526 * bytecode along with the filter expression which are sent by
3527 * liblttng-ctl and discard them.
3529 * This fixes an issue where the client may block while sending
3530 * the filter payload and encounter an error because the session
3531 * daemon closes the socket without ever handling this data.
3533 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3534 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3537 char data
[LTTNG_FILTER_MAX_LEN
];
3539 DBG("Discarding disable event command payload of size %zu", count
);
3541 ret
= lttcomm_recv_unix_sock(sock
, data
,
3542 count
> sizeof(data
) ? sizeof(data
) : count
);
3547 count
-= (size_t) ret
;
3550 /* FIXME: passing packed structure to non-packed pointer */
3551 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3552 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3553 &cmd_ctx
->lsm
->u
.disable
.event
);
3556 case LTTNG_ENABLE_CHANNEL
:
3558 cmd_ctx
->lsm
->u
.channel
.chan
.attr
.extended
.ptr
=
3559 (struct lttng_channel_extended
*) &cmd_ctx
->lsm
->u
.channel
.extended
;
3560 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3561 &cmd_ctx
->lsm
->u
.channel
.chan
,
3562 kernel_poll_pipe
[1]);
3565 case LTTNG_TRACK_PID
:
3567 ret
= cmd_track_pid(cmd_ctx
->session
,
3568 cmd_ctx
->lsm
->domain
.type
,
3569 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3572 case LTTNG_UNTRACK_PID
:
3574 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3575 cmd_ctx
->lsm
->domain
.type
,
3576 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3579 case LTTNG_ENABLE_EVENT
:
3581 struct lttng_event
*ev
= NULL
;
3582 struct lttng_event_exclusion
*exclusion
= NULL
;
3583 struct lttng_filter_bytecode
*bytecode
= NULL
;
3584 char *filter_expression
= NULL
;
3586 /* Handle exclusion events and receive it from the client. */
3587 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3588 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3590 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3591 (count
* LTTNG_SYMBOL_NAME_LEN
));
3593 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3597 DBG("Receiving var len exclusion event list from client ...");
3598 exclusion
->count
= count
;
3599 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3600 count
* LTTNG_SYMBOL_NAME_LEN
);
3602 DBG("Nothing recv() from client var len data... continuing");
3605 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3610 /* Get filter expression from client. */
3611 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3612 size_t expression_len
=
3613 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3615 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3616 ret
= LTTNG_ERR_FILTER_INVAL
;
3621 filter_expression
= zmalloc(expression_len
);
3622 if (!filter_expression
) {
3624 ret
= LTTNG_ERR_FILTER_NOMEM
;
3628 /* Receive var. len. data */
3629 DBG("Receiving var len filter's expression from client ...");
3630 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3633 DBG("Nothing recv() from client var len data... continuing");
3635 free(filter_expression
);
3637 ret
= LTTNG_ERR_FILTER_INVAL
;
3642 /* Handle filter and get bytecode from client. */
3643 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3644 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3646 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3647 ret
= LTTNG_ERR_FILTER_INVAL
;
3648 free(filter_expression
);
3653 bytecode
= zmalloc(bytecode_len
);
3655 free(filter_expression
);
3657 ret
= LTTNG_ERR_FILTER_NOMEM
;
3661 /* Receive var. len. data */
3662 DBG("Receiving var len filter's bytecode from client ...");
3663 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3665 DBG("Nothing recv() from client var len data... continuing");
3667 free(filter_expression
);
3670 ret
= LTTNG_ERR_FILTER_INVAL
;
3674 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3675 free(filter_expression
);
3678 ret
= LTTNG_ERR_FILTER_INVAL
;
3683 ev
= lttng_event_copy(&cmd_ctx
->lsm
->u
.enable
.event
);
3685 DBG("Failed to copy event: %s",
3686 cmd_ctx
->lsm
->u
.enable
.event
.name
);
3687 ret
= LTTNG_ERR_NOMEM
;
3692 if (cmd_ctx
->lsm
->u
.enable
.userspace_probe_location_len
> 0) {
3693 /* Expect a userspace probe description. */
3694 ret
= receive_userspace_probe(cmd_ctx
, sock
, sock_error
, ev
);
3696 lttng_event_destroy(ev
);
3701 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3702 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3704 filter_expression
, bytecode
, exclusion
,
3705 kernel_poll_pipe
[1]);
3706 lttng_event_destroy(ev
);
3709 case LTTNG_LIST_TRACEPOINTS
:
3711 struct lttng_event
*events
;
3714 session_lock_list();
3715 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3716 session_unlock_list();
3717 if (nb_events
< 0) {
3718 /* Return value is a negative lttng_error_code. */
3724 * Setup lttng message with payload size set to the event list size in
3725 * bytes and then copy list into the llm payload.
3727 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3728 sizeof(struct lttng_event
) * nb_events
);
3738 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3740 struct lttng_event_field
*fields
;
3743 session_lock_list();
3744 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3746 session_unlock_list();
3747 if (nb_fields
< 0) {
3748 /* Return value is a negative lttng_error_code. */
3754 * Setup lttng message with payload size set to the event list size in
3755 * bytes and then copy list into the llm payload.
3757 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3758 sizeof(struct lttng_event_field
) * nb_fields
);
3768 case LTTNG_LIST_SYSCALLS
:
3770 struct lttng_event
*events
;
3773 nb_events
= cmd_list_syscalls(&events
);
3774 if (nb_events
< 0) {
3775 /* Return value is a negative lttng_error_code. */
3781 * Setup lttng message with payload size set to the event list size in
3782 * bytes and then copy list into the llm payload.
3784 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3785 sizeof(struct lttng_event
) * nb_events
);
3795 case LTTNG_LIST_TRACKER_PIDS
:
3797 int32_t *pids
= NULL
;
3800 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3801 cmd_ctx
->lsm
->domain
.type
, &pids
);
3803 /* Return value is a negative lttng_error_code. */
3809 * Setup lttng message with payload size set to the event list size in
3810 * bytes and then copy list into the llm payload.
3812 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3813 sizeof(int32_t) * nr_pids
);
3823 case LTTNG_SET_CONSUMER_URI
:
3826 struct lttng_uri
*uris
;
3828 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3829 len
= nb_uri
* sizeof(struct lttng_uri
);
3832 ret
= LTTNG_ERR_INVALID
;
3836 uris
= zmalloc(len
);
3838 ret
= LTTNG_ERR_FATAL
;
3842 /* Receive variable len data */
3843 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3844 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3846 DBG("No URIs received from client... continuing");
3848 ret
= LTTNG_ERR_SESSION_FAIL
;
3853 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3855 if (ret
!= LTTNG_OK
) {
3862 case LTTNG_START_TRACE
:
3865 * On the first start, if we have a kernel session and we have
3866 * enabled time or size-based rotations, we have to make sure
3867 * the kernel tracer supports it.
3869 if (!cmd_ctx
->session
->has_been_started
&& \
3870 cmd_ctx
->session
->kernel_session
&& \
3871 (cmd_ctx
->session
->rotate_timer_period
|| \
3872 cmd_ctx
->session
->rotate_size
) && \
3873 !check_rotate_compatible()) {
3874 DBG("Kernel tracer version is not compatible with the rotation feature");
3875 ret
= LTTNG_ERR_ROTATION_WRONG_VERSION
;
3878 ret
= cmd_start_trace(cmd_ctx
->session
);
3881 case LTTNG_STOP_TRACE
:
3883 ret
= cmd_stop_trace(cmd_ctx
->session
);
3886 case LTTNG_CREATE_SESSION
:
3889 struct lttng_uri
*uris
= NULL
;
3891 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3892 len
= nb_uri
* sizeof(struct lttng_uri
);
3895 uris
= zmalloc(len
);
3897 ret
= LTTNG_ERR_FATAL
;
3901 /* Receive variable len data */
3902 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3903 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3905 DBG("No URIs received from client... continuing");
3907 ret
= LTTNG_ERR_SESSION_FAIL
;
3912 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3913 DBG("Creating session with ONE network URI is a bad call");
3914 ret
= LTTNG_ERR_SESSION_FAIL
;
3920 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3921 &cmd_ctx
->creds
, 0);
3927 case LTTNG_DESTROY_SESSION
:
3929 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1],
3930 notification_thread_handle
);
3932 /* Set session to NULL so we do not unlock it after free. */
3933 cmd_ctx
->session
= NULL
;
3936 case LTTNG_LIST_DOMAINS
:
3939 struct lttng_domain
*domains
= NULL
;
3941 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3943 /* Return value is a negative lttng_error_code. */
3948 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3949 nb_dom
* sizeof(struct lttng_domain
));
3959 case LTTNG_LIST_CHANNELS
:
3961 ssize_t payload_size
;
3962 struct lttng_channel
*channels
= NULL
;
3964 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3965 cmd_ctx
->session
, &channels
);
3966 if (payload_size
< 0) {
3967 /* Return value is a negative lttng_error_code. */
3968 ret
= -payload_size
;
3972 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3983 case LTTNG_LIST_EVENTS
:
3986 struct lttng_event
*events
= NULL
;
3987 struct lttcomm_event_command_header cmd_header
;
3990 memset(&cmd_header
, 0, sizeof(cmd_header
));
3991 /* Extended infos are included at the end of events */
3992 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3993 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3994 &events
, &total_size
);
3997 /* Return value is a negative lttng_error_code. */
4002 cmd_header
.nb_events
= nb_event
;
4003 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
4004 &cmd_header
, sizeof(cmd_header
));
4014 case LTTNG_LIST_SESSIONS
:
4016 unsigned int nr_sessions
;
4017 void *sessions_payload
;
4020 session_lock_list();
4021 nr_sessions
= lttng_sessions_count(
4022 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
4023 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
4024 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
4025 sessions_payload
= zmalloc(payload_len
);
4027 if (!sessions_payload
) {
4028 session_unlock_list();
4033 cmd_list_lttng_sessions(sessions_payload
,
4034 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
4035 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
4036 session_unlock_list();
4038 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
4040 free(sessions_payload
);
4049 case LTTNG_REGISTER_CONSUMER
:
4051 struct consumer_data
*cdata
;
4053 switch (cmd_ctx
->lsm
->domain
.type
) {
4054 case LTTNG_DOMAIN_KERNEL
:
4055 cdata
= &kconsumer_data
;
4058 ret
= LTTNG_ERR_UND
;
4062 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
4063 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
4066 case LTTNG_DATA_PENDING
:
4069 uint8_t pending_ret_byte
;
4071 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
4076 * This function may returns 0 or 1 to indicate whether or not
4077 * there is data pending. In case of error, it should return an
4078 * LTTNG_ERR code. However, some code paths may still return
4079 * a nondescript error code, which we handle by returning an
4082 if (pending_ret
== 0 || pending_ret
== 1) {
4084 * ret will be set to LTTNG_OK at the end of
4087 } else if (pending_ret
< 0) {
4088 ret
= LTTNG_ERR_UNK
;
4095 pending_ret_byte
= (uint8_t) pending_ret
;
4097 /* 1 byte to return whether or not data is pending */
4098 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
4099 &pending_ret_byte
, 1);
4108 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
4110 struct lttcomm_lttng_output_id reply
;
4112 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
4113 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
4114 if (ret
!= LTTNG_OK
) {
4118 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
4124 /* Copy output list into message payload */
4128 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
4130 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
4131 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
4134 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4137 struct lttng_snapshot_output
*outputs
= NULL
;
4139 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4140 if (nb_output
< 0) {
4145 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4146 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4147 nb_output
* sizeof(struct lttng_snapshot_output
));
4157 case LTTNG_SNAPSHOT_RECORD
:
4159 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4160 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4161 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4164 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4167 struct lttng_uri
*uris
= NULL
;
4169 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4170 len
= nb_uri
* sizeof(struct lttng_uri
);
4173 uris
= zmalloc(len
);
4175 ret
= LTTNG_ERR_FATAL
;
4179 /* Receive variable len data */
4180 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4181 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4183 DBG("No URIs received from client... continuing");
4185 ret
= LTTNG_ERR_SESSION_FAIL
;
4190 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4191 DBG("Creating session with ONE network URI is a bad call");
4192 ret
= LTTNG_ERR_SESSION_FAIL
;
4198 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4199 nb_uri
, &cmd_ctx
->creds
);
4203 case LTTNG_CREATE_SESSION_LIVE
:
4206 struct lttng_uri
*uris
= NULL
;
4208 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4209 len
= nb_uri
* sizeof(struct lttng_uri
);
4212 uris
= zmalloc(len
);
4214 ret
= LTTNG_ERR_FATAL
;
4218 /* Receive variable len data */
4219 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4220 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4222 DBG("No URIs received from client... continuing");
4224 ret
= LTTNG_ERR_SESSION_FAIL
;
4229 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4230 DBG("Creating session with ONE network URI is a bad call");
4231 ret
= LTTNG_ERR_SESSION_FAIL
;
4237 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4238 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4242 case LTTNG_SAVE_SESSION
:
4244 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4248 case LTTNG_SET_SESSION_SHM_PATH
:
4250 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4251 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4254 case LTTNG_REGENERATE_METADATA
:
4256 ret
= cmd_regenerate_metadata(cmd_ctx
->session
);
4259 case LTTNG_REGENERATE_STATEDUMP
:
4261 ret
= cmd_regenerate_statedump(cmd_ctx
->session
);
4264 case LTTNG_REGISTER_TRIGGER
:
4266 ret
= cmd_register_trigger(cmd_ctx
, sock
,
4267 notification_thread_handle
);
4270 case LTTNG_UNREGISTER_TRIGGER
:
4272 ret
= cmd_unregister_trigger(cmd_ctx
, sock
,
4273 notification_thread_handle
);
4276 case LTTNG_ROTATE_SESSION
:
4278 struct lttng_rotate_session_return rotate_return
;
4280 DBG("Client rotate session \"%s\"", cmd_ctx
->session
->name
);
4282 memset(&rotate_return
, 0, sizeof(rotate_return
));
4283 if (cmd_ctx
->session
->kernel_session
&& !check_rotate_compatible()) {
4284 DBG("Kernel tracer version is not compatible with the rotation feature");
4285 ret
= LTTNG_ERR_ROTATION_WRONG_VERSION
;
4289 ret
= cmd_rotate_session(cmd_ctx
->session
, &rotate_return
);
4295 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &rotate_return
,
4296 sizeof(rotate_return
));
4305 case LTTNG_ROTATION_GET_INFO
:
4307 struct lttng_rotation_get_info_return get_info_return
;
4309 memset(&get_info_return
, 0, sizeof(get_info_return
));
4310 ret
= cmd_rotate_get_info(cmd_ctx
->session
, &get_info_return
,
4311 cmd_ctx
->lsm
->u
.get_rotation_info
.rotation_id
);
4317 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &get_info_return
,
4318 sizeof(get_info_return
));
4327 case LTTNG_SESSION_GET_CURRENT_OUTPUT
:
4329 struct lttng_session_get_current_output_return output_return
;
4331 memset(&output_return
, 0, sizeof(output_return
));
4332 ret
= cmd_session_get_current_output(cmd_ctx
->session
,
4339 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &output_return
,
4340 sizeof(output_return
));
4349 case LTTNG_ROTATION_SET_SCHEDULE
:
4352 enum lttng_rotation_schedule_type schedule_type
;
4355 if (cmd_ctx
->session
->kernel_session
&& !check_rotate_compatible()) {
4356 DBG("Kernel tracer version does not support session rotations");
4357 ret
= LTTNG_ERR_ROTATION_WRONG_VERSION
;
4361 set_schedule
= cmd_ctx
->lsm
->u
.rotation_set_schedule
.set
== 1;
4362 schedule_type
= (enum lttng_rotation_schedule_type
) cmd_ctx
->lsm
->u
.rotation_set_schedule
.type
;
4363 value
= cmd_ctx
->lsm
->u
.rotation_set_schedule
.value
;
4365 ret
= cmd_rotation_set_schedule(cmd_ctx
->session
,
4369 notification_thread_handle
);
4370 if (ret
!= LTTNG_OK
) {
4376 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES
:
4378 struct lttng_session_list_schedules_return schedules
= {
4379 .periodic
.set
= !!cmd_ctx
->session
->rotate_timer_period
,
4380 .periodic
.value
= cmd_ctx
->session
->rotate_timer_period
,
4381 .size
.set
= !!cmd_ctx
->session
->rotate_size
,
4382 .size
.value
= cmd_ctx
->session
->rotate_size
,
4385 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &schedules
,
4396 ret
= LTTNG_ERR_UND
;
4401 if (cmd_ctx
->llm
== NULL
) {
4402 DBG("Missing llm structure. Allocating one.");
4403 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4407 /* Set return code */
4408 cmd_ctx
->llm
->ret_code
= ret
;
4410 if (cmd_ctx
->session
) {
4411 session_unlock(cmd_ctx
->session
);
4413 if (need_tracing_session
) {
4414 session_unlock_list();
4417 assert(!rcu_read_ongoing());
4422 * Thread managing health check socket.
4424 static void *thread_manage_health(void *data
)
4426 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4427 uint32_t revents
, nb_fd
;
4428 struct lttng_poll_event events
;
4429 struct health_comm_msg msg
;
4430 struct health_comm_reply reply
;
4432 DBG("[thread] Manage health check started");
4434 rcu_register_thread();
4436 /* We might hit an error path before this is created. */
4437 lttng_poll_init(&events
);
4439 /* Create unix socket */
4440 sock
= lttcomm_create_unix_sock(config
.health_unix_sock_path
.value
);
4442 ERR("Unable to create health check Unix socket");
4447 /* lttng health client socket path permissions */
4448 ret
= chown(config
.health_unix_sock_path
.value
, 0,
4449 utils_get_group_id(config
.tracing_group_name
.value
));
4451 ERR("Unable to set group on %s", config
.health_unix_sock_path
.value
);
4456 ret
= chmod(config
.health_unix_sock_path
.value
,
4457 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4459 ERR("Unable to set permissions on %s", config
.health_unix_sock_path
.value
);
4466 * Set the CLOEXEC flag. Return code is useless because either way, the
4469 (void) utils_set_fd_cloexec(sock
);
4471 ret
= lttcomm_listen_unix_sock(sock
);
4477 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4478 * more will be added to this poll set.
4480 ret
= sessiond_set_thread_pollset(&events
, 2);
4485 /* Add the application registration socket */
4486 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4491 sessiond_notify_ready();
4494 DBG("Health check ready");
4496 /* Inifinite blocking call, waiting for transmission */
4498 ret
= lttng_poll_wait(&events
, -1);
4501 * Restart interrupted system call.
4503 if (errno
== EINTR
) {
4511 for (i
= 0; i
< nb_fd
; i
++) {
4512 /* Fetch once the poll data */
4513 revents
= LTTNG_POLL_GETEV(&events
, i
);
4514 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4517 /* No activity for this FD (poll implementation). */
4521 /* Thread quit pipe has been closed. Killing thread. */
4522 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4528 /* Event on the registration socket */
4529 if (pollfd
== sock
) {
4530 if (revents
& LPOLLIN
) {
4532 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4533 ERR("Health socket poll error");
4536 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4542 new_sock
= lttcomm_accept_unix_sock(sock
);
4548 * Set the CLOEXEC flag. Return code is useless because either way, the
4551 (void) utils_set_fd_cloexec(new_sock
);
4553 DBG("Receiving data from client for health...");
4554 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4556 DBG("Nothing recv() from client... continuing");
4557 ret
= close(new_sock
);
4564 rcu_thread_online();
4566 memset(&reply
, 0, sizeof(reply
));
4567 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4569 * health_check_state returns 0 if health is
4572 if (!health_check_state(health_sessiond
, i
)) {
4573 reply
.ret_code
|= 1ULL << i
;
4577 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4579 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4581 ERR("Failed to send health data back to client");
4584 /* End of transmission */
4585 ret
= close(new_sock
);
4594 ERR("Health error occurred in %s", __func__
);
4596 DBG("Health check thread dying");
4597 unlink(config
.health_unix_sock_path
.value
);
4605 lttng_poll_clean(&events
);
4607 rcu_unregister_thread();
4612 * This thread manage all clients request using the unix client socket for
4615 static void *thread_manage_clients(void *data
)
4617 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4619 uint32_t revents
, nb_fd
;
4620 struct command_ctx
*cmd_ctx
= NULL
;
4621 struct lttng_poll_event events
;
4623 DBG("[thread] Manage client started");
4625 rcu_register_thread();
4627 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4629 health_code_update();
4631 ret
= lttcomm_listen_unix_sock(client_sock
);
4637 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4638 * more will be added to this poll set.
4640 ret
= sessiond_set_thread_pollset(&events
, 2);
4642 goto error_create_poll
;
4645 /* Add the application registration socket */
4646 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4651 ret
= sem_post(&load_info
->message_thread_ready
);
4653 PERROR("sem_post message_thread_ready");
4658 * Wait until all support threads are initialized before accepting
4661 while (uatomic_read(<tng_sessiond_ready
) != 0) {
4663 struct timeval timeout
;
4666 FD_SET(thread_quit_pipe
[0], &read_fds
);
4667 memset(&timeout
, 0, sizeof(timeout
));
4668 timeout
.tv_usec
= 1000;
4671 * If a support thread failed to launch, it may signal that
4672 * we must exit and the sessiond would never be marked as
4675 * The timeout is set to 1ms, which serves as a way to
4676 * pace down this check.
4678 ret
= select(thread_quit_pipe
[0] + 1, &read_fds
, NULL
, NULL
,
4680 if (ret
> 0 || (ret
< 0 && errno
!= EINTR
)) {
4685 * This barrier is paired with the one in sessiond_notify_ready() to
4686 * ensure that loads accessing data initialized by the other threads,
4687 * on which this thread was waiting, are not performed before this point.
4689 * Note that this could be a 'read' memory barrier, but a full barrier
4690 * is used in case the code changes. The performance implications of
4691 * this choice are minimal since this is a slow path.
4695 /* This testpoint is after we signal readiness to the parent. */
4696 if (testpoint(sessiond_thread_manage_clients
)) {
4700 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4704 health_code_update();
4707 const struct cmd_completion_handler
*cmd_completion_handler
;
4709 DBG("Accepting client command ...");
4711 /* Inifinite blocking call, waiting for transmission */
4713 health_poll_entry();
4714 ret
= lttng_poll_wait(&events
, -1);
4718 * Restart interrupted system call.
4720 if (errno
== EINTR
) {
4728 for (i
= 0; i
< nb_fd
; i
++) {
4729 /* Fetch once the poll data */
4730 revents
= LTTNG_POLL_GETEV(&events
, i
);
4731 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4733 health_code_update();
4736 /* No activity for this FD (poll implementation). */
4740 /* Thread quit pipe has been closed. Killing thread. */
4741 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4747 /* Event on the registration socket */
4748 if (pollfd
== client_sock
) {
4749 if (revents
& LPOLLIN
) {
4751 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4752 ERR("Client socket poll error");
4755 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4761 DBG("Wait for client response");
4763 health_code_update();
4765 sock
= lttcomm_accept_unix_sock(client_sock
);
4771 * Set the CLOEXEC flag. Return code is useless because either way, the
4774 (void) utils_set_fd_cloexec(sock
);
4776 /* Set socket option for credentials retrieval */
4777 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4782 /* Allocate context command to process the client request */
4783 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4784 if (cmd_ctx
== NULL
) {
4785 PERROR("zmalloc cmd_ctx");
4789 /* Allocate data buffer for reception */
4790 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4791 if (cmd_ctx
->lsm
== NULL
) {
4792 PERROR("zmalloc cmd_ctx->lsm");
4796 cmd_ctx
->llm
= NULL
;
4797 cmd_ctx
->session
= NULL
;
4799 health_code_update();
4802 * Data is received from the lttng client. The struct
4803 * lttcomm_session_msg (lsm) contains the command and data request of
4806 DBG("Receiving data from client ...");
4807 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4808 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4810 DBG("Nothing recv() from client... continuing");
4816 clean_command_ctx(&cmd_ctx
);
4820 health_code_update();
4822 // TODO: Validate cmd_ctx including sanity check for
4823 // security purpose.
4825 rcu_thread_online();
4827 * This function dispatch the work to the kernel or userspace tracer
4828 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4829 * informations for the client. The command context struct contains
4830 * everything this function may needs.
4832 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4833 rcu_thread_offline();
4841 * TODO: Inform client somehow of the fatal error. At
4842 * this point, ret < 0 means that a zmalloc failed
4843 * (ENOMEM). Error detected but still accept
4844 * command, unless a socket error has been
4847 clean_command_ctx(&cmd_ctx
);
4851 cmd_completion_handler
= cmd_pop_completion_handler();
4852 if (cmd_completion_handler
) {
4853 enum lttng_error_code completion_code
;
4855 completion_code
= cmd_completion_handler
->run(
4856 cmd_completion_handler
->data
);
4857 if (completion_code
!= LTTNG_OK
) {
4858 clean_command_ctx(&cmd_ctx
);
4863 health_code_update();
4865 DBG("Sending response (size: %d, retcode: %s (%d))",
4866 cmd_ctx
->lttng_msg_size
,
4867 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4868 cmd_ctx
->llm
->ret_code
);
4869 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4871 ERR("Failed to send data back to client");
4874 /* End of transmission */
4881 clean_command_ctx(&cmd_ctx
);
4883 health_code_update();
4895 lttng_poll_clean(&events
);
4896 clean_command_ctx(&cmd_ctx
);
4900 unlink(config
.client_unix_sock_path
.value
);
4901 if (client_sock
>= 0) {
4902 ret
= close(client_sock
);
4910 ERR("Health error occurred in %s", __func__
);
4913 health_unregister(health_sessiond
);
4915 DBG("Client thread dying");
4917 rcu_unregister_thread();
4920 * Since we are creating the consumer threads, we own them, so we need
4921 * to join them before our thread exits.
4923 ret
= join_consumer_thread(&kconsumer_data
);
4926 PERROR("join_consumer");
4929 ret
= join_consumer_thread(&ustconsumer32_data
);
4932 PERROR("join_consumer ust32");
4935 ret
= join_consumer_thread(&ustconsumer64_data
);
4938 PERROR("join_consumer ust64");
4943 static int string_match(const char *str1
, const char *str2
)
4945 return (str1
&& str2
) && !strcmp(str1
, str2
);
4949 * Take an option from the getopt output and set it in the right variable to be
4952 * Return 0 on success else a negative value.
4954 static int set_option(int opt
, const char *arg
, const char *optname
)
4958 if (string_match(optname
, "client-sock") || opt
== 'c') {
4959 if (!arg
|| *arg
== '\0') {
4963 if (lttng_is_setuid_setgid()) {
4964 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4965 "-c, --client-sock");
4967 config_string_set(&config
.client_unix_sock_path
,
4969 if (!config
.client_unix_sock_path
.value
) {
4974 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4975 if (!arg
|| *arg
== '\0') {
4979 if (lttng_is_setuid_setgid()) {
4980 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4983 config_string_set(&config
.apps_unix_sock_path
,
4985 if (!config
.apps_unix_sock_path
.value
) {
4990 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4991 config
.daemonize
= true;
4992 } else if (string_match(optname
, "background") || opt
== 'b') {
4993 config
.background
= true;
4994 } else if (string_match(optname
, "group") || opt
== 'g') {
4995 if (!arg
|| *arg
== '\0') {
4999 if (lttng_is_setuid_setgid()) {
5000 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5003 config_string_set(&config
.tracing_group_name
,
5005 if (!config
.tracing_group_name
.value
) {
5010 } else if (string_match(optname
, "help") || opt
== 'h') {
5011 ret
= utils_show_help(8, "lttng-sessiond", help_msg
);
5013 ERR("Cannot show --help for `lttng-sessiond`");
5016 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
5017 } else if (string_match(optname
, "version") || opt
== 'V') {
5018 fprintf(stdout
, "%s\n", VERSION
);
5020 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
5021 config
.sig_parent
= true;
5022 } else if (string_match(optname
, "kconsumerd-err-sock")) {
5023 if (!arg
|| *arg
== '\0') {
5027 if (lttng_is_setuid_setgid()) {
5028 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5029 "--kconsumerd-err-sock");
5031 config_string_set(&config
.kconsumerd_err_unix_sock_path
,
5033 if (!config
.kconsumerd_err_unix_sock_path
.value
) {
5038 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
5039 if (!arg
|| *arg
== '\0') {
5043 if (lttng_is_setuid_setgid()) {
5044 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5045 "--kconsumerd-cmd-sock");
5047 config_string_set(&config
.kconsumerd_cmd_unix_sock_path
,
5049 if (!config
.kconsumerd_cmd_unix_sock_path
.value
) {
5054 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
5055 if (!arg
|| *arg
== '\0') {
5059 if (lttng_is_setuid_setgid()) {
5060 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5061 "--ustconsumerd64-err-sock");
5063 config_string_set(&config
.consumerd64_err_unix_sock_path
,
5065 if (!config
.consumerd64_err_unix_sock_path
.value
) {
5070 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
5071 if (!arg
|| *arg
== '\0') {
5075 if (lttng_is_setuid_setgid()) {
5076 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5077 "--ustconsumerd64-cmd-sock");
5079 config_string_set(&config
.consumerd64_cmd_unix_sock_path
,
5081 if (!config
.consumerd64_cmd_unix_sock_path
.value
) {
5086 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
5087 if (!arg
|| *arg
== '\0') {
5091 if (lttng_is_setuid_setgid()) {
5092 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5093 "--ustconsumerd32-err-sock");
5095 config_string_set(&config
.consumerd32_err_unix_sock_path
,
5097 if (!config
.consumerd32_err_unix_sock_path
.value
) {
5102 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
5103 if (!arg
|| *arg
== '\0') {
5107 if (lttng_is_setuid_setgid()) {
5108 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5109 "--ustconsumerd32-cmd-sock");
5111 config_string_set(&config
.consumerd32_cmd_unix_sock_path
,
5113 if (!config
.consumerd32_cmd_unix_sock_path
.value
) {
5118 } else if (string_match(optname
, "no-kernel")) {
5119 config
.no_kernel
= true;
5120 } else if (string_match(optname
, "quiet") || opt
== 'q') {
5121 config
.quiet
= true;
5122 } else if (string_match(optname
, "verbose") || opt
== 'v') {
5123 /* Verbose level can increase using multiple -v */
5125 /* Value obtained from config file */
5126 config
.verbose
= config_parse_value(arg
);
5128 /* -v used on command line */
5131 /* Clamp value to [0, 3] */
5132 config
.verbose
= config
.verbose
< 0 ? 0 :
5133 (config
.verbose
<= 3 ? config
.verbose
: 3);
5134 } else if (string_match(optname
, "verbose-consumer")) {
5136 config
.verbose_consumer
= config_parse_value(arg
);
5138 config
.verbose_consumer
++;
5140 } else if (string_match(optname
, "consumerd32-path")) {
5141 if (!arg
|| *arg
== '\0') {
5145 if (lttng_is_setuid_setgid()) {
5146 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5147 "--consumerd32-path");
5149 config_string_set(&config
.consumerd32_bin_path
,
5151 if (!config
.consumerd32_bin_path
.value
) {
5156 } else if (string_match(optname
, "consumerd32-libdir")) {
5157 if (!arg
|| *arg
== '\0') {
5161 if (lttng_is_setuid_setgid()) {
5162 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5163 "--consumerd32-libdir");
5165 config_string_set(&config
.consumerd32_lib_dir
,
5167 if (!config
.consumerd32_lib_dir
.value
) {
5172 } else if (string_match(optname
, "consumerd64-path")) {
5173 if (!arg
|| *arg
== '\0') {
5177 if (lttng_is_setuid_setgid()) {
5178 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5179 "--consumerd64-path");
5181 config_string_set(&config
.consumerd64_bin_path
,
5183 if (!config
.consumerd64_bin_path
.value
) {
5188 } else if (string_match(optname
, "consumerd64-libdir")) {
5189 if (!arg
|| *arg
== '\0') {
5193 if (lttng_is_setuid_setgid()) {
5194 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5195 "--consumerd64-libdir");
5197 config_string_set(&config
.consumerd64_lib_dir
,
5199 if (!config
.consumerd64_lib_dir
.value
) {
5204 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
5205 if (!arg
|| *arg
== '\0') {
5209 if (lttng_is_setuid_setgid()) {
5210 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5213 config_string_set(&config
.pid_file_path
, strdup(arg
));
5214 if (!config
.pid_file_path
.value
) {
5219 } else if (string_match(optname
, "agent-tcp-port")) {
5220 if (!arg
|| *arg
== '\0') {
5224 if (lttng_is_setuid_setgid()) {
5225 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5226 "--agent-tcp-port");
5231 v
= strtoul(arg
, NULL
, 0);
5232 if (errno
!= 0 || !isdigit(arg
[0])) {
5233 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
5236 if (v
== 0 || v
>= 65535) {
5237 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
5240 config
.agent_tcp_port
.begin
= config
.agent_tcp_port
.end
= (int) v
;
5241 DBG3("Agent TCP port set to non default: %i", (int) v
);
5243 } else if (string_match(optname
, "load") || opt
== 'l') {
5244 if (!arg
|| *arg
== '\0') {
5248 if (lttng_is_setuid_setgid()) {
5249 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5252 config_string_set(&config
.load_session_path
, strdup(arg
));
5253 if (!config
.load_session_path
.value
) {
5258 } else if (string_match(optname
, "kmod-probes")) {
5259 if (!arg
|| *arg
== '\0') {
5263 if (lttng_is_setuid_setgid()) {
5264 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5267 config_string_set(&config
.kmod_probes_list
, strdup(arg
));
5268 if (!config
.kmod_probes_list
.value
) {
5273 } else if (string_match(optname
, "extra-kmod-probes")) {
5274 if (!arg
|| *arg
== '\0') {
5278 if (lttng_is_setuid_setgid()) {
5279 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5280 "--extra-kmod-probes");
5282 config_string_set(&config
.kmod_extra_probes_list
,
5284 if (!config
.kmod_extra_probes_list
.value
) {
5289 } else if (string_match(optname
, "config") || opt
== 'f') {
5290 /* This is handled in set_options() thus silent skip. */
5293 /* Unknown option or other error.
5294 * Error is printed by getopt, just return */
5299 if (ret
== -EINVAL
) {
5300 const char *opt_name
= "unknown";
5303 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
5305 if (opt
== long_options
[i
].val
) {
5306 opt_name
= long_options
[i
].name
;
5311 WARN("Invalid argument provided for option \"%s\", using default value.",
5319 * config_entry_handler_cb used to handle options read from a config file.
5320 * See config_entry_handler_cb comment in common/config/session-config.h for the
5321 * return value conventions.
5323 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
5327 if (!entry
|| !entry
->name
|| !entry
->value
) {
5332 /* Check if the option is to be ignored */
5333 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5334 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5339 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5342 /* Ignore if not fully matched. */
5343 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5348 * If the option takes no argument on the command line, we have to
5349 * check if the value is "true". We support non-zero numeric values,
5352 if (!long_options
[i
].has_arg
) {
5353 ret
= config_parse_value(entry
->value
);
5356 WARN("Invalid configuration value \"%s\" for option %s",
5357 entry
->value
, entry
->name
);
5359 /* False, skip boolean config option. */
5364 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5368 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5375 * daemon configuration loading and argument parsing
5377 static int set_options(int argc
, char **argv
)
5379 int ret
= 0, c
= 0, option_index
= 0;
5380 int orig_optopt
= optopt
, orig_optind
= optind
;
5382 const char *config_path
= NULL
;
5384 optstring
= utils_generate_optstring(long_options
,
5385 sizeof(long_options
) / sizeof(struct option
));
5391 /* Check for the --config option */
5392 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5393 &option_index
)) != -1) {
5397 } else if (c
!= 'f') {
5398 /* if not equal to --config option. */
5402 if (lttng_is_setuid_setgid()) {
5403 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5406 config_path
= utils_expand_path(optarg
);
5408 ERR("Failed to resolve path: %s", optarg
);
5413 ret
= config_get_section_entries(config_path
, config_section_name
,
5414 config_entry_handler
, NULL
);
5417 ERR("Invalid configuration option at line %i", ret
);
5423 /* Reset getopt's global state */
5424 optopt
= orig_optopt
;
5425 optind
= orig_optind
;
5429 * getopt_long() will not set option_index if it encounters a
5432 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5439 * Pass NULL as the long option name if popt left the index
5442 ret
= set_option(c
, optarg
,
5443 option_index
< 0 ? NULL
:
5444 long_options
[option_index
].name
);
5456 * Creates the two needed socket by the daemon.
5457 * apps_sock - The communication socket for all UST apps.
5458 * client_sock - The communication of the cli tool (lttng).
5460 static int init_daemon_socket(void)
5465 old_umask
= umask(0);
5467 /* Create client tool unix socket */
5468 client_sock
= lttcomm_create_unix_sock(config
.client_unix_sock_path
.value
);
5469 if (client_sock
< 0) {
5470 ERR("Create unix sock failed: %s", config
.client_unix_sock_path
.value
);
5475 /* Set the cloexec flag */
5476 ret
= utils_set_fd_cloexec(client_sock
);
5478 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5479 "Continuing but note that the consumer daemon will have a "
5480 "reference to this socket on exec()", client_sock
);
5483 /* File permission MUST be 660 */
5484 ret
= chmod(config
.client_unix_sock_path
.value
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5486 ERR("Set file permissions failed: %s", config
.client_unix_sock_path
.value
);
5491 /* Create the application unix socket */
5492 apps_sock
= lttcomm_create_unix_sock(config
.apps_unix_sock_path
.value
);
5493 if (apps_sock
< 0) {
5494 ERR("Create unix sock failed: %s", config
.apps_unix_sock_path
.value
);
5499 /* Set the cloexec flag */
5500 ret
= utils_set_fd_cloexec(apps_sock
);
5502 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5503 "Continuing but note that the consumer daemon will have a "
5504 "reference to this socket on exec()", apps_sock
);
5507 /* File permission MUST be 666 */
5508 ret
= chmod(config
.apps_unix_sock_path
.value
,
5509 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5511 ERR("Set file permissions failed: %s", config
.apps_unix_sock_path
.value
);
5516 DBG3("Session daemon client socket %d and application socket %d created",
5517 client_sock
, apps_sock
);
5525 * Create lockfile using the rundir and return its fd.
5527 static int create_lockfile(void)
5529 return utils_create_lock_file(config
.lock_file_path
.value
);
5533 * Check if the global socket is available, and if a daemon is answering at the
5534 * other side. If yes, error is returned.
5536 * Also attempts to create and hold the lock file.
5538 static int check_existing_daemon(void)
5542 /* Is there anybody out there ? */
5543 if (lttng_session_daemon_alive()) {
5548 lockfile_fd
= create_lockfile();
5549 if (lockfile_fd
< 0) {
5557 static void sessiond_cleanup_lock_file(void)
5562 * Cleanup lock file by deleting it and finaly closing it which will
5563 * release the file system lock.
5565 if (lockfile_fd
>= 0) {
5566 ret
= remove(config
.lock_file_path
.value
);
5568 PERROR("remove lock file");
5570 ret
= close(lockfile_fd
);
5572 PERROR("close lock file");
5578 * Set the tracing group gid onto the client socket.
5580 * Race window between mkdir and chown is OK because we are going from more
5581 * permissive (root.root) to less permissive (root.tracing).
5583 static int set_permissions(char *rundir
)
5588 gid
= utils_get_group_id(config
.tracing_group_name
.value
);
5590 /* Set lttng run dir */
5591 ret
= chown(rundir
, 0, gid
);
5593 ERR("Unable to set group on %s", rundir
);
5598 * Ensure all applications and tracing group can search the run
5599 * dir. Allow everyone to read the directory, since it does not
5600 * buy us anything to hide its content.
5602 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5604 ERR("Unable to set permissions on %s", rundir
);
5608 /* lttng client socket path */
5609 ret
= chown(config
.client_unix_sock_path
.value
, 0, gid
);
5611 ERR("Unable to set group on %s", config
.client_unix_sock_path
.value
);
5615 /* kconsumer error socket path */
5616 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5618 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5622 /* 64-bit ustconsumer error socket path */
5623 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5625 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5629 /* 32-bit ustconsumer compat32 error socket path */
5630 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5632 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5636 DBG("All permissions are set");
5642 * Create the lttng run directory needed for all global sockets and pipe.
5644 static int create_lttng_rundir(void)
5648 DBG3("Creating LTTng run directory: %s", config
.rundir
.value
);
5650 ret
= mkdir(config
.rundir
.value
, S_IRWXU
);
5652 if (errno
!= EEXIST
) {
5653 ERR("Unable to create %s", config
.rundir
.value
);
5665 * Setup sockets and directory needed by the consumerds' communication with the
5668 static int set_consumer_sockets(struct consumer_data
*consumer_data
)
5673 switch (consumer_data
->type
) {
5674 case LTTNG_CONSUMER_KERNEL
:
5675 path
= config
.kconsumerd_path
.value
;
5677 case LTTNG_CONSUMER64_UST
:
5678 path
= config
.consumerd64_path
.value
;
5680 case LTTNG_CONSUMER32_UST
:
5681 path
= config
.consumerd32_path
.value
;
5684 ERR("Consumer type unknown");
5690 DBG2("Creating consumer directory: %s", path
);
5692 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5693 if (ret
< 0 && errno
!= EEXIST
) {
5695 ERR("Failed to create %s", path
);
5699 ret
= chown(path
, 0, utils_get_group_id(config
.tracing_group_name
.value
));
5701 ERR("Unable to set group on %s", path
);
5707 /* Create the consumerd error unix socket */
5708 consumer_data
->err_sock
=
5709 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5710 if (consumer_data
->err_sock
< 0) {
5711 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5717 * Set the CLOEXEC flag. Return code is useless because either way, the
5720 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5722 PERROR("utils_set_fd_cloexec");
5723 /* continue anyway */
5726 /* File permission MUST be 660 */
5727 ret
= chmod(consumer_data
->err_unix_sock_path
,
5728 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5730 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5740 * Signal handler for the daemon
5742 * Simply stop all worker threads, leaving main() return gracefully after
5743 * joining all threads and calling cleanup().
5745 static void sighandler(int sig
)
5749 DBG("SIGINT caught");
5753 DBG("SIGTERM caught");
5757 CMM_STORE_SHARED(recv_child_signal
, 1);
5765 * Setup signal handler for :
5766 * SIGINT, SIGTERM, SIGPIPE
5768 static int set_signal_handler(void)
5771 struct sigaction sa
;
5774 if ((ret
= sigemptyset(&sigset
)) < 0) {
5775 PERROR("sigemptyset");
5779 sa
.sa_mask
= sigset
;
5782 sa
.sa_handler
= sighandler
;
5783 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5784 PERROR("sigaction");
5788 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5789 PERROR("sigaction");
5793 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5794 PERROR("sigaction");
5798 sa
.sa_handler
= SIG_IGN
;
5799 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5800 PERROR("sigaction");
5804 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5810 * Set open files limit to unlimited. This daemon can open a large number of
5811 * file descriptors in order to consume multiple kernel traces.
5813 static void set_ulimit(void)
5818 /* The kernel does not allow an infinite limit for open files */
5819 lim
.rlim_cur
= 65535;
5820 lim
.rlim_max
= 65535;
5822 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5824 PERROR("failed to set open files limit");
5828 static int write_pidfile(void)
5830 return utils_create_pid_file(getpid(), config
.pid_file_path
.value
);
5833 static int set_clock_plugin_env(void)
5836 char *env_value
= NULL
;
5838 if (!config
.lttng_ust_clock_plugin
.value
) {
5842 ret
= asprintf(&env_value
, "LTTNG_UST_CLOCK_PLUGIN=%s",
5843 config
.lttng_ust_clock_plugin
.value
);
5849 ret
= putenv(env_value
);
5852 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5856 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5857 config
.lttng_ust_clock_plugin
.value
);
5863 struct rotation_thread_timer_queue
*create_rotate_timer_queue(void)
5865 struct rotation_thread_timer_queue
*queue
= NULL
;
5867 queue
= zmalloc(sizeof(struct rotation_thread_timer_queue
));
5869 PERROR("Failed to allocate timer rotate queue");
5873 queue
->event_pipe
= lttng_pipe_open(FD_CLOEXEC
| O_NONBLOCK
);
5874 CDS_INIT_LIST_HEAD(&queue
->list
);
5875 pthread_mutex_init(&queue
->lock
, NULL
);
5882 void destroy_rotate_timer_queue(struct rotation_thread_timer_queue
*queue
)
5884 struct sessiond_rotation_timer
*node
, *tmp_node
;
5890 lttng_pipe_destroy(queue
->event_pipe
);
5892 pthread_mutex_lock(&queue
->lock
);
5893 /* Empty wait queue. */
5894 cds_list_for_each_entry_safe(node
, tmp_node
, &queue
->list
, head
) {
5895 cds_list_del(&node
->head
);
5898 pthread_mutex_unlock(&queue
->lock
);
5900 pthread_mutex_destroy(&queue
->lock
);
5907 int main(int argc
, char **argv
)
5909 int ret
= 0, retval
= 0;
5911 const char *env_app_timeout
;
5912 struct lttng_pipe
*ust32_channel_monitor_pipe
= NULL
,
5913 *ust64_channel_monitor_pipe
= NULL
,
5914 *kernel_channel_monitor_pipe
= NULL
;
5915 bool notification_thread_launched
= false;
5916 bool rotation_thread_launched
= false;
5917 bool timer_thread_launched
= false;
5918 struct lttng_pipe
*ust32_channel_rotate_pipe
= NULL
,
5919 *ust64_channel_rotate_pipe
= NULL
,
5920 *kernel_channel_rotate_pipe
= NULL
;
5921 struct timer_thread_parameters timer_thread_ctx
;
5922 /* Queue of rotation jobs populated by the sessiond-timer. */
5923 struct rotation_thread_timer_queue
*rotation_timer_queue
= NULL
;
5924 sem_t notification_thread_ready
;
5926 init_kernel_workarounds();
5928 rcu_register_thread();
5930 if (set_signal_handler()) {
5932 goto exit_set_signal_handler
;
5935 if (sessiond_timer_signal_init()) {
5937 goto exit_set_signal_handler
;
5940 page_size
= sysconf(_SC_PAGESIZE
);
5941 if (page_size
< 0) {
5942 PERROR("sysconf _SC_PAGESIZE");
5943 page_size
= LONG_MAX
;
5944 WARN("Fallback page size to %ld", page_size
);
5947 ret
= sessiond_config_init(&config
);
5950 goto exit_set_signal_handler
;
5954 * Init config from environment variables.
5955 * Command line option override env configuration per-doc. Do env first.
5957 sessiond_config_apply_env_config(&config
);
5960 * Parse arguments and load the daemon configuration file.
5962 * We have an exit_options exit path to free memory reserved by
5963 * set_options. This is needed because the rest of sessiond_cleanup()
5964 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5965 * depends on set_options.
5968 if (set_options(argc
, argv
)) {
5974 * Resolve all paths received as arguments, configuration option, or
5975 * through environment variable as absolute paths. This is necessary
5976 * since daemonizing causes the sessiond's current working directory
5979 ret
= sessiond_config_resolve_paths(&config
);
5985 lttng_opt_verbose
= config
.verbose
;
5986 lttng_opt_quiet
= config
.quiet
;
5987 kconsumer_data
.err_unix_sock_path
=
5988 config
.kconsumerd_err_unix_sock_path
.value
;
5989 kconsumer_data
.cmd_unix_sock_path
=
5990 config
.kconsumerd_cmd_unix_sock_path
.value
;
5991 ustconsumer32_data
.err_unix_sock_path
=
5992 config
.consumerd32_err_unix_sock_path
.value
;
5993 ustconsumer32_data
.cmd_unix_sock_path
=
5994 config
.consumerd32_cmd_unix_sock_path
.value
;
5995 ustconsumer64_data
.err_unix_sock_path
=
5996 config
.consumerd64_err_unix_sock_path
.value
;
5997 ustconsumer64_data
.cmd_unix_sock_path
=
5998 config
.consumerd64_cmd_unix_sock_path
.value
;
5999 set_clock_plugin_env();
6001 sessiond_config_log(&config
);
6003 if (create_lttng_rundir()) {
6008 /* Abort launch if a session daemon is already running. */
6009 if (check_existing_daemon()) {
6010 ERR("A session daemon is already running.");
6016 if (config
.daemonize
|| config
.background
) {
6019 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
6020 !config
.background
);
6027 * We are in the child. Make sure all other file descriptors are
6028 * closed, in case we are called with more opened file
6029 * descriptors than the standard ones and the lock file.
6031 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
6032 if (i
== lockfile_fd
) {
6039 if (run_as_create_worker(argv
[0]) < 0) {
6040 goto exit_create_run_as_worker_cleanup
;
6044 * Starting from here, we can create threads. This needs to be after
6045 * lttng_daemonize due to RCU.
6049 * Initialize the health check subsystem. This call should set the
6050 * appropriate time values.
6052 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
6053 if (!health_sessiond
) {
6054 PERROR("health_app_create error");
6056 goto exit_health_sessiond_cleanup
;
6059 /* Create thread to clean up RCU hash tables */
6060 if (init_ht_cleanup_thread(&ht_cleanup_thread
)) {
6062 goto exit_ht_cleanup
;
6065 /* Create thread quit pipe */
6066 if (init_thread_quit_pipe()) {
6068 goto exit_init_data
;
6071 /* Check if daemon is UID = 0 */
6072 is_root
= !getuid();
6074 /* Create global run dir with root access */
6076 kernel_channel_monitor_pipe
= lttng_pipe_open(0);
6077 if (!kernel_channel_monitor_pipe
) {
6078 ERR("Failed to create kernel consumer channel monitor pipe");
6080 goto exit_init_data
;
6082 kconsumer_data
.channel_monitor_pipe
=
6083 lttng_pipe_release_writefd(
6084 kernel_channel_monitor_pipe
);
6085 if (kconsumer_data
.channel_monitor_pipe
< 0) {
6087 goto exit_init_data
;
6089 kernel_channel_rotate_pipe
= lttng_pipe_open(0);
6090 if (!kernel_channel_rotate_pipe
) {
6091 ERR("Failed to create kernel consumer channel rotate pipe");
6093 goto exit_init_data
;
6095 kconsumer_data
.channel_rotate_pipe
=
6096 lttng_pipe_release_writefd(
6097 kernel_channel_rotate_pipe
);
6098 if (kconsumer_data
.channel_rotate_pipe
< 0) {
6100 goto exit_init_data
;
6104 /* Set consumer initial state */
6105 kernel_consumerd_state
= CONSUMER_STOPPED
;
6106 ust_consumerd_state
= CONSUMER_STOPPED
;
6108 ust32_channel_monitor_pipe
= lttng_pipe_open(0);
6109 if (!ust32_channel_monitor_pipe
) {
6110 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
6112 goto exit_init_data
;
6114 ustconsumer32_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
6115 ust32_channel_monitor_pipe
);
6116 if (ustconsumer32_data
.channel_monitor_pipe
< 0) {
6118 goto exit_init_data
;
6120 ust32_channel_rotate_pipe
= lttng_pipe_open(0);
6121 if (!ust32_channel_rotate_pipe
) {
6122 ERR("Failed to create 32-bit user space consumer channel rotate pipe");
6124 goto exit_init_data
;
6126 ustconsumer32_data
.channel_rotate_pipe
= lttng_pipe_release_writefd(
6127 ust32_channel_rotate_pipe
);
6128 if (ustconsumer32_data
.channel_rotate_pipe
< 0) {
6130 goto exit_init_data
;
6134 * The rotation_timer_queue structure is shared between the sessiond timer
6135 * thread and the rotation thread. The main() keeps the ownership and
6136 * destroys it when both threads have quit.
6138 rotation_timer_queue
= create_rotate_timer_queue();
6139 if (!rotation_timer_queue
) {
6141 goto exit_init_data
;
6143 timer_thread_ctx
.rotation_timer_queue
= rotation_timer_queue
;
6145 ust64_channel_monitor_pipe
= lttng_pipe_open(0);
6146 if (!ust64_channel_monitor_pipe
) {
6147 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
6149 goto exit_init_data
;
6151 ustconsumer64_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
6152 ust64_channel_monitor_pipe
);
6153 if (ustconsumer64_data
.channel_monitor_pipe
< 0) {
6155 goto exit_init_data
;
6157 ust64_channel_rotate_pipe
= lttng_pipe_open(0);
6158 if (!ust64_channel_rotate_pipe
) {
6159 ERR("Failed to create 64-bit user space consumer channel rotate pipe");
6161 goto exit_init_data
;
6163 ustconsumer64_data
.channel_rotate_pipe
= lttng_pipe_release_writefd(
6164 ust64_channel_rotate_pipe
);
6165 if (ustconsumer64_data
.channel_rotate_pipe
< 0) {
6167 goto exit_init_data
;
6171 * Init UST app hash table. Alloc hash table before this point since
6172 * cleanup() can get called after that point.
6174 if (ust_app_ht_alloc()) {
6175 ERR("Failed to allocate UST app hash table");
6177 goto exit_init_data
;
6181 * Initialize agent app hash table. We allocate the hash table here
6182 * since cleanup() can get called after this point.
6184 if (agent_app_ht_alloc()) {
6185 ERR("Failed to allocate Agent app hash table");
6187 goto exit_init_data
;
6191 * These actions must be executed as root. We do that *after* setting up
6192 * the sockets path because we MUST make the check for another daemon using
6193 * those paths *before* trying to set the kernel consumer sockets and init
6197 if (set_consumer_sockets(&kconsumer_data
)) {
6199 goto exit_init_data
;
6202 /* Setup kernel tracer */
6203 if (!config
.no_kernel
) {
6204 init_kernel_tracer();
6205 if (kernel_tracer_fd
>= 0) {
6206 ret
= syscall_init_table();
6208 ERR("Unable to populate syscall table. "
6209 "Syscall tracing won't work "
6210 "for this session daemon.");
6215 /* Set ulimit for open files */
6218 /* init lttng_fd tracking must be done after set_ulimit. */
6221 if (set_consumer_sockets(&ustconsumer64_data
)) {
6223 goto exit_init_data
;
6226 if (set_consumer_sockets(&ustconsumer32_data
)) {
6228 goto exit_init_data
;
6231 /* Setup the needed unix socket */
6232 if (init_daemon_socket()) {
6234 goto exit_init_data
;
6237 /* Set credentials to socket */
6238 if (is_root
&& set_permissions(config
.rundir
.value
)) {
6240 goto exit_init_data
;
6243 /* Get parent pid if -S, --sig-parent is specified. */
6244 if (config
.sig_parent
) {
6248 /* Setup the kernel pipe for waking up the kernel thread */
6249 if (is_root
&& !config
.no_kernel
) {
6250 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
6252 goto exit_init_data
;
6256 /* Setup the thread apps communication pipe. */
6257 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
6259 goto exit_init_data
;
6262 /* Setup the thread apps notify communication pipe. */
6263 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
6265 goto exit_init_data
;
6268 /* Initialize global buffer per UID and PID registry. */
6269 buffer_reg_init_uid_registry();
6270 buffer_reg_init_pid_registry();
6272 /* Init UST command queue. */
6273 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
6276 * Get session list pointer. This pointer MUST NOT be free'd. This list
6277 * is statically declared in session.c
6279 session_list_ptr
= session_get_list();
6283 /* Check for the application socket timeout env variable. */
6284 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
6285 if (env_app_timeout
) {
6286 config
.app_socket_timeout
= atoi(env_app_timeout
);
6288 config
.app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
6291 ret
= write_pidfile();
6293 ERR("Error in write_pidfile");
6295 goto exit_init_data
;
6298 /* Initialize communication library */
6300 /* Initialize TCP timeout values */
6301 lttcomm_inet_init();
6303 if (load_session_init_data(&load_info
) < 0) {
6305 goto exit_init_data
;
6307 load_info
->path
= config
.load_session_path
.value
;
6309 /* Create health-check thread. */
6310 ret
= pthread_create(&health_thread
, default_pthread_attr(),
6311 thread_manage_health
, (void *) NULL
);
6314 PERROR("pthread_create health");
6320 * The rotation thread needs the notification thread to be ready before
6321 * creating the rotate_notification_channel, so we use this semaphore as
6322 * a rendez-vous point.
6324 sem_init(¬ification_thread_ready
, 0, 0);
6326 /* notification_thread_data acquires the pipes' read side. */
6327 notification_thread_handle
= notification_thread_handle_create(
6328 ust32_channel_monitor_pipe
,
6329 ust64_channel_monitor_pipe
,
6330 kernel_channel_monitor_pipe
,
6331 ¬ification_thread_ready
);
6332 if (!notification_thread_handle
) {
6334 ERR("Failed to create notification thread shared data");
6336 goto exit_notification
;
6339 /* Create notification thread. */
6340 ret
= pthread_create(¬ification_thread
, default_pthread_attr(),
6341 thread_notification
, notification_thread_handle
);
6344 PERROR("pthread_create notification");
6347 goto exit_notification
;
6349 notification_thread_launched
= true;
6351 /* Create timer thread. */
6352 ret
= pthread_create(&timer_thread
, default_pthread_attr(),
6353 sessiond_timer_thread
, &timer_thread_ctx
);
6356 PERROR("pthread_create timer");
6359 goto exit_notification
;
6361 timer_thread_launched
= true;
6363 /* rotation_thread_data acquires the pipes' read side. */
6364 rotation_thread_handle
= rotation_thread_handle_create(
6365 ust32_channel_rotate_pipe
,
6366 ust64_channel_rotate_pipe
,
6367 kernel_channel_rotate_pipe
,
6368 thread_quit_pipe
[0],
6369 rotation_timer_queue
,
6370 notification_thread_handle
,
6371 ¬ification_thread_ready
);
6372 if (!rotation_thread_handle
) {
6374 ERR("Failed to create rotation thread shared data");
6379 /* Create rotation thread. */
6380 ret
= pthread_create(&rotation_thread
, default_pthread_attr(),
6381 thread_rotation
, rotation_thread_handle
);
6384 PERROR("pthread_create rotation");
6389 rotation_thread_launched
= true;
6391 /* Create thread to manage the client socket */
6392 ret
= pthread_create(&client_thread
, default_pthread_attr(),
6393 thread_manage_clients
, (void *) NULL
);
6396 PERROR("pthread_create clients");
6402 /* Create thread to dispatch registration */
6403 ret
= pthread_create(&dispatch_thread
, default_pthread_attr(),
6404 thread_dispatch_ust_registration
, (void *) NULL
);
6407 PERROR("pthread_create dispatch");
6413 /* Create thread to manage application registration. */
6414 ret
= pthread_create(®_apps_thread
, default_pthread_attr(),
6415 thread_registration_apps
, (void *) NULL
);
6418 PERROR("pthread_create registration");
6424 /* Create thread to manage application socket */
6425 ret
= pthread_create(&apps_thread
, default_pthread_attr(),
6426 thread_manage_apps
, (void *) NULL
);
6429 PERROR("pthread_create apps");
6435 /* Create thread to manage application notify socket */
6436 ret
= pthread_create(&apps_notify_thread
, default_pthread_attr(),
6437 ust_thread_manage_notify
, (void *) NULL
);
6440 PERROR("pthread_create notify");
6443 goto exit_apps_notify
;
6446 /* Create agent registration thread. */
6447 ret
= pthread_create(&agent_reg_thread
, default_pthread_attr(),
6448 agent_thread_manage_registration
, (void *) NULL
);
6451 PERROR("pthread_create agent");
6454 goto exit_agent_reg
;
6457 /* Don't start this thread if kernel tracing is not requested nor root */
6458 if (is_root
&& !config
.no_kernel
) {
6459 /* Create kernel thread to manage kernel event */
6460 ret
= pthread_create(&kernel_thread
, default_pthread_attr(),
6461 thread_manage_kernel
, (void *) NULL
);
6464 PERROR("pthread_create kernel");
6471 /* Create session loading thread. */
6472 ret
= pthread_create(&load_session_thread
, default_pthread_attr(),
6473 thread_load_session
, load_info
);
6476 PERROR("pthread_create load_session_thread");
6479 goto exit_load_session
;
6483 * This is where we start awaiting program completion (e.g. through
6484 * signal that asks threads to teardown).
6487 ret
= pthread_join(load_session_thread
, &status
);
6490 PERROR("pthread_join load_session_thread");
6495 if (is_root
&& !config
.no_kernel
) {
6496 ret
= pthread_join(kernel_thread
, &status
);
6499 PERROR("pthread_join");
6505 ret
= pthread_join(agent_reg_thread
, &status
);
6508 PERROR("pthread_join agent");
6513 ret
= pthread_join(apps_notify_thread
, &status
);
6516 PERROR("pthread_join apps notify");
6521 ret
= pthread_join(apps_thread
, &status
);
6524 PERROR("pthread_join apps");
6529 ret
= pthread_join(reg_apps_thread
, &status
);
6532 PERROR("pthread_join");
6538 * Join dispatch thread after joining reg_apps_thread to ensure
6539 * we don't leak applications in the queue.
6541 ret
= pthread_join(dispatch_thread
, &status
);
6544 PERROR("pthread_join");
6549 ret
= pthread_join(client_thread
, &status
);
6552 PERROR("pthread_join");
6559 sem_destroy(¬ification_thread_ready
);
6560 ret
= pthread_join(health_thread
, &status
);
6563 PERROR("pthread_join health thread");
6570 * Wait for all pending call_rcu work to complete before tearing
6571 * down data structures. call_rcu worker may be trying to
6572 * perform lookups in those structures.
6576 * sessiond_cleanup() is called when no other thread is running, except
6577 * the ht_cleanup thread, which is needed to destroy the hash tables.
6579 rcu_thread_online();
6583 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6584 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6585 * the queue is empty before shutting down the clean-up thread.
6590 * The teardown of the notification system is performed after the
6591 * session daemon's teardown in order to allow it to be notified
6592 * of the active session and channels at the moment of the teardown.
6594 if (notification_thread_handle
) {
6595 if (notification_thread_launched
) {
6596 notification_thread_command_quit(
6597 notification_thread_handle
);
6598 ret
= pthread_join(notification_thread
, &status
);
6601 PERROR("pthread_join notification thread");
6605 notification_thread_handle_destroy(notification_thread_handle
);
6608 if (rotation_thread_handle
) {
6609 if (rotation_thread_launched
) {
6610 ret
= pthread_join(rotation_thread
, &status
);
6613 PERROR("pthread_join rotation thread");
6617 rotation_thread_handle_destroy(rotation_thread_handle
);
6620 if (timer_thread_launched
) {
6621 kill(getpid(), LTTNG_SESSIOND_SIG_EXIT
);
6622 ret
= pthread_join(timer_thread
, &status
);
6625 PERROR("pthread_join timer thread");
6631 * After the rotation and timer thread have quit, we can safely destroy
6632 * the rotation_timer_queue.
6634 destroy_rotate_timer_queue(rotation_timer_queue
);
6636 rcu_thread_offline();
6637 rcu_unregister_thread();
6639 ret
= fini_ht_cleanup_thread(&ht_cleanup_thread
);
6643 lttng_pipe_destroy(ust32_channel_monitor_pipe
);
6644 lttng_pipe_destroy(ust64_channel_monitor_pipe
);
6645 lttng_pipe_destroy(kernel_channel_monitor_pipe
);
6646 lttng_pipe_destroy(ust32_channel_rotate_pipe
);
6647 lttng_pipe_destroy(ust64_channel_rotate_pipe
);
6648 lttng_pipe_destroy(kernel_channel_rotate_pipe
);
6651 health_app_destroy(health_sessiond
);
6652 exit_health_sessiond_cleanup
:
6653 exit_create_run_as_worker_cleanup
:
6656 sessiond_cleanup_lock_file();
6657 sessiond_cleanup_options();
6659 exit_set_signal_handler
: