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>
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
61 #include "kernel-consumer.h"
65 #include "ust-consumer.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
73 #include "load-session-thread.h"
74 #include "notification-thread.h"
75 #include "notification-thread-commands.h"
78 #include "ht-cleanup.h"
79 #include "sessiond-config.h"
81 static const char *help_msg
=
82 #ifdef LTTNG_EMBED_HELP
83 #include <lttng-sessiond.8.h>
90 static pid_t ppid
; /* Parent PID for --sig-parent option */
91 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
92 static int lockfile_fd
= -1;
94 /* Set to 1 when a SIGUSR1 signal is received. */
95 static int recv_child_signal
;
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
101 static struct consumer_data kconsumer_data
= {
102 .type
= LTTNG_CONSUMER_KERNEL
,
105 .channel_monitor_pipe
= -1,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer64_data
= {
112 .type
= LTTNG_CONSUMER64_UST
,
115 .channel_monitor_pipe
= -1,
116 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
117 .lock
= PTHREAD_MUTEX_INITIALIZER
,
118 .cond
= PTHREAD_COND_INITIALIZER
,
119 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
121 static struct consumer_data ustconsumer32_data
= {
122 .type
= LTTNG_CONSUMER32_UST
,
125 .channel_monitor_pipe
= -1,
126 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
127 .lock
= PTHREAD_MUTEX_INITIALIZER
,
128 .cond
= PTHREAD_COND_INITIALIZER
,
129 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
132 /* Command line options */
133 static const struct option long_options
[] = {
134 { "client-sock", required_argument
, 0, 'c' },
135 { "apps-sock", required_argument
, 0, 'a' },
136 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
137 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
138 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
139 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
140 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
142 { "consumerd32-path", required_argument
, 0, '\0' },
143 { "consumerd32-libdir", required_argument
, 0, '\0' },
144 { "consumerd64-path", required_argument
, 0, '\0' },
145 { "consumerd64-libdir", required_argument
, 0, '\0' },
146 { "daemonize", no_argument
, 0, 'd' },
147 { "background", no_argument
, 0, 'b' },
148 { "sig-parent", no_argument
, 0, 'S' },
149 { "help", no_argument
, 0, 'h' },
150 { "group", required_argument
, 0, 'g' },
151 { "version", no_argument
, 0, 'V' },
152 { "quiet", no_argument
, 0, 'q' },
153 { "verbose", no_argument
, 0, 'v' },
154 { "verbose-consumer", no_argument
, 0, '\0' },
155 { "no-kernel", no_argument
, 0, '\0' },
156 { "pidfile", required_argument
, 0, 'p' },
157 { "agent-tcp-port", required_argument
, 0, '\0' },
158 { "config", required_argument
, 0, 'f' },
159 { "load", required_argument
, 0, 'l' },
160 { "kmod-probes", required_argument
, 0, '\0' },
161 { "extra-kmod-probes", required_argument
, 0, '\0' },
165 struct sessiond_config config
;
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options
[] = { "help", "version", "config" };
170 /* Shared between threads */
171 static int dispatch_thread_exit
;
173 /* Sockets and FDs */
174 static int client_sock
= -1;
175 static int apps_sock
= -1;
176 int kernel_tracer_fd
= -1;
177 static int kernel_poll_pipe
[2] = { -1, -1 };
180 * Quit pipe for all threads. This permits a single cancellation point
181 * for all threads when receiving an event on the pipe.
183 static int thread_quit_pipe
[2] = { -1, -1 };
186 * This pipe is used to inform the thread managing application communication
187 * that a command is queued and ready to be processed.
189 static int apps_cmd_pipe
[2] = { -1, -1 };
191 int apps_cmd_notify_pipe
[2] = { -1, -1 };
193 /* Pthread, Mutexes and Semaphores */
194 static pthread_t apps_thread
;
195 static pthread_t apps_notify_thread
;
196 static pthread_t reg_apps_thread
;
197 static pthread_t client_thread
;
198 static pthread_t kernel_thread
;
199 static pthread_t dispatch_thread
;
200 static pthread_t health_thread
;
201 static pthread_t ht_cleanup_thread
;
202 static pthread_t agent_reg_thread
;
203 static pthread_t load_session_thread
;
204 static pthread_t notification_thread
;
207 * UST registration command queue. This queue is tied with a futex and uses a N
208 * wakers / 1 waiter implemented and detailed in futex.c/.h
210 * The thread_registration_apps and thread_dispatch_ust_registration uses this
211 * queue along with the wait/wake scheme. The thread_manage_apps receives down
212 * the line new application socket and monitors it for any I/O error or clean
213 * close that triggers an unregistration of the application.
215 static struct ust_cmd_queue ust_cmd_queue
;
218 * Pointer initialized before thread creation.
220 * This points to the tracing session list containing the session count and a
221 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
222 * MUST NOT be taken if you call a public function in session.c.
224 * The lock is nested inside the structure: session_list_ptr->lock. Please use
225 * session_lock_list and session_unlock_list for lock acquisition.
227 static struct ltt_session_list
*session_list_ptr
;
229 int ust_consumerd64_fd
= -1;
230 int ust_consumerd32_fd
= -1;
232 static const char *module_proc_lttng
= "/proc/lttng";
235 * Consumer daemon state which is changed when spawning it, killing it or in
236 * case of a fatal error.
238 enum consumerd_state
{
239 CONSUMER_STARTED
= 1,
240 CONSUMER_STOPPED
= 2,
245 * This consumer daemon state is used to validate if a client command will be
246 * able to reach the consumer. If not, the client is informed. For instance,
247 * doing a "lttng start" when the consumer state is set to ERROR will return an
248 * error to the client.
250 * The following example shows a possible race condition of this scheme:
252 * consumer thread error happens
254 * client cmd checks state -> still OK
255 * consumer thread exit, sets error
256 * client cmd try to talk to consumer
259 * However, since the consumer is a different daemon, we have no way of making
260 * sure the command will reach it safely even with this state flag. This is why
261 * we consider that up to the state validation during command processing, the
262 * command is safe. After that, we can not guarantee the correctness of the
263 * client request vis-a-vis the consumer.
265 static enum consumerd_state ust_consumerd_state
;
266 static enum consumerd_state kernel_consumerd_state
;
268 /* Set in main() with the current page size. */
271 /* Application health monitoring */
272 struct health_app
*health_sessiond
;
274 /* Am I root or not. */
275 int is_root
; /* Set to 1 if the daemon is running as root */
277 const char * const config_section_name
= "sessiond";
279 /* Load session thread information to operate. */
280 struct load_session_thread_data
*load_info
;
282 /* Notification thread handle. */
283 struct notification_thread_handle
*notification_thread_handle
;
285 /* Global hash tables */
286 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
289 * Whether sessiond is ready for commands/notification channel/health check
291 * NR_LTTNG_SESSIOND_READY must match the number of calls to
292 * sessiond_notify_ready().
294 #define NR_LTTNG_SESSIOND_READY 5
295 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
297 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
299 return (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) ? 1 : 0;
302 /* Notify parents that we are ready for cmd and health check */
304 void sessiond_notify_ready(void)
306 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
308 * Notify parent pid that we are ready to accept command
309 * for client side. This ppid is the one from the
310 * external process that spawned us.
312 if (config
.sig_parent
) {
317 * Notify the parent of the fork() process that we are
320 if (config
.daemonize
|| config
.background
) {
321 kill(child_ppid
, SIGUSR1
);
327 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
334 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
340 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
352 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
354 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
356 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
360 * Init thread quit pipe.
362 * Return -1 on error or 0 if all pipes are created.
364 static int __init_thread_quit_pipe(int *a_pipe
)
370 PERROR("thread quit pipe");
374 for (i
= 0; i
< 2; i
++) {
375 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
386 static int init_thread_quit_pipe(void)
388 return __init_thread_quit_pipe(thread_quit_pipe
);
392 * Stop all threads by closing the thread quit pipe.
394 static void stop_threads(void)
398 /* Stopping all threads */
399 DBG("Terminating all threads");
400 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
402 ERR("write error on thread quit pipe");
405 /* Dispatch thread */
406 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
407 futex_nto1_wake(&ust_cmd_queue
.futex
);
411 * Close every consumer sockets.
413 static void close_consumer_sockets(void)
417 if (kconsumer_data
.err_sock
>= 0) {
418 ret
= close(kconsumer_data
.err_sock
);
420 PERROR("kernel consumer err_sock close");
423 if (ustconsumer32_data
.err_sock
>= 0) {
424 ret
= close(ustconsumer32_data
.err_sock
);
426 PERROR("UST consumerd32 err_sock close");
429 if (ustconsumer64_data
.err_sock
>= 0) {
430 ret
= close(ustconsumer64_data
.err_sock
);
432 PERROR("UST consumerd64 err_sock close");
435 if (kconsumer_data
.cmd_sock
>= 0) {
436 ret
= close(kconsumer_data
.cmd_sock
);
438 PERROR("kernel consumer cmd_sock close");
441 if (ustconsumer32_data
.cmd_sock
>= 0) {
442 ret
= close(ustconsumer32_data
.cmd_sock
);
444 PERROR("UST consumerd32 cmd_sock close");
447 if (ustconsumer64_data
.cmd_sock
>= 0) {
448 ret
= close(ustconsumer64_data
.cmd_sock
);
450 PERROR("UST consumerd64 cmd_sock close");
453 if (kconsumer_data
.channel_monitor_pipe
>= 0) {
454 ret
= close(kconsumer_data
.channel_monitor_pipe
);
456 PERROR("kernel consumer channel monitor pipe close");
459 if (ustconsumer32_data
.channel_monitor_pipe
>= 0) {
460 ret
= close(ustconsumer32_data
.channel_monitor_pipe
);
462 PERROR("UST consumerd32 channel monitor pipe close");
465 if (ustconsumer64_data
.channel_monitor_pipe
>= 0) {
466 ret
= close(ustconsumer64_data
.channel_monitor_pipe
);
468 PERROR("UST consumerd64 channel monitor pipe close");
474 * Wait on consumer process termination.
476 * Need to be called with the consumer data lock held or from a context
477 * ensuring no concurrent access to data (e.g: cleanup).
479 static void wait_consumer(struct consumer_data
*consumer_data
)
484 if (consumer_data
->pid
<= 0) {
488 DBG("Waiting for complete teardown of consumerd (PID: %d)",
490 ret
= waitpid(consumer_data
->pid
, &status
, 0);
492 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
493 } else if (!WIFEXITED(status
)) {
494 ERR("consumerd termination with error: %d",
497 consumer_data
->pid
= 0;
501 * Cleanup the session daemon's data structures.
503 static void sessiond_cleanup(void)
506 struct ltt_session
*sess
, *stmp
;
508 DBG("Cleanup sessiond");
511 * Close the thread quit pipe. It has already done its job,
512 * since we are now called.
514 utils_close_pipe(thread_quit_pipe
);
517 * If config.pid_file_path.value is undefined, the default file will be
518 * wiped when removing the rundir.
520 if (config
.pid_file_path
.value
) {
521 ret
= remove(config
.pid_file_path
.value
);
523 PERROR("remove pidfile %s", config
.pid_file_path
.value
);
527 DBG("Removing sessiond and consumerd content of directory %s",
528 config
.rundir
.value
);
531 DBG("Removing %s", config
.pid_file_path
.value
);
532 (void) unlink(config
.pid_file_path
.value
);
534 DBG("Removing %s", config
.agent_port_file_path
.value
);
535 (void) unlink(config
.agent_port_file_path
.value
);
538 DBG("Removing %s", kconsumer_data
.err_unix_sock_path
);
539 (void) unlink(kconsumer_data
.err_unix_sock_path
);
541 DBG("Removing directory %s", config
.kconsumerd_path
.value
);
542 (void) rmdir(config
.kconsumerd_path
.value
);
544 /* ust consumerd 32 */
545 DBG("Removing %s", config
.consumerd32_err_unix_sock_path
.value
);
546 (void) unlink(config
.consumerd32_err_unix_sock_path
.value
);
548 DBG("Removing directory %s", config
.consumerd32_path
.value
);
549 (void) rmdir(config
.consumerd32_path
.value
);
551 /* ust consumerd 64 */
552 DBG("Removing %s", config
.consumerd64_err_unix_sock_path
.value
);
553 (void) unlink(config
.consumerd64_err_unix_sock_path
.value
);
555 DBG("Removing directory %s", config
.consumerd64_path
.value
);
556 (void) rmdir(config
.consumerd64_path
.value
);
558 DBG("Cleaning up all sessions");
560 /* Destroy session list mutex */
561 if (session_list_ptr
!= NULL
) {
562 pthread_mutex_destroy(&session_list_ptr
->lock
);
564 /* Cleanup ALL session */
565 cds_list_for_each_entry_safe(sess
, stmp
,
566 &session_list_ptr
->head
, list
) {
567 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
571 wait_consumer(&kconsumer_data
);
572 wait_consumer(&ustconsumer64_data
);
573 wait_consumer(&ustconsumer32_data
);
575 DBG("Cleaning up all agent apps");
576 agent_app_ht_clean();
578 DBG("Closing all UST sockets");
579 ust_app_clean_list();
580 buffer_reg_destroy_registries();
582 if (is_root
&& !config
.no_kernel
) {
583 DBG2("Closing kernel fd");
584 if (kernel_tracer_fd
>= 0) {
585 ret
= close(kernel_tracer_fd
);
590 DBG("Unloading kernel modules");
591 modprobe_remove_lttng_all();
595 close_consumer_sockets();
598 load_session_destroy_data(load_info
);
603 * We do NOT rmdir rundir because there are other processes
604 * using it, for instance lttng-relayd, which can start in
605 * parallel with this teardown.
610 * Cleanup the daemon's option data structures.
612 static void sessiond_cleanup_options(void)
614 DBG("Cleaning up options");
616 sessiond_config_fini(&config
);
618 run_as_destroy_worker();
622 * Send data on a unix socket using the liblttsessiondcomm API.
624 * Return lttcomm error code.
626 static int send_unix_sock(int sock
, void *buf
, size_t len
)
628 /* Check valid length */
633 return lttcomm_send_unix_sock(sock
, buf
, len
);
637 * Free memory of a command context structure.
639 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
641 DBG("Clean command context structure");
643 if ((*cmd_ctx
)->llm
) {
644 free((*cmd_ctx
)->llm
);
646 if ((*cmd_ctx
)->lsm
) {
647 free((*cmd_ctx
)->lsm
);
655 * Notify UST applications using the shm mmap futex.
657 static int notify_ust_apps(int active
)
661 DBG("Notifying applications of session daemon state: %d", active
);
663 /* See shm.c for this call implying mmap, shm and futex calls */
664 wait_shm_mmap
= shm_ust_get_mmap(config
.wait_shm_path
.value
, is_root
);
665 if (wait_shm_mmap
== NULL
) {
669 /* Wake waiting process */
670 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
672 /* Apps notified successfully */
680 * Setup the outgoing data buffer for the response (llm) by allocating the
681 * right amount of memory and copying the original information from the lsm
684 * Return 0 on success, negative value on error.
686 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
687 const void *payload_buf
, size_t payload_len
,
688 const void *cmd_header_buf
, size_t cmd_header_len
)
691 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
692 const size_t cmd_header_offset
= header_len
;
693 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
694 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
696 cmd_ctx
->llm
= zmalloc(total_msg_size
);
698 if (cmd_ctx
->llm
== NULL
) {
704 /* Copy common data */
705 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
706 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
707 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
708 cmd_ctx
->llm
->data_size
= payload_len
;
709 cmd_ctx
->lttng_msg_size
= total_msg_size
;
711 /* Copy command header */
712 if (cmd_header_len
) {
713 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
719 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
728 * Version of setup_lttng_msg() without command header.
730 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
731 void *payload_buf
, size_t payload_len
)
733 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
736 * Update the kernel poll set of all channel fd available over all tracing
737 * session. Add the wakeup pipe at the end of the set.
739 static int update_kernel_poll(struct lttng_poll_event
*events
)
742 struct ltt_session
*session
;
743 struct ltt_kernel_channel
*channel
;
745 DBG("Updating kernel poll set");
748 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
749 session_lock(session
);
750 if (session
->kernel_session
== NULL
) {
751 session_unlock(session
);
755 cds_list_for_each_entry(channel
,
756 &session
->kernel_session
->channel_list
.head
, list
) {
757 /* Add channel fd to the kernel poll set */
758 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
760 session_unlock(session
);
763 DBG("Channel fd %d added to kernel set", channel
->fd
);
765 session_unlock(session
);
767 session_unlock_list();
772 session_unlock_list();
777 * Find the channel fd from 'fd' over all tracing session. When found, check
778 * for new channel stream and send those stream fds to the kernel consumer.
780 * Useful for CPU hotplug feature.
782 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
785 struct ltt_session
*session
;
786 struct ltt_kernel_session
*ksess
;
787 struct ltt_kernel_channel
*channel
;
789 DBG("Updating kernel streams for channel fd %d", fd
);
792 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
793 session_lock(session
);
794 if (session
->kernel_session
== NULL
) {
795 session_unlock(session
);
798 ksess
= session
->kernel_session
;
800 cds_list_for_each_entry(channel
,
801 &ksess
->channel_list
.head
, list
) {
802 struct lttng_ht_iter iter
;
803 struct consumer_socket
*socket
;
805 if (channel
->fd
!= fd
) {
808 DBG("Channel found, updating kernel streams");
809 ret
= kernel_open_channel_stream(channel
);
813 /* Update the stream global counter */
814 ksess
->stream_count_global
+= ret
;
817 * Have we already sent fds to the consumer? If yes, it
818 * means that tracing is started so it is safe to send
819 * our updated stream fds.
821 if (ksess
->consumer_fds_sent
!= 1
822 || ksess
->consumer
== NULL
) {
828 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
829 &iter
.iter
, socket
, node
.node
) {
830 pthread_mutex_lock(socket
->lock
);
831 ret
= kernel_consumer_send_channel_stream(socket
,
833 session
->output_traces
? 1 : 0);
834 pthread_mutex_unlock(socket
->lock
);
842 session_unlock(session
);
844 session_unlock_list();
848 session_unlock(session
);
849 session_unlock_list();
854 * For each tracing session, update newly registered apps. The session list
855 * lock MUST be acquired before calling this.
857 static void update_ust_app(int app_sock
)
859 struct ltt_session
*sess
, *stmp
;
861 /* Consumer is in an ERROR state. Stop any application update. */
862 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
863 /* Stop the update process since the consumer is dead. */
867 /* For all tracing session(s) */
868 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
872 if (!sess
->ust_session
) {
877 assert(app_sock
>= 0);
878 app
= ust_app_find_by_sock(app_sock
);
881 * Application can be unregistered before so
882 * this is possible hence simply stopping the
885 DBG3("UST app update failed to find app sock %d",
889 ust_app_global_update(sess
->ust_session
, app
);
893 session_unlock(sess
);
898 * This thread manage event coming from the kernel.
900 * Features supported in this thread:
903 static void *thread_manage_kernel(void *data
)
905 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
906 uint32_t revents
, nb_fd
;
908 struct lttng_poll_event events
;
910 DBG("[thread] Thread manage kernel started");
912 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
915 * This first step of the while is to clean this structure which could free
916 * non NULL pointers so initialize it before the loop.
918 lttng_poll_init(&events
);
920 if (testpoint(sessiond_thread_manage_kernel
)) {
921 goto error_testpoint
;
924 health_code_update();
926 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
927 goto error_testpoint
;
931 health_code_update();
933 if (update_poll_flag
== 1) {
934 /* Clean events object. We are about to populate it again. */
935 lttng_poll_clean(&events
);
937 ret
= sessiond_set_thread_pollset(&events
, 2);
939 goto error_poll_create
;
942 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
947 /* This will add the available kernel channel if any. */
948 ret
= update_kernel_poll(&events
);
952 update_poll_flag
= 0;
955 DBG("Thread kernel polling");
957 /* Poll infinite value of time */
960 ret
= lttng_poll_wait(&events
, -1);
961 DBG("Thread kernel return from poll on %d fds",
962 LTTNG_POLL_GETNB(&events
));
966 * Restart interrupted system call.
968 if (errno
== EINTR
) {
972 } else if (ret
== 0) {
973 /* Should not happen since timeout is infinite */
974 ERR("Return value of poll is 0 with an infinite timeout.\n"
975 "This should not have happened! Continuing...");
981 for (i
= 0; i
< nb_fd
; i
++) {
982 /* Fetch once the poll data */
983 revents
= LTTNG_POLL_GETEV(&events
, i
);
984 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
986 health_code_update();
989 /* No activity for this FD (poll implementation). */
993 /* Thread quit pipe has been closed. Killing thread. */
994 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1000 /* Check for data on kernel pipe */
1001 if (revents
& LPOLLIN
) {
1002 if (pollfd
== kernel_poll_pipe
[0]) {
1003 (void) lttng_read(kernel_poll_pipe
[0],
1006 * Ret value is useless here, if this pipe gets any actions an
1007 * update is required anyway.
1009 update_poll_flag
= 1;
1013 * New CPU detected by the kernel. Adding kernel stream to
1014 * kernel session and updating the kernel consumer
1016 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1022 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1023 update_poll_flag
= 1;
1026 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1034 lttng_poll_clean(&events
);
1037 utils_close_pipe(kernel_poll_pipe
);
1038 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1041 ERR("Health error occurred in %s", __func__
);
1042 WARN("Kernel thread died unexpectedly. "
1043 "Kernel tracing can continue but CPU hotplug is disabled.");
1045 health_unregister(health_sessiond
);
1046 DBG("Kernel thread dying");
1051 * Signal pthread condition of the consumer data that the thread.
1053 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1055 pthread_mutex_lock(&data
->cond_mutex
);
1058 * The state is set before signaling. It can be any value, it's the waiter
1059 * job to correctly interpret this condition variable associated to the
1060 * consumer pthread_cond.
1062 * A value of 0 means that the corresponding thread of the consumer data
1063 * was not started. 1 indicates that the thread has started and is ready
1064 * for action. A negative value means that there was an error during the
1067 data
->consumer_thread_is_ready
= state
;
1068 (void) pthread_cond_signal(&data
->cond
);
1070 pthread_mutex_unlock(&data
->cond_mutex
);
1074 * This thread manage the consumer error sent back to the session daemon.
1076 static void *thread_manage_consumer(void *data
)
1078 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1079 uint32_t revents
, nb_fd
;
1080 enum lttcomm_return_code code
;
1081 struct lttng_poll_event events
;
1082 struct consumer_data
*consumer_data
= data
;
1083 struct consumer_socket
*cmd_socket_wrapper
= NULL
;
1085 DBG("[thread] Manage consumer started");
1087 rcu_register_thread();
1088 rcu_thread_online();
1090 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1092 health_code_update();
1095 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1096 * metadata_sock. Nothing more will be added to this poll set.
1098 ret
= sessiond_set_thread_pollset(&events
, 3);
1104 * The error socket here is already in a listening state which was done
1105 * just before spawning this thread to avoid a race between the consumer
1106 * daemon exec trying to connect and the listen() call.
1108 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1113 health_code_update();
1115 /* Infinite blocking call, waiting for transmission */
1117 health_poll_entry();
1119 if (testpoint(sessiond_thread_manage_consumer
)) {
1123 ret
= lttng_poll_wait(&events
, -1);
1127 * Restart interrupted system call.
1129 if (errno
== EINTR
) {
1137 for (i
= 0; i
< nb_fd
; i
++) {
1138 /* Fetch once the poll data */
1139 revents
= LTTNG_POLL_GETEV(&events
, i
);
1140 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1142 health_code_update();
1145 /* No activity for this FD (poll implementation). */
1149 /* Thread quit pipe has been closed. Killing thread. */
1150 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1156 /* Event on the registration socket */
1157 if (pollfd
== consumer_data
->err_sock
) {
1158 if (revents
& LPOLLIN
) {
1160 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1161 ERR("consumer err socket poll error");
1164 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1170 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1176 * Set the CLOEXEC flag. Return code is useless because either way, the
1179 (void) utils_set_fd_cloexec(sock
);
1181 health_code_update();
1183 DBG2("Receiving code from consumer err_sock");
1185 /* Getting status code from kconsumerd */
1186 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1187 sizeof(enum lttcomm_return_code
));
1192 health_code_update();
1193 if (code
!= LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1194 ERR("consumer error when waiting for SOCK_READY : %s",
1195 lttcomm_get_readable_code(-code
));
1199 /* Connect both command and metadata sockets. */
1200 consumer_data
->cmd_sock
=
1201 lttcomm_connect_unix_sock(
1202 consumer_data
->cmd_unix_sock_path
);
1203 consumer_data
->metadata_fd
=
1204 lttcomm_connect_unix_sock(
1205 consumer_data
->cmd_unix_sock_path
);
1206 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
1207 PERROR("consumer connect cmd socket");
1208 /* On error, signal condition and quit. */
1209 signal_consumer_condition(consumer_data
, -1);
1213 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1215 /* Create metadata socket lock. */
1216 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1217 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1218 PERROR("zmalloc pthread mutex");
1221 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1223 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1224 DBG("Consumer metadata socket ready (fd: %d)",
1225 consumer_data
->metadata_fd
);
1228 * Remove the consumerd error sock since we've established a connection.
1230 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1235 /* Add new accepted error socket. */
1236 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1241 /* Add metadata socket that is successfully connected. */
1242 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1243 LPOLLIN
| LPOLLRDHUP
);
1248 health_code_update();
1251 * Transfer the write-end of the channel monitoring pipe to the
1252 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1254 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
1255 if (!cmd_socket_wrapper
) {
1258 cmd_socket_wrapper
->lock
= &consumer_data
->lock
;
1260 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
1261 consumer_data
->channel_monitor_pipe
);
1265 /* Discard the socket wrapper as it is no longer needed. */
1266 consumer_destroy_socket(cmd_socket_wrapper
);
1267 cmd_socket_wrapper
= NULL
;
1269 /* The thread is completely initialized, signal that it is ready. */
1270 signal_consumer_condition(consumer_data
, 1);
1272 /* Infinite blocking call, waiting for transmission */
1275 health_code_update();
1277 /* Exit the thread because the thread quit pipe has been triggered. */
1279 /* Not a health error. */
1284 health_poll_entry();
1285 ret
= lttng_poll_wait(&events
, -1);
1289 * Restart interrupted system call.
1291 if (errno
== EINTR
) {
1299 for (i
= 0; i
< nb_fd
; i
++) {
1300 /* Fetch once the poll data */
1301 revents
= LTTNG_POLL_GETEV(&events
, i
);
1302 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1304 health_code_update();
1307 /* No activity for this FD (poll implementation). */
1312 * Thread quit pipe has been triggered, flag that we should stop
1313 * but continue the current loop to handle potential data from
1316 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1318 if (pollfd
== sock
) {
1319 /* Event on the consumerd socket */
1320 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1321 && !(revents
& LPOLLIN
)) {
1322 ERR("consumer err socket second poll error");
1325 health_code_update();
1326 /* Wait for any kconsumerd error */
1327 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1328 sizeof(enum lttcomm_return_code
));
1330 ERR("consumer closed the command socket");
1334 ERR("consumer return code : %s",
1335 lttcomm_get_readable_code(-code
));
1338 } else if (pollfd
== consumer_data
->metadata_fd
) {
1339 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1340 && !(revents
& LPOLLIN
)) {
1341 ERR("consumer err metadata socket second poll error");
1344 /* UST metadata requests */
1345 ret
= ust_consumer_metadata_request(
1346 &consumer_data
->metadata_sock
);
1348 ERR("Handling metadata request");
1352 /* No need for an else branch all FDs are tested prior. */
1354 health_code_update();
1360 * We lock here because we are about to close the sockets and some other
1361 * thread might be using them so get exclusive access which will abort all
1362 * other consumer command by other threads.
1364 pthread_mutex_lock(&consumer_data
->lock
);
1366 /* Immediately set the consumerd state to stopped */
1367 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1368 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1369 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1370 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1371 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1373 /* Code flow error... */
1377 if (consumer_data
->err_sock
>= 0) {
1378 ret
= close(consumer_data
->err_sock
);
1382 consumer_data
->err_sock
= -1;
1384 if (consumer_data
->cmd_sock
>= 0) {
1385 ret
= close(consumer_data
->cmd_sock
);
1389 consumer_data
->cmd_sock
= -1;
1391 if (consumer_data
->metadata_sock
.fd_ptr
&&
1392 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1393 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1405 unlink(consumer_data
->err_unix_sock_path
);
1406 unlink(consumer_data
->cmd_unix_sock_path
);
1407 pthread_mutex_unlock(&consumer_data
->lock
);
1409 /* Cleanup metadata socket mutex. */
1410 if (consumer_data
->metadata_sock
.lock
) {
1411 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1412 free(consumer_data
->metadata_sock
.lock
);
1414 lttng_poll_clean(&events
);
1416 if (cmd_socket_wrapper
) {
1417 consumer_destroy_socket(cmd_socket_wrapper
);
1422 ERR("Health error occurred in %s", __func__
);
1424 health_unregister(health_sessiond
);
1425 DBG("consumer thread cleanup completed");
1427 rcu_thread_offline();
1428 rcu_unregister_thread();
1434 * This thread manage application communication.
1436 static void *thread_manage_apps(void *data
)
1438 int i
, ret
, pollfd
, err
= -1;
1440 uint32_t revents
, nb_fd
;
1441 struct lttng_poll_event events
;
1443 DBG("[thread] Manage application started");
1445 rcu_register_thread();
1446 rcu_thread_online();
1448 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1450 if (testpoint(sessiond_thread_manage_apps
)) {
1451 goto error_testpoint
;
1454 health_code_update();
1456 ret
= sessiond_set_thread_pollset(&events
, 2);
1458 goto error_poll_create
;
1461 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1466 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1470 health_code_update();
1473 DBG("Apps thread polling");
1475 /* Inifinite blocking call, waiting for transmission */
1477 health_poll_entry();
1478 ret
= lttng_poll_wait(&events
, -1);
1479 DBG("Apps thread return from poll on %d fds",
1480 LTTNG_POLL_GETNB(&events
));
1484 * Restart interrupted system call.
1486 if (errno
== EINTR
) {
1494 for (i
= 0; i
< nb_fd
; i
++) {
1495 /* Fetch once the poll data */
1496 revents
= LTTNG_POLL_GETEV(&events
, i
);
1497 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1499 health_code_update();
1502 /* No activity for this FD (poll implementation). */
1506 /* Thread quit pipe has been closed. Killing thread. */
1507 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1513 /* Inspect the apps cmd pipe */
1514 if (pollfd
== apps_cmd_pipe
[0]) {
1515 if (revents
& LPOLLIN
) {
1519 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1520 if (size_ret
< sizeof(sock
)) {
1521 PERROR("read apps cmd pipe");
1525 health_code_update();
1528 * Since this is a command socket (write then read),
1529 * we only monitor the error events of the socket.
1531 ret
= lttng_poll_add(&events
, sock
,
1532 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1537 DBG("Apps with sock %d added to poll set", sock
);
1538 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1539 ERR("Apps command pipe error");
1542 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1547 * At this point, we know that a registered application made
1548 * the event at poll_wait.
1550 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1551 /* Removing from the poll set */
1552 ret
= lttng_poll_del(&events
, pollfd
);
1557 /* Socket closed on remote end. */
1558 ust_app_unregister(pollfd
);
1560 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1565 health_code_update();
1571 lttng_poll_clean(&events
);
1574 utils_close_pipe(apps_cmd_pipe
);
1575 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1578 * We don't clean the UST app hash table here since already registered
1579 * applications can still be controlled so let them be until the session
1580 * daemon dies or the applications stop.
1585 ERR("Health error occurred in %s", __func__
);
1587 health_unregister(health_sessiond
);
1588 DBG("Application communication apps thread cleanup complete");
1589 rcu_thread_offline();
1590 rcu_unregister_thread();
1595 * Send a socket to a thread This is called from the dispatch UST registration
1596 * thread once all sockets are set for the application.
1598 * The sock value can be invalid, we don't really care, the thread will handle
1599 * it and make the necessary cleanup if so.
1601 * On success, return 0 else a negative value being the errno message of the
1604 static int send_socket_to_thread(int fd
, int sock
)
1609 * It's possible that the FD is set as invalid with -1 concurrently just
1610 * before calling this function being a shutdown state of the thread.
1617 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1618 if (ret
< sizeof(sock
)) {
1619 PERROR("write apps pipe %d", fd
);
1626 /* All good. Don't send back the write positive ret value. */
1633 * Sanitize the wait queue of the dispatch registration thread meaning removing
1634 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1635 * notify socket is never received.
1637 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1639 int ret
, nb_fd
= 0, i
;
1640 unsigned int fd_added
= 0;
1641 struct lttng_poll_event events
;
1642 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1646 lttng_poll_init(&events
);
1648 /* Just skip everything for an empty queue. */
1649 if (!wait_queue
->count
) {
1653 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1658 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1659 &wait_queue
->head
, head
) {
1660 assert(wait_node
->app
);
1661 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1662 LPOLLHUP
| LPOLLERR
);
1675 * Poll but don't block so we can quickly identify the faulty events and
1676 * clean them afterwards from the wait queue.
1678 ret
= lttng_poll_wait(&events
, 0);
1684 for (i
= 0; i
< nb_fd
; i
++) {
1685 /* Get faulty FD. */
1686 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1687 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1690 /* No activity for this FD (poll implementation). */
1694 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1695 &wait_queue
->head
, head
) {
1696 if (pollfd
== wait_node
->app
->sock
&&
1697 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1698 cds_list_del(&wait_node
->head
);
1699 wait_queue
->count
--;
1700 ust_app_destroy(wait_node
->app
);
1703 * Silence warning of use-after-free in
1704 * cds_list_for_each_entry_safe which uses
1705 * __typeof__(*wait_node).
1710 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1717 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1721 lttng_poll_clean(&events
);
1725 lttng_poll_clean(&events
);
1727 ERR("Unable to sanitize wait queue");
1732 * Dispatch request from the registration threads to the application
1733 * communication thread.
1735 static void *thread_dispatch_ust_registration(void *data
)
1738 struct cds_wfcq_node
*node
;
1739 struct ust_command
*ust_cmd
= NULL
;
1740 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1741 struct ust_reg_wait_queue wait_queue
= {
1745 rcu_register_thread();
1747 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1749 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1750 goto error_testpoint
;
1753 health_code_update();
1755 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1757 DBG("[thread] Dispatch UST command started");
1760 health_code_update();
1762 /* Atomically prepare the queue futex */
1763 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1765 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1770 struct ust_app
*app
= NULL
;
1774 * Make sure we don't have node(s) that have hung up before receiving
1775 * the notify socket. This is to clean the list in order to avoid
1776 * memory leaks from notify socket that are never seen.
1778 sanitize_wait_queue(&wait_queue
);
1780 health_code_update();
1781 /* Dequeue command for registration */
1782 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1784 DBG("Woken up but nothing in the UST command queue");
1785 /* Continue thread execution */
1789 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1791 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1792 " gid:%d sock:%d name:%s (version %d.%d)",
1793 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1794 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1795 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1796 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1798 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1799 wait_node
= zmalloc(sizeof(*wait_node
));
1801 PERROR("zmalloc wait_node dispatch");
1802 ret
= close(ust_cmd
->sock
);
1804 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1806 lttng_fd_put(LTTNG_FD_APPS
, 1);
1810 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1812 /* Create application object if socket is CMD. */
1813 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1815 if (!wait_node
->app
) {
1816 ret
= close(ust_cmd
->sock
);
1818 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1820 lttng_fd_put(LTTNG_FD_APPS
, 1);
1826 * Add application to the wait queue so we can set the notify
1827 * socket before putting this object in the global ht.
1829 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1834 * We have to continue here since we don't have the notify
1835 * socket and the application MUST be added to the hash table
1836 * only at that moment.
1841 * Look for the application in the local wait queue and set the
1842 * notify socket if found.
1844 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1845 &wait_queue
.head
, head
) {
1846 health_code_update();
1847 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1848 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1849 cds_list_del(&wait_node
->head
);
1851 app
= wait_node
->app
;
1853 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1859 * With no application at this stage the received socket is
1860 * basically useless so close it before we free the cmd data
1861 * structure for good.
1864 ret
= close(ust_cmd
->sock
);
1866 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1868 lttng_fd_put(LTTNG_FD_APPS
, 1);
1875 * @session_lock_list
1877 * Lock the global session list so from the register up to the
1878 * registration done message, no thread can see the application
1879 * and change its state.
1881 session_lock_list();
1885 * Add application to the global hash table. This needs to be
1886 * done before the update to the UST registry can locate the
1891 /* Set app version. This call will print an error if needed. */
1892 (void) ust_app_version(app
);
1894 /* Send notify socket through the notify pipe. */
1895 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1899 session_unlock_list();
1901 * No notify thread, stop the UST tracing. However, this is
1902 * not an internal error of the this thread thus setting
1903 * the health error code to a normal exit.
1910 * Update newly registered application with the tracing
1911 * registry info already enabled information.
1913 update_ust_app(app
->sock
);
1916 * Don't care about return value. Let the manage apps threads
1917 * handle app unregistration upon socket close.
1919 (void) ust_app_register_done(app
);
1922 * Even if the application socket has been closed, send the app
1923 * to the thread and unregistration will take place at that
1926 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1929 session_unlock_list();
1931 * No apps. thread, stop the UST tracing. However, this is
1932 * not an internal error of the this thread thus setting
1933 * the health error code to a normal exit.
1940 session_unlock_list();
1942 } while (node
!= NULL
);
1944 health_poll_entry();
1945 /* Futex wait on queue. Blocking call on futex() */
1946 futex_nto1_wait(&ust_cmd_queue
.futex
);
1949 /* Normal exit, no error */
1953 /* Clean up wait queue. */
1954 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1955 &wait_queue
.head
, head
) {
1956 cds_list_del(&wait_node
->head
);
1961 /* Empty command queue. */
1963 /* Dequeue command for registration */
1964 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1968 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1969 ret
= close(ust_cmd
->sock
);
1971 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
1973 lttng_fd_put(LTTNG_FD_APPS
, 1);
1978 DBG("Dispatch thread dying");
1981 ERR("Health error occurred in %s", __func__
);
1983 health_unregister(health_sessiond
);
1984 rcu_unregister_thread();
1989 * This thread manage application registration.
1991 static void *thread_registration_apps(void *data
)
1993 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1994 uint32_t revents
, nb_fd
;
1995 struct lttng_poll_event events
;
1997 * Get allocated in this thread, enqueued to a global queue, dequeued and
1998 * freed in the manage apps thread.
2000 struct ust_command
*ust_cmd
= NULL
;
2002 DBG("[thread] Manage application registration started");
2004 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2006 if (testpoint(sessiond_thread_registration_apps
)) {
2007 goto error_testpoint
;
2010 ret
= lttcomm_listen_unix_sock(apps_sock
);
2016 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2017 * more will be added to this poll set.
2019 ret
= sessiond_set_thread_pollset(&events
, 2);
2021 goto error_create_poll
;
2024 /* Add the application registration socket */
2025 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2027 goto error_poll_add
;
2030 /* Notify all applications to register */
2031 ret
= notify_ust_apps(1);
2033 ERR("Failed to notify applications or create the wait shared memory.\n"
2034 "Execution continues but there might be problem for already\n"
2035 "running applications that wishes to register.");
2039 DBG("Accepting application registration");
2041 /* Inifinite blocking call, waiting for transmission */
2043 health_poll_entry();
2044 ret
= lttng_poll_wait(&events
, -1);
2048 * Restart interrupted system call.
2050 if (errno
== EINTR
) {
2058 for (i
= 0; i
< nb_fd
; i
++) {
2059 health_code_update();
2061 /* Fetch once the poll data */
2062 revents
= LTTNG_POLL_GETEV(&events
, i
);
2063 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2066 /* No activity for this FD (poll implementation). */
2070 /* Thread quit pipe has been closed. Killing thread. */
2071 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2077 /* Event on the registration socket */
2078 if (pollfd
== apps_sock
) {
2079 if (revents
& LPOLLIN
) {
2080 sock
= lttcomm_accept_unix_sock(apps_sock
);
2086 * Set socket timeout for both receiving and ending.
2087 * app_socket_timeout is in seconds, whereas
2088 * lttcomm_setsockopt_rcv_timeout and
2089 * lttcomm_setsockopt_snd_timeout expect msec as
2092 if (config
.app_socket_timeout
>= 0) {
2093 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2094 config
.app_socket_timeout
* 1000);
2095 (void) lttcomm_setsockopt_snd_timeout(sock
,
2096 config
.app_socket_timeout
* 1000);
2100 * Set the CLOEXEC flag. Return code is useless because
2101 * either way, the show must go on.
2103 (void) utils_set_fd_cloexec(sock
);
2105 /* Create UST registration command for enqueuing */
2106 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2107 if (ust_cmd
== NULL
) {
2108 PERROR("ust command zmalloc");
2117 * Using message-based transmissions to ensure we don't
2118 * have to deal with partially received messages.
2120 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2122 ERR("Exhausted file descriptors allowed for applications.");
2132 health_code_update();
2133 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2136 /* Close socket of the application. */
2141 lttng_fd_put(LTTNG_FD_APPS
, 1);
2145 health_code_update();
2147 ust_cmd
->sock
= sock
;
2150 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2151 " gid:%d sock:%d name:%s (version %d.%d)",
2152 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2153 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2154 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2155 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2158 * Lock free enqueue the registration request. The red pill
2159 * has been taken! This apps will be part of the *system*.
2161 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2164 * Wake the registration queue futex. Implicit memory
2165 * barrier with the exchange in cds_wfcq_enqueue.
2167 futex_nto1_wake(&ust_cmd_queue
.futex
);
2168 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2169 ERR("Register apps socket poll error");
2172 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2181 /* Notify that the registration thread is gone */
2184 if (apps_sock
>= 0) {
2185 ret
= close(apps_sock
);
2195 lttng_fd_put(LTTNG_FD_APPS
, 1);
2197 unlink(config
.apps_unix_sock_path
.value
);
2200 lttng_poll_clean(&events
);
2204 DBG("UST Registration thread cleanup complete");
2207 ERR("Health error occurred in %s", __func__
);
2209 health_unregister(health_sessiond
);
2215 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2216 * exec or it will fails.
2218 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2221 struct timespec timeout
;
2224 * Make sure we set the readiness flag to 0 because we are NOT ready.
2225 * This access to consumer_thread_is_ready does not need to be
2226 * protected by consumer_data.cond_mutex (yet) since the consumer
2227 * management thread has not been started at this point.
2229 consumer_data
->consumer_thread_is_ready
= 0;
2231 /* Setup pthread condition */
2232 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2235 PERROR("pthread_condattr_init consumer data");
2240 * Set the monotonic clock in order to make sure we DO NOT jump in time
2241 * between the clock_gettime() call and the timedwait call. See bug #324
2242 * for a more details and how we noticed it.
2244 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2247 PERROR("pthread_condattr_setclock consumer data");
2251 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2254 PERROR("pthread_cond_init consumer data");
2258 ret
= pthread_create(&consumer_data
->thread
, default_pthread_attr(),
2259 thread_manage_consumer
, consumer_data
);
2262 PERROR("pthread_create consumer");
2267 /* We are about to wait on a pthread condition */
2268 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2270 /* Get time for sem_timedwait absolute timeout */
2271 clock_ret
= lttng_clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2273 * Set the timeout for the condition timed wait even if the clock gettime
2274 * call fails since we might loop on that call and we want to avoid to
2275 * increment the timeout too many times.
2277 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2280 * The following loop COULD be skipped in some conditions so this is why we
2281 * set ret to 0 in order to make sure at least one round of the loop is
2287 * Loop until the condition is reached or when a timeout is reached. Note
2288 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2289 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2290 * possible. This loop does not take any chances and works with both of
2293 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2294 if (clock_ret
< 0) {
2295 PERROR("clock_gettime spawn consumer");
2296 /* Infinite wait for the consumerd thread to be ready */
2297 ret
= pthread_cond_wait(&consumer_data
->cond
,
2298 &consumer_data
->cond_mutex
);
2300 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2301 &consumer_data
->cond_mutex
, &timeout
);
2305 /* Release the pthread condition */
2306 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2310 if (ret
== ETIMEDOUT
) {
2314 * Call has timed out so we kill the kconsumerd_thread and return
2317 ERR("Condition timed out. The consumer thread was never ready."
2319 pth_ret
= pthread_cancel(consumer_data
->thread
);
2321 PERROR("pthread_cancel consumer thread");
2324 PERROR("pthread_cond_wait failed consumer thread");
2326 /* Caller is expecting a negative value on failure. */
2331 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2332 if (consumer_data
->pid
== 0) {
2333 ERR("Consumerd did not start");
2334 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2337 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2346 * Join consumer thread
2348 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2352 /* Consumer pid must be a real one. */
2353 if (consumer_data
->pid
> 0) {
2355 ret
= kill(consumer_data
->pid
, SIGTERM
);
2357 PERROR("Error killing consumer daemon");
2360 return pthread_join(consumer_data
->thread
, &status
);
2367 * Fork and exec a consumer daemon (consumerd).
2369 * Return pid if successful else -1.
2371 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2375 const char *consumer_to_use
;
2376 const char *verbosity
;
2379 DBG("Spawning consumerd");
2386 if (config
.verbose_consumer
) {
2387 verbosity
= "--verbose";
2388 } else if (lttng_opt_quiet
) {
2389 verbosity
= "--quiet";
2394 switch (consumer_data
->type
) {
2395 case LTTNG_CONSUMER_KERNEL
:
2397 * Find out which consumerd to execute. We will first try the
2398 * 64-bit path, then the sessiond's installation directory, and
2399 * fallback on the 32-bit one,
2401 DBG3("Looking for a kernel consumer at these locations:");
2402 DBG3(" 1) %s", config
.consumerd64_bin_path
.value
? : "NULL");
2403 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, DEFAULT_CONSUMERD_FILE
);
2404 DBG3(" 3) %s", config
.consumerd32_bin_path
.value
? : "NULL");
2405 if (stat(config
.consumerd64_bin_path
.value
, &st
) == 0) {
2406 DBG3("Found location #1");
2407 consumer_to_use
= config
.consumerd64_bin_path
.value
;
2408 } else if (stat(INSTALL_BIN_PATH
"/" DEFAULT_CONSUMERD_FILE
, &st
) == 0) {
2409 DBG3("Found location #2");
2410 consumer_to_use
= INSTALL_BIN_PATH
"/" DEFAULT_CONSUMERD_FILE
;
2411 } else if (stat(config
.consumerd32_bin_path
.value
, &st
) == 0) {
2412 DBG3("Found location #3");
2413 consumer_to_use
= config
.consumerd32_bin_path
.value
;
2415 DBG("Could not find any valid consumerd executable");
2419 DBG("Using kernel consumer at: %s", consumer_to_use
);
2420 (void) execl(consumer_to_use
,
2421 "lttng-consumerd", verbosity
, "-k",
2422 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2423 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2424 "--group", config
.tracing_group_name
.value
,
2427 case LTTNG_CONSUMER64_UST
:
2429 char *tmpnew
= NULL
;
2431 if (config
.consumerd64_lib_dir
.value
) {
2435 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2439 tmplen
= strlen(config
.consumerd64_lib_dir
.value
) + 1 /* : */ + strlen(tmp
);
2440 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2445 strcat(tmpnew
, config
.consumerd64_lib_dir
.value
);
2446 if (tmp
[0] != '\0') {
2447 strcat(tmpnew
, ":");
2448 strcat(tmpnew
, tmp
);
2450 ret
= setenv("LD_LIBRARY_PATH", tmpnew
, 1);
2457 DBG("Using 64-bit UST consumer at: %s", config
.consumerd64_bin_path
.value
);
2458 (void) execl(config
.consumerd64_bin_path
.value
, "lttng-consumerd", verbosity
, "-u",
2459 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2460 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2461 "--group", config
.tracing_group_name
.value
,
2466 case LTTNG_CONSUMER32_UST
:
2468 char *tmpnew
= NULL
;
2470 if (config
.consumerd32_lib_dir
.value
) {
2474 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2478 tmplen
= strlen(config
.consumerd32_lib_dir
.value
) + 1 /* : */ + strlen(tmp
);
2479 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2484 strcat(tmpnew
, config
.consumerd32_lib_dir
.value
);
2485 if (tmp
[0] != '\0') {
2486 strcat(tmpnew
, ":");
2487 strcat(tmpnew
, tmp
);
2489 ret
= setenv("LD_LIBRARY_PATH", tmpnew
, 1);
2496 DBG("Using 32-bit UST consumer at: %s", config
.consumerd32_bin_path
.value
);
2497 (void) execl(config
.consumerd32_bin_path
.value
, "lttng-consumerd", verbosity
, "-u",
2498 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2499 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2500 "--group", config
.tracing_group_name
.value
,
2506 ERR("unknown consumer type");
2510 PERROR("Consumer execl()");
2512 /* Reaching this point, we got a failure on our execl(). */
2514 } else if (pid
> 0) {
2517 PERROR("start consumer fork");
2525 * Spawn the consumerd daemon and session daemon thread.
2527 static int start_consumerd(struct consumer_data
*consumer_data
)
2532 * Set the listen() state on the socket since there is a possible race
2533 * between the exec() of the consumer daemon and this call if place in the
2534 * consumer thread. See bug #366 for more details.
2536 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2541 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2542 if (consumer_data
->pid
!= 0) {
2543 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2547 ret
= spawn_consumerd(consumer_data
);
2549 ERR("Spawning consumerd failed");
2550 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2554 /* Setting up the consumer_data pid */
2555 consumer_data
->pid
= ret
;
2556 DBG2("Consumer pid %d", consumer_data
->pid
);
2557 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2559 DBG2("Spawning consumer control thread");
2560 ret
= spawn_consumer_thread(consumer_data
);
2562 ERR("Fatal error spawning consumer control thread");
2570 /* Cleanup already created sockets on error. */
2571 if (consumer_data
->err_sock
>= 0) {
2574 err
= close(consumer_data
->err_sock
);
2576 PERROR("close consumer data error socket");
2583 * Setup necessary data for kernel tracer action.
2585 static int init_kernel_tracer(void)
2589 /* Modprobe lttng kernel modules */
2590 ret
= modprobe_lttng_control();
2595 /* Open debugfs lttng */
2596 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2597 if (kernel_tracer_fd
< 0) {
2598 DBG("Failed to open %s", module_proc_lttng
);
2602 /* Validate kernel version */
2603 ret
= kernel_validate_version(kernel_tracer_fd
);
2608 ret
= modprobe_lttng_data();
2613 ret
= kernel_supports_ring_buffer_snapshot_sample_positions(
2620 WARN("Kernel tracer does not support buffer monitoring. "
2621 "The monitoring timer of channels in the kernel domain "
2622 "will be set to 0 (disabled).");
2625 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2629 modprobe_remove_lttng_control();
2630 ret
= close(kernel_tracer_fd
);
2634 kernel_tracer_fd
= -1;
2635 return LTTNG_ERR_KERN_VERSION
;
2638 ret
= close(kernel_tracer_fd
);
2644 modprobe_remove_lttng_control();
2647 WARN("No kernel tracer available");
2648 kernel_tracer_fd
= -1;
2650 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2652 return LTTNG_ERR_KERN_NA
;
2658 * Copy consumer output from the tracing session to the domain session. The
2659 * function also applies the right modification on a per domain basis for the
2660 * trace files destination directory.
2662 * Should *NOT* be called with RCU read-side lock held.
2664 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2667 const char *dir_name
;
2668 struct consumer_output
*consumer
;
2671 assert(session
->consumer
);
2674 case LTTNG_DOMAIN_KERNEL
:
2675 DBG3("Copying tracing session consumer output in kernel session");
2677 * XXX: We should audit the session creation and what this function
2678 * does "extra" in order to avoid a destroy since this function is used
2679 * in the domain session creation (kernel and ust) only. Same for UST
2682 if (session
->kernel_session
->consumer
) {
2683 consumer_output_put(session
->kernel_session
->consumer
);
2685 session
->kernel_session
->consumer
=
2686 consumer_copy_output(session
->consumer
);
2687 /* Ease our life a bit for the next part */
2688 consumer
= session
->kernel_session
->consumer
;
2689 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2691 case LTTNG_DOMAIN_JUL
:
2692 case LTTNG_DOMAIN_LOG4J
:
2693 case LTTNG_DOMAIN_PYTHON
:
2694 case LTTNG_DOMAIN_UST
:
2695 DBG3("Copying tracing session consumer output in UST session");
2696 if (session
->ust_session
->consumer
) {
2697 consumer_output_put(session
->ust_session
->consumer
);
2699 session
->ust_session
->consumer
=
2700 consumer_copy_output(session
->consumer
);
2701 /* Ease our life a bit for the next part */
2702 consumer
= session
->ust_session
->consumer
;
2703 dir_name
= DEFAULT_UST_TRACE_DIR
;
2706 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2710 /* Append correct directory to subdir */
2711 strncat(consumer
->subdir
, dir_name
,
2712 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2713 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2722 * Create an UST session and add it to the session ust list.
2724 * Should *NOT* be called with RCU read-side lock held.
2726 static int create_ust_session(struct ltt_session
*session
,
2727 struct lttng_domain
*domain
)
2730 struct ltt_ust_session
*lus
= NULL
;
2734 assert(session
->consumer
);
2736 switch (domain
->type
) {
2737 case LTTNG_DOMAIN_JUL
:
2738 case LTTNG_DOMAIN_LOG4J
:
2739 case LTTNG_DOMAIN_PYTHON
:
2740 case LTTNG_DOMAIN_UST
:
2743 ERR("Unknown UST domain on create session %d", domain
->type
);
2744 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2748 DBG("Creating UST session");
2750 lus
= trace_ust_create_session(session
->id
);
2752 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2756 lus
->uid
= session
->uid
;
2757 lus
->gid
= session
->gid
;
2758 lus
->output_traces
= session
->output_traces
;
2759 lus
->snapshot_mode
= session
->snapshot_mode
;
2760 lus
->live_timer_interval
= session
->live_timer
;
2761 session
->ust_session
= lus
;
2762 if (session
->shm_path
[0]) {
2763 strncpy(lus
->root_shm_path
, session
->shm_path
,
2764 sizeof(lus
->root_shm_path
));
2765 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2766 strncpy(lus
->shm_path
, session
->shm_path
,
2767 sizeof(lus
->shm_path
));
2768 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2769 strncat(lus
->shm_path
, "/ust",
2770 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2772 /* Copy session output to the newly created UST session */
2773 ret
= copy_session_consumer(domain
->type
, session
);
2774 if (ret
!= LTTNG_OK
) {
2782 session
->ust_session
= NULL
;
2787 * Create a kernel tracer session then create the default channel.
2789 static int create_kernel_session(struct ltt_session
*session
)
2793 DBG("Creating kernel session");
2795 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2797 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2801 /* Code flow safety */
2802 assert(session
->kernel_session
);
2804 /* Copy session output to the newly created Kernel session */
2805 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2806 if (ret
!= LTTNG_OK
) {
2810 /* Create directory(ies) on local filesystem. */
2811 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2812 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2813 ret
= run_as_mkdir_recursive(
2814 session
->kernel_session
->consumer
->dst
.trace_path
,
2815 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2817 if (errno
!= EEXIST
) {
2818 ERR("Trace directory creation error");
2824 session
->kernel_session
->uid
= session
->uid
;
2825 session
->kernel_session
->gid
= session
->gid
;
2826 session
->kernel_session
->output_traces
= session
->output_traces
;
2827 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2832 trace_kernel_destroy_session(session
->kernel_session
);
2833 session
->kernel_session
= NULL
;
2838 * Count number of session permitted by uid/gid.
2840 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2843 struct ltt_session
*session
;
2845 DBG("Counting number of available session for UID %d GID %d",
2847 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2849 * Only list the sessions the user can control.
2851 if (!session_access_ok(session
, uid
, gid
)) {
2860 * Process the command requested by the lttng client within the command
2861 * context structure. This function make sure that the return structure (llm)
2862 * is set and ready for transmission before returning.
2864 * Return any error encountered or 0 for success.
2866 * "sock" is only used for special-case var. len data.
2868 * Should *NOT* be called with RCU read-side lock held.
2870 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2874 int need_tracing_session
= 1;
2877 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2879 assert(!rcu_read_ongoing());
2883 switch (cmd_ctx
->lsm
->cmd_type
) {
2884 case LTTNG_CREATE_SESSION
:
2885 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2886 case LTTNG_CREATE_SESSION_LIVE
:
2887 case LTTNG_DESTROY_SESSION
:
2888 case LTTNG_LIST_SESSIONS
:
2889 case LTTNG_LIST_DOMAINS
:
2890 case LTTNG_START_TRACE
:
2891 case LTTNG_STOP_TRACE
:
2892 case LTTNG_DATA_PENDING
:
2893 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2894 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2895 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2896 case LTTNG_SNAPSHOT_RECORD
:
2897 case LTTNG_SAVE_SESSION
:
2898 case LTTNG_SET_SESSION_SHM_PATH
:
2899 case LTTNG_REGENERATE_METADATA
:
2900 case LTTNG_REGENERATE_STATEDUMP
:
2901 case LTTNG_REGISTER_TRIGGER
:
2902 case LTTNG_UNREGISTER_TRIGGER
:
2909 if (config
.no_kernel
&& need_domain
2910 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2912 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2914 ret
= LTTNG_ERR_KERN_NA
;
2919 /* Deny register consumer if we already have a spawned consumer. */
2920 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2921 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2922 if (kconsumer_data
.pid
> 0) {
2923 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2924 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2927 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2931 * Check for command that don't needs to allocate a returned payload. We do
2932 * this here so we don't have to make the call for no payload at each
2935 switch(cmd_ctx
->lsm
->cmd_type
) {
2936 case LTTNG_LIST_SESSIONS
:
2937 case LTTNG_LIST_TRACEPOINTS
:
2938 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2939 case LTTNG_LIST_DOMAINS
:
2940 case LTTNG_LIST_CHANNELS
:
2941 case LTTNG_LIST_EVENTS
:
2942 case LTTNG_LIST_SYSCALLS
:
2943 case LTTNG_LIST_TRACKER_PIDS
:
2944 case LTTNG_DATA_PENDING
:
2947 /* Setup lttng message with no payload */
2948 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
2950 /* This label does not try to unlock the session */
2951 goto init_setup_error
;
2955 /* Commands that DO NOT need a session. */
2956 switch (cmd_ctx
->lsm
->cmd_type
) {
2957 case LTTNG_CREATE_SESSION
:
2958 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2959 case LTTNG_CREATE_SESSION_LIVE
:
2960 case LTTNG_LIST_SESSIONS
:
2961 case LTTNG_LIST_TRACEPOINTS
:
2962 case LTTNG_LIST_SYSCALLS
:
2963 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2964 case LTTNG_SAVE_SESSION
:
2965 case LTTNG_REGISTER_TRIGGER
:
2966 case LTTNG_UNREGISTER_TRIGGER
:
2967 need_tracing_session
= 0;
2970 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2972 * We keep the session list lock across _all_ commands
2973 * for now, because the per-session lock does not
2974 * handle teardown properly.
2976 session_lock_list();
2977 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2978 if (cmd_ctx
->session
== NULL
) {
2979 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2982 /* Acquire lock for the session */
2983 session_lock(cmd_ctx
->session
);
2989 * Commands that need a valid session but should NOT create one if none
2990 * exists. Instead of creating one and destroying it when the command is
2991 * handled, process that right before so we save some round trip in useless
2994 switch (cmd_ctx
->lsm
->cmd_type
) {
2995 case LTTNG_DISABLE_CHANNEL
:
2996 case LTTNG_DISABLE_EVENT
:
2997 switch (cmd_ctx
->lsm
->domain
.type
) {
2998 case LTTNG_DOMAIN_KERNEL
:
2999 if (!cmd_ctx
->session
->kernel_session
) {
3000 ret
= LTTNG_ERR_NO_CHANNEL
;
3004 case LTTNG_DOMAIN_JUL
:
3005 case LTTNG_DOMAIN_LOG4J
:
3006 case LTTNG_DOMAIN_PYTHON
:
3007 case LTTNG_DOMAIN_UST
:
3008 if (!cmd_ctx
->session
->ust_session
) {
3009 ret
= LTTNG_ERR_NO_CHANNEL
;
3014 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3026 * Check domain type for specific "pre-action".
3028 switch (cmd_ctx
->lsm
->domain
.type
) {
3029 case LTTNG_DOMAIN_KERNEL
:
3031 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3035 /* Kernel tracer check */
3036 if (kernel_tracer_fd
== -1) {
3037 /* Basically, load kernel tracer modules */
3038 ret
= init_kernel_tracer();
3044 /* Consumer is in an ERROR state. Report back to client */
3045 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3046 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3050 /* Need a session for kernel command */
3051 if (need_tracing_session
) {
3052 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3053 ret
= create_kernel_session(cmd_ctx
->session
);
3055 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3060 /* Start the kernel consumer daemon */
3061 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3062 if (kconsumer_data
.pid
== 0 &&
3063 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3064 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3065 ret
= start_consumerd(&kconsumer_data
);
3067 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3070 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3072 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3076 * The consumer was just spawned so we need to add the socket to
3077 * the consumer output of the session if exist.
3079 ret
= consumer_create_socket(&kconsumer_data
,
3080 cmd_ctx
->session
->kernel_session
->consumer
);
3087 case LTTNG_DOMAIN_JUL
:
3088 case LTTNG_DOMAIN_LOG4J
:
3089 case LTTNG_DOMAIN_PYTHON
:
3090 case LTTNG_DOMAIN_UST
:
3092 if (!ust_app_supported()) {
3093 ret
= LTTNG_ERR_NO_UST
;
3096 /* Consumer is in an ERROR state. Report back to client */
3097 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3098 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3102 if (need_tracing_session
) {
3103 /* Create UST session if none exist. */
3104 if (cmd_ctx
->session
->ust_session
== NULL
) {
3105 ret
= create_ust_session(cmd_ctx
->session
,
3106 &cmd_ctx
->lsm
->domain
);
3107 if (ret
!= LTTNG_OK
) {
3112 /* Start the UST consumer daemons */
3114 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3115 if (config
.consumerd64_bin_path
.value
&&
3116 ustconsumer64_data
.pid
== 0 &&
3117 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3118 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3119 ret
= start_consumerd(&ustconsumer64_data
);
3121 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3122 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3126 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3127 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3129 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3133 * Setup socket for consumer 64 bit. No need for atomic access
3134 * since it was set above and can ONLY be set in this thread.
3136 ret
= consumer_create_socket(&ustconsumer64_data
,
3137 cmd_ctx
->session
->ust_session
->consumer
);
3143 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3144 if (config
.consumerd32_bin_path
.value
&&
3145 ustconsumer32_data
.pid
== 0 &&
3146 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3147 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3148 ret
= start_consumerd(&ustconsumer32_data
);
3150 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3151 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3155 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3156 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3158 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3162 * Setup socket for consumer 64 bit. No need for atomic access
3163 * since it was set above and can ONLY be set in this thread.
3165 ret
= consumer_create_socket(&ustconsumer32_data
,
3166 cmd_ctx
->session
->ust_session
->consumer
);
3178 /* Validate consumer daemon state when start/stop trace command */
3179 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3180 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3181 switch (cmd_ctx
->lsm
->domain
.type
) {
3182 case LTTNG_DOMAIN_NONE
:
3184 case LTTNG_DOMAIN_JUL
:
3185 case LTTNG_DOMAIN_LOG4J
:
3186 case LTTNG_DOMAIN_PYTHON
:
3187 case LTTNG_DOMAIN_UST
:
3188 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3189 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3193 case LTTNG_DOMAIN_KERNEL
:
3194 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3195 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3200 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3206 * Check that the UID or GID match that of the tracing session.
3207 * The root user can interact with all sessions.
3209 if (need_tracing_session
) {
3210 if (!session_access_ok(cmd_ctx
->session
,
3211 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3212 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3213 ret
= LTTNG_ERR_EPERM
;
3219 * Send relayd information to consumer as soon as we have a domain and a
3222 if (cmd_ctx
->session
&& need_domain
) {
3224 * Setup relayd if not done yet. If the relayd information was already
3225 * sent to the consumer, this call will gracefully return.
3227 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3228 if (ret
!= LTTNG_OK
) {
3233 /* Process by command type */
3234 switch (cmd_ctx
->lsm
->cmd_type
) {
3235 case LTTNG_ADD_CONTEXT
:
3238 * An LTTNG_ADD_CONTEXT command might have a supplementary
3239 * payload if the context being added is an application context.
3241 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3242 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3243 char *provider_name
= NULL
, *context_name
= NULL
;
3244 size_t provider_name_len
=
3245 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3246 size_t context_name_len
=
3247 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3249 if (provider_name_len
== 0 || context_name_len
== 0) {
3251 * Application provider and context names MUST
3254 ret
= -LTTNG_ERR_INVALID
;
3258 provider_name
= zmalloc(provider_name_len
+ 1);
3259 if (!provider_name
) {
3260 ret
= -LTTNG_ERR_NOMEM
;
3263 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3266 context_name
= zmalloc(context_name_len
+ 1);
3267 if (!context_name
) {
3268 ret
= -LTTNG_ERR_NOMEM
;
3269 goto error_add_context
;
3271 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3274 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3277 goto error_add_context
;
3280 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3283 goto error_add_context
;
3288 * cmd_add_context assumes ownership of the provider and context
3291 ret
= cmd_add_context(cmd_ctx
->session
,
3292 cmd_ctx
->lsm
->domain
.type
,
3293 cmd_ctx
->lsm
->u
.context
.channel_name
,
3294 &cmd_ctx
->lsm
->u
.context
.ctx
,
3295 kernel_poll_pipe
[1]);
3297 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3298 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3300 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3301 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3307 case LTTNG_DISABLE_CHANNEL
:
3309 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3310 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3313 case LTTNG_DISABLE_EVENT
:
3317 * FIXME: handle filter; for now we just receive the filter's
3318 * bytecode along with the filter expression which are sent by
3319 * liblttng-ctl and discard them.
3321 * This fixes an issue where the client may block while sending
3322 * the filter payload and encounter an error because the session
3323 * daemon closes the socket without ever handling this data.
3325 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3326 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3329 char data
[LTTNG_FILTER_MAX_LEN
];
3331 DBG("Discarding disable event command payload of size %zu", count
);
3333 ret
= lttcomm_recv_unix_sock(sock
, data
,
3334 count
> sizeof(data
) ? sizeof(data
) : count
);
3339 count
-= (size_t) ret
;
3342 /* FIXME: passing packed structure to non-packed pointer */
3343 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3344 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3345 &cmd_ctx
->lsm
->u
.disable
.event
);
3348 case LTTNG_ENABLE_CHANNEL
:
3350 cmd_ctx
->lsm
->u
.channel
.chan
.attr
.extended
.ptr
=
3351 (struct lttng_channel_extended
*) &cmd_ctx
->lsm
->u
.channel
.extended
;
3352 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3353 &cmd_ctx
->lsm
->u
.channel
.chan
,
3354 kernel_poll_pipe
[1]);
3357 case LTTNG_TRACK_PID
:
3359 ret
= cmd_track_pid(cmd_ctx
->session
,
3360 cmd_ctx
->lsm
->domain
.type
,
3361 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3364 case LTTNG_UNTRACK_PID
:
3366 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3367 cmd_ctx
->lsm
->domain
.type
,
3368 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3371 case LTTNG_ENABLE_EVENT
:
3373 struct lttng_event_exclusion
*exclusion
= NULL
;
3374 struct lttng_filter_bytecode
*bytecode
= NULL
;
3375 char *filter_expression
= NULL
;
3377 /* Handle exclusion events and receive it from the client. */
3378 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3379 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3381 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3382 (count
* LTTNG_SYMBOL_NAME_LEN
));
3384 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3388 DBG("Receiving var len exclusion event list from client ...");
3389 exclusion
->count
= count
;
3390 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3391 count
* LTTNG_SYMBOL_NAME_LEN
);
3393 DBG("Nothing recv() from client var len data... continuing");
3396 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3401 /* Get filter expression from client. */
3402 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3403 size_t expression_len
=
3404 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3406 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3407 ret
= LTTNG_ERR_FILTER_INVAL
;
3412 filter_expression
= zmalloc(expression_len
);
3413 if (!filter_expression
) {
3415 ret
= LTTNG_ERR_FILTER_NOMEM
;
3419 /* Receive var. len. data */
3420 DBG("Receiving var len filter's expression from client ...");
3421 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3424 DBG("Nothing recv() from client car len data... continuing");
3426 free(filter_expression
);
3428 ret
= LTTNG_ERR_FILTER_INVAL
;
3433 /* Handle filter and get bytecode from client. */
3434 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3435 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3437 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3438 ret
= LTTNG_ERR_FILTER_INVAL
;
3439 free(filter_expression
);
3444 bytecode
= zmalloc(bytecode_len
);
3446 free(filter_expression
);
3448 ret
= LTTNG_ERR_FILTER_NOMEM
;
3452 /* Receive var. len. data */
3453 DBG("Receiving var len filter's bytecode from client ...");
3454 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3456 DBG("Nothing recv() from client car len data... continuing");
3458 free(filter_expression
);
3461 ret
= LTTNG_ERR_FILTER_INVAL
;
3465 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3466 free(filter_expression
);
3469 ret
= LTTNG_ERR_FILTER_INVAL
;
3474 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3475 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3476 &cmd_ctx
->lsm
->u
.enable
.event
,
3477 filter_expression
, bytecode
, exclusion
,
3478 kernel_poll_pipe
[1]);
3481 case LTTNG_LIST_TRACEPOINTS
:
3483 struct lttng_event
*events
;
3486 session_lock_list();
3487 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3488 session_unlock_list();
3489 if (nb_events
< 0) {
3490 /* Return value is a negative lttng_error_code. */
3496 * Setup lttng message with payload size set to the event list size in
3497 * bytes and then copy list into the llm payload.
3499 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3500 sizeof(struct lttng_event
) * nb_events
);
3510 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3512 struct lttng_event_field
*fields
;
3515 session_lock_list();
3516 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3518 session_unlock_list();
3519 if (nb_fields
< 0) {
3520 /* Return value is a negative lttng_error_code. */
3526 * Setup lttng message with payload size set to the event list size in
3527 * bytes and then copy list into the llm payload.
3529 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3530 sizeof(struct lttng_event_field
) * nb_fields
);
3540 case LTTNG_LIST_SYSCALLS
:
3542 struct lttng_event
*events
;
3545 nb_events
= cmd_list_syscalls(&events
);
3546 if (nb_events
< 0) {
3547 /* Return value is a negative lttng_error_code. */
3553 * Setup lttng message with payload size set to the event list size in
3554 * bytes and then copy list into the llm payload.
3556 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3557 sizeof(struct lttng_event
) * nb_events
);
3567 case LTTNG_LIST_TRACKER_PIDS
:
3569 int32_t *pids
= NULL
;
3572 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3573 cmd_ctx
->lsm
->domain
.type
, &pids
);
3575 /* Return value is a negative lttng_error_code. */
3581 * Setup lttng message with payload size set to the event list size in
3582 * bytes and then copy list into the llm payload.
3584 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3585 sizeof(int32_t) * nr_pids
);
3595 case LTTNG_SET_CONSUMER_URI
:
3598 struct lttng_uri
*uris
;
3600 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3601 len
= nb_uri
* sizeof(struct lttng_uri
);
3604 ret
= LTTNG_ERR_INVALID
;
3608 uris
= zmalloc(len
);
3610 ret
= LTTNG_ERR_FATAL
;
3614 /* Receive variable len data */
3615 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3616 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3618 DBG("No URIs received from client... continuing");
3620 ret
= LTTNG_ERR_SESSION_FAIL
;
3625 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3627 if (ret
!= LTTNG_OK
) {
3634 case LTTNG_START_TRACE
:
3636 ret
= cmd_start_trace(cmd_ctx
->session
);
3639 case LTTNG_STOP_TRACE
:
3641 ret
= cmd_stop_trace(cmd_ctx
->session
);
3644 case LTTNG_CREATE_SESSION
:
3647 struct lttng_uri
*uris
= NULL
;
3649 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3650 len
= nb_uri
* sizeof(struct lttng_uri
);
3653 uris
= zmalloc(len
);
3655 ret
= LTTNG_ERR_FATAL
;
3659 /* Receive variable len data */
3660 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3661 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3663 DBG("No URIs received from client... continuing");
3665 ret
= LTTNG_ERR_SESSION_FAIL
;
3670 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3671 DBG("Creating session with ONE network URI is a bad call");
3672 ret
= LTTNG_ERR_SESSION_FAIL
;
3678 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3679 &cmd_ctx
->creds
, 0);
3685 case LTTNG_DESTROY_SESSION
:
3687 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3689 /* Set session to NULL so we do not unlock it after free. */
3690 cmd_ctx
->session
= NULL
;
3693 case LTTNG_LIST_DOMAINS
:
3696 struct lttng_domain
*domains
= NULL
;
3698 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3700 /* Return value is a negative lttng_error_code. */
3705 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3706 nb_dom
* sizeof(struct lttng_domain
));
3716 case LTTNG_LIST_CHANNELS
:
3718 ssize_t payload_size
;
3719 struct lttng_channel
*channels
= NULL
;
3721 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3722 cmd_ctx
->session
, &channels
);
3723 if (payload_size
< 0) {
3724 /* Return value is a negative lttng_error_code. */
3725 ret
= -payload_size
;
3729 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3740 case LTTNG_LIST_EVENTS
:
3743 struct lttng_event
*events
= NULL
;
3744 struct lttcomm_event_command_header cmd_header
;
3747 memset(&cmd_header
, 0, sizeof(cmd_header
));
3748 /* Extended infos are included at the end of events */
3749 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3750 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3751 &events
, &total_size
);
3754 /* Return value is a negative lttng_error_code. */
3759 cmd_header
.nb_events
= nb_event
;
3760 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3761 &cmd_header
, sizeof(cmd_header
));
3771 case LTTNG_LIST_SESSIONS
:
3773 unsigned int nr_sessions
;
3774 void *sessions_payload
;
3777 session_lock_list();
3778 nr_sessions
= lttng_sessions_count(
3779 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3780 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3781 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3782 sessions_payload
= zmalloc(payload_len
);
3784 if (!sessions_payload
) {
3785 session_unlock_list();
3790 cmd_list_lttng_sessions(sessions_payload
,
3791 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3792 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3793 session_unlock_list();
3795 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3797 free(sessions_payload
);
3806 case LTTNG_REGISTER_CONSUMER
:
3808 struct consumer_data
*cdata
;
3810 switch (cmd_ctx
->lsm
->domain
.type
) {
3811 case LTTNG_DOMAIN_KERNEL
:
3812 cdata
= &kconsumer_data
;
3815 ret
= LTTNG_ERR_UND
;
3819 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3820 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3823 case LTTNG_DATA_PENDING
:
3826 uint8_t pending_ret_byte
;
3828 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3833 * This function may returns 0 or 1 to indicate whether or not
3834 * there is data pending. In case of error, it should return an
3835 * LTTNG_ERR code. However, some code paths may still return
3836 * a nondescript error code, which we handle by returning an
3839 if (pending_ret
== 0 || pending_ret
== 1) {
3841 * ret will be set to LTTNG_OK at the end of
3844 } else if (pending_ret
< 0) {
3845 ret
= LTTNG_ERR_UNK
;
3852 pending_ret_byte
= (uint8_t) pending_ret
;
3854 /* 1 byte to return whether or not data is pending */
3855 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
3856 &pending_ret_byte
, 1);
3865 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3867 struct lttcomm_lttng_output_id reply
;
3869 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3870 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3871 if (ret
!= LTTNG_OK
) {
3875 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
3881 /* Copy output list into message payload */
3885 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3887 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3888 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3891 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3894 struct lttng_snapshot_output
*outputs
= NULL
;
3896 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3897 if (nb_output
< 0) {
3902 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
3903 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
3904 nb_output
* sizeof(struct lttng_snapshot_output
));
3914 case LTTNG_SNAPSHOT_RECORD
:
3916 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3917 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3918 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3921 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3924 struct lttng_uri
*uris
= NULL
;
3926 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3927 len
= nb_uri
* sizeof(struct lttng_uri
);
3930 uris
= zmalloc(len
);
3932 ret
= LTTNG_ERR_FATAL
;
3936 /* Receive variable len data */
3937 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3938 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3940 DBG("No URIs received from client... continuing");
3942 ret
= LTTNG_ERR_SESSION_FAIL
;
3947 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3948 DBG("Creating session with ONE network URI is a bad call");
3949 ret
= LTTNG_ERR_SESSION_FAIL
;
3955 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3956 nb_uri
, &cmd_ctx
->creds
);
3960 case LTTNG_CREATE_SESSION_LIVE
:
3963 struct lttng_uri
*uris
= NULL
;
3965 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3966 len
= nb_uri
* sizeof(struct lttng_uri
);
3969 uris
= zmalloc(len
);
3971 ret
= LTTNG_ERR_FATAL
;
3975 /* Receive variable len data */
3976 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3977 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3979 DBG("No URIs received from client... continuing");
3981 ret
= LTTNG_ERR_SESSION_FAIL
;
3986 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3987 DBG("Creating session with ONE network URI is a bad call");
3988 ret
= LTTNG_ERR_SESSION_FAIL
;
3994 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3995 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3999 case LTTNG_SAVE_SESSION
:
4001 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4005 case LTTNG_SET_SESSION_SHM_PATH
:
4007 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4008 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4011 case LTTNG_REGENERATE_METADATA
:
4013 ret
= cmd_regenerate_metadata(cmd_ctx
->session
);
4016 case LTTNG_REGENERATE_STATEDUMP
:
4018 ret
= cmd_regenerate_statedump(cmd_ctx
->session
);
4021 case LTTNG_REGISTER_TRIGGER
:
4023 ret
= cmd_register_trigger(cmd_ctx
, sock
,
4024 notification_thread_handle
);
4027 case LTTNG_UNREGISTER_TRIGGER
:
4029 ret
= cmd_unregister_trigger(cmd_ctx
, sock
,
4030 notification_thread_handle
);
4034 ret
= LTTNG_ERR_UND
;
4039 if (cmd_ctx
->llm
== NULL
) {
4040 DBG("Missing llm structure. Allocating one.");
4041 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4045 /* Set return code */
4046 cmd_ctx
->llm
->ret_code
= ret
;
4048 if (cmd_ctx
->session
) {
4049 session_unlock(cmd_ctx
->session
);
4051 if (need_tracing_session
) {
4052 session_unlock_list();
4055 assert(!rcu_read_ongoing());
4060 * Thread managing health check socket.
4062 static void *thread_manage_health(void *data
)
4064 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4065 uint32_t revents
, nb_fd
;
4066 struct lttng_poll_event events
;
4067 struct health_comm_msg msg
;
4068 struct health_comm_reply reply
;
4070 DBG("[thread] Manage health check started");
4072 rcu_register_thread();
4074 /* We might hit an error path before this is created. */
4075 lttng_poll_init(&events
);
4077 /* Create unix socket */
4078 sock
= lttcomm_create_unix_sock(config
.health_unix_sock_path
.value
);
4080 ERR("Unable to create health check Unix socket");
4085 /* lttng health client socket path permissions */
4086 ret
= chown(config
.health_unix_sock_path
.value
, 0,
4087 utils_get_group_id(config
.tracing_group_name
.value
));
4089 ERR("Unable to set group on %s", config
.health_unix_sock_path
.value
);
4094 ret
= chmod(config
.health_unix_sock_path
.value
,
4095 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4097 ERR("Unable to set permissions on %s", config
.health_unix_sock_path
.value
);
4104 * Set the CLOEXEC flag. Return code is useless because either way, the
4107 (void) utils_set_fd_cloexec(sock
);
4109 ret
= lttcomm_listen_unix_sock(sock
);
4115 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4116 * more will be added to this poll set.
4118 ret
= sessiond_set_thread_pollset(&events
, 2);
4123 /* Add the application registration socket */
4124 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4129 sessiond_notify_ready();
4132 DBG("Health check ready");
4134 /* Inifinite blocking call, waiting for transmission */
4136 ret
= lttng_poll_wait(&events
, -1);
4139 * Restart interrupted system call.
4141 if (errno
== EINTR
) {
4149 for (i
= 0; i
< nb_fd
; i
++) {
4150 /* Fetch once the poll data */
4151 revents
= LTTNG_POLL_GETEV(&events
, i
);
4152 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4155 /* No activity for this FD (poll implementation). */
4159 /* Thread quit pipe has been closed. Killing thread. */
4160 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4166 /* Event on the registration socket */
4167 if (pollfd
== sock
) {
4168 if (revents
& LPOLLIN
) {
4170 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4171 ERR("Health socket poll error");
4174 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4180 new_sock
= lttcomm_accept_unix_sock(sock
);
4186 * Set the CLOEXEC flag. Return code is useless because either way, the
4189 (void) utils_set_fd_cloexec(new_sock
);
4191 DBG("Receiving data from client for health...");
4192 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4194 DBG("Nothing recv() from client... continuing");
4195 ret
= close(new_sock
);
4202 rcu_thread_online();
4204 memset(&reply
, 0, sizeof(reply
));
4205 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4207 * health_check_state returns 0 if health is
4210 if (!health_check_state(health_sessiond
, i
)) {
4211 reply
.ret_code
|= 1ULL << i
;
4215 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4217 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4219 ERR("Failed to send health data back to client");
4222 /* End of transmission */
4223 ret
= close(new_sock
);
4232 ERR("Health error occurred in %s", __func__
);
4234 DBG("Health check thread dying");
4235 unlink(config
.health_unix_sock_path
.value
);
4243 lttng_poll_clean(&events
);
4245 rcu_unregister_thread();
4250 * This thread manage all clients request using the unix client socket for
4253 static void *thread_manage_clients(void *data
)
4255 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4257 uint32_t revents
, nb_fd
;
4258 struct command_ctx
*cmd_ctx
= NULL
;
4259 struct lttng_poll_event events
;
4261 DBG("[thread] Manage client started");
4263 rcu_register_thread();
4265 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4267 health_code_update();
4269 ret
= lttcomm_listen_unix_sock(client_sock
);
4275 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4276 * more will be added to this poll set.
4278 ret
= sessiond_set_thread_pollset(&events
, 2);
4280 goto error_create_poll
;
4283 /* Add the application registration socket */
4284 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4289 sessiond_notify_ready();
4291 ret
= sem_post(&load_info
->message_thread_ready
);
4293 PERROR("sem_post message_thread_ready");
4298 * Wait until all support threads are initialized before accepting
4301 while (uatomic_read(<tng_sessiond_ready
) != 0) {
4303 struct timeval timeout
;
4306 FD_SET(thread_quit_pipe
[0], &read_fds
);
4307 memset(&timeout
, 0, sizeof(timeout
));
4308 timeout
.tv_usec
= 1000;
4311 * If a support thread failed to launch, it may signal that
4312 * we must exit and the sessiond would never be marked as
4315 * The timeout is set to 1ms, which serves as a way to
4316 * pace down this check.
4318 ret
= select(thread_quit_pipe
[0] + 1, &read_fds
, NULL
, NULL
,
4320 if (ret
> 0 || (ret
< 0 && errno
!= EINTR
)) {
4325 /* This testpoint is after we signal readiness to the parent. */
4326 if (testpoint(sessiond_thread_manage_clients
)) {
4330 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4334 health_code_update();
4337 DBG("Accepting client command ...");
4339 /* Inifinite blocking call, waiting for transmission */
4341 health_poll_entry();
4342 ret
= lttng_poll_wait(&events
, -1);
4346 * Restart interrupted system call.
4348 if (errno
== EINTR
) {
4356 for (i
= 0; i
< nb_fd
; i
++) {
4357 /* Fetch once the poll data */
4358 revents
= LTTNG_POLL_GETEV(&events
, i
);
4359 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4361 health_code_update();
4364 /* No activity for this FD (poll implementation). */
4368 /* Thread quit pipe has been closed. Killing thread. */
4369 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4375 /* Event on the registration socket */
4376 if (pollfd
== client_sock
) {
4377 if (revents
& LPOLLIN
) {
4379 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4380 ERR("Client socket poll error");
4383 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4389 DBG("Wait for client response");
4391 health_code_update();
4393 sock
= lttcomm_accept_unix_sock(client_sock
);
4399 * Set the CLOEXEC flag. Return code is useless because either way, the
4402 (void) utils_set_fd_cloexec(sock
);
4404 /* Set socket option for credentials retrieval */
4405 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4410 /* Allocate context command to process the client request */
4411 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4412 if (cmd_ctx
== NULL
) {
4413 PERROR("zmalloc cmd_ctx");
4417 /* Allocate data buffer for reception */
4418 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4419 if (cmd_ctx
->lsm
== NULL
) {
4420 PERROR("zmalloc cmd_ctx->lsm");
4424 cmd_ctx
->llm
= NULL
;
4425 cmd_ctx
->session
= NULL
;
4427 health_code_update();
4430 * Data is received from the lttng client. The struct
4431 * lttcomm_session_msg (lsm) contains the command and data request of
4434 DBG("Receiving data from client ...");
4435 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4436 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4438 DBG("Nothing recv() from client... continuing");
4444 clean_command_ctx(&cmd_ctx
);
4448 health_code_update();
4450 // TODO: Validate cmd_ctx including sanity check for
4451 // security purpose.
4453 rcu_thread_online();
4455 * This function dispatch the work to the kernel or userspace tracer
4456 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4457 * informations for the client. The command context struct contains
4458 * everything this function may needs.
4460 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4461 rcu_thread_offline();
4469 * TODO: Inform client somehow of the fatal error. At
4470 * this point, ret < 0 means that a zmalloc failed
4471 * (ENOMEM). Error detected but still accept
4472 * command, unless a socket error has been
4475 clean_command_ctx(&cmd_ctx
);
4479 health_code_update();
4481 DBG("Sending response (size: %d, retcode: %s (%d))",
4482 cmd_ctx
->lttng_msg_size
,
4483 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4484 cmd_ctx
->llm
->ret_code
);
4485 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4487 ERR("Failed to send data back to client");
4490 /* End of transmission */
4497 clean_command_ctx(&cmd_ctx
);
4499 health_code_update();
4511 lttng_poll_clean(&events
);
4512 clean_command_ctx(&cmd_ctx
);
4516 unlink(config
.client_unix_sock_path
.value
);
4517 if (client_sock
>= 0) {
4518 ret
= close(client_sock
);
4526 ERR("Health error occurred in %s", __func__
);
4529 health_unregister(health_sessiond
);
4531 DBG("Client thread dying");
4533 rcu_unregister_thread();
4536 * Since we are creating the consumer threads, we own them, so we need
4537 * to join them before our thread exits.
4539 ret
= join_consumer_thread(&kconsumer_data
);
4542 PERROR("join_consumer");
4545 ret
= join_consumer_thread(&ustconsumer32_data
);
4548 PERROR("join_consumer ust32");
4551 ret
= join_consumer_thread(&ustconsumer64_data
);
4554 PERROR("join_consumer ust64");
4559 static int string_match(const char *str1
, const char *str2
)
4561 return (str1
&& str2
) && !strcmp(str1
, str2
);
4565 * Take an option from the getopt output and set it in the right variable to be
4568 * Return 0 on success else a negative value.
4570 static int set_option(int opt
, const char *arg
, const char *optname
)
4574 if (string_match(optname
, "client-sock") || opt
== 'c') {
4575 if (!arg
|| *arg
== '\0') {
4579 if (lttng_is_setuid_setgid()) {
4580 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4581 "-c, --client-sock");
4583 config_string_set(&config
.client_unix_sock_path
,
4585 if (!config
.client_unix_sock_path
.value
) {
4590 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4591 if (!arg
|| *arg
== '\0') {
4595 if (lttng_is_setuid_setgid()) {
4596 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4599 config_string_set(&config
.apps_unix_sock_path
,
4601 if (!config
.apps_unix_sock_path
.value
) {
4606 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4607 config
.daemonize
= true;
4608 } else if (string_match(optname
, "background") || opt
== 'b') {
4609 config
.background
= true;
4610 } else if (string_match(optname
, "group") || opt
== 'g') {
4611 if (!arg
|| *arg
== '\0') {
4615 if (lttng_is_setuid_setgid()) {
4616 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4619 config_string_set(&config
.tracing_group_name
,
4621 if (!config
.tracing_group_name
.value
) {
4626 } else if (string_match(optname
, "help") || opt
== 'h') {
4627 ret
= utils_show_help(8, "lttng-sessiond", help_msg
);
4629 ERR("Cannot show --help for `lttng-sessiond`");
4632 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4633 } else if (string_match(optname
, "version") || opt
== 'V') {
4634 fprintf(stdout
, "%s\n", VERSION
);
4636 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4637 config
.sig_parent
= true;
4638 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4639 if (!arg
|| *arg
== '\0') {
4643 if (lttng_is_setuid_setgid()) {
4644 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4645 "--kconsumerd-err-sock");
4647 config_string_set(&config
.kconsumerd_err_unix_sock_path
,
4649 if (!config
.kconsumerd_err_unix_sock_path
.value
) {
4654 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4655 if (!arg
|| *arg
== '\0') {
4659 if (lttng_is_setuid_setgid()) {
4660 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4661 "--kconsumerd-cmd-sock");
4663 config_string_set(&config
.kconsumerd_cmd_unix_sock_path
,
4665 if (!config
.kconsumerd_cmd_unix_sock_path
.value
) {
4670 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4671 if (!arg
|| *arg
== '\0') {
4675 if (lttng_is_setuid_setgid()) {
4676 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4677 "--ustconsumerd64-err-sock");
4679 config_string_set(&config
.consumerd64_err_unix_sock_path
,
4681 if (!config
.consumerd64_err_unix_sock_path
.value
) {
4686 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4687 if (!arg
|| *arg
== '\0') {
4691 if (lttng_is_setuid_setgid()) {
4692 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4693 "--ustconsumerd64-cmd-sock");
4695 config_string_set(&config
.consumerd64_cmd_unix_sock_path
,
4697 if (!config
.consumerd64_cmd_unix_sock_path
.value
) {
4702 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4703 if (!arg
|| *arg
== '\0') {
4707 if (lttng_is_setuid_setgid()) {
4708 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4709 "--ustconsumerd32-err-sock");
4711 config_string_set(&config
.consumerd32_err_unix_sock_path
,
4713 if (!config
.consumerd32_err_unix_sock_path
.value
) {
4718 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4719 if (!arg
|| *arg
== '\0') {
4723 if (lttng_is_setuid_setgid()) {
4724 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4725 "--ustconsumerd32-cmd-sock");
4727 config_string_set(&config
.consumerd32_cmd_unix_sock_path
,
4729 if (!config
.consumerd32_cmd_unix_sock_path
.value
) {
4734 } else if (string_match(optname
, "no-kernel")) {
4735 config
.no_kernel
= true;
4736 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4737 config
.quiet
= true;
4738 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4739 /* Verbose level can increase using multiple -v */
4741 /* Value obtained from config file */
4742 config
.verbose
= config_parse_value(arg
);
4744 /* -v used on command line */
4747 /* Clamp value to [0, 3] */
4748 config
.verbose
= config
.verbose
< 0 ? 0 :
4749 (config
.verbose
<= 3 ? config
.verbose
: 3);
4750 } else if (string_match(optname
, "verbose-consumer")) {
4752 config
.verbose_consumer
= config_parse_value(arg
);
4754 config
.verbose_consumer
++;
4756 } else if (string_match(optname
, "consumerd32-path")) {
4757 if (!arg
|| *arg
== '\0') {
4761 if (lttng_is_setuid_setgid()) {
4762 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4763 "--consumerd32-path");
4765 config_string_set(&config
.consumerd32_bin_path
,
4767 if (!config
.consumerd32_bin_path
.value
) {
4772 } else if (string_match(optname
, "consumerd32-libdir")) {
4773 if (!arg
|| *arg
== '\0') {
4777 if (lttng_is_setuid_setgid()) {
4778 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4779 "--consumerd32-libdir");
4781 config_string_set(&config
.consumerd32_lib_dir
,
4783 if (!config
.consumerd32_lib_dir
.value
) {
4788 } else if (string_match(optname
, "consumerd64-path")) {
4789 if (!arg
|| *arg
== '\0') {
4793 if (lttng_is_setuid_setgid()) {
4794 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4795 "--consumerd64-path");
4797 config_string_set(&config
.consumerd64_bin_path
,
4799 if (!config
.consumerd64_bin_path
.value
) {
4804 } else if (string_match(optname
, "consumerd64-libdir")) {
4805 if (!arg
|| *arg
== '\0') {
4809 if (lttng_is_setuid_setgid()) {
4810 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4811 "--consumerd64-libdir");
4813 config_string_set(&config
.consumerd64_lib_dir
,
4815 if (!config
.consumerd64_lib_dir
.value
) {
4820 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4821 if (!arg
|| *arg
== '\0') {
4825 if (lttng_is_setuid_setgid()) {
4826 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4829 config_string_set(&config
.pid_file_path
, strdup(arg
));
4830 if (!config
.pid_file_path
.value
) {
4835 } else if (string_match(optname
, "agent-tcp-port")) {
4836 if (!arg
|| *arg
== '\0') {
4840 if (lttng_is_setuid_setgid()) {
4841 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4842 "--agent-tcp-port");
4847 v
= strtoul(arg
, NULL
, 0);
4848 if (errno
!= 0 || !isdigit(arg
[0])) {
4849 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4852 if (v
== 0 || v
>= 65535) {
4853 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4856 config
.agent_tcp_port
= (uint32_t) v
;
4857 DBG3("Agent TCP port set to non default: %u", config
.agent_tcp_port
);
4859 } else if (string_match(optname
, "load") || opt
== 'l') {
4860 if (!arg
|| *arg
== '\0') {
4864 if (lttng_is_setuid_setgid()) {
4865 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4868 config_string_set(&config
.load_session_path
, strdup(arg
));
4869 if (!config
.load_session_path
.value
) {
4874 } else if (string_match(optname
, "kmod-probes")) {
4875 if (!arg
|| *arg
== '\0') {
4879 if (lttng_is_setuid_setgid()) {
4880 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4883 config_string_set(&config
.kmod_probes_list
, strdup(arg
));
4884 if (!config
.kmod_probes_list
.value
) {
4889 } else if (string_match(optname
, "extra-kmod-probes")) {
4890 if (!arg
|| *arg
== '\0') {
4894 if (lttng_is_setuid_setgid()) {
4895 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4896 "--extra-kmod-probes");
4898 config_string_set(&config
.kmod_extra_probes_list
,
4900 if (!config
.kmod_extra_probes_list
.value
) {
4905 } else if (string_match(optname
, "config") || opt
== 'f') {
4906 /* This is handled in set_options() thus silent skip. */
4909 /* Unknown option or other error.
4910 * Error is printed by getopt, just return */
4915 if (ret
== -EINVAL
) {
4916 const char *opt_name
= "unknown";
4919 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4921 if (opt
== long_options
[i
].val
) {
4922 opt_name
= long_options
[i
].name
;
4927 WARN("Invalid argument provided for option \"%s\", using default value.",
4935 * config_entry_handler_cb used to handle options read from a config file.
4936 * See config_entry_handler_cb comment in common/config/session-config.h for the
4937 * return value conventions.
4939 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4943 if (!entry
|| !entry
->name
|| !entry
->value
) {
4948 /* Check if the option is to be ignored */
4949 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4950 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4955 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4958 /* Ignore if not fully matched. */
4959 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4964 * If the option takes no argument on the command line, we have to
4965 * check if the value is "true". We support non-zero numeric values,
4968 if (!long_options
[i
].has_arg
) {
4969 ret
= config_parse_value(entry
->value
);
4972 WARN("Invalid configuration value \"%s\" for option %s",
4973 entry
->value
, entry
->name
);
4975 /* False, skip boolean config option. */
4980 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4984 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4991 * daemon configuration loading and argument parsing
4993 static int set_options(int argc
, char **argv
)
4995 int ret
= 0, c
= 0, option_index
= 0;
4996 int orig_optopt
= optopt
, orig_optind
= optind
;
4998 const char *config_path
= NULL
;
5000 optstring
= utils_generate_optstring(long_options
,
5001 sizeof(long_options
) / sizeof(struct option
));
5007 /* Check for the --config option */
5008 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5009 &option_index
)) != -1) {
5013 } else if (c
!= 'f') {
5014 /* if not equal to --config option. */
5018 if (lttng_is_setuid_setgid()) {
5019 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5022 config_path
= utils_expand_path(optarg
);
5024 ERR("Failed to resolve path: %s", optarg
);
5029 ret
= config_get_section_entries(config_path
, config_section_name
,
5030 config_entry_handler
, NULL
);
5033 ERR("Invalid configuration option at line %i", ret
);
5039 /* Reset getopt's global state */
5040 optopt
= orig_optopt
;
5041 optind
= orig_optind
;
5045 * getopt_long() will not set option_index if it encounters a
5048 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5055 * Pass NULL as the long option name if popt left the index
5058 ret
= set_option(c
, optarg
,
5059 option_index
< 0 ? NULL
:
5060 long_options
[option_index
].name
);
5072 * Creates the two needed socket by the daemon.
5073 * apps_sock - The communication socket for all UST apps.
5074 * client_sock - The communication of the cli tool (lttng).
5076 static int init_daemon_socket(void)
5081 old_umask
= umask(0);
5083 /* Create client tool unix socket */
5084 client_sock
= lttcomm_create_unix_sock(config
.client_unix_sock_path
.value
);
5085 if (client_sock
< 0) {
5086 ERR("Create unix sock failed: %s", config
.client_unix_sock_path
.value
);
5091 /* Set the cloexec flag */
5092 ret
= utils_set_fd_cloexec(client_sock
);
5094 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5095 "Continuing but note that the consumer daemon will have a "
5096 "reference to this socket on exec()", client_sock
);
5099 /* File permission MUST be 660 */
5100 ret
= chmod(config
.client_unix_sock_path
.value
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5102 ERR("Set file permissions failed: %s", config
.client_unix_sock_path
.value
);
5107 /* Create the application unix socket */
5108 apps_sock
= lttcomm_create_unix_sock(config
.apps_unix_sock_path
.value
);
5109 if (apps_sock
< 0) {
5110 ERR("Create unix sock failed: %s", config
.apps_unix_sock_path
.value
);
5115 /* Set the cloexec flag */
5116 ret
= utils_set_fd_cloexec(apps_sock
);
5118 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5119 "Continuing but note that the consumer daemon will have a "
5120 "reference to this socket on exec()", apps_sock
);
5123 /* File permission MUST be 666 */
5124 ret
= chmod(config
.apps_unix_sock_path
.value
,
5125 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5127 ERR("Set file permissions failed: %s", config
.apps_unix_sock_path
.value
);
5132 DBG3("Session daemon client socket %d and application socket %d created",
5133 client_sock
, apps_sock
);
5141 * Create lockfile using the rundir and return its fd.
5143 static int create_lockfile(void)
5145 return utils_create_lock_file(config
.lock_file_path
.value
);
5149 * Check if the global socket is available, and if a daemon is answering at the
5150 * other side. If yes, error is returned.
5152 * Also attempts to create and hold the lock file.
5154 static int check_existing_daemon(void)
5158 /* Is there anybody out there ? */
5159 if (lttng_session_daemon_alive()) {
5164 lockfile_fd
= create_lockfile();
5165 if (lockfile_fd
< 0) {
5173 static void sessiond_cleanup_lock_file(void)
5178 * Cleanup lock file by deleting it and finaly closing it which will
5179 * release the file system lock.
5181 if (lockfile_fd
>= 0) {
5182 ret
= remove(config
.lock_file_path
.value
);
5184 PERROR("remove lock file");
5186 ret
= close(lockfile_fd
);
5188 PERROR("close lock file");
5194 * Set the tracing group gid onto the client socket.
5196 * Race window between mkdir and chown is OK because we are going from more
5197 * permissive (root.root) to less permissive (root.tracing).
5199 static int set_permissions(char *rundir
)
5204 gid
= utils_get_group_id(config
.tracing_group_name
.value
);
5206 /* Set lttng run dir */
5207 ret
= chown(rundir
, 0, gid
);
5209 ERR("Unable to set group on %s", rundir
);
5214 * Ensure all applications and tracing group can search the run
5215 * dir. Allow everyone to read the directory, since it does not
5216 * buy us anything to hide its content.
5218 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5220 ERR("Unable to set permissions on %s", rundir
);
5224 /* lttng client socket path */
5225 ret
= chown(config
.client_unix_sock_path
.value
, 0, gid
);
5227 ERR("Unable to set group on %s", config
.client_unix_sock_path
.value
);
5231 /* kconsumer error socket path */
5232 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5234 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5238 /* 64-bit ustconsumer error socket path */
5239 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5241 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5245 /* 32-bit ustconsumer compat32 error socket path */
5246 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5248 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5252 DBG("All permissions are set");
5258 * Create the lttng run directory needed for all global sockets and pipe.
5260 static int create_lttng_rundir(void)
5264 DBG3("Creating LTTng run directory: %s", config
.rundir
.value
);
5266 ret
= mkdir(config
.rundir
.value
, S_IRWXU
);
5268 if (errno
!= EEXIST
) {
5269 ERR("Unable to create %s", config
.rundir
.value
);
5281 * Setup sockets and directory needed by the consumerds' communication with the
5284 static int set_consumer_sockets(struct consumer_data
*consumer_data
)
5289 switch (consumer_data
->type
) {
5290 case LTTNG_CONSUMER_KERNEL
:
5291 path
= config
.kconsumerd_path
.value
;
5293 case LTTNG_CONSUMER64_UST
:
5294 path
= config
.consumerd64_path
.value
;
5296 case LTTNG_CONSUMER32_UST
:
5297 path
= config
.consumerd32_path
.value
;
5300 ERR("Consumer type unknown");
5306 DBG2("Creating consumer directory: %s", path
);
5308 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5309 if (ret
< 0 && errno
!= EEXIST
) {
5311 ERR("Failed to create %s", path
);
5315 ret
= chown(path
, 0, utils_get_group_id(config
.tracing_group_name
.value
));
5317 ERR("Unable to set group on %s", path
);
5323 /* Create the consumerd error unix socket */
5324 consumer_data
->err_sock
=
5325 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5326 if (consumer_data
->err_sock
< 0) {
5327 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5333 * Set the CLOEXEC flag. Return code is useless because either way, the
5336 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5338 PERROR("utils_set_fd_cloexec");
5339 /* continue anyway */
5342 /* File permission MUST be 660 */
5343 ret
= chmod(consumer_data
->err_unix_sock_path
,
5344 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5346 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5356 * Signal handler for the daemon
5358 * Simply stop all worker threads, leaving main() return gracefully after
5359 * joining all threads and calling cleanup().
5361 static void sighandler(int sig
)
5365 DBG("SIGINT caught");
5369 DBG("SIGTERM caught");
5373 CMM_STORE_SHARED(recv_child_signal
, 1);
5381 * Setup signal handler for :
5382 * SIGINT, SIGTERM, SIGPIPE
5384 static int set_signal_handler(void)
5387 struct sigaction sa
;
5390 if ((ret
= sigemptyset(&sigset
)) < 0) {
5391 PERROR("sigemptyset");
5395 sa
.sa_mask
= sigset
;
5398 sa
.sa_handler
= sighandler
;
5399 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5400 PERROR("sigaction");
5404 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5405 PERROR("sigaction");
5409 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5410 PERROR("sigaction");
5414 sa
.sa_handler
= SIG_IGN
;
5415 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5416 PERROR("sigaction");
5420 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5426 * Set open files limit to unlimited. This daemon can open a large number of
5427 * file descriptors in order to consume multiple kernel traces.
5429 static void set_ulimit(void)
5434 /* The kernel does not allow an infinite limit for open files */
5435 lim
.rlim_cur
= 65535;
5436 lim
.rlim_max
= 65535;
5438 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5440 PERROR("failed to set open files limit");
5444 static int write_pidfile(void)
5446 return utils_create_pid_file(getpid(), config
.pid_file_path
.value
);
5450 * Write agent TCP port using the rundir.
5452 static int write_agent_port(void)
5454 return utils_create_pid_file(config
.agent_tcp_port
,
5455 config
.agent_port_file_path
.value
);
5458 static int set_clock_plugin_env(void)
5461 char *env_value
= NULL
;
5463 if (!config
.lttng_ust_clock_plugin
.value
) {
5467 ret
= asprintf(&env_value
, "LTTNG_UST_CLOCK_PLUGIN=%s",
5468 config
.lttng_ust_clock_plugin
.value
);
5474 ret
= putenv(env_value
);
5477 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5481 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5482 config
.lttng_ust_clock_plugin
.value
);
5490 int main(int argc
, char **argv
)
5492 int ret
= 0, retval
= 0;
5494 const char *env_app_timeout
;
5495 struct lttng_pipe
*ust32_channel_monitor_pipe
= NULL
,
5496 *ust64_channel_monitor_pipe
= NULL
,
5497 *kernel_channel_monitor_pipe
= NULL
;
5498 bool notification_thread_running
= false;
5500 init_kernel_workarounds();
5502 rcu_register_thread();
5504 if (set_signal_handler()) {
5506 goto exit_set_signal_handler
;
5509 page_size
= sysconf(_SC_PAGESIZE
);
5510 if (page_size
< 0) {
5511 PERROR("sysconf _SC_PAGESIZE");
5512 page_size
= LONG_MAX
;
5513 WARN("Fallback page size to %ld", page_size
);
5516 ret
= sessiond_config_init(&config
);
5519 goto exit_set_signal_handler
;
5523 * Parse arguments and load the daemon configuration file.
5525 * We have an exit_options exit path to free memory reserved by
5526 * set_options. This is needed because the rest of sessiond_cleanup()
5527 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5528 * depends on set_options.
5531 if (set_options(argc
, argv
)) {
5536 /* Init config from environment variables. */
5537 sessiond_config_apply_env_config(&config
);
5540 * Resolve all paths received as arguments, configuration option, or
5541 * through environment variable as absolute paths. This is necessary
5542 * since daemonizing causes the sessiond's current working directory
5545 ret
= sessiond_config_resolve_paths(&config
);
5551 lttng_opt_verbose
= config
.verbose
;
5552 lttng_opt_quiet
= config
.quiet
;
5553 kconsumer_data
.err_unix_sock_path
=
5554 config
.kconsumerd_err_unix_sock_path
.value
;
5555 kconsumer_data
.cmd_unix_sock_path
=
5556 config
.kconsumerd_cmd_unix_sock_path
.value
;
5557 ustconsumer32_data
.err_unix_sock_path
=
5558 config
.consumerd32_err_unix_sock_path
.value
;
5559 ustconsumer32_data
.cmd_unix_sock_path
=
5560 config
.consumerd32_cmd_unix_sock_path
.value
;
5561 ustconsumer64_data
.err_unix_sock_path
=
5562 config
.consumerd64_err_unix_sock_path
.value
;
5563 ustconsumer64_data
.cmd_unix_sock_path
=
5564 config
.consumerd64_cmd_unix_sock_path
.value
;
5565 set_clock_plugin_env();
5567 sessiond_config_log(&config
);
5569 if (create_lttng_rundir()) {
5574 /* Abort launch if a session daemon is already running. */
5575 if (check_existing_daemon()) {
5576 ERR("A session daemon is already running.");
5582 if (config
.daemonize
|| config
.background
) {
5585 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5586 !config
.background
);
5593 * We are in the child. Make sure all other file descriptors are
5594 * closed, in case we are called with more opened file
5595 * descriptors than the standard ones and the lock file.
5597 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5598 if (i
== lockfile_fd
) {
5605 if (run_as_create_worker(argv
[0]) < 0) {
5606 goto exit_create_run_as_worker_cleanup
;
5610 * Starting from here, we can create threads. This needs to be after
5611 * lttng_daemonize due to RCU.
5615 * Initialize the health check subsystem. This call should set the
5616 * appropriate time values.
5618 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5619 if (!health_sessiond
) {
5620 PERROR("health_app_create error");
5622 goto exit_health_sessiond_cleanup
;
5625 /* Create thread to clean up RCU hash tables */
5626 if (init_ht_cleanup_thread(&ht_cleanup_thread
)) {
5628 goto exit_ht_cleanup
;
5631 /* Create thread quit pipe */
5632 if (init_thread_quit_pipe()) {
5634 goto exit_init_data
;
5637 /* Check if daemon is UID = 0 */
5638 is_root
= !getuid();
5640 /* Create global run dir with root access */
5642 kernel_channel_monitor_pipe
= lttng_pipe_open(0);
5643 if (!kernel_channel_monitor_pipe
) {
5644 ERR("Failed to create kernel consumer channel monitor pipe");
5646 goto exit_init_data
;
5648 kconsumer_data
.channel_monitor_pipe
=
5649 lttng_pipe_release_writefd(
5650 kernel_channel_monitor_pipe
);
5651 if (kconsumer_data
.channel_monitor_pipe
< 0) {
5653 goto exit_init_data
;
5657 /* Set consumer initial state */
5658 kernel_consumerd_state
= CONSUMER_STOPPED
;
5659 ust_consumerd_state
= CONSUMER_STOPPED
;
5661 ust32_channel_monitor_pipe
= lttng_pipe_open(0);
5662 if (!ust32_channel_monitor_pipe
) {
5663 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5665 goto exit_init_data
;
5667 ustconsumer32_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
5668 ust32_channel_monitor_pipe
);
5669 if (ustconsumer32_data
.channel_monitor_pipe
< 0) {
5671 goto exit_init_data
;
5674 ust64_channel_monitor_pipe
= lttng_pipe_open(0);
5675 if (!ust64_channel_monitor_pipe
) {
5676 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5678 goto exit_init_data
;
5680 ustconsumer64_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
5681 ust64_channel_monitor_pipe
);
5682 if (ustconsumer64_data
.channel_monitor_pipe
< 0) {
5684 goto exit_init_data
;
5688 * Init UST app hash table. Alloc hash table before this point since
5689 * cleanup() can get called after that point.
5691 if (ust_app_ht_alloc()) {
5692 ERR("Failed to allocate UST app hash table");
5694 goto exit_init_data
;
5698 * Initialize agent app hash table. We allocate the hash table here
5699 * since cleanup() can get called after this point.
5701 if (agent_app_ht_alloc()) {
5702 ERR("Failed to allocate Agent app hash table");
5704 goto exit_init_data
;
5708 * These actions must be executed as root. We do that *after* setting up
5709 * the sockets path because we MUST make the check for another daemon using
5710 * those paths *before* trying to set the kernel consumer sockets and init
5714 if (set_consumer_sockets(&kconsumer_data
)) {
5716 goto exit_init_data
;
5719 /* Setup kernel tracer */
5720 if (!config
.no_kernel
) {
5721 init_kernel_tracer();
5722 if (kernel_tracer_fd
>= 0) {
5723 ret
= syscall_init_table();
5725 ERR("Unable to populate syscall table. "
5726 "Syscall tracing won't work "
5727 "for this session daemon.");
5732 /* Set ulimit for open files */
5735 /* init lttng_fd tracking must be done after set_ulimit. */
5738 if (set_consumer_sockets(&ustconsumer64_data
)) {
5740 goto exit_init_data
;
5743 if (set_consumer_sockets(&ustconsumer32_data
)) {
5745 goto exit_init_data
;
5748 /* Setup the needed unix socket */
5749 if (init_daemon_socket()) {
5751 goto exit_init_data
;
5754 /* Set credentials to socket */
5755 if (is_root
&& set_permissions(config
.rundir
.value
)) {
5757 goto exit_init_data
;
5760 /* Get parent pid if -S, --sig-parent is specified. */
5761 if (config
.sig_parent
) {
5765 /* Setup the kernel pipe for waking up the kernel thread */
5766 if (is_root
&& !config
.no_kernel
) {
5767 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5769 goto exit_init_data
;
5773 /* Setup the thread apps communication pipe. */
5774 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5776 goto exit_init_data
;
5779 /* Setup the thread apps notify communication pipe. */
5780 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5782 goto exit_init_data
;
5785 /* Initialize global buffer per UID and PID registry. */
5786 buffer_reg_init_uid_registry();
5787 buffer_reg_init_pid_registry();
5789 /* Init UST command queue. */
5790 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5793 * Get session list pointer. This pointer MUST NOT be free'd. This list
5794 * is statically declared in session.c
5796 session_list_ptr
= session_get_list();
5800 /* Check for the application socket timeout env variable. */
5801 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5802 if (env_app_timeout
) {
5803 config
.app_socket_timeout
= atoi(env_app_timeout
);
5805 config
.app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5808 ret
= write_pidfile();
5810 ERR("Error in write_pidfile");
5812 goto exit_init_data
;
5814 ret
= write_agent_port();
5816 ERR("Error in write_agent_port");
5818 goto exit_init_data
;
5821 /* Initialize communication library */
5823 /* Initialize TCP timeout values */
5824 lttcomm_inet_init();
5826 if (load_session_init_data(&load_info
) < 0) {
5828 goto exit_init_data
;
5830 load_info
->path
= config
.load_session_path
.value
;
5832 /* Create health-check thread. */
5833 ret
= pthread_create(&health_thread
, default_pthread_attr(),
5834 thread_manage_health
, (void *) NULL
);
5837 PERROR("pthread_create health");
5842 /* notification_thread_data acquires the pipes' read side. */
5843 notification_thread_handle
= notification_thread_handle_create(
5844 ust32_channel_monitor_pipe
,
5845 ust64_channel_monitor_pipe
,
5846 kernel_channel_monitor_pipe
);
5847 if (!notification_thread_handle
) {
5849 ERR("Failed to create notification thread shared data");
5851 goto exit_notification
;
5854 /* Create notification thread. */
5855 ret
= pthread_create(¬ification_thread
, default_pthread_attr(),
5856 thread_notification
, notification_thread_handle
);
5859 PERROR("pthread_create notification");
5862 goto exit_notification
;
5864 notification_thread_running
= true;
5866 /* Create thread to manage the client socket */
5867 ret
= pthread_create(&client_thread
, default_pthread_attr(),
5868 thread_manage_clients
, (void *) NULL
);
5871 PERROR("pthread_create clients");
5877 /* Create thread to dispatch registration */
5878 ret
= pthread_create(&dispatch_thread
, default_pthread_attr(),
5879 thread_dispatch_ust_registration
, (void *) NULL
);
5882 PERROR("pthread_create dispatch");
5888 /* Create thread to manage application registration. */
5889 ret
= pthread_create(®_apps_thread
, default_pthread_attr(),
5890 thread_registration_apps
, (void *) NULL
);
5893 PERROR("pthread_create registration");
5899 /* Create thread to manage application socket */
5900 ret
= pthread_create(&apps_thread
, default_pthread_attr(),
5901 thread_manage_apps
, (void *) NULL
);
5904 PERROR("pthread_create apps");
5910 /* Create thread to manage application notify socket */
5911 ret
= pthread_create(&apps_notify_thread
, default_pthread_attr(),
5912 ust_thread_manage_notify
, (void *) NULL
);
5915 PERROR("pthread_create notify");
5918 goto exit_apps_notify
;
5921 /* Create agent registration thread. */
5922 ret
= pthread_create(&agent_reg_thread
, default_pthread_attr(),
5923 agent_thread_manage_registration
, (void *) NULL
);
5926 PERROR("pthread_create agent");
5929 goto exit_agent_reg
;
5932 /* Don't start this thread if kernel tracing is not requested nor root */
5933 if (is_root
&& !config
.no_kernel
) {
5934 /* Create kernel thread to manage kernel event */
5935 ret
= pthread_create(&kernel_thread
, default_pthread_attr(),
5936 thread_manage_kernel
, (void *) NULL
);
5939 PERROR("pthread_create kernel");
5946 /* Create session loading thread. */
5947 ret
= pthread_create(&load_session_thread
, default_pthread_attr(),
5948 thread_load_session
, load_info
);
5951 PERROR("pthread_create load_session_thread");
5954 goto exit_load_session
;
5958 * This is where we start awaiting program completion (e.g. through
5959 * signal that asks threads to teardown).
5962 ret
= pthread_join(load_session_thread
, &status
);
5965 PERROR("pthread_join load_session_thread");
5970 if (is_root
&& !config
.no_kernel
) {
5971 ret
= pthread_join(kernel_thread
, &status
);
5974 PERROR("pthread_join");
5980 ret
= pthread_join(agent_reg_thread
, &status
);
5983 PERROR("pthread_join agent");
5988 ret
= pthread_join(apps_notify_thread
, &status
);
5991 PERROR("pthread_join apps notify");
5996 ret
= pthread_join(apps_thread
, &status
);
5999 PERROR("pthread_join apps");
6004 ret
= pthread_join(reg_apps_thread
, &status
);
6007 PERROR("pthread_join");
6013 * Join dispatch thread after joining reg_apps_thread to ensure
6014 * we don't leak applications in the queue.
6016 ret
= pthread_join(dispatch_thread
, &status
);
6019 PERROR("pthread_join");
6024 ret
= pthread_join(client_thread
, &status
);
6027 PERROR("pthread_join");
6033 ret
= pthread_join(health_thread
, &status
);
6036 PERROR("pthread_join health thread");
6043 * Wait for all pending call_rcu work to complete before tearing
6044 * down data structures. call_rcu worker may be trying to
6045 * perform lookups in those structures.
6049 * sessiond_cleanup() is called when no other thread is running, except
6050 * the ht_cleanup thread, which is needed to destroy the hash tables.
6052 rcu_thread_online();
6056 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6057 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6058 * the queue is empty before shutting down the clean-up thread.
6063 * The teardown of the notification system is performed after the
6064 * session daemon's teardown in order to allow it to be notified
6065 * of the active session and channels at the moment of the teardown.
6067 if (notification_thread_handle
) {
6068 if (notification_thread_running
) {
6069 notification_thread_command_quit(
6070 notification_thread_handle
);
6071 ret
= pthread_join(notification_thread
, &status
);
6074 PERROR("pthread_join notification thread");
6078 notification_thread_handle_destroy(notification_thread_handle
);
6081 rcu_thread_offline();
6082 rcu_unregister_thread();
6084 ret
= fini_ht_cleanup_thread(&ht_cleanup_thread
);
6088 lttng_pipe_destroy(ust32_channel_monitor_pipe
);
6089 lttng_pipe_destroy(ust64_channel_monitor_pipe
);
6090 lttng_pipe_destroy(kernel_channel_monitor_pipe
);
6093 health_app_destroy(health_sessiond
);
6094 exit_health_sessiond_cleanup
:
6095 exit_create_run_as_worker_cleanup
:
6098 sessiond_cleanup_lock_file();
6099 sessiond_cleanup_options();
6101 exit_set_signal_handler
: