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"
76 #include "rotation-thread.h"
79 #include "ht-cleanup.h"
80 #include "sessiond-timer.h"
82 #define CONSUMERD_FILE "lttng-consumerd"
84 static const char *help_msg
=
85 #ifdef LTTNG_EMBED_HELP
86 #include <lttng-sessiond.8.h>
93 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
94 static int tracing_group_name_override
;
95 static char *opt_pidfile
;
96 static int opt_sig_parent
;
97 static int opt_verbose_consumer
;
98 static int opt_daemon
, opt_background
;
99 static int opt_no_kernel
;
100 static char *opt_load_session_path
;
101 static pid_t ppid
; /* Parent PID for --sig-parent option */
102 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
104 static int lockfile_fd
= -1;
106 /* Set to 1 when a SIGUSR1 signal is received. */
107 static int recv_child_signal
;
110 * Consumer daemon specific control data. Every value not initialized here is
111 * set to 0 by the static definition.
113 static struct consumer_data kconsumer_data
= {
114 .type
= LTTNG_CONSUMER_KERNEL
,
115 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
116 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
119 .channel_monitor_pipe
= -1,
120 .channel_rotate_pipe
= -1,
121 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
122 .lock
= PTHREAD_MUTEX_INITIALIZER
,
123 .cond
= PTHREAD_COND_INITIALIZER
,
124 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
126 static struct consumer_data ustconsumer64_data
= {
127 .type
= LTTNG_CONSUMER64_UST
,
128 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
129 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
132 .channel_monitor_pipe
= -1,
133 .channel_rotate_pipe
= -1,
134 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
135 .lock
= PTHREAD_MUTEX_INITIALIZER
,
136 .cond
= PTHREAD_COND_INITIALIZER
,
137 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
139 static struct consumer_data ustconsumer32_data
= {
140 .type
= LTTNG_CONSUMER32_UST
,
141 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
142 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
145 .channel_monitor_pipe
= -1,
146 .channel_rotate_pipe
= -1,
147 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
148 .lock
= PTHREAD_MUTEX_INITIALIZER
,
149 .cond
= PTHREAD_COND_INITIALIZER
,
150 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
153 /* Command line options */
154 static const struct option long_options
[] = {
155 { "client-sock", required_argument
, 0, 'c' },
156 { "apps-sock", required_argument
, 0, 'a' },
157 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
158 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
159 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
160 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
161 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
162 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
163 { "consumerd32-path", required_argument
, 0, '\0' },
164 { "consumerd32-libdir", required_argument
, 0, '\0' },
165 { "consumerd64-path", required_argument
, 0, '\0' },
166 { "consumerd64-libdir", required_argument
, 0, '\0' },
167 { "daemonize", no_argument
, 0, 'd' },
168 { "background", no_argument
, 0, 'b' },
169 { "sig-parent", no_argument
, 0, 'S' },
170 { "help", no_argument
, 0, 'h' },
171 { "group", required_argument
, 0, 'g' },
172 { "version", no_argument
, 0, 'V' },
173 { "quiet", no_argument
, 0, 'q' },
174 { "verbose", no_argument
, 0, 'v' },
175 { "verbose-consumer", no_argument
, 0, '\0' },
176 { "no-kernel", no_argument
, 0, '\0' },
177 { "pidfile", required_argument
, 0, 'p' },
178 { "agent-tcp-port", required_argument
, 0, '\0' },
179 { "config", required_argument
, 0, 'f' },
180 { "load", required_argument
, 0, 'l' },
181 { "kmod-probes", required_argument
, 0, '\0' },
182 { "extra-kmod-probes", required_argument
, 0, '\0' },
186 /* Command line options to ignore from configuration file */
187 static const char *config_ignore_options
[] = { "help", "version", "config" };
189 /* Shared between threads */
190 static int dispatch_thread_exit
;
192 /* Global application Unix socket path */
193 static char apps_unix_sock_path
[PATH_MAX
];
194 /* Global client Unix socket path */
195 static char client_unix_sock_path
[PATH_MAX
];
196 /* global wait shm path for UST */
197 static char wait_shm_path
[PATH_MAX
];
198 /* Global health check unix path */
199 static char health_unix_sock_path
[PATH_MAX
];
201 /* Sockets and FDs */
202 static int client_sock
= -1;
203 static int apps_sock
= -1;
204 int kernel_tracer_fd
= -1;
205 static int kernel_poll_pipe
[2] = { -1, -1 };
208 * Quit pipe for all threads. This permits a single cancellation point
209 * for all threads when receiving an event on the pipe.
211 static int thread_quit_pipe
[2] = { -1, -1 };
214 * This pipe is used to inform the thread managing application communication
215 * that a command is queued and ready to be processed.
217 static int apps_cmd_pipe
[2] = { -1, -1 };
219 int apps_cmd_notify_pipe
[2] = { -1, -1 };
222 * Pipe to wakeup the rotation thread when a timer related to the session
223 * rotation feature fires.
225 static int rotate_timer_pipe
[2] = { -1, -1 };
227 /* Pthread, Mutexes and Semaphores */
228 static pthread_t apps_thread
;
229 static pthread_t apps_notify_thread
;
230 static pthread_t reg_apps_thread
;
231 static pthread_t client_thread
;
232 static pthread_t kernel_thread
;
233 static pthread_t dispatch_thread
;
234 static pthread_t health_thread
;
235 static pthread_t ht_cleanup_thread
;
236 static pthread_t agent_reg_thread
;
237 static pthread_t load_session_thread
;
238 static pthread_t notification_thread
;
239 static pthread_t rotation_thread
;
240 static pthread_t timer_thread
;
243 * UST registration command queue. This queue is tied with a futex and uses a N
244 * wakers / 1 waiter implemented and detailed in futex.c/.h
246 * The thread_registration_apps and thread_dispatch_ust_registration uses this
247 * queue along with the wait/wake scheme. The thread_manage_apps receives down
248 * the line new application socket and monitors it for any I/O error or clean
249 * close that triggers an unregistration of the application.
251 static struct ust_cmd_queue ust_cmd_queue
;
254 * Pointer initialized before thread creation.
256 * This points to the tracing session list containing the session count and a
257 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
258 * MUST NOT be taken if you call a public function in session.c.
260 * The lock is nested inside the structure: session_list_ptr->lock. Please use
261 * session_lock_list and session_unlock_list for lock acquisition.
263 static struct ltt_session_list
*session_list_ptr
;
265 int ust_consumerd64_fd
= -1;
266 int ust_consumerd32_fd
= -1;
268 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
269 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
270 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
271 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
272 static int consumerd32_bin_override
;
273 static int consumerd64_bin_override
;
274 static int consumerd32_libdir_override
;
275 static int consumerd64_libdir_override
;
277 static const char *module_proc_lttng
= "/proc/lttng";
280 * Consumer daemon state which is changed when spawning it, killing it or in
281 * case of a fatal error.
283 enum consumerd_state
{
284 CONSUMER_STARTED
= 1,
285 CONSUMER_STOPPED
= 2,
290 * This consumer daemon state is used to validate if a client command will be
291 * able to reach the consumer. If not, the client is informed. For instance,
292 * doing a "lttng start" when the consumer state is set to ERROR will return an
293 * error to the client.
295 * The following example shows a possible race condition of this scheme:
297 * consumer thread error happens
299 * client cmd checks state -> still OK
300 * consumer thread exit, sets error
301 * client cmd try to talk to consumer
304 * However, since the consumer is a different daemon, we have no way of making
305 * sure the command will reach it safely even with this state flag. This is why
306 * we consider that up to the state validation during command processing, the
307 * command is safe. After that, we can not guarantee the correctness of the
308 * client request vis-a-vis the consumer.
310 static enum consumerd_state ust_consumerd_state
;
311 static enum consumerd_state kernel_consumerd_state
;
314 * Socket timeout for receiving and sending in seconds.
316 static int app_socket_timeout
;
318 /* Set in main() with the current page size. */
321 /* Application health monitoring */
322 struct health_app
*health_sessiond
;
324 /* Agent TCP port for registration. Used by the agent thread. */
325 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
327 /* Am I root or not. */
328 int is_root
; /* Set to 1 if the daemon is running as root */
330 const char * const config_section_name
= "sessiond";
332 /* Load session thread information to operate. */
333 struct load_session_thread_data
*load_info
;
335 /* Notification thread handle. */
336 struct notification_thread_handle
*notification_thread_handle
;
338 /* Rotation thread handle. */
339 struct rotation_thread_handle
*rotation_thread_handle
;
341 /* Global hash tables */
342 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
345 * Whether sessiond is ready for commands/notification channel/health check
347 * NR_LTTNG_SESSIOND_READY must match the number of calls to
348 * sessiond_notify_ready().
350 #define NR_LTTNG_SESSIOND_READY 5
351 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
353 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
355 return (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) ? 1 : 0;
358 /* Notify parents that we are ready for cmd and health check */
360 void sessiond_notify_ready(void)
362 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
364 * Notify parent pid that we are ready to accept command
365 * for client side. This ppid is the one from the
366 * external process that spawned us.
368 if (opt_sig_parent
) {
373 * Notify the parent of the fork() process that we are
376 if (opt_daemon
|| opt_background
) {
377 kill(child_ppid
, SIGUSR1
);
383 void setup_consumerd_path(void)
385 const char *bin
, *libdir
;
388 * Allow INSTALL_BIN_PATH to be used as a target path for the
389 * native architecture size consumer if CONFIG_CONSUMER*_PATH
390 * has not been defined.
392 #if (CAA_BITS_PER_LONG == 32)
393 if (!consumerd32_bin
[0]) {
394 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
396 if (!consumerd32_libdir
[0]) {
397 consumerd32_libdir
= INSTALL_LIB_PATH
;
399 #elif (CAA_BITS_PER_LONG == 64)
400 if (!consumerd64_bin
[0]) {
401 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
403 if (!consumerd64_libdir
[0]) {
404 consumerd64_libdir
= INSTALL_LIB_PATH
;
407 #error "Unknown bitness"
411 * runtime env. var. overrides the build default.
413 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
415 consumerd32_bin
= bin
;
417 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
419 consumerd64_bin
= bin
;
421 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
423 consumerd32_libdir
= libdir
;
425 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
427 consumerd64_libdir
= libdir
;
432 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
439 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
445 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
457 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
459 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
461 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
465 * Init thread quit pipe.
467 * Return -1 on error or 0 if all pipes are created.
469 static int __init_thread_quit_pipe(int *a_pipe
)
475 PERROR("thread quit pipe");
479 for (i
= 0; i
< 2; i
++) {
480 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
491 static int init_thread_quit_pipe(void)
493 return __init_thread_quit_pipe(thread_quit_pipe
);
497 * Stop all threads by closing the thread quit pipe.
499 static void stop_threads(void)
503 /* Stopping all threads */
504 DBG("Terminating all threads");
505 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
507 ERR("write error on thread quit pipe");
510 /* Dispatch thread */
511 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
512 futex_nto1_wake(&ust_cmd_queue
.futex
);
516 * Close every consumer sockets.
518 static void close_consumer_sockets(void)
522 if (kconsumer_data
.err_sock
>= 0) {
523 ret
= close(kconsumer_data
.err_sock
);
525 PERROR("kernel consumer err_sock close");
528 if (ustconsumer32_data
.err_sock
>= 0) {
529 ret
= close(ustconsumer32_data
.err_sock
);
531 PERROR("UST consumerd32 err_sock close");
534 if (ustconsumer64_data
.err_sock
>= 0) {
535 ret
= close(ustconsumer64_data
.err_sock
);
537 PERROR("UST consumerd64 err_sock close");
540 if (kconsumer_data
.cmd_sock
>= 0) {
541 ret
= close(kconsumer_data
.cmd_sock
);
543 PERROR("kernel consumer cmd_sock close");
546 if (ustconsumer32_data
.cmd_sock
>= 0) {
547 ret
= close(ustconsumer32_data
.cmd_sock
);
549 PERROR("UST consumerd32 cmd_sock close");
552 if (ustconsumer64_data
.cmd_sock
>= 0) {
553 ret
= close(ustconsumer64_data
.cmd_sock
);
555 PERROR("UST consumerd64 cmd_sock close");
558 if (kconsumer_data
.channel_monitor_pipe
>= 0) {
559 ret
= close(kconsumer_data
.channel_monitor_pipe
);
561 PERROR("kernel consumer channel monitor pipe close");
564 if (ustconsumer32_data
.channel_monitor_pipe
>= 0) {
565 ret
= close(ustconsumer32_data
.channel_monitor_pipe
);
567 PERROR("UST consumerd32 channel monitor pipe close");
570 if (ustconsumer64_data
.channel_monitor_pipe
>= 0) {
571 ret
= close(ustconsumer64_data
.channel_monitor_pipe
);
573 PERROR("UST consumerd64 channel monitor pipe close");
576 if (kconsumer_data
.channel_rotate_pipe
>= 0) {
577 ret
= close(kconsumer_data
.channel_rotate_pipe
);
579 PERROR("kernel consumer channel rotate pipe close");
582 if (ustconsumer32_data
.channel_rotate_pipe
>= 0) {
583 ret
= close(ustconsumer32_data
.channel_rotate_pipe
);
585 PERROR("UST consumerd32 channel rotate pipe close");
588 if (ustconsumer64_data
.channel_rotate_pipe
>= 0) {
589 ret
= close(ustconsumer64_data
.channel_rotate_pipe
);
591 PERROR("UST consumerd64 channel rotate pipe close");
597 * Generate the full lock file path using the rundir.
599 * Return the snprintf() return value thus a negative value is an error.
601 static int generate_lock_file_path(char *path
, size_t len
)
608 /* Build lockfile path from rundir. */
609 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
611 PERROR("snprintf lockfile path");
618 * Wait on consumer process termination.
620 * Need to be called with the consumer data lock held or from a context
621 * ensuring no concurrent access to data (e.g: cleanup).
623 static void wait_consumer(struct consumer_data
*consumer_data
)
628 if (consumer_data
->pid
<= 0) {
632 DBG("Waiting for complete teardown of consumerd (PID: %d)",
634 ret
= waitpid(consumer_data
->pid
, &status
, 0);
636 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
637 } else if (!WIFEXITED(status
)) {
638 ERR("consumerd termination with error: %d",
641 consumer_data
->pid
= 0;
645 * Cleanup the session daemon's data structures.
647 static void sessiond_cleanup(void)
650 struct ltt_session
*sess
, *stmp
;
653 DBG("Cleanup sessiond");
656 * Close the thread quit pipe. It has already done its job,
657 * since we are now called.
659 utils_close_pipe(thread_quit_pipe
);
662 * If opt_pidfile is undefined, the default file will be wiped when
663 * removing the rundir.
666 ret
= remove(opt_pidfile
);
668 PERROR("remove pidfile %s", opt_pidfile
);
672 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
675 snprintf(path
, PATH_MAX
,
677 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
678 DBG("Removing %s", path
);
681 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
682 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
683 DBG("Removing %s", path
);
687 snprintf(path
, PATH_MAX
,
688 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
690 DBG("Removing %s", path
);
693 snprintf(path
, PATH_MAX
,
694 DEFAULT_KCONSUMERD_PATH
,
696 DBG("Removing directory %s", path
);
699 /* ust consumerd 32 */
700 snprintf(path
, PATH_MAX
,
701 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
703 DBG("Removing %s", path
);
706 snprintf(path
, PATH_MAX
,
707 DEFAULT_USTCONSUMERD32_PATH
,
709 DBG("Removing directory %s", path
);
712 /* ust consumerd 64 */
713 snprintf(path
, PATH_MAX
,
714 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
716 DBG("Removing %s", path
);
719 snprintf(path
, PATH_MAX
,
720 DEFAULT_USTCONSUMERD64_PATH
,
722 DBG("Removing directory %s", path
);
725 DBG("Cleaning up all sessions");
727 /* Destroy session list mutex */
728 if (session_list_ptr
!= NULL
) {
729 pthread_mutex_destroy(&session_list_ptr
->lock
);
731 /* Cleanup ALL session */
732 cds_list_for_each_entry_safe(sess
, stmp
,
733 &session_list_ptr
->head
, list
) {
734 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
738 wait_consumer(&kconsumer_data
);
739 wait_consumer(&ustconsumer64_data
);
740 wait_consumer(&ustconsumer32_data
);
742 DBG("Cleaning up all agent apps");
743 agent_app_ht_clean();
745 DBG("Closing all UST sockets");
746 ust_app_clean_list();
747 buffer_reg_destroy_registries();
749 if (is_root
&& !opt_no_kernel
) {
750 DBG2("Closing kernel fd");
751 if (kernel_tracer_fd
>= 0) {
752 ret
= close(kernel_tracer_fd
);
757 DBG("Unloading kernel modules");
758 modprobe_remove_lttng_all();
762 close_consumer_sockets();
765 load_session_destroy_data(load_info
);
770 * Cleanup lock file by deleting it and finaly closing it which will
771 * release the file system lock.
773 if (lockfile_fd
>= 0) {
774 char lockfile_path
[PATH_MAX
];
776 ret
= generate_lock_file_path(lockfile_path
,
777 sizeof(lockfile_path
));
779 ret
= remove(lockfile_path
);
781 PERROR("remove lock file");
783 ret
= close(lockfile_fd
);
785 PERROR("close lock file");
791 * We do NOT rmdir rundir because there are other processes
792 * using it, for instance lttng-relayd, which can start in
793 * parallel with this teardown.
800 * Cleanup the daemon's option data structures.
802 static void sessiond_cleanup_options(void)
804 DBG("Cleaning up options");
807 * If the override option is set, the pointer points to a *non* const
808 * thus freeing it even though the variable type is set to const.
810 if (tracing_group_name_override
) {
811 free((void *) tracing_group_name
);
813 if (consumerd32_bin_override
) {
814 free((void *) consumerd32_bin
);
816 if (consumerd64_bin_override
) {
817 free((void *) consumerd64_bin
);
819 if (consumerd32_libdir_override
) {
820 free((void *) consumerd32_libdir
);
822 if (consumerd64_libdir_override
) {
823 free((void *) consumerd64_libdir
);
827 free(opt_load_session_path
);
828 free(kmod_probes_list
);
829 free(kmod_extra_probes_list
);
831 run_as_destroy_worker();
835 * Send data on a unix socket using the liblttsessiondcomm API.
837 * Return lttcomm error code.
839 static int send_unix_sock(int sock
, void *buf
, size_t len
)
841 /* Check valid length */
846 return lttcomm_send_unix_sock(sock
, buf
, len
);
850 * Free memory of a command context structure.
852 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
854 DBG("Clean command context structure");
856 if ((*cmd_ctx
)->llm
) {
857 free((*cmd_ctx
)->llm
);
859 if ((*cmd_ctx
)->lsm
) {
860 free((*cmd_ctx
)->lsm
);
868 * Notify UST applications using the shm mmap futex.
870 static int notify_ust_apps(int active
)
874 DBG("Notifying applications of session daemon state: %d", active
);
876 /* See shm.c for this call implying mmap, shm and futex calls */
877 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
878 if (wait_shm_mmap
== NULL
) {
882 /* Wake waiting process */
883 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
885 /* Apps notified successfully */
893 * Setup the outgoing data buffer for the response (llm) by allocating the
894 * right amount of memory and copying the original information from the lsm
897 * Return 0 on success, negative value on error.
899 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
900 const void *payload_buf
, size_t payload_len
,
901 const void *cmd_header_buf
, size_t cmd_header_len
)
904 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
905 const size_t cmd_header_offset
= header_len
;
906 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
907 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
909 cmd_ctx
->llm
= zmalloc(total_msg_size
);
911 if (cmd_ctx
->llm
== NULL
) {
917 /* Copy common data */
918 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
919 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
920 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
921 cmd_ctx
->llm
->data_size
= payload_len
;
922 cmd_ctx
->lttng_msg_size
= total_msg_size
;
924 /* Copy command header */
925 if (cmd_header_len
) {
926 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
932 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
941 * Version of setup_lttng_msg() without command header.
943 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
944 void *payload_buf
, size_t payload_len
)
946 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
949 * Update the kernel poll set of all channel fd available over all tracing
950 * session. Add the wakeup pipe at the end of the set.
952 static int update_kernel_poll(struct lttng_poll_event
*events
)
955 struct ltt_session
*session
;
956 struct ltt_kernel_channel
*channel
;
958 DBG("Updating kernel poll set");
961 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
962 session_lock(session
);
963 if (session
->kernel_session
== NULL
) {
964 session_unlock(session
);
968 cds_list_for_each_entry(channel
,
969 &session
->kernel_session
->channel_list
.head
, list
) {
970 /* Add channel fd to the kernel poll set */
971 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
973 session_unlock(session
);
976 DBG("Channel fd %d added to kernel set", channel
->fd
);
978 session_unlock(session
);
980 session_unlock_list();
985 session_unlock_list();
990 * Find the channel fd from 'fd' over all tracing session. When found, check
991 * for new channel stream and send those stream fds to the kernel consumer.
993 * Useful for CPU hotplug feature.
995 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
998 struct ltt_session
*session
;
999 struct ltt_kernel_session
*ksess
;
1000 struct ltt_kernel_channel
*channel
;
1002 DBG("Updating kernel streams for channel fd %d", fd
);
1004 session_lock_list();
1005 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
1006 session_lock(session
);
1007 if (session
->kernel_session
== NULL
) {
1008 session_unlock(session
);
1011 ksess
= session
->kernel_session
;
1013 cds_list_for_each_entry(channel
,
1014 &ksess
->channel_list
.head
, list
) {
1015 struct lttng_ht_iter iter
;
1016 struct consumer_socket
*socket
;
1018 if (channel
->fd
!= fd
) {
1021 DBG("Channel found, updating kernel streams");
1022 ret
= kernel_open_channel_stream(channel
);
1026 /* Update the stream global counter */
1027 ksess
->stream_count_global
+= ret
;
1030 * Have we already sent fds to the consumer? If yes, it
1031 * means that tracing is started so it is safe to send
1032 * our updated stream fds.
1034 if (ksess
->consumer_fds_sent
!= 1
1035 || ksess
->consumer
== NULL
) {
1041 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
1042 &iter
.iter
, socket
, node
.node
) {
1043 pthread_mutex_lock(socket
->lock
);
1044 ret
= kernel_consumer_send_channel_stream(socket
,
1046 session
->output_traces
? 1 : 0);
1047 pthread_mutex_unlock(socket
->lock
);
1055 session_unlock(session
);
1057 session_unlock_list();
1061 session_unlock(session
);
1062 session_unlock_list();
1067 * For each tracing session, update newly registered apps. The session list
1068 * lock MUST be acquired before calling this.
1070 static void update_ust_app(int app_sock
)
1072 struct ltt_session
*sess
, *stmp
;
1074 /* Consumer is in an ERROR state. Stop any application update. */
1075 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1076 /* Stop the update process since the consumer is dead. */
1080 /* For all tracing session(s) */
1081 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1082 struct ust_app
*app
;
1085 if (!sess
->ust_session
) {
1086 goto unlock_session
;
1090 assert(app_sock
>= 0);
1091 app
= ust_app_find_by_sock(app_sock
);
1094 * Application can be unregistered before so
1095 * this is possible hence simply stopping the
1098 DBG3("UST app update failed to find app sock %d",
1102 ust_app_global_update(sess
->ust_session
, app
);
1106 session_unlock(sess
);
1111 * This thread manage event coming from the kernel.
1113 * Features supported in this thread:
1116 static void *thread_manage_kernel(void *data
)
1118 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1119 uint32_t revents
, nb_fd
;
1121 struct lttng_poll_event events
;
1123 DBG("[thread] Thread manage kernel started");
1125 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1128 * This first step of the while is to clean this structure which could free
1129 * non NULL pointers so initialize it before the loop.
1131 lttng_poll_init(&events
);
1133 if (testpoint(sessiond_thread_manage_kernel
)) {
1134 goto error_testpoint
;
1137 health_code_update();
1139 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1140 goto error_testpoint
;
1144 health_code_update();
1146 if (update_poll_flag
== 1) {
1147 /* Clean events object. We are about to populate it again. */
1148 lttng_poll_clean(&events
);
1150 ret
= sessiond_set_thread_pollset(&events
, 2);
1152 goto error_poll_create
;
1155 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1160 /* This will add the available kernel channel if any. */
1161 ret
= update_kernel_poll(&events
);
1165 update_poll_flag
= 0;
1168 DBG("Thread kernel polling");
1170 /* Poll infinite value of time */
1172 health_poll_entry();
1173 ret
= lttng_poll_wait(&events
, -1);
1174 DBG("Thread kernel return from poll on %d fds",
1175 LTTNG_POLL_GETNB(&events
));
1179 * Restart interrupted system call.
1181 if (errno
== EINTR
) {
1185 } else if (ret
== 0) {
1186 /* Should not happen since timeout is infinite */
1187 ERR("Return value of poll is 0 with an infinite timeout.\n"
1188 "This should not have happened! Continuing...");
1194 for (i
= 0; i
< nb_fd
; i
++) {
1195 /* Fetch once the poll data */
1196 revents
= LTTNG_POLL_GETEV(&events
, i
);
1197 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1199 health_code_update();
1202 /* No activity for this FD (poll implementation). */
1206 /* Thread quit pipe has been closed. Killing thread. */
1207 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1213 /* Check for data on kernel pipe */
1214 if (revents
& LPOLLIN
) {
1215 if (pollfd
== kernel_poll_pipe
[0]) {
1216 (void) lttng_read(kernel_poll_pipe
[0],
1219 * Ret value is useless here, if this pipe gets any actions an
1220 * update is required anyway.
1222 update_poll_flag
= 1;
1226 * New CPU detected by the kernel. Adding kernel stream to
1227 * kernel session and updating the kernel consumer
1229 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1235 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1236 update_poll_flag
= 1;
1239 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1247 lttng_poll_clean(&events
);
1250 utils_close_pipe(kernel_poll_pipe
);
1251 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1254 ERR("Health error occurred in %s", __func__
);
1255 WARN("Kernel thread died unexpectedly. "
1256 "Kernel tracing can continue but CPU hotplug is disabled.");
1258 health_unregister(health_sessiond
);
1259 DBG("Kernel thread dying");
1264 * Signal pthread condition of the consumer data that the thread.
1266 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1268 pthread_mutex_lock(&data
->cond_mutex
);
1271 * The state is set before signaling. It can be any value, it's the waiter
1272 * job to correctly interpret this condition variable associated to the
1273 * consumer pthread_cond.
1275 * A value of 0 means that the corresponding thread of the consumer data
1276 * was not started. 1 indicates that the thread has started and is ready
1277 * for action. A negative value means that there was an error during the
1280 data
->consumer_thread_is_ready
= state
;
1281 (void) pthread_cond_signal(&data
->cond
);
1283 pthread_mutex_unlock(&data
->cond_mutex
);
1287 * This thread manage the consumer error sent back to the session daemon.
1289 static void *thread_manage_consumer(void *data
)
1291 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1292 uint32_t revents
, nb_fd
;
1293 enum lttcomm_return_code code
;
1294 struct lttng_poll_event events
;
1295 struct consumer_data
*consumer_data
= data
;
1296 struct consumer_socket
*cmd_socket_wrapper
= NULL
;
1298 DBG("[thread] Manage consumer started");
1300 rcu_register_thread();
1301 rcu_thread_online();
1303 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1305 health_code_update();
1308 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1309 * metadata_sock. Nothing more will be added to this poll set.
1311 ret
= sessiond_set_thread_pollset(&events
, 3);
1317 * The error socket here is already in a listening state which was done
1318 * just before spawning this thread to avoid a race between the consumer
1319 * daemon exec trying to connect and the listen() call.
1321 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1326 health_code_update();
1328 /* Infinite blocking call, waiting for transmission */
1330 health_poll_entry();
1332 if (testpoint(sessiond_thread_manage_consumer
)) {
1336 ret
= lttng_poll_wait(&events
, -1);
1340 * Restart interrupted system call.
1342 if (errno
== EINTR
) {
1350 for (i
= 0; i
< nb_fd
; i
++) {
1351 /* Fetch once the poll data */
1352 revents
= LTTNG_POLL_GETEV(&events
, i
);
1353 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1355 health_code_update();
1358 /* No activity for this FD (poll implementation). */
1362 /* Thread quit pipe has been closed. Killing thread. */
1363 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1369 /* Event on the registration socket */
1370 if (pollfd
== consumer_data
->err_sock
) {
1371 if (revents
& LPOLLIN
) {
1373 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1374 ERR("consumer err socket poll error");
1377 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1383 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1389 * Set the CLOEXEC flag. Return code is useless because either way, the
1392 (void) utils_set_fd_cloexec(sock
);
1394 health_code_update();
1396 DBG2("Receiving code from consumer err_sock");
1398 /* Getting status code from kconsumerd */
1399 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1400 sizeof(enum lttcomm_return_code
));
1405 health_code_update();
1406 if (code
!= LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1407 ERR("consumer error when waiting for SOCK_READY : %s",
1408 lttcomm_get_readable_code(-code
));
1412 /* Connect both command and metadata sockets. */
1413 consumer_data
->cmd_sock
=
1414 lttcomm_connect_unix_sock(
1415 consumer_data
->cmd_unix_sock_path
);
1416 consumer_data
->metadata_fd
=
1417 lttcomm_connect_unix_sock(
1418 consumer_data
->cmd_unix_sock_path
);
1419 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
1420 PERROR("consumer connect cmd socket");
1421 /* On error, signal condition and quit. */
1422 signal_consumer_condition(consumer_data
, -1);
1426 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1428 /* Create metadata socket lock. */
1429 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1430 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1431 PERROR("zmalloc pthread mutex");
1434 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1436 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1437 DBG("Consumer metadata socket ready (fd: %d)",
1438 consumer_data
->metadata_fd
);
1441 * Remove the consumerd error sock since we've established a connection.
1443 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1448 /* Add new accepted error socket. */
1449 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1454 /* Add metadata socket that is successfully connected. */
1455 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1456 LPOLLIN
| LPOLLRDHUP
);
1461 health_code_update();
1464 * Transfer the write-end of the channel monitoring and rotate pipe
1465 * to the consumer by issuing a SET_CHANNEL_MONITOR_PIPE and
1466 * SET_CHANNEL_ROTATE_PIPE commands.
1468 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
1469 if (!cmd_socket_wrapper
) {
1473 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
1474 consumer_data
->channel_monitor_pipe
);
1479 ret
= consumer_send_channel_rotate_pipe(cmd_socket_wrapper
,
1480 consumer_data
->channel_rotate_pipe
);
1485 /* Discard the socket wrapper as it is no longer needed. */
1486 consumer_destroy_socket(cmd_socket_wrapper
);
1487 cmd_socket_wrapper
= NULL
;
1489 /* The thread is completely initialized, signal that it is ready. */
1490 signal_consumer_condition(consumer_data
, 1);
1492 /* Infinite blocking call, waiting for transmission */
1495 health_code_update();
1497 /* Exit the thread because the thread quit pipe has been triggered. */
1499 /* Not a health error. */
1504 health_poll_entry();
1505 ret
= lttng_poll_wait(&events
, -1);
1509 * Restart interrupted system call.
1511 if (errno
== EINTR
) {
1519 for (i
= 0; i
< nb_fd
; i
++) {
1520 /* Fetch once the poll data */
1521 revents
= LTTNG_POLL_GETEV(&events
, i
);
1522 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1524 health_code_update();
1527 /* No activity for this FD (poll implementation). */
1532 * Thread quit pipe has been triggered, flag that we should stop
1533 * but continue the current loop to handle potential data from
1536 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1538 if (pollfd
== sock
) {
1539 /* Event on the consumerd socket */
1540 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1541 && !(revents
& LPOLLIN
)) {
1542 ERR("consumer err socket second poll error");
1545 health_code_update();
1546 /* Wait for any kconsumerd error */
1547 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1548 sizeof(enum lttcomm_return_code
));
1550 ERR("consumer closed the command socket");
1554 ERR("consumer return code : %s",
1555 lttcomm_get_readable_code(-code
));
1558 } else if (pollfd
== consumer_data
->metadata_fd
) {
1559 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1560 && !(revents
& LPOLLIN
)) {
1561 ERR("consumer err metadata socket second poll error");
1564 /* UST metadata requests */
1565 ret
= ust_consumer_metadata_request(
1566 &consumer_data
->metadata_sock
);
1568 ERR("Handling metadata request");
1572 /* No need for an else branch all FDs are tested prior. */
1574 health_code_update();
1580 * We lock here because we are about to close the sockets and some other
1581 * thread might be using them so get exclusive access which will abort all
1582 * other consumer command by other threads.
1584 pthread_mutex_lock(&consumer_data
->lock
);
1586 /* Immediately set the consumerd state to stopped */
1587 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1588 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1589 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1590 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1591 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1593 /* Code flow error... */
1597 if (consumer_data
->err_sock
>= 0) {
1598 ret
= close(consumer_data
->err_sock
);
1602 consumer_data
->err_sock
= -1;
1604 if (consumer_data
->cmd_sock
>= 0) {
1605 ret
= close(consumer_data
->cmd_sock
);
1609 consumer_data
->cmd_sock
= -1;
1611 if (consumer_data
->metadata_sock
.fd_ptr
&&
1612 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1613 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1625 unlink(consumer_data
->err_unix_sock_path
);
1626 unlink(consumer_data
->cmd_unix_sock_path
);
1627 pthread_mutex_unlock(&consumer_data
->lock
);
1629 /* Cleanup metadata socket mutex. */
1630 if (consumer_data
->metadata_sock
.lock
) {
1631 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1632 free(consumer_data
->metadata_sock
.lock
);
1634 lttng_poll_clean(&events
);
1636 if (cmd_socket_wrapper
) {
1637 consumer_destroy_socket(cmd_socket_wrapper
);
1642 ERR("Health error occurred in %s", __func__
);
1644 health_unregister(health_sessiond
);
1645 DBG("consumer thread cleanup completed");
1647 rcu_thread_offline();
1648 rcu_unregister_thread();
1654 * This thread manage application communication.
1656 static void *thread_manage_apps(void *data
)
1658 int i
, ret
, pollfd
, err
= -1;
1660 uint32_t revents
, nb_fd
;
1661 struct lttng_poll_event events
;
1663 DBG("[thread] Manage application started");
1665 rcu_register_thread();
1666 rcu_thread_online();
1668 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1670 if (testpoint(sessiond_thread_manage_apps
)) {
1671 goto error_testpoint
;
1674 health_code_update();
1676 ret
= sessiond_set_thread_pollset(&events
, 2);
1678 goto error_poll_create
;
1681 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1686 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1690 health_code_update();
1693 DBG("Apps thread polling");
1695 /* Inifinite blocking call, waiting for transmission */
1697 health_poll_entry();
1698 ret
= lttng_poll_wait(&events
, -1);
1699 DBG("Apps thread return from poll on %d fds",
1700 LTTNG_POLL_GETNB(&events
));
1704 * Restart interrupted system call.
1706 if (errno
== EINTR
) {
1714 for (i
= 0; i
< nb_fd
; i
++) {
1715 /* Fetch once the poll data */
1716 revents
= LTTNG_POLL_GETEV(&events
, i
);
1717 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1719 health_code_update();
1722 /* No activity for this FD (poll implementation). */
1726 /* Thread quit pipe has been closed. Killing thread. */
1727 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1733 /* Inspect the apps cmd pipe */
1734 if (pollfd
== apps_cmd_pipe
[0]) {
1735 if (revents
& LPOLLIN
) {
1739 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1740 if (size_ret
< sizeof(sock
)) {
1741 PERROR("read apps cmd pipe");
1745 health_code_update();
1748 * Since this is a command socket (write then read),
1749 * we only monitor the error events of the socket.
1751 ret
= lttng_poll_add(&events
, sock
,
1752 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1757 DBG("Apps with sock %d added to poll set", sock
);
1758 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1759 ERR("Apps command pipe error");
1762 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1767 * At this point, we know that a registered application made
1768 * the event at poll_wait.
1770 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1771 /* Removing from the poll set */
1772 ret
= lttng_poll_del(&events
, pollfd
);
1777 /* Socket closed on remote end. */
1778 ust_app_unregister(pollfd
);
1780 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1785 health_code_update();
1791 lttng_poll_clean(&events
);
1794 utils_close_pipe(apps_cmd_pipe
);
1795 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1798 * We don't clean the UST app hash table here since already registered
1799 * applications can still be controlled so let them be until the session
1800 * daemon dies or the applications stop.
1805 ERR("Health error occurred in %s", __func__
);
1807 health_unregister(health_sessiond
);
1808 DBG("Application communication apps thread cleanup complete");
1809 rcu_thread_offline();
1810 rcu_unregister_thread();
1815 * Send a socket to a thread This is called from the dispatch UST registration
1816 * thread once all sockets are set for the application.
1818 * The sock value can be invalid, we don't really care, the thread will handle
1819 * it and make the necessary cleanup if so.
1821 * On success, return 0 else a negative value being the errno message of the
1824 static int send_socket_to_thread(int fd
, int sock
)
1829 * It's possible that the FD is set as invalid with -1 concurrently just
1830 * before calling this function being a shutdown state of the thread.
1837 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1838 if (ret
< sizeof(sock
)) {
1839 PERROR("write apps pipe %d", fd
);
1846 /* All good. Don't send back the write positive ret value. */
1853 * Sanitize the wait queue of the dispatch registration thread meaning removing
1854 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1855 * notify socket is never received.
1857 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1859 int ret
, nb_fd
= 0, i
;
1860 unsigned int fd_added
= 0;
1861 struct lttng_poll_event events
;
1862 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1866 lttng_poll_init(&events
);
1868 /* Just skip everything for an empty queue. */
1869 if (!wait_queue
->count
) {
1873 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1878 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1879 &wait_queue
->head
, head
) {
1880 assert(wait_node
->app
);
1881 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1882 LPOLLHUP
| LPOLLERR
);
1895 * Poll but don't block so we can quickly identify the faulty events and
1896 * clean them afterwards from the wait queue.
1898 ret
= lttng_poll_wait(&events
, 0);
1904 for (i
= 0; i
< nb_fd
; i
++) {
1905 /* Get faulty FD. */
1906 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1907 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1910 /* No activity for this FD (poll implementation). */
1914 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1915 &wait_queue
->head
, head
) {
1916 if (pollfd
== wait_node
->app
->sock
&&
1917 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1918 cds_list_del(&wait_node
->head
);
1919 wait_queue
->count
--;
1920 ust_app_destroy(wait_node
->app
);
1923 * Silence warning of use-after-free in
1924 * cds_list_for_each_entry_safe which uses
1925 * __typeof__(*wait_node).
1930 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1937 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1941 lttng_poll_clean(&events
);
1945 lttng_poll_clean(&events
);
1947 ERR("Unable to sanitize wait queue");
1952 * Dispatch request from the registration threads to the application
1953 * communication thread.
1955 static void *thread_dispatch_ust_registration(void *data
)
1958 struct cds_wfcq_node
*node
;
1959 struct ust_command
*ust_cmd
= NULL
;
1960 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1961 struct ust_reg_wait_queue wait_queue
= {
1965 rcu_register_thread();
1967 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1969 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1970 goto error_testpoint
;
1973 health_code_update();
1975 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1977 DBG("[thread] Dispatch UST command started");
1980 health_code_update();
1982 /* Atomically prepare the queue futex */
1983 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1985 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1990 struct ust_app
*app
= NULL
;
1994 * Make sure we don't have node(s) that have hung up before receiving
1995 * the notify socket. This is to clean the list in order to avoid
1996 * memory leaks from notify socket that are never seen.
1998 sanitize_wait_queue(&wait_queue
);
2000 health_code_update();
2001 /* Dequeue command for registration */
2002 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2004 DBG("Woken up but nothing in the UST command queue");
2005 /* Continue thread execution */
2009 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2011 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
2012 " gid:%d sock:%d name:%s (version %d.%d)",
2013 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2014 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2015 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2016 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2018 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
2019 wait_node
= zmalloc(sizeof(*wait_node
));
2021 PERROR("zmalloc wait_node dispatch");
2022 ret
= close(ust_cmd
->sock
);
2024 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
2026 lttng_fd_put(LTTNG_FD_APPS
, 1);
2030 CDS_INIT_LIST_HEAD(&wait_node
->head
);
2032 /* Create application object if socket is CMD. */
2033 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
2035 if (!wait_node
->app
) {
2036 ret
= close(ust_cmd
->sock
);
2038 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
2040 lttng_fd_put(LTTNG_FD_APPS
, 1);
2046 * Add application to the wait queue so we can set the notify
2047 * socket before putting this object in the global ht.
2049 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
2054 * We have to continue here since we don't have the notify
2055 * socket and the application MUST be added to the hash table
2056 * only at that moment.
2061 * Look for the application in the local wait queue and set the
2062 * notify socket if found.
2064 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2065 &wait_queue
.head
, head
) {
2066 health_code_update();
2067 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
2068 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
2069 cds_list_del(&wait_node
->head
);
2071 app
= wait_node
->app
;
2073 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
2079 * With no application at this stage the received socket is
2080 * basically useless so close it before we free the cmd data
2081 * structure for good.
2084 ret
= close(ust_cmd
->sock
);
2086 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
2088 lttng_fd_put(LTTNG_FD_APPS
, 1);
2095 * @session_lock_list
2097 * Lock the global session list so from the register up to the
2098 * registration done message, no thread can see the application
2099 * and change its state.
2101 session_lock_list();
2105 * Add application to the global hash table. This needs to be
2106 * done before the update to the UST registry can locate the
2111 /* Set app version. This call will print an error if needed. */
2112 (void) ust_app_version(app
);
2114 /* Send notify socket through the notify pipe. */
2115 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2119 session_unlock_list();
2121 * No notify thread, stop the UST tracing. However, this is
2122 * not an internal error of the this thread thus setting
2123 * the health error code to a normal exit.
2130 * Update newly registered application with the tracing
2131 * registry info already enabled information.
2133 update_ust_app(app
->sock
);
2136 * Don't care about return value. Let the manage apps threads
2137 * handle app unregistration upon socket close.
2139 (void) ust_app_register_done(app
);
2142 * Even if the application socket has been closed, send the app
2143 * to the thread and unregistration will take place at that
2146 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2149 session_unlock_list();
2151 * No apps. thread, stop the UST tracing. However, this is
2152 * not an internal error of the this thread thus setting
2153 * the health error code to a normal exit.
2160 session_unlock_list();
2162 } while (node
!= NULL
);
2164 health_poll_entry();
2165 /* Futex wait on queue. Blocking call on futex() */
2166 futex_nto1_wait(&ust_cmd_queue
.futex
);
2169 /* Normal exit, no error */
2173 /* Clean up wait queue. */
2174 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2175 &wait_queue
.head
, head
) {
2176 cds_list_del(&wait_node
->head
);
2181 /* Empty command queue. */
2183 /* Dequeue command for registration */
2184 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2188 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2189 ret
= close(ust_cmd
->sock
);
2191 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2193 lttng_fd_put(LTTNG_FD_APPS
, 1);
2198 DBG("Dispatch thread dying");
2201 ERR("Health error occurred in %s", __func__
);
2203 health_unregister(health_sessiond
);
2204 rcu_unregister_thread();
2209 * This thread manage application registration.
2211 static void *thread_registration_apps(void *data
)
2213 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2214 uint32_t revents
, nb_fd
;
2215 struct lttng_poll_event events
;
2217 * Get allocated in this thread, enqueued to a global queue, dequeued and
2218 * freed in the manage apps thread.
2220 struct ust_command
*ust_cmd
= NULL
;
2222 DBG("[thread] Manage application registration started");
2224 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2226 if (testpoint(sessiond_thread_registration_apps
)) {
2227 goto error_testpoint
;
2230 ret
= lttcomm_listen_unix_sock(apps_sock
);
2236 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2237 * more will be added to this poll set.
2239 ret
= sessiond_set_thread_pollset(&events
, 2);
2241 goto error_create_poll
;
2244 /* Add the application registration socket */
2245 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2247 goto error_poll_add
;
2250 /* Notify all applications to register */
2251 ret
= notify_ust_apps(1);
2253 ERR("Failed to notify applications or create the wait shared memory.\n"
2254 "Execution continues but there might be problem for already\n"
2255 "running applications that wishes to register.");
2259 DBG("Accepting application registration");
2261 /* Inifinite blocking call, waiting for transmission */
2263 health_poll_entry();
2264 ret
= lttng_poll_wait(&events
, -1);
2268 * Restart interrupted system call.
2270 if (errno
== EINTR
) {
2278 for (i
= 0; i
< nb_fd
; i
++) {
2279 health_code_update();
2281 /* Fetch once the poll data */
2282 revents
= LTTNG_POLL_GETEV(&events
, i
);
2283 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2286 /* No activity for this FD (poll implementation). */
2290 /* Thread quit pipe has been closed. Killing thread. */
2291 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2297 /* Event on the registration socket */
2298 if (pollfd
== apps_sock
) {
2299 if (revents
& LPOLLIN
) {
2300 sock
= lttcomm_accept_unix_sock(apps_sock
);
2306 * Set socket timeout for both receiving and ending.
2307 * app_socket_timeout is in seconds, whereas
2308 * lttcomm_setsockopt_rcv_timeout and
2309 * lttcomm_setsockopt_snd_timeout expect msec as
2312 if (app_socket_timeout
>= 0) {
2313 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2314 app_socket_timeout
* 1000);
2315 (void) lttcomm_setsockopt_snd_timeout(sock
,
2316 app_socket_timeout
* 1000);
2320 * Set the CLOEXEC flag. Return code is useless because
2321 * either way, the show must go on.
2323 (void) utils_set_fd_cloexec(sock
);
2325 /* Create UST registration command for enqueuing */
2326 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2327 if (ust_cmd
== NULL
) {
2328 PERROR("ust command zmalloc");
2337 * Using message-based transmissions to ensure we don't
2338 * have to deal with partially received messages.
2340 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2342 ERR("Exhausted file descriptors allowed for applications.");
2352 health_code_update();
2353 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2356 /* Close socket of the application. */
2361 lttng_fd_put(LTTNG_FD_APPS
, 1);
2365 health_code_update();
2367 ust_cmd
->sock
= sock
;
2370 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2371 " gid:%d sock:%d name:%s (version %d.%d)",
2372 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2373 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2374 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2375 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2378 * Lock free enqueue the registration request. The red pill
2379 * has been taken! This apps will be part of the *system*.
2381 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2384 * Wake the registration queue futex. Implicit memory
2385 * barrier with the exchange in cds_wfcq_enqueue.
2387 futex_nto1_wake(&ust_cmd_queue
.futex
);
2388 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2389 ERR("Register apps socket poll error");
2392 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2401 /* Notify that the registration thread is gone */
2404 if (apps_sock
>= 0) {
2405 ret
= close(apps_sock
);
2415 lttng_fd_put(LTTNG_FD_APPS
, 1);
2417 unlink(apps_unix_sock_path
);
2420 lttng_poll_clean(&events
);
2424 DBG("UST Registration thread cleanup complete");
2427 ERR("Health error occurred in %s", __func__
);
2429 health_unregister(health_sessiond
);
2435 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2436 * exec or it will fails.
2438 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2441 struct timespec timeout
;
2444 * Make sure we set the readiness flag to 0 because we are NOT ready.
2445 * This access to consumer_thread_is_ready does not need to be
2446 * protected by consumer_data.cond_mutex (yet) since the consumer
2447 * management thread has not been started at this point.
2449 consumer_data
->consumer_thread_is_ready
= 0;
2451 /* Setup pthread condition */
2452 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2455 PERROR("pthread_condattr_init consumer data");
2460 * Set the monotonic clock in order to make sure we DO NOT jump in time
2461 * between the clock_gettime() call and the timedwait call. See bug #324
2462 * for a more details and how we noticed it.
2464 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2467 PERROR("pthread_condattr_setclock consumer data");
2471 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2474 PERROR("pthread_cond_init consumer data");
2478 ret
= pthread_create(&consumer_data
->thread
, default_pthread_attr(),
2479 thread_manage_consumer
, consumer_data
);
2482 PERROR("pthread_create consumer");
2487 /* We are about to wait on a pthread condition */
2488 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2490 /* Get time for sem_timedwait absolute timeout */
2491 clock_ret
= lttng_clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2493 * Set the timeout for the condition timed wait even if the clock gettime
2494 * call fails since we might loop on that call and we want to avoid to
2495 * increment the timeout too many times.
2497 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2500 * The following loop COULD be skipped in some conditions so this is why we
2501 * set ret to 0 in order to make sure at least one round of the loop is
2507 * Loop until the condition is reached or when a timeout is reached. Note
2508 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2509 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2510 * possible. This loop does not take any chances and works with both of
2513 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2514 if (clock_ret
< 0) {
2515 PERROR("clock_gettime spawn consumer");
2516 /* Infinite wait for the consumerd thread to be ready */
2517 ret
= pthread_cond_wait(&consumer_data
->cond
,
2518 &consumer_data
->cond_mutex
);
2520 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2521 &consumer_data
->cond_mutex
, &timeout
);
2525 /* Release the pthread condition */
2526 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2530 if (ret
== ETIMEDOUT
) {
2534 * Call has timed out so we kill the kconsumerd_thread and return
2537 ERR("Condition timed out. The consumer thread was never ready."
2539 pth_ret
= pthread_cancel(consumer_data
->thread
);
2541 PERROR("pthread_cancel consumer thread");
2544 PERROR("pthread_cond_wait failed consumer thread");
2546 /* Caller is expecting a negative value on failure. */
2551 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2552 if (consumer_data
->pid
== 0) {
2553 ERR("Consumerd did not start");
2554 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2557 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2566 * Join consumer thread
2568 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2572 /* Consumer pid must be a real one. */
2573 if (consumer_data
->pid
> 0) {
2575 ret
= kill(consumer_data
->pid
, SIGTERM
);
2577 PERROR("Error killing consumer daemon");
2580 return pthread_join(consumer_data
->thread
, &status
);
2587 * Fork and exec a consumer daemon (consumerd).
2589 * Return pid if successful else -1.
2591 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2595 const char *consumer_to_use
;
2596 const char *verbosity
;
2599 DBG("Spawning consumerd");
2606 if (opt_verbose_consumer
) {
2607 verbosity
= "--verbose";
2608 } else if (lttng_opt_quiet
) {
2609 verbosity
= "--quiet";
2614 switch (consumer_data
->type
) {
2615 case LTTNG_CONSUMER_KERNEL
:
2617 * Find out which consumerd to execute. We will first try the
2618 * 64-bit path, then the sessiond's installation directory, and
2619 * fallback on the 32-bit one,
2621 DBG3("Looking for a kernel consumer at these locations:");
2622 DBG3(" 1) %s", consumerd64_bin
);
2623 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2624 DBG3(" 3) %s", consumerd32_bin
);
2625 if (stat(consumerd64_bin
, &st
) == 0) {
2626 DBG3("Found location #1");
2627 consumer_to_use
= consumerd64_bin
;
2628 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2629 DBG3("Found location #2");
2630 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2631 } else if (stat(consumerd32_bin
, &st
) == 0) {
2632 DBG3("Found location #3");
2633 consumer_to_use
= consumerd32_bin
;
2635 DBG("Could not find any valid consumerd executable");
2639 DBG("Using kernel consumer at: %s", consumer_to_use
);
2640 (void) execl(consumer_to_use
,
2641 "lttng-consumerd", verbosity
, "-k",
2642 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2643 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2644 "--group", tracing_group_name
,
2647 case LTTNG_CONSUMER64_UST
:
2649 char *tmpnew
= NULL
;
2651 if (consumerd64_libdir
[0] != '\0') {
2655 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2659 tmplen
= strlen("LD_LIBRARY_PATH=")
2660 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2661 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2666 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2667 strcat(tmpnew
, consumerd64_libdir
);
2668 if (tmp
[0] != '\0') {
2669 strcat(tmpnew
, ":");
2670 strcat(tmpnew
, tmp
);
2672 ret
= putenv(tmpnew
);
2679 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2680 (void) execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2681 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2682 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2683 "--group", tracing_group_name
,
2685 if (consumerd64_libdir
[0] != '\0') {
2690 case LTTNG_CONSUMER32_UST
:
2692 char *tmpnew
= NULL
;
2694 if (consumerd32_libdir
[0] != '\0') {
2698 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2702 tmplen
= strlen("LD_LIBRARY_PATH=")
2703 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2704 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2709 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2710 strcat(tmpnew
, consumerd32_libdir
);
2711 if (tmp
[0] != '\0') {
2712 strcat(tmpnew
, ":");
2713 strcat(tmpnew
, tmp
);
2715 ret
= putenv(tmpnew
);
2722 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2723 (void) execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2724 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2725 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2726 "--group", tracing_group_name
,
2728 if (consumerd32_libdir
[0] != '\0') {
2734 PERROR("unknown consumer type");
2738 PERROR("Consumer execl()");
2740 /* Reaching this point, we got a failure on our execl(). */
2742 } else if (pid
> 0) {
2745 PERROR("start consumer fork");
2753 * Spawn the consumerd daemon and session daemon thread.
2755 static int start_consumerd(struct consumer_data
*consumer_data
)
2760 * Set the listen() state on the socket since there is a possible race
2761 * between the exec() of the consumer daemon and this call if place in the
2762 * consumer thread. See bug #366 for more details.
2764 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2769 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2770 if (consumer_data
->pid
!= 0) {
2771 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2775 ret
= spawn_consumerd(consumer_data
);
2777 ERR("Spawning consumerd failed");
2778 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2782 /* Setting up the consumer_data pid */
2783 consumer_data
->pid
= ret
;
2784 DBG2("Consumer pid %d", consumer_data
->pid
);
2785 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2787 DBG2("Spawning consumer control thread");
2788 ret
= spawn_consumer_thread(consumer_data
);
2790 ERR("Fatal error spawning consumer control thread");
2798 /* Cleanup already created sockets on error. */
2799 if (consumer_data
->err_sock
>= 0) {
2802 err
= close(consumer_data
->err_sock
);
2804 PERROR("close consumer data error socket");
2811 * Setup necessary data for kernel tracer action.
2813 static int init_kernel_tracer(void)
2817 /* Modprobe lttng kernel modules */
2818 ret
= modprobe_lttng_control();
2823 /* Open debugfs lttng */
2824 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2825 if (kernel_tracer_fd
< 0) {
2826 DBG("Failed to open %s", module_proc_lttng
);
2830 /* Validate kernel version */
2831 ret
= kernel_validate_version(kernel_tracer_fd
);
2836 ret
= modprobe_lttng_data();
2841 ret
= kernel_supports_ring_buffer_snapshot_sample_positions(
2848 WARN("Kernel tracer does not support buffer monitoring. "
2849 "The monitoring timer of channels in the kernel domain "
2850 "will be set to 0 (disabled).");
2853 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2857 modprobe_remove_lttng_control();
2858 ret
= close(kernel_tracer_fd
);
2862 kernel_tracer_fd
= -1;
2863 return LTTNG_ERR_KERN_VERSION
;
2866 ret
= close(kernel_tracer_fd
);
2872 modprobe_remove_lttng_control();
2875 WARN("No kernel tracer available");
2876 kernel_tracer_fd
= -1;
2878 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2880 return LTTNG_ERR_KERN_NA
;
2886 * Copy consumer output from the tracing session to the domain session. The
2887 * function also applies the right modification on a per domain basis for the
2888 * trace files destination directory.
2890 * Should *NOT* be called with RCU read-side lock held.
2892 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2895 const char *dir_name
;
2896 struct consumer_output
*consumer
;
2899 assert(session
->consumer
);
2902 case LTTNG_DOMAIN_KERNEL
:
2903 DBG3("Copying tracing session consumer output in kernel session");
2905 * XXX: We should audit the session creation and what this function
2906 * does "extra" in order to avoid a destroy since this function is used
2907 * in the domain session creation (kernel and ust) only. Same for UST
2910 if (session
->kernel_session
->consumer
) {
2911 consumer_output_put(session
->kernel_session
->consumer
);
2913 session
->kernel_session
->consumer
=
2914 consumer_copy_output(session
->consumer
);
2915 /* Ease our life a bit for the next part */
2916 consumer
= session
->kernel_session
->consumer
;
2917 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2919 case LTTNG_DOMAIN_JUL
:
2920 case LTTNG_DOMAIN_LOG4J
:
2921 case LTTNG_DOMAIN_PYTHON
:
2922 case LTTNG_DOMAIN_UST
:
2923 DBG3("Copying tracing session consumer output in UST session");
2924 if (session
->ust_session
->consumer
) {
2925 consumer_output_put(session
->ust_session
->consumer
);
2927 session
->ust_session
->consumer
=
2928 consumer_copy_output(session
->consumer
);
2929 /* Ease our life a bit for the next part */
2930 consumer
= session
->ust_session
->consumer
;
2931 dir_name
= DEFAULT_UST_TRACE_DIR
;
2934 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2938 /* Append correct directory to subdir */
2939 strncat(consumer
->subdir
, dir_name
,
2940 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2941 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2950 * Create an UST session and add it to the session ust list.
2952 * Should *NOT* be called with RCU read-side lock held.
2954 static int create_ust_session(struct ltt_session
*session
,
2955 struct lttng_domain
*domain
)
2958 struct ltt_ust_session
*lus
= NULL
;
2962 assert(session
->consumer
);
2964 switch (domain
->type
) {
2965 case LTTNG_DOMAIN_JUL
:
2966 case LTTNG_DOMAIN_LOG4J
:
2967 case LTTNG_DOMAIN_PYTHON
:
2968 case LTTNG_DOMAIN_UST
:
2971 ERR("Unknown UST domain on create session %d", domain
->type
);
2972 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2976 DBG("Creating UST session");
2978 lus
= trace_ust_create_session(session
->id
);
2980 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2984 lus
->uid
= session
->uid
;
2985 lus
->gid
= session
->gid
;
2986 lus
->output_traces
= session
->output_traces
;
2987 lus
->snapshot_mode
= session
->snapshot_mode
;
2988 lus
->live_timer_interval
= session
->live_timer
;
2989 session
->ust_session
= lus
;
2990 if (session
->shm_path
[0]) {
2991 strncpy(lus
->root_shm_path
, session
->shm_path
,
2992 sizeof(lus
->root_shm_path
));
2993 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2994 strncpy(lus
->shm_path
, session
->shm_path
,
2995 sizeof(lus
->shm_path
));
2996 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2997 strncat(lus
->shm_path
, "/ust",
2998 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
3000 /* Copy session output to the newly created UST session */
3001 ret
= copy_session_consumer(domain
->type
, session
);
3002 if (ret
!= LTTNG_OK
) {
3010 session
->ust_session
= NULL
;
3015 * Create a kernel tracer session then create the default channel.
3017 static int create_kernel_session(struct ltt_session
*session
)
3021 DBG("Creating kernel session");
3023 ret
= kernel_create_session(session
, kernel_tracer_fd
);
3025 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3029 /* Code flow safety */
3030 assert(session
->kernel_session
);
3032 /* Copy session output to the newly created Kernel session */
3033 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
3034 if (ret
!= LTTNG_OK
) {
3038 /* Create directory(ies) on local filesystem. */
3039 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
3040 strlen(session
->kernel_session
->consumer
->dst
.session_root_path
) > 0) {
3041 ret
= run_as_mkdir_recursive(
3042 session
->kernel_session
->consumer
->dst
.session_root_path
,
3043 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
3045 if (errno
!= EEXIST
) {
3046 ERR("Trace directory creation error");
3052 session
->kernel_session
->uid
= session
->uid
;
3053 session
->kernel_session
->gid
= session
->gid
;
3054 session
->kernel_session
->output_traces
= session
->output_traces
;
3055 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
3060 trace_kernel_destroy_session(session
->kernel_session
);
3061 session
->kernel_session
= NULL
;
3066 * Count number of session permitted by uid/gid.
3068 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
3071 struct ltt_session
*session
;
3073 DBG("Counting number of available session for UID %d GID %d",
3075 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
3077 * Only list the sessions the user can control.
3079 if (!session_access_ok(session
, uid
, gid
)) {
3088 * Process the command requested by the lttng client within the command
3089 * context structure. This function make sure that the return structure (llm)
3090 * is set and ready for transmission before returning.
3092 * Return any error encountered or 0 for success.
3094 * "sock" is only used for special-case var. len data.
3096 * Should *NOT* be called with RCU read-side lock held.
3098 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
3102 int need_tracing_session
= 1;
3105 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
3107 assert(!rcu_read_ongoing());
3111 switch (cmd_ctx
->lsm
->cmd_type
) {
3112 case LTTNG_CREATE_SESSION
:
3113 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3114 case LTTNG_CREATE_SESSION_LIVE
:
3115 case LTTNG_DESTROY_SESSION
:
3116 case LTTNG_LIST_SESSIONS
:
3117 case LTTNG_LIST_DOMAINS
:
3118 case LTTNG_START_TRACE
:
3119 case LTTNG_STOP_TRACE
:
3120 case LTTNG_DATA_PENDING
:
3121 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3122 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3123 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3124 case LTTNG_SNAPSHOT_RECORD
:
3125 case LTTNG_SAVE_SESSION
:
3126 case LTTNG_SET_SESSION_SHM_PATH
:
3127 case LTTNG_REGENERATE_METADATA
:
3128 case LTTNG_REGENERATE_STATEDUMP
:
3129 case LTTNG_REGISTER_TRIGGER
:
3130 case LTTNG_UNREGISTER_TRIGGER
:
3131 case LTTNG_ROTATE_SESSION
:
3132 case LTTNG_ROTATE_PENDING
:
3133 case LTTNG_ROTATE_SETUP
:
3140 if (opt_no_kernel
&& need_domain
3141 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3143 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3145 ret
= LTTNG_ERR_KERN_NA
;
3150 /* Deny register consumer if we already have a spawned consumer. */
3151 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3152 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3153 if (kconsumer_data
.pid
> 0) {
3154 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3155 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3158 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3162 * Check for command that don't needs to allocate a returned payload. We do
3163 * this here so we don't have to make the call for no payload at each
3166 switch(cmd_ctx
->lsm
->cmd_type
) {
3167 case LTTNG_LIST_SESSIONS
:
3168 case LTTNG_LIST_TRACEPOINTS
:
3169 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3170 case LTTNG_LIST_DOMAINS
:
3171 case LTTNG_LIST_CHANNELS
:
3172 case LTTNG_LIST_EVENTS
:
3173 case LTTNG_LIST_SYSCALLS
:
3174 case LTTNG_LIST_TRACKER_PIDS
:
3175 case LTTNG_DATA_PENDING
:
3176 case LTTNG_ROTATE_SESSION
:
3177 case LTTNG_ROTATE_PENDING
:
3180 /* Setup lttng message with no payload */
3181 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3183 /* This label does not try to unlock the session */
3184 goto init_setup_error
;
3188 /* Commands that DO NOT need a session. */
3189 switch (cmd_ctx
->lsm
->cmd_type
) {
3190 case LTTNG_CREATE_SESSION
:
3191 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3192 case LTTNG_CREATE_SESSION_LIVE
:
3193 case LTTNG_LIST_SESSIONS
:
3194 case LTTNG_LIST_TRACEPOINTS
:
3195 case LTTNG_LIST_SYSCALLS
:
3196 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3197 case LTTNG_SAVE_SESSION
:
3198 case LTTNG_REGISTER_TRIGGER
:
3199 case LTTNG_UNREGISTER_TRIGGER
:
3200 need_tracing_session
= 0;
3203 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3205 * We keep the session list lock across _all_ commands
3206 * for now, because the per-session lock does not
3207 * handle teardown properly.
3209 session_lock_list();
3210 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3211 if (cmd_ctx
->session
== NULL
) {
3212 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3215 /* Acquire lock for the session */
3216 session_lock(cmd_ctx
->session
);
3222 * Commands that need a valid session but should NOT create one if none
3223 * exists. Instead of creating one and destroying it when the command is
3224 * handled, process that right before so we save some round trip in useless
3227 switch (cmd_ctx
->lsm
->cmd_type
) {
3228 case LTTNG_DISABLE_CHANNEL
:
3229 case LTTNG_DISABLE_EVENT
:
3230 switch (cmd_ctx
->lsm
->domain
.type
) {
3231 case LTTNG_DOMAIN_KERNEL
:
3232 if (!cmd_ctx
->session
->kernel_session
) {
3233 ret
= LTTNG_ERR_NO_CHANNEL
;
3237 case LTTNG_DOMAIN_JUL
:
3238 case LTTNG_DOMAIN_LOG4J
:
3239 case LTTNG_DOMAIN_PYTHON
:
3240 case LTTNG_DOMAIN_UST
:
3241 if (!cmd_ctx
->session
->ust_session
) {
3242 ret
= LTTNG_ERR_NO_CHANNEL
;
3247 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3259 * Check domain type for specific "pre-action".
3261 switch (cmd_ctx
->lsm
->domain
.type
) {
3262 case LTTNG_DOMAIN_KERNEL
:
3264 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3268 /* Kernel tracer check */
3269 if (kernel_tracer_fd
== -1) {
3270 /* Basically, load kernel tracer modules */
3271 ret
= init_kernel_tracer();
3277 /* Consumer is in an ERROR state. Report back to client */
3278 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3279 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3283 /* Need a session for kernel command */
3284 if (need_tracing_session
) {
3285 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3286 ret
= create_kernel_session(cmd_ctx
->session
);
3288 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3293 /* Start the kernel consumer daemon */
3294 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3295 if (kconsumer_data
.pid
== 0 &&
3296 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3297 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3298 ret
= start_consumerd(&kconsumer_data
);
3300 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3303 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3305 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3309 * The consumer was just spawned so we need to add the socket to
3310 * the consumer output of the session if exist.
3312 ret
= consumer_create_socket(&kconsumer_data
,
3313 cmd_ctx
->session
->kernel_session
->consumer
);
3320 case LTTNG_DOMAIN_JUL
:
3321 case LTTNG_DOMAIN_LOG4J
:
3322 case LTTNG_DOMAIN_PYTHON
:
3323 case LTTNG_DOMAIN_UST
:
3325 if (!ust_app_supported()) {
3326 ret
= LTTNG_ERR_NO_UST
;
3329 /* Consumer is in an ERROR state. Report back to client */
3330 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3331 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3335 if (need_tracing_session
) {
3336 /* Create UST session if none exist. */
3337 if (cmd_ctx
->session
->ust_session
== NULL
) {
3338 ret
= create_ust_session(cmd_ctx
->session
,
3339 &cmd_ctx
->lsm
->domain
);
3340 if (ret
!= LTTNG_OK
) {
3345 /* Start the UST consumer daemons */
3347 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3348 if (consumerd64_bin
[0] != '\0' &&
3349 ustconsumer64_data
.pid
== 0 &&
3350 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3351 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3352 ret
= start_consumerd(&ustconsumer64_data
);
3354 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3355 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3359 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3360 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3362 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3366 * Setup socket for consumer 64 bit. No need for atomic access
3367 * since it was set above and can ONLY be set in this thread.
3369 ret
= consumer_create_socket(&ustconsumer64_data
,
3370 cmd_ctx
->session
->ust_session
->consumer
);
3376 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3377 if (consumerd32_bin
[0] != '\0' &&
3378 ustconsumer32_data
.pid
== 0 &&
3379 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3380 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3381 ret
= start_consumerd(&ustconsumer32_data
);
3383 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3384 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3388 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3389 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3391 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3395 * Setup socket for consumer 64 bit. No need for atomic access
3396 * since it was set above and can ONLY be set in this thread.
3398 ret
= consumer_create_socket(&ustconsumer32_data
,
3399 cmd_ctx
->session
->ust_session
->consumer
);
3411 /* Validate consumer daemon state when start/stop trace command */
3412 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3413 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3414 switch (cmd_ctx
->lsm
->domain
.type
) {
3415 case LTTNG_DOMAIN_NONE
:
3417 case LTTNG_DOMAIN_JUL
:
3418 case LTTNG_DOMAIN_LOG4J
:
3419 case LTTNG_DOMAIN_PYTHON
:
3420 case LTTNG_DOMAIN_UST
:
3421 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3422 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3426 case LTTNG_DOMAIN_KERNEL
:
3427 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3428 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3433 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3439 * Check that the UID or GID match that of the tracing session.
3440 * The root user can interact with all sessions.
3442 if (need_tracing_session
) {
3443 if (!session_access_ok(cmd_ctx
->session
,
3444 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3445 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3446 ret
= LTTNG_ERR_EPERM
;
3452 * Send relayd information to consumer as soon as we have a domain and a
3455 if (cmd_ctx
->session
&& need_domain
) {
3457 * Setup relayd if not done yet. If the relayd information was already
3458 * sent to the consumer, this call will gracefully return.
3460 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3461 if (ret
!= LTTNG_OK
) {
3466 /* Process by command type */
3467 switch (cmd_ctx
->lsm
->cmd_type
) {
3468 case LTTNG_ADD_CONTEXT
:
3471 * An LTTNG_ADD_CONTEXT command might have a supplementary
3472 * payload if the context being added is an application context.
3474 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3475 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3476 char *provider_name
= NULL
, *context_name
= NULL
;
3477 size_t provider_name_len
=
3478 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3479 size_t context_name_len
=
3480 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3482 if (provider_name_len
== 0 || context_name_len
== 0) {
3484 * Application provider and context names MUST
3487 ret
= -LTTNG_ERR_INVALID
;
3491 provider_name
= zmalloc(provider_name_len
+ 1);
3492 if (!provider_name
) {
3493 ret
= -LTTNG_ERR_NOMEM
;
3496 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3499 context_name
= zmalloc(context_name_len
+ 1);
3500 if (!context_name
) {
3501 ret
= -LTTNG_ERR_NOMEM
;
3502 goto error_add_context
;
3504 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3507 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3510 goto error_add_context
;
3513 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3516 goto error_add_context
;
3521 * cmd_add_context assumes ownership of the provider and context
3524 ret
= cmd_add_context(cmd_ctx
->session
,
3525 cmd_ctx
->lsm
->domain
.type
,
3526 cmd_ctx
->lsm
->u
.context
.channel_name
,
3527 &cmd_ctx
->lsm
->u
.context
.ctx
,
3528 kernel_poll_pipe
[1]);
3530 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3531 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3533 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3534 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3540 case LTTNG_DISABLE_CHANNEL
:
3542 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3543 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3546 case LTTNG_DISABLE_EVENT
:
3550 * FIXME: handle filter; for now we just receive the filter's
3551 * bytecode along with the filter expression which are sent by
3552 * liblttng-ctl and discard them.
3554 * This fixes an issue where the client may block while sending
3555 * the filter payload and encounter an error because the session
3556 * daemon closes the socket without ever handling this data.
3558 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3559 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3562 char data
[LTTNG_FILTER_MAX_LEN
];
3564 DBG("Discarding disable event command payload of size %zu", count
);
3566 ret
= lttcomm_recv_unix_sock(sock
, data
,
3567 count
> sizeof(data
) ? sizeof(data
) : count
);
3572 count
-= (size_t) ret
;
3575 /* FIXME: passing packed structure to non-packed pointer */
3576 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3577 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3578 &cmd_ctx
->lsm
->u
.disable
.event
);
3581 case LTTNG_ENABLE_CHANNEL
:
3583 cmd_ctx
->lsm
->u
.channel
.chan
.attr
.extended
.ptr
=
3584 (struct lttng_channel_extended
*) &cmd_ctx
->lsm
->u
.channel
.extended
;
3585 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3586 &cmd_ctx
->lsm
->u
.channel
.chan
,
3587 kernel_poll_pipe
[1]);
3590 case LTTNG_TRACK_PID
:
3592 ret
= cmd_track_pid(cmd_ctx
->session
,
3593 cmd_ctx
->lsm
->domain
.type
,
3594 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3597 case LTTNG_UNTRACK_PID
:
3599 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3600 cmd_ctx
->lsm
->domain
.type
,
3601 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3604 case LTTNG_ENABLE_EVENT
:
3606 struct lttng_event_exclusion
*exclusion
= NULL
;
3607 struct lttng_filter_bytecode
*bytecode
= NULL
;
3608 char *filter_expression
= NULL
;
3610 /* Handle exclusion events and receive it from the client. */
3611 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3612 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3614 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3615 (count
* LTTNG_SYMBOL_NAME_LEN
));
3617 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3621 DBG("Receiving var len exclusion event list from client ...");
3622 exclusion
->count
= count
;
3623 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3624 count
* LTTNG_SYMBOL_NAME_LEN
);
3626 DBG("Nothing recv() from client var len data... continuing");
3629 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3634 /* Get filter expression from client. */
3635 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3636 size_t expression_len
=
3637 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3639 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3640 ret
= LTTNG_ERR_FILTER_INVAL
;
3645 filter_expression
= zmalloc(expression_len
);
3646 if (!filter_expression
) {
3648 ret
= LTTNG_ERR_FILTER_NOMEM
;
3652 /* Receive var. len. data */
3653 DBG("Receiving var len filter's expression from client ...");
3654 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3657 DBG("Nothing recv() from client car len data... continuing");
3659 free(filter_expression
);
3661 ret
= LTTNG_ERR_FILTER_INVAL
;
3666 /* Handle filter and get bytecode from client. */
3667 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3668 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3670 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3671 ret
= LTTNG_ERR_FILTER_INVAL
;
3672 free(filter_expression
);
3677 bytecode
= zmalloc(bytecode_len
);
3679 free(filter_expression
);
3681 ret
= LTTNG_ERR_FILTER_NOMEM
;
3685 /* Receive var. len. data */
3686 DBG("Receiving var len filter's bytecode from client ...");
3687 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3689 DBG("Nothing recv() from client car len data... continuing");
3691 free(filter_expression
);
3694 ret
= LTTNG_ERR_FILTER_INVAL
;
3698 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3699 free(filter_expression
);
3702 ret
= LTTNG_ERR_FILTER_INVAL
;
3707 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3708 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3709 &cmd_ctx
->lsm
->u
.enable
.event
,
3710 filter_expression
, bytecode
, exclusion
,
3711 kernel_poll_pipe
[1]);
3714 case LTTNG_LIST_TRACEPOINTS
:
3716 struct lttng_event
*events
;
3719 session_lock_list();
3720 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3721 session_unlock_list();
3722 if (nb_events
< 0) {
3723 /* Return value is a negative lttng_error_code. */
3729 * Setup lttng message with payload size set to the event list size in
3730 * bytes and then copy list into the llm payload.
3732 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3733 sizeof(struct lttng_event
) * nb_events
);
3743 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3745 struct lttng_event_field
*fields
;
3748 session_lock_list();
3749 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3751 session_unlock_list();
3752 if (nb_fields
< 0) {
3753 /* Return value is a negative lttng_error_code. */
3759 * Setup lttng message with payload size set to the event list size in
3760 * bytes and then copy list into the llm payload.
3762 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3763 sizeof(struct lttng_event_field
) * nb_fields
);
3773 case LTTNG_LIST_SYSCALLS
:
3775 struct lttng_event
*events
;
3778 nb_events
= cmd_list_syscalls(&events
);
3779 if (nb_events
< 0) {
3780 /* Return value is a negative lttng_error_code. */
3786 * Setup lttng message with payload size set to the event list size in
3787 * bytes and then copy list into the llm payload.
3789 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3790 sizeof(struct lttng_event
) * nb_events
);
3800 case LTTNG_LIST_TRACKER_PIDS
:
3802 int32_t *pids
= NULL
;
3805 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3806 cmd_ctx
->lsm
->domain
.type
, &pids
);
3808 /* Return value is a negative lttng_error_code. */
3814 * Setup lttng message with payload size set to the event list size in
3815 * bytes and then copy list into the llm payload.
3817 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3818 sizeof(int32_t) * nr_pids
);
3828 case LTTNG_SET_CONSUMER_URI
:
3831 struct lttng_uri
*uris
;
3833 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3834 len
= nb_uri
* sizeof(struct lttng_uri
);
3837 ret
= LTTNG_ERR_INVALID
;
3841 uris
= zmalloc(len
);
3843 ret
= LTTNG_ERR_FATAL
;
3847 /* Receive variable len data */
3848 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3849 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3851 DBG("No URIs received from client... continuing");
3853 ret
= LTTNG_ERR_SESSION_FAIL
;
3858 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3860 if (ret
!= LTTNG_OK
) {
3867 case LTTNG_START_TRACE
:
3869 ret
= cmd_start_trace(cmd_ctx
->session
);
3872 case LTTNG_STOP_TRACE
:
3874 ret
= cmd_stop_trace(cmd_ctx
->session
);
3877 case LTTNG_CREATE_SESSION
:
3880 struct lttng_uri
*uris
= NULL
;
3882 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3883 len
= nb_uri
* sizeof(struct lttng_uri
);
3886 uris
= zmalloc(len
);
3888 ret
= LTTNG_ERR_FATAL
;
3892 /* Receive variable len data */
3893 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3894 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3896 DBG("No URIs received from client... continuing");
3898 ret
= LTTNG_ERR_SESSION_FAIL
;
3903 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3904 DBG("Creating session with ONE network URI is a bad call");
3905 ret
= LTTNG_ERR_SESSION_FAIL
;
3911 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3912 &cmd_ctx
->creds
, 0);
3918 case LTTNG_DESTROY_SESSION
:
3920 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3922 /* Set session to NULL so we do not unlock it after free. */
3923 cmd_ctx
->session
= NULL
;
3926 case LTTNG_LIST_DOMAINS
:
3929 struct lttng_domain
*domains
= NULL
;
3931 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3933 /* Return value is a negative lttng_error_code. */
3938 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3939 nb_dom
* sizeof(struct lttng_domain
));
3949 case LTTNG_LIST_CHANNELS
:
3951 ssize_t payload_size
;
3952 struct lttng_channel
*channels
= NULL
;
3954 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3955 cmd_ctx
->session
, &channels
);
3956 if (payload_size
< 0) {
3957 /* Return value is a negative lttng_error_code. */
3958 ret
= -payload_size
;
3962 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3973 case LTTNG_LIST_EVENTS
:
3976 struct lttng_event
*events
= NULL
;
3977 struct lttcomm_event_command_header cmd_header
;
3980 memset(&cmd_header
, 0, sizeof(cmd_header
));
3981 /* Extended infos are included at the end of events */
3982 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3983 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3984 &events
, &total_size
);
3987 /* Return value is a negative lttng_error_code. */
3992 cmd_header
.nb_events
= nb_event
;
3993 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3994 &cmd_header
, sizeof(cmd_header
));
4004 case LTTNG_LIST_SESSIONS
:
4006 unsigned int nr_sessions
;
4007 void *sessions_payload
;
4010 session_lock_list();
4011 nr_sessions
= lttng_sessions_count(
4012 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
4013 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
4014 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
4015 sessions_payload
= zmalloc(payload_len
);
4017 if (!sessions_payload
) {
4018 session_unlock_list();
4023 cmd_list_lttng_sessions(sessions_payload
,
4024 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
4025 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
4026 session_unlock_list();
4028 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
4030 free(sessions_payload
);
4039 case LTTNG_REGISTER_CONSUMER
:
4041 struct consumer_data
*cdata
;
4043 switch (cmd_ctx
->lsm
->domain
.type
) {
4044 case LTTNG_DOMAIN_KERNEL
:
4045 cdata
= &kconsumer_data
;
4048 ret
= LTTNG_ERR_UND
;
4052 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
4053 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
4056 case LTTNG_DATA_PENDING
:
4059 uint8_t pending_ret_byte
;
4061 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
4066 * This function may returns 0 or 1 to indicate whether or not
4067 * there is data pending. In case of error, it should return an
4068 * LTTNG_ERR code. However, some code paths may still return
4069 * a nondescript error code, which we handle by returning an
4072 if (pending_ret
== 0 || pending_ret
== 1) {
4074 * ret will be set to LTTNG_OK at the end of
4077 } else if (pending_ret
< 0) {
4078 ret
= LTTNG_ERR_UNK
;
4085 pending_ret_byte
= (uint8_t) pending_ret
;
4087 /* 1 byte to return whether or not data is pending */
4088 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
4089 &pending_ret_byte
, 1);
4098 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
4100 struct lttcomm_lttng_output_id reply
;
4102 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
4103 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
4104 if (ret
!= LTTNG_OK
) {
4108 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
4114 /* Copy output list into message payload */
4118 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
4120 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
4121 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
4124 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4127 struct lttng_snapshot_output
*outputs
= NULL
;
4129 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4130 if (nb_output
< 0) {
4135 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4136 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4137 nb_output
* sizeof(struct lttng_snapshot_output
));
4147 case LTTNG_SNAPSHOT_RECORD
:
4149 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4150 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4151 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4154 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4157 struct lttng_uri
*uris
= NULL
;
4159 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4160 len
= nb_uri
* sizeof(struct lttng_uri
);
4163 uris
= zmalloc(len
);
4165 ret
= LTTNG_ERR_FATAL
;
4169 /* Receive variable len data */
4170 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4171 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4173 DBG("No URIs received from client... continuing");
4175 ret
= LTTNG_ERR_SESSION_FAIL
;
4180 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4181 DBG("Creating session with ONE network URI is a bad call");
4182 ret
= LTTNG_ERR_SESSION_FAIL
;
4188 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4189 nb_uri
, &cmd_ctx
->creds
);
4193 case LTTNG_CREATE_SESSION_LIVE
:
4196 struct lttng_uri
*uris
= NULL
;
4198 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4199 len
= nb_uri
* sizeof(struct lttng_uri
);
4202 uris
= zmalloc(len
);
4204 ret
= LTTNG_ERR_FATAL
;
4208 /* Receive variable len data */
4209 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4210 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4212 DBG("No URIs received from client... continuing");
4214 ret
= LTTNG_ERR_SESSION_FAIL
;
4219 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4220 DBG("Creating session with ONE network URI is a bad call");
4221 ret
= LTTNG_ERR_SESSION_FAIL
;
4227 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4228 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4232 case LTTNG_SAVE_SESSION
:
4234 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4238 case LTTNG_SET_SESSION_SHM_PATH
:
4240 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4241 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4244 case LTTNG_REGENERATE_METADATA
:
4246 ret
= cmd_regenerate_metadata(cmd_ctx
->session
);
4249 case LTTNG_REGENERATE_STATEDUMP
:
4251 ret
= cmd_regenerate_statedump(cmd_ctx
->session
);
4254 case LTTNG_REGISTER_TRIGGER
:
4256 ret
= cmd_register_trigger(cmd_ctx
, sock
,
4257 notification_thread_handle
);
4260 case LTTNG_UNREGISTER_TRIGGER
:
4262 ret
= cmd_unregister_trigger(cmd_ctx
, sock
,
4263 notification_thread_handle
);
4266 case LTTNG_ROTATE_SESSION
:
4268 struct lttng_rotate_session_return
*rotate_return
= NULL
;
4270 ret
= cmd_rotate_session(cmd_ctx
->session
, &rotate_return
);
4276 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, rotate_return
,
4277 sizeof(struct lttng_rotate_session_return
));
4278 free(rotate_return
);
4287 case LTTNG_ROTATE_PENDING
:
4289 struct lttng_rotate_pending_return
*pending_return
= NULL
;
4291 ret
= cmd_rotate_pending(cmd_ctx
->session
, &pending_return
,
4292 cmd_ctx
->lsm
->u
.rotate_pending
.rotate_id
);
4298 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pending_return
,
4299 sizeof(struct lttng_rotate_session_handle
));
4300 free(pending_return
);
4309 case LTTNG_ROTATE_SETUP
:
4311 ret
= cmd_rotate_setup(cmd_ctx
->session
,
4312 cmd_ctx
->lsm
->u
.rotate_setup
.timer_us
,
4313 cmd_ctx
->lsm
->u
.rotate_setup
.size
);
4317 ret
= LTTNG_ERR_UND
;
4322 if (cmd_ctx
->llm
== NULL
) {
4323 DBG("Missing llm structure. Allocating one.");
4324 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4328 /* Set return code */
4329 cmd_ctx
->llm
->ret_code
= ret
;
4331 if (cmd_ctx
->session
) {
4332 session_unlock(cmd_ctx
->session
);
4334 if (need_tracing_session
) {
4335 session_unlock_list();
4338 assert(!rcu_read_ongoing());
4343 * Thread managing health check socket.
4345 static void *thread_manage_health(void *data
)
4347 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4348 uint32_t revents
, nb_fd
;
4349 struct lttng_poll_event events
;
4350 struct health_comm_msg msg
;
4351 struct health_comm_reply reply
;
4353 DBG("[thread] Manage health check started");
4355 rcu_register_thread();
4357 /* We might hit an error path before this is created. */
4358 lttng_poll_init(&events
);
4360 /* Create unix socket */
4361 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4363 ERR("Unable to create health check Unix socket");
4368 /* lttng health client socket path permissions */
4369 ret
= chown(health_unix_sock_path
, 0,
4370 utils_get_group_id(tracing_group_name
));
4372 ERR("Unable to set group on %s", health_unix_sock_path
);
4377 ret
= chmod(health_unix_sock_path
,
4378 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4380 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4387 * Set the CLOEXEC flag. Return code is useless because either way, the
4390 (void) utils_set_fd_cloexec(sock
);
4392 ret
= lttcomm_listen_unix_sock(sock
);
4398 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4399 * more will be added to this poll set.
4401 ret
= sessiond_set_thread_pollset(&events
, 2);
4406 /* Add the application registration socket */
4407 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4412 sessiond_notify_ready();
4415 DBG("Health check ready");
4417 /* Inifinite blocking call, waiting for transmission */
4419 ret
= lttng_poll_wait(&events
, -1);
4422 * Restart interrupted system call.
4424 if (errno
== EINTR
) {
4432 for (i
= 0; i
< nb_fd
; i
++) {
4433 /* Fetch once the poll data */
4434 revents
= LTTNG_POLL_GETEV(&events
, i
);
4435 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4438 /* No activity for this FD (poll implementation). */
4442 /* Thread quit pipe has been closed. Killing thread. */
4443 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4449 /* Event on the registration socket */
4450 if (pollfd
== sock
) {
4451 if (revents
& LPOLLIN
) {
4453 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4454 ERR("Health socket poll error");
4457 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4463 new_sock
= lttcomm_accept_unix_sock(sock
);
4469 * Set the CLOEXEC flag. Return code is useless because either way, the
4472 (void) utils_set_fd_cloexec(new_sock
);
4474 DBG("Receiving data from client for health...");
4475 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4477 DBG("Nothing recv() from client... continuing");
4478 ret
= close(new_sock
);
4485 rcu_thread_online();
4487 memset(&reply
, 0, sizeof(reply
));
4488 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4490 * health_check_state returns 0 if health is
4493 if (!health_check_state(health_sessiond
, i
)) {
4494 reply
.ret_code
|= 1ULL << i
;
4498 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4500 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4502 ERR("Failed to send health data back to client");
4505 /* End of transmission */
4506 ret
= close(new_sock
);
4515 ERR("Health error occurred in %s", __func__
);
4517 DBG("Health check thread dying");
4518 unlink(health_unix_sock_path
);
4526 lttng_poll_clean(&events
);
4528 rcu_unregister_thread();
4533 * This thread manage all clients request using the unix client socket for
4536 static void *thread_manage_clients(void *data
)
4538 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4540 uint32_t revents
, nb_fd
;
4541 struct command_ctx
*cmd_ctx
= NULL
;
4542 struct lttng_poll_event events
;
4544 DBG("[thread] Manage client started");
4546 rcu_register_thread();
4548 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4550 health_code_update();
4552 ret
= lttcomm_listen_unix_sock(client_sock
);
4558 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4559 * more will be added to this poll set.
4561 ret
= sessiond_set_thread_pollset(&events
, 2);
4563 goto error_create_poll
;
4566 /* Add the application registration socket */
4567 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4572 sessiond_notify_ready();
4573 ret
= sem_post(&load_info
->message_thread_ready
);
4575 PERROR("sem_post message_thread_ready");
4579 /* This testpoint is after we signal readiness to the parent. */
4580 if (testpoint(sessiond_thread_manage_clients
)) {
4584 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4588 health_code_update();
4591 DBG("Accepting client command ...");
4593 /* Inifinite blocking call, waiting for transmission */
4595 health_poll_entry();
4596 ret
= lttng_poll_wait(&events
, -1);
4600 * Restart interrupted system call.
4602 if (errno
== EINTR
) {
4610 for (i
= 0; i
< nb_fd
; i
++) {
4611 /* Fetch once the poll data */
4612 revents
= LTTNG_POLL_GETEV(&events
, i
);
4613 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4615 health_code_update();
4618 /* No activity for this FD (poll implementation). */
4622 /* Thread quit pipe has been closed. Killing thread. */
4623 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4629 /* Event on the registration socket */
4630 if (pollfd
== client_sock
) {
4631 if (revents
& LPOLLIN
) {
4633 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4634 ERR("Client socket poll error");
4637 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4643 DBG("Wait for client response");
4645 health_code_update();
4647 sock
= lttcomm_accept_unix_sock(client_sock
);
4653 * Set the CLOEXEC flag. Return code is useless because either way, the
4656 (void) utils_set_fd_cloexec(sock
);
4658 /* Set socket option for credentials retrieval */
4659 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4664 /* Allocate context command to process the client request */
4665 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4666 if (cmd_ctx
== NULL
) {
4667 PERROR("zmalloc cmd_ctx");
4671 /* Allocate data buffer for reception */
4672 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4673 if (cmd_ctx
->lsm
== NULL
) {
4674 PERROR("zmalloc cmd_ctx->lsm");
4678 cmd_ctx
->llm
= NULL
;
4679 cmd_ctx
->session
= NULL
;
4681 health_code_update();
4684 * Data is received from the lttng client. The struct
4685 * lttcomm_session_msg (lsm) contains the command and data request of
4688 DBG("Receiving data from client ...");
4689 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4690 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4692 DBG("Nothing recv() from client... continuing");
4698 clean_command_ctx(&cmd_ctx
);
4702 health_code_update();
4704 // TODO: Validate cmd_ctx including sanity check for
4705 // security purpose.
4707 rcu_thread_online();
4709 * This function dispatch the work to the kernel or userspace tracer
4710 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4711 * informations for the client. The command context struct contains
4712 * everything this function may needs.
4714 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4715 rcu_thread_offline();
4723 * TODO: Inform client somehow of the fatal error. At
4724 * this point, ret < 0 means that a zmalloc failed
4725 * (ENOMEM). Error detected but still accept
4726 * command, unless a socket error has been
4729 clean_command_ctx(&cmd_ctx
);
4733 health_code_update();
4735 DBG("Sending response (size: %d, retcode: %s (%d))",
4736 cmd_ctx
->lttng_msg_size
,
4737 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4738 cmd_ctx
->llm
->ret_code
);
4739 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4741 ERR("Failed to send data back to client");
4744 /* End of transmission */
4751 clean_command_ctx(&cmd_ctx
);
4753 health_code_update();
4765 lttng_poll_clean(&events
);
4766 clean_command_ctx(&cmd_ctx
);
4770 unlink(client_unix_sock_path
);
4771 if (client_sock
>= 0) {
4772 ret
= close(client_sock
);
4780 ERR("Health error occurred in %s", __func__
);
4783 health_unregister(health_sessiond
);
4785 DBG("Client thread dying");
4787 rcu_unregister_thread();
4790 * Since we are creating the consumer threads, we own them, so we need
4791 * to join them before our thread exits.
4793 ret
= join_consumer_thread(&kconsumer_data
);
4796 PERROR("join_consumer");
4799 ret
= join_consumer_thread(&ustconsumer32_data
);
4802 PERROR("join_consumer ust32");
4805 ret
= join_consumer_thread(&ustconsumer64_data
);
4808 PERROR("join_consumer ust64");
4813 static int string_match(const char *str1
, const char *str2
)
4815 return (str1
&& str2
) && !strcmp(str1
, str2
);
4819 * Take an option from the getopt output and set it in the right variable to be
4822 * Return 0 on success else a negative value.
4824 static int set_option(int opt
, const char *arg
, const char *optname
)
4828 if (string_match(optname
, "client-sock") || opt
== 'c') {
4829 if (!arg
|| *arg
== '\0') {
4833 if (lttng_is_setuid_setgid()) {
4834 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4835 "-c, --client-sock");
4837 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4839 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4840 if (!arg
|| *arg
== '\0') {
4844 if (lttng_is_setuid_setgid()) {
4845 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4848 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4850 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4852 } else if (string_match(optname
, "background") || opt
== 'b') {
4854 } else if (string_match(optname
, "group") || opt
== 'g') {
4855 if (!arg
|| *arg
== '\0') {
4859 if (lttng_is_setuid_setgid()) {
4860 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4864 * If the override option is set, the pointer points to a
4865 * *non* const thus freeing it even though the variable type is
4868 if (tracing_group_name_override
) {
4869 free((void *) tracing_group_name
);
4871 tracing_group_name
= strdup(arg
);
4872 if (!tracing_group_name
) {
4876 tracing_group_name_override
= 1;
4878 } else if (string_match(optname
, "help") || opt
== 'h') {
4879 ret
= utils_show_help(8, "lttng-sessiond", help_msg
);
4881 ERR("Cannot show --help for `lttng-sessiond`");
4884 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4885 } else if (string_match(optname
, "version") || opt
== 'V') {
4886 fprintf(stdout
, "%s\n", VERSION
);
4888 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4890 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4891 if (!arg
|| *arg
== '\0') {
4895 if (lttng_is_setuid_setgid()) {
4896 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4897 "--kconsumerd-err-sock");
4899 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4901 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4902 if (!arg
|| *arg
== '\0') {
4906 if (lttng_is_setuid_setgid()) {
4907 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4908 "--kconsumerd-cmd-sock");
4910 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4912 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4913 if (!arg
|| *arg
== '\0') {
4917 if (lttng_is_setuid_setgid()) {
4918 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4919 "--ustconsumerd64-err-sock");
4921 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4923 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4924 if (!arg
|| *arg
== '\0') {
4928 if (lttng_is_setuid_setgid()) {
4929 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4930 "--ustconsumerd64-cmd-sock");
4932 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4934 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4935 if (!arg
|| *arg
== '\0') {
4939 if (lttng_is_setuid_setgid()) {
4940 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4941 "--ustconsumerd32-err-sock");
4943 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4945 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4946 if (!arg
|| *arg
== '\0') {
4950 if (lttng_is_setuid_setgid()) {
4951 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4952 "--ustconsumerd32-cmd-sock");
4954 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4956 } else if (string_match(optname
, "no-kernel")) {
4958 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4959 lttng_opt_quiet
= 1;
4960 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4961 /* Verbose level can increase using multiple -v */
4963 /* Value obtained from config file */
4964 lttng_opt_verbose
= config_parse_value(arg
);
4966 /* -v used on command line */
4967 lttng_opt_verbose
++;
4969 /* Clamp value to [0, 3] */
4970 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4971 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4972 } else if (string_match(optname
, "verbose-consumer")) {
4974 opt_verbose_consumer
= config_parse_value(arg
);
4976 opt_verbose_consumer
++;
4978 } else if (string_match(optname
, "consumerd32-path")) {
4979 if (!arg
|| *arg
== '\0') {
4983 if (lttng_is_setuid_setgid()) {
4984 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4985 "--consumerd32-path");
4987 if (consumerd32_bin_override
) {
4988 free((void *) consumerd32_bin
);
4990 consumerd32_bin
= strdup(arg
);
4991 if (!consumerd32_bin
) {
4995 consumerd32_bin_override
= 1;
4997 } else if (string_match(optname
, "consumerd32-libdir")) {
4998 if (!arg
|| *arg
== '\0') {
5002 if (lttng_is_setuid_setgid()) {
5003 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5004 "--consumerd32-libdir");
5006 if (consumerd32_libdir_override
) {
5007 free((void *) consumerd32_libdir
);
5009 consumerd32_libdir
= strdup(arg
);
5010 if (!consumerd32_libdir
) {
5014 consumerd32_libdir_override
= 1;
5016 } else if (string_match(optname
, "consumerd64-path")) {
5017 if (!arg
|| *arg
== '\0') {
5021 if (lttng_is_setuid_setgid()) {
5022 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5023 "--consumerd64-path");
5025 if (consumerd64_bin_override
) {
5026 free((void *) consumerd64_bin
);
5028 consumerd64_bin
= strdup(arg
);
5029 if (!consumerd64_bin
) {
5033 consumerd64_bin_override
= 1;
5035 } else if (string_match(optname
, "consumerd64-libdir")) {
5036 if (!arg
|| *arg
== '\0') {
5040 if (lttng_is_setuid_setgid()) {
5041 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5042 "--consumerd64-libdir");
5044 if (consumerd64_libdir_override
) {
5045 free((void *) consumerd64_libdir
);
5047 consumerd64_libdir
= strdup(arg
);
5048 if (!consumerd64_libdir
) {
5052 consumerd64_libdir_override
= 1;
5054 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
5055 if (!arg
|| *arg
== '\0') {
5059 if (lttng_is_setuid_setgid()) {
5060 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5064 opt_pidfile
= strdup(arg
);
5070 } else if (string_match(optname
, "agent-tcp-port")) {
5071 if (!arg
|| *arg
== '\0') {
5075 if (lttng_is_setuid_setgid()) {
5076 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5077 "--agent-tcp-port");
5082 v
= strtoul(arg
, NULL
, 0);
5083 if (errno
!= 0 || !isdigit(arg
[0])) {
5084 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
5087 if (v
== 0 || v
>= 65535) {
5088 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
5091 agent_tcp_port
= (uint32_t) v
;
5092 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
5094 } else if (string_match(optname
, "load") || opt
== 'l') {
5095 if (!arg
|| *arg
== '\0') {
5099 if (lttng_is_setuid_setgid()) {
5100 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5103 free(opt_load_session_path
);
5104 opt_load_session_path
= strdup(arg
);
5105 if (!opt_load_session_path
) {
5110 } else if (string_match(optname
, "kmod-probes")) {
5111 if (!arg
|| *arg
== '\0') {
5115 if (lttng_is_setuid_setgid()) {
5116 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5119 free(kmod_probes_list
);
5120 kmod_probes_list
= strdup(arg
);
5121 if (!kmod_probes_list
) {
5126 } else if (string_match(optname
, "extra-kmod-probes")) {
5127 if (!arg
|| *arg
== '\0') {
5131 if (lttng_is_setuid_setgid()) {
5132 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5133 "--extra-kmod-probes");
5135 free(kmod_extra_probes_list
);
5136 kmod_extra_probes_list
= strdup(arg
);
5137 if (!kmod_extra_probes_list
) {
5142 } else if (string_match(optname
, "config") || opt
== 'f') {
5143 /* This is handled in set_options() thus silent skip. */
5146 /* Unknown option or other error.
5147 * Error is printed by getopt, just return */
5152 if (ret
== -EINVAL
) {
5153 const char *opt_name
= "unknown";
5156 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
5158 if (opt
== long_options
[i
].val
) {
5159 opt_name
= long_options
[i
].name
;
5164 WARN("Invalid argument provided for option \"%s\", using default value.",
5172 * config_entry_handler_cb used to handle options read from a config file.
5173 * See config_entry_handler_cb comment in common/config/session-config.h for the
5174 * return value conventions.
5176 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
5180 if (!entry
|| !entry
->name
|| !entry
->value
) {
5185 /* Check if the option is to be ignored */
5186 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5187 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5192 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5195 /* Ignore if not fully matched. */
5196 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5201 * If the option takes no argument on the command line, we have to
5202 * check if the value is "true". We support non-zero numeric values,
5205 if (!long_options
[i
].has_arg
) {
5206 ret
= config_parse_value(entry
->value
);
5209 WARN("Invalid configuration value \"%s\" for option %s",
5210 entry
->value
, entry
->name
);
5212 /* False, skip boolean config option. */
5217 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5221 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5228 * daemon configuration loading and argument parsing
5230 static int set_options(int argc
, char **argv
)
5232 int ret
= 0, c
= 0, option_index
= 0;
5233 int orig_optopt
= optopt
, orig_optind
= optind
;
5235 const char *config_path
= NULL
;
5237 optstring
= utils_generate_optstring(long_options
,
5238 sizeof(long_options
) / sizeof(struct option
));
5244 /* Check for the --config option */
5245 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5246 &option_index
)) != -1) {
5250 } else if (c
!= 'f') {
5251 /* if not equal to --config option. */
5255 if (lttng_is_setuid_setgid()) {
5256 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5259 config_path
= utils_expand_path(optarg
);
5261 ERR("Failed to resolve path: %s", optarg
);
5266 ret
= config_get_section_entries(config_path
, config_section_name
,
5267 config_entry_handler
, NULL
);
5270 ERR("Invalid configuration option at line %i", ret
);
5276 /* Reset getopt's global state */
5277 optopt
= orig_optopt
;
5278 optind
= orig_optind
;
5282 * getopt_long() will not set option_index if it encounters a
5285 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5292 * Pass NULL as the long option name if popt left the index
5295 ret
= set_option(c
, optarg
,
5296 option_index
< 0 ? NULL
:
5297 long_options
[option_index
].name
);
5309 * Creates the two needed socket by the daemon.
5310 * apps_sock - The communication socket for all UST apps.
5311 * client_sock - The communication of the cli tool (lttng).
5313 static int init_daemon_socket(void)
5318 old_umask
= umask(0);
5320 /* Create client tool unix socket */
5321 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5322 if (client_sock
< 0) {
5323 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5328 /* Set the cloexec flag */
5329 ret
= utils_set_fd_cloexec(client_sock
);
5331 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5332 "Continuing but note that the consumer daemon will have a "
5333 "reference to this socket on exec()", client_sock
);
5336 /* File permission MUST be 660 */
5337 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5339 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5344 /* Create the application unix socket */
5345 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5346 if (apps_sock
< 0) {
5347 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5352 /* Set the cloexec flag */
5353 ret
= utils_set_fd_cloexec(apps_sock
);
5355 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5356 "Continuing but note that the consumer daemon will have a "
5357 "reference to this socket on exec()", apps_sock
);
5360 /* File permission MUST be 666 */
5361 ret
= chmod(apps_unix_sock_path
,
5362 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5364 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5369 DBG3("Session daemon client socket %d and application socket %d created",
5370 client_sock
, apps_sock
);
5378 * Check if the global socket is available, and if a daemon is answering at the
5379 * other side. If yes, error is returned.
5381 static int check_existing_daemon(void)
5383 /* Is there anybody out there ? */
5384 if (lttng_session_daemon_alive()) {
5392 * Set the tracing group gid onto the client socket.
5394 * Race window between mkdir and chown is OK because we are going from more
5395 * permissive (root.root) to less permissive (root.tracing).
5397 static int set_permissions(char *rundir
)
5402 gid
= utils_get_group_id(tracing_group_name
);
5404 /* Set lttng run dir */
5405 ret
= chown(rundir
, 0, gid
);
5407 ERR("Unable to set group on %s", rundir
);
5412 * Ensure all applications and tracing group can search the run
5413 * dir. Allow everyone to read the directory, since it does not
5414 * buy us anything to hide its content.
5416 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5418 ERR("Unable to set permissions on %s", rundir
);
5422 /* lttng client socket path */
5423 ret
= chown(client_unix_sock_path
, 0, gid
);
5425 ERR("Unable to set group on %s", client_unix_sock_path
);
5429 /* kconsumer error socket path */
5430 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5432 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5436 /* 64-bit ustconsumer error socket path */
5437 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5439 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5443 /* 32-bit ustconsumer compat32 error socket path */
5444 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5446 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5450 DBG("All permissions are set");
5456 * Create the lttng run directory needed for all global sockets and pipe.
5458 static int create_lttng_rundir(const char *rundir
)
5462 DBG3("Creating LTTng run directory: %s", rundir
);
5464 ret
= mkdir(rundir
, S_IRWXU
);
5466 if (errno
!= EEXIST
) {
5467 ERR("Unable to create %s", rundir
);
5479 * Setup sockets and directory needed by the kconsumerd communication with the
5482 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5486 char path
[PATH_MAX
];
5488 switch (consumer_data
->type
) {
5489 case LTTNG_CONSUMER_KERNEL
:
5490 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5492 case LTTNG_CONSUMER64_UST
:
5493 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5495 case LTTNG_CONSUMER32_UST
:
5496 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5499 ERR("Consumer type unknown");
5504 DBG2("Creating consumer directory: %s", path
);
5506 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5507 if (ret
< 0 && errno
!= EEXIST
) {
5509 ERR("Failed to create %s", path
);
5513 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5515 ERR("Unable to set group on %s", path
);
5521 /* Create the kconsumerd error unix socket */
5522 consumer_data
->err_sock
=
5523 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5524 if (consumer_data
->err_sock
< 0) {
5525 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5531 * Set the CLOEXEC flag. Return code is useless because either way, the
5534 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5536 PERROR("utils_set_fd_cloexec");
5537 /* continue anyway */
5540 /* File permission MUST be 660 */
5541 ret
= chmod(consumer_data
->err_unix_sock_path
,
5542 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5544 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5554 * Signal handler for the daemon
5556 * Simply stop all worker threads, leaving main() return gracefully after
5557 * joining all threads and calling cleanup().
5559 static void sighandler(int sig
)
5563 DBG("SIGINT caught");
5567 DBG("SIGTERM caught");
5571 CMM_STORE_SHARED(recv_child_signal
, 1);
5579 * Setup signal handler for :
5580 * SIGINT, SIGTERM, SIGPIPE
5582 static int set_signal_handler(void)
5585 struct sigaction sa
;
5588 if ((ret
= sigemptyset(&sigset
)) < 0) {
5589 PERROR("sigemptyset");
5593 sa
.sa_mask
= sigset
;
5596 sa
.sa_handler
= sighandler
;
5597 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5598 PERROR("sigaction");
5602 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5603 PERROR("sigaction");
5607 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5608 PERROR("sigaction");
5612 sa
.sa_handler
= SIG_IGN
;
5613 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5614 PERROR("sigaction");
5618 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5624 * Set open files limit to unlimited. This daemon can open a large number of
5625 * file descriptors in order to consume multiple kernel traces.
5627 static void set_ulimit(void)
5632 /* The kernel does not allow an infinite limit for open files */
5633 lim
.rlim_cur
= 65535;
5634 lim
.rlim_max
= 65535;
5636 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5638 PERROR("failed to set open files limit");
5643 * Write pidfile using the rundir and opt_pidfile.
5645 static int write_pidfile(void)
5648 char pidfile_path
[PATH_MAX
];
5653 if (lttng_strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
))) {
5658 /* Build pidfile path from rundir and opt_pidfile. */
5659 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5660 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5662 PERROR("snprintf pidfile path");
5668 * Create pid file in rundir.
5670 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5676 * Create lockfile using the rundir and return its fd.
5678 static int create_lockfile(void)
5681 char lockfile_path
[PATH_MAX
];
5683 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5688 ret
= utils_create_lock_file(lockfile_path
);
5694 * Write agent TCP port using the rundir.
5696 static int write_agent_port(void)
5699 char path
[PATH_MAX
];
5703 ret
= snprintf(path
, sizeof(path
), "%s/"
5704 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5706 PERROR("snprintf agent port path");
5711 * Create TCP agent port file in rundir.
5713 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5722 int main(int argc
, char **argv
)
5724 int ret
= 0, retval
= 0;
5726 const char *home_path
, *env_app_timeout
;
5727 struct lttng_pipe
*ust32_channel_monitor_pipe
= NULL
,
5728 *ust64_channel_monitor_pipe
= NULL
,
5729 *kernel_channel_monitor_pipe
= NULL
;
5730 bool notification_thread_running
= false;
5731 bool timer_thread_running
= false;
5732 struct lttng_pipe
*ust32_channel_rotate_pipe
= NULL
,
5733 *ust64_channel_rotate_pipe
= NULL
,
5734 *kernel_channel_rotate_pipe
= NULL
;
5735 struct timer_thread_parameters timer_thread_ctx
;
5737 init_kernel_workarounds();
5739 rcu_register_thread();
5741 if (set_signal_handler()) {
5743 goto exit_set_signal_handler
;
5746 if (sessiond_timer_signal_init()) {
5748 goto exit_set_signal_handler
;
5751 setup_consumerd_path();
5753 page_size
= sysconf(_SC_PAGESIZE
);
5754 if (page_size
< 0) {
5755 PERROR("sysconf _SC_PAGESIZE");
5756 page_size
= LONG_MAX
;
5757 WARN("Fallback page size to %ld", page_size
);
5761 * Parse arguments and load the daemon configuration file.
5763 * We have an exit_options exit path to free memory reserved by
5764 * set_options. This is needed because the rest of sessiond_cleanup()
5765 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5766 * depends on set_options.
5769 if (set_options(argc
, argv
)) {
5775 if (opt_daemon
|| opt_background
) {
5778 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5786 * We are in the child. Make sure all other file descriptors are
5787 * closed, in case we are called with more opened file
5788 * descriptors than the standard ones.
5790 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5795 if (run_as_create_worker(argv
[0]) < 0) {
5796 goto exit_create_run_as_worker_cleanup
;
5800 * Starting from here, we can create threads. This needs to be after
5801 * lttng_daemonize due to RCU.
5805 * Initialize the health check subsystem. This call should set the
5806 * appropriate time values.
5808 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5809 if (!health_sessiond
) {
5810 PERROR("health_app_create error");
5812 goto exit_health_sessiond_cleanup
;
5815 /* Create thread to clean up RCU hash tables */
5816 if (init_ht_cleanup_thread(&ht_cleanup_thread
)) {
5818 goto exit_ht_cleanup
;
5821 /* Create thread quit pipe */
5822 if (init_thread_quit_pipe()) {
5824 goto exit_init_data
;
5827 /* Check if daemon is UID = 0 */
5828 is_root
= !getuid();
5831 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5834 goto exit_init_data
;
5837 /* Create global run dir with root access */
5838 if (create_lttng_rundir(rundir
)) {
5840 goto exit_init_data
;
5843 if (strlen(apps_unix_sock_path
) == 0) {
5844 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5845 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5848 goto exit_init_data
;
5852 if (strlen(client_unix_sock_path
) == 0) {
5853 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5854 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5857 goto exit_init_data
;
5861 /* Set global SHM for ust */
5862 if (strlen(wait_shm_path
) == 0) {
5863 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5864 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5867 goto exit_init_data
;
5871 if (strlen(health_unix_sock_path
) == 0) {
5872 ret
= snprintf(health_unix_sock_path
,
5873 sizeof(health_unix_sock_path
),
5874 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5877 goto exit_init_data
;
5881 /* Setup kernel consumerd path */
5882 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5883 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5886 goto exit_init_data
;
5888 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5889 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5892 goto exit_init_data
;
5895 DBG2("Kernel consumer err path: %s",
5896 kconsumer_data
.err_unix_sock_path
);
5897 DBG2("Kernel consumer cmd path: %s",
5898 kconsumer_data
.cmd_unix_sock_path
);
5899 kernel_channel_monitor_pipe
= lttng_pipe_open(0);
5900 if (!kernel_channel_monitor_pipe
) {
5901 ERR("Failed to create kernel consumer channel monitor pipe");
5903 goto exit_init_data
;
5905 kconsumer_data
.channel_monitor_pipe
=
5906 lttng_pipe_release_writefd(
5907 kernel_channel_monitor_pipe
);
5908 if (kconsumer_data
.channel_monitor_pipe
< 0) {
5910 goto exit_init_data
;
5912 kernel_channel_rotate_pipe
= lttng_pipe_open(0);
5913 if (!kernel_channel_rotate_pipe
) {
5914 ERR("Failed to create kernel consumer channel rotate pipe");
5916 goto exit_init_data
;
5918 kconsumer_data
.channel_rotate_pipe
=
5919 lttng_pipe_release_writefd(
5920 kernel_channel_rotate_pipe
);
5921 if (kconsumer_data
.channel_rotate_pipe
< 0) {
5923 goto exit_init_data
;
5926 home_path
= utils_get_home_dir();
5927 if (home_path
== NULL
) {
5928 /* TODO: Add --socket PATH option */
5929 ERR("Can't get HOME directory for sockets creation.");
5931 goto exit_init_data
;
5935 * Create rundir from home path. This will create something like
5938 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5941 goto exit_init_data
;
5944 if (create_lttng_rundir(rundir
)) {
5946 goto exit_init_data
;
5949 if (strlen(apps_unix_sock_path
) == 0) {
5950 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5951 DEFAULT_HOME_APPS_UNIX_SOCK
,
5955 goto exit_init_data
;
5959 /* Set the cli tool unix socket path */
5960 if (strlen(client_unix_sock_path
) == 0) {
5961 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5962 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5966 goto exit_init_data
;
5970 /* Set global SHM for ust */
5971 if (strlen(wait_shm_path
) == 0) {
5972 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5973 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5977 goto exit_init_data
;
5981 /* Set health check Unix path */
5982 if (strlen(health_unix_sock_path
) == 0) {
5983 ret
= snprintf(health_unix_sock_path
,
5984 sizeof(health_unix_sock_path
),
5985 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5989 goto exit_init_data
;
5994 lockfile_fd
= create_lockfile();
5995 if (lockfile_fd
< 0) {
5997 goto exit_init_data
;
6000 /* Set consumer initial state */
6001 kernel_consumerd_state
= CONSUMER_STOPPED
;
6002 ust_consumerd_state
= CONSUMER_STOPPED
;
6004 DBG("Client socket path %s", client_unix_sock_path
);
6005 DBG("Application socket path %s", apps_unix_sock_path
);
6006 DBG("Application wait path %s", wait_shm_path
);
6007 DBG("LTTng run directory path: %s", rundir
);
6009 /* 32 bits consumerd path setup */
6010 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
6011 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
6013 PERROR("snprintf 32-bit consumer error socket path");
6015 goto exit_init_data
;
6017 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
6018 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
6020 PERROR("snprintf 32-bit consumer command socket path");
6022 goto exit_init_data
;
6025 DBG2("UST consumer 32 bits err path: %s",
6026 ustconsumer32_data
.err_unix_sock_path
);
6027 DBG2("UST consumer 32 bits cmd path: %s",
6028 ustconsumer32_data
.cmd_unix_sock_path
);
6029 ust32_channel_monitor_pipe
= lttng_pipe_open(0);
6030 if (!ust32_channel_monitor_pipe
) {
6031 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
6033 goto exit_init_data
;
6035 ustconsumer32_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
6036 ust32_channel_monitor_pipe
);
6037 if (ustconsumer32_data
.channel_monitor_pipe
< 0) {
6039 goto exit_init_data
;
6041 ust32_channel_rotate_pipe
= lttng_pipe_open(0);
6042 if (!ust32_channel_rotate_pipe
) {
6043 ERR("Failed to create 32-bit user space consumer channel rotate pipe");
6045 goto exit_init_data
;
6047 ustconsumer32_data
.channel_rotate_pipe
= lttng_pipe_release_writefd(
6048 ust32_channel_rotate_pipe
);
6049 if (ustconsumer32_data
.channel_rotate_pipe
< 0) {
6051 goto exit_init_data
;
6055 * Create the rotate_timer_pipe as non blocking because we have to
6056 * write in it from the sighandler of the timer thread.
6058 ret
= utils_create_pipe_cloexec_nonblock(rotate_timer_pipe
);
6060 ERR("Failed to create rotate pending pipe");
6062 goto exit_init_data
;
6065 * The write-side of the pipe is used by the timer thread to wakeup
6066 * the rotation thread when needed.
6068 timer_thread_ctx
.rotate_timer_pipe
= rotate_timer_pipe
[1];
6070 /* 64 bits consumerd path setup */
6071 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
6072 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
6074 PERROR("snprintf 64-bit consumer error socket path");
6076 goto exit_init_data
;
6078 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
6079 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
6081 PERROR("snprintf 64-bit consumer command socket path");
6083 goto exit_init_data
;
6086 DBG2("UST consumer 64 bits err path: %s",
6087 ustconsumer64_data
.err_unix_sock_path
);
6088 DBG2("UST consumer 64 bits cmd path: %s",
6089 ustconsumer64_data
.cmd_unix_sock_path
);
6090 ust64_channel_monitor_pipe
= lttng_pipe_open(0);
6091 if (!ust64_channel_monitor_pipe
) {
6092 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
6094 goto exit_init_data
;
6096 ustconsumer64_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
6097 ust64_channel_monitor_pipe
);
6098 if (ustconsumer64_data
.channel_monitor_pipe
< 0) {
6100 goto exit_init_data
;
6102 ust64_channel_rotate_pipe
= lttng_pipe_open(0);
6103 if (!ust64_channel_rotate_pipe
) {
6104 ERR("Failed to create 64-bit user space consumer channel rotate pipe");
6106 goto exit_init_data
;
6108 ustconsumer64_data
.channel_rotate_pipe
= lttng_pipe_release_writefd(
6109 ust64_channel_rotate_pipe
);
6110 if (ustconsumer64_data
.channel_rotate_pipe
< 0) {
6112 goto exit_init_data
;
6116 * See if daemon already exist.
6118 if (check_existing_daemon()) {
6119 ERR("Already running daemon.\n");
6121 * We do not goto exit because we must not cleanup()
6122 * because a daemon is already running.
6125 goto exit_init_data
;
6129 * Init UST app hash table. Alloc hash table before this point since
6130 * cleanup() can get called after that point.
6132 if (ust_app_ht_alloc()) {
6133 ERR("Failed to allocate UST app hash table");
6135 goto exit_init_data
;
6139 * Initialize agent app hash table. We allocate the hash table here
6140 * since cleanup() can get called after this point.
6142 if (agent_app_ht_alloc()) {
6143 ERR("Failed to allocate Agent app hash table");
6145 goto exit_init_data
;
6149 * These actions must be executed as root. We do that *after* setting up
6150 * the sockets path because we MUST make the check for another daemon using
6151 * those paths *before* trying to set the kernel consumer sockets and init
6155 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
6157 goto exit_init_data
;
6160 /* Setup kernel tracer */
6161 if (!opt_no_kernel
) {
6162 init_kernel_tracer();
6163 if (kernel_tracer_fd
>= 0) {
6164 ret
= syscall_init_table();
6166 ERR("Unable to populate syscall table. "
6167 "Syscall tracing won't work "
6168 "for this session daemon.");
6173 /* Set ulimit for open files */
6176 /* init lttng_fd tracking must be done after set_ulimit. */
6179 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
6181 goto exit_init_data
;
6184 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
6186 goto exit_init_data
;
6189 /* Setup the needed unix socket */
6190 if (init_daemon_socket()) {
6192 goto exit_init_data
;
6195 /* Set credentials to socket */
6196 if (is_root
&& set_permissions(rundir
)) {
6198 goto exit_init_data
;
6201 /* Get parent pid if -S, --sig-parent is specified. */
6202 if (opt_sig_parent
) {
6206 /* Setup the kernel pipe for waking up the kernel thread */
6207 if (is_root
&& !opt_no_kernel
) {
6208 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
6210 goto exit_init_data
;
6214 /* Setup the thread apps communication pipe. */
6215 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
6217 goto exit_init_data
;
6220 /* Setup the thread apps notify communication pipe. */
6221 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
6223 goto exit_init_data
;
6226 /* Initialize global buffer per UID and PID registry. */
6227 buffer_reg_init_uid_registry();
6228 buffer_reg_init_pid_registry();
6230 /* Init UST command queue. */
6231 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
6234 * Get session list pointer. This pointer MUST NOT be free'd. This list
6235 * is statically declared in session.c
6237 session_list_ptr
= session_get_list();
6241 /* Check for the application socket timeout env variable. */
6242 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
6243 if (env_app_timeout
) {
6244 app_socket_timeout
= atoi(env_app_timeout
);
6246 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
6249 ret
= write_pidfile();
6251 ERR("Error in write_pidfile");
6253 goto exit_init_data
;
6255 ret
= write_agent_port();
6257 ERR("Error in write_agent_port");
6259 goto exit_init_data
;
6262 /* Initialize communication library */
6264 /* Initialize TCP timeout values */
6265 lttcomm_inet_init();
6267 if (load_session_init_data(&load_info
) < 0) {
6269 goto exit_init_data
;
6271 load_info
->path
= opt_load_session_path
;
6273 /* Create health-check thread. */
6274 ret
= pthread_create(&health_thread
, default_pthread_attr(),
6275 thread_manage_health
, (void *) NULL
);
6278 PERROR("pthread_create health");
6283 /* notification_thread_data acquires the pipes' read side. */
6284 notification_thread_handle
= notification_thread_handle_create(
6285 ust32_channel_monitor_pipe
,
6286 ust64_channel_monitor_pipe
,
6287 kernel_channel_monitor_pipe
);
6288 if (!notification_thread_handle
) {
6290 ERR("Failed to create notification thread shared data");
6292 goto exit_notification
;
6295 /* Create notification thread. */
6296 ret
= pthread_create(¬ification_thread
, default_pthread_attr(),
6297 thread_notification
, notification_thread_handle
);
6300 PERROR("pthread_create notification");
6303 goto exit_notification
;
6305 notification_thread_running
= true;
6307 /* rotation_thread_data acquires the pipes' read side. */
6308 rotation_thread_handle
= rotation_thread_handle_create(
6309 ust32_channel_rotate_pipe
,
6310 ust64_channel_rotate_pipe
,
6311 kernel_channel_rotate_pipe
,
6312 thread_quit_pipe
[0],
6313 rotate_timer_pipe
[0]);
6314 if (!rotation_thread_handle
) {
6316 ERR("Failed to create rotation thread shared data");
6321 /* Create timer thread. */
6322 ret
= pthread_create(&timer_thread
, default_pthread_attr(),
6323 sessiond_timer_thread
, &timer_thread_ctx
);
6326 PERROR("pthread_create timer");
6329 goto exit_notification
;
6331 timer_thread_running
= true;
6333 /* Create rotation thread. */
6334 ret
= pthread_create(&rotation_thread
, default_pthread_attr(),
6335 thread_rotation
, rotation_thread_handle
);
6338 PERROR("pthread_create rotation");
6344 /* Create thread to manage the client socket */
6345 ret
= pthread_create(&client_thread
, default_pthread_attr(),
6346 thread_manage_clients
, (void *) NULL
);
6349 PERROR("pthread_create clients");
6355 /* Create thread to dispatch registration */
6356 ret
= pthread_create(&dispatch_thread
, default_pthread_attr(),
6357 thread_dispatch_ust_registration
, (void *) NULL
);
6360 PERROR("pthread_create dispatch");
6366 /* Create thread to manage application registration. */
6367 ret
= pthread_create(®_apps_thread
, default_pthread_attr(),
6368 thread_registration_apps
, (void *) NULL
);
6371 PERROR("pthread_create registration");
6377 /* Create thread to manage application socket */
6378 ret
= pthread_create(&apps_thread
, default_pthread_attr(),
6379 thread_manage_apps
, (void *) NULL
);
6382 PERROR("pthread_create apps");
6388 /* Create thread to manage application notify socket */
6389 ret
= pthread_create(&apps_notify_thread
, default_pthread_attr(),
6390 ust_thread_manage_notify
, (void *) NULL
);
6393 PERROR("pthread_create notify");
6396 goto exit_apps_notify
;
6399 /* Create agent registration thread. */
6400 ret
= pthread_create(&agent_reg_thread
, default_pthread_attr(),
6401 agent_thread_manage_registration
, (void *) NULL
);
6404 PERROR("pthread_create agent");
6407 goto exit_agent_reg
;
6410 /* Don't start this thread if kernel tracing is not requested nor root */
6411 if (is_root
&& !opt_no_kernel
) {
6412 /* Create kernel thread to manage kernel event */
6413 ret
= pthread_create(&kernel_thread
, default_pthread_attr(),
6414 thread_manage_kernel
, (void *) NULL
);
6417 PERROR("pthread_create kernel");
6424 /* Create session loading thread. */
6425 ret
= pthread_create(&load_session_thread
, default_pthread_attr(),
6426 thread_load_session
, load_info
);
6429 PERROR("pthread_create load_session_thread");
6432 goto exit_load_session
;
6436 * This is where we start awaiting program completion (e.g. through
6437 * signal that asks threads to teardown).
6440 ret
= pthread_join(load_session_thread
, &status
);
6443 PERROR("pthread_join load_session_thread");
6448 if (is_root
&& !opt_no_kernel
) {
6449 ret
= pthread_join(kernel_thread
, &status
);
6452 PERROR("pthread_join");
6458 ret
= pthread_join(agent_reg_thread
, &status
);
6461 PERROR("pthread_join agent");
6466 ret
= pthread_join(apps_notify_thread
, &status
);
6469 PERROR("pthread_join apps notify");
6474 ret
= pthread_join(apps_thread
, &status
);
6477 PERROR("pthread_join apps");
6482 ret
= pthread_join(reg_apps_thread
, &status
);
6485 PERROR("pthread_join");
6491 * Join dispatch thread after joining reg_apps_thread to ensure
6492 * we don't leak applications in the queue.
6494 ret
= pthread_join(dispatch_thread
, &status
);
6497 PERROR("pthread_join");
6502 ret
= pthread_join(client_thread
, &status
);
6505 PERROR("pthread_join");
6512 ret
= pthread_join(health_thread
, &status
);
6515 PERROR("pthread_join health thread");
6522 * Wait for all pending call_rcu work to complete before tearing
6523 * down data structures. call_rcu worker may be trying to
6524 * perform lookups in those structures.
6528 * sessiond_cleanup() is called when no other thread is running, except
6529 * the ht_cleanup thread, which is needed to destroy the hash tables.
6531 rcu_thread_online();
6535 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6536 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6537 * the queue is empty before shutting down the clean-up thread.
6542 * The teardown of the notification system is performed after the
6543 * session daemon's teardown in order to allow it to be notified
6544 * of the active session and channels at the moment of the teardown.
6546 if (notification_thread_handle
) {
6547 if (notification_thread_running
) {
6548 notification_thread_command_quit(
6549 notification_thread_handle
);
6550 ret
= pthread_join(notification_thread
, &status
);
6553 PERROR("pthread_join notification thread");
6557 notification_thread_handle_destroy(notification_thread_handle
);
6560 if (rotation_thread_handle
) {
6561 rotation_thread_handle_destroy(rotation_thread_handle
);
6562 ret
= close(rotate_timer_pipe
[0]);
6564 PERROR("Close rotate pending pipe");
6567 ret
= close(rotate_timer_pipe
[1]);
6569 PERROR("Close rotate pending pipe");
6574 ret
= pthread_join(rotation_thread
, &status
);
6577 PERROR("pthread_join rotation thread");
6581 if (timer_thread_running
) {
6582 kill(getpid(), LTTNG_SESSIOND_SIG_EXIT
);
6583 ret
= pthread_join(timer_thread
, &status
);
6586 PERROR("pthread_join timer thread");
6591 rcu_thread_offline();
6592 rcu_unregister_thread();
6594 ret
= fini_ht_cleanup_thread(&ht_cleanup_thread
);
6598 lttng_pipe_destroy(ust32_channel_monitor_pipe
);
6599 lttng_pipe_destroy(ust64_channel_monitor_pipe
);
6600 lttng_pipe_destroy(kernel_channel_monitor_pipe
);
6601 lttng_pipe_destroy(ust32_channel_rotate_pipe
);
6602 lttng_pipe_destroy(ust64_channel_rotate_pipe
);
6603 lttng_pipe_destroy(kernel_channel_rotate_pipe
);
6606 health_app_destroy(health_sessiond
);
6607 exit_health_sessiond_cleanup
:
6608 exit_create_run_as_worker_cleanup
:
6611 sessiond_cleanup_options();
6613 exit_set_signal_handler
: