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
;
661 /* For all tracing session(s) */
662 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
664 if (sess
->ust_session
) {
665 ust_app_global_update(sess
->ust_session
, app_sock
);
667 session_unlock(sess
);
672 * This thread manage event coming from the kernel.
674 * Features supported in this thread:
677 static void *thread_manage_kernel(void *data
)
679 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
680 uint32_t revents
, nb_fd
;
682 struct lttng_poll_event events
;
684 DBG("[thread] Thread manage kernel started");
686 health_register(HEALTH_TYPE_KERNEL
);
689 * This first step of the while is to clean this structure which could free
690 * non NULL pointers so zero it before the loop.
692 memset(&events
, 0, sizeof(events
));
694 if (testpoint(thread_manage_kernel
)) {
695 goto error_testpoint
;
698 health_code_update();
700 if (testpoint(thread_manage_kernel_before_loop
)) {
701 goto error_testpoint
;
705 health_code_update();
707 if (update_poll_flag
== 1) {
708 /* Clean events object. We are about to populate it again. */
709 lttng_poll_clean(&events
);
711 ret
= create_thread_poll_set(&events
, 2);
713 goto error_poll_create
;
716 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
721 /* This will add the available kernel channel if any. */
722 ret
= update_kernel_poll(&events
);
726 update_poll_flag
= 0;
729 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
731 /* Poll infinite value of time */
734 ret
= lttng_poll_wait(&events
, -1);
738 * Restart interrupted system call.
740 if (errno
== EINTR
) {
744 } else if (ret
== 0) {
745 /* Should not happen since timeout is infinite */
746 ERR("Return value of poll is 0 with an infinite timeout.\n"
747 "This should not have happened! Continuing...");
753 for (i
= 0; i
< nb_fd
; i
++) {
754 /* Fetch once the poll data */
755 revents
= LTTNG_POLL_GETEV(&events
, i
);
756 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
758 health_code_update();
760 /* Thread quit pipe has been closed. Killing thread. */
761 ret
= check_thread_quit_pipe(pollfd
, revents
);
767 /* Check for data on kernel pipe */
768 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
770 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
771 } while (ret
< 0 && errno
== EINTR
);
773 * Ret value is useless here, if this pipe gets any actions an
774 * update is required anyway.
776 update_poll_flag
= 1;
780 * New CPU detected by the kernel. Adding kernel stream to
781 * kernel session and updating the kernel consumer
783 if (revents
& LPOLLIN
) {
784 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
790 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
791 * and unregister kernel stream at this point.
800 lttng_poll_clean(&events
);
803 utils_close_pipe(kernel_poll_pipe
);
804 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
807 ERR("Health error occurred in %s", __func__
);
808 WARN("Kernel thread died unexpectedly. "
809 "Kernel tracing can continue but CPU hotplug is disabled.");
812 DBG("Kernel thread dying");
817 * Signal pthread condition of the consumer data that the thread.
819 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
821 pthread_mutex_lock(&data
->cond_mutex
);
824 * The state is set before signaling. It can be any value, it's the waiter
825 * job to correctly interpret this condition variable associated to the
826 * consumer pthread_cond.
828 * A value of 0 means that the corresponding thread of the consumer data
829 * was not started. 1 indicates that the thread has started and is ready
830 * for action. A negative value means that there was an error during the
833 data
->consumer_thread_is_ready
= state
;
834 (void) pthread_cond_signal(&data
->cond
);
836 pthread_mutex_unlock(&data
->cond_mutex
);
840 * This thread manage the consumer error sent back to the session daemon.
842 static void *thread_manage_consumer(void *data
)
844 int sock
= -1, i
, ret
, pollfd
, err
= -1;
845 uint32_t revents
, nb_fd
;
846 enum lttcomm_return_code code
;
847 struct lttng_poll_event events
;
848 struct consumer_data
*consumer_data
= data
;
850 DBG("[thread] Manage consumer started");
852 health_register(HEALTH_TYPE_CONSUMER
);
854 health_code_update();
857 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
858 * Nothing more will be added to this poll set.
860 ret
= create_thread_poll_set(&events
, 2);
866 * The error socket here is already in a listening state which was done
867 * just before spawning this thread to avoid a race between the consumer
868 * daemon exec trying to connect and the listen() call.
870 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
875 health_code_update();
877 /* Inifinite blocking call, waiting for transmission */
881 if (testpoint(thread_manage_consumer
)) {
885 ret
= lttng_poll_wait(&events
, -1);
889 * Restart interrupted system call.
891 if (errno
== EINTR
) {
899 for (i
= 0; i
< nb_fd
; i
++) {
900 /* Fetch once the poll data */
901 revents
= LTTNG_POLL_GETEV(&events
, i
);
902 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
904 health_code_update();
906 /* Thread quit pipe has been closed. Killing thread. */
907 ret
= check_thread_quit_pipe(pollfd
, revents
);
913 /* Event on the registration socket */
914 if (pollfd
== consumer_data
->err_sock
) {
915 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
916 ERR("consumer err socket poll error");
922 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
928 * Set the CLOEXEC flag. Return code is useless because either way, the
931 (void) utils_set_fd_cloexec(sock
);
933 health_code_update();
935 DBG2("Receiving code from consumer err_sock");
937 /* Getting status code from kconsumerd */
938 ret
= lttcomm_recv_unix_sock(sock
, &code
,
939 sizeof(enum lttcomm_return_code
));
944 health_code_update();
946 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
947 consumer_data
->cmd_sock
=
948 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
949 if (consumer_data
->cmd_sock
< 0) {
950 /* On error, signal condition and quit. */
951 signal_consumer_condition(consumer_data
, -1);
952 PERROR("consumer connect");
955 signal_consumer_condition(consumer_data
, 1);
956 DBG("Consumer command socket ready");
958 ERR("consumer error when waiting for SOCK_READY : %s",
959 lttcomm_get_readable_code(-code
));
963 /* Remove the kconsumerd error sock since we've established a connexion */
964 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
969 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
974 health_code_update();
976 /* Inifinite blocking call, waiting for transmission */
979 ret
= lttng_poll_wait(&events
, -1);
983 * Restart interrupted system call.
985 if (errno
== EINTR
) {
993 for (i
= 0; i
< nb_fd
; i
++) {
994 /* Fetch once the poll data */
995 revents
= LTTNG_POLL_GETEV(&events
, i
);
996 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
998 health_code_update();
1000 /* Thread quit pipe has been closed. Killing thread. */
1001 ret
= check_thread_quit_pipe(pollfd
, revents
);
1007 /* Event on the kconsumerd socket */
1008 if (pollfd
== sock
) {
1009 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1010 ERR("consumer err socket second poll error");
1016 health_code_update();
1018 /* Wait for any kconsumerd error */
1019 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1020 sizeof(enum lttcomm_return_code
));
1022 ERR("consumer closed the command socket");
1026 ERR("consumer return code : %s", lttcomm_get_readable_code(-code
));
1030 /* Immediately set the consumerd state to stopped */
1031 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1032 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1033 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1034 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1035 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1037 /* Code flow error... */
1041 if (consumer_data
->err_sock
>= 0) {
1042 ret
= close(consumer_data
->err_sock
);
1047 if (consumer_data
->cmd_sock
>= 0) {
1048 ret
= close(consumer_data
->cmd_sock
);
1060 unlink(consumer_data
->err_unix_sock_path
);
1061 unlink(consumer_data
->cmd_unix_sock_path
);
1062 consumer_data
->pid
= 0;
1064 lttng_poll_clean(&events
);
1068 ERR("Health error occurred in %s", __func__
);
1070 health_unregister();
1071 DBG("consumer thread cleanup completed");
1077 * This thread manage application communication.
1079 static void *thread_manage_apps(void *data
)
1081 int i
, ret
, pollfd
, err
= -1;
1082 uint32_t revents
, nb_fd
;
1083 struct ust_command ust_cmd
;
1084 struct lttng_poll_event events
;
1086 DBG("[thread] Manage application started");
1088 rcu_register_thread();
1089 rcu_thread_online();
1091 health_register(HEALTH_TYPE_APP_MANAGE
);
1093 if (testpoint(thread_manage_apps
)) {
1094 goto error_testpoint
;
1097 health_code_update();
1099 ret
= create_thread_poll_set(&events
, 2);
1101 goto error_poll_create
;
1104 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1109 if (testpoint(thread_manage_apps_before_loop
)) {
1113 health_code_update();
1116 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1118 /* Inifinite blocking call, waiting for transmission */
1120 health_poll_entry();
1121 ret
= lttng_poll_wait(&events
, -1);
1125 * Restart interrupted system call.
1127 if (errno
== EINTR
) {
1135 for (i
= 0; i
< nb_fd
; i
++) {
1136 /* Fetch once the poll data */
1137 revents
= LTTNG_POLL_GETEV(&events
, i
);
1138 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1140 health_code_update();
1142 /* Thread quit pipe has been closed. Killing thread. */
1143 ret
= check_thread_quit_pipe(pollfd
, revents
);
1149 /* Inspect the apps cmd pipe */
1150 if (pollfd
== apps_cmd_pipe
[0]) {
1151 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1152 ERR("Apps command pipe error");
1154 } else if (revents
& LPOLLIN
) {
1157 ret
= read(apps_cmd_pipe
[0], &ust_cmd
, sizeof(ust_cmd
));
1158 } while (ret
< 0 && errno
== EINTR
);
1159 if (ret
< 0 || ret
< sizeof(ust_cmd
)) {
1160 PERROR("read apps cmd pipe");
1164 health_code_update();
1168 * Lock the global session list so from the register up to
1169 * the registration done message, no thread can see the
1170 * application and change its state.
1172 session_lock_list();
1174 /* Register applicaton to the session daemon */
1175 ret
= ust_app_register(&ust_cmd
.reg_msg
,
1177 if (ret
== -ENOMEM
) {
1178 session_unlock_list();
1180 } else if (ret
< 0) {
1181 session_unlock_list();
1185 health_code_update();
1188 * Validate UST version compatibility.
1190 ret
= ust_app_validate_version(ust_cmd
.sock
);
1193 * Add channel(s) and event(s) to newly registered apps
1194 * from lttng global UST domain.
1196 update_ust_app(ust_cmd
.sock
);
1199 health_code_update();
1201 ret
= ust_app_register_done(ust_cmd
.sock
);
1204 * If the registration is not possible, we simply
1205 * unregister the apps and continue
1207 ust_app_unregister(ust_cmd
.sock
);
1210 * We only monitor the error events of the socket. This
1211 * thread does not handle any incoming data from UST
1214 ret
= lttng_poll_add(&events
, ust_cmd
.sock
,
1215 LPOLLERR
& LPOLLHUP
& LPOLLRDHUP
);
1217 session_unlock_list();
1221 /* Set socket timeout for both receiving and ending */
1222 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd
.sock
,
1223 app_socket_timeout
);
1224 (void) lttcomm_setsockopt_snd_timeout(ust_cmd
.sock
,
1225 app_socket_timeout
);
1227 DBG("Apps with sock %d added to poll set",
1230 session_unlock_list();
1232 health_code_update();
1238 * At this point, we know that a registered application made
1239 * the event at poll_wait.
1241 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1242 /* Removing from the poll set */
1243 ret
= lttng_poll_del(&events
, pollfd
);
1248 /* Socket closed on remote end. */
1249 ust_app_unregister(pollfd
);
1254 health_code_update();
1260 lttng_poll_clean(&events
);
1263 utils_close_pipe(apps_cmd_pipe
);
1264 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1267 * We don't clean the UST app hash table here since already registered
1268 * applications can still be controlled so let them be until the session
1269 * daemon dies or the applications stop.
1274 ERR("Health error occurred in %s", __func__
);
1276 health_unregister();
1277 DBG("Application communication apps thread cleanup complete");
1278 rcu_thread_offline();
1279 rcu_unregister_thread();
1284 * Dispatch request from the registration threads to the application
1285 * communication thread.
1287 static void *thread_dispatch_ust_registration(void *data
)
1290 struct cds_wfq_node
*node
;
1291 struct ust_command
*ust_cmd
= NULL
;
1293 DBG("[thread] Dispatch UST command started");
1295 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1296 /* Atomically prepare the queue futex */
1297 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1300 /* Dequeue command for registration */
1301 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1303 DBG("Woken up but nothing in the UST command queue");
1304 /* Continue thread execution */
1308 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1310 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1311 " gid:%d sock:%d name:%s (version %d.%d)",
1312 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1313 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1314 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1315 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1317 * Inform apps thread of the new application registration. This
1318 * call is blocking so we can be assured that the data will be read
1319 * at some point in time or wait to the end of the world :)
1321 if (apps_cmd_pipe
[1] >= 0) {
1323 ret
= write(apps_cmd_pipe
[1], ust_cmd
,
1324 sizeof(struct ust_command
));
1325 } while (ret
< 0 && errno
== EINTR
);
1326 if (ret
< 0 || ret
!= sizeof(struct ust_command
)) {
1327 PERROR("write apps cmd pipe");
1328 if (errno
== EBADF
) {
1330 * We can't inform the application thread to process
1331 * registration. We will exit or else application
1332 * registration will not occur and tracing will never
1339 /* Application manager thread is not available. */
1340 ret
= close(ust_cmd
->sock
);
1342 PERROR("close ust_cmd sock");
1346 } while (node
!= NULL
);
1348 /* Futex wait on queue. Blocking call on futex() */
1349 futex_nto1_wait(&ust_cmd_queue
.futex
);
1353 DBG("Dispatch thread dying");
1358 * This thread manage application registration.
1360 static void *thread_registration_apps(void *data
)
1362 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1363 uint32_t revents
, nb_fd
;
1364 struct lttng_poll_event events
;
1366 * Get allocated in this thread, enqueued to a global queue, dequeued and
1367 * freed in the manage apps thread.
1369 struct ust_command
*ust_cmd
= NULL
;
1371 DBG("[thread] Manage application registration started");
1373 health_register(HEALTH_TYPE_APP_REG
);
1375 if (testpoint(thread_registration_apps
)) {
1376 goto error_testpoint
;
1379 ret
= lttcomm_listen_unix_sock(apps_sock
);
1385 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1386 * more will be added to this poll set.
1388 ret
= create_thread_poll_set(&events
, 2);
1390 goto error_create_poll
;
1393 /* Add the application registration socket */
1394 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1396 goto error_poll_add
;
1399 /* Notify all applications to register */
1400 ret
= notify_ust_apps(1);
1402 ERR("Failed to notify applications or create the wait shared memory.\n"
1403 "Execution continues but there might be problem for already\n"
1404 "running applications that wishes to register.");
1408 DBG("Accepting application registration");
1410 /* Inifinite blocking call, waiting for transmission */
1412 health_poll_entry();
1413 ret
= lttng_poll_wait(&events
, -1);
1417 * Restart interrupted system call.
1419 if (errno
== EINTR
) {
1427 for (i
= 0; i
< nb_fd
; i
++) {
1428 health_code_update();
1430 /* Fetch once the poll data */
1431 revents
= LTTNG_POLL_GETEV(&events
, i
);
1432 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1434 /* Thread quit pipe has been closed. Killing thread. */
1435 ret
= check_thread_quit_pipe(pollfd
, revents
);
1441 /* Event on the registration socket */
1442 if (pollfd
== apps_sock
) {
1443 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1444 ERR("Register apps socket poll error");
1446 } else if (revents
& LPOLLIN
) {
1447 sock
= lttcomm_accept_unix_sock(apps_sock
);
1453 * Set the CLOEXEC flag. Return code is useless because
1454 * either way, the show must go on.
1456 (void) utils_set_fd_cloexec(sock
);
1458 /* Create UST registration command for enqueuing */
1459 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1460 if (ust_cmd
== NULL
) {
1461 PERROR("ust command zmalloc");
1466 * Using message-based transmissions to ensure we don't
1467 * have to deal with partially received messages.
1469 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1471 ERR("Exhausted file descriptors allowed for applications.");
1480 health_code_update();
1481 ret
= lttcomm_recv_unix_sock(sock
, &ust_cmd
->reg_msg
,
1482 sizeof(struct ust_register_msg
));
1483 if (ret
< 0 || ret
< sizeof(struct ust_register_msg
)) {
1485 PERROR("lttcomm_recv_unix_sock register apps");
1487 ERR("Wrong size received on apps register");
1494 lttng_fd_put(LTTNG_FD_APPS
, 1);
1498 health_code_update();
1500 ust_cmd
->sock
= sock
;
1503 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1504 " gid:%d sock:%d name:%s (version %d.%d)",
1505 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1506 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1507 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1508 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1511 * Lock free enqueue the registration request. The red pill
1512 * has been taken! This apps will be part of the *system*.
1514 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1517 * Wake the registration queue futex. Implicit memory
1518 * barrier with the exchange in cds_wfq_enqueue.
1520 futex_nto1_wake(&ust_cmd_queue
.futex
);
1530 ERR("Health error occurred in %s", __func__
);
1533 /* Notify that the registration thread is gone */
1536 if (apps_sock
>= 0) {
1537 ret
= close(apps_sock
);
1547 lttng_fd_put(LTTNG_FD_APPS
, 1);
1549 unlink(apps_unix_sock_path
);
1552 lttng_poll_clean(&events
);
1556 DBG("UST Registration thread cleanup complete");
1557 health_unregister();
1563 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1564 * exec or it will fails.
1566 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1569 struct timespec timeout
;
1571 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1572 consumer_data
->consumer_thread_is_ready
= 0;
1574 /* Setup pthread condition */
1575 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1578 PERROR("pthread_condattr_init consumer data");
1583 * Set the monotonic clock in order to make sure we DO NOT jump in time
1584 * between the clock_gettime() call and the timedwait call. See bug #324
1585 * for a more details and how we noticed it.
1587 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1590 PERROR("pthread_condattr_setclock consumer data");
1594 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1597 PERROR("pthread_cond_init consumer data");
1601 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1604 PERROR("pthread_create consumer");
1609 /* We are about to wait on a pthread condition */
1610 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1612 /* Get time for sem_timedwait absolute timeout */
1613 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1615 * Set the timeout for the condition timed wait even if the clock gettime
1616 * call fails since we might loop on that call and we want to avoid to
1617 * increment the timeout too many times.
1619 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1622 * The following loop COULD be skipped in some conditions so this is why we
1623 * set ret to 0 in order to make sure at least one round of the loop is
1629 * Loop until the condition is reached or when a timeout is reached. Note
1630 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1631 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1632 * possible. This loop does not take any chances and works with both of
1635 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1636 if (clock_ret
< 0) {
1637 PERROR("clock_gettime spawn consumer");
1638 /* Infinite wait for the consumerd thread to be ready */
1639 ret
= pthread_cond_wait(&consumer_data
->cond
,
1640 &consumer_data
->cond_mutex
);
1642 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1643 &consumer_data
->cond_mutex
, &timeout
);
1647 /* Release the pthread condition */
1648 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1652 if (ret
== ETIMEDOUT
) {
1654 * Call has timed out so we kill the kconsumerd_thread and return
1657 ERR("Condition timed out. The consumer thread was never ready."
1659 ret
= pthread_cancel(consumer_data
->thread
);
1661 PERROR("pthread_cancel consumer thread");
1664 PERROR("pthread_cond_wait failed consumer thread");
1669 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1670 if (consumer_data
->pid
== 0) {
1671 ERR("Consumerd did not start");
1672 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1675 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1684 * Join consumer thread
1686 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1690 /* Consumer pid must be a real one. */
1691 if (consumer_data
->pid
> 0) {
1693 ret
= kill(consumer_data
->pid
, SIGTERM
);
1695 ERR("Error killing consumer daemon");
1698 return pthread_join(consumer_data
->thread
, &status
);
1705 * Fork and exec a consumer daemon (consumerd).
1707 * Return pid if successful else -1.
1709 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1713 const char *consumer_to_use
;
1714 const char *verbosity
;
1717 DBG("Spawning consumerd");
1724 if (opt_verbose_consumer
) {
1725 verbosity
= "--verbose";
1727 verbosity
= "--quiet";
1729 switch (consumer_data
->type
) {
1730 case LTTNG_CONSUMER_KERNEL
:
1732 * Find out which consumerd to execute. We will first try the
1733 * 64-bit path, then the sessiond's installation directory, and
1734 * fallback on the 32-bit one,
1736 DBG3("Looking for a kernel consumer at these locations:");
1737 DBG3(" 1) %s", consumerd64_bin
);
1738 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1739 DBG3(" 3) %s", consumerd32_bin
);
1740 if (stat(consumerd64_bin
, &st
) == 0) {
1741 DBG3("Found location #1");
1742 consumer_to_use
= consumerd64_bin
;
1743 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1744 DBG3("Found location #2");
1745 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1746 } else if (stat(consumerd32_bin
, &st
) == 0) {
1747 DBG3("Found location #3");
1748 consumer_to_use
= consumerd32_bin
;
1750 DBG("Could not find any valid consumerd executable");
1753 DBG("Using kernel consumer at: %s", consumer_to_use
);
1754 execl(consumer_to_use
,
1755 "lttng-consumerd", verbosity
, "-k",
1756 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1757 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1760 case LTTNG_CONSUMER64_UST
:
1762 char *tmpnew
= NULL
;
1764 if (consumerd64_libdir
[0] != '\0') {
1768 tmp
= getenv("LD_LIBRARY_PATH");
1772 tmplen
= strlen("LD_LIBRARY_PATH=")
1773 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1774 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1779 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1780 strcat(tmpnew
, consumerd64_libdir
);
1781 if (tmp
[0] != '\0') {
1782 strcat(tmpnew
, ":");
1783 strcat(tmpnew
, tmp
);
1785 ret
= putenv(tmpnew
);
1791 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1792 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1793 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1794 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1796 if (consumerd64_libdir
[0] != '\0') {
1804 case LTTNG_CONSUMER32_UST
:
1806 char *tmpnew
= NULL
;
1808 if (consumerd32_libdir
[0] != '\0') {
1812 tmp
= getenv("LD_LIBRARY_PATH");
1816 tmplen
= strlen("LD_LIBRARY_PATH=")
1817 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1818 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1823 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1824 strcat(tmpnew
, consumerd32_libdir
);
1825 if (tmp
[0] != '\0') {
1826 strcat(tmpnew
, ":");
1827 strcat(tmpnew
, tmp
);
1829 ret
= putenv(tmpnew
);
1835 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
1836 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
1837 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1838 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1840 if (consumerd32_libdir
[0] != '\0') {
1849 PERROR("unknown consumer type");
1853 PERROR("kernel start consumer exec");
1856 } else if (pid
> 0) {
1859 PERROR("start consumer fork");
1867 * Spawn the consumerd daemon and session daemon thread.
1869 static int start_consumerd(struct consumer_data
*consumer_data
)
1874 * Set the listen() state on the socket since there is a possible race
1875 * between the exec() of the consumer daemon and this call if place in the
1876 * consumer thread. See bug #366 for more details.
1878 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
1883 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1884 if (consumer_data
->pid
!= 0) {
1885 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1889 ret
= spawn_consumerd(consumer_data
);
1891 ERR("Spawning consumerd failed");
1892 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1896 /* Setting up the consumer_data pid */
1897 consumer_data
->pid
= ret
;
1898 DBG2("Consumer pid %d", consumer_data
->pid
);
1899 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1901 DBG2("Spawning consumer control thread");
1902 ret
= spawn_consumer_thread(consumer_data
);
1904 ERR("Fatal error spawning consumer control thread");
1912 /* Cleanup already created socket on error. */
1913 if (consumer_data
->err_sock
>= 0) {
1916 err
= close(consumer_data
->err_sock
);
1918 PERROR("close consumer data error socket");
1925 * Compute health status of each consumer. If one of them is zero (bad
1926 * state), we return 0.
1928 static int check_consumer_health(void)
1932 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
1934 DBG3("Health consumer check %d", ret
);
1940 * Setup necessary data for kernel tracer action.
1942 static int init_kernel_tracer(void)
1946 /* Modprobe lttng kernel modules */
1947 ret
= modprobe_lttng_control();
1952 /* Open debugfs lttng */
1953 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1954 if (kernel_tracer_fd
< 0) {
1955 DBG("Failed to open %s", module_proc_lttng
);
1960 /* Validate kernel version */
1961 ret
= kernel_validate_version(kernel_tracer_fd
);
1966 ret
= modprobe_lttng_data();
1971 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1975 modprobe_remove_lttng_control();
1976 ret
= close(kernel_tracer_fd
);
1980 kernel_tracer_fd
= -1;
1981 return LTTNG_ERR_KERN_VERSION
;
1984 ret
= close(kernel_tracer_fd
);
1990 modprobe_remove_lttng_control();
1993 WARN("No kernel tracer available");
1994 kernel_tracer_fd
= -1;
1996 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
1998 return LTTNG_ERR_KERN_NA
;
2004 * Copy consumer output from the tracing session to the domain session. The
2005 * function also applies the right modification on a per domain basis for the
2006 * trace files destination directory.
2008 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2011 const char *dir_name
;
2012 struct consumer_output
*consumer
;
2015 assert(session
->consumer
);
2018 case LTTNG_DOMAIN_KERNEL
:
2019 DBG3("Copying tracing session consumer output in kernel session");
2021 * XXX: We should audit the session creation and what this function
2022 * does "extra" in order to avoid a destroy since this function is used
2023 * in the domain session creation (kernel and ust) only. Same for UST
2026 if (session
->kernel_session
->consumer
) {
2027 consumer_destroy_output(session
->kernel_session
->consumer
);
2029 session
->kernel_session
->consumer
=
2030 consumer_copy_output(session
->consumer
);
2031 /* Ease our life a bit for the next part */
2032 consumer
= session
->kernel_session
->consumer
;
2033 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2035 case LTTNG_DOMAIN_UST
:
2036 DBG3("Copying tracing session consumer output in UST session");
2037 if (session
->ust_session
->consumer
) {
2038 consumer_destroy_output(session
->ust_session
->consumer
);
2040 session
->ust_session
->consumer
=
2041 consumer_copy_output(session
->consumer
);
2042 /* Ease our life a bit for the next part */
2043 consumer
= session
->ust_session
->consumer
;
2044 dir_name
= DEFAULT_UST_TRACE_DIR
;
2047 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2051 /* Append correct directory to subdir */
2052 strncat(consumer
->subdir
, dir_name
,
2053 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2054 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2063 * Create an UST session and add it to the session ust list.
2065 static int create_ust_session(struct ltt_session
*session
,
2066 struct lttng_domain
*domain
)
2069 struct ltt_ust_session
*lus
= NULL
;
2073 assert(session
->consumer
);
2075 switch (domain
->type
) {
2076 case LTTNG_DOMAIN_UST
:
2079 ERR("Unknown UST domain on create session %d", domain
->type
);
2080 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2084 DBG("Creating UST session");
2086 lus
= trace_ust_create_session(session
->path
, session
->id
, domain
);
2088 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2092 lus
->uid
= session
->uid
;
2093 lus
->gid
= session
->gid
;
2094 session
->ust_session
= lus
;
2096 /* Copy session output to the newly created UST session */
2097 ret
= copy_session_consumer(domain
->type
, session
);
2098 if (ret
!= LTTNG_OK
) {
2106 session
->ust_session
= NULL
;
2111 * Create a kernel tracer session then create the default channel.
2113 static int create_kernel_session(struct ltt_session
*session
)
2117 DBG("Creating kernel session");
2119 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2121 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2125 /* Code flow safety */
2126 assert(session
->kernel_session
);
2128 /* Copy session output to the newly created Kernel session */
2129 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2130 if (ret
!= LTTNG_OK
) {
2134 /* Create directory(ies) on local filesystem. */
2135 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2136 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2137 ret
= run_as_mkdir_recursive(
2138 session
->kernel_session
->consumer
->dst
.trace_path
,
2139 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2141 if (ret
!= -EEXIST
) {
2142 ERR("Trace directory creation error");
2148 session
->kernel_session
->uid
= session
->uid
;
2149 session
->kernel_session
->gid
= session
->gid
;
2154 trace_kernel_destroy_session(session
->kernel_session
);
2155 session
->kernel_session
= NULL
;
2160 * Count number of session permitted by uid/gid.
2162 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2165 struct ltt_session
*session
;
2167 DBG("Counting number of available session for UID %d GID %d",
2169 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2171 * Only list the sessions the user can control.
2173 if (!session_access_ok(session
, uid
, gid
)) {
2182 * Process the command requested by the lttng client within the command
2183 * context structure. This function make sure that the return structure (llm)
2184 * is set and ready for transmission before returning.
2186 * Return any error encountered or 0 for success.
2188 * "sock" is only used for special-case var. len data.
2190 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2194 int need_tracing_session
= 1;
2197 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2201 switch (cmd_ctx
->lsm
->cmd_type
) {
2202 case LTTNG_CREATE_SESSION
:
2203 case LTTNG_DESTROY_SESSION
:
2204 case LTTNG_LIST_SESSIONS
:
2205 case LTTNG_LIST_DOMAINS
:
2206 case LTTNG_START_TRACE
:
2207 case LTTNG_STOP_TRACE
:
2208 case LTTNG_DATA_PENDING
:
2215 if (opt_no_kernel
&& need_domain
2216 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2218 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2220 ret
= LTTNG_ERR_KERN_NA
;
2225 /* Deny register consumer if we already have a spawned consumer. */
2226 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2227 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2228 if (kconsumer_data
.pid
> 0) {
2229 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2230 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2233 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2237 * Check for command that don't needs to allocate a returned payload. We do
2238 * this here so we don't have to make the call for no payload at each
2241 switch(cmd_ctx
->lsm
->cmd_type
) {
2242 case LTTNG_LIST_SESSIONS
:
2243 case LTTNG_LIST_TRACEPOINTS
:
2244 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2245 case LTTNG_LIST_DOMAINS
:
2246 case LTTNG_LIST_CHANNELS
:
2247 case LTTNG_LIST_EVENTS
:
2250 /* Setup lttng message with no payload */
2251 ret
= setup_lttng_msg(cmd_ctx
, 0);
2253 /* This label does not try to unlock the session */
2254 goto init_setup_error
;
2258 /* Commands that DO NOT need a session. */
2259 switch (cmd_ctx
->lsm
->cmd_type
) {
2260 case LTTNG_CREATE_SESSION
:
2261 case LTTNG_CALIBRATE
:
2262 case LTTNG_LIST_SESSIONS
:
2263 case LTTNG_LIST_TRACEPOINTS
:
2264 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2265 need_tracing_session
= 0;
2268 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2270 * We keep the session list lock across _all_ commands
2271 * for now, because the per-session lock does not
2272 * handle teardown properly.
2274 session_lock_list();
2275 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2276 if (cmd_ctx
->session
== NULL
) {
2277 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2278 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2280 /* If no session name specified */
2281 ret
= LTTNG_ERR_SELECT_SESS
;
2285 /* Acquire lock for the session */
2286 session_lock(cmd_ctx
->session
);
2296 * Check domain type for specific "pre-action".
2298 switch (cmd_ctx
->lsm
->domain
.type
) {
2299 case LTTNG_DOMAIN_KERNEL
:
2301 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2305 /* Kernel tracer check */
2306 if (kernel_tracer_fd
== -1) {
2307 /* Basically, load kernel tracer modules */
2308 ret
= init_kernel_tracer();
2314 /* Consumer is in an ERROR state. Report back to client */
2315 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2316 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2320 /* Need a session for kernel command */
2321 if (need_tracing_session
) {
2322 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2323 ret
= create_kernel_session(cmd_ctx
->session
);
2325 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2330 /* Start the kernel consumer daemon */
2331 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2332 if (kconsumer_data
.pid
== 0 &&
2333 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2334 cmd_ctx
->session
->start_consumer
) {
2335 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2336 ret
= start_consumerd(&kconsumer_data
);
2338 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2341 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2343 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2347 * The consumer was just spawned so we need to add the socket to
2348 * the consumer output of the session if exist.
2350 ret
= consumer_create_socket(&kconsumer_data
,
2351 cmd_ctx
->session
->kernel_session
->consumer
);
2358 case LTTNG_DOMAIN_UST
:
2360 /* Consumer is in an ERROR state. Report back to client */
2361 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2362 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2366 if (need_tracing_session
) {
2367 /* Create UST session if none exist. */
2368 if (cmd_ctx
->session
->ust_session
== NULL
) {
2369 ret
= create_ust_session(cmd_ctx
->session
,
2370 &cmd_ctx
->lsm
->domain
);
2371 if (ret
!= LTTNG_OK
) {
2376 /* Start the UST consumer daemons */
2378 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2379 if (consumerd64_bin
[0] != '\0' &&
2380 ustconsumer64_data
.pid
== 0 &&
2381 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2382 cmd_ctx
->session
->start_consumer
) {
2383 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2384 ret
= start_consumerd(&ustconsumer64_data
);
2386 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2387 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2391 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2392 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2394 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2398 * Setup socket for consumer 64 bit. No need for atomic access
2399 * since it was set above and can ONLY be set in this thread.
2401 ret
= consumer_create_socket(&ustconsumer64_data
,
2402 cmd_ctx
->session
->ust_session
->consumer
);
2408 if (consumerd32_bin
[0] != '\0' &&
2409 ustconsumer32_data
.pid
== 0 &&
2410 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2411 cmd_ctx
->session
->start_consumer
) {
2412 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2413 ret
= start_consumerd(&ustconsumer32_data
);
2415 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2416 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2420 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2421 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2423 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2427 * Setup socket for consumer 64 bit. No need for atomic access
2428 * since it was set above and can ONLY be set in this thread.
2430 ret
= consumer_create_socket(&ustconsumer32_data
,
2431 cmd_ctx
->session
->ust_session
->consumer
);
2443 /* Validate consumer daemon state when start/stop trace command */
2444 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2445 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2446 switch (cmd_ctx
->lsm
->domain
.type
) {
2447 case LTTNG_DOMAIN_UST
:
2448 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2449 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2453 case LTTNG_DOMAIN_KERNEL
:
2454 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2455 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2463 * Check that the UID or GID match that of the tracing session.
2464 * The root user can interact with all sessions.
2466 if (need_tracing_session
) {
2467 if (!session_access_ok(cmd_ctx
->session
,
2468 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2469 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2470 ret
= LTTNG_ERR_EPERM
;
2475 /* Process by command type */
2476 switch (cmd_ctx
->lsm
->cmd_type
) {
2477 case LTTNG_ADD_CONTEXT
:
2479 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2480 cmd_ctx
->lsm
->u
.context
.channel_name
,
2481 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2484 case LTTNG_DISABLE_CHANNEL
:
2486 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2487 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2490 case LTTNG_DISABLE_EVENT
:
2492 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2493 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2494 cmd_ctx
->lsm
->u
.disable
.name
);
2497 case LTTNG_DISABLE_ALL_EVENT
:
2499 DBG("Disabling all events");
2501 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2502 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2505 case LTTNG_DISABLE_CONSUMER
:
2507 ret
= cmd_disable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2510 case LTTNG_ENABLE_CHANNEL
:
2512 ret
= cmd_enable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2513 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2516 case LTTNG_ENABLE_CONSUMER
:
2519 * XXX: 0 means that this URI should be applied on the session. Should
2520 * be a DOMAIN enuam.
2522 ret
= cmd_enable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2523 if (ret
!= LTTNG_OK
) {
2527 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2528 /* Add the URI for the UST session if a consumer is present. */
2529 if (cmd_ctx
->session
->ust_session
&&
2530 cmd_ctx
->session
->ust_session
->consumer
) {
2531 ret
= cmd_enable_consumer(LTTNG_DOMAIN_UST
, cmd_ctx
->session
);
2532 } else if (cmd_ctx
->session
->kernel_session
&&
2533 cmd_ctx
->session
->kernel_session
->consumer
) {
2534 ret
= cmd_enable_consumer(LTTNG_DOMAIN_KERNEL
,
2540 case LTTNG_ENABLE_EVENT
:
2542 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2543 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2544 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2547 case LTTNG_ENABLE_ALL_EVENT
:
2549 DBG("Enabling all events");
2551 ret
= cmd_enable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2552 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2553 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2556 case LTTNG_LIST_TRACEPOINTS
:
2558 struct lttng_event
*events
;
2561 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2562 if (nb_events
< 0) {
2563 /* Return value is a negative lttng_error_code. */
2569 * Setup lttng message with payload size set to the event list size in
2570 * bytes and then copy list into the llm payload.
2572 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2578 /* Copy event list into message payload */
2579 memcpy(cmd_ctx
->llm
->payload
, events
,
2580 sizeof(struct lttng_event
) * nb_events
);
2587 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2589 struct lttng_event_field
*fields
;
2592 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2594 if (nb_fields
< 0) {
2595 /* Return value is a negative lttng_error_code. */
2601 * Setup lttng message with payload size set to the event list size in
2602 * bytes and then copy list into the llm payload.
2604 ret
= setup_lttng_msg(cmd_ctx
,
2605 sizeof(struct lttng_event_field
) * nb_fields
);
2611 /* Copy event list into message payload */
2612 memcpy(cmd_ctx
->llm
->payload
, fields
,
2613 sizeof(struct lttng_event_field
) * nb_fields
);
2620 case LTTNG_SET_CONSUMER_URI
:
2623 struct lttng_uri
*uris
;
2625 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2626 len
= nb_uri
* sizeof(struct lttng_uri
);
2629 ret
= LTTNG_ERR_INVALID
;
2633 uris
= zmalloc(len
);
2635 ret
= LTTNG_ERR_FATAL
;
2639 /* Receive variable len data */
2640 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2641 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2643 DBG("No URIs received from client... continuing");
2645 ret
= LTTNG_ERR_SESSION_FAIL
;
2650 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2652 if (ret
!= LTTNG_OK
) {
2658 * XXX: 0 means that this URI should be applied on the session. Should
2659 * be a DOMAIN enuam.
2661 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2662 /* Add the URI for the UST session if a consumer is present. */
2663 if (cmd_ctx
->session
->ust_session
&&
2664 cmd_ctx
->session
->ust_session
->consumer
) {
2665 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2667 } else if (cmd_ctx
->session
->kernel_session
&&
2668 cmd_ctx
->session
->kernel_session
->consumer
) {
2669 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2670 cmd_ctx
->session
, nb_uri
, uris
);
2678 case LTTNG_START_TRACE
:
2680 ret
= cmd_start_trace(cmd_ctx
->session
);
2683 case LTTNG_STOP_TRACE
:
2685 ret
= cmd_stop_trace(cmd_ctx
->session
);
2688 case LTTNG_CREATE_SESSION
:
2691 struct lttng_uri
*uris
= NULL
;
2693 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2694 len
= nb_uri
* sizeof(struct lttng_uri
);
2697 uris
= zmalloc(len
);
2699 ret
= LTTNG_ERR_FATAL
;
2703 /* Receive variable len data */
2704 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2705 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2707 DBG("No URIs received from client... continuing");
2709 ret
= LTTNG_ERR_SESSION_FAIL
;
2714 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2715 DBG("Creating session with ONE network URI is a bad call");
2716 ret
= LTTNG_ERR_SESSION_FAIL
;
2722 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2729 case LTTNG_DESTROY_SESSION
:
2731 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2733 /* Set session to NULL so we do not unlock it after free. */
2734 cmd_ctx
->session
= NULL
;
2737 case LTTNG_LIST_DOMAINS
:
2740 struct lttng_domain
*domains
;
2742 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2744 /* Return value is a negative lttng_error_code. */
2749 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2754 /* Copy event list into message payload */
2755 memcpy(cmd_ctx
->llm
->payload
, domains
,
2756 nb_dom
* sizeof(struct lttng_domain
));
2763 case LTTNG_LIST_CHANNELS
:
2766 struct lttng_channel
*channels
;
2768 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2769 cmd_ctx
->session
, &channels
);
2771 /* Return value is a negative lttng_error_code. */
2776 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2781 /* Copy event list into message payload */
2782 memcpy(cmd_ctx
->llm
->payload
, channels
,
2783 nb_chan
* sizeof(struct lttng_channel
));
2790 case LTTNG_LIST_EVENTS
:
2793 struct lttng_event
*events
= NULL
;
2795 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2796 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2798 /* Return value is a negative lttng_error_code. */
2803 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2808 /* Copy event list into message payload */
2809 memcpy(cmd_ctx
->llm
->payload
, events
,
2810 nb_event
* sizeof(struct lttng_event
));
2817 case LTTNG_LIST_SESSIONS
:
2819 unsigned int nr_sessions
;
2821 session_lock_list();
2822 nr_sessions
= lttng_sessions_count(
2823 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2824 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2826 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2828 session_unlock_list();
2832 /* Filled the session array */
2833 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2834 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2835 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2837 session_unlock_list();
2842 case LTTNG_CALIBRATE
:
2844 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
2845 &cmd_ctx
->lsm
->u
.calibrate
);
2848 case LTTNG_REGISTER_CONSUMER
:
2850 struct consumer_data
*cdata
;
2852 switch (cmd_ctx
->lsm
->domain
.type
) {
2853 case LTTNG_DOMAIN_KERNEL
:
2854 cdata
= &kconsumer_data
;
2857 ret
= LTTNG_ERR_UND
;
2861 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2862 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
2865 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
2867 struct lttng_filter_bytecode
*bytecode
;
2869 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
2870 ret
= LTTNG_ERR_FILTER_INVAL
;
2873 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
2874 ret
= LTTNG_ERR_FILTER_INVAL
;
2877 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2879 ret
= LTTNG_ERR_FILTER_NOMEM
;
2882 /* Receive var. len. data */
2883 DBG("Receiving var len data from client ...");
2884 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
2885 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2887 DBG("Nothing recv() from client var len data... continuing");
2889 ret
= LTTNG_ERR_FILTER_INVAL
;
2893 if (bytecode
->len
+ sizeof(*bytecode
)
2894 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
2896 ret
= LTTNG_ERR_FILTER_INVAL
;
2900 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2901 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2902 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
2905 case LTTNG_DATA_PENDING
:
2907 ret
= cmd_data_pending(cmd_ctx
->session
);
2911 ret
= LTTNG_ERR_UND
;
2916 if (cmd_ctx
->llm
== NULL
) {
2917 DBG("Missing llm structure. Allocating one.");
2918 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
2922 /* Set return code */
2923 cmd_ctx
->llm
->ret_code
= ret
;
2925 if (cmd_ctx
->session
) {
2926 session_unlock(cmd_ctx
->session
);
2928 if (need_tracing_session
) {
2929 session_unlock_list();
2936 * Thread managing health check socket.
2938 static void *thread_manage_health(void *data
)
2940 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
2941 uint32_t revents
, nb_fd
;
2942 struct lttng_poll_event events
;
2943 struct lttcomm_health_msg msg
;
2944 struct lttcomm_health_data reply
;
2946 DBG("[thread] Manage health check started");
2948 rcu_register_thread();
2950 /* Create unix socket */
2951 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
2953 ERR("Unable to create health check Unix socket");
2959 * Set the CLOEXEC flag. Return code is useless because either way, the
2962 (void) utils_set_fd_cloexec(sock
);
2964 ret
= lttcomm_listen_unix_sock(sock
);
2970 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2971 * more will be added to this poll set.
2973 ret
= create_thread_poll_set(&events
, 2);
2978 /* Add the application registration socket */
2979 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
2985 DBG("Health check ready");
2987 /* Inifinite blocking call, waiting for transmission */
2989 ret
= lttng_poll_wait(&events
, -1);
2992 * Restart interrupted system call.
2994 if (errno
== EINTR
) {
3002 for (i
= 0; i
< nb_fd
; i
++) {
3003 /* Fetch once the poll data */
3004 revents
= LTTNG_POLL_GETEV(&events
, i
);
3005 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3007 /* Thread quit pipe has been closed. Killing thread. */
3008 ret
= check_thread_quit_pipe(pollfd
, revents
);
3014 /* Event on the registration socket */
3015 if (pollfd
== sock
) {
3016 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3017 ERR("Health socket poll error");
3023 new_sock
= lttcomm_accept_unix_sock(sock
);
3029 * Set the CLOEXEC flag. Return code is useless because either way, the
3032 (void) utils_set_fd_cloexec(new_sock
);
3034 DBG("Receiving data from client for health...");
3035 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3037 DBG("Nothing recv() from client... continuing");
3038 ret
= close(new_sock
);
3046 rcu_thread_online();
3048 switch (msg
.component
) {
3049 case LTTNG_HEALTH_CMD
:
3050 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3052 case LTTNG_HEALTH_APP_MANAGE
:
3053 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3055 case LTTNG_HEALTH_APP_REG
:
3056 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3058 case LTTNG_HEALTH_KERNEL
:
3059 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3061 case LTTNG_HEALTH_CONSUMER
:
3062 reply
.ret_code
= check_consumer_health();
3064 case LTTNG_HEALTH_ALL
:
3066 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3067 health_check_state(HEALTH_TYPE_APP_REG
) &&
3068 health_check_state(HEALTH_TYPE_CMD
) &&
3069 health_check_state(HEALTH_TYPE_KERNEL
) &&
3070 check_consumer_health();
3073 reply
.ret_code
= LTTNG_ERR_UND
;
3078 * Flip ret value since 0 is a success and 1 indicates a bad health for
3079 * the client where in the sessiond it is the opposite. Again, this is
3080 * just to make things easier for us poor developer which enjoy a lot
3083 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3084 reply
.ret_code
= !reply
.ret_code
;
3087 DBG2("Health check return value %d", reply
.ret_code
);
3089 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3091 ERR("Failed to send health data back to client");
3094 /* End of transmission */
3095 ret
= close(new_sock
);
3105 ERR("Health error occurred in %s", __func__
);
3107 DBG("Health check thread dying");
3108 unlink(health_unix_sock_path
);
3115 if (new_sock
>= 0) {
3116 ret
= close(new_sock
);
3122 lttng_poll_clean(&events
);
3124 rcu_unregister_thread();
3129 * This thread manage all clients request using the unix client socket for
3132 static void *thread_manage_clients(void *data
)
3134 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3136 uint32_t revents
, nb_fd
;
3137 struct command_ctx
*cmd_ctx
= NULL
;
3138 struct lttng_poll_event events
;
3140 DBG("[thread] Manage client started");
3142 rcu_register_thread();
3144 health_register(HEALTH_TYPE_CMD
);
3146 if (testpoint(thread_manage_clients
)) {
3147 goto error_testpoint
;
3150 health_code_update();
3152 ret
= lttcomm_listen_unix_sock(client_sock
);
3158 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3159 * more will be added to this poll set.
3161 ret
= create_thread_poll_set(&events
, 2);
3163 goto error_create_poll
;
3166 /* Add the application registration socket */
3167 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3173 * Notify parent pid that we are ready to accept command for client side.
3175 if (opt_sig_parent
) {
3176 kill(ppid
, SIGUSR1
);
3179 if (testpoint(thread_manage_clients_before_loop
)) {
3183 health_code_update();
3186 DBG("Accepting client command ...");
3188 /* Inifinite blocking call, waiting for transmission */
3190 health_poll_entry();
3191 ret
= lttng_poll_wait(&events
, -1);
3195 * Restart interrupted system call.
3197 if (errno
== EINTR
) {
3205 for (i
= 0; i
< nb_fd
; i
++) {
3206 /* Fetch once the poll data */
3207 revents
= LTTNG_POLL_GETEV(&events
, i
);
3208 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3210 health_code_update();
3212 /* Thread quit pipe has been closed. Killing thread. */
3213 ret
= check_thread_quit_pipe(pollfd
, revents
);
3219 /* Event on the registration socket */
3220 if (pollfd
== client_sock
) {
3221 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3222 ERR("Client socket poll error");
3228 DBG("Wait for client response");
3230 health_code_update();
3232 sock
= lttcomm_accept_unix_sock(client_sock
);
3238 * Set the CLOEXEC flag. Return code is useless because either way, the
3241 (void) utils_set_fd_cloexec(sock
);
3243 /* Set socket option for credentials retrieval */
3244 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3249 /* Allocate context command to process the client request */
3250 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3251 if (cmd_ctx
== NULL
) {
3252 PERROR("zmalloc cmd_ctx");
3256 /* Allocate data buffer for reception */
3257 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3258 if (cmd_ctx
->lsm
== NULL
) {
3259 PERROR("zmalloc cmd_ctx->lsm");
3263 cmd_ctx
->llm
= NULL
;
3264 cmd_ctx
->session
= NULL
;
3266 health_code_update();
3269 * Data is received from the lttng client. The struct
3270 * lttcomm_session_msg (lsm) contains the command and data request of
3273 DBG("Receiving data from client ...");
3274 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3275 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3277 DBG("Nothing recv() from client... continuing");
3283 clean_command_ctx(&cmd_ctx
);
3287 health_code_update();
3289 // TODO: Validate cmd_ctx including sanity check for
3290 // security purpose.
3292 rcu_thread_online();
3294 * This function dispatch the work to the kernel or userspace tracer
3295 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3296 * informations for the client. The command context struct contains
3297 * everything this function may needs.
3299 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3300 rcu_thread_offline();
3310 * TODO: Inform client somehow of the fatal error. At
3311 * this point, ret < 0 means that a zmalloc failed
3312 * (ENOMEM). Error detected but still accept
3313 * command, unless a socket error has been
3316 clean_command_ctx(&cmd_ctx
);
3320 health_code_update();
3322 DBG("Sending response (size: %d, retcode: %s)",
3323 cmd_ctx
->lttng_msg_size
,
3324 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3325 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3327 ERR("Failed to send data back to client");
3330 /* End of transmission */
3337 clean_command_ctx(&cmd_ctx
);
3339 health_code_update();
3351 lttng_poll_clean(&events
);
3352 clean_command_ctx(&cmd_ctx
);
3357 unlink(client_unix_sock_path
);
3358 if (client_sock
>= 0) {
3359 ret
= close(client_sock
);
3367 ERR("Health error occurred in %s", __func__
);
3370 health_unregister();
3372 DBG("Client thread dying");
3374 rcu_unregister_thread();
3380 * usage function on stderr
3382 static void usage(void)
3384 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3385 fprintf(stderr
, " -h, --help Display this usage.\n");
3386 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3387 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3388 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3389 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3390 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3391 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3392 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3393 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3394 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3395 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3396 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3397 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3398 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3399 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3400 fprintf(stderr
, " -V, --version Show version number.\n");
3401 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3402 fprintf(stderr
, " -q, --quiet No output at all.\n");
3403 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3404 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3405 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3409 * daemon argument parsing
3411 static int parse_args(int argc
, char **argv
)
3415 static struct option long_options
[] = {
3416 { "client-sock", 1, 0, 'c' },
3417 { "apps-sock", 1, 0, 'a' },
3418 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3419 { "kconsumerd-err-sock", 1, 0, 'E' },
3420 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3421 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3422 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3423 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3424 { "consumerd32-path", 1, 0, 'u' },
3425 { "consumerd32-libdir", 1, 0, 'U' },
3426 { "consumerd64-path", 1, 0, 't' },
3427 { "consumerd64-libdir", 1, 0, 'T' },
3428 { "daemonize", 0, 0, 'd' },
3429 { "sig-parent", 0, 0, 'S' },
3430 { "help", 0, 0, 'h' },
3431 { "group", 1, 0, 'g' },
3432 { "version", 0, 0, 'V' },
3433 { "quiet", 0, 0, 'q' },
3434 { "verbose", 0, 0, 'v' },
3435 { "verbose-consumer", 0, 0, 'Z' },
3436 { "no-kernel", 0, 0, 'N' },
3441 int option_index
= 0;
3442 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3443 long_options
, &option_index
);
3450 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3452 fprintf(stderr
, " with arg %s\n", optarg
);
3456 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3459 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3465 opt_tracing_group
= optarg
;
3471 fprintf(stdout
, "%s\n", VERSION
);
3477 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3480 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3483 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3486 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3489 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3492 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3498 lttng_opt_quiet
= 1;
3501 /* Verbose level can increase using multiple -v */
3502 lttng_opt_verbose
+= 1;
3505 opt_verbose_consumer
+= 1;
3508 consumerd32_bin
= optarg
;
3511 consumerd32_libdir
= optarg
;
3514 consumerd64_bin
= optarg
;
3517 consumerd64_libdir
= optarg
;
3520 /* Unknown option or other error.
3521 * Error is printed by getopt, just return */
3530 * Creates the two needed socket by the daemon.
3531 * apps_sock - The communication socket for all UST apps.
3532 * client_sock - The communication of the cli tool (lttng).
3534 static int init_daemon_socket(void)
3539 old_umask
= umask(0);
3541 /* Create client tool unix socket */
3542 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3543 if (client_sock
< 0) {
3544 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3549 /* Set the cloexec flag */
3550 ret
= utils_set_fd_cloexec(client_sock
);
3552 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3553 "Continuing but note that the consumer daemon will have a "
3554 "reference to this socket on exec()", client_sock
);
3557 /* File permission MUST be 660 */
3558 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3560 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3565 /* Create the application unix socket */
3566 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3567 if (apps_sock
< 0) {
3568 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3573 /* Set the cloexec flag */
3574 ret
= utils_set_fd_cloexec(apps_sock
);
3576 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3577 "Continuing but note that the consumer daemon will have a "
3578 "reference to this socket on exec()", apps_sock
);
3581 /* File permission MUST be 666 */
3582 ret
= chmod(apps_unix_sock_path
,
3583 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3585 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3590 DBG3("Session daemon client socket %d and application socket %d created",
3591 client_sock
, apps_sock
);
3599 * Check if the global socket is available, and if a daemon is answering at the
3600 * other side. If yes, error is returned.
3602 static int check_existing_daemon(void)
3604 /* Is there anybody out there ? */
3605 if (lttng_session_daemon_alive()) {
3613 * Set the tracing group gid onto the client socket.
3615 * Race window between mkdir and chown is OK because we are going from more
3616 * permissive (root.root) to less permissive (root.tracing).
3618 static int set_permissions(char *rundir
)
3623 ret
= allowed_group();
3625 WARN("No tracing group detected");
3632 /* Set lttng run dir */
3633 ret
= chown(rundir
, 0, gid
);
3635 ERR("Unable to set group on %s", rundir
);
3639 /* Ensure tracing group can search the run dir */
3640 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3642 ERR("Unable to set permissions on %s", rundir
);
3646 /* lttng client socket path */
3647 ret
= chown(client_unix_sock_path
, 0, gid
);
3649 ERR("Unable to set group on %s", client_unix_sock_path
);
3653 /* kconsumer error socket path */
3654 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3656 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3660 /* 64-bit ustconsumer error socket path */
3661 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3663 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3667 /* 32-bit ustconsumer compat32 error socket path */
3668 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3670 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3674 DBG("All permissions are set");
3681 * Create the lttng run directory needed for all global sockets and pipe.
3683 static int create_lttng_rundir(const char *rundir
)
3687 DBG3("Creating LTTng run directory: %s", rundir
);
3689 ret
= mkdir(rundir
, S_IRWXU
);
3691 if (errno
!= EEXIST
) {
3692 ERR("Unable to create %s", rundir
);
3704 * Setup sockets and directory needed by the kconsumerd communication with the
3707 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3711 char path
[PATH_MAX
];
3713 switch (consumer_data
->type
) {
3714 case LTTNG_CONSUMER_KERNEL
:
3715 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3717 case LTTNG_CONSUMER64_UST
:
3718 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3720 case LTTNG_CONSUMER32_UST
:
3721 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3724 ERR("Consumer type unknown");
3729 DBG2("Creating consumer directory: %s", path
);
3731 ret
= mkdir(path
, S_IRWXU
);
3733 if (errno
!= EEXIST
) {
3735 ERR("Failed to create %s", path
);
3741 /* Create the kconsumerd error unix socket */
3742 consumer_data
->err_sock
=
3743 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3744 if (consumer_data
->err_sock
< 0) {
3745 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3750 /* File permission MUST be 660 */
3751 ret
= chmod(consumer_data
->err_unix_sock_path
,
3752 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3754 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3764 * Signal handler for the daemon
3766 * Simply stop all worker threads, leaving main() return gracefully after
3767 * joining all threads and calling cleanup().
3769 static void sighandler(int sig
)
3773 DBG("SIGPIPE caught");
3776 DBG("SIGINT caught");
3780 DBG("SIGTERM caught");
3789 * Setup signal handler for :
3790 * SIGINT, SIGTERM, SIGPIPE
3792 static int set_signal_handler(void)
3795 struct sigaction sa
;
3798 if ((ret
= sigemptyset(&sigset
)) < 0) {
3799 PERROR("sigemptyset");
3803 sa
.sa_handler
= sighandler
;
3804 sa
.sa_mask
= sigset
;
3806 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3807 PERROR("sigaction");
3811 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3812 PERROR("sigaction");
3816 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3817 PERROR("sigaction");
3821 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3827 * Set open files limit to unlimited. This daemon can open a large number of
3828 * file descriptors in order to consumer multiple kernel traces.
3830 static void set_ulimit(void)
3835 /* The kernel does not allowed an infinite limit for open files */
3836 lim
.rlim_cur
= 65535;
3837 lim
.rlim_max
= 65535;
3839 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
3841 PERROR("failed to set open files limit");
3848 int main(int argc
, char **argv
)
3852 const char *home_path
, *env_app_timeout
;
3854 init_kernel_workarounds();
3856 rcu_register_thread();
3858 setup_consumerd_path();
3860 /* Parse arguments */
3862 if ((ret
= parse_args(argc
, argv
)) < 0) {
3872 * child: setsid, close FD 0, 1, 2, chdir /
3873 * parent: exit (if fork is successful)
3881 * We are in the child. Make sure all other file
3882 * descriptors are closed, in case we are called with
3883 * more opened file descriptors than the standard ones.
3885 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3890 /* Create thread quit pipe */
3891 if ((ret
= init_thread_quit_pipe()) < 0) {
3895 /* Check if daemon is UID = 0 */
3896 is_root
= !getuid();
3899 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
3901 /* Create global run dir with root access */
3902 ret
= create_lttng_rundir(rundir
);
3907 if (strlen(apps_unix_sock_path
) == 0) {
3908 snprintf(apps_unix_sock_path
, PATH_MAX
,
3909 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
3912 if (strlen(client_unix_sock_path
) == 0) {
3913 snprintf(client_unix_sock_path
, PATH_MAX
,
3914 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
3917 /* Set global SHM for ust */
3918 if (strlen(wait_shm_path
) == 0) {
3919 snprintf(wait_shm_path
, PATH_MAX
,
3920 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
3923 if (strlen(health_unix_sock_path
) == 0) {
3924 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3925 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
3928 /* Setup kernel consumerd path */
3929 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
3930 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
3931 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
3932 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
3934 DBG2("Kernel consumer err path: %s",
3935 kconsumer_data
.err_unix_sock_path
);
3936 DBG2("Kernel consumer cmd path: %s",
3937 kconsumer_data
.cmd_unix_sock_path
);
3939 home_path
= get_home_dir();
3940 if (home_path
== NULL
) {
3941 /* TODO: Add --socket PATH option */
3942 ERR("Can't get HOME directory for sockets creation.");
3948 * Create rundir from home path. This will create something like
3951 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
3957 ret
= create_lttng_rundir(rundir
);
3962 if (strlen(apps_unix_sock_path
) == 0) {
3963 snprintf(apps_unix_sock_path
, PATH_MAX
,
3964 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
3967 /* Set the cli tool unix socket path */
3968 if (strlen(client_unix_sock_path
) == 0) {
3969 snprintf(client_unix_sock_path
, PATH_MAX
,
3970 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
3973 /* Set global SHM for ust */
3974 if (strlen(wait_shm_path
) == 0) {
3975 snprintf(wait_shm_path
, PATH_MAX
,
3976 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, geteuid());
3979 /* Set health check Unix path */
3980 if (strlen(health_unix_sock_path
) == 0) {
3981 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3982 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
3986 /* Set consumer initial state */
3987 kernel_consumerd_state
= CONSUMER_STOPPED
;
3988 ust_consumerd_state
= CONSUMER_STOPPED
;
3990 DBG("Client socket path %s", client_unix_sock_path
);
3991 DBG("Application socket path %s", apps_unix_sock_path
);
3992 DBG("LTTng run directory path: %s", rundir
);
3994 /* 32 bits consumerd path setup */
3995 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
3996 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
3997 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
3998 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4000 DBG2("UST consumer 32 bits err path: %s",
4001 ustconsumer32_data
.err_unix_sock_path
);
4002 DBG2("UST consumer 32 bits cmd path: %s",
4003 ustconsumer32_data
.cmd_unix_sock_path
);
4005 /* 64 bits consumerd path setup */
4006 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4007 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4008 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4009 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4011 DBG2("UST consumer 64 bits err path: %s",
4012 ustconsumer64_data
.err_unix_sock_path
);
4013 DBG2("UST consumer 64 bits cmd path: %s",
4014 ustconsumer64_data
.cmd_unix_sock_path
);
4017 * See if daemon already exist.
4019 if ((ret
= check_existing_daemon()) < 0) {
4020 ERR("Already running daemon.\n");
4022 * We do not goto exit because we must not cleanup()
4023 * because a daemon is already running.
4029 * Init UST app hash table. Alloc hash table before this point since
4030 * cleanup() can get called after that point.
4034 /* After this point, we can safely call cleanup() with "goto exit" */
4037 * These actions must be executed as root. We do that *after* setting up
4038 * the sockets path because we MUST make the check for another daemon using
4039 * those paths *before* trying to set the kernel consumer sockets and init
4043 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4048 /* Setup kernel tracer */
4049 if (!opt_no_kernel
) {
4050 init_kernel_tracer();
4053 /* Set ulimit for open files */
4056 /* init lttng_fd tracking must be done after set_ulimit. */
4059 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4064 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4069 if ((ret
= set_signal_handler()) < 0) {
4073 /* Setup the needed unix socket */
4074 if ((ret
= init_daemon_socket()) < 0) {
4078 /* Set credentials to socket */
4079 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4083 /* Get parent pid if -S, --sig-parent is specified. */
4084 if (opt_sig_parent
) {
4088 /* Setup the kernel pipe for waking up the kernel thread */
4089 if (is_root
&& !opt_no_kernel
) {
4090 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4095 /* Setup the thread apps communication pipe. */
4096 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4100 /* Init UST command queue. */
4101 cds_wfq_init(&ust_cmd_queue
.queue
);
4104 * Get session list pointer. This pointer MUST NOT be free(). This list is
4105 * statically declared in session.c
4107 session_list_ptr
= session_get_list();
4109 /* Set up max poll set size */
4110 lttng_poll_set_max_size();
4114 /* Check for the application socket timeout env variable. */
4115 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4116 if (env_app_timeout
) {
4117 app_socket_timeout
= atoi(env_app_timeout
);
4119 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4122 /* Create thread to manage the client socket */
4123 ret
= pthread_create(&health_thread
, NULL
,
4124 thread_manage_health
, (void *) NULL
);
4126 PERROR("pthread_create health");
4130 /* Create thread to manage the client socket */
4131 ret
= pthread_create(&client_thread
, NULL
,
4132 thread_manage_clients
, (void *) NULL
);
4134 PERROR("pthread_create clients");
4138 /* Create thread to dispatch registration */
4139 ret
= pthread_create(&dispatch_thread
, NULL
,
4140 thread_dispatch_ust_registration
, (void *) NULL
);
4142 PERROR("pthread_create dispatch");
4146 /* Create thread to manage application registration. */
4147 ret
= pthread_create(®_apps_thread
, NULL
,
4148 thread_registration_apps
, (void *) NULL
);
4150 PERROR("pthread_create registration");
4154 /* Create thread to manage application socket */
4155 ret
= pthread_create(&apps_thread
, NULL
,
4156 thread_manage_apps
, (void *) NULL
);
4158 PERROR("pthread_create apps");
4162 /* Don't start this thread if kernel tracing is not requested nor root */
4163 if (is_root
&& !opt_no_kernel
) {
4164 /* Create kernel thread to manage kernel event */
4165 ret
= pthread_create(&kernel_thread
, NULL
,
4166 thread_manage_kernel
, (void *) NULL
);
4168 PERROR("pthread_create kernel");
4172 ret
= pthread_join(kernel_thread
, &status
);
4174 PERROR("pthread_join");
4175 goto error
; /* join error, exit without cleanup */
4180 ret
= pthread_join(apps_thread
, &status
);
4182 PERROR("pthread_join");
4183 goto error
; /* join error, exit without cleanup */
4187 ret
= pthread_join(reg_apps_thread
, &status
);
4189 PERROR("pthread_join");
4190 goto error
; /* join error, exit without cleanup */
4194 ret
= pthread_join(dispatch_thread
, &status
);
4196 PERROR("pthread_join");
4197 goto error
; /* join error, exit without cleanup */
4201 ret
= pthread_join(client_thread
, &status
);
4203 PERROR("pthread_join");
4204 goto error
; /* join error, exit without cleanup */
4207 ret
= join_consumer_thread(&kconsumer_data
);
4209 PERROR("join_consumer");
4210 goto error
; /* join error, exit without cleanup */
4213 ret
= join_consumer_thread(&ustconsumer32_data
);
4215 PERROR("join_consumer ust32");
4216 goto error
; /* join error, exit without cleanup */
4219 ret
= join_consumer_thread(&ustconsumer64_data
);
4221 PERROR("join_consumer ust64");
4222 goto error
; /* join error, exit without cleanup */
4226 ret
= pthread_join(health_thread
, &status
);
4228 PERROR("pthread_join health thread");
4229 goto error
; /* join error, exit without cleanup */
4235 * cleanup() is called when no other thread is running.
4237 rcu_thread_online();
4239 rcu_thread_offline();
4240 rcu_unregister_thread();