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/defaults.h>
45 #include <common/kernel-consumer/kernel-consumer.h>
46 #include <common/futex.h>
47 #include <common/relayd/relayd.h>
48 #include <common/utils.h>
49 #include <common/daemonize.h>
50 #include <common/config/config.h>
52 #include "lttng-sessiond.h"
53 #include "buffer-registry.h"
60 #include "kernel-consumer.h"
64 #include "ust-consumer.h"
67 #include "health-sessiond.h"
68 #include "testpoint.h"
69 #include "ust-thread.h"
70 #include "jul-thread.h"
72 #define CONSUMERD_FILE "lttng-consumerd"
75 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
76 static int tracing_group_name_override
;
77 static char *opt_pidfile
;
78 static int opt_sig_parent
;
79 static int opt_verbose_consumer
;
80 static int opt_daemon
, opt_background
;
81 static int opt_no_kernel
;
82 static pid_t ppid
; /* Parent PID for --sig-parent option */
83 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
86 /* Set to 1 when a SIGUSR1 signal is received. */
87 static int recv_child_signal
;
90 * Consumer daemon specific control data. Every value not initialized here is
91 * set to 0 by the static definition.
93 static struct consumer_data kconsumer_data
= {
94 .type
= LTTNG_CONSUMER_KERNEL
,
95 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
96 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
99 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
100 .lock
= PTHREAD_MUTEX_INITIALIZER
,
101 .cond
= PTHREAD_COND_INITIALIZER
,
102 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
104 static struct consumer_data ustconsumer64_data
= {
105 .type
= LTTNG_CONSUMER64_UST
,
106 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
107 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
110 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 .lock
= PTHREAD_MUTEX_INITIALIZER
,
112 .cond
= PTHREAD_COND_INITIALIZER
,
113 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
115 static struct consumer_data ustconsumer32_data
= {
116 .type
= LTTNG_CONSUMER32_UST
,
117 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
118 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
121 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
122 .lock
= PTHREAD_MUTEX_INITIALIZER
,
123 .cond
= PTHREAD_COND_INITIALIZER
,
124 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
127 /* Command line options */
128 static const struct option long_options
[] = {
129 { "client-sock", 1, 0, 'c' },
130 { "apps-sock", 1, 0, 'a' },
131 { "kconsumerd-cmd-sock", 1, 0, 'C' },
132 { "kconsumerd-err-sock", 1, 0, 'E' },
133 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
134 { "ustconsumerd32-err-sock", 1, 0, 'H' },
135 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
136 { "ustconsumerd64-err-sock", 1, 0, 'F' },
137 { "consumerd32-path", 1, 0, 'u' },
138 { "consumerd32-libdir", 1, 0, 'U' },
139 { "consumerd64-path", 1, 0, 't' },
140 { "consumerd64-libdir", 1, 0, 'T' },
141 { "daemonize", 0, 0, 'd' },
142 { "background", 0, 0, 'b' },
143 { "sig-parent", 0, 0, 'S' },
144 { "help", 0, 0, 'h' },
145 { "group", 1, 0, 'g' },
146 { "version", 0, 0, 'V' },
147 { "quiet", 0, 0, 'q' },
148 { "verbose", 0, 0, 'v' },
149 { "verbose-consumer", 0, 0, 'Z' },
150 { "no-kernel", 0, 0, 'N' },
151 { "pidfile", 1, 0, 'p' },
152 { "jul-tcp-port", 1, 0, 'J' },
153 { "config", 1, 0, 'f' },
157 /* Command line options to ignore from configuration file */
158 static const char *config_ignore_options
[] = { "help", "version", "config" };
160 /* Shared between threads */
161 static int dispatch_thread_exit
;
163 /* Global application Unix socket path */
164 static char apps_unix_sock_path
[PATH_MAX
];
165 /* Global client Unix socket path */
166 static char client_unix_sock_path
[PATH_MAX
];
167 /* global wait shm path for UST */
168 static char wait_shm_path
[PATH_MAX
];
169 /* Global health check unix path */
170 static char health_unix_sock_path
[PATH_MAX
];
172 /* Sockets and FDs */
173 static int client_sock
= -1;
174 static int apps_sock
= -1;
175 int kernel_tracer_fd
= -1;
176 static int kernel_poll_pipe
[2] = { -1, -1 };
179 * Quit pipe for all threads. This permits a single cancellation point
180 * for all threads when receiving an event on the pipe.
182 static int thread_quit_pipe
[2] = { -1, -1 };
185 * This pipe is used to inform the thread managing application communication
186 * that a command is queued and ready to be processed.
188 static int apps_cmd_pipe
[2] = { -1, -1 };
190 int apps_cmd_notify_pipe
[2] = { -1, -1 };
192 /* Pthread, Mutexes and Semaphores */
193 static pthread_t apps_thread
;
194 static pthread_t apps_notify_thread
;
195 static pthread_t reg_apps_thread
;
196 static pthread_t client_thread
;
197 static pthread_t kernel_thread
;
198 static pthread_t dispatch_thread
;
199 static pthread_t health_thread
;
200 static pthread_t ht_cleanup_thread
;
201 static pthread_t jul_reg_thread
;
204 * UST registration command queue. This queue is tied with a futex and uses a N
205 * wakers / 1 waiter implemented and detailed in futex.c/.h
207 * The thread_manage_apps and thread_dispatch_ust_registration interact with
208 * this queue and the wait/wake scheme.
210 static struct ust_cmd_queue ust_cmd_queue
;
213 * Pointer initialized before thread creation.
215 * This points to the tracing session list containing the session count and a
216 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
217 * MUST NOT be taken if you call a public function in session.c.
219 * The lock is nested inside the structure: session_list_ptr->lock. Please use
220 * session_lock_list and session_unlock_list for lock acquisition.
222 static struct ltt_session_list
*session_list_ptr
;
224 int ust_consumerd64_fd
= -1;
225 int ust_consumerd32_fd
= -1;
227 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
228 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
229 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
230 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
231 static int consumerd32_bin_override
;
232 static int consumerd64_bin_override
;
233 static int consumerd32_libdir_override
;
234 static int consumerd64_libdir_override
;
236 static const char *module_proc_lttng
= "/proc/lttng";
239 * Consumer daemon state which is changed when spawning it, killing it or in
240 * case of a fatal error.
242 enum consumerd_state
{
243 CONSUMER_STARTED
= 1,
244 CONSUMER_STOPPED
= 2,
249 * This consumer daemon state is used to validate if a client command will be
250 * able to reach the consumer. If not, the client is informed. For instance,
251 * doing a "lttng start" when the consumer state is set to ERROR will return an
252 * error to the client.
254 * The following example shows a possible race condition of this scheme:
256 * consumer thread error happens
258 * client cmd checks state -> still OK
259 * consumer thread exit, sets error
260 * client cmd try to talk to consumer
263 * However, since the consumer is a different daemon, we have no way of making
264 * sure the command will reach it safely even with this state flag. This is why
265 * we consider that up to the state validation during command processing, the
266 * command is safe. After that, we can not guarantee the correctness of the
267 * client request vis-a-vis the consumer.
269 static enum consumerd_state ust_consumerd_state
;
270 static enum consumerd_state kernel_consumerd_state
;
273 * Socket timeout for receiving and sending in seconds.
275 static int app_socket_timeout
;
277 /* Set in main() with the current page size. */
280 /* Application health monitoring */
281 struct health_app
*health_sessiond
;
283 /* JUL TCP port for registration. Used by the JUL thread. */
284 unsigned int jul_tcp_port
= DEFAULT_JUL_TCP_PORT
;
286 /* Am I root or not. */
287 int is_root
; /* Set to 1 if the daemon is running as root */
289 const char * const config_section_name
= "sessiond";
292 * Whether sessiond is ready for commands/health check requests.
293 * NR_LTTNG_SESSIOND_READY must match the number of calls to
294 * lttng_sessiond_notify_ready().
296 #define NR_LTTNG_SESSIOND_READY 2
297 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
299 /* Notify parents that we are ready for cmd and health check */
301 void lttng_sessiond_notify_ready(void)
303 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
305 * Notify parent pid that we are ready to accept command
306 * for client side. This ppid is the one from the
307 * external process that spawned us.
309 if (opt_sig_parent
) {
314 * Notify the parent of the fork() process that we are
317 if (opt_daemon
|| opt_background
) {
318 kill(child_ppid
, SIGUSR1
);
324 void setup_consumerd_path(void)
326 const char *bin
, *libdir
;
329 * Allow INSTALL_BIN_PATH to be used as a target path for the
330 * native architecture size consumer if CONFIG_CONSUMER*_PATH
331 * has not been defined.
333 #if (CAA_BITS_PER_LONG == 32)
334 if (!consumerd32_bin
[0]) {
335 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
337 if (!consumerd32_libdir
[0]) {
338 consumerd32_libdir
= INSTALL_LIB_PATH
;
340 #elif (CAA_BITS_PER_LONG == 64)
341 if (!consumerd64_bin
[0]) {
342 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
344 if (!consumerd64_libdir
[0]) {
345 consumerd64_libdir
= INSTALL_LIB_PATH
;
348 #error "Unknown bitness"
352 * runtime env. var. overrides the build default.
354 bin
= getenv("LTTNG_CONSUMERD32_BIN");
356 consumerd32_bin
= bin
;
358 bin
= getenv("LTTNG_CONSUMERD64_BIN");
360 consumerd64_bin
= bin
;
362 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
364 consumerd32_libdir
= libdir
;
366 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
368 consumerd64_libdir
= libdir
;
373 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
375 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
381 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
387 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
399 * Check if the thread quit pipe was triggered.
401 * Return 1 if it was triggered else 0;
403 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
405 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
413 * Init thread quit pipe.
415 * Return -1 on error or 0 if all pipes are created.
417 static int init_thread_quit_pipe(void)
421 ret
= pipe(thread_quit_pipe
);
423 PERROR("thread quit pipe");
427 for (i
= 0; i
< 2; i
++) {
428 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
440 * Stop all threads by closing the thread quit pipe.
442 static void stop_threads(void)
446 /* Stopping all threads */
447 DBG("Terminating all threads");
448 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
450 ERR("write error on thread quit pipe");
453 /* Dispatch thread */
454 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
455 futex_nto1_wake(&ust_cmd_queue
.futex
);
459 * Close every consumer sockets.
461 static void close_consumer_sockets(void)
465 if (kconsumer_data
.err_sock
>= 0) {
466 ret
= close(kconsumer_data
.err_sock
);
468 PERROR("kernel consumer err_sock close");
471 if (ustconsumer32_data
.err_sock
>= 0) {
472 ret
= close(ustconsumer32_data
.err_sock
);
474 PERROR("UST consumerd32 err_sock close");
477 if (ustconsumer64_data
.err_sock
>= 0) {
478 ret
= close(ustconsumer64_data
.err_sock
);
480 PERROR("UST consumerd64 err_sock close");
483 if (kconsumer_data
.cmd_sock
>= 0) {
484 ret
= close(kconsumer_data
.cmd_sock
);
486 PERROR("kernel consumer cmd_sock close");
489 if (ustconsumer32_data
.cmd_sock
>= 0) {
490 ret
= close(ustconsumer32_data
.cmd_sock
);
492 PERROR("UST consumerd32 cmd_sock close");
495 if (ustconsumer64_data
.cmd_sock
>= 0) {
496 ret
= close(ustconsumer64_data
.cmd_sock
);
498 PERROR("UST consumerd64 cmd_sock close");
506 static void cleanup(void)
509 struct ltt_session
*sess
, *stmp
;
515 * Close the thread quit pipe. It has already done its job,
516 * since we are now called.
518 utils_close_pipe(thread_quit_pipe
);
521 * If opt_pidfile is undefined, the default file will be wiped when
522 * removing the rundir.
525 ret
= remove(opt_pidfile
);
527 PERROR("remove pidfile %s", opt_pidfile
);
531 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
534 snprintf(path
, PATH_MAX
,
536 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
537 DBG("Removing %s", path
);
540 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
541 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
);
542 DBG("Removing %s", path
);
546 snprintf(path
, PATH_MAX
,
547 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
549 DBG("Removing %s", path
);
552 snprintf(path
, PATH_MAX
,
553 DEFAULT_KCONSUMERD_PATH
,
555 DBG("Removing directory %s", path
);
558 /* ust consumerd 32 */
559 snprintf(path
, PATH_MAX
,
560 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
562 DBG("Removing %s", path
);
565 snprintf(path
, PATH_MAX
,
566 DEFAULT_USTCONSUMERD32_PATH
,
568 DBG("Removing directory %s", path
);
571 /* ust consumerd 64 */
572 snprintf(path
, PATH_MAX
,
573 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
575 DBG("Removing %s", path
);
578 snprintf(path
, PATH_MAX
,
579 DEFAULT_USTCONSUMERD64_PATH
,
581 DBG("Removing directory %s", path
);
585 * We do NOT rmdir rundir because there are other processes
586 * using it, for instance lttng-relayd, which can start in
587 * parallel with this teardown.
592 DBG("Cleaning up all sessions");
594 /* Destroy session list mutex */
595 if (session_list_ptr
!= NULL
) {
596 pthread_mutex_destroy(&session_list_ptr
->lock
);
598 /* Cleanup ALL session */
599 cds_list_for_each_entry_safe(sess
, stmp
,
600 &session_list_ptr
->head
, list
) {
601 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
605 DBG("Closing all UST sockets");
606 ust_app_clean_list();
607 buffer_reg_destroy_registries();
609 if (is_root
&& !opt_no_kernel
) {
610 DBG2("Closing kernel fd");
611 if (kernel_tracer_fd
>= 0) {
612 ret
= close(kernel_tracer_fd
);
617 DBG("Unloading kernel modules");
618 modprobe_remove_lttng_all();
621 close_consumer_sockets();
624 * If the override option is set, the pointer points to a *non* const thus
625 * freeing it even though the variable type is set to const.
627 if (tracing_group_name_override
) {
628 free((void *) tracing_group_name
);
630 if (consumerd32_bin_override
) {
631 free((void *) consumerd32_bin
);
633 if (consumerd64_bin_override
) {
634 free((void *) consumerd64_bin
);
636 if (consumerd32_libdir_override
) {
637 free((void *) consumerd32_libdir
);
639 if (consumerd64_libdir_override
) {
640 free((void *) consumerd64_libdir
);
648 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
649 "Matthew, BEET driven development works!%c[%dm",
650 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
655 * Send data on a unix socket using the liblttsessiondcomm API.
657 * Return lttcomm error code.
659 static int send_unix_sock(int sock
, void *buf
, size_t len
)
661 /* Check valid length */
666 return lttcomm_send_unix_sock(sock
, buf
, len
);
670 * Free memory of a command context structure.
672 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
674 DBG("Clean command context structure");
676 if ((*cmd_ctx
)->llm
) {
677 free((*cmd_ctx
)->llm
);
679 if ((*cmd_ctx
)->lsm
) {
680 free((*cmd_ctx
)->lsm
);
688 * Notify UST applications using the shm mmap futex.
690 static int notify_ust_apps(int active
)
694 DBG("Notifying applications of session daemon state: %d", active
);
696 /* See shm.c for this call implying mmap, shm and futex calls */
697 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
698 if (wait_shm_mmap
== NULL
) {
702 /* Wake waiting process */
703 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
705 /* Apps notified successfully */
713 * Setup the outgoing data buffer for the response (llm) by allocating the
714 * right amount of memory and copying the original information from the lsm
717 * Return total size of the buffer pointed by buf.
719 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
725 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
726 if (cmd_ctx
->llm
== NULL
) {
732 /* Copy common data */
733 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
734 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
736 cmd_ctx
->llm
->data_size
= size
;
737 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
746 * Update the kernel poll set of all channel fd available over all tracing
747 * session. Add the wakeup pipe at the end of the set.
749 static int update_kernel_poll(struct lttng_poll_event
*events
)
752 struct ltt_session
*session
;
753 struct ltt_kernel_channel
*channel
;
755 DBG("Updating kernel poll set");
758 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
759 session_lock(session
);
760 if (session
->kernel_session
== NULL
) {
761 session_unlock(session
);
765 cds_list_for_each_entry(channel
,
766 &session
->kernel_session
->channel_list
.head
, list
) {
767 /* Add channel fd to the kernel poll set */
768 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
770 session_unlock(session
);
773 DBG("Channel fd %d added to kernel set", channel
->fd
);
775 session_unlock(session
);
777 session_unlock_list();
782 session_unlock_list();
787 * Find the channel fd from 'fd' over all tracing session. When found, check
788 * for new channel stream and send those stream fds to the kernel consumer.
790 * Useful for CPU hotplug feature.
792 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
795 struct ltt_session
*session
;
796 struct ltt_kernel_session
*ksess
;
797 struct ltt_kernel_channel
*channel
;
799 DBG("Updating kernel streams for channel fd %d", fd
);
802 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
803 session_lock(session
);
804 if (session
->kernel_session
== NULL
) {
805 session_unlock(session
);
808 ksess
= session
->kernel_session
;
810 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
811 if (channel
->fd
== fd
) {
812 DBG("Channel found, updating kernel streams");
813 ret
= kernel_open_channel_stream(channel
);
817 /* Update the stream global counter */
818 ksess
->stream_count_global
+= ret
;
821 * Have we already sent fds to the consumer? If yes, it means
822 * that tracing is started so it is safe to send our updated
825 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
826 struct lttng_ht_iter iter
;
827 struct consumer_socket
*socket
;
830 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
831 &iter
.iter
, socket
, node
.node
) {
832 pthread_mutex_lock(socket
->lock
);
833 ret
= kernel_consumer_send_channel_stream(socket
,
835 session
->output_traces
? 1 : 0);
836 pthread_mutex_unlock(socket
->lock
);
847 session_unlock(session
);
849 session_unlock_list();
853 session_unlock(session
);
854 session_unlock_list();
859 * For each tracing session, update newly registered apps. The session list
860 * lock MUST be acquired before calling this.
862 static void update_ust_app(int app_sock
)
864 struct ltt_session
*sess
, *stmp
;
866 /* Consumer is in an ERROR state. Stop any application update. */
867 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
868 /* Stop the update process since the consumer is dead. */
872 /* For all tracing session(s) */
873 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
875 if (sess
->ust_session
) {
876 ust_app_global_update(sess
->ust_session
, app_sock
);
878 session_unlock(sess
);
883 * This thread manage event coming from the kernel.
885 * Features supported in this thread:
888 static void *thread_manage_kernel(void *data
)
890 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
891 uint32_t revents
, nb_fd
;
893 struct lttng_poll_event events
;
895 DBG("[thread] Thread manage kernel started");
897 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
900 * This first step of the while is to clean this structure which could free
901 * non NULL pointers so initialize it before the loop.
903 lttng_poll_init(&events
);
905 if (testpoint(sessiond_thread_manage_kernel
)) {
906 goto error_testpoint
;
909 health_code_update();
911 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
912 goto error_testpoint
;
916 health_code_update();
918 if (update_poll_flag
== 1) {
919 /* Clean events object. We are about to populate it again. */
920 lttng_poll_clean(&events
);
922 ret
= sessiond_set_thread_pollset(&events
, 2);
924 goto error_poll_create
;
927 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
932 /* This will add the available kernel channel if any. */
933 ret
= update_kernel_poll(&events
);
937 update_poll_flag
= 0;
940 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
942 /* Poll infinite value of time */
945 ret
= lttng_poll_wait(&events
, -1);
949 * Restart interrupted system call.
951 if (errno
== EINTR
) {
955 } else if (ret
== 0) {
956 /* Should not happen since timeout is infinite */
957 ERR("Return value of poll is 0 with an infinite timeout.\n"
958 "This should not have happened! Continuing...");
964 for (i
= 0; i
< nb_fd
; i
++) {
965 /* Fetch once the poll data */
966 revents
= LTTNG_POLL_GETEV(&events
, i
);
967 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
969 health_code_update();
971 /* Thread quit pipe has been closed. Killing thread. */
972 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
978 /* Check for data on kernel pipe */
979 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
980 (void) lttng_read(kernel_poll_pipe
[0],
983 * Ret value is useless here, if this pipe gets any actions an
984 * update is required anyway.
986 update_poll_flag
= 1;
990 * New CPU detected by the kernel. Adding kernel stream to
991 * kernel session and updating the kernel consumer
993 if (revents
& LPOLLIN
) {
994 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1000 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1001 * and unregister kernel stream at this point.
1010 lttng_poll_clean(&events
);
1013 utils_close_pipe(kernel_poll_pipe
);
1014 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1017 ERR("Health error occurred in %s", __func__
);
1018 WARN("Kernel thread died unexpectedly. "
1019 "Kernel tracing can continue but CPU hotplug is disabled.");
1021 health_unregister(health_sessiond
);
1022 DBG("Kernel thread dying");
1027 * Signal pthread condition of the consumer data that the thread.
1029 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1031 pthread_mutex_lock(&data
->cond_mutex
);
1034 * The state is set before signaling. It can be any value, it's the waiter
1035 * job to correctly interpret this condition variable associated to the
1036 * consumer pthread_cond.
1038 * A value of 0 means that the corresponding thread of the consumer data
1039 * was not started. 1 indicates that the thread has started and is ready
1040 * for action. A negative value means that there was an error during the
1043 data
->consumer_thread_is_ready
= state
;
1044 (void) pthread_cond_signal(&data
->cond
);
1046 pthread_mutex_unlock(&data
->cond_mutex
);
1050 * This thread manage the consumer error sent back to the session daemon.
1052 static void *thread_manage_consumer(void *data
)
1054 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1055 uint32_t revents
, nb_fd
;
1056 enum lttcomm_return_code code
;
1057 struct lttng_poll_event events
;
1058 struct consumer_data
*consumer_data
= data
;
1060 DBG("[thread] Manage consumer started");
1062 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1064 health_code_update();
1067 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1068 * metadata_sock. Nothing more will be added to this poll set.
1070 ret
= sessiond_set_thread_pollset(&events
, 3);
1076 * The error socket here is already in a listening state which was done
1077 * just before spawning this thread to avoid a race between the consumer
1078 * daemon exec trying to connect and the listen() call.
1080 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1085 health_code_update();
1087 /* Infinite blocking call, waiting for transmission */
1089 health_poll_entry();
1091 if (testpoint(sessiond_thread_manage_consumer
)) {
1095 ret
= lttng_poll_wait(&events
, -1);
1099 * Restart interrupted system call.
1101 if (errno
== EINTR
) {
1109 for (i
= 0; i
< nb_fd
; i
++) {
1110 /* Fetch once the poll data */
1111 revents
= LTTNG_POLL_GETEV(&events
, i
);
1112 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1114 health_code_update();
1116 /* Thread quit pipe has been closed. Killing thread. */
1117 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1123 /* Event on the registration socket */
1124 if (pollfd
== consumer_data
->err_sock
) {
1125 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1126 ERR("consumer err socket poll error");
1132 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1138 * Set the CLOEXEC flag. Return code is useless because either way, the
1141 (void) utils_set_fd_cloexec(sock
);
1143 health_code_update();
1145 DBG2("Receiving code from consumer err_sock");
1147 /* Getting status code from kconsumerd */
1148 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1149 sizeof(enum lttcomm_return_code
));
1154 health_code_update();
1155 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1156 /* Connect both socket, command and metadata. */
1157 consumer_data
->cmd_sock
=
1158 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1159 consumer_data
->metadata_fd
=
1160 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1161 if (consumer_data
->cmd_sock
< 0
1162 || consumer_data
->metadata_fd
< 0) {
1163 PERROR("consumer connect cmd socket");
1164 /* On error, signal condition and quit. */
1165 signal_consumer_condition(consumer_data
, -1);
1168 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1169 /* Create metadata socket lock. */
1170 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1171 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1172 PERROR("zmalloc pthread mutex");
1176 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1178 signal_consumer_condition(consumer_data
, 1);
1179 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1180 DBG("Consumer metadata socket ready (fd: %d)",
1181 consumer_data
->metadata_fd
);
1183 ERR("consumer error when waiting for SOCK_READY : %s",
1184 lttcomm_get_readable_code(-code
));
1188 /* Remove the consumerd error sock since we've established a connexion */
1189 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1194 /* Add new accepted error socket. */
1195 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1200 /* Add metadata socket that is successfully connected. */
1201 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1202 LPOLLIN
| LPOLLRDHUP
);
1207 health_code_update();
1209 /* Infinite blocking call, waiting for transmission */
1212 health_poll_entry();
1213 ret
= lttng_poll_wait(&events
, -1);
1217 * Restart interrupted system call.
1219 if (errno
== EINTR
) {
1227 for (i
= 0; i
< nb_fd
; i
++) {
1228 /* Fetch once the poll data */
1229 revents
= LTTNG_POLL_GETEV(&events
, i
);
1230 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1232 health_code_update();
1234 /* Thread quit pipe has been closed. Killing thread. */
1235 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1241 if (pollfd
== sock
) {
1242 /* Event on the consumerd socket */
1243 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1244 ERR("consumer err socket second poll error");
1247 health_code_update();
1248 /* Wait for any kconsumerd error */
1249 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1250 sizeof(enum lttcomm_return_code
));
1252 ERR("consumer closed the command socket");
1256 ERR("consumer return code : %s",
1257 lttcomm_get_readable_code(-code
));
1260 } else if (pollfd
== consumer_data
->metadata_fd
) {
1261 /* UST metadata requests */
1262 ret
= ust_consumer_metadata_request(
1263 &consumer_data
->metadata_sock
);
1265 ERR("Handling metadata request");
1270 ERR("Unknown pollfd");
1274 health_code_update();
1280 * We lock here because we are about to close the sockets and some other
1281 * thread might be using them so get exclusive access which will abort all
1282 * other consumer command by other threads.
1284 pthread_mutex_lock(&consumer_data
->lock
);
1286 /* Immediately set the consumerd state to stopped */
1287 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1288 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1289 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1290 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1291 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1293 /* Code flow error... */
1297 if (consumer_data
->err_sock
>= 0) {
1298 ret
= close(consumer_data
->err_sock
);
1302 consumer_data
->err_sock
= -1;
1304 if (consumer_data
->cmd_sock
>= 0) {
1305 ret
= close(consumer_data
->cmd_sock
);
1309 consumer_data
->cmd_sock
= -1;
1311 if (consumer_data
->metadata_sock
.fd_ptr
&&
1312 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1313 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1325 unlink(consumer_data
->err_unix_sock_path
);
1326 unlink(consumer_data
->cmd_unix_sock_path
);
1327 consumer_data
->pid
= 0;
1328 pthread_mutex_unlock(&consumer_data
->lock
);
1330 /* Cleanup metadata socket mutex. */
1331 if (consumer_data
->metadata_sock
.lock
) {
1332 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1333 free(consumer_data
->metadata_sock
.lock
);
1335 lttng_poll_clean(&events
);
1339 ERR("Health error occurred in %s", __func__
);
1341 health_unregister(health_sessiond
);
1342 DBG("consumer thread cleanup completed");
1348 * This thread manage application communication.
1350 static void *thread_manage_apps(void *data
)
1352 int i
, ret
, pollfd
, err
= -1;
1354 uint32_t revents
, nb_fd
;
1355 struct lttng_poll_event events
;
1357 DBG("[thread] Manage application started");
1359 rcu_register_thread();
1360 rcu_thread_online();
1362 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1364 if (testpoint(sessiond_thread_manage_apps
)) {
1365 goto error_testpoint
;
1368 health_code_update();
1370 ret
= sessiond_set_thread_pollset(&events
, 2);
1372 goto error_poll_create
;
1375 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1380 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1384 health_code_update();
1387 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1389 /* Inifinite blocking call, waiting for transmission */
1391 health_poll_entry();
1392 ret
= lttng_poll_wait(&events
, -1);
1396 * Restart interrupted system call.
1398 if (errno
== EINTR
) {
1406 for (i
= 0; i
< nb_fd
; i
++) {
1407 /* Fetch once the poll data */
1408 revents
= LTTNG_POLL_GETEV(&events
, i
);
1409 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1411 health_code_update();
1413 /* Thread quit pipe has been closed. Killing thread. */
1414 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1420 /* Inspect the apps cmd pipe */
1421 if (pollfd
== apps_cmd_pipe
[0]) {
1422 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1423 ERR("Apps command pipe error");
1425 } else if (revents
& LPOLLIN
) {
1429 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1430 if (size_ret
< sizeof(sock
)) {
1431 PERROR("read apps cmd pipe");
1435 health_code_update();
1438 * We only monitor the error events of the socket. This
1439 * thread does not handle any incoming data from UST
1442 ret
= lttng_poll_add(&events
, sock
,
1443 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1448 DBG("Apps with sock %d added to poll set", sock
);
1452 * At this point, we know that a registered application made
1453 * the event at poll_wait.
1455 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1456 /* Removing from the poll set */
1457 ret
= lttng_poll_del(&events
, pollfd
);
1462 /* Socket closed on remote end. */
1463 ust_app_unregister(pollfd
);
1467 health_code_update();
1473 lttng_poll_clean(&events
);
1476 utils_close_pipe(apps_cmd_pipe
);
1477 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1480 * We don't clean the UST app hash table here since already registered
1481 * applications can still be controlled so let them be until the session
1482 * daemon dies or the applications stop.
1487 ERR("Health error occurred in %s", __func__
);
1489 health_unregister(health_sessiond
);
1490 DBG("Application communication apps thread cleanup complete");
1491 rcu_thread_offline();
1492 rcu_unregister_thread();
1497 * Send a socket to a thread This is called from the dispatch UST registration
1498 * thread once all sockets are set for the application.
1500 * The sock value can be invalid, we don't really care, the thread will handle
1501 * it and make the necessary cleanup if so.
1503 * On success, return 0 else a negative value being the errno message of the
1506 static int send_socket_to_thread(int fd
, int sock
)
1511 * It's possible that the FD is set as invalid with -1 concurrently just
1512 * before calling this function being a shutdown state of the thread.
1519 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1520 if (ret
< sizeof(sock
)) {
1521 PERROR("write apps pipe %d", fd
);
1528 /* All good. Don't send back the write positive ret value. */
1535 * Sanitize the wait queue of the dispatch registration thread meaning removing
1536 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1537 * notify socket is never received.
1539 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1541 int ret
, nb_fd
= 0, i
;
1542 unsigned int fd_added
= 0;
1543 struct lttng_poll_event events
;
1544 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1548 lttng_poll_init(&events
);
1550 /* Just skip everything for an empty queue. */
1551 if (!wait_queue
->count
) {
1555 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1560 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1561 &wait_queue
->head
, head
) {
1562 assert(wait_node
->app
);
1563 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1564 LPOLLHUP
| LPOLLERR
);
1577 * Poll but don't block so we can quickly identify the faulty events and
1578 * clean them afterwards from the wait queue.
1580 ret
= lttng_poll_wait(&events
, 0);
1586 for (i
= 0; i
< nb_fd
; i
++) {
1587 /* Get faulty FD. */
1588 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1589 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1591 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1592 &wait_queue
->head
, head
) {
1593 if (pollfd
== wait_node
->app
->sock
&&
1594 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1595 cds_list_del(&wait_node
->head
);
1596 wait_queue
->count
--;
1597 ust_app_destroy(wait_node
->app
);
1605 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1609 lttng_poll_clean(&events
);
1613 lttng_poll_clean(&events
);
1615 ERR("Unable to sanitize wait queue");
1620 * Dispatch request from the registration threads to the application
1621 * communication thread.
1623 static void *thread_dispatch_ust_registration(void *data
)
1626 struct cds_wfq_node
*node
;
1627 struct ust_command
*ust_cmd
= NULL
;
1628 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1629 struct ust_reg_wait_queue wait_queue
= {
1633 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1635 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1636 goto error_testpoint
;
1639 health_code_update();
1641 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1643 DBG("[thread] Dispatch UST command started");
1645 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1646 health_code_update();
1648 /* Atomically prepare the queue futex */
1649 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1652 struct ust_app
*app
= NULL
;
1656 * Make sure we don't have node(s) that have hung up before receiving
1657 * the notify socket. This is to clean the list in order to avoid
1658 * memory leaks from notify socket that are never seen.
1660 sanitize_wait_queue(&wait_queue
);
1662 health_code_update();
1663 /* Dequeue command for registration */
1664 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1666 DBG("Woken up but nothing in the UST command queue");
1667 /* Continue thread execution */
1671 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1673 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1674 " gid:%d sock:%d name:%s (version %d.%d)",
1675 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1676 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1677 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1678 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1680 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1681 wait_node
= zmalloc(sizeof(*wait_node
));
1683 PERROR("zmalloc wait_node dispatch");
1684 ret
= close(ust_cmd
->sock
);
1686 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1688 lttng_fd_put(LTTNG_FD_APPS
, 1);
1692 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1694 /* Create application object if socket is CMD. */
1695 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1697 if (!wait_node
->app
) {
1698 ret
= close(ust_cmd
->sock
);
1700 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1702 lttng_fd_put(LTTNG_FD_APPS
, 1);
1708 * Add application to the wait queue so we can set the notify
1709 * socket before putting this object in the global ht.
1711 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1716 * We have to continue here since we don't have the notify
1717 * socket and the application MUST be added to the hash table
1718 * only at that moment.
1723 * Look for the application in the local wait queue and set the
1724 * notify socket if found.
1726 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1727 &wait_queue
.head
, head
) {
1728 health_code_update();
1729 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1730 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1731 cds_list_del(&wait_node
->head
);
1733 app
= wait_node
->app
;
1735 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1741 * With no application at this stage the received socket is
1742 * basically useless so close it before we free the cmd data
1743 * structure for good.
1746 ret
= close(ust_cmd
->sock
);
1748 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1750 lttng_fd_put(LTTNG_FD_APPS
, 1);
1757 * @session_lock_list
1759 * Lock the global session list so from the register up to the
1760 * registration done message, no thread can see the application
1761 * and change its state.
1763 session_lock_list();
1767 * Add application to the global hash table. This needs to be
1768 * done before the update to the UST registry can locate the
1773 /* Set app version. This call will print an error if needed. */
1774 (void) ust_app_version(app
);
1776 /* Send notify socket through the notify pipe. */
1777 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1781 session_unlock_list();
1783 * No notify thread, stop the UST tracing. However, this is
1784 * not an internal error of the this thread thus setting
1785 * the health error code to a normal exit.
1792 * Update newly registered application with the tracing
1793 * registry info already enabled information.
1795 update_ust_app(app
->sock
);
1798 * Don't care about return value. Let the manage apps threads
1799 * handle app unregistration upon socket close.
1801 (void) ust_app_register_done(app
->sock
);
1804 * Even if the application socket has been closed, send the app
1805 * to the thread and unregistration will take place at that
1808 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1811 session_unlock_list();
1813 * No apps. thread, stop the UST tracing. However, this is
1814 * not an internal error of the this thread thus setting
1815 * the health error code to a normal exit.
1822 session_unlock_list();
1824 } while (node
!= NULL
);
1826 health_poll_entry();
1827 /* Futex wait on queue. Blocking call on futex() */
1828 futex_nto1_wait(&ust_cmd_queue
.futex
);
1831 /* Normal exit, no error */
1835 /* Clean up wait queue. */
1836 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1837 &wait_queue
.head
, head
) {
1838 cds_list_del(&wait_node
->head
);
1844 DBG("Dispatch thread dying");
1847 ERR("Health error occurred in %s", __func__
);
1849 health_unregister(health_sessiond
);
1854 * This thread manage application registration.
1856 static void *thread_registration_apps(void *data
)
1858 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1859 uint32_t revents
, nb_fd
;
1860 struct lttng_poll_event events
;
1862 * Get allocated in this thread, enqueued to a global queue, dequeued and
1863 * freed in the manage apps thread.
1865 struct ust_command
*ust_cmd
= NULL
;
1867 DBG("[thread] Manage application registration started");
1869 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1871 if (testpoint(sessiond_thread_registration_apps
)) {
1872 goto error_testpoint
;
1875 ret
= lttcomm_listen_unix_sock(apps_sock
);
1881 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1882 * more will be added to this poll set.
1884 ret
= sessiond_set_thread_pollset(&events
, 2);
1886 goto error_create_poll
;
1889 /* Add the application registration socket */
1890 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1892 goto error_poll_add
;
1895 /* Notify all applications to register */
1896 ret
= notify_ust_apps(1);
1898 ERR("Failed to notify applications or create the wait shared memory.\n"
1899 "Execution continues but there might be problem for already\n"
1900 "running applications that wishes to register.");
1904 DBG("Accepting application registration");
1906 /* Inifinite blocking call, waiting for transmission */
1908 health_poll_entry();
1909 ret
= lttng_poll_wait(&events
, -1);
1913 * Restart interrupted system call.
1915 if (errno
== EINTR
) {
1923 for (i
= 0; i
< nb_fd
; i
++) {
1924 health_code_update();
1926 /* Fetch once the poll data */
1927 revents
= LTTNG_POLL_GETEV(&events
, i
);
1928 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1930 /* Thread quit pipe has been closed. Killing thread. */
1931 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1937 /* Event on the registration socket */
1938 if (pollfd
== apps_sock
) {
1939 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1940 ERR("Register apps socket poll error");
1942 } else if (revents
& LPOLLIN
) {
1943 sock
= lttcomm_accept_unix_sock(apps_sock
);
1949 * Set socket timeout for both receiving and ending.
1950 * app_socket_timeout is in seconds, whereas
1951 * lttcomm_setsockopt_rcv_timeout and
1952 * lttcomm_setsockopt_snd_timeout expect msec as
1955 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1956 app_socket_timeout
* 1000);
1957 (void) lttcomm_setsockopt_snd_timeout(sock
,
1958 app_socket_timeout
* 1000);
1961 * Set the CLOEXEC flag. Return code is useless because
1962 * either way, the show must go on.
1964 (void) utils_set_fd_cloexec(sock
);
1966 /* Create UST registration command for enqueuing */
1967 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1968 if (ust_cmd
== NULL
) {
1969 PERROR("ust command zmalloc");
1974 * Using message-based transmissions to ensure we don't
1975 * have to deal with partially received messages.
1977 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1979 ERR("Exhausted file descriptors allowed for applications.");
1989 health_code_update();
1990 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1993 /* Close socket of the application. */
1998 lttng_fd_put(LTTNG_FD_APPS
, 1);
2002 health_code_update();
2004 ust_cmd
->sock
= sock
;
2007 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2008 " gid:%d sock:%d name:%s (version %d.%d)",
2009 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2010 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2011 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2012 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2015 * Lock free enqueue the registration request. The red pill
2016 * has been taken! This apps will be part of the *system*.
2018 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
2021 * Wake the registration queue futex. Implicit memory
2022 * barrier with the exchange in cds_wfq_enqueue.
2024 futex_nto1_wake(&ust_cmd_queue
.futex
);
2032 /* Notify that the registration thread is gone */
2035 if (apps_sock
>= 0) {
2036 ret
= close(apps_sock
);
2046 lttng_fd_put(LTTNG_FD_APPS
, 1);
2048 unlink(apps_unix_sock_path
);
2051 lttng_poll_clean(&events
);
2055 DBG("UST Registration thread cleanup complete");
2058 ERR("Health error occurred in %s", __func__
);
2060 health_unregister(health_sessiond
);
2066 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2067 * exec or it will fails.
2069 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2072 struct timespec timeout
;
2074 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2075 consumer_data
->consumer_thread_is_ready
= 0;
2077 /* Setup pthread condition */
2078 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2081 PERROR("pthread_condattr_init consumer data");
2086 * Set the monotonic clock in order to make sure we DO NOT jump in time
2087 * between the clock_gettime() call and the timedwait call. See bug #324
2088 * for a more details and how we noticed it.
2090 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2093 PERROR("pthread_condattr_setclock consumer data");
2097 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2100 PERROR("pthread_cond_init consumer data");
2104 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2107 PERROR("pthread_create consumer");
2112 /* We are about to wait on a pthread condition */
2113 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2115 /* Get time for sem_timedwait absolute timeout */
2116 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2118 * Set the timeout for the condition timed wait even if the clock gettime
2119 * call fails since we might loop on that call and we want to avoid to
2120 * increment the timeout too many times.
2122 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2125 * The following loop COULD be skipped in some conditions so this is why we
2126 * set ret to 0 in order to make sure at least one round of the loop is
2132 * Loop until the condition is reached or when a timeout is reached. Note
2133 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2134 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2135 * possible. This loop does not take any chances and works with both of
2138 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2139 if (clock_ret
< 0) {
2140 PERROR("clock_gettime spawn consumer");
2141 /* Infinite wait for the consumerd thread to be ready */
2142 ret
= pthread_cond_wait(&consumer_data
->cond
,
2143 &consumer_data
->cond_mutex
);
2145 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2146 &consumer_data
->cond_mutex
, &timeout
);
2150 /* Release the pthread condition */
2151 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2155 if (ret
== ETIMEDOUT
) {
2159 * Call has timed out so we kill the kconsumerd_thread and return
2162 ERR("Condition timed out. The consumer thread was never ready."
2164 pth_ret
= pthread_cancel(consumer_data
->thread
);
2166 PERROR("pthread_cancel consumer thread");
2169 PERROR("pthread_cond_wait failed consumer thread");
2171 /* Caller is expecting a negative value on failure. */
2176 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2177 if (consumer_data
->pid
== 0) {
2178 ERR("Consumerd did not start");
2179 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2182 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2191 * Join consumer thread
2193 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2197 /* Consumer pid must be a real one. */
2198 if (consumer_data
->pid
> 0) {
2200 ret
= kill(consumer_data
->pid
, SIGTERM
);
2202 ERR("Error killing consumer daemon");
2205 return pthread_join(consumer_data
->thread
, &status
);
2212 * Fork and exec a consumer daemon (consumerd).
2214 * Return pid if successful else -1.
2216 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2220 const char *consumer_to_use
;
2221 const char *verbosity
;
2224 DBG("Spawning consumerd");
2231 if (opt_verbose_consumer
) {
2232 verbosity
= "--verbose";
2234 verbosity
= "--quiet";
2236 switch (consumer_data
->type
) {
2237 case LTTNG_CONSUMER_KERNEL
:
2239 * Find out which consumerd to execute. We will first try the
2240 * 64-bit path, then the sessiond's installation directory, and
2241 * fallback on the 32-bit one,
2243 DBG3("Looking for a kernel consumer at these locations:");
2244 DBG3(" 1) %s", consumerd64_bin
);
2245 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2246 DBG3(" 3) %s", consumerd32_bin
);
2247 if (stat(consumerd64_bin
, &st
) == 0) {
2248 DBG3("Found location #1");
2249 consumer_to_use
= consumerd64_bin
;
2250 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2251 DBG3("Found location #2");
2252 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2253 } else if (stat(consumerd32_bin
, &st
) == 0) {
2254 DBG3("Found location #3");
2255 consumer_to_use
= consumerd32_bin
;
2257 DBG("Could not find any valid consumerd executable");
2261 DBG("Using kernel consumer at: %s", consumer_to_use
);
2262 ret
= execl(consumer_to_use
,
2263 "lttng-consumerd", verbosity
, "-k",
2264 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2265 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2266 "--group", tracing_group_name
,
2269 case LTTNG_CONSUMER64_UST
:
2271 char *tmpnew
= NULL
;
2273 if (consumerd64_libdir
[0] != '\0') {
2277 tmp
= getenv("LD_LIBRARY_PATH");
2281 tmplen
= strlen("LD_LIBRARY_PATH=")
2282 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2283 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2288 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2289 strcat(tmpnew
, consumerd64_libdir
);
2290 if (tmp
[0] != '\0') {
2291 strcat(tmpnew
, ":");
2292 strcat(tmpnew
, tmp
);
2294 ret
= putenv(tmpnew
);
2301 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2302 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2303 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2304 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2305 "--group", tracing_group_name
,
2307 if (consumerd64_libdir
[0] != '\0') {
2312 case LTTNG_CONSUMER32_UST
:
2314 char *tmpnew
= NULL
;
2316 if (consumerd32_libdir
[0] != '\0') {
2320 tmp
= getenv("LD_LIBRARY_PATH");
2324 tmplen
= strlen("LD_LIBRARY_PATH=")
2325 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2326 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2331 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2332 strcat(tmpnew
, consumerd32_libdir
);
2333 if (tmp
[0] != '\0') {
2334 strcat(tmpnew
, ":");
2335 strcat(tmpnew
, tmp
);
2337 ret
= putenv(tmpnew
);
2344 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2345 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2346 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2347 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2348 "--group", tracing_group_name
,
2350 if (consumerd32_libdir
[0] != '\0') {
2356 PERROR("unknown consumer type");
2360 PERROR("Consumer execl()");
2362 /* Reaching this point, we got a failure on our execl(). */
2364 } else if (pid
> 0) {
2367 PERROR("start consumer fork");
2375 * Spawn the consumerd daemon and session daemon thread.
2377 static int start_consumerd(struct consumer_data
*consumer_data
)
2382 * Set the listen() state on the socket since there is a possible race
2383 * between the exec() of the consumer daemon and this call if place in the
2384 * consumer thread. See bug #366 for more details.
2386 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2391 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2392 if (consumer_data
->pid
!= 0) {
2393 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2397 ret
= spawn_consumerd(consumer_data
);
2399 ERR("Spawning consumerd failed");
2400 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2404 /* Setting up the consumer_data pid */
2405 consumer_data
->pid
= ret
;
2406 DBG2("Consumer pid %d", consumer_data
->pid
);
2407 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2409 DBG2("Spawning consumer control thread");
2410 ret
= spawn_consumer_thread(consumer_data
);
2412 ERR("Fatal error spawning consumer control thread");
2420 /* Cleanup already created sockets on error. */
2421 if (consumer_data
->err_sock
>= 0) {
2424 err
= close(consumer_data
->err_sock
);
2426 PERROR("close consumer data error socket");
2433 * Setup necessary data for kernel tracer action.
2435 static int init_kernel_tracer(void)
2439 /* Modprobe lttng kernel modules */
2440 ret
= modprobe_lttng_control();
2445 /* Open debugfs lttng */
2446 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2447 if (kernel_tracer_fd
< 0) {
2448 DBG("Failed to open %s", module_proc_lttng
);
2453 /* Validate kernel version */
2454 ret
= kernel_validate_version(kernel_tracer_fd
);
2459 ret
= modprobe_lttng_data();
2464 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2468 modprobe_remove_lttng_control();
2469 ret
= close(kernel_tracer_fd
);
2473 kernel_tracer_fd
= -1;
2474 return LTTNG_ERR_KERN_VERSION
;
2477 ret
= close(kernel_tracer_fd
);
2483 modprobe_remove_lttng_control();
2486 WARN("No kernel tracer available");
2487 kernel_tracer_fd
= -1;
2489 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2491 return LTTNG_ERR_KERN_NA
;
2497 * Copy consumer output from the tracing session to the domain session. The
2498 * function also applies the right modification on a per domain basis for the
2499 * trace files destination directory.
2501 * Should *NOT* be called with RCU read-side lock held.
2503 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2506 const char *dir_name
;
2507 struct consumer_output
*consumer
;
2510 assert(session
->consumer
);
2513 case LTTNG_DOMAIN_KERNEL
:
2514 DBG3("Copying tracing session consumer output in kernel session");
2516 * XXX: We should audit the session creation and what this function
2517 * does "extra" in order to avoid a destroy since this function is used
2518 * in the domain session creation (kernel and ust) only. Same for UST
2521 if (session
->kernel_session
->consumer
) {
2522 consumer_destroy_output(session
->kernel_session
->consumer
);
2524 session
->kernel_session
->consumer
=
2525 consumer_copy_output(session
->consumer
);
2526 /* Ease our life a bit for the next part */
2527 consumer
= session
->kernel_session
->consumer
;
2528 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2530 case LTTNG_DOMAIN_JUL
:
2531 case LTTNG_DOMAIN_UST
:
2532 DBG3("Copying tracing session consumer output in UST session");
2533 if (session
->ust_session
->consumer
) {
2534 consumer_destroy_output(session
->ust_session
->consumer
);
2536 session
->ust_session
->consumer
=
2537 consumer_copy_output(session
->consumer
);
2538 /* Ease our life a bit for the next part */
2539 consumer
= session
->ust_session
->consumer
;
2540 dir_name
= DEFAULT_UST_TRACE_DIR
;
2543 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2547 /* Append correct directory to subdir */
2548 strncat(consumer
->subdir
, dir_name
,
2549 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2550 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2559 * Create an UST session and add it to the session ust list.
2561 * Should *NOT* be called with RCU read-side lock held.
2563 static int create_ust_session(struct ltt_session
*session
,
2564 struct lttng_domain
*domain
)
2567 struct ltt_ust_session
*lus
= NULL
;
2571 assert(session
->consumer
);
2573 switch (domain
->type
) {
2574 case LTTNG_DOMAIN_JUL
:
2575 case LTTNG_DOMAIN_UST
:
2578 ERR("Unknown UST domain on create session %d", domain
->type
);
2579 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2583 DBG("Creating UST session");
2585 lus
= trace_ust_create_session(session
->id
);
2587 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2591 lus
->uid
= session
->uid
;
2592 lus
->gid
= session
->gid
;
2593 lus
->output_traces
= session
->output_traces
;
2594 lus
->snapshot_mode
= session
->snapshot_mode
;
2595 lus
->live_timer_interval
= session
->live_timer
;
2596 session
->ust_session
= lus
;
2598 /* Copy session output to the newly created UST session */
2599 ret
= copy_session_consumer(domain
->type
, session
);
2600 if (ret
!= LTTNG_OK
) {
2608 session
->ust_session
= NULL
;
2613 * Create a kernel tracer session then create the default channel.
2615 static int create_kernel_session(struct ltt_session
*session
)
2619 DBG("Creating kernel session");
2621 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2623 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2627 /* Code flow safety */
2628 assert(session
->kernel_session
);
2630 /* Copy session output to the newly created Kernel session */
2631 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2632 if (ret
!= LTTNG_OK
) {
2636 /* Create directory(ies) on local filesystem. */
2637 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2638 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2639 ret
= run_as_mkdir_recursive(
2640 session
->kernel_session
->consumer
->dst
.trace_path
,
2641 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2643 if (ret
!= -EEXIST
) {
2644 ERR("Trace directory creation error");
2650 session
->kernel_session
->uid
= session
->uid
;
2651 session
->kernel_session
->gid
= session
->gid
;
2652 session
->kernel_session
->output_traces
= session
->output_traces
;
2653 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2658 trace_kernel_destroy_session(session
->kernel_session
);
2659 session
->kernel_session
= NULL
;
2664 * Count number of session permitted by uid/gid.
2666 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2669 struct ltt_session
*session
;
2671 DBG("Counting number of available session for UID %d GID %d",
2673 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2675 * Only list the sessions the user can control.
2677 if (!session_access_ok(session
, uid
, gid
)) {
2686 * Process the command requested by the lttng client within the command
2687 * context structure. This function make sure that the return structure (llm)
2688 * is set and ready for transmission before returning.
2690 * Return any error encountered or 0 for success.
2692 * "sock" is only used for special-case var. len data.
2694 * Should *NOT* be called with RCU read-side lock held.
2696 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2700 int need_tracing_session
= 1;
2703 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2707 switch (cmd_ctx
->lsm
->cmd_type
) {
2708 case LTTNG_CREATE_SESSION
:
2709 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2710 case LTTNG_CREATE_SESSION_LIVE
:
2711 case LTTNG_DESTROY_SESSION
:
2712 case LTTNG_LIST_SESSIONS
:
2713 case LTTNG_LIST_DOMAINS
:
2714 case LTTNG_START_TRACE
:
2715 case LTTNG_STOP_TRACE
:
2716 case LTTNG_DATA_PENDING
:
2717 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2718 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2719 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2720 case LTTNG_SNAPSHOT_RECORD
:
2727 if (opt_no_kernel
&& need_domain
2728 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2730 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2732 ret
= LTTNG_ERR_KERN_NA
;
2737 /* Deny register consumer if we already have a spawned consumer. */
2738 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2739 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2740 if (kconsumer_data
.pid
> 0) {
2741 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2742 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2745 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2749 * Check for command that don't needs to allocate a returned payload. We do
2750 * this here so we don't have to make the call for no payload at each
2753 switch(cmd_ctx
->lsm
->cmd_type
) {
2754 case LTTNG_LIST_SESSIONS
:
2755 case LTTNG_LIST_TRACEPOINTS
:
2756 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2757 case LTTNG_LIST_DOMAINS
:
2758 case LTTNG_LIST_CHANNELS
:
2759 case LTTNG_LIST_EVENTS
:
2762 /* Setup lttng message with no payload */
2763 ret
= setup_lttng_msg(cmd_ctx
, 0);
2765 /* This label does not try to unlock the session */
2766 goto init_setup_error
;
2770 /* Commands that DO NOT need a session. */
2771 switch (cmd_ctx
->lsm
->cmd_type
) {
2772 case LTTNG_CREATE_SESSION
:
2773 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2774 case LTTNG_CREATE_SESSION_LIVE
:
2775 case LTTNG_CALIBRATE
:
2776 case LTTNG_LIST_SESSIONS
:
2777 case LTTNG_LIST_TRACEPOINTS
:
2778 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2779 need_tracing_session
= 0;
2782 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2784 * We keep the session list lock across _all_ commands
2785 * for now, because the per-session lock does not
2786 * handle teardown properly.
2788 session_lock_list();
2789 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2790 if (cmd_ctx
->session
== NULL
) {
2791 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2794 /* Acquire lock for the session */
2795 session_lock(cmd_ctx
->session
);
2805 * Check domain type for specific "pre-action".
2807 switch (cmd_ctx
->lsm
->domain
.type
) {
2808 case LTTNG_DOMAIN_KERNEL
:
2810 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2814 /* Kernel tracer check */
2815 if (kernel_tracer_fd
== -1) {
2816 /* Basically, load kernel tracer modules */
2817 ret
= init_kernel_tracer();
2823 /* Consumer is in an ERROR state. Report back to client */
2824 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2825 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2829 /* Need a session for kernel command */
2830 if (need_tracing_session
) {
2831 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2832 ret
= create_kernel_session(cmd_ctx
->session
);
2834 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2839 /* Start the kernel consumer daemon */
2840 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2841 if (kconsumer_data
.pid
== 0 &&
2842 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2843 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2844 ret
= start_consumerd(&kconsumer_data
);
2846 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2849 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2851 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2855 * The consumer was just spawned so we need to add the socket to
2856 * the consumer output of the session if exist.
2858 ret
= consumer_create_socket(&kconsumer_data
,
2859 cmd_ctx
->session
->kernel_session
->consumer
);
2866 case LTTNG_DOMAIN_JUL
:
2867 case LTTNG_DOMAIN_UST
:
2869 if (!ust_app_supported()) {
2870 ret
= LTTNG_ERR_NO_UST
;
2873 /* Consumer is in an ERROR state. Report back to client */
2874 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2875 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2879 if (need_tracing_session
) {
2880 /* Create UST session if none exist. */
2881 if (cmd_ctx
->session
->ust_session
== NULL
) {
2882 ret
= create_ust_session(cmd_ctx
->session
,
2883 &cmd_ctx
->lsm
->domain
);
2884 if (ret
!= LTTNG_OK
) {
2889 /* Start the UST consumer daemons */
2891 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2892 if (consumerd64_bin
[0] != '\0' &&
2893 ustconsumer64_data
.pid
== 0 &&
2894 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2895 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2896 ret
= start_consumerd(&ustconsumer64_data
);
2898 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2899 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2903 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2904 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2906 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2910 * Setup socket for consumer 64 bit. No need for atomic access
2911 * since it was set above and can ONLY be set in this thread.
2913 ret
= consumer_create_socket(&ustconsumer64_data
,
2914 cmd_ctx
->session
->ust_session
->consumer
);
2920 if (consumerd32_bin
[0] != '\0' &&
2921 ustconsumer32_data
.pid
== 0 &&
2922 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2923 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2924 ret
= start_consumerd(&ustconsumer32_data
);
2926 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2927 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2931 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2932 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2934 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2938 * Setup socket for consumer 64 bit. No need for atomic access
2939 * since it was set above and can ONLY be set in this thread.
2941 ret
= consumer_create_socket(&ustconsumer32_data
,
2942 cmd_ctx
->session
->ust_session
->consumer
);
2954 /* Validate consumer daemon state when start/stop trace command */
2955 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2956 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2957 switch (cmd_ctx
->lsm
->domain
.type
) {
2958 case LTTNG_DOMAIN_JUL
:
2959 case LTTNG_DOMAIN_UST
:
2960 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2961 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2965 case LTTNG_DOMAIN_KERNEL
:
2966 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2967 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2975 * Check that the UID or GID match that of the tracing session.
2976 * The root user can interact with all sessions.
2978 if (need_tracing_session
) {
2979 if (!session_access_ok(cmd_ctx
->session
,
2980 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2981 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2982 ret
= LTTNG_ERR_EPERM
;
2988 * Send relayd information to consumer as soon as we have a domain and a
2991 if (cmd_ctx
->session
&& need_domain
) {
2993 * Setup relayd if not done yet. If the relayd information was already
2994 * sent to the consumer, this call will gracefully return.
2996 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2997 if (ret
!= LTTNG_OK
) {
3002 /* Process by command type */
3003 switch (cmd_ctx
->lsm
->cmd_type
) {
3004 case LTTNG_ADD_CONTEXT
:
3006 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3007 cmd_ctx
->lsm
->u
.context
.channel_name
,
3008 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3011 case LTTNG_DISABLE_CHANNEL
:
3013 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3014 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3017 case LTTNG_DISABLE_EVENT
:
3019 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3020 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3021 cmd_ctx
->lsm
->u
.disable
.name
);
3024 case LTTNG_DISABLE_ALL_EVENT
:
3026 DBG("Disabling all events");
3028 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3029 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3032 case LTTNG_ENABLE_CHANNEL
:
3034 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3035 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3038 case LTTNG_ENABLE_EVENT
:
3040 struct lttng_event_exclusion
*exclusion
= NULL
;
3041 struct lttng_filter_bytecode
*bytecode
= NULL
;
3043 /* Handle exclusion events and receive it from the client. */
3044 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3045 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3047 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3048 (count
* LTTNG_SYMBOL_NAME_LEN
));
3050 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3054 DBG("Receiving var len exclusion event list from client ...");
3055 exclusion
->count
= count
;
3056 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3057 count
* LTTNG_SYMBOL_NAME_LEN
);
3059 DBG("Nothing recv() from client var len data... continuing");
3062 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3067 /* Handle filter and get bytecode from client. */
3068 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3069 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3071 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3072 ret
= LTTNG_ERR_FILTER_INVAL
;
3077 bytecode
= zmalloc(bytecode_len
);
3080 ret
= LTTNG_ERR_FILTER_NOMEM
;
3084 /* Receive var. len. data */
3085 DBG("Receiving var len filter's bytecode from client ...");
3086 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3088 DBG("Nothing recv() from client car len data... continuing");
3092 ret
= LTTNG_ERR_FILTER_INVAL
;
3096 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3099 ret
= LTTNG_ERR_FILTER_INVAL
;
3104 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3105 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3106 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, exclusion
,
3107 kernel_poll_pipe
[1]);
3110 case LTTNG_ENABLE_ALL_EVENT
:
3112 DBG("Enabling all events");
3114 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3115 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3116 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
3119 case LTTNG_LIST_TRACEPOINTS
:
3121 struct lttng_event
*events
;
3124 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3125 if (nb_events
< 0) {
3126 /* Return value is a negative lttng_error_code. */
3132 * Setup lttng message with payload size set to the event list size in
3133 * bytes and then copy list into the llm payload.
3135 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3141 /* Copy event list into message payload */
3142 memcpy(cmd_ctx
->llm
->payload
, events
,
3143 sizeof(struct lttng_event
) * nb_events
);
3150 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3152 struct lttng_event_field
*fields
;
3155 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3157 if (nb_fields
< 0) {
3158 /* Return value is a negative lttng_error_code. */
3164 * Setup lttng message with payload size set to the event list size in
3165 * bytes and then copy list into the llm payload.
3167 ret
= setup_lttng_msg(cmd_ctx
,
3168 sizeof(struct lttng_event_field
) * nb_fields
);
3174 /* Copy event list into message payload */
3175 memcpy(cmd_ctx
->llm
->payload
, fields
,
3176 sizeof(struct lttng_event_field
) * nb_fields
);
3183 case LTTNG_SET_CONSUMER_URI
:
3186 struct lttng_uri
*uris
;
3188 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3189 len
= nb_uri
* sizeof(struct lttng_uri
);
3192 ret
= LTTNG_ERR_INVALID
;
3196 uris
= zmalloc(len
);
3198 ret
= LTTNG_ERR_FATAL
;
3202 /* Receive variable len data */
3203 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3204 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3206 DBG("No URIs received from client... continuing");
3208 ret
= LTTNG_ERR_SESSION_FAIL
;
3213 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3215 if (ret
!= LTTNG_OK
) {
3221 * XXX: 0 means that this URI should be applied on the session. Should
3222 * be a DOMAIN enuam.
3224 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3225 /* Add the URI for the UST session if a consumer is present. */
3226 if (cmd_ctx
->session
->ust_session
&&
3227 cmd_ctx
->session
->ust_session
->consumer
) {
3228 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3230 } else if (cmd_ctx
->session
->kernel_session
&&
3231 cmd_ctx
->session
->kernel_session
->consumer
) {
3232 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3233 cmd_ctx
->session
, nb_uri
, uris
);
3241 case LTTNG_START_TRACE
:
3243 ret
= cmd_start_trace(cmd_ctx
->session
);
3246 case LTTNG_STOP_TRACE
:
3248 ret
= cmd_stop_trace(cmd_ctx
->session
);
3251 case LTTNG_CREATE_SESSION
:
3254 struct lttng_uri
*uris
= NULL
;
3256 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3257 len
= nb_uri
* sizeof(struct lttng_uri
);
3260 uris
= zmalloc(len
);
3262 ret
= LTTNG_ERR_FATAL
;
3266 /* Receive variable len data */
3267 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3268 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3270 DBG("No URIs received from client... continuing");
3272 ret
= LTTNG_ERR_SESSION_FAIL
;
3277 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3278 DBG("Creating session with ONE network URI is a bad call");
3279 ret
= LTTNG_ERR_SESSION_FAIL
;
3285 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3286 &cmd_ctx
->creds
, 0);
3292 case LTTNG_DESTROY_SESSION
:
3294 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3296 /* Set session to NULL so we do not unlock it after free. */
3297 cmd_ctx
->session
= NULL
;
3300 case LTTNG_LIST_DOMAINS
:
3303 struct lttng_domain
*domains
;
3305 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3307 /* Return value is a negative lttng_error_code. */
3312 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3318 /* Copy event list into message payload */
3319 memcpy(cmd_ctx
->llm
->payload
, domains
,
3320 nb_dom
* sizeof(struct lttng_domain
));
3327 case LTTNG_LIST_CHANNELS
:
3330 struct lttng_channel
*channels
;
3332 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3333 cmd_ctx
->session
, &channels
);
3335 /* Return value is a negative lttng_error_code. */
3340 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3346 /* Copy event list into message payload */
3347 memcpy(cmd_ctx
->llm
->payload
, channels
,
3348 nb_chan
* sizeof(struct lttng_channel
));
3355 case LTTNG_LIST_EVENTS
:
3358 struct lttng_event
*events
= NULL
;
3360 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3361 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3363 /* Return value is a negative lttng_error_code. */
3368 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3374 /* Copy event list into message payload */
3375 memcpy(cmd_ctx
->llm
->payload
, events
,
3376 nb_event
* sizeof(struct lttng_event
));
3383 case LTTNG_LIST_SESSIONS
:
3385 unsigned int nr_sessions
;
3387 session_lock_list();
3388 nr_sessions
= lttng_sessions_count(
3389 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3390 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3392 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3394 session_unlock_list();
3398 /* Filled the session array */
3399 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3400 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3401 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3403 session_unlock_list();
3408 case LTTNG_CALIBRATE
:
3410 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3411 &cmd_ctx
->lsm
->u
.calibrate
);
3414 case LTTNG_REGISTER_CONSUMER
:
3416 struct consumer_data
*cdata
;
3418 switch (cmd_ctx
->lsm
->domain
.type
) {
3419 case LTTNG_DOMAIN_KERNEL
:
3420 cdata
= &kconsumer_data
;
3423 ret
= LTTNG_ERR_UND
;
3427 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3428 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3431 case LTTNG_DATA_PENDING
:
3433 ret
= cmd_data_pending(cmd_ctx
->session
);
3436 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3438 struct lttcomm_lttng_output_id reply
;
3440 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3441 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3442 if (ret
!= LTTNG_OK
) {
3446 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3451 /* Copy output list into message payload */
3452 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3456 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3458 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3459 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3462 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3465 struct lttng_snapshot_output
*outputs
= NULL
;
3467 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3468 if (nb_output
< 0) {
3473 ret
= setup_lttng_msg(cmd_ctx
,
3474 nb_output
* sizeof(struct lttng_snapshot_output
));
3481 /* Copy output list into message payload */
3482 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3483 nb_output
* sizeof(struct lttng_snapshot_output
));
3490 case LTTNG_SNAPSHOT_RECORD
:
3492 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3493 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3494 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3497 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3500 struct lttng_uri
*uris
= NULL
;
3502 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3503 len
= nb_uri
* sizeof(struct lttng_uri
);
3506 uris
= zmalloc(len
);
3508 ret
= LTTNG_ERR_FATAL
;
3512 /* Receive variable len data */
3513 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3514 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3516 DBG("No URIs received from client... continuing");
3518 ret
= LTTNG_ERR_SESSION_FAIL
;
3523 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3524 DBG("Creating session with ONE network URI is a bad call");
3525 ret
= LTTNG_ERR_SESSION_FAIL
;
3531 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3532 nb_uri
, &cmd_ctx
->creds
);
3536 case LTTNG_CREATE_SESSION_LIVE
:
3539 struct lttng_uri
*uris
= NULL
;
3541 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3542 len
= nb_uri
* sizeof(struct lttng_uri
);
3545 uris
= zmalloc(len
);
3547 ret
= LTTNG_ERR_FATAL
;
3551 /* Receive variable len data */
3552 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3553 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3555 DBG("No URIs received from client... continuing");
3557 ret
= LTTNG_ERR_SESSION_FAIL
;
3562 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3563 DBG("Creating session with ONE network URI is a bad call");
3564 ret
= LTTNG_ERR_SESSION_FAIL
;
3570 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3571 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3576 ret
= LTTNG_ERR_UND
;
3581 if (cmd_ctx
->llm
== NULL
) {
3582 DBG("Missing llm structure. Allocating one.");
3583 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3587 /* Set return code */
3588 cmd_ctx
->llm
->ret_code
= ret
;
3590 if (cmd_ctx
->session
) {
3591 session_unlock(cmd_ctx
->session
);
3593 if (need_tracing_session
) {
3594 session_unlock_list();
3601 * Thread managing health check socket.
3603 static void *thread_manage_health(void *data
)
3605 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3606 uint32_t revents
, nb_fd
;
3607 struct lttng_poll_event events
;
3608 struct health_comm_msg msg
;
3609 struct health_comm_reply reply
;
3611 DBG("[thread] Manage health check started");
3613 rcu_register_thread();
3615 /* We might hit an error path before this is created. */
3616 lttng_poll_init(&events
);
3618 /* Create unix socket */
3619 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3621 ERR("Unable to create health check Unix socket");
3627 /* lttng health client socket path permissions */
3628 ret
= chown(health_unix_sock_path
, 0,
3629 utils_get_group_id(tracing_group_name
));
3631 ERR("Unable to set group on %s", health_unix_sock_path
);
3637 ret
= chmod(health_unix_sock_path
,
3638 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3640 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3648 * Set the CLOEXEC flag. Return code is useless because either way, the
3651 (void) utils_set_fd_cloexec(sock
);
3653 ret
= lttcomm_listen_unix_sock(sock
);
3659 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3660 * more will be added to this poll set.
3662 ret
= sessiond_set_thread_pollset(&events
, 2);
3667 /* Add the application registration socket */
3668 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3673 lttng_sessiond_notify_ready();
3676 DBG("Health check ready");
3678 /* Inifinite blocking call, waiting for transmission */
3680 ret
= lttng_poll_wait(&events
, -1);
3683 * Restart interrupted system call.
3685 if (errno
== EINTR
) {
3693 for (i
= 0; i
< nb_fd
; i
++) {
3694 /* Fetch once the poll data */
3695 revents
= LTTNG_POLL_GETEV(&events
, i
);
3696 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3698 /* Thread quit pipe has been closed. Killing thread. */
3699 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3705 /* Event on the registration socket */
3706 if (pollfd
== sock
) {
3707 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3708 ERR("Health socket poll error");
3714 new_sock
= lttcomm_accept_unix_sock(sock
);
3720 * Set the CLOEXEC flag. Return code is useless because either way, the
3723 (void) utils_set_fd_cloexec(new_sock
);
3725 DBG("Receiving data from client for health...");
3726 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3728 DBG("Nothing recv() from client... continuing");
3729 ret
= close(new_sock
);
3737 rcu_thread_online();
3739 memset(&reply
, 0, sizeof(reply
));
3740 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3742 * health_check_state returns 0 if health is
3745 if (!health_check_state(health_sessiond
, i
)) {
3746 reply
.ret_code
|= 1ULL << i
;
3750 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3752 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3754 ERR("Failed to send health data back to client");
3757 /* End of transmission */
3758 ret
= close(new_sock
);
3768 ERR("Health error occurred in %s", __func__
);
3770 DBG("Health check thread dying");
3771 unlink(health_unix_sock_path
);
3779 lttng_poll_clean(&events
);
3781 rcu_unregister_thread();
3786 * This thread manage all clients request using the unix client socket for
3789 static void *thread_manage_clients(void *data
)
3791 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3793 uint32_t revents
, nb_fd
;
3794 struct command_ctx
*cmd_ctx
= NULL
;
3795 struct lttng_poll_event events
;
3797 DBG("[thread] Manage client started");
3799 rcu_register_thread();
3801 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3803 health_code_update();
3805 ret
= lttcomm_listen_unix_sock(client_sock
);
3811 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3812 * more will be added to this poll set.
3814 ret
= sessiond_set_thread_pollset(&events
, 2);
3816 goto error_create_poll
;
3819 /* Add the application registration socket */
3820 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3825 lttng_sessiond_notify_ready();
3827 /* This testpoint is after we signal readiness to the parent. */
3828 if (testpoint(sessiond_thread_manage_clients
)) {
3832 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
3836 health_code_update();
3839 DBG("Accepting client command ...");
3841 /* Inifinite blocking call, waiting for transmission */
3843 health_poll_entry();
3844 ret
= lttng_poll_wait(&events
, -1);
3848 * Restart interrupted system call.
3850 if (errno
== EINTR
) {
3858 for (i
= 0; i
< nb_fd
; i
++) {
3859 /* Fetch once the poll data */
3860 revents
= LTTNG_POLL_GETEV(&events
, i
);
3861 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3863 health_code_update();
3865 /* Thread quit pipe has been closed. Killing thread. */
3866 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3872 /* Event on the registration socket */
3873 if (pollfd
== client_sock
) {
3874 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3875 ERR("Client socket poll error");
3881 DBG("Wait for client response");
3883 health_code_update();
3885 sock
= lttcomm_accept_unix_sock(client_sock
);
3891 * Set the CLOEXEC flag. Return code is useless because either way, the
3894 (void) utils_set_fd_cloexec(sock
);
3896 /* Set socket option for credentials retrieval */
3897 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3902 /* Allocate context command to process the client request */
3903 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3904 if (cmd_ctx
== NULL
) {
3905 PERROR("zmalloc cmd_ctx");
3909 /* Allocate data buffer for reception */
3910 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3911 if (cmd_ctx
->lsm
== NULL
) {
3912 PERROR("zmalloc cmd_ctx->lsm");
3916 cmd_ctx
->llm
= NULL
;
3917 cmd_ctx
->session
= NULL
;
3919 health_code_update();
3922 * Data is received from the lttng client. The struct
3923 * lttcomm_session_msg (lsm) contains the command and data request of
3926 DBG("Receiving data from client ...");
3927 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3928 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3930 DBG("Nothing recv() from client... continuing");
3936 clean_command_ctx(&cmd_ctx
);
3940 health_code_update();
3942 // TODO: Validate cmd_ctx including sanity check for
3943 // security purpose.
3945 rcu_thread_online();
3947 * This function dispatch the work to the kernel or userspace tracer
3948 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3949 * informations for the client. The command context struct contains
3950 * everything this function may needs.
3952 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3953 rcu_thread_offline();
3961 * TODO: Inform client somehow of the fatal error. At
3962 * this point, ret < 0 means that a zmalloc failed
3963 * (ENOMEM). Error detected but still accept
3964 * command, unless a socket error has been
3967 clean_command_ctx(&cmd_ctx
);
3971 health_code_update();
3973 DBG("Sending response (size: %d, retcode: %s)",
3974 cmd_ctx
->lttng_msg_size
,
3975 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3976 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3978 ERR("Failed to send data back to client");
3981 /* End of transmission */
3988 clean_command_ctx(&cmd_ctx
);
3990 health_code_update();
4002 lttng_poll_clean(&events
);
4003 clean_command_ctx(&cmd_ctx
);
4007 unlink(client_unix_sock_path
);
4008 if (client_sock
>= 0) {
4009 ret
= close(client_sock
);
4017 ERR("Health error occurred in %s", __func__
);
4020 health_unregister(health_sessiond
);
4022 DBG("Client thread dying");
4024 rcu_unregister_thread();
4030 * usage function on stderr
4032 static void usage(void)
4034 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4035 fprintf(stderr
, " -h, --help Display this usage.\n");
4036 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4037 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4038 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4039 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4040 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4041 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4042 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4043 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4044 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4045 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4046 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4047 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4048 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4049 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4050 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4051 fprintf(stderr
, " -V, --version Show version number.\n");
4052 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4053 fprintf(stderr
, " -q, --quiet No output at all.\n");
4054 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4055 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4056 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4057 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4058 fprintf(stderr
, " --jul-tcp-port JUL application registration TCP port\n");
4059 fprintf(stderr
, " -f --config Load daemon configuration file\n");
4063 * Take an option from the getopt output and set it in the right variable to be
4066 * Return 0 on success else a negative value.
4068 static int set_option(int opt
, const char *arg
, const char *optname
)
4074 fprintf(stderr
, "option %s", optname
);
4076 fprintf(stderr
, " with arg %s\n", arg
);
4080 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4083 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4092 tracing_group_name
= strdup(arg
);
4098 fprintf(stdout
, "%s\n", VERSION
);
4104 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4107 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4110 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4113 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4116 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4119 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4125 lttng_opt_quiet
= 1;
4128 /* Verbose level can increase using multiple -v */
4130 lttng_opt_verbose
= config_parse_value(arg
);
4132 lttng_opt_verbose
+= 1;
4137 opt_verbose_consumer
= config_parse_value(arg
);
4139 opt_verbose_consumer
+= 1;
4143 consumerd32_bin
= strdup(arg
);
4144 consumerd32_bin_override
= 1;
4147 consumerd32_libdir
= strdup(arg
);
4148 consumerd32_libdir_override
= 1;
4151 consumerd64_bin
= strdup(arg
);
4152 consumerd64_bin_override
= 1;
4155 consumerd64_libdir
= strdup(arg
);
4156 consumerd64_libdir_override
= 1;
4159 opt_pidfile
= strdup(arg
);
4161 case 'J': /* JUL TCP port. */
4166 v
= strtoul(arg
, NULL
, 0);
4167 if (errno
!= 0 || !isdigit(arg
[0])) {
4168 ERR("Wrong value in --jul-tcp-port parameter: %s", arg
);
4171 if (v
== 0 || v
>= 65535) {
4172 ERR("Port overflow in --jul-tcp-port parameter: %s", arg
);
4175 jul_tcp_port
= (uint32_t) v
;
4176 DBG3("JUL TCP port set to non default: %u", jul_tcp_port
);
4180 /* Unknown option or other error.
4181 * Error is printed by getopt, just return */
4189 * config_entry_handler_cb used to handle options read from a config file.
4190 * See config_entry_handler_cb comment in common/config/config.h for the
4191 * return value conventions.
4193 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4197 if (!entry
|| !entry
->name
|| !entry
->value
) {
4202 /* Check if the option is to be ignored */
4203 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4204 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4209 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4212 /* Ignore if not fully matched. */
4213 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4218 * If the option takes no argument on the command line, we have to
4219 * check if the value is "true". We support non-zero numeric values,
4222 if (!long_options
[i
].has_arg
) {
4223 ret
= config_parse_value(entry
->value
);
4226 WARN("Invalid configuration value \"%s\" for option %s",
4227 entry
->value
, entry
->name
);
4229 /* False, skip boolean config option. */
4234 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4238 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4245 * daemon configuration loading and argument parsing
4247 static int set_options(int argc
, char **argv
)
4249 int ret
= 0, c
= 0, option_index
= 0;
4250 int orig_optopt
= optopt
, orig_optind
= optind
;
4252 const char *config_path
= NULL
;
4254 optstring
= utils_generate_optstring(long_options
,
4255 sizeof(long_options
) / sizeof(struct option
));
4261 /* Check for the --config option */
4262 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4263 &option_index
)) != -1) {
4267 } else if (c
!= 'f') {
4268 /* if not equal to --config option. */
4272 config_path
= utils_expand_path(optarg
);
4274 ERR("Failed to resolve path: %s", optarg
);
4278 ret
= config_get_section_entries(config_path
, config_section_name
,
4279 config_entry_handler
, NULL
);
4282 ERR("Invalid configuration option at line %i", ret
);
4288 /* Reset getopt's global state */
4289 optopt
= orig_optopt
;
4290 optind
= orig_optind
;
4292 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
4297 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
4309 * Creates the two needed socket by the daemon.
4310 * apps_sock - The communication socket for all UST apps.
4311 * client_sock - The communication of the cli tool (lttng).
4313 static int init_daemon_socket(void)
4318 old_umask
= umask(0);
4320 /* Create client tool unix socket */
4321 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4322 if (client_sock
< 0) {
4323 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4328 /* Set the cloexec flag */
4329 ret
= utils_set_fd_cloexec(client_sock
);
4331 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4332 "Continuing but note that the consumer daemon will have a "
4333 "reference to this socket on exec()", client_sock
);
4336 /* File permission MUST be 660 */
4337 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4339 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4344 /* Create the application unix socket */
4345 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4346 if (apps_sock
< 0) {
4347 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4352 /* Set the cloexec flag */
4353 ret
= utils_set_fd_cloexec(apps_sock
);
4355 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4356 "Continuing but note that the consumer daemon will have a "
4357 "reference to this socket on exec()", apps_sock
);
4360 /* File permission MUST be 666 */
4361 ret
= chmod(apps_unix_sock_path
,
4362 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4364 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4369 DBG3("Session daemon client socket %d and application socket %d created",
4370 client_sock
, apps_sock
);
4378 * Check if the global socket is available, and if a daemon is answering at the
4379 * other side. If yes, error is returned.
4381 static int check_existing_daemon(void)
4383 /* Is there anybody out there ? */
4384 if (lttng_session_daemon_alive()) {
4392 * Set the tracing group gid onto the client socket.
4394 * Race window between mkdir and chown is OK because we are going from more
4395 * permissive (root.root) to less permissive (root.tracing).
4397 static int set_permissions(char *rundir
)
4402 gid
= utils_get_group_id(tracing_group_name
);
4404 /* Set lttng run dir */
4405 ret
= chown(rundir
, 0, gid
);
4407 ERR("Unable to set group on %s", rundir
);
4412 * Ensure all applications and tracing group can search the run
4413 * dir. Allow everyone to read the directory, since it does not
4414 * buy us anything to hide its content.
4416 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4418 ERR("Unable to set permissions on %s", rundir
);
4422 /* lttng client socket path */
4423 ret
= chown(client_unix_sock_path
, 0, gid
);
4425 ERR("Unable to set group on %s", client_unix_sock_path
);
4429 /* kconsumer error socket path */
4430 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4432 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4436 /* 64-bit ustconsumer error socket path */
4437 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4439 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4443 /* 32-bit ustconsumer compat32 error socket path */
4444 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4446 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4450 DBG("All permissions are set");
4456 * Create the lttng run directory needed for all global sockets and pipe.
4458 static int create_lttng_rundir(const char *rundir
)
4462 DBG3("Creating LTTng run directory: %s", rundir
);
4464 ret
= mkdir(rundir
, S_IRWXU
);
4466 if (errno
!= EEXIST
) {
4467 ERR("Unable to create %s", rundir
);
4479 * Setup sockets and directory needed by the kconsumerd communication with the
4482 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4486 char path
[PATH_MAX
];
4488 switch (consumer_data
->type
) {
4489 case LTTNG_CONSUMER_KERNEL
:
4490 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4492 case LTTNG_CONSUMER64_UST
:
4493 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4495 case LTTNG_CONSUMER32_UST
:
4496 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4499 ERR("Consumer type unknown");
4504 DBG2("Creating consumer directory: %s", path
);
4506 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4508 if (errno
!= EEXIST
) {
4510 ERR("Failed to create %s", path
);
4516 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4518 ERR("Unable to set group on %s", path
);
4524 /* Create the kconsumerd error unix socket */
4525 consumer_data
->err_sock
=
4526 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4527 if (consumer_data
->err_sock
< 0) {
4528 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4534 * Set the CLOEXEC flag. Return code is useless because either way, the
4537 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4539 PERROR("utils_set_fd_cloexec");
4540 /* continue anyway */
4543 /* File permission MUST be 660 */
4544 ret
= chmod(consumer_data
->err_unix_sock_path
,
4545 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4547 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4557 * Signal handler for the daemon
4559 * Simply stop all worker threads, leaving main() return gracefully after
4560 * joining all threads and calling cleanup().
4562 static void sighandler(int sig
)
4566 DBG("SIGPIPE caught");
4569 DBG("SIGINT caught");
4573 DBG("SIGTERM caught");
4577 CMM_STORE_SHARED(recv_child_signal
, 1);
4585 * Setup signal handler for :
4586 * SIGINT, SIGTERM, SIGPIPE
4588 static int set_signal_handler(void)
4591 struct sigaction sa
;
4594 if ((ret
= sigemptyset(&sigset
)) < 0) {
4595 PERROR("sigemptyset");
4599 sa
.sa_handler
= sighandler
;
4600 sa
.sa_mask
= sigset
;
4602 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4603 PERROR("sigaction");
4607 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4608 PERROR("sigaction");
4612 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4613 PERROR("sigaction");
4617 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4618 PERROR("sigaction");
4622 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4628 * Set open files limit to unlimited. This daemon can open a large number of
4629 * file descriptors in order to consumer multiple kernel traces.
4631 static void set_ulimit(void)
4636 /* The kernel does not allowed an infinite limit for open files */
4637 lim
.rlim_cur
= 65535;
4638 lim
.rlim_max
= 65535;
4640 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4642 PERROR("failed to set open files limit");
4647 * Write pidfile using the rundir and opt_pidfile.
4649 static void write_pidfile(void)
4652 char pidfile_path
[PATH_MAX
];
4657 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4659 /* Build pidfile path from rundir and opt_pidfile. */
4660 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4661 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4663 PERROR("snprintf pidfile path");
4669 * Create pid file in rundir. Return value is of no importance. The
4670 * execution will continue even though we are not able to write the file.
4672 (void) utils_create_pid_file(getpid(), pidfile_path
);
4679 * Write JUL TCP port using the rundir.
4681 static void write_julport(void)
4684 char path
[PATH_MAX
];
4688 ret
= snprintf(path
, sizeof(path
), "%s/"
4689 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
, rundir
);
4691 PERROR("snprintf julport path");
4696 * Create TCP JUL port file in rundir. Return value is of no importance.
4697 * The execution will continue even though we are not able to write the
4700 (void) utils_create_pid_file(jul_tcp_port
, path
);
4709 int main(int argc
, char **argv
)
4713 const char *home_path
, *env_app_timeout
;
4715 init_kernel_workarounds();
4717 rcu_register_thread();
4719 if ((ret
= set_signal_handler()) < 0) {
4723 setup_consumerd_path();
4725 page_size
= sysconf(_SC_PAGESIZE
);
4726 if (page_size
< 0) {
4727 PERROR("sysconf _SC_PAGESIZE");
4728 page_size
= LONG_MAX
;
4729 WARN("Fallback page size to %ld", page_size
);
4732 /* Parse arguments and load the daemon configuration file */
4734 if ((ret
= set_options(argc
, argv
)) < 0) {
4739 if (opt_daemon
|| opt_background
) {
4742 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4749 * We are in the child. Make sure all other file descriptors are
4750 * closed, in case we are called with more opened file descriptors than
4751 * the standard ones.
4753 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4758 /* Create thread quit pipe */
4759 if ((ret
= init_thread_quit_pipe()) < 0) {
4763 /* Check if daemon is UID = 0 */
4764 is_root
= !getuid();
4767 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4769 /* Create global run dir with root access */
4770 ret
= create_lttng_rundir(rundir
);
4775 if (strlen(apps_unix_sock_path
) == 0) {
4776 snprintf(apps_unix_sock_path
, PATH_MAX
,
4777 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4780 if (strlen(client_unix_sock_path
) == 0) {
4781 snprintf(client_unix_sock_path
, PATH_MAX
,
4782 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4785 /* Set global SHM for ust */
4786 if (strlen(wait_shm_path
) == 0) {
4787 snprintf(wait_shm_path
, PATH_MAX
,
4788 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4791 if (strlen(health_unix_sock_path
) == 0) {
4792 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4793 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4796 /* Setup kernel consumerd path */
4797 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4798 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4799 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4800 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4802 DBG2("Kernel consumer err path: %s",
4803 kconsumer_data
.err_unix_sock_path
);
4804 DBG2("Kernel consumer cmd path: %s",
4805 kconsumer_data
.cmd_unix_sock_path
);
4807 home_path
= utils_get_home_dir();
4808 if (home_path
== NULL
) {
4809 /* TODO: Add --socket PATH option */
4810 ERR("Can't get HOME directory for sockets creation.");
4816 * Create rundir from home path. This will create something like
4819 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4825 ret
= create_lttng_rundir(rundir
);
4830 if (strlen(apps_unix_sock_path
) == 0) {
4831 snprintf(apps_unix_sock_path
, PATH_MAX
,
4832 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4835 /* Set the cli tool unix socket path */
4836 if (strlen(client_unix_sock_path
) == 0) {
4837 snprintf(client_unix_sock_path
, PATH_MAX
,
4838 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4841 /* Set global SHM for ust */
4842 if (strlen(wait_shm_path
) == 0) {
4843 snprintf(wait_shm_path
, PATH_MAX
,
4844 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4847 /* Set health check Unix path */
4848 if (strlen(health_unix_sock_path
) == 0) {
4849 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4850 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4854 /* Set consumer initial state */
4855 kernel_consumerd_state
= CONSUMER_STOPPED
;
4856 ust_consumerd_state
= CONSUMER_STOPPED
;
4858 DBG("Client socket path %s", client_unix_sock_path
);
4859 DBG("Application socket path %s", apps_unix_sock_path
);
4860 DBG("Application wait path %s", wait_shm_path
);
4861 DBG("LTTng run directory path: %s", rundir
);
4863 /* 32 bits consumerd path setup */
4864 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4865 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4866 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4867 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4869 DBG2("UST consumer 32 bits err path: %s",
4870 ustconsumer32_data
.err_unix_sock_path
);
4871 DBG2("UST consumer 32 bits cmd path: %s",
4872 ustconsumer32_data
.cmd_unix_sock_path
);
4874 /* 64 bits consumerd path setup */
4875 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4876 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4877 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4878 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4880 DBG2("UST consumer 64 bits err path: %s",
4881 ustconsumer64_data
.err_unix_sock_path
);
4882 DBG2("UST consumer 64 bits cmd path: %s",
4883 ustconsumer64_data
.cmd_unix_sock_path
);
4886 * See if daemon already exist.
4888 if ((ret
= check_existing_daemon()) < 0) {
4889 ERR("Already running daemon.\n");
4891 * We do not goto exit because we must not cleanup()
4892 * because a daemon is already running.
4898 * Init UST app hash table. Alloc hash table before this point since
4899 * cleanup() can get called after that point.
4903 /* Initialize JUL domain subsystem. */
4904 if ((ret
= jul_init()) < 0) {
4905 /* ENOMEM at this point. */
4909 /* After this point, we can safely call cleanup() with "goto exit" */
4912 * These actions must be executed as root. We do that *after* setting up
4913 * the sockets path because we MUST make the check for another daemon using
4914 * those paths *before* trying to set the kernel consumer sockets and init
4918 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4923 /* Setup kernel tracer */
4924 if (!opt_no_kernel
) {
4925 init_kernel_tracer();
4928 /* Set ulimit for open files */
4931 /* init lttng_fd tracking must be done after set_ulimit. */
4934 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4939 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4944 /* Setup the needed unix socket */
4945 if ((ret
= init_daemon_socket()) < 0) {
4949 /* Set credentials to socket */
4950 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4954 /* Get parent pid if -S, --sig-parent is specified. */
4955 if (opt_sig_parent
) {
4959 /* Setup the kernel pipe for waking up the kernel thread */
4960 if (is_root
&& !opt_no_kernel
) {
4961 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4966 /* Setup the thread ht_cleanup communication pipe. */
4967 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4971 /* Setup the thread apps communication pipe. */
4972 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4976 /* Setup the thread apps notify communication pipe. */
4977 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4981 /* Initialize global buffer per UID and PID registry. */
4982 buffer_reg_init_uid_registry();
4983 buffer_reg_init_pid_registry();
4985 /* Init UST command queue. */
4986 cds_wfq_init(&ust_cmd_queue
.queue
);
4989 * Get session list pointer. This pointer MUST NOT be free(). This list is
4990 * statically declared in session.c
4992 session_list_ptr
= session_get_list();
4994 /* Set up max poll set size */
4995 lttng_poll_set_max_size();
4999 /* Check for the application socket timeout env variable. */
5000 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5001 if (env_app_timeout
) {
5002 app_socket_timeout
= atoi(env_app_timeout
);
5004 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5010 /* Initialize communication library */
5012 /* This is to get the TCP timeout value. */
5013 lttcomm_inet_init();
5016 * Initialize the health check subsystem. This call should set the
5017 * appropriate time values.
5019 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5020 if (!health_sessiond
) {
5021 PERROR("health_app_create error");
5022 goto exit_health_sessiond_cleanup
;
5025 /* Create thread to clean up RCU hash tables */
5026 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5027 thread_ht_cleanup
, (void *) NULL
);
5029 PERROR("pthread_create ht_cleanup");
5030 goto exit_ht_cleanup
;
5033 /* Create health-check thread */
5034 ret
= pthread_create(&health_thread
, NULL
,
5035 thread_manage_health
, (void *) NULL
);
5037 PERROR("pthread_create health");
5041 /* Create thread to manage the client socket */
5042 ret
= pthread_create(&client_thread
, NULL
,
5043 thread_manage_clients
, (void *) NULL
);
5045 PERROR("pthread_create clients");
5049 /* Create thread to dispatch registration */
5050 ret
= pthread_create(&dispatch_thread
, NULL
,
5051 thread_dispatch_ust_registration
, (void *) NULL
);
5053 PERROR("pthread_create dispatch");
5057 /* Create thread to manage application registration. */
5058 ret
= pthread_create(®_apps_thread
, NULL
,
5059 thread_registration_apps
, (void *) NULL
);
5061 PERROR("pthread_create registration");
5065 /* Create thread to manage application socket */
5066 ret
= pthread_create(&apps_thread
, NULL
,
5067 thread_manage_apps
, (void *) NULL
);
5069 PERROR("pthread_create apps");
5073 /* Create thread to manage application notify socket */
5074 ret
= pthread_create(&apps_notify_thread
, NULL
,
5075 ust_thread_manage_notify
, (void *) NULL
);
5077 PERROR("pthread_create apps");
5078 goto exit_apps_notify
;
5081 /* Create JUL registration thread. */
5082 ret
= pthread_create(&jul_reg_thread
, NULL
,
5083 jul_thread_manage_registration
, (void *) NULL
);
5085 PERROR("pthread_create apps");
5089 /* Don't start this thread if kernel tracing is not requested nor root */
5090 if (is_root
&& !opt_no_kernel
) {
5091 /* Create kernel thread to manage kernel event */
5092 ret
= pthread_create(&kernel_thread
, NULL
,
5093 thread_manage_kernel
, (void *) NULL
);
5095 PERROR("pthread_create kernel");
5099 ret
= pthread_join(kernel_thread
, &status
);
5101 PERROR("pthread_join");
5102 goto error
; /* join error, exit without cleanup */
5107 ret
= pthread_join(jul_reg_thread
, &status
);
5109 PERROR("pthread_join JUL");
5110 goto error
; /* join error, exit without cleanup */
5114 ret
= pthread_join(apps_notify_thread
, &status
);
5116 PERROR("pthread_join apps notify");
5117 goto error
; /* join error, exit without cleanup */
5121 ret
= pthread_join(apps_thread
, &status
);
5123 PERROR("pthread_join apps");
5124 goto error
; /* join error, exit without cleanup */
5129 ret
= pthread_join(reg_apps_thread
, &status
);
5131 PERROR("pthread_join");
5132 goto error
; /* join error, exit without cleanup */
5136 ret
= pthread_join(dispatch_thread
, &status
);
5138 PERROR("pthread_join");
5139 goto error
; /* join error, exit without cleanup */
5143 ret
= pthread_join(client_thread
, &status
);
5145 PERROR("pthread_join");
5146 goto error
; /* join error, exit without cleanup */
5149 ret
= join_consumer_thread(&kconsumer_data
);
5151 PERROR("join_consumer");
5152 goto error
; /* join error, exit without cleanup */
5155 ret
= join_consumer_thread(&ustconsumer32_data
);
5157 PERROR("join_consumer ust32");
5158 goto error
; /* join error, exit without cleanup */
5161 ret
= join_consumer_thread(&ustconsumer64_data
);
5163 PERROR("join_consumer ust64");
5164 goto error
; /* join error, exit without cleanup */
5168 ret
= pthread_join(health_thread
, &status
);
5170 PERROR("pthread_join health thread");
5171 goto error
; /* join error, exit without cleanup */
5175 ret
= pthread_join(ht_cleanup_thread
, &status
);
5177 PERROR("pthread_join ht cleanup thread");
5178 goto error
; /* join error, exit without cleanup */
5181 health_app_destroy(health_sessiond
);
5182 exit_health_sessiond_cleanup
:
5185 * cleanup() is called when no other thread is running.
5187 rcu_thread_online();
5189 rcu_thread_offline();
5190 rcu_unregister_thread();