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/config/config.h>
51 #include "lttng-sessiond.h"
52 #include "buffer-registry.h"
59 #include "kernel-consumer.h"
63 #include "ust-consumer.h"
66 #include "health-sessiond.h"
67 #include "testpoint.h"
68 #include "ust-thread.h"
69 #include "jul-thread.h"
71 #define CONSUMERD_FILE "lttng-consumerd"
74 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
75 static int tracing_group_name_override
;
76 static char *opt_pidfile
;
77 static int opt_sig_parent
;
78 static int opt_verbose_consumer
;
79 static int opt_daemon
;
80 static int opt_no_kernel
;
81 static int is_root
; /* Set to 1 if the daemon is running as root */
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 { "sig-parent", 0, 0, 'S' },
143 { "help", 0, 0, 'h' },
144 { "group", 1, 0, 'g' },
145 { "version", 0, 0, 'V' },
146 { "quiet", 0, 0, 'q' },
147 { "verbose", 0, 0, 'v' },
148 { "verbose-consumer", 0, 0, 'Z' },
149 { "no-kernel", 0, 0, 'N' },
150 { "pidfile", 1, 0, 'p' },
151 { "jul-tcp-port", 1, 0, 'J' },
152 { "config", 1, 0, 'f' },
156 /* Command line options to ignore from configuration file */
157 static const char *config_ignore_options
[] = { "help", "version", "config" };
159 /* Shared between threads */
160 static int dispatch_thread_exit
;
162 /* Global application Unix socket path */
163 static char apps_unix_sock_path
[PATH_MAX
];
164 /* Global client Unix socket path */
165 static char client_unix_sock_path
[PATH_MAX
];
166 /* global wait shm path for UST */
167 static char wait_shm_path
[PATH_MAX
];
168 /* Global health check unix path */
169 static char health_unix_sock_path
[PATH_MAX
];
171 /* Sockets and FDs */
172 static int client_sock
= -1;
173 static int apps_sock
= -1;
174 int kernel_tracer_fd
= -1;
175 static int kernel_poll_pipe
[2] = { -1, -1 };
178 * Quit pipe for all threads. This permits a single cancellation point
179 * for all threads when receiving an event on the pipe.
181 static int thread_quit_pipe
[2] = { -1, -1 };
184 * This pipe is used to inform the thread managing application communication
185 * that a command is queued and ready to be processed.
187 static int apps_cmd_pipe
[2] = { -1, -1 };
189 int apps_cmd_notify_pipe
[2] = { -1, -1 };
191 /* Pthread, Mutexes and Semaphores */
192 static pthread_t apps_thread
;
193 static pthread_t apps_notify_thread
;
194 static pthread_t reg_apps_thread
;
195 static pthread_t client_thread
;
196 static pthread_t kernel_thread
;
197 static pthread_t dispatch_thread
;
198 static pthread_t health_thread
;
199 static pthread_t ht_cleanup_thread
;
200 static pthread_t jul_reg_thread
;
203 * UST registration command queue. This queue is tied with a futex and uses a N
204 * wakers / 1 waiter implemented and detailed in futex.c/.h
206 * The thread_manage_apps and thread_dispatch_ust_registration interact with
207 * this queue and the wait/wake scheme.
209 static struct ust_cmd_queue ust_cmd_queue
;
212 * Pointer initialized before thread creation.
214 * This points to the tracing session list containing the session count and a
215 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
216 * MUST NOT be taken if you call a public function in session.c.
218 * The lock is nested inside the structure: session_list_ptr->lock. Please use
219 * session_lock_list and session_unlock_list for lock acquisition.
221 static struct ltt_session_list
*session_list_ptr
;
223 int ust_consumerd64_fd
= -1;
224 int ust_consumerd32_fd
= -1;
226 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
227 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
228 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
229 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
230 static int consumerd32_bin_override
;
231 static int consumerd64_bin_override
;
232 static int consumerd32_libdir_override
;
233 static int consumerd64_libdir_override
;
235 static const char *module_proc_lttng
= "/proc/lttng";
238 * Consumer daemon state which is changed when spawning it, killing it or in
239 * case of a fatal error.
241 enum consumerd_state
{
242 CONSUMER_STARTED
= 1,
243 CONSUMER_STOPPED
= 2,
248 * This consumer daemon state is used to validate if a client command will be
249 * able to reach the consumer. If not, the client is informed. For instance,
250 * doing a "lttng start" when the consumer state is set to ERROR will return an
251 * error to the client.
253 * The following example shows a possible race condition of this scheme:
255 * consumer thread error happens
257 * client cmd checks state -> still OK
258 * consumer thread exit, sets error
259 * client cmd try to talk to consumer
262 * However, since the consumer is a different daemon, we have no way of making
263 * sure the command will reach it safely even with this state flag. This is why
264 * we consider that up to the state validation during command processing, the
265 * command is safe. After that, we can not guarantee the correctness of the
266 * client request vis-a-vis the consumer.
268 static enum consumerd_state ust_consumerd_state
;
269 static enum consumerd_state kernel_consumerd_state
;
272 * Socket timeout for receiving and sending in seconds.
274 static int app_socket_timeout
;
276 /* Set in main() with the current page size. */
279 /* Application health monitoring */
280 struct health_app
*health_sessiond
;
282 /* JUL TCP port for registration. Used by the JUL thread. */
283 unsigned int jul_tcp_port
= DEFAULT_JUL_TCP_PORT
;
285 const char * const config_section_name
= "sessiond";
288 void setup_consumerd_path(void)
290 const char *bin
, *libdir
;
293 * Allow INSTALL_BIN_PATH to be used as a target path for the
294 * native architecture size consumer if CONFIG_CONSUMER*_PATH
295 * has not been defined.
297 #if (CAA_BITS_PER_LONG == 32)
298 if (!consumerd32_bin
[0]) {
299 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
301 if (!consumerd32_libdir
[0]) {
302 consumerd32_libdir
= INSTALL_LIB_PATH
;
304 #elif (CAA_BITS_PER_LONG == 64)
305 if (!consumerd64_bin
[0]) {
306 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
308 if (!consumerd64_libdir
[0]) {
309 consumerd64_libdir
= INSTALL_LIB_PATH
;
312 #error "Unknown bitness"
316 * runtime env. var. overrides the build default.
318 bin
= getenv("LTTNG_CONSUMERD32_BIN");
320 consumerd32_bin
= bin
;
322 bin
= getenv("LTTNG_CONSUMERD64_BIN");
324 consumerd64_bin
= bin
;
326 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
328 consumerd32_libdir
= libdir
;
330 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
332 consumerd64_libdir
= libdir
;
337 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
339 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
345 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
351 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
363 * Check if the thread quit pipe was triggered.
365 * Return 1 if it was triggered else 0;
367 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
369 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
377 * Init thread quit pipe.
379 * Return -1 on error or 0 if all pipes are created.
381 static int init_thread_quit_pipe(void)
385 ret
= pipe(thread_quit_pipe
);
387 PERROR("thread quit pipe");
391 for (i
= 0; i
< 2; i
++) {
392 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
404 * Stop all threads by closing the thread quit pipe.
406 static void stop_threads(void)
410 /* Stopping all threads */
411 DBG("Terminating all threads");
412 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
414 ERR("write error on thread quit pipe");
417 /* Dispatch thread */
418 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
419 futex_nto1_wake(&ust_cmd_queue
.futex
);
423 * Close every consumer sockets.
425 static void close_consumer_sockets(void)
429 if (kconsumer_data
.err_sock
>= 0) {
430 ret
= close(kconsumer_data
.err_sock
);
432 PERROR("kernel consumer err_sock close");
435 if (ustconsumer32_data
.err_sock
>= 0) {
436 ret
= close(ustconsumer32_data
.err_sock
);
438 PERROR("UST consumerd32 err_sock close");
441 if (ustconsumer64_data
.err_sock
>= 0) {
442 ret
= close(ustconsumer64_data
.err_sock
);
444 PERROR("UST consumerd64 err_sock close");
447 if (kconsumer_data
.cmd_sock
>= 0) {
448 ret
= close(kconsumer_data
.cmd_sock
);
450 PERROR("kernel consumer cmd_sock close");
453 if (ustconsumer32_data
.cmd_sock
>= 0) {
454 ret
= close(ustconsumer32_data
.cmd_sock
);
456 PERROR("UST consumerd32 cmd_sock close");
459 if (ustconsumer64_data
.cmd_sock
>= 0) {
460 ret
= close(ustconsumer64_data
.cmd_sock
);
462 PERROR("UST consumerd64 cmd_sock close");
470 static void cleanup(void)
473 struct ltt_session
*sess
, *stmp
;
479 * Close the thread quit pipe. It has already done its job,
480 * since we are now called.
482 utils_close_pipe(thread_quit_pipe
);
485 * If opt_pidfile is undefined, the default file will be wiped when
486 * removing the rundir.
489 ret
= remove(opt_pidfile
);
491 PERROR("remove pidfile %s", opt_pidfile
);
495 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
498 snprintf(path
, PATH_MAX
,
500 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
501 DBG("Removing %s", path
);
504 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
505 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
);
506 DBG("Removing %s", path
);
510 snprintf(path
, PATH_MAX
,
511 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
513 DBG("Removing %s", path
);
516 snprintf(path
, PATH_MAX
,
517 DEFAULT_KCONSUMERD_PATH
,
519 DBG("Removing directory %s", path
);
522 /* ust consumerd 32 */
523 snprintf(path
, PATH_MAX
,
524 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
526 DBG("Removing %s", path
);
529 snprintf(path
, PATH_MAX
,
530 DEFAULT_USTCONSUMERD32_PATH
,
532 DBG("Removing directory %s", path
);
535 /* ust consumerd 64 */
536 snprintf(path
, PATH_MAX
,
537 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
539 DBG("Removing %s", path
);
542 snprintf(path
, PATH_MAX
,
543 DEFAULT_USTCONSUMERD64_PATH
,
545 DBG("Removing directory %s", path
);
549 * We do NOT rmdir rundir because there are other processes
550 * using it, for instance lttng-relayd, which can start in
551 * parallel with this teardown.
556 DBG("Cleaning up all sessions");
558 /* Destroy session list mutex */
559 if (session_list_ptr
!= NULL
) {
560 pthread_mutex_destroy(&session_list_ptr
->lock
);
562 /* Cleanup ALL session */
563 cds_list_for_each_entry_safe(sess
, stmp
,
564 &session_list_ptr
->head
, list
) {
565 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
569 DBG("Closing all UST sockets");
570 ust_app_clean_list();
571 buffer_reg_destroy_registries();
573 if (is_root
&& !opt_no_kernel
) {
574 DBG2("Closing kernel fd");
575 if (kernel_tracer_fd
>= 0) {
576 ret
= close(kernel_tracer_fd
);
581 DBG("Unloading kernel modules");
582 modprobe_remove_lttng_all();
585 close_consumer_sockets();
588 * If the override option is set, the pointer points to a *non* const thus
589 * freeing it even though the variable type is set to const.
591 if (tracing_group_name_override
) {
592 free((void *) tracing_group_name
);
594 if (consumerd32_bin_override
) {
595 free((void *) consumerd32_bin
);
597 if (consumerd64_bin_override
) {
598 free((void *) consumerd64_bin
);
600 if (consumerd32_libdir_override
) {
601 free((void *) consumerd32_libdir
);
603 if (consumerd64_libdir_override
) {
604 free((void *) consumerd64_libdir
);
612 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
613 "Matthew, BEET driven development works!%c[%dm",
614 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
619 * Send data on a unix socket using the liblttsessiondcomm API.
621 * Return lttcomm error code.
623 static int send_unix_sock(int sock
, void *buf
, size_t len
)
625 /* Check valid length */
630 return lttcomm_send_unix_sock(sock
, buf
, len
);
634 * Free memory of a command context structure.
636 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
638 DBG("Clean command context structure");
640 if ((*cmd_ctx
)->llm
) {
641 free((*cmd_ctx
)->llm
);
643 if ((*cmd_ctx
)->lsm
) {
644 free((*cmd_ctx
)->lsm
);
652 * Notify UST applications using the shm mmap futex.
654 static int notify_ust_apps(int active
)
658 DBG("Notifying applications of session daemon state: %d", active
);
660 /* See shm.c for this call implying mmap, shm and futex calls */
661 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
662 if (wait_shm_mmap
== NULL
) {
666 /* Wake waiting process */
667 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
669 /* Apps notified successfully */
677 * Setup the outgoing data buffer for the response (llm) by allocating the
678 * right amount of memory and copying the original information from the lsm
681 * Return total size of the buffer pointed by buf.
683 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
689 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
690 if (cmd_ctx
->llm
== NULL
) {
696 /* Copy common data */
697 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
698 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
700 cmd_ctx
->llm
->data_size
= size
;
701 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
710 * Update the kernel poll set of all channel fd available over all tracing
711 * session. Add the wakeup pipe at the end of the set.
713 static int update_kernel_poll(struct lttng_poll_event
*events
)
716 struct ltt_session
*session
;
717 struct ltt_kernel_channel
*channel
;
719 DBG("Updating kernel poll set");
722 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
723 session_lock(session
);
724 if (session
->kernel_session
== NULL
) {
725 session_unlock(session
);
729 cds_list_for_each_entry(channel
,
730 &session
->kernel_session
->channel_list
.head
, list
) {
731 /* Add channel fd to the kernel poll set */
732 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
734 session_unlock(session
);
737 DBG("Channel fd %d added to kernel set", channel
->fd
);
739 session_unlock(session
);
741 session_unlock_list();
746 session_unlock_list();
751 * Find the channel fd from 'fd' over all tracing session. When found, check
752 * for new channel stream and send those stream fds to the kernel consumer.
754 * Useful for CPU hotplug feature.
756 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
759 struct ltt_session
*session
;
760 struct ltt_kernel_session
*ksess
;
761 struct ltt_kernel_channel
*channel
;
763 DBG("Updating kernel streams for channel fd %d", fd
);
766 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
767 session_lock(session
);
768 if (session
->kernel_session
== NULL
) {
769 session_unlock(session
);
772 ksess
= session
->kernel_session
;
774 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
775 if (channel
->fd
== fd
) {
776 DBG("Channel found, updating kernel streams");
777 ret
= kernel_open_channel_stream(channel
);
781 /* Update the stream global counter */
782 ksess
->stream_count_global
+= ret
;
785 * Have we already sent fds to the consumer? If yes, it means
786 * that tracing is started so it is safe to send our updated
789 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
790 struct lttng_ht_iter iter
;
791 struct consumer_socket
*socket
;
794 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
795 &iter
.iter
, socket
, node
.node
) {
796 pthread_mutex_lock(socket
->lock
);
797 ret
= kernel_consumer_send_channel_stream(socket
,
799 session
->output_traces
? 1 : 0);
800 pthread_mutex_unlock(socket
->lock
);
811 session_unlock(session
);
813 session_unlock_list();
817 session_unlock(session
);
818 session_unlock_list();
823 * For each tracing session, update newly registered apps. The session list
824 * lock MUST be acquired before calling this.
826 static void update_ust_app(int app_sock
)
828 struct ltt_session
*sess
, *stmp
;
830 /* Consumer is in an ERROR state. Stop any application update. */
831 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
832 /* Stop the update process since the consumer is dead. */
836 /* For all tracing session(s) */
837 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
839 if (sess
->ust_session
) {
840 ust_app_global_update(sess
->ust_session
, app_sock
);
842 session_unlock(sess
);
847 * This thread manage event coming from the kernel.
849 * Features supported in this thread:
852 static void *thread_manage_kernel(void *data
)
854 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
855 uint32_t revents
, nb_fd
;
857 struct lttng_poll_event events
;
859 DBG("[thread] Thread manage kernel started");
861 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
864 * This first step of the while is to clean this structure which could free
865 * non NULL pointers so initialize it before the loop.
867 lttng_poll_init(&events
);
869 if (testpoint(sessiond_thread_manage_kernel
)) {
870 goto error_testpoint
;
873 health_code_update();
875 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
876 goto error_testpoint
;
880 health_code_update();
882 if (update_poll_flag
== 1) {
883 /* Clean events object. We are about to populate it again. */
884 lttng_poll_clean(&events
);
886 ret
= sessiond_set_thread_pollset(&events
, 2);
888 goto error_poll_create
;
891 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
896 /* This will add the available kernel channel if any. */
897 ret
= update_kernel_poll(&events
);
901 update_poll_flag
= 0;
904 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
906 /* Poll infinite value of time */
909 ret
= lttng_poll_wait(&events
, -1);
913 * Restart interrupted system call.
915 if (errno
== EINTR
) {
919 } else if (ret
== 0) {
920 /* Should not happen since timeout is infinite */
921 ERR("Return value of poll is 0 with an infinite timeout.\n"
922 "This should not have happened! Continuing...");
928 for (i
= 0; i
< nb_fd
; i
++) {
929 /* Fetch once the poll data */
930 revents
= LTTNG_POLL_GETEV(&events
, i
);
931 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
933 health_code_update();
935 /* Thread quit pipe has been closed. Killing thread. */
936 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
942 /* Check for data on kernel pipe */
943 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
944 (void) lttng_read(kernel_poll_pipe
[0],
947 * Ret value is useless here, if this pipe gets any actions an
948 * update is required anyway.
950 update_poll_flag
= 1;
954 * New CPU detected by the kernel. Adding kernel stream to
955 * kernel session and updating the kernel consumer
957 if (revents
& LPOLLIN
) {
958 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
964 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
965 * and unregister kernel stream at this point.
974 lttng_poll_clean(&events
);
977 utils_close_pipe(kernel_poll_pipe
);
978 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
981 ERR("Health error occurred in %s", __func__
);
982 WARN("Kernel thread died unexpectedly. "
983 "Kernel tracing can continue but CPU hotplug is disabled.");
985 health_unregister(health_sessiond
);
986 DBG("Kernel thread dying");
991 * Signal pthread condition of the consumer data that the thread.
993 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
995 pthread_mutex_lock(&data
->cond_mutex
);
998 * The state is set before signaling. It can be any value, it's the waiter
999 * job to correctly interpret this condition variable associated to the
1000 * consumer pthread_cond.
1002 * A value of 0 means that the corresponding thread of the consumer data
1003 * was not started. 1 indicates that the thread has started and is ready
1004 * for action. A negative value means that there was an error during the
1007 data
->consumer_thread_is_ready
= state
;
1008 (void) pthread_cond_signal(&data
->cond
);
1010 pthread_mutex_unlock(&data
->cond_mutex
);
1014 * This thread manage the consumer error sent back to the session daemon.
1016 static void *thread_manage_consumer(void *data
)
1018 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1019 uint32_t revents
, nb_fd
;
1020 enum lttcomm_return_code code
;
1021 struct lttng_poll_event events
;
1022 struct consumer_data
*consumer_data
= data
;
1024 DBG("[thread] Manage consumer started");
1026 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1028 health_code_update();
1031 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1032 * metadata_sock. Nothing more will be added to this poll set.
1034 ret
= sessiond_set_thread_pollset(&events
, 3);
1040 * The error socket here is already in a listening state which was done
1041 * just before spawning this thread to avoid a race between the consumer
1042 * daemon exec trying to connect and the listen() call.
1044 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1049 health_code_update();
1051 /* Infinite blocking call, waiting for transmission */
1053 health_poll_entry();
1055 if (testpoint(sessiond_thread_manage_consumer
)) {
1059 ret
= lttng_poll_wait(&events
, -1);
1063 * Restart interrupted system call.
1065 if (errno
== EINTR
) {
1073 for (i
= 0; i
< nb_fd
; i
++) {
1074 /* Fetch once the poll data */
1075 revents
= LTTNG_POLL_GETEV(&events
, i
);
1076 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1078 health_code_update();
1080 /* Thread quit pipe has been closed. Killing thread. */
1081 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1087 /* Event on the registration socket */
1088 if (pollfd
== consumer_data
->err_sock
) {
1089 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1090 ERR("consumer err socket poll error");
1096 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1102 * Set the CLOEXEC flag. Return code is useless because either way, the
1105 (void) utils_set_fd_cloexec(sock
);
1107 health_code_update();
1109 DBG2("Receiving code from consumer err_sock");
1111 /* Getting status code from kconsumerd */
1112 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1113 sizeof(enum lttcomm_return_code
));
1118 health_code_update();
1120 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1121 /* Connect both socket, command and metadata. */
1122 consumer_data
->cmd_sock
=
1123 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1124 consumer_data
->metadata_fd
=
1125 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1126 if (consumer_data
->cmd_sock
< 0
1127 || consumer_data
->metadata_fd
< 0) {
1128 PERROR("consumer connect cmd socket");
1129 /* On error, signal condition and quit. */
1130 signal_consumer_condition(consumer_data
, -1);
1133 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1134 /* Create metadata socket lock. */
1135 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1136 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1137 PERROR("zmalloc pthread mutex");
1141 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1143 signal_consumer_condition(consumer_data
, 1);
1144 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1145 DBG("Consumer metadata socket ready (fd: %d)",
1146 consumer_data
->metadata_fd
);
1148 ERR("consumer error when waiting for SOCK_READY : %s",
1149 lttcomm_get_readable_code(-code
));
1153 /* Remove the consumerd error sock since we've established a connexion */
1154 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1159 /* Add new accepted error socket. */
1160 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1165 /* Add metadata socket that is successfully connected. */
1166 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1167 LPOLLIN
| LPOLLRDHUP
);
1172 health_code_update();
1174 /* Infinite blocking call, waiting for transmission */
1177 health_poll_entry();
1178 ret
= lttng_poll_wait(&events
, -1);
1182 * Restart interrupted system call.
1184 if (errno
== EINTR
) {
1192 for (i
= 0; i
< nb_fd
; i
++) {
1193 /* Fetch once the poll data */
1194 revents
= LTTNG_POLL_GETEV(&events
, i
);
1195 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1197 health_code_update();
1199 /* Thread quit pipe has been closed. Killing thread. */
1200 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1206 if (pollfd
== sock
) {
1207 /* Event on the consumerd socket */
1208 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1209 ERR("consumer err socket second poll error");
1212 health_code_update();
1213 /* Wait for any kconsumerd error */
1214 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1215 sizeof(enum lttcomm_return_code
));
1217 ERR("consumer closed the command socket");
1221 ERR("consumer return code : %s",
1222 lttcomm_get_readable_code(-code
));
1225 } else if (pollfd
== consumer_data
->metadata_fd
) {
1226 /* UST metadata requests */
1227 ret
= ust_consumer_metadata_request(
1228 &consumer_data
->metadata_sock
);
1230 ERR("Handling metadata request");
1235 ERR("Unknown pollfd");
1239 health_code_update();
1245 * We lock here because we are about to close the sockets and some other
1246 * thread might be using them so get exclusive access which will abort all
1247 * other consumer command by other threads.
1249 pthread_mutex_lock(&consumer_data
->lock
);
1251 /* Immediately set the consumerd state to stopped */
1252 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1253 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1254 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1255 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1256 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1258 /* Code flow error... */
1262 if (consumer_data
->err_sock
>= 0) {
1263 ret
= close(consumer_data
->err_sock
);
1267 consumer_data
->err_sock
= -1;
1269 if (consumer_data
->cmd_sock
>= 0) {
1270 ret
= close(consumer_data
->cmd_sock
);
1274 consumer_data
->cmd_sock
= -1;
1276 if (*consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1277 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1290 unlink(consumer_data
->err_unix_sock_path
);
1291 unlink(consumer_data
->cmd_unix_sock_path
);
1292 consumer_data
->pid
= 0;
1293 pthread_mutex_unlock(&consumer_data
->lock
);
1295 /* Cleanup metadata socket mutex. */
1296 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1297 free(consumer_data
->metadata_sock
.lock
);
1299 lttng_poll_clean(&events
);
1303 ERR("Health error occurred in %s", __func__
);
1305 health_unregister(health_sessiond
);
1306 DBG("consumer thread cleanup completed");
1312 * This thread manage application communication.
1314 static void *thread_manage_apps(void *data
)
1316 int i
, ret
, pollfd
, err
= -1;
1318 uint32_t revents
, nb_fd
;
1319 struct lttng_poll_event events
;
1321 DBG("[thread] Manage application started");
1323 rcu_register_thread();
1324 rcu_thread_online();
1326 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1328 if (testpoint(sessiond_thread_manage_apps
)) {
1329 goto error_testpoint
;
1332 health_code_update();
1334 ret
= sessiond_set_thread_pollset(&events
, 2);
1336 goto error_poll_create
;
1339 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1344 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1348 health_code_update();
1351 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1353 /* Inifinite blocking call, waiting for transmission */
1355 health_poll_entry();
1356 ret
= lttng_poll_wait(&events
, -1);
1360 * Restart interrupted system call.
1362 if (errno
== EINTR
) {
1370 for (i
= 0; i
< nb_fd
; i
++) {
1371 /* Fetch once the poll data */
1372 revents
= LTTNG_POLL_GETEV(&events
, i
);
1373 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1375 health_code_update();
1377 /* Thread quit pipe has been closed. Killing thread. */
1378 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1384 /* Inspect the apps cmd pipe */
1385 if (pollfd
== apps_cmd_pipe
[0]) {
1386 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1387 ERR("Apps command pipe error");
1389 } else if (revents
& LPOLLIN
) {
1393 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1394 if (size_ret
< sizeof(sock
)) {
1395 PERROR("read apps cmd pipe");
1399 health_code_update();
1402 * We only monitor the error events of the socket. This
1403 * thread does not handle any incoming data from UST
1406 ret
= lttng_poll_add(&events
, sock
,
1407 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1412 DBG("Apps with sock %d added to poll set", sock
);
1416 * At this point, we know that a registered application made
1417 * the event at poll_wait.
1419 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1420 /* Removing from the poll set */
1421 ret
= lttng_poll_del(&events
, pollfd
);
1426 /* Socket closed on remote end. */
1427 ust_app_unregister(pollfd
);
1431 health_code_update();
1437 lttng_poll_clean(&events
);
1440 utils_close_pipe(apps_cmd_pipe
);
1441 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1444 * We don't clean the UST app hash table here since already registered
1445 * applications can still be controlled so let them be until the session
1446 * daemon dies or the applications stop.
1451 ERR("Health error occurred in %s", __func__
);
1453 health_unregister(health_sessiond
);
1454 DBG("Application communication apps thread cleanup complete");
1455 rcu_thread_offline();
1456 rcu_unregister_thread();
1461 * Send a socket to a thread This is called from the dispatch UST registration
1462 * thread once all sockets are set for the application.
1464 * The sock value can be invalid, we don't really care, the thread will handle
1465 * it and make the necessary cleanup if so.
1467 * On success, return 0 else a negative value being the errno message of the
1470 static int send_socket_to_thread(int fd
, int sock
)
1475 * It's possible that the FD is set as invalid with -1 concurrently just
1476 * before calling this function being a shutdown state of the thread.
1483 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1484 if (ret
< sizeof(sock
)) {
1485 PERROR("write apps pipe %d", fd
);
1492 /* All good. Don't send back the write positive ret value. */
1499 * Sanitize the wait queue of the dispatch registration thread meaning removing
1500 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1501 * notify socket is never received.
1503 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1505 int ret
, nb_fd
= 0, i
;
1506 unsigned int fd_added
= 0;
1507 struct lttng_poll_event events
;
1508 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1512 lttng_poll_init(&events
);
1514 /* Just skip everything for an empty queue. */
1515 if (!wait_queue
->count
) {
1519 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1524 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1525 &wait_queue
->head
, head
) {
1526 assert(wait_node
->app
);
1527 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1528 LPOLLHUP
| LPOLLERR
);
1541 * Poll but don't block so we can quickly identify the faulty events and
1542 * clean them afterwards from the wait queue.
1544 ret
= lttng_poll_wait(&events
, 0);
1550 for (i
= 0; i
< nb_fd
; i
++) {
1551 /* Get faulty FD. */
1552 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1553 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1555 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1556 &wait_queue
->head
, head
) {
1557 if (pollfd
== wait_node
->app
->sock
&&
1558 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1559 cds_list_del(&wait_node
->head
);
1560 wait_queue
->count
--;
1561 ust_app_destroy(wait_node
->app
);
1569 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1573 lttng_poll_clean(&events
);
1577 lttng_poll_clean(&events
);
1579 ERR("Unable to sanitize wait queue");
1584 * Dispatch request from the registration threads to the application
1585 * communication thread.
1587 static void *thread_dispatch_ust_registration(void *data
)
1590 struct cds_wfq_node
*node
;
1591 struct ust_command
*ust_cmd
= NULL
;
1592 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1593 struct ust_reg_wait_queue wait_queue
= {
1597 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1599 health_code_update();
1601 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1603 DBG("[thread] Dispatch UST command started");
1605 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1606 health_code_update();
1608 /* Atomically prepare the queue futex */
1609 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1612 struct ust_app
*app
= NULL
;
1616 * Make sure we don't have node(s) that have hung up before receiving
1617 * the notify socket. This is to clean the list in order to avoid
1618 * memory leaks from notify socket that are never seen.
1620 sanitize_wait_queue(&wait_queue
);
1622 health_code_update();
1623 /* Dequeue command for registration */
1624 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1626 DBG("Woken up but nothing in the UST command queue");
1627 /* Continue thread execution */
1631 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1633 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1634 " gid:%d sock:%d name:%s (version %d.%d)",
1635 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1636 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1637 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1638 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1640 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1641 wait_node
= zmalloc(sizeof(*wait_node
));
1643 PERROR("zmalloc wait_node dispatch");
1644 ret
= close(ust_cmd
->sock
);
1646 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1648 lttng_fd_put(LTTNG_FD_APPS
, 1);
1652 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1654 /* Create application object if socket is CMD. */
1655 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1657 if (!wait_node
->app
) {
1658 ret
= close(ust_cmd
->sock
);
1660 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1662 lttng_fd_put(LTTNG_FD_APPS
, 1);
1668 * Add application to the wait queue so we can set the notify
1669 * socket before putting this object in the global ht.
1671 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1676 * We have to continue here since we don't have the notify
1677 * socket and the application MUST be added to the hash table
1678 * only at that moment.
1683 * Look for the application in the local wait queue and set the
1684 * notify socket if found.
1686 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1687 &wait_queue
.head
, head
) {
1688 health_code_update();
1689 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1690 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1691 cds_list_del(&wait_node
->head
);
1693 app
= wait_node
->app
;
1695 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1701 * With no application at this stage the received socket is
1702 * basically useless so close it before we free the cmd data
1703 * structure for good.
1706 ret
= close(ust_cmd
->sock
);
1708 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1710 lttng_fd_put(LTTNG_FD_APPS
, 1);
1717 * @session_lock_list
1719 * Lock the global session list so from the register up to the
1720 * registration done message, no thread can see the application
1721 * and change its state.
1723 session_lock_list();
1727 * Add application to the global hash table. This needs to be
1728 * done before the update to the UST registry can locate the
1733 /* Set app version. This call will print an error if needed. */
1734 (void) ust_app_version(app
);
1736 /* Send notify socket through the notify pipe. */
1737 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1741 session_unlock_list();
1743 * No notify thread, stop the UST tracing. However, this is
1744 * not an internal error of the this thread thus setting
1745 * the health error code to a normal exit.
1752 * Update newly registered application with the tracing
1753 * registry info already enabled information.
1755 update_ust_app(app
->sock
);
1758 * Don't care about return value. Let the manage apps threads
1759 * handle app unregistration upon socket close.
1761 (void) ust_app_register_done(app
->sock
);
1764 * Even if the application socket has been closed, send the app
1765 * to the thread and unregistration will take place at that
1768 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1771 session_unlock_list();
1773 * No apps. thread, stop the UST tracing. However, this is
1774 * not an internal error of the this thread thus setting
1775 * the health error code to a normal exit.
1782 session_unlock_list();
1784 } while (node
!= NULL
);
1786 health_poll_entry();
1787 /* Futex wait on queue. Blocking call on futex() */
1788 futex_nto1_wait(&ust_cmd_queue
.futex
);
1791 /* Normal exit, no error */
1795 /* Clean up wait queue. */
1796 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1797 &wait_queue
.head
, head
) {
1798 cds_list_del(&wait_node
->head
);
1803 DBG("Dispatch thread dying");
1806 ERR("Health error occurred in %s", __func__
);
1808 health_unregister(health_sessiond
);
1813 * This thread manage application registration.
1815 static void *thread_registration_apps(void *data
)
1817 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1818 uint32_t revents
, nb_fd
;
1819 struct lttng_poll_event events
;
1821 * Get allocated in this thread, enqueued to a global queue, dequeued and
1822 * freed in the manage apps thread.
1824 struct ust_command
*ust_cmd
= NULL
;
1826 DBG("[thread] Manage application registration started");
1828 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1830 if (testpoint(sessiond_thread_registration_apps
)) {
1831 goto error_testpoint
;
1834 ret
= lttcomm_listen_unix_sock(apps_sock
);
1840 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1841 * more will be added to this poll set.
1843 ret
= sessiond_set_thread_pollset(&events
, 2);
1845 goto error_create_poll
;
1848 /* Add the application registration socket */
1849 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1851 goto error_poll_add
;
1854 /* Notify all applications to register */
1855 ret
= notify_ust_apps(1);
1857 ERR("Failed to notify applications or create the wait shared memory.\n"
1858 "Execution continues but there might be problem for already\n"
1859 "running applications that wishes to register.");
1863 DBG("Accepting application registration");
1865 /* Inifinite blocking call, waiting for transmission */
1867 health_poll_entry();
1868 ret
= lttng_poll_wait(&events
, -1);
1872 * Restart interrupted system call.
1874 if (errno
== EINTR
) {
1882 for (i
= 0; i
< nb_fd
; i
++) {
1883 health_code_update();
1885 /* Fetch once the poll data */
1886 revents
= LTTNG_POLL_GETEV(&events
, i
);
1887 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1889 /* Thread quit pipe has been closed. Killing thread. */
1890 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1896 /* Event on the registration socket */
1897 if (pollfd
== apps_sock
) {
1898 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1899 ERR("Register apps socket poll error");
1901 } else if (revents
& LPOLLIN
) {
1902 sock
= lttcomm_accept_unix_sock(apps_sock
);
1908 * Set socket timeout for both receiving and ending.
1909 * app_socket_timeout is in seconds, whereas
1910 * lttcomm_setsockopt_rcv_timeout and
1911 * lttcomm_setsockopt_snd_timeout expect msec as
1914 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1915 app_socket_timeout
* 1000);
1916 (void) lttcomm_setsockopt_snd_timeout(sock
,
1917 app_socket_timeout
* 1000);
1920 * Set the CLOEXEC flag. Return code is useless because
1921 * either way, the show must go on.
1923 (void) utils_set_fd_cloexec(sock
);
1925 /* Create UST registration command for enqueuing */
1926 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1927 if (ust_cmd
== NULL
) {
1928 PERROR("ust command zmalloc");
1933 * Using message-based transmissions to ensure we don't
1934 * have to deal with partially received messages.
1936 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1938 ERR("Exhausted file descriptors allowed for applications.");
1948 health_code_update();
1949 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1952 /* Close socket of the application. */
1957 lttng_fd_put(LTTNG_FD_APPS
, 1);
1961 health_code_update();
1963 ust_cmd
->sock
= sock
;
1966 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1967 " gid:%d sock:%d name:%s (version %d.%d)",
1968 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1969 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1970 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1971 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1974 * Lock free enqueue the registration request. The red pill
1975 * has been taken! This apps will be part of the *system*.
1977 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1980 * Wake the registration queue futex. Implicit memory
1981 * barrier with the exchange in cds_wfq_enqueue.
1983 futex_nto1_wake(&ust_cmd_queue
.futex
);
1993 ERR("Health error occurred in %s", __func__
);
1996 /* Notify that the registration thread is gone */
1999 if (apps_sock
>= 0) {
2000 ret
= close(apps_sock
);
2010 lttng_fd_put(LTTNG_FD_APPS
, 1);
2012 unlink(apps_unix_sock_path
);
2015 lttng_poll_clean(&events
);
2019 DBG("UST Registration thread cleanup complete");
2020 health_unregister(health_sessiond
);
2026 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2027 * exec or it will fails.
2029 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2032 struct timespec timeout
;
2034 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2035 consumer_data
->consumer_thread_is_ready
= 0;
2037 /* Setup pthread condition */
2038 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2041 PERROR("pthread_condattr_init consumer data");
2046 * Set the monotonic clock in order to make sure we DO NOT jump in time
2047 * between the clock_gettime() call and the timedwait call. See bug #324
2048 * for a more details and how we noticed it.
2050 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2053 PERROR("pthread_condattr_setclock consumer data");
2057 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2060 PERROR("pthread_cond_init consumer data");
2064 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2067 PERROR("pthread_create consumer");
2072 /* We are about to wait on a pthread condition */
2073 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2075 /* Get time for sem_timedwait absolute timeout */
2076 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2078 * Set the timeout for the condition timed wait even if the clock gettime
2079 * call fails since we might loop on that call and we want to avoid to
2080 * increment the timeout too many times.
2082 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2085 * The following loop COULD be skipped in some conditions so this is why we
2086 * set ret to 0 in order to make sure at least one round of the loop is
2092 * Loop until the condition is reached or when a timeout is reached. Note
2093 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2094 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2095 * possible. This loop does not take any chances and works with both of
2098 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2099 if (clock_ret
< 0) {
2100 PERROR("clock_gettime spawn consumer");
2101 /* Infinite wait for the consumerd thread to be ready */
2102 ret
= pthread_cond_wait(&consumer_data
->cond
,
2103 &consumer_data
->cond_mutex
);
2105 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2106 &consumer_data
->cond_mutex
, &timeout
);
2110 /* Release the pthread condition */
2111 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2115 if (ret
== ETIMEDOUT
) {
2117 * Call has timed out so we kill the kconsumerd_thread and return
2120 ERR("Condition timed out. The consumer thread was never ready."
2122 ret
= pthread_cancel(consumer_data
->thread
);
2124 PERROR("pthread_cancel consumer thread");
2127 PERROR("pthread_cond_wait failed consumer thread");
2132 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2133 if (consumer_data
->pid
== 0) {
2134 ERR("Consumerd did not start");
2135 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2138 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2147 * Join consumer thread
2149 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2153 /* Consumer pid must be a real one. */
2154 if (consumer_data
->pid
> 0) {
2156 ret
= kill(consumer_data
->pid
, SIGTERM
);
2158 ERR("Error killing consumer daemon");
2161 return pthread_join(consumer_data
->thread
, &status
);
2168 * Fork and exec a consumer daemon (consumerd).
2170 * Return pid if successful else -1.
2172 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2176 const char *consumer_to_use
;
2177 const char *verbosity
;
2180 DBG("Spawning consumerd");
2187 if (opt_verbose_consumer
) {
2188 verbosity
= "--verbose";
2190 verbosity
= "--quiet";
2192 switch (consumer_data
->type
) {
2193 case LTTNG_CONSUMER_KERNEL
:
2195 * Find out which consumerd to execute. We will first try the
2196 * 64-bit path, then the sessiond's installation directory, and
2197 * fallback on the 32-bit one,
2199 DBG3("Looking for a kernel consumer at these locations:");
2200 DBG3(" 1) %s", consumerd64_bin
);
2201 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2202 DBG3(" 3) %s", consumerd32_bin
);
2203 if (stat(consumerd64_bin
, &st
) == 0) {
2204 DBG3("Found location #1");
2205 consumer_to_use
= consumerd64_bin
;
2206 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2207 DBG3("Found location #2");
2208 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2209 } else if (stat(consumerd32_bin
, &st
) == 0) {
2210 DBG3("Found location #3");
2211 consumer_to_use
= consumerd32_bin
;
2213 DBG("Could not find any valid consumerd executable");
2216 DBG("Using kernel consumer at: %s", consumer_to_use
);
2217 execl(consumer_to_use
,
2218 "lttng-consumerd", verbosity
, "-k",
2219 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2220 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2221 "--group", tracing_group_name
,
2224 case LTTNG_CONSUMER64_UST
:
2226 char *tmpnew
= NULL
;
2228 if (consumerd64_libdir
[0] != '\0') {
2232 tmp
= getenv("LD_LIBRARY_PATH");
2236 tmplen
= strlen("LD_LIBRARY_PATH=")
2237 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2238 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2243 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2244 strcat(tmpnew
, consumerd64_libdir
);
2245 if (tmp
[0] != '\0') {
2246 strcat(tmpnew
, ":");
2247 strcat(tmpnew
, tmp
);
2249 ret
= putenv(tmpnew
);
2256 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2257 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2258 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2259 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2260 "--group", tracing_group_name
,
2262 if (consumerd64_libdir
[0] != '\0') {
2270 case LTTNG_CONSUMER32_UST
:
2272 char *tmpnew
= NULL
;
2274 if (consumerd32_libdir
[0] != '\0') {
2278 tmp
= getenv("LD_LIBRARY_PATH");
2282 tmplen
= strlen("LD_LIBRARY_PATH=")
2283 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2284 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2289 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2290 strcat(tmpnew
, consumerd32_libdir
);
2291 if (tmp
[0] != '\0') {
2292 strcat(tmpnew
, ":");
2293 strcat(tmpnew
, tmp
);
2295 ret
= putenv(tmpnew
);
2302 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2303 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2304 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2305 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2306 "--group", tracing_group_name
,
2308 if (consumerd32_libdir
[0] != '\0') {
2317 PERROR("unknown consumer type");
2321 PERROR("kernel start consumer exec");
2324 } else if (pid
> 0) {
2327 PERROR("start consumer fork");
2335 * Spawn the consumerd daemon and session daemon thread.
2337 static int start_consumerd(struct consumer_data
*consumer_data
)
2342 * Set the listen() state on the socket since there is a possible race
2343 * between the exec() of the consumer daemon and this call if place in the
2344 * consumer thread. See bug #366 for more details.
2346 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2351 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2352 if (consumer_data
->pid
!= 0) {
2353 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2357 ret
= spawn_consumerd(consumer_data
);
2359 ERR("Spawning consumerd failed");
2360 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2364 /* Setting up the consumer_data pid */
2365 consumer_data
->pid
= ret
;
2366 DBG2("Consumer pid %d", consumer_data
->pid
);
2367 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2369 DBG2("Spawning consumer control thread");
2370 ret
= spawn_consumer_thread(consumer_data
);
2372 ERR("Fatal error spawning consumer control thread");
2380 /* Cleanup already created sockets on error. */
2381 if (consumer_data
->err_sock
>= 0) {
2384 err
= close(consumer_data
->err_sock
);
2386 PERROR("close consumer data error socket");
2393 * Setup necessary data for kernel tracer action.
2395 static int init_kernel_tracer(void)
2399 /* Modprobe lttng kernel modules */
2400 ret
= modprobe_lttng_control();
2405 /* Open debugfs lttng */
2406 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2407 if (kernel_tracer_fd
< 0) {
2408 DBG("Failed to open %s", module_proc_lttng
);
2413 /* Validate kernel version */
2414 ret
= kernel_validate_version(kernel_tracer_fd
);
2419 ret
= modprobe_lttng_data();
2424 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2428 modprobe_remove_lttng_control();
2429 ret
= close(kernel_tracer_fd
);
2433 kernel_tracer_fd
= -1;
2434 return LTTNG_ERR_KERN_VERSION
;
2437 ret
= close(kernel_tracer_fd
);
2443 modprobe_remove_lttng_control();
2446 WARN("No kernel tracer available");
2447 kernel_tracer_fd
= -1;
2449 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2451 return LTTNG_ERR_KERN_NA
;
2457 * Copy consumer output from the tracing session to the domain session. The
2458 * function also applies the right modification on a per domain basis for the
2459 * trace files destination directory.
2461 * Should *NOT* be called with RCU read-side lock held.
2463 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2466 const char *dir_name
;
2467 struct consumer_output
*consumer
;
2470 assert(session
->consumer
);
2473 case LTTNG_DOMAIN_KERNEL
:
2474 DBG3("Copying tracing session consumer output in kernel session");
2476 * XXX: We should audit the session creation and what this function
2477 * does "extra" in order to avoid a destroy since this function is used
2478 * in the domain session creation (kernel and ust) only. Same for UST
2481 if (session
->kernel_session
->consumer
) {
2482 consumer_destroy_output(session
->kernel_session
->consumer
);
2484 session
->kernel_session
->consumer
=
2485 consumer_copy_output(session
->consumer
);
2486 /* Ease our life a bit for the next part */
2487 consumer
= session
->kernel_session
->consumer
;
2488 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2490 case LTTNG_DOMAIN_JUL
:
2491 case LTTNG_DOMAIN_UST
:
2492 DBG3("Copying tracing session consumer output in UST session");
2493 if (session
->ust_session
->consumer
) {
2494 consumer_destroy_output(session
->ust_session
->consumer
);
2496 session
->ust_session
->consumer
=
2497 consumer_copy_output(session
->consumer
);
2498 /* Ease our life a bit for the next part */
2499 consumer
= session
->ust_session
->consumer
;
2500 dir_name
= DEFAULT_UST_TRACE_DIR
;
2503 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2507 /* Append correct directory to subdir */
2508 strncat(consumer
->subdir
, dir_name
,
2509 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2510 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2519 * Create an UST session and add it to the session ust list.
2521 * Should *NOT* be called with RCU read-side lock held.
2523 static int create_ust_session(struct ltt_session
*session
,
2524 struct lttng_domain
*domain
)
2527 struct ltt_ust_session
*lus
= NULL
;
2531 assert(session
->consumer
);
2533 switch (domain
->type
) {
2534 case LTTNG_DOMAIN_JUL
:
2535 case LTTNG_DOMAIN_UST
:
2538 ERR("Unknown UST domain on create session %d", domain
->type
);
2539 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2543 DBG("Creating UST session");
2545 lus
= trace_ust_create_session(session
->id
);
2547 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2551 lus
->uid
= session
->uid
;
2552 lus
->gid
= session
->gid
;
2553 lus
->output_traces
= session
->output_traces
;
2554 lus
->snapshot_mode
= session
->snapshot_mode
;
2555 lus
->live_timer_interval
= session
->live_timer
;
2556 session
->ust_session
= lus
;
2558 /* Copy session output to the newly created UST session */
2559 ret
= copy_session_consumer(domain
->type
, session
);
2560 if (ret
!= LTTNG_OK
) {
2568 session
->ust_session
= NULL
;
2573 * Create a kernel tracer session then create the default channel.
2575 static int create_kernel_session(struct ltt_session
*session
)
2579 DBG("Creating kernel session");
2581 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2583 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2587 /* Code flow safety */
2588 assert(session
->kernel_session
);
2590 /* Copy session output to the newly created Kernel session */
2591 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2592 if (ret
!= LTTNG_OK
) {
2596 /* Create directory(ies) on local filesystem. */
2597 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2598 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2599 ret
= run_as_mkdir_recursive(
2600 session
->kernel_session
->consumer
->dst
.trace_path
,
2601 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2603 if (ret
!= -EEXIST
) {
2604 ERR("Trace directory creation error");
2610 session
->kernel_session
->uid
= session
->uid
;
2611 session
->kernel_session
->gid
= session
->gid
;
2612 session
->kernel_session
->output_traces
= session
->output_traces
;
2613 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2618 trace_kernel_destroy_session(session
->kernel_session
);
2619 session
->kernel_session
= NULL
;
2624 * Count number of session permitted by uid/gid.
2626 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2629 struct ltt_session
*session
;
2631 DBG("Counting number of available session for UID %d GID %d",
2633 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2635 * Only list the sessions the user can control.
2637 if (!session_access_ok(session
, uid
, gid
)) {
2646 * Process the command requested by the lttng client within the command
2647 * context structure. This function make sure that the return structure (llm)
2648 * is set and ready for transmission before returning.
2650 * Return any error encountered or 0 for success.
2652 * "sock" is only used for special-case var. len data.
2654 * Should *NOT* be called with RCU read-side lock held.
2656 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2660 int need_tracing_session
= 1;
2663 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2667 switch (cmd_ctx
->lsm
->cmd_type
) {
2668 case LTTNG_CREATE_SESSION
:
2669 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2670 case LTTNG_CREATE_SESSION_LIVE
:
2671 case LTTNG_DESTROY_SESSION
:
2672 case LTTNG_LIST_SESSIONS
:
2673 case LTTNG_LIST_DOMAINS
:
2674 case LTTNG_START_TRACE
:
2675 case LTTNG_STOP_TRACE
:
2676 case LTTNG_DATA_PENDING
:
2677 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2678 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2679 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2680 case LTTNG_SNAPSHOT_RECORD
:
2687 if (opt_no_kernel
&& need_domain
2688 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2690 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2692 ret
= LTTNG_ERR_KERN_NA
;
2697 /* Deny register consumer if we already have a spawned consumer. */
2698 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2699 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2700 if (kconsumer_data
.pid
> 0) {
2701 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2702 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2705 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2709 * Check for command that don't needs to allocate a returned payload. We do
2710 * this here so we don't have to make the call for no payload at each
2713 switch(cmd_ctx
->lsm
->cmd_type
) {
2714 case LTTNG_LIST_SESSIONS
:
2715 case LTTNG_LIST_TRACEPOINTS
:
2716 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2717 case LTTNG_LIST_DOMAINS
:
2718 case LTTNG_LIST_CHANNELS
:
2719 case LTTNG_LIST_EVENTS
:
2722 /* Setup lttng message with no payload */
2723 ret
= setup_lttng_msg(cmd_ctx
, 0);
2725 /* This label does not try to unlock the session */
2726 goto init_setup_error
;
2730 /* Commands that DO NOT need a session. */
2731 switch (cmd_ctx
->lsm
->cmd_type
) {
2732 case LTTNG_CREATE_SESSION
:
2733 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2734 case LTTNG_CREATE_SESSION_LIVE
:
2735 case LTTNG_CALIBRATE
:
2736 case LTTNG_LIST_SESSIONS
:
2737 case LTTNG_LIST_TRACEPOINTS
:
2738 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2739 need_tracing_session
= 0;
2742 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2744 * We keep the session list lock across _all_ commands
2745 * for now, because the per-session lock does not
2746 * handle teardown properly.
2748 session_lock_list();
2749 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2750 if (cmd_ctx
->session
== NULL
) {
2751 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2754 /* Acquire lock for the session */
2755 session_lock(cmd_ctx
->session
);
2765 * Check domain type for specific "pre-action".
2767 switch (cmd_ctx
->lsm
->domain
.type
) {
2768 case LTTNG_DOMAIN_KERNEL
:
2770 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2774 /* Kernel tracer check */
2775 if (kernel_tracer_fd
== -1) {
2776 /* Basically, load kernel tracer modules */
2777 ret
= init_kernel_tracer();
2783 /* Consumer is in an ERROR state. Report back to client */
2784 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2785 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2789 /* Need a session for kernel command */
2790 if (need_tracing_session
) {
2791 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2792 ret
= create_kernel_session(cmd_ctx
->session
);
2794 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2799 /* Start the kernel consumer daemon */
2800 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2801 if (kconsumer_data
.pid
== 0 &&
2802 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2803 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2804 ret
= start_consumerd(&kconsumer_data
);
2806 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2809 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2811 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2815 * The consumer was just spawned so we need to add the socket to
2816 * the consumer output of the session if exist.
2818 ret
= consumer_create_socket(&kconsumer_data
,
2819 cmd_ctx
->session
->kernel_session
->consumer
);
2826 case LTTNG_DOMAIN_JUL
:
2827 case LTTNG_DOMAIN_UST
:
2829 if (!ust_app_supported()) {
2830 ret
= LTTNG_ERR_NO_UST
;
2833 /* Consumer is in an ERROR state. Report back to client */
2834 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2835 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2839 if (need_tracing_session
) {
2840 /* Create UST session if none exist. */
2841 if (cmd_ctx
->session
->ust_session
== NULL
) {
2842 ret
= create_ust_session(cmd_ctx
->session
,
2843 &cmd_ctx
->lsm
->domain
);
2844 if (ret
!= LTTNG_OK
) {
2849 /* Start the UST consumer daemons */
2851 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2852 if (consumerd64_bin
[0] != '\0' &&
2853 ustconsumer64_data
.pid
== 0 &&
2854 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2855 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2856 ret
= start_consumerd(&ustconsumer64_data
);
2858 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2859 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2863 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2864 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2866 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2870 * Setup socket for consumer 64 bit. No need for atomic access
2871 * since it was set above and can ONLY be set in this thread.
2873 ret
= consumer_create_socket(&ustconsumer64_data
,
2874 cmd_ctx
->session
->ust_session
->consumer
);
2880 if (consumerd32_bin
[0] != '\0' &&
2881 ustconsumer32_data
.pid
== 0 &&
2882 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2883 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2884 ret
= start_consumerd(&ustconsumer32_data
);
2886 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2887 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2891 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2892 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2894 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2898 * Setup socket for consumer 64 bit. No need for atomic access
2899 * since it was set above and can ONLY be set in this thread.
2901 ret
= consumer_create_socket(&ustconsumer32_data
,
2902 cmd_ctx
->session
->ust_session
->consumer
);
2914 /* Validate consumer daemon state when start/stop trace command */
2915 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2916 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2917 switch (cmd_ctx
->lsm
->domain
.type
) {
2918 case LTTNG_DOMAIN_JUL
:
2919 case LTTNG_DOMAIN_UST
:
2920 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2921 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2925 case LTTNG_DOMAIN_KERNEL
:
2926 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2927 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2935 * Check that the UID or GID match that of the tracing session.
2936 * The root user can interact with all sessions.
2938 if (need_tracing_session
) {
2939 if (!session_access_ok(cmd_ctx
->session
,
2940 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2941 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2942 ret
= LTTNG_ERR_EPERM
;
2948 * Send relayd information to consumer as soon as we have a domain and a
2951 if (cmd_ctx
->session
&& need_domain
) {
2953 * Setup relayd if not done yet. If the relayd information was already
2954 * sent to the consumer, this call will gracefully return.
2956 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2957 if (ret
!= LTTNG_OK
) {
2962 /* Process by command type */
2963 switch (cmd_ctx
->lsm
->cmd_type
) {
2964 case LTTNG_ADD_CONTEXT
:
2966 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2967 cmd_ctx
->lsm
->u
.context
.channel_name
,
2968 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2971 case LTTNG_DISABLE_CHANNEL
:
2973 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2974 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2977 case LTTNG_DISABLE_EVENT
:
2979 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2980 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2981 cmd_ctx
->lsm
->u
.disable
.name
);
2984 case LTTNG_DISABLE_ALL_EVENT
:
2986 DBG("Disabling all events");
2988 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2989 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2992 case LTTNG_ENABLE_CHANNEL
:
2994 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2995 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2998 case LTTNG_ENABLE_EVENT
:
3000 struct lttng_event_exclusion
*exclusion
= NULL
;
3001 struct lttng_filter_bytecode
*bytecode
= NULL
;
3003 /* Handle exclusion events and receive it from the client. */
3004 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3005 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3007 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3008 (count
* LTTNG_SYMBOL_NAME_LEN
));
3010 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3014 DBG("Receiving var len exclusion event list from client ...");
3015 exclusion
->count
= count
;
3016 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3017 count
* LTTNG_SYMBOL_NAME_LEN
);
3019 DBG("Nothing recv() from client var len data... continuing");
3022 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3027 /* Handle filter and get bytecode from client. */
3028 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3029 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3031 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3032 ret
= LTTNG_ERR_FILTER_INVAL
;
3037 bytecode
= zmalloc(bytecode_len
);
3040 ret
= LTTNG_ERR_FILTER_NOMEM
;
3044 /* Receive var. len. data */
3045 DBG("Receiving var len filter's bytecode from client ...");
3046 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3048 DBG("Nothing recv() from client car len data... continuing");
3052 ret
= LTTNG_ERR_FILTER_INVAL
;
3056 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3059 ret
= LTTNG_ERR_FILTER_INVAL
;
3064 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3065 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3066 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, exclusion
,
3067 kernel_poll_pipe
[1]);
3070 case LTTNG_ENABLE_ALL_EVENT
:
3072 DBG("Enabling all events");
3074 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3075 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3076 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
3079 case LTTNG_LIST_TRACEPOINTS
:
3081 struct lttng_event
*events
;
3084 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3085 if (nb_events
< 0) {
3086 /* Return value is a negative lttng_error_code. */
3092 * Setup lttng message with payload size set to the event list size in
3093 * bytes and then copy list into the llm payload.
3095 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3101 /* Copy event list into message payload */
3102 memcpy(cmd_ctx
->llm
->payload
, events
,
3103 sizeof(struct lttng_event
) * nb_events
);
3110 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3112 struct lttng_event_field
*fields
;
3115 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3117 if (nb_fields
< 0) {
3118 /* Return value is a negative lttng_error_code. */
3124 * Setup lttng message with payload size set to the event list size in
3125 * bytes and then copy list into the llm payload.
3127 ret
= setup_lttng_msg(cmd_ctx
,
3128 sizeof(struct lttng_event_field
) * nb_fields
);
3134 /* Copy event list into message payload */
3135 memcpy(cmd_ctx
->llm
->payload
, fields
,
3136 sizeof(struct lttng_event_field
) * nb_fields
);
3143 case LTTNG_SET_CONSUMER_URI
:
3146 struct lttng_uri
*uris
;
3148 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3149 len
= nb_uri
* sizeof(struct lttng_uri
);
3152 ret
= LTTNG_ERR_INVALID
;
3156 uris
= zmalloc(len
);
3158 ret
= LTTNG_ERR_FATAL
;
3162 /* Receive variable len data */
3163 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3164 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3166 DBG("No URIs received from client... continuing");
3168 ret
= LTTNG_ERR_SESSION_FAIL
;
3173 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3175 if (ret
!= LTTNG_OK
) {
3181 * XXX: 0 means that this URI should be applied on the session. Should
3182 * be a DOMAIN enuam.
3184 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3185 /* Add the URI for the UST session if a consumer is present. */
3186 if (cmd_ctx
->session
->ust_session
&&
3187 cmd_ctx
->session
->ust_session
->consumer
) {
3188 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3190 } else if (cmd_ctx
->session
->kernel_session
&&
3191 cmd_ctx
->session
->kernel_session
->consumer
) {
3192 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3193 cmd_ctx
->session
, nb_uri
, uris
);
3201 case LTTNG_START_TRACE
:
3203 ret
= cmd_start_trace(cmd_ctx
->session
);
3206 case LTTNG_STOP_TRACE
:
3208 ret
= cmd_stop_trace(cmd_ctx
->session
);
3211 case LTTNG_CREATE_SESSION
:
3214 struct lttng_uri
*uris
= NULL
;
3216 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3217 len
= nb_uri
* sizeof(struct lttng_uri
);
3220 uris
= zmalloc(len
);
3222 ret
= LTTNG_ERR_FATAL
;
3226 /* Receive variable len data */
3227 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3228 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3230 DBG("No URIs received from client... continuing");
3232 ret
= LTTNG_ERR_SESSION_FAIL
;
3237 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3238 DBG("Creating session with ONE network URI is a bad call");
3239 ret
= LTTNG_ERR_SESSION_FAIL
;
3245 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3246 &cmd_ctx
->creds
, 0);
3252 case LTTNG_DESTROY_SESSION
:
3254 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3256 /* Set session to NULL so we do not unlock it after free. */
3257 cmd_ctx
->session
= NULL
;
3260 case LTTNG_LIST_DOMAINS
:
3263 struct lttng_domain
*domains
;
3265 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3267 /* Return value is a negative lttng_error_code. */
3272 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3278 /* Copy event list into message payload */
3279 memcpy(cmd_ctx
->llm
->payload
, domains
,
3280 nb_dom
* sizeof(struct lttng_domain
));
3287 case LTTNG_LIST_CHANNELS
:
3290 struct lttng_channel
*channels
;
3292 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3293 cmd_ctx
->session
, &channels
);
3295 /* Return value is a negative lttng_error_code. */
3300 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3306 /* Copy event list into message payload */
3307 memcpy(cmd_ctx
->llm
->payload
, channels
,
3308 nb_chan
* sizeof(struct lttng_channel
));
3315 case LTTNG_LIST_EVENTS
:
3318 struct lttng_event
*events
= NULL
;
3320 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3321 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3323 /* Return value is a negative lttng_error_code. */
3328 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3334 /* Copy event list into message payload */
3335 memcpy(cmd_ctx
->llm
->payload
, events
,
3336 nb_event
* sizeof(struct lttng_event
));
3343 case LTTNG_LIST_SESSIONS
:
3345 unsigned int nr_sessions
;
3347 session_lock_list();
3348 nr_sessions
= lttng_sessions_count(
3349 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3350 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3352 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3354 session_unlock_list();
3358 /* Filled the session array */
3359 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3360 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3361 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3363 session_unlock_list();
3368 case LTTNG_CALIBRATE
:
3370 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3371 &cmd_ctx
->lsm
->u
.calibrate
);
3374 case LTTNG_REGISTER_CONSUMER
:
3376 struct consumer_data
*cdata
;
3378 switch (cmd_ctx
->lsm
->domain
.type
) {
3379 case LTTNG_DOMAIN_KERNEL
:
3380 cdata
= &kconsumer_data
;
3383 ret
= LTTNG_ERR_UND
;
3387 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3388 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3391 case LTTNG_DATA_PENDING
:
3393 ret
= cmd_data_pending(cmd_ctx
->session
);
3396 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3398 struct lttcomm_lttng_output_id reply
;
3400 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3401 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3402 if (ret
!= LTTNG_OK
) {
3406 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3411 /* Copy output list into message payload */
3412 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3416 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3418 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3419 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3422 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3425 struct lttng_snapshot_output
*outputs
= NULL
;
3427 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3428 if (nb_output
< 0) {
3433 ret
= setup_lttng_msg(cmd_ctx
,
3434 nb_output
* sizeof(struct lttng_snapshot_output
));
3441 /* Copy output list into message payload */
3442 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3443 nb_output
* sizeof(struct lttng_snapshot_output
));
3450 case LTTNG_SNAPSHOT_RECORD
:
3452 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3453 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3454 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3457 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3460 struct lttng_uri
*uris
= NULL
;
3462 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3463 len
= nb_uri
* sizeof(struct lttng_uri
);
3466 uris
= zmalloc(len
);
3468 ret
= LTTNG_ERR_FATAL
;
3472 /* Receive variable len data */
3473 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3474 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3476 DBG("No URIs received from client... continuing");
3478 ret
= LTTNG_ERR_SESSION_FAIL
;
3483 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3484 DBG("Creating session with ONE network URI is a bad call");
3485 ret
= LTTNG_ERR_SESSION_FAIL
;
3491 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3492 nb_uri
, &cmd_ctx
->creds
);
3496 case LTTNG_CREATE_SESSION_LIVE
:
3499 struct lttng_uri
*uris
= NULL
;
3501 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3502 len
= nb_uri
* sizeof(struct lttng_uri
);
3505 uris
= zmalloc(len
);
3507 ret
= LTTNG_ERR_FATAL
;
3511 /* Receive variable len data */
3512 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3513 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3515 DBG("No URIs received from client... continuing");
3517 ret
= LTTNG_ERR_SESSION_FAIL
;
3522 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3523 DBG("Creating session with ONE network URI is a bad call");
3524 ret
= LTTNG_ERR_SESSION_FAIL
;
3530 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3531 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3536 ret
= LTTNG_ERR_UND
;
3541 if (cmd_ctx
->llm
== NULL
) {
3542 DBG("Missing llm structure. Allocating one.");
3543 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3547 /* Set return code */
3548 cmd_ctx
->llm
->ret_code
= ret
;
3550 if (cmd_ctx
->session
) {
3551 session_unlock(cmd_ctx
->session
);
3553 if (need_tracing_session
) {
3554 session_unlock_list();
3561 * Thread managing health check socket.
3563 static void *thread_manage_health(void *data
)
3565 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3566 uint32_t revents
, nb_fd
;
3567 struct lttng_poll_event events
;
3568 struct health_comm_msg msg
;
3569 struct health_comm_reply reply
;
3571 DBG("[thread] Manage health check started");
3573 rcu_register_thread();
3575 /* We might hit an error path before this is created. */
3576 lttng_poll_init(&events
);
3578 /* Create unix socket */
3579 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3581 ERR("Unable to create health check Unix socket");
3587 /* lttng health client socket path permissions */
3588 ret
= chown(health_unix_sock_path
, 0,
3589 utils_get_group_id(tracing_group_name
));
3591 ERR("Unable to set group on %s", health_unix_sock_path
);
3597 ret
= chmod(health_unix_sock_path
,
3598 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3600 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3608 * Set the CLOEXEC flag. Return code is useless because either way, the
3611 (void) utils_set_fd_cloexec(sock
);
3613 ret
= lttcomm_listen_unix_sock(sock
);
3619 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3620 * more will be added to this poll set.
3622 ret
= sessiond_set_thread_pollset(&events
, 2);
3627 /* Add the application registration socket */
3628 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3634 DBG("Health check ready");
3636 /* Inifinite blocking call, waiting for transmission */
3638 ret
= lttng_poll_wait(&events
, -1);
3641 * Restart interrupted system call.
3643 if (errno
== EINTR
) {
3651 for (i
= 0; i
< nb_fd
; i
++) {
3652 /* Fetch once the poll data */
3653 revents
= LTTNG_POLL_GETEV(&events
, i
);
3654 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3656 /* Thread quit pipe has been closed. Killing thread. */
3657 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3663 /* Event on the registration socket */
3664 if (pollfd
== sock
) {
3665 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3666 ERR("Health socket poll error");
3672 new_sock
= lttcomm_accept_unix_sock(sock
);
3678 * Set the CLOEXEC flag. Return code is useless because either way, the
3681 (void) utils_set_fd_cloexec(new_sock
);
3683 DBG("Receiving data from client for health...");
3684 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3686 DBG("Nothing recv() from client... continuing");
3687 ret
= close(new_sock
);
3695 rcu_thread_online();
3698 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3700 * health_check_state returns 0 if health is
3703 if (!health_check_state(health_sessiond
, i
)) {
3704 reply
.ret_code
|= 1ULL << i
;
3708 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3710 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3712 ERR("Failed to send health data back to client");
3715 /* End of transmission */
3716 ret
= close(new_sock
);
3726 ERR("Health error occurred in %s", __func__
);
3728 DBG("Health check thread dying");
3729 unlink(health_unix_sock_path
);
3737 lttng_poll_clean(&events
);
3739 rcu_unregister_thread();
3744 * This thread manage all clients request using the unix client socket for
3747 static void *thread_manage_clients(void *data
)
3749 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3751 uint32_t revents
, nb_fd
;
3752 struct command_ctx
*cmd_ctx
= NULL
;
3753 struct lttng_poll_event events
;
3755 DBG("[thread] Manage client started");
3757 rcu_register_thread();
3759 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3761 health_code_update();
3763 ret
= lttcomm_listen_unix_sock(client_sock
);
3769 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3770 * more will be added to this poll set.
3772 ret
= sessiond_set_thread_pollset(&events
, 2);
3774 goto error_create_poll
;
3777 /* Add the application registration socket */
3778 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3784 * Notify parent pid that we are ready to accept command for client side.
3785 * This ppid is the one from the external process that spawned us.
3787 if (opt_sig_parent
) {
3788 kill(ppid
, SIGUSR1
);
3791 /* Notify the parent of the fork() process that we are ready. */
3793 kill(child_ppid
, SIGUSR1
);
3796 /* This testpoint is after we signal readiness to the parent. */
3797 if (testpoint(sessiond_thread_manage_clients
)) {
3801 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
3805 health_code_update();
3808 DBG("Accepting client command ...");
3810 /* Inifinite blocking call, waiting for transmission */
3812 health_poll_entry();
3813 ret
= lttng_poll_wait(&events
, -1);
3817 * Restart interrupted system call.
3819 if (errno
== EINTR
) {
3827 for (i
= 0; i
< nb_fd
; i
++) {
3828 /* Fetch once the poll data */
3829 revents
= LTTNG_POLL_GETEV(&events
, i
);
3830 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3832 health_code_update();
3834 /* Thread quit pipe has been closed. Killing thread. */
3835 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3841 /* Event on the registration socket */
3842 if (pollfd
== client_sock
) {
3843 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3844 ERR("Client socket poll error");
3850 DBG("Wait for client response");
3852 health_code_update();
3854 sock
= lttcomm_accept_unix_sock(client_sock
);
3860 * Set the CLOEXEC flag. Return code is useless because either way, the
3863 (void) utils_set_fd_cloexec(sock
);
3865 /* Set socket option for credentials retrieval */
3866 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3871 /* Allocate context command to process the client request */
3872 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3873 if (cmd_ctx
== NULL
) {
3874 PERROR("zmalloc cmd_ctx");
3878 /* Allocate data buffer for reception */
3879 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3880 if (cmd_ctx
->lsm
== NULL
) {
3881 PERROR("zmalloc cmd_ctx->lsm");
3885 cmd_ctx
->llm
= NULL
;
3886 cmd_ctx
->session
= NULL
;
3888 health_code_update();
3891 * Data is received from the lttng client. The struct
3892 * lttcomm_session_msg (lsm) contains the command and data request of
3895 DBG("Receiving data from client ...");
3896 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3897 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3899 DBG("Nothing recv() from client... continuing");
3905 clean_command_ctx(&cmd_ctx
);
3909 health_code_update();
3911 // TODO: Validate cmd_ctx including sanity check for
3912 // security purpose.
3914 rcu_thread_online();
3916 * This function dispatch the work to the kernel or userspace tracer
3917 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3918 * informations for the client. The command context struct contains
3919 * everything this function may needs.
3921 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3922 rcu_thread_offline();
3930 * TODO: Inform client somehow of the fatal error. At
3931 * this point, ret < 0 means that a zmalloc failed
3932 * (ENOMEM). Error detected but still accept
3933 * command, unless a socket error has been
3936 clean_command_ctx(&cmd_ctx
);
3940 health_code_update();
3942 DBG("Sending response (size: %d, retcode: %s)",
3943 cmd_ctx
->lttng_msg_size
,
3944 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3945 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3947 ERR("Failed to send data back to client");
3950 /* End of transmission */
3957 clean_command_ctx(&cmd_ctx
);
3959 health_code_update();
3971 lttng_poll_clean(&events
);
3972 clean_command_ctx(&cmd_ctx
);
3976 unlink(client_unix_sock_path
);
3977 if (client_sock
>= 0) {
3978 ret
= close(client_sock
);
3986 ERR("Health error occurred in %s", __func__
);
3989 health_unregister(health_sessiond
);
3991 DBG("Client thread dying");
3993 rcu_unregister_thread();
3999 * usage function on stderr
4001 static void usage(void)
4003 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4004 fprintf(stderr
, " -h, --help Display this usage.\n");
4005 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4006 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4007 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4008 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4009 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4010 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4011 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4012 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4013 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4014 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4015 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4016 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4017 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4018 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4019 fprintf(stderr
, " -V, --version Show version number.\n");
4020 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4021 fprintf(stderr
, " -q, --quiet No output at all.\n");
4022 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4023 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4024 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4025 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4026 fprintf(stderr
, " --jul-tcp-port JUL application registration TCP port\n");
4027 fprintf(stderr
, " -f --config Load daemon configuration file\n");
4031 * Take an option from the getopt output and set it in the right variable to be
4034 * Return 0 on success else a negative value.
4036 static int set_option(int opt
, const char *arg
, const char *optname
)
4042 fprintf(stderr
, "option %s", optname
);
4044 fprintf(stderr
, " with arg %s\n", arg
);
4048 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4051 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4057 tracing_group_name
= strdup(arg
);
4063 fprintf(stdout
, "%s\n", VERSION
);
4069 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4072 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4075 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4078 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4081 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4084 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4090 lttng_opt_quiet
= 1;
4093 /* Verbose level can increase using multiple -v */
4095 lttng_opt_verbose
= config_parse_value(arg
);
4097 lttng_opt_verbose
+= 1;
4102 opt_verbose_consumer
= config_parse_value(arg
);
4104 opt_verbose_consumer
+= 1;
4108 consumerd32_bin
= strdup(arg
);
4109 consumerd32_bin_override
= 1;
4112 consumerd32_libdir
= strdup(arg
);
4113 consumerd32_libdir_override
= 1;
4116 consumerd64_bin
= strdup(arg
);
4117 consumerd64_bin_override
= 1;
4120 consumerd64_libdir
= strdup(arg
);
4121 consumerd64_libdir_override
= 1;
4124 opt_pidfile
= strdup(arg
);
4126 case 'J': /* JUL TCP port. */
4131 v
= strtoul(arg
, NULL
, 0);
4132 if (errno
!= 0 || !isdigit(arg
[0])) {
4133 ERR("Wrong value in --jul-tcp-port parameter: %s", arg
);
4136 if (v
== 0 || v
>= 65535) {
4137 ERR("Port overflow in --jul-tcp-port parameter: %s", arg
);
4140 jul_tcp_port
= (uint32_t) v
;
4141 DBG3("JUL TCP port set to non default: %u", jul_tcp_port
);
4145 /* Unknown option or other error.
4146 * Error is printed by getopt, just return */
4154 * config_entry_handler_cb used to handle options read from a config file.
4155 * See config_entry_handler_cb comment in common/config/config.h for the
4156 * return value conventions.
4158 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4162 if (!entry
|| !entry
->name
|| !entry
->value
) {
4167 /* Check if the option is to be ignored */
4168 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4169 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4174 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4177 /* Ignore if not fully matched. */
4178 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4183 * If the option takes no argument on the command line, we have to
4184 * check if the value is "true". We support non-zero numeric values,
4187 if (!long_options
[i
].has_arg
) {
4188 ret
= config_parse_value(entry
->value
);
4191 WARN("Invalid configuration value \"%s\" for option %s",
4192 entry
->value
, entry
->name
);
4194 /* False, skip boolean config option. */
4199 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4203 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4210 * daemon configuration loading and argument parsing
4212 static int set_options(int argc
, char **argv
)
4214 int ret
= 0, c
= 0, option_index
= 0;
4215 int orig_optopt
= optopt
, orig_optind
= optind
;
4217 const char *config_path
= NULL
;
4219 optstring
= utils_generate_optstring(long_options
,
4220 sizeof(long_options
) / sizeof(struct option
));
4226 /* Check for the --config option */
4227 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4228 &option_index
)) != -1) {
4232 } else if (c
!= 'f') {
4233 /* if not equal to --config option. */
4237 config_path
= utils_expand_path(optarg
);
4239 ERR("Failed to resolve path: %s", optarg
);
4243 ret
= config_get_section_entries(config_path
, config_section_name
,
4244 config_entry_handler
, NULL
);
4247 ERR("Invalid configuration option at line %i", ret
);
4253 /* Reset getopt's global state */
4254 optopt
= orig_optopt
;
4255 optind
= orig_optind
;
4257 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
4262 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
4274 * Creates the two needed socket by the daemon.
4275 * apps_sock - The communication socket for all UST apps.
4276 * client_sock - The communication of the cli tool (lttng).
4278 static int init_daemon_socket(void)
4283 old_umask
= umask(0);
4285 /* Create client tool unix socket */
4286 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4287 if (client_sock
< 0) {
4288 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4293 /* Set the cloexec flag */
4294 ret
= utils_set_fd_cloexec(client_sock
);
4296 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4297 "Continuing but note that the consumer daemon will have a "
4298 "reference to this socket on exec()", client_sock
);
4301 /* File permission MUST be 660 */
4302 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4304 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4309 /* Create the application unix socket */
4310 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4311 if (apps_sock
< 0) {
4312 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4317 /* Set the cloexec flag */
4318 ret
= utils_set_fd_cloexec(apps_sock
);
4320 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4321 "Continuing but note that the consumer daemon will have a "
4322 "reference to this socket on exec()", apps_sock
);
4325 /* File permission MUST be 666 */
4326 ret
= chmod(apps_unix_sock_path
,
4327 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4329 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4334 DBG3("Session daemon client socket %d and application socket %d created",
4335 client_sock
, apps_sock
);
4343 * Check if the global socket is available, and if a daemon is answering at the
4344 * other side. If yes, error is returned.
4346 static int check_existing_daemon(void)
4348 /* Is there anybody out there ? */
4349 if (lttng_session_daemon_alive()) {
4357 * Set the tracing group gid onto the client socket.
4359 * Race window between mkdir and chown is OK because we are going from more
4360 * permissive (root.root) to less permissive (root.tracing).
4362 static int set_permissions(char *rundir
)
4367 gid
= utils_get_group_id(tracing_group_name
);
4369 /* Set lttng run dir */
4370 ret
= chown(rundir
, 0, gid
);
4372 ERR("Unable to set group on %s", rundir
);
4377 * Ensure all applications and tracing group can search the run
4378 * dir. Allow everyone to read the directory, since it does not
4379 * buy us anything to hide its content.
4381 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4383 ERR("Unable to set permissions on %s", rundir
);
4387 /* lttng client socket path */
4388 ret
= chown(client_unix_sock_path
, 0, gid
);
4390 ERR("Unable to set group on %s", client_unix_sock_path
);
4394 /* kconsumer error socket path */
4395 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4397 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4401 /* 64-bit ustconsumer error socket path */
4402 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4404 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4408 /* 32-bit ustconsumer compat32 error socket path */
4409 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4411 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4415 DBG("All permissions are set");
4421 * Create the lttng run directory needed for all global sockets and pipe.
4423 static int create_lttng_rundir(const char *rundir
)
4427 DBG3("Creating LTTng run directory: %s", rundir
);
4429 ret
= mkdir(rundir
, S_IRWXU
);
4431 if (errno
!= EEXIST
) {
4432 ERR("Unable to create %s", rundir
);
4444 * Setup sockets and directory needed by the kconsumerd communication with the
4447 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4451 char path
[PATH_MAX
];
4453 switch (consumer_data
->type
) {
4454 case LTTNG_CONSUMER_KERNEL
:
4455 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4457 case LTTNG_CONSUMER64_UST
:
4458 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4460 case LTTNG_CONSUMER32_UST
:
4461 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4464 ERR("Consumer type unknown");
4469 DBG2("Creating consumer directory: %s", path
);
4471 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4473 if (errno
!= EEXIST
) {
4475 ERR("Failed to create %s", path
);
4481 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4483 ERR("Unable to set group on %s", path
);
4489 /* Create the kconsumerd error unix socket */
4490 consumer_data
->err_sock
=
4491 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4492 if (consumer_data
->err_sock
< 0) {
4493 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4499 * Set the CLOEXEC flag. Return code is useless because either way, the
4502 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4504 PERROR("utils_set_fd_cloexec");
4505 /* continue anyway */
4508 /* File permission MUST be 660 */
4509 ret
= chmod(consumer_data
->err_unix_sock_path
,
4510 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4512 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4522 * Signal handler for the daemon
4524 * Simply stop all worker threads, leaving main() return gracefully after
4525 * joining all threads and calling cleanup().
4527 static void sighandler(int sig
)
4531 DBG("SIGPIPE caught");
4534 DBG("SIGINT caught");
4538 DBG("SIGTERM caught");
4542 CMM_STORE_SHARED(recv_child_signal
, 1);
4550 * Setup signal handler for :
4551 * SIGINT, SIGTERM, SIGPIPE
4553 static int set_signal_handler(void)
4556 struct sigaction sa
;
4559 if ((ret
= sigemptyset(&sigset
)) < 0) {
4560 PERROR("sigemptyset");
4564 sa
.sa_handler
= sighandler
;
4565 sa
.sa_mask
= sigset
;
4567 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4568 PERROR("sigaction");
4572 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4573 PERROR("sigaction");
4577 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4578 PERROR("sigaction");
4582 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4583 PERROR("sigaction");
4587 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4593 * Set open files limit to unlimited. This daemon can open a large number of
4594 * file descriptors in order to consumer multiple kernel traces.
4596 static void set_ulimit(void)
4601 /* The kernel does not allowed an infinite limit for open files */
4602 lim
.rlim_cur
= 65535;
4603 lim
.rlim_max
= 65535;
4605 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4607 PERROR("failed to set open files limit");
4612 * Write pidfile using the rundir and opt_pidfile.
4614 static void write_pidfile(void)
4617 char pidfile_path
[PATH_MAX
];
4622 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4624 /* Build pidfile path from rundir and opt_pidfile. */
4625 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4626 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4628 PERROR("snprintf pidfile path");
4634 * Create pid file in rundir. Return value is of no importance. The
4635 * execution will continue even though we are not able to write the file.
4637 (void) utils_create_pid_file(getpid(), pidfile_path
);
4644 * Write JUL TCP port using the rundir.
4646 static void write_julport(void)
4649 char path
[PATH_MAX
];
4653 ret
= snprintf(path
, sizeof(path
), "%s/"
4654 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
, rundir
);
4656 PERROR("snprintf julport path");
4661 * Create TCP JUL port file in rundir. Return value is of no importance.
4662 * The execution will continue even though we are not able to write the
4665 (void) utils_create_pid_file(jul_tcp_port
, path
);
4672 * Daemonize this process by forking and making the parent wait for the child
4673 * to signal it indicating readiness. Once received, the parent successfully
4676 * The child process undergoes the same action that daemon(3) does meaning
4677 * setsid, chdir, and dup /dev/null into 0, 1 and 2.
4679 * Return 0 on success else -1 on error.
4681 static int daemonize(void)
4686 /* Get parent pid of this process. */
4687 child_ppid
= getppid();
4693 } else if (pid
== 0) {
4700 * Get the newly created parent pid so we can signal that process when
4701 * we are ready to operate.
4703 child_ppid
= getppid();
4711 /* Try to change directory to /. If we can't well at least notify. */
4717 fd
= open(_PATH_DEVNULL
, O_RDWR
, 0);
4719 PERROR("open %s", _PATH_DEVNULL
);
4720 /* Let 0, 1 and 2 open since we can't bind them to /dev/null. */
4722 (void) dup2(fd
, STDIN_FILENO
);
4723 (void) dup2(fd
, STDOUT_FILENO
);
4724 (void) dup2(fd
, STDERR_FILENO
);
4737 * Waiting for child to notify this parent that it can exit. Note that
4738 * sleep() is interrupted before the 1 second delay as soon as the
4739 * signal is received, so it will not cause visible delay for the
4742 while (!CMM_LOAD_SHARED(recv_child_signal
)) {
4747 * Check if child exists without blocking. If so, we have to stop
4748 * this parent process and return an error.
4750 ret
= waitpid(pid
, &status
, WNOHANG
);
4751 if (ret
< 0 || (ret
!= 0 && WIFEXITED(status
))) {
4752 /* The child exited somehow or was not valid. */
4759 * From this point on, the parent can exit and the child is now an
4760 * operationnal session daemon ready to serve clients and applications.
4775 int main(int argc
, char **argv
)
4779 const char *home_path
, *env_app_timeout
;
4781 init_kernel_workarounds();
4783 rcu_register_thread();
4785 if ((ret
= set_signal_handler()) < 0) {
4789 setup_consumerd_path();
4791 page_size
= sysconf(_SC_PAGESIZE
);
4792 if (page_size
< 0) {
4793 PERROR("sysconf _SC_PAGESIZE");
4794 page_size
= LONG_MAX
;
4795 WARN("Fallback page size to %ld", page_size
);
4798 /* Parse arguments and load the daemon configuration file */
4800 if ((ret
= set_options(argc
, argv
)) < 0) {
4814 * We are in the child. Make sure all other file descriptors are
4815 * closed, in case we are called with more opened file descriptors than
4816 * the standard ones.
4818 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4823 /* Create thread quit pipe */
4824 if ((ret
= init_thread_quit_pipe()) < 0) {
4828 /* Check if daemon is UID = 0 */
4829 is_root
= !getuid();
4832 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4834 /* Create global run dir with root access */
4835 ret
= create_lttng_rundir(rundir
);
4840 if (strlen(apps_unix_sock_path
) == 0) {
4841 snprintf(apps_unix_sock_path
, PATH_MAX
,
4842 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4845 if (strlen(client_unix_sock_path
) == 0) {
4846 snprintf(client_unix_sock_path
, PATH_MAX
,
4847 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4850 /* Set global SHM for ust */
4851 if (strlen(wait_shm_path
) == 0) {
4852 snprintf(wait_shm_path
, PATH_MAX
,
4853 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4856 if (strlen(health_unix_sock_path
) == 0) {
4857 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4858 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4861 /* Setup kernel consumerd path */
4862 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4863 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4864 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4865 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4867 DBG2("Kernel consumer err path: %s",
4868 kconsumer_data
.err_unix_sock_path
);
4869 DBG2("Kernel consumer cmd path: %s",
4870 kconsumer_data
.cmd_unix_sock_path
);
4872 home_path
= utils_get_home_dir();
4873 if (home_path
== NULL
) {
4874 /* TODO: Add --socket PATH option */
4875 ERR("Can't get HOME directory for sockets creation.");
4881 * Create rundir from home path. This will create something like
4884 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4890 ret
= create_lttng_rundir(rundir
);
4895 if (strlen(apps_unix_sock_path
) == 0) {
4896 snprintf(apps_unix_sock_path
, PATH_MAX
,
4897 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4900 /* Set the cli tool unix socket path */
4901 if (strlen(client_unix_sock_path
) == 0) {
4902 snprintf(client_unix_sock_path
, PATH_MAX
,
4903 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4906 /* Set global SHM for ust */
4907 if (strlen(wait_shm_path
) == 0) {
4908 snprintf(wait_shm_path
, PATH_MAX
,
4909 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4912 /* Set health check Unix path */
4913 if (strlen(health_unix_sock_path
) == 0) {
4914 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4915 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4919 /* Set consumer initial state */
4920 kernel_consumerd_state
= CONSUMER_STOPPED
;
4921 ust_consumerd_state
= CONSUMER_STOPPED
;
4923 DBG("Client socket path %s", client_unix_sock_path
);
4924 DBG("Application socket path %s", apps_unix_sock_path
);
4925 DBG("Application wait path %s", wait_shm_path
);
4926 DBG("LTTng run directory path: %s", rundir
);
4928 /* 32 bits consumerd path setup */
4929 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4930 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4931 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4932 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4934 DBG2("UST consumer 32 bits err path: %s",
4935 ustconsumer32_data
.err_unix_sock_path
);
4936 DBG2("UST consumer 32 bits cmd path: %s",
4937 ustconsumer32_data
.cmd_unix_sock_path
);
4939 /* 64 bits consumerd path setup */
4940 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4941 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4942 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4943 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4945 DBG2("UST consumer 64 bits err path: %s",
4946 ustconsumer64_data
.err_unix_sock_path
);
4947 DBG2("UST consumer 64 bits cmd path: %s",
4948 ustconsumer64_data
.cmd_unix_sock_path
);
4951 * See if daemon already exist.
4953 if ((ret
= check_existing_daemon()) < 0) {
4954 ERR("Already running daemon.\n");
4956 * We do not goto exit because we must not cleanup()
4957 * because a daemon is already running.
4963 * Init UST app hash table. Alloc hash table before this point since
4964 * cleanup() can get called after that point.
4968 /* Initialize JUL domain subsystem. */
4969 if ((ret
= jul_init()) < 0) {
4970 /* ENOMEM at this point. */
4974 /* After this point, we can safely call cleanup() with "goto exit" */
4977 * These actions must be executed as root. We do that *after* setting up
4978 * the sockets path because we MUST make the check for another daemon using
4979 * those paths *before* trying to set the kernel consumer sockets and init
4983 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4988 /* Setup kernel tracer */
4989 if (!opt_no_kernel
) {
4990 init_kernel_tracer();
4993 /* Set ulimit for open files */
4996 /* init lttng_fd tracking must be done after set_ulimit. */
4999 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
5004 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
5009 /* Setup the needed unix socket */
5010 if ((ret
= init_daemon_socket()) < 0) {
5014 /* Set credentials to socket */
5015 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
5019 /* Get parent pid if -S, --sig-parent is specified. */
5020 if (opt_sig_parent
) {
5024 /* Setup the kernel pipe for waking up the kernel thread */
5025 if (is_root
&& !opt_no_kernel
) {
5026 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
5031 /* Setup the thread ht_cleanup communication pipe. */
5032 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
5036 /* Setup the thread apps communication pipe. */
5037 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
5041 /* Setup the thread apps notify communication pipe. */
5042 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
5046 /* Initialize global buffer per UID and PID registry. */
5047 buffer_reg_init_uid_registry();
5048 buffer_reg_init_pid_registry();
5050 /* Init UST command queue. */
5051 cds_wfq_init(&ust_cmd_queue
.queue
);
5054 * Get session list pointer. This pointer MUST NOT be free(). This list is
5055 * statically declared in session.c
5057 session_list_ptr
= session_get_list();
5059 /* Set up max poll set size */
5060 lttng_poll_set_max_size();
5064 /* Check for the application socket timeout env variable. */
5065 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5066 if (env_app_timeout
) {
5067 app_socket_timeout
= atoi(env_app_timeout
);
5069 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5075 /* Initialize communication library */
5077 /* This is to get the TCP timeout value. */
5078 lttcomm_inet_init();
5081 * Initialize the health check subsystem. This call should set the
5082 * appropriate time values.
5084 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5085 if (!health_sessiond
) {
5086 PERROR("health_app_create error");
5087 goto exit_health_sessiond_cleanup
;
5090 /* Create thread to clean up RCU hash tables */
5091 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5092 thread_ht_cleanup
, (void *) NULL
);
5094 PERROR("pthread_create ht_cleanup");
5095 goto exit_ht_cleanup
;
5098 /* Create health-check thread */
5099 ret
= pthread_create(&health_thread
, NULL
,
5100 thread_manage_health
, (void *) NULL
);
5102 PERROR("pthread_create health");
5106 /* Create thread to manage the client socket */
5107 ret
= pthread_create(&client_thread
, NULL
,
5108 thread_manage_clients
, (void *) NULL
);
5110 PERROR("pthread_create clients");
5114 /* Create thread to dispatch registration */
5115 ret
= pthread_create(&dispatch_thread
, NULL
,
5116 thread_dispatch_ust_registration
, (void *) NULL
);
5118 PERROR("pthread_create dispatch");
5122 /* Create thread to manage application registration. */
5123 ret
= pthread_create(®_apps_thread
, NULL
,
5124 thread_registration_apps
, (void *) NULL
);
5126 PERROR("pthread_create registration");
5130 /* Create thread to manage application socket */
5131 ret
= pthread_create(&apps_thread
, NULL
,
5132 thread_manage_apps
, (void *) NULL
);
5134 PERROR("pthread_create apps");
5138 /* Create thread to manage application notify socket */
5139 ret
= pthread_create(&apps_notify_thread
, NULL
,
5140 ust_thread_manage_notify
, (void *) NULL
);
5142 PERROR("pthread_create apps");
5143 goto exit_apps_notify
;
5146 /* Create JUL registration thread. */
5147 ret
= pthread_create(&jul_reg_thread
, NULL
,
5148 jul_thread_manage_registration
, (void *) NULL
);
5150 PERROR("pthread_create apps");
5154 /* Don't start this thread if kernel tracing is not requested nor root */
5155 if (is_root
&& !opt_no_kernel
) {
5156 /* Create kernel thread to manage kernel event */
5157 ret
= pthread_create(&kernel_thread
, NULL
,
5158 thread_manage_kernel
, (void *) NULL
);
5160 PERROR("pthread_create kernel");
5164 ret
= pthread_join(kernel_thread
, &status
);
5166 PERROR("pthread_join");
5167 goto error
; /* join error, exit without cleanup */
5172 ret
= pthread_join(jul_reg_thread
, &status
);
5174 PERROR("pthread_join JUL");
5175 goto error
; /* join error, exit without cleanup */
5179 ret
= pthread_join(apps_notify_thread
, &status
);
5181 PERROR("pthread_join apps notify");
5182 goto error
; /* join error, exit without cleanup */
5186 ret
= pthread_join(apps_thread
, &status
);
5188 PERROR("pthread_join apps");
5189 goto error
; /* join error, exit without cleanup */
5194 ret
= pthread_join(reg_apps_thread
, &status
);
5196 PERROR("pthread_join");
5197 goto error
; /* join error, exit without cleanup */
5201 ret
= pthread_join(dispatch_thread
, &status
);
5203 PERROR("pthread_join");
5204 goto error
; /* join error, exit without cleanup */
5208 ret
= pthread_join(client_thread
, &status
);
5210 PERROR("pthread_join");
5211 goto error
; /* join error, exit without cleanup */
5214 ret
= join_consumer_thread(&kconsumer_data
);
5216 PERROR("join_consumer");
5217 goto error
; /* join error, exit without cleanup */
5220 ret
= join_consumer_thread(&ustconsumer32_data
);
5222 PERROR("join_consumer ust32");
5223 goto error
; /* join error, exit without cleanup */
5226 ret
= join_consumer_thread(&ustconsumer64_data
);
5228 PERROR("join_consumer ust64");
5229 goto error
; /* join error, exit without cleanup */
5233 ret
= pthread_join(health_thread
, &status
);
5235 PERROR("pthread_join health thread");
5236 goto error
; /* join error, exit without cleanup */
5240 ret
= pthread_join(ht_cleanup_thread
, &status
);
5242 PERROR("pthread_join ht cleanup thread");
5243 goto error
; /* join error, exit without cleanup */
5246 health_app_destroy(health_sessiond
);
5247 exit_health_sessiond_cleanup
:
5250 * cleanup() is called when no other thread is running.
5252 rcu_thread_online();
5254 rcu_thread_offline();
5255 rcu_unregister_thread();