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.
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
33 #include <sys/types.h>
35 #include <urcu/uatomic.h>
39 #include <common/common.h>
40 #include <common/compat/poll.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"
55 #include "kernel-consumer.h"
59 #include "ust-consumer.h"
63 #include "testpoint.h"
65 #define CONSUMERD_FILE "lttng-consumerd"
68 const char default_home_dir
[] = DEFAULT_HOME_DIR
;
69 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
70 const char default_ust_sock_dir
[] = DEFAULT_UST_SOCK_DIR
;
71 const char default_global_apps_pipe
[] = DEFAULT_GLOBAL_APPS_PIPE
;
74 const char *opt_tracing_group
;
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 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
94 .lock
= PTHREAD_MUTEX_INITIALIZER
,
95 .cond
= PTHREAD_COND_INITIALIZER
,
96 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
98 static struct consumer_data ustconsumer64_data
= {
99 .type
= LTTNG_CONSUMER64_UST
,
100 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
101 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
104 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
105 .lock
= PTHREAD_MUTEX_INITIALIZER
,
106 .cond
= PTHREAD_COND_INITIALIZER
,
107 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
109 static struct consumer_data ustconsumer32_data
= {
110 .type
= LTTNG_CONSUMER32_UST
,
111 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
112 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
115 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
116 .lock
= PTHREAD_MUTEX_INITIALIZER
,
117 .cond
= PTHREAD_COND_INITIALIZER
,
118 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
121 /* Shared between threads */
122 static int dispatch_thread_exit
;
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path
[PATH_MAX
];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path
[PATH_MAX
];
128 /* global wait shm path for UST */
129 static char wait_shm_path
[PATH_MAX
];
130 /* Global health check unix path */
131 static char health_unix_sock_path
[PATH_MAX
];
133 /* Sockets and FDs */
134 static int client_sock
= -1;
135 static int apps_sock
= -1;
136 int kernel_tracer_fd
= -1;
137 static int kernel_poll_pipe
[2] = { -1, -1 };
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
143 static int thread_quit_pipe
[2] = { -1, -1 };
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
149 static int apps_cmd_pipe
[2] = { -1, -1 };
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread
;
153 static pthread_t reg_apps_thread
;
154 static pthread_t client_thread
;
155 static pthread_t kernel_thread
;
156 static pthread_t dispatch_thread
;
157 static pthread_t health_thread
;
160 * UST registration command queue. This queue is tied with a futex and uses a N
161 * wakers / 1 waiter implemented and detailed in futex.c/.h
163 * The thread_manage_apps and thread_dispatch_ust_registration interact with
164 * this queue and the wait/wake scheme.
166 static struct ust_cmd_queue ust_cmd_queue
;
169 * Pointer initialized before thread creation.
171 * This points to the tracing session list containing the session count and a
172 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
173 * MUST NOT be taken if you call a public function in session.c.
175 * The lock is nested inside the structure: session_list_ptr->lock. Please use
176 * session_lock_list and session_unlock_list for lock acquisition.
178 static struct ltt_session_list
*session_list_ptr
;
180 int ust_consumerd64_fd
= -1;
181 int ust_consumerd32_fd
= -1;
183 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
184 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
185 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
186 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
188 static const char *module_proc_lttng
= "/proc/lttng";
191 * Consumer daemon state which is changed when spawning it, killing it or in
192 * case of a fatal error.
194 enum consumerd_state
{
195 CONSUMER_STARTED
= 1,
196 CONSUMER_STOPPED
= 2,
201 * This consumer daemon state is used to validate if a client command will be
202 * able to reach the consumer. If not, the client is informed. For instance,
203 * doing a "lttng start" when the consumer state is set to ERROR will return an
204 * error to the client.
206 * The following example shows a possible race condition of this scheme:
208 * consumer thread error happens
210 * client cmd checks state -> still OK
211 * consumer thread exit, sets error
212 * client cmd try to talk to consumer
215 * However, since the consumer is a different daemon, we have no way of making
216 * sure the command will reach it safely even with this state flag. This is why
217 * we consider that up to the state validation during command processing, the
218 * command is safe. After that, we can not guarantee the correctness of the
219 * client request vis-a-vis the consumer.
221 static enum consumerd_state ust_consumerd_state
;
222 static enum consumerd_state kernel_consumerd_state
;
225 * Socket timeout for receiving and sending in seconds.
227 static int app_socket_timeout
;
230 void setup_consumerd_path(void)
232 const char *bin
, *libdir
;
235 * Allow INSTALL_BIN_PATH to be used as a target path for the
236 * native architecture size consumer if CONFIG_CONSUMER*_PATH
237 * has not been defined.
239 #if (CAA_BITS_PER_LONG == 32)
240 if (!consumerd32_bin
[0]) {
241 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
243 if (!consumerd32_libdir
[0]) {
244 consumerd32_libdir
= INSTALL_LIB_PATH
;
246 #elif (CAA_BITS_PER_LONG == 64)
247 if (!consumerd64_bin
[0]) {
248 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
250 if (!consumerd64_libdir
[0]) {
251 consumerd64_libdir
= INSTALL_LIB_PATH
;
254 #error "Unknown bitness"
258 * runtime env. var. overrides the build default.
260 bin
= getenv("LTTNG_CONSUMERD32_BIN");
262 consumerd32_bin
= bin
;
264 bin
= getenv("LTTNG_CONSUMERD64_BIN");
266 consumerd64_bin
= bin
;
268 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
270 consumerd32_libdir
= libdir
;
272 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
274 consumerd64_libdir
= libdir
;
279 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
281 static int create_thread_poll_set(struct lttng_poll_event
*events
,
286 if (events
== NULL
|| size
== 0) {
291 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
297 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
309 * Check if the thread quit pipe was triggered.
311 * Return 1 if it was triggered else 0;
313 static int check_thread_quit_pipe(int fd
, uint32_t events
)
315 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
323 * Return group ID of the tracing group or -1 if not found.
325 static gid_t
allowed_group(void)
329 if (opt_tracing_group
) {
330 grp
= getgrnam(opt_tracing_group
);
332 grp
= getgrnam(default_tracing_group
);
342 * Init thread quit pipe.
344 * Return -1 on error or 0 if all pipes are created.
346 static int init_thread_quit_pipe(void)
350 ret
= pipe(thread_quit_pipe
);
352 PERROR("thread quit pipe");
356 for (i
= 0; i
< 2; i
++) {
357 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
369 * Stop all threads by closing the thread quit pipe.
371 static void stop_threads(void)
375 /* Stopping all threads */
376 DBG("Terminating all threads");
377 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
379 ERR("write error on thread quit pipe");
382 /* Dispatch thread */
383 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
384 futex_nto1_wake(&ust_cmd_queue
.futex
);
390 static void cleanup(void)
394 struct ltt_session
*sess
, *stmp
;
398 /* First thing first, stop all threads */
399 utils_close_pipe(thread_quit_pipe
);
401 DBG("Removing %s directory", rundir
);
402 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
404 ERR("asprintf failed. Something is really wrong!");
407 /* Remove lttng run directory */
410 ERR("Unable to clean %s", rundir
);
415 DBG("Cleaning up all sessions");
417 /* Destroy session list mutex */
418 if (session_list_ptr
!= NULL
) {
419 pthread_mutex_destroy(&session_list_ptr
->lock
);
421 /* Cleanup ALL session */
422 cds_list_for_each_entry_safe(sess
, stmp
,
423 &session_list_ptr
->head
, list
) {
424 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
428 DBG("Closing all UST sockets");
429 ust_app_clean_list();
431 if (is_root
&& !opt_no_kernel
) {
432 DBG2("Closing kernel fd");
433 if (kernel_tracer_fd
>= 0) {
434 ret
= close(kernel_tracer_fd
);
439 DBG("Unloading kernel modules");
440 modprobe_remove_lttng_all();
444 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
445 "Matthew, BEET driven development works!%c[%dm",
446 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
451 * Send data on a unix socket using the liblttsessiondcomm API.
453 * Return lttcomm error code.
455 static int send_unix_sock(int sock
, void *buf
, size_t len
)
457 /* Check valid length */
462 return lttcomm_send_unix_sock(sock
, buf
, len
);
466 * Free memory of a command context structure.
468 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
470 DBG("Clean command context structure");
472 if ((*cmd_ctx
)->llm
) {
473 free((*cmd_ctx
)->llm
);
475 if ((*cmd_ctx
)->lsm
) {
476 free((*cmd_ctx
)->lsm
);
484 * Notify UST applications using the shm mmap futex.
486 static int notify_ust_apps(int active
)
490 DBG("Notifying applications of session daemon state: %d", active
);
492 /* See shm.c for this call implying mmap, shm and futex calls */
493 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
494 if (wait_shm_mmap
== NULL
) {
498 /* Wake waiting process */
499 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
501 /* Apps notified successfully */
509 * Setup the outgoing data buffer for the response (llm) by allocating the
510 * right amount of memory and copying the original information from the lsm
513 * Return total size of the buffer pointed by buf.
515 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
521 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
522 if (cmd_ctx
->llm
== NULL
) {
528 /* Copy common data */
529 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
530 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
532 cmd_ctx
->llm
->data_size
= size
;
533 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
542 * Update the kernel poll set of all channel fd available over all tracing
543 * session. Add the wakeup pipe at the end of the set.
545 static int update_kernel_poll(struct lttng_poll_event
*events
)
548 struct ltt_session
*session
;
549 struct ltt_kernel_channel
*channel
;
551 DBG("Updating kernel poll set");
554 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
555 session_lock(session
);
556 if (session
->kernel_session
== NULL
) {
557 session_unlock(session
);
561 cds_list_for_each_entry(channel
,
562 &session
->kernel_session
->channel_list
.head
, list
) {
563 /* Add channel fd to the kernel poll set */
564 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
566 session_unlock(session
);
569 DBG("Channel fd %d added to kernel set", channel
->fd
);
571 session_unlock(session
);
573 session_unlock_list();
578 session_unlock_list();
583 * Find the channel fd from 'fd' over all tracing session. When found, check
584 * for new channel stream and send those stream fds to the kernel consumer.
586 * Useful for CPU hotplug feature.
588 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
591 struct ltt_session
*session
;
592 struct ltt_kernel_session
*ksess
;
593 struct ltt_kernel_channel
*channel
;
595 DBG("Updating kernel streams for channel fd %d", fd
);
598 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
599 session_lock(session
);
600 if (session
->kernel_session
== NULL
) {
601 session_unlock(session
);
604 ksess
= session
->kernel_session
;
606 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
607 if (channel
->fd
== fd
) {
608 DBG("Channel found, updating kernel streams");
609 ret
= kernel_open_channel_stream(channel
);
615 * Have we already sent fds to the consumer? If yes, it means
616 * that tracing is started so it is safe to send our updated
619 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
620 struct lttng_ht_iter iter
;
621 struct consumer_socket
*socket
;
624 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
625 &iter
.iter
, socket
, node
.node
) {
626 /* Code flow error */
627 assert(socket
->fd
>= 0);
629 pthread_mutex_lock(socket
->lock
);
630 ret
= kernel_consumer_send_channel_stream(socket
,
632 pthread_mutex_unlock(socket
->lock
);
643 session_unlock(session
);
645 session_unlock_list();
649 session_unlock(session
);
650 session_unlock_list();
655 * For each tracing session, update newly registered apps.
657 static void update_ust_app(int app_sock
)
659 struct ltt_session
*sess
, *stmp
;
663 /* For all tracing session(s) */
664 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
666 if (sess
->ust_session
) {
667 ust_app_global_update(sess
->ust_session
, app_sock
);
669 session_unlock(sess
);
672 session_unlock_list();
676 * This thread manage event coming from the kernel.
678 * Features supported in this thread:
681 static void *thread_manage_kernel(void *data
)
683 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
684 uint32_t revents
, nb_fd
;
686 struct lttng_poll_event events
;
688 DBG("[thread] Thread manage kernel started");
690 health_register(HEALTH_TYPE_KERNEL
);
693 * This first step of the while is to clean this structure which could free
694 * non NULL pointers so zero it before the loop.
696 memset(&events
, 0, sizeof(events
));
698 if (testpoint(thread_manage_kernel
)) {
699 goto error_testpoint
;
702 health_code_update();
704 if (testpoint(thread_manage_kernel_before_loop
)) {
705 goto error_testpoint
;
709 health_code_update();
711 if (update_poll_flag
== 1) {
712 /* Clean events object. We are about to populate it again. */
713 lttng_poll_clean(&events
);
715 ret
= create_thread_poll_set(&events
, 2);
717 goto error_poll_create
;
720 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
725 /* This will add the available kernel channel if any. */
726 ret
= update_kernel_poll(&events
);
730 update_poll_flag
= 0;
733 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
735 /* Poll infinite value of time */
738 ret
= lttng_poll_wait(&events
, -1);
742 * Restart interrupted system call.
744 if (errno
== EINTR
) {
748 } else if (ret
== 0) {
749 /* Should not happen since timeout is infinite */
750 ERR("Return value of poll is 0 with an infinite timeout.\n"
751 "This should not have happened! Continuing...");
757 for (i
= 0; i
< nb_fd
; i
++) {
758 /* Fetch once the poll data */
759 revents
= LTTNG_POLL_GETEV(&events
, i
);
760 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
762 health_code_update();
764 /* Thread quit pipe has been closed. Killing thread. */
765 ret
= check_thread_quit_pipe(pollfd
, revents
);
771 /* Check for data on kernel pipe */
772 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
774 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
775 } while (ret
< 0 && errno
== EINTR
);
777 * Ret value is useless here, if this pipe gets any actions an
778 * update is required anyway.
780 update_poll_flag
= 1;
784 * New CPU detected by the kernel. Adding kernel stream to
785 * kernel session and updating the kernel consumer
787 if (revents
& LPOLLIN
) {
788 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
794 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
795 * and unregister kernel stream at this point.
804 lttng_poll_clean(&events
);
807 utils_close_pipe(kernel_poll_pipe
);
808 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
811 ERR("Health error occurred in %s", __func__
);
812 WARN("Kernel thread died unexpectedly. "
813 "Kernel tracing can continue but CPU hotplug is disabled.");
816 DBG("Kernel thread dying");
821 * Signal pthread condition of the consumer data that the thread.
823 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
825 pthread_mutex_lock(&data
->cond_mutex
);
828 * The state is set before signaling. It can be any value, it's the waiter
829 * job to correctly interpret this condition variable associated to the
830 * consumer pthread_cond.
832 * A value of 0 means that the corresponding thread of the consumer data
833 * was not started. 1 indicates that the thread has started and is ready
834 * for action. A negative value means that there was an error during the
837 data
->consumer_thread_is_ready
= state
;
838 (void) pthread_cond_signal(&data
->cond
);
840 pthread_mutex_unlock(&data
->cond_mutex
);
844 * This thread manage the consumer error sent back to the session daemon.
846 static void *thread_manage_consumer(void *data
)
848 int sock
= -1, i
, ret
, pollfd
, err
= -1;
849 uint32_t revents
, nb_fd
;
850 enum lttcomm_return_code code
;
851 struct lttng_poll_event events
;
852 struct consumer_data
*consumer_data
= data
;
854 DBG("[thread] Manage consumer started");
856 health_register(HEALTH_TYPE_CONSUMER
);
858 health_code_update();
861 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
862 * Nothing more will be added to this poll set.
864 ret
= create_thread_poll_set(&events
, 2);
870 * The error socket here is already in a listening state which was done
871 * just before spawning this thread to avoid a race between the consumer
872 * daemon exec trying to connect and the listen() call.
874 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
879 health_code_update();
881 /* Inifinite blocking call, waiting for transmission */
885 if (testpoint(thread_manage_consumer
)) {
889 ret
= lttng_poll_wait(&events
, -1);
893 * Restart interrupted system call.
895 if (errno
== EINTR
) {
903 for (i
= 0; i
< nb_fd
; i
++) {
904 /* Fetch once the poll data */
905 revents
= LTTNG_POLL_GETEV(&events
, i
);
906 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
908 health_code_update();
910 /* Thread quit pipe has been closed. Killing thread. */
911 ret
= check_thread_quit_pipe(pollfd
, revents
);
917 /* Event on the registration socket */
918 if (pollfd
== consumer_data
->err_sock
) {
919 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
920 ERR("consumer err socket poll error");
926 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
932 * Set the CLOEXEC flag. Return code is useless because either way, the
935 (void) utils_set_fd_cloexec(sock
);
937 health_code_update();
939 DBG2("Receiving code from consumer err_sock");
941 /* Getting status code from kconsumerd */
942 ret
= lttcomm_recv_unix_sock(sock
, &code
,
943 sizeof(enum lttcomm_return_code
));
948 health_code_update();
950 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
951 consumer_data
->cmd_sock
=
952 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
953 if (consumer_data
->cmd_sock
< 0) {
954 /* On error, signal condition and quit. */
955 signal_consumer_condition(consumer_data
, -1);
956 PERROR("consumer connect");
959 signal_consumer_condition(consumer_data
, 1);
960 DBG("Consumer command socket ready");
962 ERR("consumer error when waiting for SOCK_READY : %s",
963 lttcomm_get_readable_code(-code
));
967 /* Remove the kconsumerd error sock since we've established a connexion */
968 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
973 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
978 health_code_update();
980 /* Inifinite blocking call, waiting for transmission */
983 ret
= lttng_poll_wait(&events
, -1);
987 * Restart interrupted system call.
989 if (errno
== EINTR
) {
997 for (i
= 0; i
< nb_fd
; i
++) {
998 /* Fetch once the poll data */
999 revents
= LTTNG_POLL_GETEV(&events
, i
);
1000 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1002 health_code_update();
1004 /* Thread quit pipe has been closed. Killing thread. */
1005 ret
= check_thread_quit_pipe(pollfd
, revents
);
1011 /* Event on the kconsumerd socket */
1012 if (pollfd
== sock
) {
1013 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1014 ERR("consumer err socket second poll error");
1020 health_code_update();
1022 /* Wait for any kconsumerd error */
1023 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1024 sizeof(enum lttcomm_return_code
));
1026 ERR("consumer closed the command socket");
1030 ERR("consumer return code : %s", lttcomm_get_readable_code(-code
));
1034 /* Immediately set the consumerd state to stopped */
1035 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1036 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1037 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1038 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1039 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1041 /* Code flow error... */
1045 if (consumer_data
->err_sock
>= 0) {
1046 ret
= close(consumer_data
->err_sock
);
1051 if (consumer_data
->cmd_sock
>= 0) {
1052 ret
= close(consumer_data
->cmd_sock
);
1064 unlink(consumer_data
->err_unix_sock_path
);
1065 unlink(consumer_data
->cmd_unix_sock_path
);
1066 consumer_data
->pid
= 0;
1068 lttng_poll_clean(&events
);
1072 ERR("Health error occurred in %s", __func__
);
1074 health_unregister();
1075 DBG("consumer thread cleanup completed");
1081 * This thread manage application communication.
1083 static void *thread_manage_apps(void *data
)
1085 int i
, ret
, pollfd
, err
= -1;
1086 uint32_t revents
, nb_fd
;
1087 struct ust_command ust_cmd
;
1088 struct lttng_poll_event events
;
1090 DBG("[thread] Manage application started");
1092 rcu_register_thread();
1093 rcu_thread_online();
1095 health_register(HEALTH_TYPE_APP_MANAGE
);
1097 if (testpoint(thread_manage_apps
)) {
1098 goto error_testpoint
;
1101 health_code_update();
1103 ret
= create_thread_poll_set(&events
, 2);
1105 goto error_poll_create
;
1108 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1113 if (testpoint(thread_manage_apps_before_loop
)) {
1117 health_code_update();
1120 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1122 /* Inifinite blocking call, waiting for transmission */
1124 health_poll_entry();
1125 ret
= lttng_poll_wait(&events
, -1);
1129 * Restart interrupted system call.
1131 if (errno
== EINTR
) {
1139 for (i
= 0; i
< nb_fd
; i
++) {
1140 /* Fetch once the poll data */
1141 revents
= LTTNG_POLL_GETEV(&events
, i
);
1142 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1144 health_code_update();
1146 /* Thread quit pipe has been closed. Killing thread. */
1147 ret
= check_thread_quit_pipe(pollfd
, revents
);
1153 /* Inspect the apps cmd pipe */
1154 if (pollfd
== apps_cmd_pipe
[0]) {
1155 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1156 ERR("Apps command pipe error");
1158 } else if (revents
& LPOLLIN
) {
1161 ret
= read(apps_cmd_pipe
[0], &ust_cmd
, sizeof(ust_cmd
));
1162 } while (ret
< 0 && errno
== EINTR
);
1163 if (ret
< 0 || ret
< sizeof(ust_cmd
)) {
1164 PERROR("read apps cmd pipe");
1168 health_code_update();
1170 /* Register applicaton to the session daemon */
1171 ret
= ust_app_register(&ust_cmd
.reg_msg
,
1173 if (ret
== -ENOMEM
) {
1175 } else if (ret
< 0) {
1179 health_code_update();
1182 * Validate UST version compatibility.
1184 ret
= ust_app_validate_version(ust_cmd
.sock
);
1187 * Add channel(s) and event(s) to newly registered apps
1188 * from lttng global UST domain.
1190 update_ust_app(ust_cmd
.sock
);
1193 health_code_update();
1195 ret
= ust_app_register_done(ust_cmd
.sock
);
1198 * If the registration is not possible, we simply
1199 * unregister the apps and continue
1201 ust_app_unregister(ust_cmd
.sock
);
1204 * We only monitor the error events of the socket. This
1205 * thread does not handle any incoming data from UST
1208 ret
= lttng_poll_add(&events
, ust_cmd
.sock
,
1209 LPOLLERR
& LPOLLHUP
& LPOLLRDHUP
);
1214 /* Set socket timeout for both receiving and ending */
1215 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd
.sock
,
1216 app_socket_timeout
);
1217 (void) lttcomm_setsockopt_snd_timeout(ust_cmd
.sock
,
1218 app_socket_timeout
);
1220 DBG("Apps with sock %d added to poll set",
1224 health_code_update();
1230 * At this point, we know that a registered application made
1231 * the event at poll_wait.
1233 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1234 /* Removing from the poll set */
1235 ret
= lttng_poll_del(&events
, pollfd
);
1240 /* Socket closed on remote end. */
1241 ust_app_unregister(pollfd
);
1246 health_code_update();
1252 lttng_poll_clean(&events
);
1255 utils_close_pipe(apps_cmd_pipe
);
1256 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1259 * We don't clean the UST app hash table here since already registered
1260 * applications can still be controlled so let them be until the session
1261 * daemon dies or the applications stop.
1266 ERR("Health error occurred in %s", __func__
);
1268 health_unregister();
1269 DBG("Application communication apps thread cleanup complete");
1270 rcu_thread_offline();
1271 rcu_unregister_thread();
1276 * Dispatch request from the registration threads to the application
1277 * communication thread.
1279 static void *thread_dispatch_ust_registration(void *data
)
1282 struct cds_wfq_node
*node
;
1283 struct ust_command
*ust_cmd
= NULL
;
1285 DBG("[thread] Dispatch UST command started");
1287 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1288 /* Atomically prepare the queue futex */
1289 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1292 /* Dequeue command for registration */
1293 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1295 DBG("Woken up but nothing in the UST command queue");
1296 /* Continue thread execution */
1300 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1302 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1303 " gid:%d sock:%d name:%s (version %d.%d)",
1304 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1305 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1306 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1307 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1309 * Inform apps thread of the new application registration. This
1310 * call is blocking so we can be assured that the data will be read
1311 * at some point in time or wait to the end of the world :)
1313 if (apps_cmd_pipe
[1] >= 0) {
1315 ret
= write(apps_cmd_pipe
[1], ust_cmd
,
1316 sizeof(struct ust_command
));
1317 } while (ret
< 0 && errno
== EINTR
);
1318 if (ret
< 0 || ret
!= sizeof(struct ust_command
)) {
1319 PERROR("write apps cmd pipe");
1320 if (errno
== EBADF
) {
1322 * We can't inform the application thread to process
1323 * registration. We will exit or else application
1324 * registration will not occur and tracing will never
1331 /* Application manager thread is not available. */
1332 ret
= close(ust_cmd
->sock
);
1334 PERROR("close ust_cmd sock");
1338 } while (node
!= NULL
);
1340 /* Futex wait on queue. Blocking call on futex() */
1341 futex_nto1_wait(&ust_cmd_queue
.futex
);
1345 DBG("Dispatch thread dying");
1350 * This thread manage application registration.
1352 static void *thread_registration_apps(void *data
)
1354 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1355 uint32_t revents
, nb_fd
;
1356 struct lttng_poll_event events
;
1358 * Get allocated in this thread, enqueued to a global queue, dequeued and
1359 * freed in the manage apps thread.
1361 struct ust_command
*ust_cmd
= NULL
;
1363 DBG("[thread] Manage application registration started");
1365 health_register(HEALTH_TYPE_APP_REG
);
1367 if (testpoint(thread_registration_apps
)) {
1368 goto error_testpoint
;
1371 ret
= lttcomm_listen_unix_sock(apps_sock
);
1377 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1378 * more will be added to this poll set.
1380 ret
= create_thread_poll_set(&events
, 2);
1382 goto error_create_poll
;
1385 /* Add the application registration socket */
1386 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1388 goto error_poll_add
;
1391 /* Notify all applications to register */
1392 ret
= notify_ust_apps(1);
1394 ERR("Failed to notify applications or create the wait shared memory.\n"
1395 "Execution continues but there might be problem for already\n"
1396 "running applications that wishes to register.");
1400 DBG("Accepting application registration");
1402 /* Inifinite blocking call, waiting for transmission */
1404 health_poll_entry();
1405 ret
= lttng_poll_wait(&events
, -1);
1409 * Restart interrupted system call.
1411 if (errno
== EINTR
) {
1419 for (i
= 0; i
< nb_fd
; i
++) {
1420 health_code_update();
1422 /* Fetch once the poll data */
1423 revents
= LTTNG_POLL_GETEV(&events
, i
);
1424 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1426 /* Thread quit pipe has been closed. Killing thread. */
1427 ret
= check_thread_quit_pipe(pollfd
, revents
);
1433 /* Event on the registration socket */
1434 if (pollfd
== apps_sock
) {
1435 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1436 ERR("Register apps socket poll error");
1438 } else if (revents
& LPOLLIN
) {
1439 sock
= lttcomm_accept_unix_sock(apps_sock
);
1445 * Set the CLOEXEC flag. Return code is useless because
1446 * either way, the show must go on.
1448 (void) utils_set_fd_cloexec(sock
);
1450 /* Create UST registration command for enqueuing */
1451 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1452 if (ust_cmd
== NULL
) {
1453 PERROR("ust command zmalloc");
1458 * Using message-based transmissions to ensure we don't
1459 * have to deal with partially received messages.
1461 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1463 ERR("Exhausted file descriptors allowed for applications.");
1472 health_code_update();
1473 ret
= lttcomm_recv_unix_sock(sock
, &ust_cmd
->reg_msg
,
1474 sizeof(struct ust_register_msg
));
1475 if (ret
< 0 || ret
< sizeof(struct ust_register_msg
)) {
1477 PERROR("lttcomm_recv_unix_sock register apps");
1479 ERR("Wrong size received on apps register");
1486 lttng_fd_put(LTTNG_FD_APPS
, 1);
1490 health_code_update();
1492 ust_cmd
->sock
= sock
;
1495 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1496 " gid:%d sock:%d name:%s (version %d.%d)",
1497 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1498 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1499 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1500 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1503 * Lock free enqueue the registration request. The red pill
1504 * has been taken! This apps will be part of the *system*.
1506 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1509 * Wake the registration queue futex. Implicit memory
1510 * barrier with the exchange in cds_wfq_enqueue.
1512 futex_nto1_wake(&ust_cmd_queue
.futex
);
1522 ERR("Health error occurred in %s", __func__
);
1525 /* Notify that the registration thread is gone */
1528 if (apps_sock
>= 0) {
1529 ret
= close(apps_sock
);
1539 lttng_fd_put(LTTNG_FD_APPS
, 1);
1541 unlink(apps_unix_sock_path
);
1544 lttng_poll_clean(&events
);
1548 DBG("UST Registration thread cleanup complete");
1549 health_unregister();
1555 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1556 * exec or it will fails.
1558 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1561 struct timespec timeout
;
1563 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1564 consumer_data
->consumer_thread_is_ready
= 0;
1566 /* Setup pthread condition */
1567 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1570 PERROR("pthread_condattr_init consumer data");
1575 * Set the monotonic clock in order to make sure we DO NOT jump in time
1576 * between the clock_gettime() call and the timedwait call. See bug #324
1577 * for a more details and how we noticed it.
1579 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1582 PERROR("pthread_condattr_setclock consumer data");
1586 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1589 PERROR("pthread_cond_init consumer data");
1593 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1596 PERROR("pthread_create consumer");
1601 /* We are about to wait on a pthread condition */
1602 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1604 /* Get time for sem_timedwait absolute timeout */
1605 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1607 * Set the timeout for the condition timed wait even if the clock gettime
1608 * call fails since we might loop on that call and we want to avoid to
1609 * increment the timeout too many times.
1611 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1614 * The following loop COULD be skipped in some conditions so this is why we
1615 * set ret to 0 in order to make sure at least one round of the loop is
1621 * Loop until the condition is reached or when a timeout is reached. Note
1622 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1623 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1624 * possible. This loop does not take any chances and works with both of
1627 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1628 if (clock_ret
< 0) {
1629 PERROR("clock_gettime spawn consumer");
1630 /* Infinite wait for the consumerd thread to be ready */
1631 ret
= pthread_cond_wait(&consumer_data
->cond
,
1632 &consumer_data
->cond_mutex
);
1634 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1635 &consumer_data
->cond_mutex
, &timeout
);
1639 /* Release the pthread condition */
1640 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1644 if (ret
== ETIMEDOUT
) {
1646 * Call has timed out so we kill the kconsumerd_thread and return
1649 ERR("Condition timed out. The consumer thread was never ready."
1651 ret
= pthread_cancel(consumer_data
->thread
);
1653 PERROR("pthread_cancel consumer thread");
1656 PERROR("pthread_cond_wait failed consumer thread");
1661 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1662 if (consumer_data
->pid
== 0) {
1663 ERR("Consumerd did not start");
1664 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1667 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1676 * Join consumer thread
1678 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1682 /* Consumer pid must be a real one. */
1683 if (consumer_data
->pid
> 0) {
1685 ret
= kill(consumer_data
->pid
, SIGTERM
);
1687 ERR("Error killing consumer daemon");
1690 return pthread_join(consumer_data
->thread
, &status
);
1697 * Fork and exec a consumer daemon (consumerd).
1699 * Return pid if successful else -1.
1701 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1705 const char *consumer_to_use
;
1706 const char *verbosity
;
1709 DBG("Spawning consumerd");
1716 if (opt_verbose_consumer
) {
1717 verbosity
= "--verbose";
1719 verbosity
= "--quiet";
1721 switch (consumer_data
->type
) {
1722 case LTTNG_CONSUMER_KERNEL
:
1724 * Find out which consumerd to execute. We will first try the
1725 * 64-bit path, then the sessiond's installation directory, and
1726 * fallback on the 32-bit one,
1728 DBG3("Looking for a kernel consumer at these locations:");
1729 DBG3(" 1) %s", consumerd64_bin
);
1730 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1731 DBG3(" 3) %s", consumerd32_bin
);
1732 if (stat(consumerd64_bin
, &st
) == 0) {
1733 DBG3("Found location #1");
1734 consumer_to_use
= consumerd64_bin
;
1735 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1736 DBG3("Found location #2");
1737 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1738 } else if (stat(consumerd32_bin
, &st
) == 0) {
1739 DBG3("Found location #3");
1740 consumer_to_use
= consumerd32_bin
;
1742 DBG("Could not find any valid consumerd executable");
1745 DBG("Using kernel consumer at: %s", consumer_to_use
);
1746 execl(consumer_to_use
,
1747 "lttng-consumerd", verbosity
, "-k",
1748 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1749 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1752 case LTTNG_CONSUMER64_UST
:
1754 char *tmpnew
= NULL
;
1756 if (consumerd64_libdir
[0] != '\0') {
1760 tmp
= getenv("LD_LIBRARY_PATH");
1764 tmplen
= strlen("LD_LIBRARY_PATH=")
1765 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1766 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1771 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1772 strcat(tmpnew
, consumerd64_libdir
);
1773 if (tmp
[0] != '\0') {
1774 strcat(tmpnew
, ":");
1775 strcat(tmpnew
, tmp
);
1777 ret
= putenv(tmpnew
);
1783 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1784 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1785 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1786 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1788 if (consumerd64_libdir
[0] != '\0') {
1796 case LTTNG_CONSUMER32_UST
:
1798 char *tmpnew
= NULL
;
1800 if (consumerd32_libdir
[0] != '\0') {
1804 tmp
= getenv("LD_LIBRARY_PATH");
1808 tmplen
= strlen("LD_LIBRARY_PATH=")
1809 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1810 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1815 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1816 strcat(tmpnew
, consumerd32_libdir
);
1817 if (tmp
[0] != '\0') {
1818 strcat(tmpnew
, ":");
1819 strcat(tmpnew
, tmp
);
1821 ret
= putenv(tmpnew
);
1827 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
1828 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
1829 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1830 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1832 if (consumerd32_libdir
[0] != '\0') {
1841 PERROR("unknown consumer type");
1845 PERROR("kernel start consumer exec");
1848 } else if (pid
> 0) {
1851 PERROR("start consumer fork");
1859 * Spawn the consumerd daemon and session daemon thread.
1861 static int start_consumerd(struct consumer_data
*consumer_data
)
1866 * Set the listen() state on the socket since there is a possible race
1867 * between the exec() of the consumer daemon and this call if place in the
1868 * consumer thread. See bug #366 for more details.
1870 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
1875 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1876 if (consumer_data
->pid
!= 0) {
1877 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1881 ret
= spawn_consumerd(consumer_data
);
1883 ERR("Spawning consumerd failed");
1884 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1888 /* Setting up the consumer_data pid */
1889 consumer_data
->pid
= ret
;
1890 DBG2("Consumer pid %d", consumer_data
->pid
);
1891 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1893 DBG2("Spawning consumer control thread");
1894 ret
= spawn_consumer_thread(consumer_data
);
1896 ERR("Fatal error spawning consumer control thread");
1904 /* Cleanup already created socket on error. */
1905 if (consumer_data
->err_sock
>= 0) {
1908 err
= close(consumer_data
->err_sock
);
1910 PERROR("close consumer data error socket");
1917 * Compute health status of each consumer. If one of them is zero (bad
1918 * state), we return 0.
1920 static int check_consumer_health(void)
1924 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
1926 DBG3("Health consumer check %d", ret
);
1932 * Setup necessary data for kernel tracer action.
1934 static int init_kernel_tracer(void)
1938 /* Modprobe lttng kernel modules */
1939 ret
= modprobe_lttng_control();
1944 /* Open debugfs lttng */
1945 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1946 if (kernel_tracer_fd
< 0) {
1947 DBG("Failed to open %s", module_proc_lttng
);
1952 /* Validate kernel version */
1953 ret
= kernel_validate_version(kernel_tracer_fd
);
1958 ret
= modprobe_lttng_data();
1963 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1967 modprobe_remove_lttng_control();
1968 ret
= close(kernel_tracer_fd
);
1972 kernel_tracer_fd
= -1;
1973 return LTTNG_ERR_KERN_VERSION
;
1976 ret
= close(kernel_tracer_fd
);
1982 modprobe_remove_lttng_control();
1985 WARN("No kernel tracer available");
1986 kernel_tracer_fd
= -1;
1988 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
1990 return LTTNG_ERR_KERN_NA
;
1996 * Copy consumer output from the tracing session to the domain session. The
1997 * function also applies the right modification on a per domain basis for the
1998 * trace files destination directory.
2000 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2003 const char *dir_name
;
2004 struct consumer_output
*consumer
;
2007 assert(session
->consumer
);
2010 case LTTNG_DOMAIN_KERNEL
:
2011 DBG3("Copying tracing session consumer output in kernel session");
2013 * XXX: We should audit the session creation and what this function
2014 * does "extra" in order to avoid a destroy since this function is used
2015 * in the domain session creation (kernel and ust) only. Same for UST
2018 if (session
->kernel_session
->consumer
) {
2019 consumer_destroy_output(session
->kernel_session
->consumer
);
2021 session
->kernel_session
->consumer
=
2022 consumer_copy_output(session
->consumer
);
2023 /* Ease our life a bit for the next part */
2024 consumer
= session
->kernel_session
->consumer
;
2025 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2027 case LTTNG_DOMAIN_UST
:
2028 DBG3("Copying tracing session consumer output in UST session");
2029 if (session
->ust_session
->consumer
) {
2030 consumer_destroy_output(session
->ust_session
->consumer
);
2032 session
->ust_session
->consumer
=
2033 consumer_copy_output(session
->consumer
);
2034 /* Ease our life a bit for the next part */
2035 consumer
= session
->ust_session
->consumer
;
2036 dir_name
= DEFAULT_UST_TRACE_DIR
;
2039 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2043 /* Append correct directory to subdir */
2044 strncat(consumer
->subdir
, dir_name
,
2045 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2046 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2055 * Create an UST session and add it to the session ust list.
2057 static int create_ust_session(struct ltt_session
*session
,
2058 struct lttng_domain
*domain
)
2061 struct ltt_ust_session
*lus
= NULL
;
2065 assert(session
->consumer
);
2067 switch (domain
->type
) {
2068 case LTTNG_DOMAIN_UST
:
2071 ERR("Unknown UST domain on create session %d", domain
->type
);
2072 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2076 DBG("Creating UST session");
2078 lus
= trace_ust_create_session(session
->path
, session
->id
, domain
);
2080 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2084 lus
->uid
= session
->uid
;
2085 lus
->gid
= session
->gid
;
2086 session
->ust_session
= lus
;
2088 /* Copy session output to the newly created UST session */
2089 ret
= copy_session_consumer(domain
->type
, session
);
2090 if (ret
!= LTTNG_OK
) {
2098 session
->ust_session
= NULL
;
2103 * Create a kernel tracer session then create the default channel.
2105 static int create_kernel_session(struct ltt_session
*session
)
2109 DBG("Creating kernel session");
2111 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2113 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2117 /* Code flow safety */
2118 assert(session
->kernel_session
);
2120 /* Copy session output to the newly created Kernel session */
2121 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2122 if (ret
!= LTTNG_OK
) {
2126 /* Create directory(ies) on local filesystem. */
2127 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2128 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2129 ret
= run_as_mkdir_recursive(
2130 session
->kernel_session
->consumer
->dst
.trace_path
,
2131 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2133 if (ret
!= -EEXIST
) {
2134 ERR("Trace directory creation error");
2140 session
->kernel_session
->uid
= session
->uid
;
2141 session
->kernel_session
->gid
= session
->gid
;
2146 trace_kernel_destroy_session(session
->kernel_session
);
2147 session
->kernel_session
= NULL
;
2152 * Count number of session permitted by uid/gid.
2154 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2157 struct ltt_session
*session
;
2159 DBG("Counting number of available session for UID %d GID %d",
2161 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2163 * Only list the sessions the user can control.
2165 if (!session_access_ok(session
, uid
, gid
)) {
2174 * Process the command requested by the lttng client within the command
2175 * context structure. This function make sure that the return structure (llm)
2176 * is set and ready for transmission before returning.
2178 * Return any error encountered or 0 for success.
2180 * "sock" is only used for special-case var. len data.
2182 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2186 int need_tracing_session
= 1;
2189 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2193 switch (cmd_ctx
->lsm
->cmd_type
) {
2194 case LTTNG_CREATE_SESSION
:
2195 case LTTNG_DESTROY_SESSION
:
2196 case LTTNG_LIST_SESSIONS
:
2197 case LTTNG_LIST_DOMAINS
:
2198 case LTTNG_START_TRACE
:
2199 case LTTNG_STOP_TRACE
:
2200 case LTTNG_DATA_PENDING
:
2207 if (opt_no_kernel
&& need_domain
2208 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2210 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2212 ret
= LTTNG_ERR_KERN_NA
;
2217 /* Deny register consumer if we already have a spawned consumer. */
2218 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2219 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2220 if (kconsumer_data
.pid
> 0) {
2221 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2222 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2225 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2229 * Check for command that don't needs to allocate a returned payload. We do
2230 * this here so we don't have to make the call for no payload at each
2233 switch(cmd_ctx
->lsm
->cmd_type
) {
2234 case LTTNG_LIST_SESSIONS
:
2235 case LTTNG_LIST_TRACEPOINTS
:
2236 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2237 case LTTNG_LIST_DOMAINS
:
2238 case LTTNG_LIST_CHANNELS
:
2239 case LTTNG_LIST_EVENTS
:
2242 /* Setup lttng message with no payload */
2243 ret
= setup_lttng_msg(cmd_ctx
, 0);
2245 /* This label does not try to unlock the session */
2246 goto init_setup_error
;
2250 /* Commands that DO NOT need a session. */
2251 switch (cmd_ctx
->lsm
->cmd_type
) {
2252 case LTTNG_CREATE_SESSION
:
2253 case LTTNG_CALIBRATE
:
2254 case LTTNG_LIST_SESSIONS
:
2255 case LTTNG_LIST_TRACEPOINTS
:
2256 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2257 need_tracing_session
= 0;
2260 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2262 * We keep the session list lock across _all_ commands
2263 * for now, because the per-session lock does not
2264 * handle teardown properly.
2266 session_lock_list();
2267 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2268 if (cmd_ctx
->session
== NULL
) {
2269 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2270 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2272 /* If no session name specified */
2273 ret
= LTTNG_ERR_SELECT_SESS
;
2277 /* Acquire lock for the session */
2278 session_lock(cmd_ctx
->session
);
2288 * Check domain type for specific "pre-action".
2290 switch (cmd_ctx
->lsm
->domain
.type
) {
2291 case LTTNG_DOMAIN_KERNEL
:
2293 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2297 /* Kernel tracer check */
2298 if (kernel_tracer_fd
== -1) {
2299 /* Basically, load kernel tracer modules */
2300 ret
= init_kernel_tracer();
2306 /* Consumer is in an ERROR state. Report back to client */
2307 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2308 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2312 /* Need a session for kernel command */
2313 if (need_tracing_session
) {
2314 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2315 ret
= create_kernel_session(cmd_ctx
->session
);
2317 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2322 /* Start the kernel consumer daemon */
2323 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2324 if (kconsumer_data
.pid
== 0 &&
2325 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2326 cmd_ctx
->session
->start_consumer
) {
2327 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2328 ret
= start_consumerd(&kconsumer_data
);
2330 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2333 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2335 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2339 * The consumer was just spawned so we need to add the socket to
2340 * the consumer output of the session if exist.
2342 ret
= consumer_create_socket(&kconsumer_data
,
2343 cmd_ctx
->session
->kernel_session
->consumer
);
2350 case LTTNG_DOMAIN_UST
:
2352 /* Consumer is in an ERROR state. Report back to client */
2353 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2354 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2358 if (need_tracing_session
) {
2359 /* Create UST session if none exist. */
2360 if (cmd_ctx
->session
->ust_session
== NULL
) {
2361 ret
= create_ust_session(cmd_ctx
->session
,
2362 &cmd_ctx
->lsm
->domain
);
2363 if (ret
!= LTTNG_OK
) {
2368 /* Start the UST consumer daemons */
2370 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2371 if (consumerd64_bin
[0] != '\0' &&
2372 ustconsumer64_data
.pid
== 0 &&
2373 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2374 cmd_ctx
->session
->start_consumer
) {
2375 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2376 ret
= start_consumerd(&ustconsumer64_data
);
2378 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2379 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2383 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2384 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2386 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2390 * Setup socket for consumer 64 bit. No need for atomic access
2391 * since it was set above and can ONLY be set in this thread.
2393 ret
= consumer_create_socket(&ustconsumer64_data
,
2394 cmd_ctx
->session
->ust_session
->consumer
);
2400 if (consumerd32_bin
[0] != '\0' &&
2401 ustconsumer32_data
.pid
== 0 &&
2402 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2403 cmd_ctx
->session
->start_consumer
) {
2404 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2405 ret
= start_consumerd(&ustconsumer32_data
);
2407 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2408 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2412 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2413 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2415 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2419 * Setup socket for consumer 64 bit. No need for atomic access
2420 * since it was set above and can ONLY be set in this thread.
2422 ret
= consumer_create_socket(&ustconsumer32_data
,
2423 cmd_ctx
->session
->ust_session
->consumer
);
2435 /* Validate consumer daemon state when start/stop trace command */
2436 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2437 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2438 switch (cmd_ctx
->lsm
->domain
.type
) {
2439 case LTTNG_DOMAIN_UST
:
2440 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2441 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2445 case LTTNG_DOMAIN_KERNEL
:
2446 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2447 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2455 * Check that the UID or GID match that of the tracing session.
2456 * The root user can interact with all sessions.
2458 if (need_tracing_session
) {
2459 if (!session_access_ok(cmd_ctx
->session
,
2460 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2461 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2462 ret
= LTTNG_ERR_EPERM
;
2467 /* Process by command type */
2468 switch (cmd_ctx
->lsm
->cmd_type
) {
2469 case LTTNG_ADD_CONTEXT
:
2471 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2472 cmd_ctx
->lsm
->u
.context
.channel_name
,
2473 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2476 case LTTNG_DISABLE_CHANNEL
:
2478 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2479 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2482 case LTTNG_DISABLE_EVENT
:
2484 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2485 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2486 cmd_ctx
->lsm
->u
.disable
.name
);
2489 case LTTNG_DISABLE_ALL_EVENT
:
2491 DBG("Disabling all events");
2493 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2494 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2497 case LTTNG_DISABLE_CONSUMER
:
2499 ret
= cmd_disable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2502 case LTTNG_ENABLE_CHANNEL
:
2504 ret
= cmd_enable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2505 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2508 case LTTNG_ENABLE_CONSUMER
:
2511 * XXX: 0 means that this URI should be applied on the session. Should
2512 * be a DOMAIN enuam.
2514 ret
= cmd_enable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2515 if (ret
!= LTTNG_OK
) {
2519 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2520 /* Add the URI for the UST session if a consumer is present. */
2521 if (cmd_ctx
->session
->ust_session
&&
2522 cmd_ctx
->session
->ust_session
->consumer
) {
2523 ret
= cmd_enable_consumer(LTTNG_DOMAIN_UST
, cmd_ctx
->session
);
2524 } else if (cmd_ctx
->session
->kernel_session
&&
2525 cmd_ctx
->session
->kernel_session
->consumer
) {
2526 ret
= cmd_enable_consumer(LTTNG_DOMAIN_KERNEL
,
2532 case LTTNG_ENABLE_EVENT
:
2534 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2535 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2536 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2539 case LTTNG_ENABLE_ALL_EVENT
:
2541 DBG("Enabling all events");
2543 ret
= cmd_enable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2544 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2545 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2548 case LTTNG_LIST_TRACEPOINTS
:
2550 struct lttng_event
*events
;
2553 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2554 if (nb_events
< 0) {
2555 /* Return value is a negative lttng_error_code. */
2561 * Setup lttng message with payload size set to the event list size in
2562 * bytes and then copy list into the llm payload.
2564 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2570 /* Copy event list into message payload */
2571 memcpy(cmd_ctx
->llm
->payload
, events
,
2572 sizeof(struct lttng_event
) * nb_events
);
2579 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2581 struct lttng_event_field
*fields
;
2584 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2586 if (nb_fields
< 0) {
2587 /* Return value is a negative lttng_error_code. */
2593 * Setup lttng message with payload size set to the event list size in
2594 * bytes and then copy list into the llm payload.
2596 ret
= setup_lttng_msg(cmd_ctx
,
2597 sizeof(struct lttng_event_field
) * nb_fields
);
2603 /* Copy event list into message payload */
2604 memcpy(cmd_ctx
->llm
->payload
, fields
,
2605 sizeof(struct lttng_event_field
) * nb_fields
);
2612 case LTTNG_SET_CONSUMER_URI
:
2615 struct lttng_uri
*uris
;
2617 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2618 len
= nb_uri
* sizeof(struct lttng_uri
);
2621 ret
= LTTNG_ERR_INVALID
;
2625 uris
= zmalloc(len
);
2627 ret
= LTTNG_ERR_FATAL
;
2631 /* Receive variable len data */
2632 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2633 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2635 DBG("No URIs received from client... continuing");
2637 ret
= LTTNG_ERR_SESSION_FAIL
;
2642 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2644 if (ret
!= LTTNG_OK
) {
2650 * XXX: 0 means that this URI should be applied on the session. Should
2651 * be a DOMAIN enuam.
2653 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2654 /* Add the URI for the UST session if a consumer is present. */
2655 if (cmd_ctx
->session
->ust_session
&&
2656 cmd_ctx
->session
->ust_session
->consumer
) {
2657 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2659 } else if (cmd_ctx
->session
->kernel_session
&&
2660 cmd_ctx
->session
->kernel_session
->consumer
) {
2661 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2662 cmd_ctx
->session
, nb_uri
, uris
);
2670 case LTTNG_START_TRACE
:
2672 ret
= cmd_start_trace(cmd_ctx
->session
);
2675 case LTTNG_STOP_TRACE
:
2677 ret
= cmd_stop_trace(cmd_ctx
->session
);
2680 case LTTNG_CREATE_SESSION
:
2683 struct lttng_uri
*uris
= NULL
;
2685 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2686 len
= nb_uri
* sizeof(struct lttng_uri
);
2689 uris
= zmalloc(len
);
2691 ret
= LTTNG_ERR_FATAL
;
2695 /* Receive variable len data */
2696 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2697 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2699 DBG("No URIs received from client... continuing");
2701 ret
= LTTNG_ERR_SESSION_FAIL
;
2706 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2707 DBG("Creating session with ONE network URI is a bad call");
2708 ret
= LTTNG_ERR_SESSION_FAIL
;
2714 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2721 case LTTNG_DESTROY_SESSION
:
2723 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2725 /* Set session to NULL so we do not unlock it after free. */
2726 cmd_ctx
->session
= NULL
;
2729 case LTTNG_LIST_DOMAINS
:
2732 struct lttng_domain
*domains
;
2734 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2736 /* Return value is a negative lttng_error_code. */
2741 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2746 /* Copy event list into message payload */
2747 memcpy(cmd_ctx
->llm
->payload
, domains
,
2748 nb_dom
* sizeof(struct lttng_domain
));
2755 case LTTNG_LIST_CHANNELS
:
2758 struct lttng_channel
*channels
;
2760 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2761 cmd_ctx
->session
, &channels
);
2763 /* Return value is a negative lttng_error_code. */
2768 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2773 /* Copy event list into message payload */
2774 memcpy(cmd_ctx
->llm
->payload
, channels
,
2775 nb_chan
* sizeof(struct lttng_channel
));
2782 case LTTNG_LIST_EVENTS
:
2785 struct lttng_event
*events
= NULL
;
2787 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2788 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2790 /* Return value is a negative lttng_error_code. */
2795 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2800 /* Copy event list into message payload */
2801 memcpy(cmd_ctx
->llm
->payload
, events
,
2802 nb_event
* sizeof(struct lttng_event
));
2809 case LTTNG_LIST_SESSIONS
:
2811 unsigned int nr_sessions
;
2813 session_lock_list();
2814 nr_sessions
= lttng_sessions_count(
2815 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2816 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2818 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2820 session_unlock_list();
2824 /* Filled the session array */
2825 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2826 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2827 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2829 session_unlock_list();
2834 case LTTNG_CALIBRATE
:
2836 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
2837 &cmd_ctx
->lsm
->u
.calibrate
);
2840 case LTTNG_REGISTER_CONSUMER
:
2842 struct consumer_data
*cdata
;
2844 switch (cmd_ctx
->lsm
->domain
.type
) {
2845 case LTTNG_DOMAIN_KERNEL
:
2846 cdata
= &kconsumer_data
;
2849 ret
= LTTNG_ERR_UND
;
2853 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2854 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
2857 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
2859 struct lttng_filter_bytecode
*bytecode
;
2861 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
2862 ret
= LTTNG_ERR_FILTER_INVAL
;
2865 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
2866 ret
= LTTNG_ERR_FILTER_INVAL
;
2869 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2871 ret
= LTTNG_ERR_FILTER_NOMEM
;
2874 /* Receive var. len. data */
2875 DBG("Receiving var len data from client ...");
2876 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
2877 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2879 DBG("Nothing recv() from client var len data... continuing");
2881 ret
= LTTNG_ERR_FILTER_INVAL
;
2885 if (bytecode
->len
+ sizeof(*bytecode
)
2886 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
2888 ret
= LTTNG_ERR_FILTER_INVAL
;
2892 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2893 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2894 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
2897 case LTTNG_DATA_PENDING
:
2899 ret
= cmd_data_pending(cmd_ctx
->session
);
2903 ret
= LTTNG_ERR_UND
;
2908 if (cmd_ctx
->llm
== NULL
) {
2909 DBG("Missing llm structure. Allocating one.");
2910 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
2914 /* Set return code */
2915 cmd_ctx
->llm
->ret_code
= ret
;
2917 if (cmd_ctx
->session
) {
2918 session_unlock(cmd_ctx
->session
);
2920 if (need_tracing_session
) {
2921 session_unlock_list();
2928 * Thread managing health check socket.
2930 static void *thread_manage_health(void *data
)
2932 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
2933 uint32_t revents
, nb_fd
;
2934 struct lttng_poll_event events
;
2935 struct lttcomm_health_msg msg
;
2936 struct lttcomm_health_data reply
;
2938 DBG("[thread] Manage health check started");
2940 rcu_register_thread();
2942 /* Create unix socket */
2943 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
2945 ERR("Unable to create health check Unix socket");
2951 * Set the CLOEXEC flag. Return code is useless because either way, the
2954 (void) utils_set_fd_cloexec(sock
);
2956 ret
= lttcomm_listen_unix_sock(sock
);
2962 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2963 * more will be added to this poll set.
2965 ret
= create_thread_poll_set(&events
, 2);
2970 /* Add the application registration socket */
2971 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
2977 DBG("Health check ready");
2979 /* Inifinite blocking call, waiting for transmission */
2981 ret
= lttng_poll_wait(&events
, -1);
2984 * Restart interrupted system call.
2986 if (errno
== EINTR
) {
2994 for (i
= 0; i
< nb_fd
; i
++) {
2995 /* Fetch once the poll data */
2996 revents
= LTTNG_POLL_GETEV(&events
, i
);
2997 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2999 /* Thread quit pipe has been closed. Killing thread. */
3000 ret
= check_thread_quit_pipe(pollfd
, revents
);
3006 /* Event on the registration socket */
3007 if (pollfd
== sock
) {
3008 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3009 ERR("Health socket poll error");
3015 new_sock
= lttcomm_accept_unix_sock(sock
);
3021 * Set the CLOEXEC flag. Return code is useless because either way, the
3024 (void) utils_set_fd_cloexec(new_sock
);
3026 DBG("Receiving data from client for health...");
3027 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3029 DBG("Nothing recv() from client... continuing");
3030 ret
= close(new_sock
);
3038 rcu_thread_online();
3040 switch (msg
.component
) {
3041 case LTTNG_HEALTH_CMD
:
3042 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3044 case LTTNG_HEALTH_APP_MANAGE
:
3045 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3047 case LTTNG_HEALTH_APP_REG
:
3048 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3050 case LTTNG_HEALTH_KERNEL
:
3051 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3053 case LTTNG_HEALTH_CONSUMER
:
3054 reply
.ret_code
= check_consumer_health();
3056 case LTTNG_HEALTH_ALL
:
3058 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3059 health_check_state(HEALTH_TYPE_APP_REG
) &&
3060 health_check_state(HEALTH_TYPE_CMD
) &&
3061 health_check_state(HEALTH_TYPE_KERNEL
) &&
3062 check_consumer_health();
3065 reply
.ret_code
= LTTNG_ERR_UND
;
3070 * Flip ret value since 0 is a success and 1 indicates a bad health for
3071 * the client where in the sessiond it is the opposite. Again, this is
3072 * just to make things easier for us poor developer which enjoy a lot
3075 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3076 reply
.ret_code
= !reply
.ret_code
;
3079 DBG2("Health check return value %d", reply
.ret_code
);
3081 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3083 ERR("Failed to send health data back to client");
3086 /* End of transmission */
3087 ret
= close(new_sock
);
3097 ERR("Health error occurred in %s", __func__
);
3099 DBG("Health check thread dying");
3100 unlink(health_unix_sock_path
);
3107 if (new_sock
>= 0) {
3108 ret
= close(new_sock
);
3114 lttng_poll_clean(&events
);
3116 rcu_unregister_thread();
3121 * This thread manage all clients request using the unix client socket for
3124 static void *thread_manage_clients(void *data
)
3126 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3128 uint32_t revents
, nb_fd
;
3129 struct command_ctx
*cmd_ctx
= NULL
;
3130 struct lttng_poll_event events
;
3132 DBG("[thread] Manage client started");
3134 rcu_register_thread();
3136 health_register(HEALTH_TYPE_CMD
);
3138 if (testpoint(thread_manage_clients
)) {
3139 goto error_testpoint
;
3142 health_code_update();
3144 ret
= lttcomm_listen_unix_sock(client_sock
);
3150 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3151 * more will be added to this poll set.
3153 ret
= create_thread_poll_set(&events
, 2);
3155 goto error_create_poll
;
3158 /* Add the application registration socket */
3159 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3165 * Notify parent pid that we are ready to accept command for client side.
3167 if (opt_sig_parent
) {
3168 kill(ppid
, SIGUSR1
);
3171 if (testpoint(thread_manage_clients_before_loop
)) {
3175 health_code_update();
3178 DBG("Accepting client command ...");
3180 /* Inifinite blocking call, waiting for transmission */
3182 health_poll_entry();
3183 ret
= lttng_poll_wait(&events
, -1);
3187 * Restart interrupted system call.
3189 if (errno
== EINTR
) {
3197 for (i
= 0; i
< nb_fd
; i
++) {
3198 /* Fetch once the poll data */
3199 revents
= LTTNG_POLL_GETEV(&events
, i
);
3200 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3202 health_code_update();
3204 /* Thread quit pipe has been closed. Killing thread. */
3205 ret
= check_thread_quit_pipe(pollfd
, revents
);
3211 /* Event on the registration socket */
3212 if (pollfd
== client_sock
) {
3213 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3214 ERR("Client socket poll error");
3220 DBG("Wait for client response");
3222 health_code_update();
3224 sock
= lttcomm_accept_unix_sock(client_sock
);
3230 * Set the CLOEXEC flag. Return code is useless because either way, the
3233 (void) utils_set_fd_cloexec(sock
);
3235 /* Set socket option for credentials retrieval */
3236 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3241 /* Allocate context command to process the client request */
3242 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3243 if (cmd_ctx
== NULL
) {
3244 PERROR("zmalloc cmd_ctx");
3248 /* Allocate data buffer for reception */
3249 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3250 if (cmd_ctx
->lsm
== NULL
) {
3251 PERROR("zmalloc cmd_ctx->lsm");
3255 cmd_ctx
->llm
= NULL
;
3256 cmd_ctx
->session
= NULL
;
3258 health_code_update();
3261 * Data is received from the lttng client. The struct
3262 * lttcomm_session_msg (lsm) contains the command and data request of
3265 DBG("Receiving data from client ...");
3266 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3267 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3269 DBG("Nothing recv() from client... continuing");
3275 clean_command_ctx(&cmd_ctx
);
3279 health_code_update();
3281 // TODO: Validate cmd_ctx including sanity check for
3282 // security purpose.
3284 rcu_thread_online();
3286 * This function dispatch the work to the kernel or userspace tracer
3287 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3288 * informations for the client. The command context struct contains
3289 * everything this function may needs.
3291 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3292 rcu_thread_offline();
3302 * TODO: Inform client somehow of the fatal error. At
3303 * this point, ret < 0 means that a zmalloc failed
3304 * (ENOMEM). Error detected but still accept
3305 * command, unless a socket error has been
3308 clean_command_ctx(&cmd_ctx
);
3312 health_code_update();
3314 DBG("Sending response (size: %d, retcode: %s)",
3315 cmd_ctx
->lttng_msg_size
,
3316 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3317 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3319 ERR("Failed to send data back to client");
3322 /* End of transmission */
3329 clean_command_ctx(&cmd_ctx
);
3331 health_code_update();
3343 lttng_poll_clean(&events
);
3344 clean_command_ctx(&cmd_ctx
);
3349 unlink(client_unix_sock_path
);
3350 if (client_sock
>= 0) {
3351 ret
= close(client_sock
);
3359 ERR("Health error occurred in %s", __func__
);
3362 health_unregister();
3364 DBG("Client thread dying");
3366 rcu_unregister_thread();
3372 * usage function on stderr
3374 static void usage(void)
3376 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3377 fprintf(stderr
, " -h, --help Display this usage.\n");
3378 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3379 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3380 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3381 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3382 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3383 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3384 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3385 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3386 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3387 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3388 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3389 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3390 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3391 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3392 fprintf(stderr
, " -V, --version Show version number.\n");
3393 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3394 fprintf(stderr
, " -q, --quiet No output at all.\n");
3395 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3396 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3397 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3401 * daemon argument parsing
3403 static int parse_args(int argc
, char **argv
)
3407 static struct option long_options
[] = {
3408 { "client-sock", 1, 0, 'c' },
3409 { "apps-sock", 1, 0, 'a' },
3410 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3411 { "kconsumerd-err-sock", 1, 0, 'E' },
3412 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3413 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3414 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3415 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3416 { "consumerd32-path", 1, 0, 'u' },
3417 { "consumerd32-libdir", 1, 0, 'U' },
3418 { "consumerd64-path", 1, 0, 't' },
3419 { "consumerd64-libdir", 1, 0, 'T' },
3420 { "daemonize", 0, 0, 'd' },
3421 { "sig-parent", 0, 0, 'S' },
3422 { "help", 0, 0, 'h' },
3423 { "group", 1, 0, 'g' },
3424 { "version", 0, 0, 'V' },
3425 { "quiet", 0, 0, 'q' },
3426 { "verbose", 0, 0, 'v' },
3427 { "verbose-consumer", 0, 0, 'Z' },
3428 { "no-kernel", 0, 0, 'N' },
3433 int option_index
= 0;
3434 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3435 long_options
, &option_index
);
3442 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3444 fprintf(stderr
, " with arg %s\n", optarg
);
3448 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3451 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3457 opt_tracing_group
= optarg
;
3463 fprintf(stdout
, "%s\n", VERSION
);
3469 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3472 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3475 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3478 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3481 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3484 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3490 lttng_opt_quiet
= 1;
3493 /* Verbose level can increase using multiple -v */
3494 lttng_opt_verbose
+= 1;
3497 opt_verbose_consumer
+= 1;
3500 consumerd32_bin
= optarg
;
3503 consumerd32_libdir
= optarg
;
3506 consumerd64_bin
= optarg
;
3509 consumerd64_libdir
= optarg
;
3512 /* Unknown option or other error.
3513 * Error is printed by getopt, just return */
3522 * Creates the two needed socket by the daemon.
3523 * apps_sock - The communication socket for all UST apps.
3524 * client_sock - The communication of the cli tool (lttng).
3526 static int init_daemon_socket(void)
3531 old_umask
= umask(0);
3533 /* Create client tool unix socket */
3534 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3535 if (client_sock
< 0) {
3536 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3541 /* Set the cloexec flag */
3542 ret
= utils_set_fd_cloexec(client_sock
);
3544 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3545 "Continuing but note that the consumer daemon will have a "
3546 "reference to this socket on exec()", client_sock
);
3549 /* File permission MUST be 660 */
3550 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3552 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3557 /* Create the application unix socket */
3558 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3559 if (apps_sock
< 0) {
3560 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3565 /* Set the cloexec flag */
3566 ret
= utils_set_fd_cloexec(apps_sock
);
3568 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3569 "Continuing but note that the consumer daemon will have a "
3570 "reference to this socket on exec()", apps_sock
);
3573 /* File permission MUST be 666 */
3574 ret
= chmod(apps_unix_sock_path
,
3575 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3577 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3582 DBG3("Session daemon client socket %d and application socket %d created",
3583 client_sock
, apps_sock
);
3591 * Check if the global socket is available, and if a daemon is answering at the
3592 * other side. If yes, error is returned.
3594 static int check_existing_daemon(void)
3596 /* Is there anybody out there ? */
3597 if (lttng_session_daemon_alive()) {
3605 * Set the tracing group gid onto the client socket.
3607 * Race window between mkdir and chown is OK because we are going from more
3608 * permissive (root.root) to less permissive (root.tracing).
3610 static int set_permissions(char *rundir
)
3615 ret
= allowed_group();
3617 WARN("No tracing group detected");
3624 /* Set lttng run dir */
3625 ret
= chown(rundir
, 0, gid
);
3627 ERR("Unable to set group on %s", rundir
);
3631 /* Ensure tracing group can search the run dir */
3632 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3634 ERR("Unable to set permissions on %s", rundir
);
3638 /* lttng client socket path */
3639 ret
= chown(client_unix_sock_path
, 0, gid
);
3641 ERR("Unable to set group on %s", client_unix_sock_path
);
3645 /* kconsumer error socket path */
3646 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3648 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3652 /* 64-bit ustconsumer error socket path */
3653 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3655 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3659 /* 32-bit ustconsumer compat32 error socket path */
3660 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3662 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3666 DBG("All permissions are set");
3673 * Create the lttng run directory needed for all global sockets and pipe.
3675 static int create_lttng_rundir(const char *rundir
)
3679 DBG3("Creating LTTng run directory: %s", rundir
);
3681 ret
= mkdir(rundir
, S_IRWXU
);
3683 if (errno
!= EEXIST
) {
3684 ERR("Unable to create %s", rundir
);
3696 * Setup sockets and directory needed by the kconsumerd communication with the
3699 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3703 char path
[PATH_MAX
];
3705 switch (consumer_data
->type
) {
3706 case LTTNG_CONSUMER_KERNEL
:
3707 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3709 case LTTNG_CONSUMER64_UST
:
3710 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3712 case LTTNG_CONSUMER32_UST
:
3713 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3716 ERR("Consumer type unknown");
3721 DBG2("Creating consumer directory: %s", path
);
3723 ret
= mkdir(path
, S_IRWXU
);
3725 if (errno
!= EEXIST
) {
3727 ERR("Failed to create %s", path
);
3733 /* Create the kconsumerd error unix socket */
3734 consumer_data
->err_sock
=
3735 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3736 if (consumer_data
->err_sock
< 0) {
3737 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3742 /* File permission MUST be 660 */
3743 ret
= chmod(consumer_data
->err_unix_sock_path
,
3744 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3746 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3756 * Signal handler for the daemon
3758 * Simply stop all worker threads, leaving main() return gracefully after
3759 * joining all threads and calling cleanup().
3761 static void sighandler(int sig
)
3765 DBG("SIGPIPE caught");
3768 DBG("SIGINT caught");
3772 DBG("SIGTERM caught");
3781 * Setup signal handler for :
3782 * SIGINT, SIGTERM, SIGPIPE
3784 static int set_signal_handler(void)
3787 struct sigaction sa
;
3790 if ((ret
= sigemptyset(&sigset
)) < 0) {
3791 PERROR("sigemptyset");
3795 sa
.sa_handler
= sighandler
;
3796 sa
.sa_mask
= sigset
;
3798 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3799 PERROR("sigaction");
3803 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3804 PERROR("sigaction");
3808 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3809 PERROR("sigaction");
3813 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3819 * Set open files limit to unlimited. This daemon can open a large number of
3820 * file descriptors in order to consumer multiple kernel traces.
3822 static void set_ulimit(void)
3827 /* The kernel does not allowed an infinite limit for open files */
3828 lim
.rlim_cur
= 65535;
3829 lim
.rlim_max
= 65535;
3831 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
3833 PERROR("failed to set open files limit");
3840 int main(int argc
, char **argv
)
3844 const char *home_path
, *env_app_timeout
;
3846 init_kernel_workarounds();
3848 rcu_register_thread();
3850 setup_consumerd_path();
3852 /* Parse arguments */
3854 if ((ret
= parse_args(argc
, argv
)) < 0) {
3864 * child: setsid, close FD 0, 1, 2, chdir /
3865 * parent: exit (if fork is successful)
3873 * We are in the child. Make sure all other file
3874 * descriptors are closed, in case we are called with
3875 * more opened file descriptors than the standard ones.
3877 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3882 /* Create thread quit pipe */
3883 if ((ret
= init_thread_quit_pipe()) < 0) {
3887 /* Check if daemon is UID = 0 */
3888 is_root
= !getuid();
3891 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
3893 /* Create global run dir with root access */
3894 ret
= create_lttng_rundir(rundir
);
3899 if (strlen(apps_unix_sock_path
) == 0) {
3900 snprintf(apps_unix_sock_path
, PATH_MAX
,
3901 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
3904 if (strlen(client_unix_sock_path
) == 0) {
3905 snprintf(client_unix_sock_path
, PATH_MAX
,
3906 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
3909 /* Set global SHM for ust */
3910 if (strlen(wait_shm_path
) == 0) {
3911 snprintf(wait_shm_path
, PATH_MAX
,
3912 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
3915 if (strlen(health_unix_sock_path
) == 0) {
3916 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3917 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
3920 /* Setup kernel consumerd path */
3921 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
3922 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
3923 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
3924 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
3926 DBG2("Kernel consumer err path: %s",
3927 kconsumer_data
.err_unix_sock_path
);
3928 DBG2("Kernel consumer cmd path: %s",
3929 kconsumer_data
.cmd_unix_sock_path
);
3931 home_path
= get_home_dir();
3932 if (home_path
== NULL
) {
3933 /* TODO: Add --socket PATH option */
3934 ERR("Can't get HOME directory for sockets creation.");
3940 * Create rundir from home path. This will create something like
3943 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
3949 ret
= create_lttng_rundir(rundir
);
3954 if (strlen(apps_unix_sock_path
) == 0) {
3955 snprintf(apps_unix_sock_path
, PATH_MAX
,
3956 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
3959 /* Set the cli tool unix socket path */
3960 if (strlen(client_unix_sock_path
) == 0) {
3961 snprintf(client_unix_sock_path
, PATH_MAX
,
3962 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
3965 /* Set global SHM for ust */
3966 if (strlen(wait_shm_path
) == 0) {
3967 snprintf(wait_shm_path
, PATH_MAX
,
3968 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, geteuid());
3971 /* Set health check Unix path */
3972 if (strlen(health_unix_sock_path
) == 0) {
3973 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3974 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
3978 /* Set consumer initial state */
3979 kernel_consumerd_state
= CONSUMER_STOPPED
;
3980 ust_consumerd_state
= CONSUMER_STOPPED
;
3982 DBG("Client socket path %s", client_unix_sock_path
);
3983 DBG("Application socket path %s", apps_unix_sock_path
);
3984 DBG("LTTng run directory path: %s", rundir
);
3986 /* 32 bits consumerd path setup */
3987 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
3988 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
3989 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
3990 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
3992 DBG2("UST consumer 32 bits err path: %s",
3993 ustconsumer32_data
.err_unix_sock_path
);
3994 DBG2("UST consumer 32 bits cmd path: %s",
3995 ustconsumer32_data
.cmd_unix_sock_path
);
3997 /* 64 bits consumerd path setup */
3998 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
3999 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4000 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4001 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4003 DBG2("UST consumer 64 bits err path: %s",
4004 ustconsumer64_data
.err_unix_sock_path
);
4005 DBG2("UST consumer 64 bits cmd path: %s",
4006 ustconsumer64_data
.cmd_unix_sock_path
);
4009 * See if daemon already exist.
4011 if ((ret
= check_existing_daemon()) < 0) {
4012 ERR("Already running daemon.\n");
4014 * We do not goto exit because we must not cleanup()
4015 * because a daemon is already running.
4021 * Init UST app hash table. Alloc hash table before this point since
4022 * cleanup() can get called after that point.
4026 /* After this point, we can safely call cleanup() with "goto exit" */
4029 * These actions must be executed as root. We do that *after* setting up
4030 * the sockets path because we MUST make the check for another daemon using
4031 * those paths *before* trying to set the kernel consumer sockets and init
4035 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4040 /* Setup kernel tracer */
4041 if (!opt_no_kernel
) {
4042 init_kernel_tracer();
4045 /* Set ulimit for open files */
4048 /* init lttng_fd tracking must be done after set_ulimit. */
4051 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4056 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4061 if ((ret
= set_signal_handler()) < 0) {
4065 /* Setup the needed unix socket */
4066 if ((ret
= init_daemon_socket()) < 0) {
4070 /* Set credentials to socket */
4071 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4075 /* Get parent pid if -S, --sig-parent is specified. */
4076 if (opt_sig_parent
) {
4080 /* Setup the kernel pipe for waking up the kernel thread */
4081 if (is_root
&& !opt_no_kernel
) {
4082 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4087 /* Setup the thread apps communication pipe. */
4088 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4092 /* Init UST command queue. */
4093 cds_wfq_init(&ust_cmd_queue
.queue
);
4096 * Get session list pointer. This pointer MUST NOT be free(). This list is
4097 * statically declared in session.c
4099 session_list_ptr
= session_get_list();
4101 /* Set up max poll set size */
4102 lttng_poll_set_max_size();
4106 /* Check for the application socket timeout env variable. */
4107 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4108 if (env_app_timeout
) {
4109 app_socket_timeout
= atoi(env_app_timeout
);
4111 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4114 /* Create thread to manage the client socket */
4115 ret
= pthread_create(&health_thread
, NULL
,
4116 thread_manage_health
, (void *) NULL
);
4118 PERROR("pthread_create health");
4122 /* Create thread to manage the client socket */
4123 ret
= pthread_create(&client_thread
, NULL
,
4124 thread_manage_clients
, (void *) NULL
);
4126 PERROR("pthread_create clients");
4130 /* Create thread to dispatch registration */
4131 ret
= pthread_create(&dispatch_thread
, NULL
,
4132 thread_dispatch_ust_registration
, (void *) NULL
);
4134 PERROR("pthread_create dispatch");
4138 /* Create thread to manage application registration. */
4139 ret
= pthread_create(®_apps_thread
, NULL
,
4140 thread_registration_apps
, (void *) NULL
);
4142 PERROR("pthread_create registration");
4146 /* Create thread to manage application socket */
4147 ret
= pthread_create(&apps_thread
, NULL
,
4148 thread_manage_apps
, (void *) NULL
);
4150 PERROR("pthread_create apps");
4154 /* Don't start this thread if kernel tracing is not requested nor root */
4155 if (is_root
&& !opt_no_kernel
) {
4156 /* Create kernel thread to manage kernel event */
4157 ret
= pthread_create(&kernel_thread
, NULL
,
4158 thread_manage_kernel
, (void *) NULL
);
4160 PERROR("pthread_create kernel");
4164 ret
= pthread_join(kernel_thread
, &status
);
4166 PERROR("pthread_join");
4167 goto error
; /* join error, exit without cleanup */
4172 ret
= pthread_join(apps_thread
, &status
);
4174 PERROR("pthread_join");
4175 goto error
; /* join error, exit without cleanup */
4179 ret
= pthread_join(reg_apps_thread
, &status
);
4181 PERROR("pthread_join");
4182 goto error
; /* join error, exit without cleanup */
4186 ret
= pthread_join(dispatch_thread
, &status
);
4188 PERROR("pthread_join");
4189 goto error
; /* join error, exit without cleanup */
4193 ret
= pthread_join(client_thread
, &status
);
4195 PERROR("pthread_join");
4196 goto error
; /* join error, exit without cleanup */
4199 ret
= join_consumer_thread(&kconsumer_data
);
4201 PERROR("join_consumer");
4202 goto error
; /* join error, exit without cleanup */
4205 ret
= join_consumer_thread(&ustconsumer32_data
);
4207 PERROR("join_consumer ust32");
4208 goto error
; /* join error, exit without cleanup */
4211 ret
= join_consumer_thread(&ustconsumer64_data
);
4213 PERROR("join_consumer ust64");
4214 goto error
; /* join error, exit without cleanup */
4218 ret
= pthread_join(health_thread
, &status
);
4220 PERROR("pthread_join health thread");
4221 goto error
; /* join error, exit without cleanup */
4227 * cleanup() is called when no other thread is running.
4229 rcu_thread_online();
4231 rcu_thread_offline();
4232 rcu_unregister_thread();