2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
34 #include <sys/types.h>
36 #include <urcu/uatomic.h>
40 #include <common/common.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
48 #include "lttng-sessiond.h"
49 #include "buffer-registry.h"
56 #include "kernel-consumer.h"
60 #include "ust-consumer.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
67 #define CONSUMERD_FILE "lttng-consumerd"
70 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
73 const char *opt_tracing_group
;
74 static const char *opt_pidfile
;
75 static int opt_sig_parent
;
76 static int opt_verbose_consumer
;
77 static int opt_daemon
;
78 static int opt_no_kernel
;
79 static int is_root
; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid
; /* Parent PID for --sig-parent option */
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
87 static struct consumer_data kconsumer_data
= {
88 .type
= LTTNG_CONSUMER_KERNEL
,
89 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
90 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
93 .metadata_sock
.fd
= -1,
94 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
95 .lock
= PTHREAD_MUTEX_INITIALIZER
,
96 .cond
= PTHREAD_COND_INITIALIZER
,
97 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
99 static struct consumer_data ustconsumer64_data
= {
100 .type
= LTTNG_CONSUMER64_UST
,
101 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
102 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
105 .metadata_sock
.fd
= -1,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer32_data
= {
112 .type
= LTTNG_CONSUMER32_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
117 .metadata_sock
.fd
= -1,
118 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 .lock
= PTHREAD_MUTEX_INITIALIZER
,
120 .cond
= PTHREAD_COND_INITIALIZER
,
121 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
124 /* Shared between threads */
125 static int dispatch_thread_exit
;
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path
[PATH_MAX
];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path
[PATH_MAX
];
131 /* global wait shm path for UST */
132 static char wait_shm_path
[PATH_MAX
];
133 /* Global health check unix path */
134 static char health_unix_sock_path
[PATH_MAX
];
136 /* Sockets and FDs */
137 static int client_sock
= -1;
138 static int apps_sock
= -1;
139 int kernel_tracer_fd
= -1;
140 static int kernel_poll_pipe
[2] = { -1, -1 };
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
146 static int thread_quit_pipe
[2] = { -1, -1 };
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
152 static int apps_cmd_pipe
[2] = { -1, -1 };
154 int apps_cmd_notify_pipe
[2] = { -1, -1 };
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread
;
158 static pthread_t apps_notify_thread
;
159 static pthread_t reg_apps_thread
;
160 static pthread_t client_thread
;
161 static pthread_t kernel_thread
;
162 static pthread_t dispatch_thread
;
163 static pthread_t health_thread
;
164 static pthread_t ht_cleanup_thread
;
167 * UST registration command queue. This queue is tied with a futex and uses a N
168 * wakers / 1 waiter implemented and detailed in futex.c/.h
170 * The thread_manage_apps and thread_dispatch_ust_registration interact with
171 * this queue and the wait/wake scheme.
173 static struct ust_cmd_queue ust_cmd_queue
;
176 * Pointer initialized before thread creation.
178 * This points to the tracing session list containing the session count and a
179 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
180 * MUST NOT be taken if you call a public function in session.c.
182 * The lock is nested inside the structure: session_list_ptr->lock. Please use
183 * session_lock_list and session_unlock_list for lock acquisition.
185 static struct ltt_session_list
*session_list_ptr
;
187 int ust_consumerd64_fd
= -1;
188 int ust_consumerd32_fd
= -1;
190 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
191 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
192 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
193 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
195 static const char *module_proc_lttng
= "/proc/lttng";
198 * Consumer daemon state which is changed when spawning it, killing it or in
199 * case of a fatal error.
201 enum consumerd_state
{
202 CONSUMER_STARTED
= 1,
203 CONSUMER_STOPPED
= 2,
208 * This consumer daemon state is used to validate if a client command will be
209 * able to reach the consumer. If not, the client is informed. For instance,
210 * doing a "lttng start" when the consumer state is set to ERROR will return an
211 * error to the client.
213 * The following example shows a possible race condition of this scheme:
215 * consumer thread error happens
217 * client cmd checks state -> still OK
218 * consumer thread exit, sets error
219 * client cmd try to talk to consumer
222 * However, since the consumer is a different daemon, we have no way of making
223 * sure the command will reach it safely even with this state flag. This is why
224 * we consider that up to the state validation during command processing, the
225 * command is safe. After that, we can not guarantee the correctness of the
226 * client request vis-a-vis the consumer.
228 static enum consumerd_state ust_consumerd_state
;
229 static enum consumerd_state kernel_consumerd_state
;
232 * Socket timeout for receiving and sending in seconds.
234 static int app_socket_timeout
;
236 /* Set in main() with the current page size. */
240 void setup_consumerd_path(void)
242 const char *bin
, *libdir
;
245 * Allow INSTALL_BIN_PATH to be used as a target path for the
246 * native architecture size consumer if CONFIG_CONSUMER*_PATH
247 * has not been defined.
249 #if (CAA_BITS_PER_LONG == 32)
250 if (!consumerd32_bin
[0]) {
251 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
253 if (!consumerd32_libdir
[0]) {
254 consumerd32_libdir
= INSTALL_LIB_PATH
;
256 #elif (CAA_BITS_PER_LONG == 64)
257 if (!consumerd64_bin
[0]) {
258 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
260 if (!consumerd64_libdir
[0]) {
261 consumerd64_libdir
= INSTALL_LIB_PATH
;
264 #error "Unknown bitness"
268 * runtime env. var. overrides the build default.
270 bin
= getenv("LTTNG_CONSUMERD32_BIN");
272 consumerd32_bin
= bin
;
274 bin
= getenv("LTTNG_CONSUMERD64_BIN");
276 consumerd64_bin
= bin
;
278 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
280 consumerd32_libdir
= libdir
;
282 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
284 consumerd64_libdir
= libdir
;
289 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
291 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
297 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
303 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
315 * Check if the thread quit pipe was triggered.
317 * Return 1 if it was triggered else 0;
319 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
321 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
329 * Return group ID of the tracing group or -1 if not found.
331 static gid_t
allowed_group(void)
335 if (opt_tracing_group
) {
336 grp
= getgrnam(opt_tracing_group
);
338 grp
= getgrnam(default_tracing_group
);
348 * Init thread quit pipe.
350 * Return -1 on error or 0 if all pipes are created.
352 static int init_thread_quit_pipe(void)
356 ret
= pipe(thread_quit_pipe
);
358 PERROR("thread quit pipe");
362 for (i
= 0; i
< 2; i
++) {
363 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
375 * Stop all threads by closing the thread quit pipe.
377 static void stop_threads(void)
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
385 ERR("write error on thread quit pipe");
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
390 futex_nto1_wake(&ust_cmd_queue
.futex
);
394 * Close every consumer sockets.
396 static void close_consumer_sockets(void)
400 if (kconsumer_data
.err_sock
>= 0) {
401 ret
= close(kconsumer_data
.err_sock
);
403 PERROR("kernel consumer err_sock close");
406 if (ustconsumer32_data
.err_sock
>= 0) {
407 ret
= close(ustconsumer32_data
.err_sock
);
409 PERROR("UST consumer32 err_sock close");
412 if (ustconsumer64_data
.err_sock
>= 0) {
413 ret
= close(ustconsumer64_data
.err_sock
);
415 PERROR("UST consumer64 err_sock close");
418 if (kconsumer_data
.cmd_sock
>= 0) {
419 ret
= close(kconsumer_data
.cmd_sock
);
421 PERROR("kernel consumer cmd_sock close");
424 if (ustconsumer32_data
.cmd_sock
>= 0) {
425 ret
= close(ustconsumer32_data
.cmd_sock
);
427 PERROR("UST consumer32 cmd_sock close");
430 if (ustconsumer64_data
.cmd_sock
>= 0) {
431 ret
= close(ustconsumer64_data
.cmd_sock
);
433 PERROR("UST consumer64 cmd_sock close");
441 static void cleanup(void)
445 struct ltt_session
*sess
, *stmp
;
449 /* First thing first, stop all threads */
450 utils_close_pipe(thread_quit_pipe
);
453 * If opt_pidfile is undefined, the default file will be wiped when
454 * removing the rundir.
457 ret
= remove(opt_pidfile
);
459 PERROR("remove pidfile %s", opt_pidfile
);
463 DBG("Removing %s directory", rundir
);
464 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
466 ERR("asprintf failed. Something is really wrong!");
469 /* Remove lttng run directory */
472 ERR("Unable to clean %s", rundir
);
477 DBG("Cleaning up all sessions");
479 /* Destroy session list mutex */
480 if (session_list_ptr
!= NULL
) {
481 pthread_mutex_destroy(&session_list_ptr
->lock
);
483 /* Cleanup ALL session */
484 cds_list_for_each_entry_safe(sess
, stmp
,
485 &session_list_ptr
->head
, list
) {
486 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
490 DBG("Closing all UST sockets");
491 ust_app_clean_list();
492 buffer_reg_destroy_registries();
494 if (is_root
&& !opt_no_kernel
) {
495 DBG2("Closing kernel fd");
496 if (kernel_tracer_fd
>= 0) {
497 ret
= close(kernel_tracer_fd
);
502 DBG("Unloading kernel modules");
503 modprobe_remove_lttng_all();
506 close_consumer_sockets();
509 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
510 "Matthew, BEET driven development works!%c[%dm",
511 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
516 * Send data on a unix socket using the liblttsessiondcomm API.
518 * Return lttcomm error code.
520 static int send_unix_sock(int sock
, void *buf
, size_t len
)
522 /* Check valid length */
527 return lttcomm_send_unix_sock(sock
, buf
, len
);
531 * Free memory of a command context structure.
533 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
535 DBG("Clean command context structure");
537 if ((*cmd_ctx
)->llm
) {
538 free((*cmd_ctx
)->llm
);
540 if ((*cmd_ctx
)->lsm
) {
541 free((*cmd_ctx
)->lsm
);
549 * Notify UST applications using the shm mmap futex.
551 static int notify_ust_apps(int active
)
555 DBG("Notifying applications of session daemon state: %d", active
);
557 /* See shm.c for this call implying mmap, shm and futex calls */
558 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
559 if (wait_shm_mmap
== NULL
) {
563 /* Wake waiting process */
564 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
566 /* Apps notified successfully */
574 * Setup the outgoing data buffer for the response (llm) by allocating the
575 * right amount of memory and copying the original information from the lsm
578 * Return total size of the buffer pointed by buf.
580 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
586 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
587 if (cmd_ctx
->llm
== NULL
) {
593 /* Copy common data */
594 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
595 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
597 cmd_ctx
->llm
->data_size
= size
;
598 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
607 * Update the kernel poll set of all channel fd available over all tracing
608 * session. Add the wakeup pipe at the end of the set.
610 static int update_kernel_poll(struct lttng_poll_event
*events
)
613 struct ltt_session
*session
;
614 struct ltt_kernel_channel
*channel
;
616 DBG("Updating kernel poll set");
619 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
620 session_lock(session
);
621 if (session
->kernel_session
== NULL
) {
622 session_unlock(session
);
626 cds_list_for_each_entry(channel
,
627 &session
->kernel_session
->channel_list
.head
, list
) {
628 /* Add channel fd to the kernel poll set */
629 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
631 session_unlock(session
);
634 DBG("Channel fd %d added to kernel set", channel
->fd
);
636 session_unlock(session
);
638 session_unlock_list();
643 session_unlock_list();
648 * Find the channel fd from 'fd' over all tracing session. When found, check
649 * for new channel stream and send those stream fds to the kernel consumer.
651 * Useful for CPU hotplug feature.
653 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
656 struct ltt_session
*session
;
657 struct ltt_kernel_session
*ksess
;
658 struct ltt_kernel_channel
*channel
;
660 DBG("Updating kernel streams for channel fd %d", fd
);
663 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
664 session_lock(session
);
665 if (session
->kernel_session
== NULL
) {
666 session_unlock(session
);
669 ksess
= session
->kernel_session
;
671 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
672 if (channel
->fd
== fd
) {
673 DBG("Channel found, updating kernel streams");
674 ret
= kernel_open_channel_stream(channel
);
680 * Have we already sent fds to the consumer? If yes, it means
681 * that tracing is started so it is safe to send our updated
684 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
685 struct lttng_ht_iter iter
;
686 struct consumer_socket
*socket
;
689 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
690 &iter
.iter
, socket
, node
.node
) {
691 /* Code flow error */
692 assert(socket
->fd
>= 0);
694 pthread_mutex_lock(socket
->lock
);
695 ret
= kernel_consumer_send_channel_stream(socket
,
697 pthread_mutex_unlock(socket
->lock
);
708 session_unlock(session
);
710 session_unlock_list();
714 session_unlock(session
);
715 session_unlock_list();
720 * For each tracing session, update newly registered apps. The session list
721 * lock MUST be acquired before calling this.
723 static void update_ust_app(int app_sock
)
725 struct ltt_session
*sess
, *stmp
;
727 /* For all tracing session(s) */
728 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
730 if (sess
->ust_session
) {
731 ust_app_global_update(sess
->ust_session
, app_sock
);
733 session_unlock(sess
);
738 * This thread manage event coming from the kernel.
740 * Features supported in this thread:
743 static void *thread_manage_kernel(void *data
)
745 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
746 uint32_t revents
, nb_fd
;
748 struct lttng_poll_event events
;
750 DBG("[thread] Thread manage kernel started");
752 health_register(HEALTH_TYPE_KERNEL
);
755 * This first step of the while is to clean this structure which could free
756 * non NULL pointers so initialize it before the loop.
758 lttng_poll_init(&events
);
760 if (testpoint(thread_manage_kernel
)) {
761 goto error_testpoint
;
764 health_code_update();
766 if (testpoint(thread_manage_kernel_before_loop
)) {
767 goto error_testpoint
;
771 health_code_update();
773 if (update_poll_flag
== 1) {
774 /* Clean events object. We are about to populate it again. */
775 lttng_poll_clean(&events
);
777 ret
= sessiond_set_thread_pollset(&events
, 2);
779 goto error_poll_create
;
782 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
787 /* This will add the available kernel channel if any. */
788 ret
= update_kernel_poll(&events
);
792 update_poll_flag
= 0;
795 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
797 /* Poll infinite value of time */
800 ret
= lttng_poll_wait(&events
, -1);
804 * Restart interrupted system call.
806 if (errno
== EINTR
) {
810 } else if (ret
== 0) {
811 /* Should not happen since timeout is infinite */
812 ERR("Return value of poll is 0 with an infinite timeout.\n"
813 "This should not have happened! Continuing...");
819 for (i
= 0; i
< nb_fd
; i
++) {
820 /* Fetch once the poll data */
821 revents
= LTTNG_POLL_GETEV(&events
, i
);
822 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
824 health_code_update();
826 /* Thread quit pipe has been closed. Killing thread. */
827 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
833 /* Check for data on kernel pipe */
834 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
836 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
837 } while (ret
< 0 && errno
== EINTR
);
839 * Ret value is useless here, if this pipe gets any actions an
840 * update is required anyway.
842 update_poll_flag
= 1;
846 * New CPU detected by the kernel. Adding kernel stream to
847 * kernel session and updating the kernel consumer
849 if (revents
& LPOLLIN
) {
850 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
856 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
857 * and unregister kernel stream at this point.
866 lttng_poll_clean(&events
);
869 utils_close_pipe(kernel_poll_pipe
);
870 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
873 ERR("Health error occurred in %s", __func__
);
874 WARN("Kernel thread died unexpectedly. "
875 "Kernel tracing can continue but CPU hotplug is disabled.");
878 DBG("Kernel thread dying");
883 * Signal pthread condition of the consumer data that the thread.
885 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
887 pthread_mutex_lock(&data
->cond_mutex
);
890 * The state is set before signaling. It can be any value, it's the waiter
891 * job to correctly interpret this condition variable associated to the
892 * consumer pthread_cond.
894 * A value of 0 means that the corresponding thread of the consumer data
895 * was not started. 1 indicates that the thread has started and is ready
896 * for action. A negative value means that there was an error during the
899 data
->consumer_thread_is_ready
= state
;
900 (void) pthread_cond_signal(&data
->cond
);
902 pthread_mutex_unlock(&data
->cond_mutex
);
906 * This thread manage the consumer error sent back to the session daemon.
908 static void *thread_manage_consumer(void *data
)
910 int sock
= -1, i
, ret
, pollfd
, err
= -1;
911 uint32_t revents
, nb_fd
;
912 enum lttcomm_return_code code
;
913 struct lttng_poll_event events
;
914 struct consumer_data
*consumer_data
= data
;
916 DBG("[thread] Manage consumer started");
918 health_register(HEALTH_TYPE_CONSUMER
);
920 health_code_update();
923 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
924 * metadata_sock. Nothing more will be added to this poll set.
926 ret
= sessiond_set_thread_pollset(&events
, 3);
932 * The error socket here is already in a listening state which was done
933 * just before spawning this thread to avoid a race between the consumer
934 * daemon exec trying to connect and the listen() call.
936 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
941 health_code_update();
943 /* Infinite blocking call, waiting for transmission */
947 if (testpoint(thread_manage_consumer
)) {
951 ret
= lttng_poll_wait(&events
, -1);
955 * Restart interrupted system call.
957 if (errno
== EINTR
) {
965 for (i
= 0; i
< nb_fd
; i
++) {
966 /* Fetch once the poll data */
967 revents
= LTTNG_POLL_GETEV(&events
, i
);
968 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
970 health_code_update();
972 /* Thread quit pipe has been closed. Killing thread. */
973 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
979 /* Event on the registration socket */
980 if (pollfd
== consumer_data
->err_sock
) {
981 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
982 ERR("consumer err socket poll error");
988 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
994 * Set the CLOEXEC flag. Return code is useless because either way, the
997 (void) utils_set_fd_cloexec(sock
);
999 health_code_update();
1001 DBG2("Receiving code from consumer err_sock");
1003 /* Getting status code from kconsumerd */
1004 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1005 sizeof(enum lttcomm_return_code
));
1010 health_code_update();
1012 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1013 /* Connect both socket, command and metadata. */
1014 consumer_data
->cmd_sock
=
1015 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1016 consumer_data
->metadata_sock
.fd
=
1017 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1018 if (consumer_data
->cmd_sock
< 0 ||
1019 consumer_data
->metadata_sock
.fd
< 0) {
1020 PERROR("consumer connect cmd socket");
1021 /* On error, signal condition and quit. */
1022 signal_consumer_condition(consumer_data
, -1);
1025 /* Create metadata socket lock. */
1026 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1027 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1028 PERROR("zmalloc pthread mutex");
1032 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1034 signal_consumer_condition(consumer_data
, 1);
1035 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1036 DBG("Consumer metadata socket ready (fd: %d)",
1037 consumer_data
->metadata_sock
.fd
);
1039 ERR("consumer error when waiting for SOCK_READY : %s",
1040 lttcomm_get_readable_code(-code
));
1044 /* Remove the consumerd error sock since we've established a connexion */
1045 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1050 /* Add new accepted error socket. */
1051 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1056 /* Add metadata socket that is successfully connected. */
1057 ret
= lttng_poll_add(&events
, consumer_data
->metadata_sock
.fd
,
1058 LPOLLIN
| LPOLLRDHUP
);
1063 health_code_update();
1065 /* Infinite blocking call, waiting for transmission */
1068 health_poll_entry();
1069 ret
= lttng_poll_wait(&events
, -1);
1073 * Restart interrupted system call.
1075 if (errno
== EINTR
) {
1083 for (i
= 0; i
< nb_fd
; i
++) {
1084 /* Fetch once the poll data */
1085 revents
= LTTNG_POLL_GETEV(&events
, i
);
1086 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1088 health_code_update();
1090 /* Thread quit pipe has been closed. Killing thread. */
1091 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1097 if (pollfd
== sock
) {
1098 /* Event on the consumerd socket */
1099 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1100 ERR("consumer err socket second poll error");
1103 health_code_update();
1104 /* Wait for any kconsumerd error */
1105 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1106 sizeof(enum lttcomm_return_code
));
1108 ERR("consumer closed the command socket");
1112 ERR("consumer return code : %s",
1113 lttcomm_get_readable_code(-code
));
1116 } else if (pollfd
== consumer_data
->metadata_sock
.fd
) {
1117 /* UST metadata requests */
1118 ret
= ust_consumer_metadata_request(
1119 &consumer_data
->metadata_sock
);
1121 ERR("Handling metadata request");
1126 ERR("Unknown pollfd");
1130 health_code_update();
1135 /* Immediately set the consumerd state to stopped */
1136 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1137 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1138 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1139 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1140 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1142 /* Code flow error... */
1146 if (consumer_data
->err_sock
>= 0) {
1147 ret
= close(consumer_data
->err_sock
);
1152 if (consumer_data
->cmd_sock
>= 0) {
1153 ret
= close(consumer_data
->cmd_sock
);
1158 if (consumer_data
->metadata_sock
.fd
>= 0) {
1159 ret
= close(consumer_data
->metadata_sock
.fd
);
1164 /* Cleanup metadata socket mutex. */
1165 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1166 free(consumer_data
->metadata_sock
.lock
);
1175 unlink(consumer_data
->err_unix_sock_path
);
1176 unlink(consumer_data
->cmd_unix_sock_path
);
1177 consumer_data
->pid
= 0;
1179 lttng_poll_clean(&events
);
1183 ERR("Health error occurred in %s", __func__
);
1185 health_unregister();
1186 DBG("consumer thread cleanup completed");
1192 * This thread manage application communication.
1194 static void *thread_manage_apps(void *data
)
1196 int i
, ret
, pollfd
, err
= -1;
1197 uint32_t revents
, nb_fd
;
1198 struct lttng_poll_event events
;
1200 DBG("[thread] Manage application started");
1202 rcu_register_thread();
1203 rcu_thread_online();
1205 health_register(HEALTH_TYPE_APP_MANAGE
);
1207 if (testpoint(thread_manage_apps
)) {
1208 goto error_testpoint
;
1211 health_code_update();
1213 ret
= sessiond_set_thread_pollset(&events
, 2);
1215 goto error_poll_create
;
1218 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1223 if (testpoint(thread_manage_apps_before_loop
)) {
1227 health_code_update();
1230 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1232 /* Inifinite blocking call, waiting for transmission */
1234 health_poll_entry();
1235 ret
= lttng_poll_wait(&events
, -1);
1239 * Restart interrupted system call.
1241 if (errno
== EINTR
) {
1249 for (i
= 0; i
< nb_fd
; i
++) {
1250 /* Fetch once the poll data */
1251 revents
= LTTNG_POLL_GETEV(&events
, i
);
1252 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1254 health_code_update();
1256 /* Thread quit pipe has been closed. Killing thread. */
1257 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1263 /* Inspect the apps cmd pipe */
1264 if (pollfd
== apps_cmd_pipe
[0]) {
1265 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1266 ERR("Apps command pipe error");
1268 } else if (revents
& LPOLLIN
) {
1273 ret
= read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1274 } while (ret
< 0 && errno
== EINTR
);
1275 if (ret
< 0 || ret
< sizeof(sock
)) {
1276 PERROR("read apps cmd pipe");
1280 health_code_update();
1283 * We only monitor the error events of the socket. This
1284 * thread does not handle any incoming data from UST
1287 ret
= lttng_poll_add(&events
, sock
,
1288 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1293 /* Set socket timeout for both receiving and ending */
1294 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1295 app_socket_timeout
);
1296 (void) lttcomm_setsockopt_snd_timeout(sock
,
1297 app_socket_timeout
);
1299 DBG("Apps with sock %d added to poll set", sock
);
1301 health_code_update();
1307 * At this point, we know that a registered application made
1308 * the event at poll_wait.
1310 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1311 /* Removing from the poll set */
1312 ret
= lttng_poll_del(&events
, pollfd
);
1317 /* Socket closed on remote end. */
1318 ust_app_unregister(pollfd
);
1323 health_code_update();
1329 lttng_poll_clean(&events
);
1332 utils_close_pipe(apps_cmd_pipe
);
1333 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1336 * We don't clean the UST app hash table here since already registered
1337 * applications can still be controlled so let them be until the session
1338 * daemon dies or the applications stop.
1343 ERR("Health error occurred in %s", __func__
);
1345 health_unregister();
1346 DBG("Application communication apps thread cleanup complete");
1347 rcu_thread_offline();
1348 rcu_unregister_thread();
1353 * Send a socket to a thread This is called from the dispatch UST registration
1354 * thread once all sockets are set for the application.
1356 * On success, return 0 else a negative value being the errno message of the
1359 static int send_socket_to_thread(int fd
, int sock
)
1363 /* Sockets MUST be set or else this should not have been called. */
1368 ret
= write(fd
, &sock
, sizeof(sock
));
1369 } while (ret
< 0 && errno
== EINTR
);
1370 if (ret
< 0 || ret
!= sizeof(sock
)) {
1371 PERROR("write apps pipe %d", fd
);
1378 /* All good. Don't send back the write positive ret value. */
1385 * Sanitize the wait queue of the dispatch registration thread meaning removing
1386 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1387 * notify socket is never received.
1389 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1391 int ret
, nb_fd
= 0, i
;
1392 unsigned int fd_added
= 0;
1393 struct lttng_poll_event events
;
1394 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1398 lttng_poll_init(&events
);
1400 /* Just skip everything for an empty queue. */
1401 if (!wait_queue
->count
) {
1405 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1410 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1411 &wait_queue
->head
, head
) {
1412 assert(wait_node
->app
);
1413 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1414 LPOLLHUP
| LPOLLERR
);
1427 * Poll but don't block so we can quickly identify the faulty events and
1428 * clean them afterwards from the wait queue.
1430 ret
= lttng_poll_wait(&events
, 0);
1436 for (i
= 0; i
< nb_fd
; i
++) {
1437 /* Get faulty FD. */
1438 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1439 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1441 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1442 &wait_queue
->head
, head
) {
1443 if (pollfd
== wait_node
->app
->sock
&&
1444 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1445 cds_list_del(&wait_node
->head
);
1446 wait_queue
->count
--;
1447 ust_app_destroy(wait_node
->app
);
1455 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1459 lttng_poll_clean(&events
);
1463 lttng_poll_clean(&events
);
1465 ERR("Unable to sanitize wait queue");
1470 * Dispatch request from the registration threads to the application
1471 * communication thread.
1473 static void *thread_dispatch_ust_registration(void *data
)
1476 struct cds_wfq_node
*node
;
1477 struct ust_command
*ust_cmd
= NULL
;
1478 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1479 struct ust_reg_wait_queue wait_queue
= {
1483 health_register(HEALTH_TYPE_APP_REG_DISPATCH
);
1485 health_code_update();
1487 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1489 DBG("[thread] Dispatch UST command started");
1491 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1492 health_code_update();
1494 /* Atomically prepare the queue futex */
1495 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1498 struct ust_app
*app
= NULL
;
1502 * Make sure we don't have node(s) that have hung up before receiving
1503 * the notify socket. This is to clean the list in order to avoid
1504 * memory leaks from notify socket that are never seen.
1506 sanitize_wait_queue(&wait_queue
);
1508 health_code_update();
1509 /* Dequeue command for registration */
1510 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1512 DBG("Woken up but nothing in the UST command queue");
1513 /* Continue thread execution */
1517 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1519 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1520 " gid:%d sock:%d name:%s (version %d.%d)",
1521 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1522 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1523 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1524 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1526 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1527 wait_node
= zmalloc(sizeof(*wait_node
));
1529 PERROR("zmalloc wait_node dispatch");
1530 ret
= close(ust_cmd
->sock
);
1532 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1534 lttng_fd_put(1, LTTNG_FD_APPS
);
1538 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1540 /* Create application object if socket is CMD. */
1541 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1543 if (!wait_node
->app
) {
1544 ret
= close(ust_cmd
->sock
);
1546 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1548 lttng_fd_put(1, LTTNG_FD_APPS
);
1554 * Add application to the wait queue so we can set the notify
1555 * socket before putting this object in the global ht.
1557 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1562 * We have to continue here since we don't have the notify
1563 * socket and the application MUST be added to the hash table
1564 * only at that moment.
1569 * Look for the application in the local wait queue and set the
1570 * notify socket if found.
1572 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1573 &wait_queue
.head
, head
) {
1574 health_code_update();
1575 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1576 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1577 cds_list_del(&wait_node
->head
);
1579 app
= wait_node
->app
;
1581 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1587 * With no application at this stage the received socket is
1588 * basically useless so close it before we free the cmd data
1589 * structure for good.
1592 ret
= close(ust_cmd
->sock
);
1594 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1596 lttng_fd_put(1, LTTNG_FD_APPS
);
1603 * @session_lock_list
1605 * Lock the global session list so from the register up to the
1606 * registration done message, no thread can see the application
1607 * and change its state.
1609 session_lock_list();
1613 * Add application to the global hash table. This needs to be
1614 * done before the update to the UST registry can locate the
1619 /* Set app version. This call will print an error if needed. */
1620 (void) ust_app_version(app
);
1622 /* Send notify socket through the notify pipe. */
1623 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1627 session_unlock_list();
1628 /* No notify thread, stop the UST tracing. */
1633 * Update newly registered application with the tracing
1634 * registry info already enabled information.
1636 update_ust_app(app
->sock
);
1639 * Don't care about return value. Let the manage apps threads
1640 * handle app unregistration upon socket close.
1642 (void) ust_app_register_done(app
->sock
);
1645 * Even if the application socket has been closed, send the app
1646 * to the thread and unregistration will take place at that
1649 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1652 session_unlock_list();
1653 /* No apps. thread, stop the UST tracing. */
1658 session_unlock_list();
1660 } while (node
!= NULL
);
1662 health_poll_entry();
1663 /* Futex wait on queue. Blocking call on futex() */
1664 futex_nto1_wait(&ust_cmd_queue
.futex
);
1667 /* Normal exit, no error */
1671 /* Clean up wait queue. */
1672 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1673 &wait_queue
.head
, head
) {
1674 cds_list_del(&wait_node
->head
);
1679 DBG("Dispatch thread dying");
1682 ERR("Health error occurred in %s", __func__
);
1684 health_unregister();
1689 * This thread manage application registration.
1691 static void *thread_registration_apps(void *data
)
1693 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1694 uint32_t revents
, nb_fd
;
1695 struct lttng_poll_event events
;
1697 * Get allocated in this thread, enqueued to a global queue, dequeued and
1698 * freed in the manage apps thread.
1700 struct ust_command
*ust_cmd
= NULL
;
1702 DBG("[thread] Manage application registration started");
1704 health_register(HEALTH_TYPE_APP_REG
);
1706 if (testpoint(thread_registration_apps
)) {
1707 goto error_testpoint
;
1710 ret
= lttcomm_listen_unix_sock(apps_sock
);
1716 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1717 * more will be added to this poll set.
1719 ret
= sessiond_set_thread_pollset(&events
, 2);
1721 goto error_create_poll
;
1724 /* Add the application registration socket */
1725 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1727 goto error_poll_add
;
1730 /* Notify all applications to register */
1731 ret
= notify_ust_apps(1);
1733 ERR("Failed to notify applications or create the wait shared memory.\n"
1734 "Execution continues but there might be problem for already\n"
1735 "running applications that wishes to register.");
1739 DBG("Accepting application registration");
1741 /* Inifinite blocking call, waiting for transmission */
1743 health_poll_entry();
1744 ret
= lttng_poll_wait(&events
, -1);
1748 * Restart interrupted system call.
1750 if (errno
== EINTR
) {
1758 for (i
= 0; i
< nb_fd
; i
++) {
1759 health_code_update();
1761 /* Fetch once the poll data */
1762 revents
= LTTNG_POLL_GETEV(&events
, i
);
1763 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1765 /* Thread quit pipe has been closed. Killing thread. */
1766 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1772 /* Event on the registration socket */
1773 if (pollfd
== apps_sock
) {
1774 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1775 ERR("Register apps socket poll error");
1777 } else if (revents
& LPOLLIN
) {
1778 sock
= lttcomm_accept_unix_sock(apps_sock
);
1784 * Set the CLOEXEC flag. Return code is useless because
1785 * either way, the show must go on.
1787 (void) utils_set_fd_cloexec(sock
);
1789 /* Create UST registration command for enqueuing */
1790 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1791 if (ust_cmd
== NULL
) {
1792 PERROR("ust command zmalloc");
1797 * Using message-based transmissions to ensure we don't
1798 * have to deal with partially received messages.
1800 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1802 ERR("Exhausted file descriptors allowed for applications.");
1812 health_code_update();
1813 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1816 /* Close socket of the application. */
1821 lttng_fd_put(LTTNG_FD_APPS
, 1);
1825 health_code_update();
1827 ust_cmd
->sock
= sock
;
1830 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1831 " gid:%d sock:%d name:%s (version %d.%d)",
1832 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1833 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1834 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1835 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1838 * Lock free enqueue the registration request. The red pill
1839 * has been taken! This apps will be part of the *system*.
1841 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1844 * Wake the registration queue futex. Implicit memory
1845 * barrier with the exchange in cds_wfq_enqueue.
1847 futex_nto1_wake(&ust_cmd_queue
.futex
);
1857 ERR("Health error occurred in %s", __func__
);
1860 /* Notify that the registration thread is gone */
1863 if (apps_sock
>= 0) {
1864 ret
= close(apps_sock
);
1874 lttng_fd_put(LTTNG_FD_APPS
, 1);
1876 unlink(apps_unix_sock_path
);
1879 lttng_poll_clean(&events
);
1883 DBG("UST Registration thread cleanup complete");
1884 health_unregister();
1890 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1891 * exec or it will fails.
1893 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1896 struct timespec timeout
;
1898 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1899 consumer_data
->consumer_thread_is_ready
= 0;
1901 /* Setup pthread condition */
1902 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1905 PERROR("pthread_condattr_init consumer data");
1910 * Set the monotonic clock in order to make sure we DO NOT jump in time
1911 * between the clock_gettime() call and the timedwait call. See bug #324
1912 * for a more details and how we noticed it.
1914 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1917 PERROR("pthread_condattr_setclock consumer data");
1921 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1924 PERROR("pthread_cond_init consumer data");
1928 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1931 PERROR("pthread_create consumer");
1936 /* We are about to wait on a pthread condition */
1937 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1939 /* Get time for sem_timedwait absolute timeout */
1940 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1942 * Set the timeout for the condition timed wait even if the clock gettime
1943 * call fails since we might loop on that call and we want to avoid to
1944 * increment the timeout too many times.
1946 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1949 * The following loop COULD be skipped in some conditions so this is why we
1950 * set ret to 0 in order to make sure at least one round of the loop is
1956 * Loop until the condition is reached or when a timeout is reached. Note
1957 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1958 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1959 * possible. This loop does not take any chances and works with both of
1962 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1963 if (clock_ret
< 0) {
1964 PERROR("clock_gettime spawn consumer");
1965 /* Infinite wait for the consumerd thread to be ready */
1966 ret
= pthread_cond_wait(&consumer_data
->cond
,
1967 &consumer_data
->cond_mutex
);
1969 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1970 &consumer_data
->cond_mutex
, &timeout
);
1974 /* Release the pthread condition */
1975 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1979 if (ret
== ETIMEDOUT
) {
1981 * Call has timed out so we kill the kconsumerd_thread and return
1984 ERR("Condition timed out. The consumer thread was never ready."
1986 ret
= pthread_cancel(consumer_data
->thread
);
1988 PERROR("pthread_cancel consumer thread");
1991 PERROR("pthread_cond_wait failed consumer thread");
1996 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1997 if (consumer_data
->pid
== 0) {
1998 ERR("Consumerd did not start");
1999 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2002 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2011 * Join consumer thread
2013 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2017 /* Consumer pid must be a real one. */
2018 if (consumer_data
->pid
> 0) {
2020 ret
= kill(consumer_data
->pid
, SIGTERM
);
2022 ERR("Error killing consumer daemon");
2025 return pthread_join(consumer_data
->thread
, &status
);
2032 * Fork and exec a consumer daemon (consumerd).
2034 * Return pid if successful else -1.
2036 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2040 const char *consumer_to_use
;
2041 const char *verbosity
;
2044 DBG("Spawning consumerd");
2051 if (opt_verbose_consumer
) {
2052 verbosity
= "--verbose";
2054 verbosity
= "--quiet";
2056 switch (consumer_data
->type
) {
2057 case LTTNG_CONSUMER_KERNEL
:
2059 * Find out which consumerd to execute. We will first try the
2060 * 64-bit path, then the sessiond's installation directory, and
2061 * fallback on the 32-bit one,
2063 DBG3("Looking for a kernel consumer at these locations:");
2064 DBG3(" 1) %s", consumerd64_bin
);
2065 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2066 DBG3(" 3) %s", consumerd32_bin
);
2067 if (stat(consumerd64_bin
, &st
) == 0) {
2068 DBG3("Found location #1");
2069 consumer_to_use
= consumerd64_bin
;
2070 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2071 DBG3("Found location #2");
2072 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2073 } else if (stat(consumerd32_bin
, &st
) == 0) {
2074 DBG3("Found location #3");
2075 consumer_to_use
= consumerd32_bin
;
2077 DBG("Could not find any valid consumerd executable");
2080 DBG("Using kernel consumer at: %s", consumer_to_use
);
2081 execl(consumer_to_use
,
2082 "lttng-consumerd", verbosity
, "-k",
2083 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2084 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2087 case LTTNG_CONSUMER64_UST
:
2089 char *tmpnew
= NULL
;
2091 if (consumerd64_libdir
[0] != '\0') {
2095 tmp
= getenv("LD_LIBRARY_PATH");
2099 tmplen
= strlen("LD_LIBRARY_PATH=")
2100 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2101 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2106 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2107 strcat(tmpnew
, consumerd64_libdir
);
2108 if (tmp
[0] != '\0') {
2109 strcat(tmpnew
, ":");
2110 strcat(tmpnew
, tmp
);
2112 ret
= putenv(tmpnew
);
2119 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2120 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2121 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2122 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2124 if (consumerd64_libdir
[0] != '\0') {
2132 case LTTNG_CONSUMER32_UST
:
2134 char *tmpnew
= NULL
;
2136 if (consumerd32_libdir
[0] != '\0') {
2140 tmp
= getenv("LD_LIBRARY_PATH");
2144 tmplen
= strlen("LD_LIBRARY_PATH=")
2145 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2146 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2151 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2152 strcat(tmpnew
, consumerd32_libdir
);
2153 if (tmp
[0] != '\0') {
2154 strcat(tmpnew
, ":");
2155 strcat(tmpnew
, tmp
);
2157 ret
= putenv(tmpnew
);
2164 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2165 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2166 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2167 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2169 if (consumerd32_libdir
[0] != '\0') {
2178 PERROR("unknown consumer type");
2182 PERROR("kernel start consumer exec");
2185 } else if (pid
> 0) {
2188 PERROR("start consumer fork");
2196 * Spawn the consumerd daemon and session daemon thread.
2198 static int start_consumerd(struct consumer_data
*consumer_data
)
2203 * Set the listen() state on the socket since there is a possible race
2204 * between the exec() of the consumer daemon and this call if place in the
2205 * consumer thread. See bug #366 for more details.
2207 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2212 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2213 if (consumer_data
->pid
!= 0) {
2214 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2218 ret
= spawn_consumerd(consumer_data
);
2220 ERR("Spawning consumerd failed");
2221 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2225 /* Setting up the consumer_data pid */
2226 consumer_data
->pid
= ret
;
2227 DBG2("Consumer pid %d", consumer_data
->pid
);
2228 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2230 DBG2("Spawning consumer control thread");
2231 ret
= spawn_consumer_thread(consumer_data
);
2233 ERR("Fatal error spawning consumer control thread");
2241 /* Cleanup already created sockets on error. */
2242 if (consumer_data
->err_sock
>= 0) {
2245 err
= close(consumer_data
->err_sock
);
2247 PERROR("close consumer data error socket");
2254 * Compute health status of each consumer. If one of them is zero (bad
2255 * state), we return 0.
2257 static int check_consumer_health(void)
2261 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2263 DBG3("Health consumer check %d", ret
);
2269 * Setup necessary data for kernel tracer action.
2271 static int init_kernel_tracer(void)
2275 /* Modprobe lttng kernel modules */
2276 ret
= modprobe_lttng_control();
2281 /* Open debugfs lttng */
2282 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2283 if (kernel_tracer_fd
< 0) {
2284 DBG("Failed to open %s", module_proc_lttng
);
2289 /* Validate kernel version */
2290 ret
= kernel_validate_version(kernel_tracer_fd
);
2295 ret
= modprobe_lttng_data();
2300 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2304 modprobe_remove_lttng_control();
2305 ret
= close(kernel_tracer_fd
);
2309 kernel_tracer_fd
= -1;
2310 return LTTNG_ERR_KERN_VERSION
;
2313 ret
= close(kernel_tracer_fd
);
2319 modprobe_remove_lttng_control();
2322 WARN("No kernel tracer available");
2323 kernel_tracer_fd
= -1;
2325 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2327 return LTTNG_ERR_KERN_NA
;
2333 * Copy consumer output from the tracing session to the domain session. The
2334 * function also applies the right modification on a per domain basis for the
2335 * trace files destination directory.
2337 * Should *NOT* be called with RCU read-side lock held.
2339 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2342 const char *dir_name
;
2343 struct consumer_output
*consumer
;
2346 assert(session
->consumer
);
2349 case LTTNG_DOMAIN_KERNEL
:
2350 DBG3("Copying tracing session consumer output in kernel session");
2352 * XXX: We should audit the session creation and what this function
2353 * does "extra" in order to avoid a destroy since this function is used
2354 * in the domain session creation (kernel and ust) only. Same for UST
2357 if (session
->kernel_session
->consumer
) {
2358 consumer_destroy_output(session
->kernel_session
->consumer
);
2360 session
->kernel_session
->consumer
=
2361 consumer_copy_output(session
->consumer
);
2362 /* Ease our life a bit for the next part */
2363 consumer
= session
->kernel_session
->consumer
;
2364 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2366 case LTTNG_DOMAIN_UST
:
2367 DBG3("Copying tracing session consumer output in UST session");
2368 if (session
->ust_session
->consumer
) {
2369 consumer_destroy_output(session
->ust_session
->consumer
);
2371 session
->ust_session
->consumer
=
2372 consumer_copy_output(session
->consumer
);
2373 /* Ease our life a bit for the next part */
2374 consumer
= session
->ust_session
->consumer
;
2375 dir_name
= DEFAULT_UST_TRACE_DIR
;
2378 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2382 /* Append correct directory to subdir */
2383 strncat(consumer
->subdir
, dir_name
,
2384 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2385 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2394 * Create an UST session and add it to the session ust list.
2396 * Should *NOT* be called with RCU read-side lock held.
2398 static int create_ust_session(struct ltt_session
*session
,
2399 struct lttng_domain
*domain
)
2402 struct ltt_ust_session
*lus
= NULL
;
2406 assert(session
->consumer
);
2408 switch (domain
->type
) {
2409 case LTTNG_DOMAIN_UST
:
2412 ERR("Unknown UST domain on create session %d", domain
->type
);
2413 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2417 DBG("Creating UST session");
2419 lus
= trace_ust_create_session(session
->id
);
2421 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2425 lus
->uid
= session
->uid
;
2426 lus
->gid
= session
->gid
;
2427 session
->ust_session
= lus
;
2429 /* Copy session output to the newly created UST session */
2430 ret
= copy_session_consumer(domain
->type
, session
);
2431 if (ret
!= LTTNG_OK
) {
2439 session
->ust_session
= NULL
;
2444 * Create a kernel tracer session then create the default channel.
2446 static int create_kernel_session(struct ltt_session
*session
)
2450 DBG("Creating kernel session");
2452 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2454 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2458 /* Code flow safety */
2459 assert(session
->kernel_session
);
2461 /* Copy session output to the newly created Kernel session */
2462 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2463 if (ret
!= LTTNG_OK
) {
2467 /* Create directory(ies) on local filesystem. */
2468 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2469 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2470 ret
= run_as_mkdir_recursive(
2471 session
->kernel_session
->consumer
->dst
.trace_path
,
2472 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2474 if (ret
!= -EEXIST
) {
2475 ERR("Trace directory creation error");
2481 session
->kernel_session
->uid
= session
->uid
;
2482 session
->kernel_session
->gid
= session
->gid
;
2487 trace_kernel_destroy_session(session
->kernel_session
);
2488 session
->kernel_session
= NULL
;
2493 * Count number of session permitted by uid/gid.
2495 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2498 struct ltt_session
*session
;
2500 DBG("Counting number of available session for UID %d GID %d",
2502 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2504 * Only list the sessions the user can control.
2506 if (!session_access_ok(session
, uid
, gid
)) {
2515 * Process the command requested by the lttng client within the command
2516 * context structure. This function make sure that the return structure (llm)
2517 * is set and ready for transmission before returning.
2519 * Return any error encountered or 0 for success.
2521 * "sock" is only used for special-case var. len data.
2523 * Should *NOT* be called with RCU read-side lock held.
2525 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2529 int need_tracing_session
= 1;
2532 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2536 switch (cmd_ctx
->lsm
->cmd_type
) {
2537 case LTTNG_CREATE_SESSION
:
2538 case LTTNG_DESTROY_SESSION
:
2539 case LTTNG_LIST_SESSIONS
:
2540 case LTTNG_LIST_DOMAINS
:
2541 case LTTNG_START_TRACE
:
2542 case LTTNG_STOP_TRACE
:
2543 case LTTNG_DATA_PENDING
:
2550 if (opt_no_kernel
&& need_domain
2551 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2553 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2555 ret
= LTTNG_ERR_KERN_NA
;
2560 /* Deny register consumer if we already have a spawned consumer. */
2561 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2562 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2563 if (kconsumer_data
.pid
> 0) {
2564 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2565 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2568 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2572 * Check for command that don't needs to allocate a returned payload. We do
2573 * this here so we don't have to make the call for no payload at each
2576 switch(cmd_ctx
->lsm
->cmd_type
) {
2577 case LTTNG_LIST_SESSIONS
:
2578 case LTTNG_LIST_TRACEPOINTS
:
2579 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2580 case LTTNG_LIST_DOMAINS
:
2581 case LTTNG_LIST_CHANNELS
:
2582 case LTTNG_LIST_EVENTS
:
2585 /* Setup lttng message with no payload */
2586 ret
= setup_lttng_msg(cmd_ctx
, 0);
2588 /* This label does not try to unlock the session */
2589 goto init_setup_error
;
2593 /* Commands that DO NOT need a session. */
2594 switch (cmd_ctx
->lsm
->cmd_type
) {
2595 case LTTNG_CREATE_SESSION
:
2596 case LTTNG_CALIBRATE
:
2597 case LTTNG_LIST_SESSIONS
:
2598 case LTTNG_LIST_TRACEPOINTS
:
2599 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2600 need_tracing_session
= 0;
2603 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2605 * We keep the session list lock across _all_ commands
2606 * for now, because the per-session lock does not
2607 * handle teardown properly.
2609 session_lock_list();
2610 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2611 if (cmd_ctx
->session
== NULL
) {
2612 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2615 /* Acquire lock for the session */
2616 session_lock(cmd_ctx
->session
);
2626 * Check domain type for specific "pre-action".
2628 switch (cmd_ctx
->lsm
->domain
.type
) {
2629 case LTTNG_DOMAIN_KERNEL
:
2631 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2635 /* Kernel tracer check */
2636 if (kernel_tracer_fd
== -1) {
2637 /* Basically, load kernel tracer modules */
2638 ret
= init_kernel_tracer();
2644 /* Consumer is in an ERROR state. Report back to client */
2645 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2646 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2650 /* Need a session for kernel command */
2651 if (need_tracing_session
) {
2652 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2653 ret
= create_kernel_session(cmd_ctx
->session
);
2655 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2660 /* Start the kernel consumer daemon */
2661 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2662 if (kconsumer_data
.pid
== 0 &&
2663 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2664 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2665 ret
= start_consumerd(&kconsumer_data
);
2667 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2670 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2672 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2676 * The consumer was just spawned so we need to add the socket to
2677 * the consumer output of the session if exist.
2679 ret
= consumer_create_socket(&kconsumer_data
,
2680 cmd_ctx
->session
->kernel_session
->consumer
);
2687 case LTTNG_DOMAIN_UST
:
2689 /* Consumer is in an ERROR state. Report back to client */
2690 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2691 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2695 if (need_tracing_session
) {
2696 /* Create UST session if none exist. */
2697 if (cmd_ctx
->session
->ust_session
== NULL
) {
2698 ret
= create_ust_session(cmd_ctx
->session
,
2699 &cmd_ctx
->lsm
->domain
);
2700 if (ret
!= LTTNG_OK
) {
2705 /* Start the UST consumer daemons */
2707 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2708 if (consumerd64_bin
[0] != '\0' &&
2709 ustconsumer64_data
.pid
== 0 &&
2710 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2711 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2712 ret
= start_consumerd(&ustconsumer64_data
);
2714 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2715 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2719 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2720 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2722 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2726 * Setup socket for consumer 64 bit. No need for atomic access
2727 * since it was set above and can ONLY be set in this thread.
2729 ret
= consumer_create_socket(&ustconsumer64_data
,
2730 cmd_ctx
->session
->ust_session
->consumer
);
2736 if (consumerd32_bin
[0] != '\0' &&
2737 ustconsumer32_data
.pid
== 0 &&
2738 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2739 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2740 ret
= start_consumerd(&ustconsumer32_data
);
2742 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2743 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2747 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2748 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2750 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2754 * Setup socket for consumer 64 bit. No need for atomic access
2755 * since it was set above and can ONLY be set in this thread.
2757 ret
= consumer_create_socket(&ustconsumer32_data
,
2758 cmd_ctx
->session
->ust_session
->consumer
);
2770 /* Validate consumer daemon state when start/stop trace command */
2771 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2772 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2773 switch (cmd_ctx
->lsm
->domain
.type
) {
2774 case LTTNG_DOMAIN_UST
:
2775 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2776 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2780 case LTTNG_DOMAIN_KERNEL
:
2781 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2782 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2790 * Check that the UID or GID match that of the tracing session.
2791 * The root user can interact with all sessions.
2793 if (need_tracing_session
) {
2794 if (!session_access_ok(cmd_ctx
->session
,
2795 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2796 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2797 ret
= LTTNG_ERR_EPERM
;
2803 * Send relayd information to consumer as soon as we have a domain and a
2806 if (cmd_ctx
->session
&& need_domain
) {
2808 * Setup relayd if not done yet. If the relayd information was already
2809 * sent to the consumer, this call will gracefully return.
2811 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2812 if (ret
!= LTTNG_OK
) {
2817 /* Process by command type */
2818 switch (cmd_ctx
->lsm
->cmd_type
) {
2819 case LTTNG_ADD_CONTEXT
:
2821 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2822 cmd_ctx
->lsm
->u
.context
.channel_name
,
2823 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2826 case LTTNG_DISABLE_CHANNEL
:
2828 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2829 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2832 case LTTNG_DISABLE_EVENT
:
2834 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2835 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2836 cmd_ctx
->lsm
->u
.disable
.name
);
2839 case LTTNG_DISABLE_ALL_EVENT
:
2841 DBG("Disabling all events");
2843 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2844 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2847 case LTTNG_ENABLE_CHANNEL
:
2849 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2850 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2853 case LTTNG_ENABLE_EVENT
:
2855 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2856 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2857 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2860 case LTTNG_ENABLE_ALL_EVENT
:
2862 DBG("Enabling all events");
2864 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2865 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2866 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2869 case LTTNG_LIST_TRACEPOINTS
:
2871 struct lttng_event
*events
;
2874 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2875 if (nb_events
< 0) {
2876 /* Return value is a negative lttng_error_code. */
2882 * Setup lttng message with payload size set to the event list size in
2883 * bytes and then copy list into the llm payload.
2885 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2891 /* Copy event list into message payload */
2892 memcpy(cmd_ctx
->llm
->payload
, events
,
2893 sizeof(struct lttng_event
) * nb_events
);
2900 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2902 struct lttng_event_field
*fields
;
2905 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2907 if (nb_fields
< 0) {
2908 /* Return value is a negative lttng_error_code. */
2914 * Setup lttng message with payload size set to the event list size in
2915 * bytes and then copy list into the llm payload.
2917 ret
= setup_lttng_msg(cmd_ctx
,
2918 sizeof(struct lttng_event_field
) * nb_fields
);
2924 /* Copy event list into message payload */
2925 memcpy(cmd_ctx
->llm
->payload
, fields
,
2926 sizeof(struct lttng_event_field
) * nb_fields
);
2933 case LTTNG_SET_CONSUMER_URI
:
2936 struct lttng_uri
*uris
;
2938 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2939 len
= nb_uri
* sizeof(struct lttng_uri
);
2942 ret
= LTTNG_ERR_INVALID
;
2946 uris
= zmalloc(len
);
2948 ret
= LTTNG_ERR_FATAL
;
2952 /* Receive variable len data */
2953 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2954 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2956 DBG("No URIs received from client... continuing");
2958 ret
= LTTNG_ERR_SESSION_FAIL
;
2963 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2965 if (ret
!= LTTNG_OK
) {
2971 * XXX: 0 means that this URI should be applied on the session. Should
2972 * be a DOMAIN enuam.
2974 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2975 /* Add the URI for the UST session if a consumer is present. */
2976 if (cmd_ctx
->session
->ust_session
&&
2977 cmd_ctx
->session
->ust_session
->consumer
) {
2978 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2980 } else if (cmd_ctx
->session
->kernel_session
&&
2981 cmd_ctx
->session
->kernel_session
->consumer
) {
2982 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2983 cmd_ctx
->session
, nb_uri
, uris
);
2991 case LTTNG_START_TRACE
:
2993 ret
= cmd_start_trace(cmd_ctx
->session
);
2996 case LTTNG_STOP_TRACE
:
2998 ret
= cmd_stop_trace(cmd_ctx
->session
);
3001 case LTTNG_CREATE_SESSION
:
3004 struct lttng_uri
*uris
= NULL
;
3006 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3007 len
= nb_uri
* sizeof(struct lttng_uri
);
3010 uris
= zmalloc(len
);
3012 ret
= LTTNG_ERR_FATAL
;
3016 /* Receive variable len data */
3017 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3018 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3020 DBG("No URIs received from client... continuing");
3022 ret
= LTTNG_ERR_SESSION_FAIL
;
3027 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3028 DBG("Creating session with ONE network URI is a bad call");
3029 ret
= LTTNG_ERR_SESSION_FAIL
;
3035 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3042 case LTTNG_DESTROY_SESSION
:
3044 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3046 /* Set session to NULL so we do not unlock it after free. */
3047 cmd_ctx
->session
= NULL
;
3050 case LTTNG_LIST_DOMAINS
:
3053 struct lttng_domain
*domains
;
3055 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3057 /* Return value is a negative lttng_error_code. */
3062 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3068 /* Copy event list into message payload */
3069 memcpy(cmd_ctx
->llm
->payload
, domains
,
3070 nb_dom
* sizeof(struct lttng_domain
));
3077 case LTTNG_LIST_CHANNELS
:
3080 struct lttng_channel
*channels
;
3082 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3083 cmd_ctx
->session
, &channels
);
3085 /* Return value is a negative lttng_error_code. */
3090 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3096 /* Copy event list into message payload */
3097 memcpy(cmd_ctx
->llm
->payload
, channels
,
3098 nb_chan
* sizeof(struct lttng_channel
));
3105 case LTTNG_LIST_EVENTS
:
3108 struct lttng_event
*events
= NULL
;
3110 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3111 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3113 /* Return value is a negative lttng_error_code. */
3118 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3124 /* Copy event list into message payload */
3125 memcpy(cmd_ctx
->llm
->payload
, events
,
3126 nb_event
* sizeof(struct lttng_event
));
3133 case LTTNG_LIST_SESSIONS
:
3135 unsigned int nr_sessions
;
3137 session_lock_list();
3138 nr_sessions
= lttng_sessions_count(
3139 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3140 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3142 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3144 session_unlock_list();
3148 /* Filled the session array */
3149 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3150 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3151 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3153 session_unlock_list();
3158 case LTTNG_CALIBRATE
:
3160 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3161 &cmd_ctx
->lsm
->u
.calibrate
);
3164 case LTTNG_REGISTER_CONSUMER
:
3166 struct consumer_data
*cdata
;
3168 switch (cmd_ctx
->lsm
->domain
.type
) {
3169 case LTTNG_DOMAIN_KERNEL
:
3170 cdata
= &kconsumer_data
;
3173 ret
= LTTNG_ERR_UND
;
3177 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3178 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3181 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3183 struct lttng_filter_bytecode
*bytecode
;
3185 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3186 ret
= LTTNG_ERR_FILTER_INVAL
;
3189 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3190 ret
= LTTNG_ERR_FILTER_INVAL
;
3193 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3195 ret
= LTTNG_ERR_FILTER_NOMEM
;
3198 /* Receive var. len. data */
3199 DBG("Receiving var len data from client ...");
3200 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3201 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3203 DBG("Nothing recv() from client var len data... continuing");
3205 ret
= LTTNG_ERR_FILTER_INVAL
;
3209 if (bytecode
->len
+ sizeof(*bytecode
)
3210 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3212 ret
= LTTNG_ERR_FILTER_INVAL
;
3216 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3217 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3218 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3221 case LTTNG_DATA_PENDING
:
3223 ret
= cmd_data_pending(cmd_ctx
->session
);
3227 ret
= LTTNG_ERR_UND
;
3232 if (cmd_ctx
->llm
== NULL
) {
3233 DBG("Missing llm structure. Allocating one.");
3234 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3238 /* Set return code */
3239 cmd_ctx
->llm
->ret_code
= ret
;
3241 if (cmd_ctx
->session
) {
3242 session_unlock(cmd_ctx
->session
);
3244 if (need_tracing_session
) {
3245 session_unlock_list();
3252 * Thread managing health check socket.
3254 static void *thread_manage_health(void *data
)
3256 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3257 uint32_t revents
, nb_fd
;
3258 struct lttng_poll_event events
;
3259 struct lttcomm_health_msg msg
;
3260 struct lttcomm_health_data reply
;
3262 DBG("[thread] Manage health check started");
3264 rcu_register_thread();
3266 /* We might hit an error path before this is created. */
3267 lttng_poll_init(&events
);
3269 /* Create unix socket */
3270 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3272 ERR("Unable to create health check Unix socket");
3278 * Set the CLOEXEC flag. Return code is useless because either way, the
3281 (void) utils_set_fd_cloexec(sock
);
3283 ret
= lttcomm_listen_unix_sock(sock
);
3289 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3290 * more will be added to this poll set.
3292 ret
= sessiond_set_thread_pollset(&events
, 2);
3297 /* Add the application registration socket */
3298 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3304 DBG("Health check ready");
3306 /* Inifinite blocking call, waiting for transmission */
3308 ret
= lttng_poll_wait(&events
, -1);
3311 * Restart interrupted system call.
3313 if (errno
== EINTR
) {
3321 for (i
= 0; i
< nb_fd
; i
++) {
3322 /* Fetch once the poll data */
3323 revents
= LTTNG_POLL_GETEV(&events
, i
);
3324 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3326 /* Thread quit pipe has been closed. Killing thread. */
3327 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3333 /* Event on the registration socket */
3334 if (pollfd
== sock
) {
3335 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3336 ERR("Health socket poll error");
3342 new_sock
= lttcomm_accept_unix_sock(sock
);
3348 * Set the CLOEXEC flag. Return code is useless because either way, the
3351 (void) utils_set_fd_cloexec(new_sock
);
3353 DBG("Receiving data from client for health...");
3354 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3356 DBG("Nothing recv() from client... continuing");
3357 ret
= close(new_sock
);
3365 rcu_thread_online();
3367 switch (msg
.component
) {
3368 case LTTNG_HEALTH_CMD
:
3369 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3371 case LTTNG_HEALTH_APP_MANAGE
:
3372 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3374 case LTTNG_HEALTH_APP_REG
:
3375 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3377 case LTTNG_HEALTH_KERNEL
:
3378 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3380 case LTTNG_HEALTH_CONSUMER
:
3381 reply
.ret_code
= check_consumer_health();
3383 case LTTNG_HEALTH_HT_CLEANUP
:
3384 reply
.ret_code
= health_check_state(HEALTH_TYPE_HT_CLEANUP
);
3386 case LTTNG_HEALTH_APP_MANAGE_NOTIFY
:
3387 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3389 case LTTNG_HEALTH_APP_REG_DISPATCH
:
3390 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3392 case LTTNG_HEALTH_ALL
:
3394 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3395 health_check_state(HEALTH_TYPE_APP_REG
) &&
3396 health_check_state(HEALTH_TYPE_CMD
) &&
3397 health_check_state(HEALTH_TYPE_KERNEL
) &&
3398 check_consumer_health() &&
3399 health_check_state(HEALTH_TYPE_HT_CLEANUP
) &&
3400 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
) &&
3401 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3404 reply
.ret_code
= LTTNG_ERR_UND
;
3409 * Flip ret value since 0 is a success and 1 indicates a bad health for
3410 * the client where in the sessiond it is the opposite. Again, this is
3411 * just to make things easier for us poor developer which enjoy a lot
3414 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3415 reply
.ret_code
= !reply
.ret_code
;
3418 DBG2("Health check return value %d", reply
.ret_code
);
3420 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3422 ERR("Failed to send health data back to client");
3425 /* End of transmission */
3426 ret
= close(new_sock
);
3436 ERR("Health error occurred in %s", __func__
);
3438 DBG("Health check thread dying");
3439 unlink(health_unix_sock_path
);
3447 lttng_poll_clean(&events
);
3449 rcu_unregister_thread();
3454 * This thread manage all clients request using the unix client socket for
3457 static void *thread_manage_clients(void *data
)
3459 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3461 uint32_t revents
, nb_fd
;
3462 struct command_ctx
*cmd_ctx
= NULL
;
3463 struct lttng_poll_event events
;
3465 DBG("[thread] Manage client started");
3467 rcu_register_thread();
3469 health_register(HEALTH_TYPE_CMD
);
3471 if (testpoint(thread_manage_clients
)) {
3472 goto error_testpoint
;
3475 health_code_update();
3477 ret
= lttcomm_listen_unix_sock(client_sock
);
3483 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3484 * more will be added to this poll set.
3486 ret
= sessiond_set_thread_pollset(&events
, 2);
3488 goto error_create_poll
;
3491 /* Add the application registration socket */
3492 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3498 * Notify parent pid that we are ready to accept command for client side.
3500 if (opt_sig_parent
) {
3501 kill(ppid
, SIGUSR1
);
3504 if (testpoint(thread_manage_clients_before_loop
)) {
3508 health_code_update();
3511 DBG("Accepting client command ...");
3513 /* Inifinite blocking call, waiting for transmission */
3515 health_poll_entry();
3516 ret
= lttng_poll_wait(&events
, -1);
3520 * Restart interrupted system call.
3522 if (errno
== EINTR
) {
3530 for (i
= 0; i
< nb_fd
; i
++) {
3531 /* Fetch once the poll data */
3532 revents
= LTTNG_POLL_GETEV(&events
, i
);
3533 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3535 health_code_update();
3537 /* Thread quit pipe has been closed. Killing thread. */
3538 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3544 /* Event on the registration socket */
3545 if (pollfd
== client_sock
) {
3546 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3547 ERR("Client socket poll error");
3553 DBG("Wait for client response");
3555 health_code_update();
3557 sock
= lttcomm_accept_unix_sock(client_sock
);
3563 * Set the CLOEXEC flag. Return code is useless because either way, the
3566 (void) utils_set_fd_cloexec(sock
);
3568 /* Set socket option for credentials retrieval */
3569 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3574 /* Allocate context command to process the client request */
3575 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3576 if (cmd_ctx
== NULL
) {
3577 PERROR("zmalloc cmd_ctx");
3581 /* Allocate data buffer for reception */
3582 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3583 if (cmd_ctx
->lsm
== NULL
) {
3584 PERROR("zmalloc cmd_ctx->lsm");
3588 cmd_ctx
->llm
= NULL
;
3589 cmd_ctx
->session
= NULL
;
3591 health_code_update();
3594 * Data is received from the lttng client. The struct
3595 * lttcomm_session_msg (lsm) contains the command and data request of
3598 DBG("Receiving data from client ...");
3599 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3600 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3602 DBG("Nothing recv() from client... continuing");
3608 clean_command_ctx(&cmd_ctx
);
3612 health_code_update();
3614 // TODO: Validate cmd_ctx including sanity check for
3615 // security purpose.
3617 rcu_thread_online();
3619 * This function dispatch the work to the kernel or userspace tracer
3620 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3621 * informations for the client. The command context struct contains
3622 * everything this function may needs.
3624 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3625 rcu_thread_offline();
3633 * TODO: Inform client somehow of the fatal error. At
3634 * this point, ret < 0 means that a zmalloc failed
3635 * (ENOMEM). Error detected but still accept
3636 * command, unless a socket error has been
3639 clean_command_ctx(&cmd_ctx
);
3643 health_code_update();
3645 DBG("Sending response (size: %d, retcode: %s)",
3646 cmd_ctx
->lttng_msg_size
,
3647 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3648 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3650 ERR("Failed to send data back to client");
3653 /* End of transmission */
3660 clean_command_ctx(&cmd_ctx
);
3662 health_code_update();
3674 lttng_poll_clean(&events
);
3675 clean_command_ctx(&cmd_ctx
);
3680 unlink(client_unix_sock_path
);
3681 if (client_sock
>= 0) {
3682 ret
= close(client_sock
);
3690 ERR("Health error occurred in %s", __func__
);
3693 health_unregister();
3695 DBG("Client thread dying");
3697 rcu_unregister_thread();
3703 * usage function on stderr
3705 static void usage(void)
3707 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3708 fprintf(stderr
, " -h, --help Display this usage.\n");
3709 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3710 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3711 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3712 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3713 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3714 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3715 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3716 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3717 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3718 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3719 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3720 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3721 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3722 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3723 fprintf(stderr
, " -V, --version Show version number.\n");
3724 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3725 fprintf(stderr
, " -q, --quiet No output at all.\n");
3726 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3727 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3728 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3729 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3733 * daemon argument parsing
3735 static int parse_args(int argc
, char **argv
)
3739 static struct option long_options
[] = {
3740 { "client-sock", 1, 0, 'c' },
3741 { "apps-sock", 1, 0, 'a' },
3742 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3743 { "kconsumerd-err-sock", 1, 0, 'E' },
3744 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3745 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3746 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3747 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3748 { "consumerd32-path", 1, 0, 'u' },
3749 { "consumerd32-libdir", 1, 0, 'U' },
3750 { "consumerd64-path", 1, 0, 't' },
3751 { "consumerd64-libdir", 1, 0, 'T' },
3752 { "daemonize", 0, 0, 'd' },
3753 { "sig-parent", 0, 0, 'S' },
3754 { "help", 0, 0, 'h' },
3755 { "group", 1, 0, 'g' },
3756 { "version", 0, 0, 'V' },
3757 { "quiet", 0, 0, 'q' },
3758 { "verbose", 0, 0, 'v' },
3759 { "verbose-consumer", 0, 0, 'Z' },
3760 { "no-kernel", 0, 0, 'N' },
3761 { "pidfile", 1, 0, 'p' },
3766 int option_index
= 0;
3767 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3768 long_options
, &option_index
);
3775 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3777 fprintf(stderr
, " with arg %s\n", optarg
);
3781 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3784 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3790 opt_tracing_group
= optarg
;
3796 fprintf(stdout
, "%s\n", VERSION
);
3802 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3805 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3808 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3811 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3814 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3817 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3823 lttng_opt_quiet
= 1;
3826 /* Verbose level can increase using multiple -v */
3827 lttng_opt_verbose
+= 1;
3830 opt_verbose_consumer
+= 1;
3833 consumerd32_bin
= optarg
;
3836 consumerd32_libdir
= optarg
;
3839 consumerd64_bin
= optarg
;
3842 consumerd64_libdir
= optarg
;
3845 opt_pidfile
= optarg
;
3848 /* Unknown option or other error.
3849 * Error is printed by getopt, just return */
3858 * Creates the two needed socket by the daemon.
3859 * apps_sock - The communication socket for all UST apps.
3860 * client_sock - The communication of the cli tool (lttng).
3862 static int init_daemon_socket(void)
3867 old_umask
= umask(0);
3869 /* Create client tool unix socket */
3870 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3871 if (client_sock
< 0) {
3872 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3877 /* Set the cloexec flag */
3878 ret
= utils_set_fd_cloexec(client_sock
);
3880 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3881 "Continuing but note that the consumer daemon will have a "
3882 "reference to this socket on exec()", client_sock
);
3885 /* File permission MUST be 660 */
3886 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3888 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3893 /* Create the application unix socket */
3894 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3895 if (apps_sock
< 0) {
3896 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3901 /* Set the cloexec flag */
3902 ret
= utils_set_fd_cloexec(apps_sock
);
3904 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3905 "Continuing but note that the consumer daemon will have a "
3906 "reference to this socket on exec()", apps_sock
);
3909 /* File permission MUST be 666 */
3910 ret
= chmod(apps_unix_sock_path
,
3911 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3913 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3918 DBG3("Session daemon client socket %d and application socket %d created",
3919 client_sock
, apps_sock
);
3927 * Check if the global socket is available, and if a daemon is answering at the
3928 * other side. If yes, error is returned.
3930 static int check_existing_daemon(void)
3932 /* Is there anybody out there ? */
3933 if (lttng_session_daemon_alive()) {
3941 * Set the tracing group gid onto the client socket.
3943 * Race window between mkdir and chown is OK because we are going from more
3944 * permissive (root.root) to less permissive (root.tracing).
3946 static int set_permissions(char *rundir
)
3951 ret
= allowed_group();
3953 WARN("No tracing group detected");
3960 /* Set lttng run dir */
3961 ret
= chown(rundir
, 0, gid
);
3963 ERR("Unable to set group on %s", rundir
);
3967 /* Ensure tracing group can search the run dir */
3968 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3970 ERR("Unable to set permissions on %s", rundir
);
3974 /* lttng client socket path */
3975 ret
= chown(client_unix_sock_path
, 0, gid
);
3977 ERR("Unable to set group on %s", client_unix_sock_path
);
3981 /* kconsumer error socket path */
3982 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3984 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3988 /* 64-bit ustconsumer error socket path */
3989 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3991 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3995 /* 32-bit ustconsumer compat32 error socket path */
3996 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3998 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4002 DBG("All permissions are set");
4009 * Create the lttng run directory needed for all global sockets and pipe.
4011 static int create_lttng_rundir(const char *rundir
)
4015 DBG3("Creating LTTng run directory: %s", rundir
);
4017 ret
= mkdir(rundir
, S_IRWXU
);
4019 if (errno
!= EEXIST
) {
4020 ERR("Unable to create %s", rundir
);
4032 * Setup sockets and directory needed by the kconsumerd communication with the
4035 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4039 char path
[PATH_MAX
];
4041 switch (consumer_data
->type
) {
4042 case LTTNG_CONSUMER_KERNEL
:
4043 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4045 case LTTNG_CONSUMER64_UST
:
4046 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4048 case LTTNG_CONSUMER32_UST
:
4049 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4052 ERR("Consumer type unknown");
4057 DBG2("Creating consumer directory: %s", path
);
4059 ret
= mkdir(path
, S_IRWXU
);
4061 if (errno
!= EEXIST
) {
4063 ERR("Failed to create %s", path
);
4069 /* Create the kconsumerd error unix socket */
4070 consumer_data
->err_sock
=
4071 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4072 if (consumer_data
->err_sock
< 0) {
4073 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4078 /* File permission MUST be 660 */
4079 ret
= chmod(consumer_data
->err_unix_sock_path
,
4080 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4082 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4092 * Signal handler for the daemon
4094 * Simply stop all worker threads, leaving main() return gracefully after
4095 * joining all threads and calling cleanup().
4097 static void sighandler(int sig
)
4101 DBG("SIGPIPE caught");
4104 DBG("SIGINT caught");
4108 DBG("SIGTERM caught");
4117 * Setup signal handler for :
4118 * SIGINT, SIGTERM, SIGPIPE
4120 static int set_signal_handler(void)
4123 struct sigaction sa
;
4126 if ((ret
= sigemptyset(&sigset
)) < 0) {
4127 PERROR("sigemptyset");
4131 sa
.sa_handler
= sighandler
;
4132 sa
.sa_mask
= sigset
;
4134 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4135 PERROR("sigaction");
4139 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4140 PERROR("sigaction");
4144 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4145 PERROR("sigaction");
4149 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4155 * Set open files limit to unlimited. This daemon can open a large number of
4156 * file descriptors in order to consumer multiple kernel traces.
4158 static void set_ulimit(void)
4163 /* The kernel does not allowed an infinite limit for open files */
4164 lim
.rlim_cur
= 65535;
4165 lim
.rlim_max
= 65535;
4167 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4169 PERROR("failed to set open files limit");
4174 * Write pidfile using the rundir and opt_pidfile.
4176 static void write_pidfile(void)
4179 char pidfile_path
[PATH_MAX
];
4184 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4186 /* Build pidfile path from rundir and opt_pidfile. */
4187 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4188 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4190 PERROR("snprintf pidfile path");
4196 * Create pid file in rundir. Return value is of no importance. The
4197 * execution will continue even though we are not able to write the file.
4199 (void) utils_create_pid_file(getpid(), pidfile_path
);
4208 int main(int argc
, char **argv
)
4212 const char *home_path
, *env_app_timeout
;
4214 init_kernel_workarounds();
4216 rcu_register_thread();
4218 setup_consumerd_path();
4220 page_size
= sysconf(_SC_PAGESIZE
);
4221 if (page_size
< 0) {
4222 PERROR("sysconf _SC_PAGESIZE");
4223 page_size
= LONG_MAX
;
4224 WARN("Fallback page size to %ld", page_size
);
4227 /* Parse arguments */
4229 if ((ret
= parse_args(argc
, argv
)) < 0) {
4239 * child: setsid, close FD 0, 1, 2, chdir /
4240 * parent: exit (if fork is successful)
4248 * We are in the child. Make sure all other file
4249 * descriptors are closed, in case we are called with
4250 * more opened file descriptors than the standard ones.
4252 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4257 /* Create thread quit pipe */
4258 if ((ret
= init_thread_quit_pipe()) < 0) {
4262 /* Check if daemon is UID = 0 */
4263 is_root
= !getuid();
4266 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4268 /* Create global run dir with root access */
4269 ret
= create_lttng_rundir(rundir
);
4274 if (strlen(apps_unix_sock_path
) == 0) {
4275 snprintf(apps_unix_sock_path
, PATH_MAX
,
4276 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4279 if (strlen(client_unix_sock_path
) == 0) {
4280 snprintf(client_unix_sock_path
, PATH_MAX
,
4281 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4284 /* Set global SHM for ust */
4285 if (strlen(wait_shm_path
) == 0) {
4286 snprintf(wait_shm_path
, PATH_MAX
,
4287 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4290 if (strlen(health_unix_sock_path
) == 0) {
4291 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4292 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4295 /* Setup kernel consumerd path */
4296 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4297 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4298 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4299 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4301 DBG2("Kernel consumer err path: %s",
4302 kconsumer_data
.err_unix_sock_path
);
4303 DBG2("Kernel consumer cmd path: %s",
4304 kconsumer_data
.cmd_unix_sock_path
);
4306 home_path
= utils_get_home_dir();
4307 if (home_path
== NULL
) {
4308 /* TODO: Add --socket PATH option */
4309 ERR("Can't get HOME directory for sockets creation.");
4315 * Create rundir from home path. This will create something like
4318 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4324 ret
= create_lttng_rundir(rundir
);
4329 if (strlen(apps_unix_sock_path
) == 0) {
4330 snprintf(apps_unix_sock_path
, PATH_MAX
,
4331 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4334 /* Set the cli tool unix socket path */
4335 if (strlen(client_unix_sock_path
) == 0) {
4336 snprintf(client_unix_sock_path
, PATH_MAX
,
4337 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4340 /* Set global SHM for ust */
4341 if (strlen(wait_shm_path
) == 0) {
4342 snprintf(wait_shm_path
, PATH_MAX
,
4343 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4346 /* Set health check Unix path */
4347 if (strlen(health_unix_sock_path
) == 0) {
4348 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4349 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4353 /* Set consumer initial state */
4354 kernel_consumerd_state
= CONSUMER_STOPPED
;
4355 ust_consumerd_state
= CONSUMER_STOPPED
;
4357 DBG("Client socket path %s", client_unix_sock_path
);
4358 DBG("Application socket path %s", apps_unix_sock_path
);
4359 DBG("Application wait path %s", wait_shm_path
);
4360 DBG("LTTng run directory path: %s", rundir
);
4362 /* 32 bits consumerd path setup */
4363 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4364 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4365 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4366 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4368 DBG2("UST consumer 32 bits err path: %s",
4369 ustconsumer32_data
.err_unix_sock_path
);
4370 DBG2("UST consumer 32 bits cmd path: %s",
4371 ustconsumer32_data
.cmd_unix_sock_path
);
4373 /* 64 bits consumerd path setup */
4374 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4375 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4376 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4377 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4379 DBG2("UST consumer 64 bits err path: %s",
4380 ustconsumer64_data
.err_unix_sock_path
);
4381 DBG2("UST consumer 64 bits cmd path: %s",
4382 ustconsumer64_data
.cmd_unix_sock_path
);
4385 * See if daemon already exist.
4387 if ((ret
= check_existing_daemon()) < 0) {
4388 ERR("Already running daemon.\n");
4390 * We do not goto exit because we must not cleanup()
4391 * because a daemon is already running.
4397 * Init UST app hash table. Alloc hash table before this point since
4398 * cleanup() can get called after that point.
4402 /* After this point, we can safely call cleanup() with "goto exit" */
4405 * These actions must be executed as root. We do that *after* setting up
4406 * the sockets path because we MUST make the check for another daemon using
4407 * those paths *before* trying to set the kernel consumer sockets and init
4411 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4416 /* Setup kernel tracer */
4417 if (!opt_no_kernel
) {
4418 init_kernel_tracer();
4421 /* Set ulimit for open files */
4424 /* init lttng_fd tracking must be done after set_ulimit. */
4427 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4432 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4437 if ((ret
= set_signal_handler()) < 0) {
4441 /* Setup the needed unix socket */
4442 if ((ret
= init_daemon_socket()) < 0) {
4446 /* Set credentials to socket */
4447 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4451 /* Get parent pid if -S, --sig-parent is specified. */
4452 if (opt_sig_parent
) {
4456 /* Setup the kernel pipe for waking up the kernel thread */
4457 if (is_root
&& !opt_no_kernel
) {
4458 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4463 /* Setup the thread ht_cleanup communication pipe. */
4464 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4468 /* Setup the thread apps communication pipe. */
4469 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4473 /* Setup the thread apps notify communication pipe. */
4474 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4478 /* Initialize global buffer per UID and PID registry. */
4479 buffer_reg_init_uid_registry();
4480 buffer_reg_init_pid_registry();
4482 /* Init UST command queue. */
4483 cds_wfq_init(&ust_cmd_queue
.queue
);
4486 * Get session list pointer. This pointer MUST NOT be free(). This list is
4487 * statically declared in session.c
4489 session_list_ptr
= session_get_list();
4491 /* Set up max poll set size */
4492 lttng_poll_set_max_size();
4496 /* Check for the application socket timeout env variable. */
4497 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4498 if (env_app_timeout
) {
4499 app_socket_timeout
= atoi(env_app_timeout
);
4501 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4506 /* Create thread to manage the client socket */
4507 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4508 thread_ht_cleanup
, (void *) NULL
);
4510 PERROR("pthread_create ht_cleanup");
4511 goto exit_ht_cleanup
;
4514 /* Create thread to manage the client socket */
4515 ret
= pthread_create(&health_thread
, NULL
,
4516 thread_manage_health
, (void *) NULL
);
4518 PERROR("pthread_create health");
4522 /* Create thread to manage the client socket */
4523 ret
= pthread_create(&client_thread
, NULL
,
4524 thread_manage_clients
, (void *) NULL
);
4526 PERROR("pthread_create clients");
4530 /* Create thread to dispatch registration */
4531 ret
= pthread_create(&dispatch_thread
, NULL
,
4532 thread_dispatch_ust_registration
, (void *) NULL
);
4534 PERROR("pthread_create dispatch");
4538 /* Create thread to manage application registration. */
4539 ret
= pthread_create(®_apps_thread
, NULL
,
4540 thread_registration_apps
, (void *) NULL
);
4542 PERROR("pthread_create registration");
4546 /* Create thread to manage application socket */
4547 ret
= pthread_create(&apps_thread
, NULL
,
4548 thread_manage_apps
, (void *) NULL
);
4550 PERROR("pthread_create apps");
4554 /* Create thread to manage application notify socket */
4555 ret
= pthread_create(&apps_notify_thread
, NULL
,
4556 ust_thread_manage_notify
, (void *) NULL
);
4558 PERROR("pthread_create apps");
4562 /* Don't start this thread if kernel tracing is not requested nor root */
4563 if (is_root
&& !opt_no_kernel
) {
4564 /* Create kernel thread to manage kernel event */
4565 ret
= pthread_create(&kernel_thread
, NULL
,
4566 thread_manage_kernel
, (void *) NULL
);
4568 PERROR("pthread_create kernel");
4572 ret
= pthread_join(kernel_thread
, &status
);
4574 PERROR("pthread_join");
4575 goto error
; /* join error, exit without cleanup */
4580 ret
= pthread_join(apps_thread
, &status
);
4582 PERROR("pthread_join");
4583 goto error
; /* join error, exit without cleanup */
4587 ret
= pthread_join(reg_apps_thread
, &status
);
4589 PERROR("pthread_join");
4590 goto error
; /* join error, exit without cleanup */
4594 ret
= pthread_join(dispatch_thread
, &status
);
4596 PERROR("pthread_join");
4597 goto error
; /* join error, exit without cleanup */
4601 ret
= pthread_join(client_thread
, &status
);
4603 PERROR("pthread_join");
4604 goto error
; /* join error, exit without cleanup */
4607 ret
= join_consumer_thread(&kconsumer_data
);
4609 PERROR("join_consumer");
4610 goto error
; /* join error, exit without cleanup */
4613 ret
= join_consumer_thread(&ustconsumer32_data
);
4615 PERROR("join_consumer ust32");
4616 goto error
; /* join error, exit without cleanup */
4619 ret
= join_consumer_thread(&ustconsumer64_data
);
4621 PERROR("join_consumer ust64");
4622 goto error
; /* join error, exit without cleanup */
4626 ret
= pthread_join(health_thread
, &status
);
4628 PERROR("pthread_join health thread");
4629 goto error
; /* join error, exit without cleanup */
4633 ret
= pthread_join(ht_cleanup_thread
, &status
);
4635 PERROR("pthread_join ht cleanup thread");
4636 goto error
; /* join error, exit without cleanup */
4641 * cleanup() is called when no other thread is running.
4643 rcu_thread_online();
4645 rcu_thread_offline();
4646 rcu_unregister_thread();