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"
64 #include "testpoint.h"
66 #define CONSUMERD_FILE "lttng-consumerd"
69 const char default_home_dir
[] = DEFAULT_HOME_DIR
;
70 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
71 const char default_ust_sock_dir
[] = DEFAULT_UST_SOCK_DIR
;
72 const char default_global_apps_pipe
[] = DEFAULT_GLOBAL_APPS_PIPE
;
75 const char *opt_tracing_group
;
76 static int opt_sig_parent
;
77 static int opt_verbose_consumer
;
78 static int opt_daemon
;
79 static int opt_no_kernel
;
80 static int is_root
; /* Set to 1 if the daemon is running as root */
81 static pid_t ppid
; /* Parent PID for --sig-parent option */
85 * Consumer daemon specific control data. Every value not initialized here is
86 * set to 0 by the static definition.
88 static struct consumer_data kconsumer_data
= {
89 .type
= LTTNG_CONSUMER_KERNEL
,
90 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
91 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
94 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
95 .lock
= PTHREAD_MUTEX_INITIALIZER
,
96 .cond
= PTHREAD_COND_INITIALIZER
,
97 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
99 static struct consumer_data ustconsumer64_data
= {
100 .type
= LTTNG_CONSUMER64_UST
,
101 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
102 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
105 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
106 .lock
= PTHREAD_MUTEX_INITIALIZER
,
107 .cond
= PTHREAD_COND_INITIALIZER
,
108 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
110 static struct consumer_data ustconsumer32_data
= {
111 .type
= LTTNG_CONSUMER32_UST
,
112 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
113 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
116 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
117 .lock
= PTHREAD_MUTEX_INITIALIZER
,
118 .cond
= PTHREAD_COND_INITIALIZER
,
119 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
122 /* Shared between threads */
123 static int dispatch_thread_exit
;
125 /* Global application Unix socket path */
126 static char apps_unix_sock_path
[PATH_MAX
];
127 /* Global client Unix socket path */
128 static char client_unix_sock_path
[PATH_MAX
];
129 /* global wait shm path for UST */
130 static char wait_shm_path
[PATH_MAX
];
131 /* Global health check unix path */
132 static char health_unix_sock_path
[PATH_MAX
];
134 /* Sockets and FDs */
135 static int client_sock
= -1;
136 static int apps_sock
= -1;
137 int kernel_tracer_fd
= -1;
138 static int kernel_poll_pipe
[2] = { -1, -1 };
141 * Quit pipe for all threads. This permits a single cancellation point
142 * for all threads when receiving an event on the pipe.
144 static int thread_quit_pipe
[2] = { -1, -1 };
147 * This pipe is used to inform the thread managing application communication
148 * that a command is queued and ready to be processed.
150 static int apps_cmd_pipe
[2] = { -1, -1 };
152 /* Pthread, Mutexes and Semaphores */
153 static pthread_t apps_thread
;
154 static pthread_t reg_apps_thread
;
155 static pthread_t client_thread
;
156 static pthread_t kernel_thread
;
157 static pthread_t dispatch_thread
;
158 static pthread_t health_thread
;
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
167 static struct ust_cmd_queue ust_cmd_queue
;
170 * Pointer initialized before thread creation.
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
177 * session_lock_list and session_unlock_list for lock acquisition.
179 static struct ltt_session_list
*session_list_ptr
;
181 int ust_consumerd64_fd
= -1;
182 int ust_consumerd32_fd
= -1;
184 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
185 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
186 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
187 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
189 static const char *module_proc_lttng
= "/proc/lttng";
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
195 enum consumerd_state
{
196 CONSUMER_STARTED
= 1,
197 CONSUMER_STOPPED
= 2,
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
207 * The following example shows a possible race condition of this scheme:
209 * consumer thread error happens
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
222 static enum consumerd_state ust_consumerd_state
;
223 static enum consumerd_state kernel_consumerd_state
;
225 /* Used for the health monitoring of the session daemon. See health.h */
226 struct health_state health_thread_cmd
;
227 struct health_state health_thread_app_manage
;
228 struct health_state health_thread_app_reg
;
229 struct health_state health_thread_kernel
;
232 * Socket timeout for receiving and sending in seconds.
234 static int app_socket_timeout
;
237 void setup_consumerd_path(void)
239 const char *bin
, *libdir
;
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin
[0]) {
248 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
250 if (!consumerd32_libdir
[0]) {
251 consumerd32_libdir
= INSTALL_LIB_PATH
;
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin
[0]) {
255 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
257 if (!consumerd64_libdir
[0]) {
258 consumerd64_libdir
= INSTALL_LIB_PATH
;
261 #error "Unknown bitness"
265 * runtime env. var. overrides the build default.
267 bin
= getenv("LTTNG_CONSUMERD32_BIN");
269 consumerd32_bin
= bin
;
271 bin
= getenv("LTTNG_CONSUMERD64_BIN");
273 consumerd64_bin
= bin
;
275 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
277 consumerd32_libdir
= libdir
;
279 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
281 consumerd64_libdir
= libdir
;
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
288 static int create_thread_poll_set(struct lttng_poll_event
*events
,
293 if (events
== NULL
|| size
== 0) {
298 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
304 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
316 * Check if the thread quit pipe was triggered.
318 * Return 1 if it was triggered else 0;
320 static int check_thread_quit_pipe(int fd
, uint32_t events
)
322 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
330 * Return group ID of the tracing group or -1 if not found.
332 static gid_t
allowed_group(void)
336 if (opt_tracing_group
) {
337 grp
= getgrnam(opt_tracing_group
);
339 grp
= getgrnam(default_tracing_group
);
349 * Init thread quit pipe.
351 * Return -1 on error or 0 if all pipes are created.
353 static int init_thread_quit_pipe(void)
357 ret
= pipe(thread_quit_pipe
);
359 PERROR("thread quit pipe");
363 for (i
= 0; i
< 2; i
++) {
364 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
376 * Stop all threads by closing the thread quit pipe.
378 static void stop_threads(void)
382 /* Stopping all threads */
383 DBG("Terminating all threads");
384 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
386 ERR("write error on thread quit pipe");
389 /* Dispatch thread */
390 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
391 futex_nto1_wake(&ust_cmd_queue
.futex
);
397 static void cleanup(void)
401 struct ltt_session
*sess
, *stmp
;
405 /* First thing first, stop all threads */
406 utils_close_pipe(thread_quit_pipe
);
408 DBG("Removing %s directory", rundir
);
409 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
411 ERR("asprintf failed. Something is really wrong!");
414 /* Remove lttng run directory */
417 ERR("Unable to clean %s", rundir
);
422 DBG("Cleaning up all sessions");
424 /* Destroy session list mutex */
425 if (session_list_ptr
!= NULL
) {
426 pthread_mutex_destroy(&session_list_ptr
->lock
);
428 /* Cleanup ALL session */
429 cds_list_for_each_entry_safe(sess
, stmp
,
430 &session_list_ptr
->head
, list
) {
431 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
435 DBG("Closing all UST sockets");
436 ust_app_clean_list();
438 if (is_root
&& !opt_no_kernel
) {
439 DBG2("Closing kernel fd");
440 if (kernel_tracer_fd
>= 0) {
441 ret
= close(kernel_tracer_fd
);
446 DBG("Unloading kernel modules");
447 modprobe_remove_lttng_all();
451 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
452 "Matthew, BEET driven development works!%c[%dm",
453 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
458 * Send data on a unix socket using the liblttsessiondcomm API.
460 * Return lttcomm error code.
462 static int send_unix_sock(int sock
, void *buf
, size_t len
)
464 /* Check valid length */
469 return lttcomm_send_unix_sock(sock
, buf
, len
);
473 * Free memory of a command context structure.
475 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
477 DBG("Clean command context structure");
479 if ((*cmd_ctx
)->llm
) {
480 free((*cmd_ctx
)->llm
);
482 if ((*cmd_ctx
)->lsm
) {
483 free((*cmd_ctx
)->lsm
);
491 * Notify UST applications using the shm mmap futex.
493 static int notify_ust_apps(int active
)
497 DBG("Notifying applications of session daemon state: %d", active
);
499 /* See shm.c for this call implying mmap, shm and futex calls */
500 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
501 if (wait_shm_mmap
== NULL
) {
505 /* Wake waiting process */
506 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
508 /* Apps notified successfully */
516 * Setup the outgoing data buffer for the response (llm) by allocating the
517 * right amount of memory and copying the original information from the lsm
520 * Return total size of the buffer pointed by buf.
522 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
528 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
529 if (cmd_ctx
->llm
== NULL
) {
535 /* Copy common data */
536 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
537 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
539 cmd_ctx
->llm
->data_size
= size
;
540 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
549 * Update the kernel poll set of all channel fd available over all tracing
550 * session. Add the wakeup pipe at the end of the set.
552 static int update_kernel_poll(struct lttng_poll_event
*events
)
555 struct ltt_session
*session
;
556 struct ltt_kernel_channel
*channel
;
558 DBG("Updating kernel poll set");
561 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
562 session_lock(session
);
563 if (session
->kernel_session
== NULL
) {
564 session_unlock(session
);
568 cds_list_for_each_entry(channel
,
569 &session
->kernel_session
->channel_list
.head
, list
) {
570 /* Add channel fd to the kernel poll set */
571 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
573 session_unlock(session
);
576 DBG("Channel fd %d added to kernel set", channel
->fd
);
578 session_unlock(session
);
580 session_unlock_list();
585 session_unlock_list();
590 * Find the channel fd from 'fd' over all tracing session. When found, check
591 * for new channel stream and send those stream fds to the kernel consumer.
593 * Useful for CPU hotplug feature.
595 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
598 struct ltt_session
*session
;
599 struct ltt_kernel_session
*ksess
;
600 struct ltt_kernel_channel
*channel
;
602 DBG("Updating kernel streams for channel fd %d", fd
);
605 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
606 session_lock(session
);
607 if (session
->kernel_session
== NULL
) {
608 session_unlock(session
);
611 ksess
= session
->kernel_session
;
613 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
614 if (channel
->fd
== fd
) {
615 DBG("Channel found, updating kernel streams");
616 ret
= kernel_open_channel_stream(channel
);
622 * Have we already sent fds to the consumer? If yes, it means
623 * that tracing is started so it is safe to send our updated
626 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
627 struct lttng_ht_iter iter
;
628 struct consumer_socket
*socket
;
631 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
632 &iter
.iter
, socket
, node
.node
) {
633 /* Code flow error */
634 assert(socket
->fd
>= 0);
636 pthread_mutex_lock(socket
->lock
);
637 ret
= kernel_consumer_send_channel_stream(socket
->fd
,
639 pthread_mutex_unlock(socket
->lock
);
648 session_unlock(session
);
650 session_unlock_list();
654 session_unlock(session
);
655 session_unlock_list();
660 * For each tracing session, update newly registered apps.
662 static void update_ust_app(int app_sock
)
664 struct ltt_session
*sess
, *stmp
;
668 /* For all tracing session(s) */
669 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
671 if (sess
->ust_session
) {
672 ust_app_global_update(sess
->ust_session
, app_sock
);
674 session_unlock(sess
);
677 session_unlock_list();
681 * This thread manage event coming from the kernel.
683 * Features supported in this thread:
686 static void *thread_manage_kernel(void *data
)
688 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
689 uint32_t revents
, nb_fd
;
691 struct lttng_poll_event events
;
693 DBG("Thread manage kernel started");
695 testpoint(thread_manage_kernel
);
697 health_code_update(&health_thread_kernel
);
699 testpoint(thread_manage_kernel_before_loop
);
701 ret
= create_thread_poll_set(&events
, 2);
703 goto error_poll_create
;
706 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
712 health_code_update(&health_thread_kernel
);
714 if (update_poll_flag
== 1) {
716 * Reset number of fd in the poll set. Always 2 since there is the thread
717 * quit pipe and the kernel pipe.
721 ret
= update_kernel_poll(&events
);
725 update_poll_flag
= 0;
728 nb_fd
= LTTNG_POLL_GETNB(&events
);
730 DBG("Thread kernel polling on %d fds", nb_fd
);
732 /* Zeroed the poll events */
733 lttng_poll_reset(&events
);
735 /* Poll infinite value of time */
737 health_poll_update(&health_thread_kernel
);
738 ret
= lttng_poll_wait(&events
, -1);
739 health_poll_update(&health_thread_kernel
);
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...");
755 for (i
= 0; i
< nb_fd
; i
++) {
756 /* Fetch once the poll data */
757 revents
= LTTNG_POLL_GETEV(&events
, i
);
758 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
760 health_code_update(&health_thread_kernel
);
762 /* Thread quit pipe has been closed. Killing thread. */
763 ret
= check_thread_quit_pipe(pollfd
, revents
);
769 /* Check for data on kernel pipe */
770 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
771 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
772 update_poll_flag
= 1;
776 * New CPU detected by the kernel. Adding kernel stream to
777 * kernel session and updating the kernel consumer
779 if (revents
& LPOLLIN
) {
780 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
786 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
787 * and unregister kernel stream at this point.
796 lttng_poll_clean(&events
);
798 utils_close_pipe(kernel_poll_pipe
);
799 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
801 health_error(&health_thread_kernel
);
802 ERR("Health error occurred in %s", __func__
);
803 WARN("Kernel thread died unexpectedly. "
804 "Kernel tracing can continue but CPU hotplug is disabled.");
806 health_exit(&health_thread_kernel
);
807 DBG("Kernel thread dying");
812 * Signal pthread condition of the consumer data that the thread.
814 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
816 pthread_mutex_lock(&data
->cond_mutex
);
819 * The state is set before signaling. It can be any value, it's the waiter
820 * job to correctly interpret this condition variable associated to the
821 * consumer pthread_cond.
823 * A value of 0 means that the corresponding thread of the consumer data
824 * was not started. 1 indicates that the thread has started and is ready
825 * for action. A negative value means that there was an error during the
828 data
->consumer_thread_is_ready
= state
;
829 (void) pthread_cond_signal(&data
->cond
);
831 pthread_mutex_unlock(&data
->cond_mutex
);
835 * This thread manage the consumer error sent back to the session daemon.
837 static void *thread_manage_consumer(void *data
)
839 int sock
= -1, i
, ret
, pollfd
, err
= -1;
840 uint32_t revents
, nb_fd
;
841 enum lttcomm_return_code code
;
842 struct lttng_poll_event events
;
843 struct consumer_data
*consumer_data
= data
;
845 DBG("[thread] Manage consumer started");
848 * Since the consumer thread can be spawned at any moment in time, we init
849 * the health to a poll status (1, which is a valid health over time).
850 * When the thread starts, we update here the health to a "code" path being
851 * an even value so this thread, when reaching a poll wait, does not
852 * trigger an error with an even value.
854 * Here is the use case we avoid.
856 * +1: the first poll update during initialization (main())
857 * +2 * x: multiple code update once in this thread.
858 * +1: poll wait in this thread (being a good health state).
859 * == even number which after the wait period shows as a bad health.
861 * In a nutshell, the following poll update to the health state brings back
862 * the state to an even value meaning a code path.
864 health_poll_update(&consumer_data
->health
);
867 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
868 * Nothing more will be added to this poll set.
870 ret
= create_thread_poll_set(&events
, 2);
876 * The error socket here is already in a listening state which was done
877 * just before spawning this thread to avoid a race between the consumer
878 * daemon exec trying to connect and the listen() call.
880 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
885 nb_fd
= LTTNG_POLL_GETNB(&events
);
887 health_code_update(&consumer_data
->health
);
889 /* Inifinite blocking call, waiting for transmission */
891 health_poll_update(&consumer_data
->health
);
893 testpoint(thread_manage_consumer
);
895 ret
= lttng_poll_wait(&events
, -1);
896 health_poll_update(&consumer_data
->health
);
899 * Restart interrupted system call.
901 if (errno
== EINTR
) {
907 for (i
= 0; i
< nb_fd
; i
++) {
908 /* Fetch once the poll data */
909 revents
= LTTNG_POLL_GETEV(&events
, i
);
910 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
912 health_code_update(&consumer_data
->health
);
914 /* Thread quit pipe has been closed. Killing thread. */
915 ret
= check_thread_quit_pipe(pollfd
, revents
);
921 /* Event on the registration socket */
922 if (pollfd
== consumer_data
->err_sock
) {
923 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
924 ERR("consumer err socket poll error");
930 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
936 * Set the CLOEXEC flag. Return code is useless because either way, the
939 (void) utils_set_fd_cloexec(sock
);
941 health_code_update(&consumer_data
->health
);
943 DBG2("Receiving code from consumer err_sock");
945 /* Getting status code from kconsumerd */
946 ret
= lttcomm_recv_unix_sock(sock
, &code
,
947 sizeof(enum lttcomm_return_code
));
952 health_code_update(&consumer_data
->health
);
954 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
955 consumer_data
->cmd_sock
=
956 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
957 if (consumer_data
->cmd_sock
< 0) {
958 /* On error, signal condition and quit. */
959 signal_consumer_condition(consumer_data
, -1);
960 PERROR("consumer connect");
963 signal_consumer_condition(consumer_data
, 1);
964 DBG("Consumer command socket ready");
966 ERR("consumer error when waiting for SOCK_READY : %s",
967 lttcomm_get_readable_code(-code
));
971 /* Remove the kconsumerd error sock since we've established a connexion */
972 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
977 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
982 health_code_update(&consumer_data
->health
);
984 /* Update number of fd */
985 nb_fd
= LTTNG_POLL_GETNB(&events
);
987 /* Inifinite blocking call, waiting for transmission */
989 health_poll_update(&consumer_data
->health
);
990 ret
= lttng_poll_wait(&events
, -1);
991 health_poll_update(&consumer_data
->health
);
994 * Restart interrupted system call.
996 if (errno
== EINTR
) {
1002 for (i
= 0; i
< nb_fd
; i
++) {
1003 /* Fetch once the poll data */
1004 revents
= LTTNG_POLL_GETEV(&events
, i
);
1005 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1007 health_code_update(&consumer_data
->health
);
1009 /* Thread quit pipe has been closed. Killing thread. */
1010 ret
= check_thread_quit_pipe(pollfd
, revents
);
1016 /* Event on the kconsumerd socket */
1017 if (pollfd
== sock
) {
1018 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1019 ERR("consumer err socket second poll error");
1025 health_code_update(&consumer_data
->health
);
1027 /* Wait for any kconsumerd error */
1028 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1029 sizeof(enum lttcomm_return_code
));
1031 ERR("consumer closed the command socket");
1035 ERR("consumer return code : %s", lttcomm_get_readable_code(-code
));
1039 /* Immediately set the consumerd state to stopped */
1040 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1041 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1042 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1043 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1044 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1046 /* Code flow error... */
1050 if (consumer_data
->err_sock
>= 0) {
1051 ret
= close(consumer_data
->err_sock
);
1056 if (consumer_data
->cmd_sock
>= 0) {
1057 ret
= close(consumer_data
->cmd_sock
);
1069 unlink(consumer_data
->err_unix_sock_path
);
1070 unlink(consumer_data
->cmd_unix_sock_path
);
1071 consumer_data
->pid
= 0;
1073 lttng_poll_clean(&events
);
1076 health_error(&consumer_data
->health
);
1077 ERR("Health error occurred in %s", __func__
);
1079 health_exit(&consumer_data
->health
);
1080 DBG("consumer thread cleanup completed");
1086 * This thread manage application communication.
1088 static void *thread_manage_apps(void *data
)
1090 int i
, ret
, pollfd
, err
= -1;
1091 uint32_t revents
, nb_fd
;
1092 struct ust_command ust_cmd
;
1093 struct lttng_poll_event events
;
1095 DBG("[thread] Manage application started");
1097 testpoint(thread_manage_apps
);
1099 rcu_register_thread();
1100 rcu_thread_online();
1102 health_code_update(&health_thread_app_manage
);
1104 ret
= create_thread_poll_set(&events
, 2);
1106 goto error_poll_create
;
1109 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1114 testpoint(thread_manage_apps_before_loop
);
1116 health_code_update(&health_thread_app_manage
);
1119 /* Zeroed the events structure */
1120 lttng_poll_reset(&events
);
1122 nb_fd
= LTTNG_POLL_GETNB(&events
);
1124 DBG("Apps thread polling on %d fds", nb_fd
);
1126 /* Inifinite blocking call, waiting for transmission */
1128 health_poll_update(&health_thread_app_manage
);
1129 ret
= lttng_poll_wait(&events
, -1);
1130 health_poll_update(&health_thread_app_manage
);
1133 * Restart interrupted system call.
1135 if (errno
== EINTR
) {
1141 for (i
= 0; i
< nb_fd
; i
++) {
1142 /* Fetch once the poll data */
1143 revents
= LTTNG_POLL_GETEV(&events
, i
);
1144 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1146 health_code_update(&health_thread_app_manage
);
1148 /* Thread quit pipe has been closed. Killing thread. */
1149 ret
= check_thread_quit_pipe(pollfd
, revents
);
1155 /* Inspect the apps cmd pipe */
1156 if (pollfd
== apps_cmd_pipe
[0]) {
1157 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1158 ERR("Apps command pipe error");
1160 } else if (revents
& LPOLLIN
) {
1162 ret
= read(apps_cmd_pipe
[0], &ust_cmd
, sizeof(ust_cmd
));
1163 if (ret
< 0 || ret
< sizeof(ust_cmd
)) {
1164 PERROR("read apps cmd pipe");
1168 health_code_update(&health_thread_app_manage
);
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(&health_thread_app_manage
);
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(&health_thread_app_manage
);
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(&health_thread_app_manage
);
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(&health_thread_app_manage
);
1252 lttng_poll_clean(&events
);
1254 utils_close_pipe(apps_cmd_pipe
);
1255 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1258 * We don't clean the UST app hash table here since already registered
1259 * applications can still be controlled so let them be until the session
1260 * daemon dies or the applications stop.
1264 health_error(&health_thread_app_manage
);
1265 ERR("Health error occurred in %s", __func__
);
1267 health_exit(&health_thread_app_manage
);
1268 DBG("Application communication apps thread cleanup complete");
1269 rcu_thread_offline();
1270 rcu_unregister_thread();
1275 * Dispatch request from the registration threads to the application
1276 * communication thread.
1278 static void *thread_dispatch_ust_registration(void *data
)
1281 struct cds_wfq_node
*node
;
1282 struct ust_command
*ust_cmd
= NULL
;
1284 DBG("[thread] Dispatch UST command started");
1286 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1287 /* Atomically prepare the queue futex */
1288 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1291 /* Dequeue command for registration */
1292 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1294 DBG("Woken up but nothing in the UST command queue");
1295 /* Continue thread execution */
1299 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1301 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1302 " gid:%d sock:%d name:%s (version %d.%d)",
1303 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1304 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1305 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1306 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1308 * Inform apps thread of the new application registration. This
1309 * call is blocking so we can be assured that the data will be read
1310 * at some point in time or wait to the end of the world :)
1312 if (apps_cmd_pipe
[1] >= 0) {
1313 ret
= write(apps_cmd_pipe
[1], ust_cmd
,
1314 sizeof(struct ust_command
));
1316 PERROR("write apps cmd pipe");
1317 if (errno
== EBADF
) {
1319 * We can't inform the application thread to process
1320 * registration. We will exit or else application
1321 * registration will not occur and tracing will never
1328 /* Application manager thread is not available. */
1329 ret
= close(ust_cmd
->sock
);
1331 PERROR("close ust_cmd sock");
1335 } while (node
!= NULL
);
1337 /* Futex wait on queue. Blocking call on futex() */
1338 futex_nto1_wait(&ust_cmd_queue
.futex
);
1342 DBG("Dispatch thread dying");
1347 * This thread manage application registration.
1349 static void *thread_registration_apps(void *data
)
1351 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1352 uint32_t revents
, nb_fd
;
1353 struct lttng_poll_event events
;
1355 * Get allocated in this thread, enqueued to a global queue, dequeued and
1356 * freed in the manage apps thread.
1358 struct ust_command
*ust_cmd
= NULL
;
1360 DBG("[thread] Manage application registration started");
1362 testpoint(thread_registration_apps
);
1364 ret
= lttcomm_listen_unix_sock(apps_sock
);
1370 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1371 * more will be added to this poll set.
1373 ret
= create_thread_poll_set(&events
, 2);
1375 goto error_create_poll
;
1378 /* Add the application registration socket */
1379 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1381 goto error_poll_add
;
1384 /* Notify all applications to register */
1385 ret
= notify_ust_apps(1);
1387 ERR("Failed to notify applications or create the wait shared memory.\n"
1388 "Execution continues but there might be problem for already\n"
1389 "running applications that wishes to register.");
1393 DBG("Accepting application registration");
1395 nb_fd
= LTTNG_POLL_GETNB(&events
);
1397 /* Inifinite blocking call, waiting for transmission */
1399 health_poll_update(&health_thread_app_reg
);
1400 ret
= lttng_poll_wait(&events
, -1);
1401 health_poll_update(&health_thread_app_reg
);
1404 * Restart interrupted system call.
1406 if (errno
== EINTR
) {
1412 for (i
= 0; i
< nb_fd
; i
++) {
1413 health_code_update(&health_thread_app_reg
);
1415 /* Fetch once the poll data */
1416 revents
= LTTNG_POLL_GETEV(&events
, i
);
1417 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1419 /* Thread quit pipe has been closed. Killing thread. */
1420 ret
= check_thread_quit_pipe(pollfd
, revents
);
1426 /* Event on the registration socket */
1427 if (pollfd
== apps_sock
) {
1428 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1429 ERR("Register apps socket poll error");
1431 } else if (revents
& LPOLLIN
) {
1432 sock
= lttcomm_accept_unix_sock(apps_sock
);
1438 * Set the CLOEXEC flag. Return code is useless because
1439 * either way, the show must go on.
1441 (void) utils_set_fd_cloexec(sock
);
1443 /* Create UST registration command for enqueuing */
1444 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1445 if (ust_cmd
== NULL
) {
1446 PERROR("ust command zmalloc");
1451 * Using message-based transmissions to ensure we don't
1452 * have to deal with partially received messages.
1454 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1456 ERR("Exhausted file descriptors allowed for applications.");
1465 health_code_update(&health_thread_app_reg
);
1466 ret
= lttcomm_recv_unix_sock(sock
, &ust_cmd
->reg_msg
,
1467 sizeof(struct ust_register_msg
));
1468 if (ret
< 0 || ret
< sizeof(struct ust_register_msg
)) {
1470 PERROR("lttcomm_recv_unix_sock register apps");
1472 ERR("Wrong size received on apps register");
1479 lttng_fd_put(LTTNG_FD_APPS
, 1);
1483 health_code_update(&health_thread_app_reg
);
1485 ust_cmd
->sock
= sock
;
1488 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1489 " gid:%d sock:%d name:%s (version %d.%d)",
1490 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1491 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1492 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1493 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1496 * Lock free enqueue the registration request. The red pill
1497 * has been taken! This apps will be part of the *system*.
1499 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1502 * Wake the registration queue futex. Implicit memory
1503 * barrier with the exchange in cds_wfq_enqueue.
1505 futex_nto1_wake(&ust_cmd_queue
.futex
);
1514 health_error(&health_thread_app_reg
);
1515 ERR("Health error occurred in %s", __func__
);
1517 health_exit(&health_thread_app_reg
);
1519 /* Notify that the registration thread is gone */
1522 if (apps_sock
>= 0) {
1523 ret
= close(apps_sock
);
1533 lttng_fd_put(LTTNG_FD_APPS
, 1);
1535 unlink(apps_unix_sock_path
);
1538 lttng_poll_clean(&events
);
1541 DBG("UST Registration thread cleanup complete");
1547 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1548 * exec or it will fails.
1550 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1553 struct timespec timeout
;
1555 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1556 consumer_data
->consumer_thread_is_ready
= 0;
1558 /* Setup pthread condition */
1559 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1562 PERROR("pthread_condattr_init consumer data");
1567 * Set the monotonic clock in order to make sure we DO NOT jump in time
1568 * between the clock_gettime() call and the timedwait call. See bug #324
1569 * for a more details and how we noticed it.
1571 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1574 PERROR("pthread_condattr_setclock consumer data");
1578 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1581 PERROR("pthread_cond_init consumer data");
1585 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1588 PERROR("pthread_create consumer");
1593 /* We are about to wait on a pthread condition */
1594 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1596 /* Get time for sem_timedwait absolute timeout */
1597 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1599 * Set the timeout for the condition timed wait even if the clock gettime
1600 * call fails since we might loop on that call and we want to avoid to
1601 * increment the timeout too many times.
1603 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1606 * The following loop COULD be skipped in some conditions so this is why we
1607 * set ret to 0 in order to make sure at least one round of the loop is
1613 * Loop until the condition is reached or when a timeout is reached. Note
1614 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1615 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1616 * possible. This loop does not take any chances and works with both of
1619 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1620 if (clock_ret
< 0) {
1621 PERROR("clock_gettime spawn consumer");
1622 /* Infinite wait for the consumerd thread to be ready */
1623 ret
= pthread_cond_wait(&consumer_data
->cond
,
1624 &consumer_data
->cond_mutex
);
1626 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1627 &consumer_data
->cond_mutex
, &timeout
);
1631 /* Release the pthread condition */
1632 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1636 if (ret
== ETIMEDOUT
) {
1638 * Call has timed out so we kill the kconsumerd_thread and return
1641 ERR("Condition timed out. The consumer thread was never ready."
1643 ret
= pthread_cancel(consumer_data
->thread
);
1645 PERROR("pthread_cancel consumer thread");
1648 PERROR("pthread_cond_wait failed consumer thread");
1653 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1654 if (consumer_data
->pid
== 0) {
1655 ERR("Consumerd did not start");
1656 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1659 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1668 * Join consumer thread
1670 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1675 /* Consumer pid must be a real one. */
1676 if (consumer_data
->pid
> 0) {
1677 ret
= kill(consumer_data
->pid
, SIGTERM
);
1679 ERR("Error killing consumer daemon");
1682 return pthread_join(consumer_data
->thread
, &status
);
1689 * Fork and exec a consumer daemon (consumerd).
1691 * Return pid if successful else -1.
1693 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1697 const char *consumer_to_use
;
1698 const char *verbosity
;
1701 DBG("Spawning consumerd");
1708 if (opt_verbose_consumer
) {
1709 verbosity
= "--verbose";
1711 verbosity
= "--quiet";
1713 switch (consumer_data
->type
) {
1714 case LTTNG_CONSUMER_KERNEL
:
1716 * Find out which consumerd to execute. We will first try the
1717 * 64-bit path, then the sessiond's installation directory, and
1718 * fallback on the 32-bit one,
1720 DBG3("Looking for a kernel consumer at these locations:");
1721 DBG3(" 1) %s", consumerd64_bin
);
1722 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1723 DBG3(" 3) %s", consumerd32_bin
);
1724 if (stat(consumerd64_bin
, &st
) == 0) {
1725 DBG3("Found location #1");
1726 consumer_to_use
= consumerd64_bin
;
1727 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1728 DBG3("Found location #2");
1729 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1730 } else if (stat(consumerd32_bin
, &st
) == 0) {
1731 DBG3("Found location #3");
1732 consumer_to_use
= consumerd32_bin
;
1734 DBG("Could not find any valid consumerd executable");
1737 DBG("Using kernel consumer at: %s", consumer_to_use
);
1738 execl(consumer_to_use
,
1739 "lttng-consumerd", verbosity
, "-k",
1740 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1741 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1744 case LTTNG_CONSUMER64_UST
:
1746 char *tmpnew
= NULL
;
1748 if (consumerd64_libdir
[0] != '\0') {
1752 tmp
= getenv("LD_LIBRARY_PATH");
1756 tmplen
= strlen("LD_LIBRARY_PATH=")
1757 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1758 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1763 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1764 strcat(tmpnew
, consumerd64_libdir
);
1765 if (tmp
[0] != '\0') {
1766 strcat(tmpnew
, ":");
1767 strcat(tmpnew
, tmp
);
1769 ret
= putenv(tmpnew
);
1775 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1776 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1777 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1778 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1780 if (consumerd64_libdir
[0] != '\0') {
1788 case LTTNG_CONSUMER32_UST
:
1790 char *tmpnew
= NULL
;
1792 if (consumerd32_libdir
[0] != '\0') {
1796 tmp
= getenv("LD_LIBRARY_PATH");
1800 tmplen
= strlen("LD_LIBRARY_PATH=")
1801 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1802 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1807 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1808 strcat(tmpnew
, consumerd32_libdir
);
1809 if (tmp
[0] != '\0') {
1810 strcat(tmpnew
, ":");
1811 strcat(tmpnew
, tmp
);
1813 ret
= putenv(tmpnew
);
1819 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
1820 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
1821 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1822 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1824 if (consumerd32_libdir
[0] != '\0') {
1833 PERROR("unknown consumer type");
1837 PERROR("kernel start consumer exec");
1840 } else if (pid
> 0) {
1843 PERROR("start consumer fork");
1851 * Spawn the consumerd daemon and session daemon thread.
1853 static int start_consumerd(struct consumer_data
*consumer_data
)
1858 * Set the listen() state on the socket since there is a possible race
1859 * between the exec() of the consumer daemon and this call if place in the
1860 * consumer thread. See bug #366 for more details.
1862 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
1867 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1868 if (consumer_data
->pid
!= 0) {
1869 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1873 ret
= spawn_consumerd(consumer_data
);
1875 ERR("Spawning consumerd failed");
1876 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1880 /* Setting up the consumer_data pid */
1881 consumer_data
->pid
= ret
;
1882 DBG2("Consumer pid %d", consumer_data
->pid
);
1883 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1885 DBG2("Spawning consumer control thread");
1886 ret
= spawn_consumer_thread(consumer_data
);
1888 ERR("Fatal error spawning consumer control thread");
1896 /* Cleanup already created socket on error. */
1897 if (consumer_data
->err_sock
>= 0) {
1898 err
= close(consumer_data
->err_sock
);
1900 PERROR("close consumer data error socket");
1907 * Compute health status of each consumer. If one of them is zero (bad
1908 * state), we return 0.
1910 static int check_consumer_health(void)
1914 ret
= health_check_state(&kconsumer_data
.health
) &&
1915 health_check_state(&ustconsumer32_data
.health
) &&
1916 health_check_state(&ustconsumer64_data
.health
);
1918 DBG3("Health consumer check %d", ret
);
1924 * Setup necessary data for kernel tracer action.
1926 static int init_kernel_tracer(void)
1930 /* Modprobe lttng kernel modules */
1931 ret
= modprobe_lttng_control();
1936 /* Open debugfs lttng */
1937 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1938 if (kernel_tracer_fd
< 0) {
1939 DBG("Failed to open %s", module_proc_lttng
);
1944 /* Validate kernel version */
1945 ret
= kernel_validate_version(kernel_tracer_fd
);
1950 ret
= modprobe_lttng_data();
1955 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1959 modprobe_remove_lttng_control();
1960 ret
= close(kernel_tracer_fd
);
1964 kernel_tracer_fd
= -1;
1965 return LTTNG_ERR_KERN_VERSION
;
1968 ret
= close(kernel_tracer_fd
);
1974 modprobe_remove_lttng_control();
1977 WARN("No kernel tracer available");
1978 kernel_tracer_fd
= -1;
1980 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
1982 return LTTNG_ERR_KERN_NA
;
1988 * Copy consumer output from the tracing session to the domain session. The
1989 * function also applies the right modification on a per domain basis for the
1990 * trace files destination directory.
1992 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
1995 const char *dir_name
;
1996 struct consumer_output
*consumer
;
1999 assert(session
->consumer
);
2002 case LTTNG_DOMAIN_KERNEL
:
2003 DBG3("Copying tracing session consumer output in kernel session");
2005 * XXX: We should audit the session creation and what this function
2006 * does "extra" in order to avoid a destroy since this function is used
2007 * in the domain session creation (kernel and ust) only. Same for UST
2010 if (session
->kernel_session
->consumer
) {
2011 consumer_destroy_output(session
->kernel_session
->consumer
);
2013 session
->kernel_session
->consumer
=
2014 consumer_copy_output(session
->consumer
);
2015 /* Ease our life a bit for the next part */
2016 consumer
= session
->kernel_session
->consumer
;
2017 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2019 case LTTNG_DOMAIN_UST
:
2020 DBG3("Copying tracing session consumer output in UST session");
2021 if (session
->ust_session
->consumer
) {
2022 consumer_destroy_output(session
->ust_session
->consumer
);
2024 session
->ust_session
->consumer
=
2025 consumer_copy_output(session
->consumer
);
2026 /* Ease our life a bit for the next part */
2027 consumer
= session
->ust_session
->consumer
;
2028 dir_name
= DEFAULT_UST_TRACE_DIR
;
2031 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2035 /* Append correct directory to subdir */
2036 strncat(consumer
->subdir
, dir_name
,
2037 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2038 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2047 * Create an UST session and add it to the session ust list.
2049 static int create_ust_session(struct ltt_session
*session
,
2050 struct lttng_domain
*domain
)
2053 struct ltt_ust_session
*lus
= NULL
;
2057 assert(session
->consumer
);
2059 switch (domain
->type
) {
2060 case LTTNG_DOMAIN_UST
:
2063 ERR("Unknown UST domain on create session %d", domain
->type
);
2064 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2068 DBG("Creating UST session");
2070 lus
= trace_ust_create_session(session
->path
, session
->id
, domain
);
2072 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2076 lus
->uid
= session
->uid
;
2077 lus
->gid
= session
->gid
;
2078 session
->ust_session
= lus
;
2080 /* Copy session output to the newly created UST session */
2081 ret
= copy_session_consumer(domain
->type
, session
);
2082 if (ret
!= LTTNG_OK
) {
2090 session
->ust_session
= NULL
;
2095 * Create a kernel tracer session then create the default channel.
2097 static int create_kernel_session(struct ltt_session
*session
)
2101 DBG("Creating kernel session");
2103 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2105 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2109 /* Code flow safety */
2110 assert(session
->kernel_session
);
2112 /* Copy session output to the newly created Kernel session */
2113 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2114 if (ret
!= LTTNG_OK
) {
2118 /* Create directory(ies) on local filesystem. */
2119 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2120 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2121 ret
= run_as_mkdir_recursive(
2122 session
->kernel_session
->consumer
->dst
.trace_path
,
2123 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2125 if (ret
!= -EEXIST
) {
2126 ERR("Trace directory creation error");
2132 session
->kernel_session
->uid
= session
->uid
;
2133 session
->kernel_session
->gid
= session
->gid
;
2138 trace_kernel_destroy_session(session
->kernel_session
);
2139 session
->kernel_session
= NULL
;
2144 * Count number of session permitted by uid/gid.
2146 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2149 struct ltt_session
*session
;
2151 DBG("Counting number of available session for UID %d GID %d",
2153 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2155 * Only list the sessions the user can control.
2157 if (!session_access_ok(session
, uid
, gid
)) {
2166 * Process the command requested by the lttng client within the command
2167 * context structure. This function make sure that the return structure (llm)
2168 * is set and ready for transmission before returning.
2170 * Return any error encountered or 0 for success.
2172 * "sock" is only used for special-case var. len data.
2174 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2178 int need_tracing_session
= 1;
2181 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2185 switch (cmd_ctx
->lsm
->cmd_type
) {
2186 case LTTNG_CREATE_SESSION
:
2187 case LTTNG_DESTROY_SESSION
:
2188 case LTTNG_LIST_SESSIONS
:
2189 case LTTNG_LIST_DOMAINS
:
2190 case LTTNG_START_TRACE
:
2191 case LTTNG_STOP_TRACE
:
2192 case LTTNG_DATA_PENDING
:
2199 if (opt_no_kernel
&& need_domain
2200 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2202 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2204 ret
= LTTNG_ERR_KERN_NA
;
2209 /* Deny register consumer if we already have a spawned consumer. */
2210 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2211 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2212 if (kconsumer_data
.pid
> 0) {
2213 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2214 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2217 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2221 * Check for command that don't needs to allocate a returned payload. We do
2222 * this here so we don't have to make the call for no payload at each
2225 switch(cmd_ctx
->lsm
->cmd_type
) {
2226 case LTTNG_LIST_SESSIONS
:
2227 case LTTNG_LIST_TRACEPOINTS
:
2228 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2229 case LTTNG_LIST_DOMAINS
:
2230 case LTTNG_LIST_CHANNELS
:
2231 case LTTNG_LIST_EVENTS
:
2234 /* Setup lttng message with no payload */
2235 ret
= setup_lttng_msg(cmd_ctx
, 0);
2237 /* This label does not try to unlock the session */
2238 goto init_setup_error
;
2242 /* Commands that DO NOT need a session. */
2243 switch (cmd_ctx
->lsm
->cmd_type
) {
2244 case LTTNG_CREATE_SESSION
:
2245 case LTTNG_CALIBRATE
:
2246 case LTTNG_LIST_SESSIONS
:
2247 case LTTNG_LIST_TRACEPOINTS
:
2248 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2249 need_tracing_session
= 0;
2252 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2254 * We keep the session list lock across _all_ commands
2255 * for now, because the per-session lock does not
2256 * handle teardown properly.
2258 session_lock_list();
2259 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2260 if (cmd_ctx
->session
== NULL
) {
2261 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2262 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2264 /* If no session name specified */
2265 ret
= LTTNG_ERR_SELECT_SESS
;
2269 /* Acquire lock for the session */
2270 session_lock(cmd_ctx
->session
);
2280 * Check domain type for specific "pre-action".
2282 switch (cmd_ctx
->lsm
->domain
.type
) {
2283 case LTTNG_DOMAIN_KERNEL
:
2285 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2289 /* Kernel tracer check */
2290 if (kernel_tracer_fd
== -1) {
2291 /* Basically, load kernel tracer modules */
2292 ret
= init_kernel_tracer();
2298 /* Consumer is in an ERROR state. Report back to client */
2299 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2300 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2304 /* Need a session for kernel command */
2305 if (need_tracing_session
) {
2306 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2307 ret
= create_kernel_session(cmd_ctx
->session
);
2309 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2314 /* Start the kernel consumer daemon */
2315 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2316 if (kconsumer_data
.pid
== 0 &&
2317 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2318 cmd_ctx
->session
->start_consumer
) {
2319 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2320 ret
= start_consumerd(&kconsumer_data
);
2322 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2325 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2327 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2331 * The consumer was just spawned so we need to add the socket to
2332 * the consumer output of the session if exist.
2334 ret
= consumer_create_socket(&kconsumer_data
,
2335 cmd_ctx
->session
->kernel_session
->consumer
);
2342 case LTTNG_DOMAIN_UST
:
2344 /* Consumer is in an ERROR state. Report back to client */
2345 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2346 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2350 if (need_tracing_session
) {
2351 /* Create UST session if none exist. */
2352 if (cmd_ctx
->session
->ust_session
== NULL
) {
2353 ret
= create_ust_session(cmd_ctx
->session
,
2354 &cmd_ctx
->lsm
->domain
);
2355 if (ret
!= LTTNG_OK
) {
2360 /* Start the UST consumer daemons */
2362 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2363 if (consumerd64_bin
[0] != '\0' &&
2364 ustconsumer64_data
.pid
== 0 &&
2365 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2366 cmd_ctx
->session
->start_consumer
) {
2367 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2368 ret
= start_consumerd(&ustconsumer64_data
);
2370 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2371 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2375 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2376 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2378 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2382 * Setup socket for consumer 64 bit. No need for atomic access
2383 * since it was set above and can ONLY be set in this thread.
2385 ret
= consumer_create_socket(&ustconsumer64_data
,
2386 cmd_ctx
->session
->ust_session
->consumer
);
2392 if (consumerd32_bin
[0] != '\0' &&
2393 ustconsumer32_data
.pid
== 0 &&
2394 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2395 cmd_ctx
->session
->start_consumer
) {
2396 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2397 ret
= start_consumerd(&ustconsumer32_data
);
2399 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2400 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2404 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2405 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2407 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2411 * Setup socket for consumer 64 bit. No need for atomic access
2412 * since it was set above and can ONLY be set in this thread.
2414 ret
= consumer_create_socket(&ustconsumer32_data
,
2415 cmd_ctx
->session
->ust_session
->consumer
);
2427 /* Validate consumer daemon state when start/stop trace command */
2428 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2429 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2430 switch (cmd_ctx
->lsm
->domain
.type
) {
2431 case LTTNG_DOMAIN_UST
:
2432 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2433 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2437 case LTTNG_DOMAIN_KERNEL
:
2438 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2439 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2447 * Check that the UID or GID match that of the tracing session.
2448 * The root user can interact with all sessions.
2450 if (need_tracing_session
) {
2451 if (!session_access_ok(cmd_ctx
->session
,
2452 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2453 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2454 ret
= LTTNG_ERR_EPERM
;
2459 /* Process by command type */
2460 switch (cmd_ctx
->lsm
->cmd_type
) {
2461 case LTTNG_ADD_CONTEXT
:
2463 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2464 cmd_ctx
->lsm
->u
.context
.channel_name
,
2465 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2468 case LTTNG_DISABLE_CHANNEL
:
2470 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2471 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2474 case LTTNG_DISABLE_EVENT
:
2476 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2477 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2478 cmd_ctx
->lsm
->u
.disable
.name
);
2481 case LTTNG_DISABLE_ALL_EVENT
:
2483 DBG("Disabling all events");
2485 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2486 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2489 case LTTNG_DISABLE_CONSUMER
:
2491 ret
= cmd_disable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2494 case LTTNG_ENABLE_CHANNEL
:
2496 ret
= cmd_enable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2497 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2500 case LTTNG_ENABLE_CONSUMER
:
2503 * XXX: 0 means that this URI should be applied on the session. Should
2504 * be a DOMAIN enuam.
2506 ret
= cmd_enable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2507 if (ret
!= LTTNG_OK
) {
2511 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2512 /* Add the URI for the UST session if a consumer is present. */
2513 if (cmd_ctx
->session
->ust_session
&&
2514 cmd_ctx
->session
->ust_session
->consumer
) {
2515 ret
= cmd_enable_consumer(LTTNG_DOMAIN_UST
, cmd_ctx
->session
);
2516 } else if (cmd_ctx
->session
->kernel_session
&&
2517 cmd_ctx
->session
->kernel_session
->consumer
) {
2518 ret
= cmd_enable_consumer(LTTNG_DOMAIN_KERNEL
,
2524 case LTTNG_ENABLE_EVENT
:
2526 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2527 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2528 &cmd_ctx
->lsm
->u
.enable
.event
, kernel_poll_pipe
[1]);
2531 case LTTNG_ENABLE_ALL_EVENT
:
2533 DBG("Enabling all events");
2535 ret
= cmd_enable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2536 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2537 cmd_ctx
->lsm
->u
.enable
.event
.type
, kernel_poll_pipe
[1]);
2540 case LTTNG_LIST_TRACEPOINTS
:
2542 struct lttng_event
*events
;
2545 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2546 if (nb_events
< 0) {
2547 /* Return value is a negative lttng_error_code. */
2553 * Setup lttng message with payload size set to the event list size in
2554 * bytes and then copy list into the llm payload.
2556 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2562 /* Copy event list into message payload */
2563 memcpy(cmd_ctx
->llm
->payload
, events
,
2564 sizeof(struct lttng_event
) * nb_events
);
2571 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2573 struct lttng_event_field
*fields
;
2576 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2578 if (nb_fields
< 0) {
2579 /* Return value is a negative lttng_error_code. */
2585 * Setup lttng message with payload size set to the event list size in
2586 * bytes and then copy list into the llm payload.
2588 ret
= setup_lttng_msg(cmd_ctx
,
2589 sizeof(struct lttng_event_field
) * nb_fields
);
2595 /* Copy event list into message payload */
2596 memcpy(cmd_ctx
->llm
->payload
, fields
,
2597 sizeof(struct lttng_event_field
) * nb_fields
);
2604 case LTTNG_SET_CONSUMER_URI
:
2607 struct lttng_uri
*uris
;
2609 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2610 len
= nb_uri
* sizeof(struct lttng_uri
);
2613 ret
= LTTNG_ERR_INVALID
;
2617 uris
= zmalloc(len
);
2619 ret
= LTTNG_ERR_FATAL
;
2623 /* Receive variable len data */
2624 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2625 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2627 DBG("No URIs received from client... continuing");
2629 ret
= LTTNG_ERR_SESSION_FAIL
;
2634 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2636 if (ret
!= LTTNG_OK
) {
2642 * XXX: 0 means that this URI should be applied on the session. Should
2643 * be a DOMAIN enuam.
2645 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2646 /* Add the URI for the UST session if a consumer is present. */
2647 if (cmd_ctx
->session
->ust_session
&&
2648 cmd_ctx
->session
->ust_session
->consumer
) {
2649 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2651 } else if (cmd_ctx
->session
->kernel_session
&&
2652 cmd_ctx
->session
->kernel_session
->consumer
) {
2653 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2654 cmd_ctx
->session
, nb_uri
, uris
);
2662 case LTTNG_START_TRACE
:
2664 ret
= cmd_start_trace(cmd_ctx
->session
);
2667 case LTTNG_STOP_TRACE
:
2669 ret
= cmd_stop_trace(cmd_ctx
->session
);
2672 case LTTNG_CREATE_SESSION
:
2675 struct lttng_uri
*uris
= NULL
;
2677 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2678 len
= nb_uri
* sizeof(struct lttng_uri
);
2681 uris
= zmalloc(len
);
2683 ret
= LTTNG_ERR_FATAL
;
2687 /* Receive variable len data */
2688 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2689 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2691 DBG("No URIs received from client... continuing");
2693 ret
= LTTNG_ERR_SESSION_FAIL
;
2698 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2699 DBG("Creating session with ONE network URI is a bad call");
2700 ret
= LTTNG_ERR_SESSION_FAIL
;
2706 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2713 case LTTNG_DESTROY_SESSION
:
2715 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2717 /* Set session to NULL so we do not unlock it after free. */
2718 cmd_ctx
->session
= NULL
;
2721 case LTTNG_LIST_DOMAINS
:
2724 struct lttng_domain
*domains
;
2726 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2728 /* Return value is a negative lttng_error_code. */
2733 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2738 /* Copy event list into message payload */
2739 memcpy(cmd_ctx
->llm
->payload
, domains
,
2740 nb_dom
* sizeof(struct lttng_domain
));
2747 case LTTNG_LIST_CHANNELS
:
2750 struct lttng_channel
*channels
;
2752 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2753 cmd_ctx
->session
, &channels
);
2755 /* Return value is a negative lttng_error_code. */
2760 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2765 /* Copy event list into message payload */
2766 memcpy(cmd_ctx
->llm
->payload
, channels
,
2767 nb_chan
* sizeof(struct lttng_channel
));
2774 case LTTNG_LIST_EVENTS
:
2777 struct lttng_event
*events
= NULL
;
2779 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2780 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2782 /* Return value is a negative lttng_error_code. */
2787 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2792 /* Copy event list into message payload */
2793 memcpy(cmd_ctx
->llm
->payload
, events
,
2794 nb_event
* sizeof(struct lttng_event
));
2801 case LTTNG_LIST_SESSIONS
:
2803 unsigned int nr_sessions
;
2805 session_lock_list();
2806 nr_sessions
= lttng_sessions_count(
2807 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2808 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2810 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2812 session_unlock_list();
2816 /* Filled the session array */
2817 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2818 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2819 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2821 session_unlock_list();
2826 case LTTNG_CALIBRATE
:
2828 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
2829 &cmd_ctx
->lsm
->u
.calibrate
);
2832 case LTTNG_REGISTER_CONSUMER
:
2834 struct consumer_data
*cdata
;
2836 switch (cmd_ctx
->lsm
->domain
.type
) {
2837 case LTTNG_DOMAIN_KERNEL
:
2838 cdata
= &kconsumer_data
;
2841 ret
= LTTNG_ERR_UND
;
2845 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2846 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
2849 case LTTNG_SET_FILTER
:
2851 struct lttng_filter_bytecode
*bytecode
;
2853 if (cmd_ctx
->lsm
->u
.filter
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
2854 ret
= LTTNG_ERR_FILTER_INVAL
;
2857 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.filter
.bytecode_len
);
2859 ret
= LTTNG_ERR_FILTER_NOMEM
;
2862 /* Receive var. len. data */
2863 DBG("Receiving var len data from client ...");
2864 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
2865 cmd_ctx
->lsm
->u
.filter
.bytecode_len
);
2867 DBG("Nothing recv() from client var len data... continuing");
2869 ret
= LTTNG_ERR_FILTER_INVAL
;
2873 if (bytecode
->len
+ sizeof(*bytecode
)
2874 != cmd_ctx
->lsm
->u
.filter
.bytecode_len
) {
2876 ret
= LTTNG_ERR_FILTER_INVAL
;
2880 ret
= cmd_set_filter(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2881 cmd_ctx
->lsm
->u
.filter
.channel_name
,
2882 cmd_ctx
->lsm
->u
.filter
.event_name
,
2886 case LTTNG_DATA_PENDING
:
2888 ret
= cmd_data_pending(cmd_ctx
->session
);
2892 ret
= LTTNG_ERR_UND
;
2897 if (cmd_ctx
->llm
== NULL
) {
2898 DBG("Missing llm structure. Allocating one.");
2899 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
2903 /* Set return code */
2904 cmd_ctx
->llm
->ret_code
= ret
;
2906 if (cmd_ctx
->session
) {
2907 session_unlock(cmd_ctx
->session
);
2909 if (need_tracing_session
) {
2910 session_unlock_list();
2917 * Thread managing health check socket.
2919 static void *thread_manage_health(void *data
)
2921 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
2922 uint32_t revents
, nb_fd
;
2923 struct lttng_poll_event events
;
2924 struct lttcomm_health_msg msg
;
2925 struct lttcomm_health_data reply
;
2927 DBG("[thread] Manage health check started");
2929 rcu_register_thread();
2931 /* Create unix socket */
2932 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
2934 ERR("Unable to create health check Unix socket");
2940 * Set the CLOEXEC flag. Return code is useless because either way, the
2943 (void) utils_set_fd_cloexec(sock
);
2945 ret
= lttcomm_listen_unix_sock(sock
);
2951 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2952 * more will be added to this poll set.
2954 ret
= create_thread_poll_set(&events
, 2);
2959 /* Add the application registration socket */
2960 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
2966 DBG("Health check ready");
2968 nb_fd
= LTTNG_POLL_GETNB(&events
);
2970 /* Inifinite blocking call, waiting for transmission */
2972 ret
= lttng_poll_wait(&events
, -1);
2975 * Restart interrupted system call.
2977 if (errno
== EINTR
) {
2983 for (i
= 0; i
< nb_fd
; i
++) {
2984 /* Fetch once the poll data */
2985 revents
= LTTNG_POLL_GETEV(&events
, i
);
2986 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2988 /* Thread quit pipe has been closed. Killing thread. */
2989 ret
= check_thread_quit_pipe(pollfd
, revents
);
2995 /* Event on the registration socket */
2996 if (pollfd
== sock
) {
2997 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2998 ERR("Health socket poll error");
3004 new_sock
= lttcomm_accept_unix_sock(sock
);
3010 * Set the CLOEXEC flag. Return code is useless because either way, the
3013 (void) utils_set_fd_cloexec(new_sock
);
3015 DBG("Receiving data from client for health...");
3016 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3018 DBG("Nothing recv() from client... continuing");
3019 ret
= close(new_sock
);
3027 rcu_thread_online();
3029 switch (msg
.component
) {
3030 case LTTNG_HEALTH_CMD
:
3031 reply
.ret_code
= health_check_state(&health_thread_cmd
);
3033 case LTTNG_HEALTH_APP_MANAGE
:
3034 reply
.ret_code
= health_check_state(&health_thread_app_manage
);
3036 case LTTNG_HEALTH_APP_REG
:
3037 reply
.ret_code
= health_check_state(&health_thread_app_reg
);
3039 case LTTNG_HEALTH_KERNEL
:
3040 reply
.ret_code
= health_check_state(&health_thread_kernel
);
3042 case LTTNG_HEALTH_CONSUMER
:
3043 reply
.ret_code
= check_consumer_health();
3045 case LTTNG_HEALTH_ALL
:
3047 health_check_state(&health_thread_app_manage
) &&
3048 health_check_state(&health_thread_app_reg
) &&
3049 health_check_state(&health_thread_cmd
) &&
3050 health_check_state(&health_thread_kernel
) &&
3051 check_consumer_health();
3054 reply
.ret_code
= LTTNG_ERR_UND
;
3059 * Flip ret value since 0 is a success and 1 indicates a bad health for
3060 * the client where in the sessiond it is the opposite. Again, this is
3061 * just to make things easier for us poor developer which enjoy a lot
3064 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3065 reply
.ret_code
= !reply
.ret_code
;
3068 DBG2("Health check return value %d", reply
.ret_code
);
3070 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3072 ERR("Failed to send health data back to client");
3075 /* End of transmission */
3076 ret
= close(new_sock
);
3086 ERR("Health error occurred in %s", __func__
);
3088 DBG("Health check thread dying");
3089 unlink(health_unix_sock_path
);
3096 if (new_sock
>= 0) {
3097 ret
= close(new_sock
);
3103 lttng_poll_clean(&events
);
3105 rcu_unregister_thread();
3110 * This thread manage all clients request using the unix client socket for
3113 static void *thread_manage_clients(void *data
)
3115 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3117 uint32_t revents
, nb_fd
;
3118 struct command_ctx
*cmd_ctx
= NULL
;
3119 struct lttng_poll_event events
;
3121 DBG("[thread] Manage client started");
3123 testpoint(thread_manage_clients
);
3125 rcu_register_thread();
3127 health_code_update(&health_thread_cmd
);
3129 ret
= lttcomm_listen_unix_sock(client_sock
);
3135 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3136 * more will be added to this poll set.
3138 ret
= create_thread_poll_set(&events
, 2);
3140 goto error_create_poll
;
3143 /* Add the application registration socket */
3144 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3150 * Notify parent pid that we are ready to accept command for client side.
3152 if (opt_sig_parent
) {
3153 kill(ppid
, SIGUSR1
);
3156 testpoint(thread_manage_clients_before_loop
);
3158 health_code_update(&health_thread_cmd
);
3161 DBG("Accepting client command ...");
3163 nb_fd
= LTTNG_POLL_GETNB(&events
);
3165 /* Inifinite blocking call, waiting for transmission */
3167 health_poll_update(&health_thread_cmd
);
3168 ret
= lttng_poll_wait(&events
, -1);
3169 health_poll_update(&health_thread_cmd
);
3172 * Restart interrupted system call.
3174 if (errno
== EINTR
) {
3180 for (i
= 0; i
< nb_fd
; i
++) {
3181 /* Fetch once the poll data */
3182 revents
= LTTNG_POLL_GETEV(&events
, i
);
3183 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3185 health_code_update(&health_thread_cmd
);
3187 /* Thread quit pipe has been closed. Killing thread. */
3188 ret
= check_thread_quit_pipe(pollfd
, revents
);
3194 /* Event on the registration socket */
3195 if (pollfd
== client_sock
) {
3196 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3197 ERR("Client socket poll error");
3203 DBG("Wait for client response");
3205 health_code_update(&health_thread_cmd
);
3207 sock
= lttcomm_accept_unix_sock(client_sock
);
3213 * Set the CLOEXEC flag. Return code is useless because either way, the
3216 (void) utils_set_fd_cloexec(sock
);
3218 /* Set socket option for credentials retrieval */
3219 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3224 /* Allocate context command to process the client request */
3225 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3226 if (cmd_ctx
== NULL
) {
3227 PERROR("zmalloc cmd_ctx");
3231 /* Allocate data buffer for reception */
3232 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3233 if (cmd_ctx
->lsm
== NULL
) {
3234 PERROR("zmalloc cmd_ctx->lsm");
3238 cmd_ctx
->llm
= NULL
;
3239 cmd_ctx
->session
= NULL
;
3241 health_code_update(&health_thread_cmd
);
3244 * Data is received from the lttng client. The struct
3245 * lttcomm_session_msg (lsm) contains the command and data request of
3248 DBG("Receiving data from client ...");
3249 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3250 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3252 DBG("Nothing recv() from client... continuing");
3258 clean_command_ctx(&cmd_ctx
);
3262 health_code_update(&health_thread_cmd
);
3264 // TODO: Validate cmd_ctx including sanity check for
3265 // security purpose.
3267 rcu_thread_online();
3269 * This function dispatch the work to the kernel or userspace tracer
3270 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3271 * informations for the client. The command context struct contains
3272 * everything this function may needs.
3274 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3275 rcu_thread_offline();
3285 * TODO: Inform client somehow of the fatal error. At
3286 * this point, ret < 0 means that a zmalloc failed
3287 * (ENOMEM). Error detected but still accept
3288 * command, unless a socket error has been
3291 clean_command_ctx(&cmd_ctx
);
3295 health_code_update(&health_thread_cmd
);
3297 DBG("Sending response (size: %d, retcode: %s)",
3298 cmd_ctx
->lttng_msg_size
,
3299 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3300 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3302 ERR("Failed to send data back to client");
3305 /* End of transmission */
3312 clean_command_ctx(&cmd_ctx
);
3314 health_code_update(&health_thread_cmd
);
3326 lttng_poll_clean(&events
);
3327 clean_command_ctx(&cmd_ctx
);
3331 unlink(client_unix_sock_path
);
3332 if (client_sock
>= 0) {
3333 ret
= close(client_sock
);
3340 health_error(&health_thread_cmd
);
3341 ERR("Health error occurred in %s", __func__
);
3344 health_exit(&health_thread_cmd
);
3346 DBG("Client thread dying");
3348 rcu_unregister_thread();
3354 * usage function on stderr
3356 static void usage(void)
3358 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3359 fprintf(stderr
, " -h, --help Display this usage.\n");
3360 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3361 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3362 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3363 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3364 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3365 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3366 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3367 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3368 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3369 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3370 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3371 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3372 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3373 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3374 fprintf(stderr
, " -V, --version Show version number.\n");
3375 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3376 fprintf(stderr
, " -q, --quiet No output at all.\n");
3377 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3378 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3379 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3383 * daemon argument parsing
3385 static int parse_args(int argc
, char **argv
)
3389 static struct option long_options
[] = {
3390 { "client-sock", 1, 0, 'c' },
3391 { "apps-sock", 1, 0, 'a' },
3392 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3393 { "kconsumerd-err-sock", 1, 0, 'E' },
3394 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3395 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3396 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3397 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3398 { "consumerd32-path", 1, 0, 'u' },
3399 { "consumerd32-libdir", 1, 0, 'U' },
3400 { "consumerd64-path", 1, 0, 't' },
3401 { "consumerd64-libdir", 1, 0, 'T' },
3402 { "daemonize", 0, 0, 'd' },
3403 { "sig-parent", 0, 0, 'S' },
3404 { "help", 0, 0, 'h' },
3405 { "group", 1, 0, 'g' },
3406 { "version", 0, 0, 'V' },
3407 { "quiet", 0, 0, 'q' },
3408 { "verbose", 0, 0, 'v' },
3409 { "verbose-consumer", 0, 0, 'Z' },
3410 { "no-kernel", 0, 0, 'N' },
3415 int option_index
= 0;
3416 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3417 long_options
, &option_index
);
3424 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3426 fprintf(stderr
, " with arg %s\n", optarg
);
3430 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3433 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3439 opt_tracing_group
= optarg
;
3445 fprintf(stdout
, "%s\n", VERSION
);
3451 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3454 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3457 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3460 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3463 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3466 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3472 lttng_opt_quiet
= 1;
3475 /* Verbose level can increase using multiple -v */
3476 lttng_opt_verbose
+= 1;
3479 opt_verbose_consumer
+= 1;
3482 consumerd32_bin
= optarg
;
3485 consumerd32_libdir
= optarg
;
3488 consumerd64_bin
= optarg
;
3491 consumerd64_libdir
= optarg
;
3494 /* Unknown option or other error.
3495 * Error is printed by getopt, just return */
3504 * Creates the two needed socket by the daemon.
3505 * apps_sock - The communication socket for all UST apps.
3506 * client_sock - The communication of the cli tool (lttng).
3508 static int init_daemon_socket(void)
3513 old_umask
= umask(0);
3515 /* Create client tool unix socket */
3516 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3517 if (client_sock
< 0) {
3518 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3523 /* Set the cloexec flag */
3524 ret
= utils_set_fd_cloexec(client_sock
);
3526 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3527 "Continuing but note that the consumer daemon will have a "
3528 "reference to this socket on exec()", client_sock
);
3531 /* File permission MUST be 660 */
3532 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3534 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3539 /* Create the application unix socket */
3540 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3541 if (apps_sock
< 0) {
3542 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3547 /* Set the cloexec flag */
3548 ret
= utils_set_fd_cloexec(apps_sock
);
3550 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3551 "Continuing but note that the consumer daemon will have a "
3552 "reference to this socket on exec()", apps_sock
);
3555 /* File permission MUST be 666 */
3556 ret
= chmod(apps_unix_sock_path
,
3557 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3559 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3564 DBG3("Session daemon client socket %d and application socket %d created",
3565 client_sock
, apps_sock
);
3573 * Check if the global socket is available, and if a daemon is answering at the
3574 * other side. If yes, error is returned.
3576 static int check_existing_daemon(void)
3578 /* Is there anybody out there ? */
3579 if (lttng_session_daemon_alive()) {
3587 * Set the tracing group gid onto the client socket.
3589 * Race window between mkdir and chown is OK because we are going from more
3590 * permissive (root.root) to less permissive (root.tracing).
3592 static int set_permissions(char *rundir
)
3597 ret
= allowed_group();
3599 WARN("No tracing group detected");
3606 /* Set lttng run dir */
3607 ret
= chown(rundir
, 0, gid
);
3609 ERR("Unable to set group on %s", rundir
);
3613 /* Ensure tracing group can search the run dir */
3614 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3616 ERR("Unable to set permissions on %s", rundir
);
3620 /* lttng client socket path */
3621 ret
= chown(client_unix_sock_path
, 0, gid
);
3623 ERR("Unable to set group on %s", client_unix_sock_path
);
3627 /* kconsumer error socket path */
3628 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3630 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3634 /* 64-bit ustconsumer error socket path */
3635 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3637 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3641 /* 32-bit ustconsumer compat32 error socket path */
3642 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3644 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3648 DBG("All permissions are set");
3655 * Create the lttng run directory needed for all global sockets and pipe.
3657 static int create_lttng_rundir(const char *rundir
)
3661 DBG3("Creating LTTng run directory: %s", rundir
);
3663 ret
= mkdir(rundir
, S_IRWXU
);
3665 if (errno
!= EEXIST
) {
3666 ERR("Unable to create %s", rundir
);
3678 * Setup sockets and directory needed by the kconsumerd communication with the
3681 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3685 char path
[PATH_MAX
];
3687 switch (consumer_data
->type
) {
3688 case LTTNG_CONSUMER_KERNEL
:
3689 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3691 case LTTNG_CONSUMER64_UST
:
3692 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3694 case LTTNG_CONSUMER32_UST
:
3695 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3698 ERR("Consumer type unknown");
3703 DBG2("Creating consumer directory: %s", path
);
3705 ret
= mkdir(path
, S_IRWXU
);
3707 if (errno
!= EEXIST
) {
3709 ERR("Failed to create %s", path
);
3715 /* Create the kconsumerd error unix socket */
3716 consumer_data
->err_sock
=
3717 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3718 if (consumer_data
->err_sock
< 0) {
3719 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3724 /* File permission MUST be 660 */
3725 ret
= chmod(consumer_data
->err_unix_sock_path
,
3726 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3728 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3738 * Signal handler for the daemon
3740 * Simply stop all worker threads, leaving main() return gracefully after
3741 * joining all threads and calling cleanup().
3743 static void sighandler(int sig
)
3747 DBG("SIGPIPE caught");
3750 DBG("SIGINT caught");
3754 DBG("SIGTERM caught");
3763 * Setup signal handler for :
3764 * SIGINT, SIGTERM, SIGPIPE
3766 static int set_signal_handler(void)
3769 struct sigaction sa
;
3772 if ((ret
= sigemptyset(&sigset
)) < 0) {
3773 PERROR("sigemptyset");
3777 sa
.sa_handler
= sighandler
;
3778 sa
.sa_mask
= sigset
;
3780 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3781 PERROR("sigaction");
3785 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3786 PERROR("sigaction");
3790 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3791 PERROR("sigaction");
3795 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3801 * Set open files limit to unlimited. This daemon can open a large number of
3802 * file descriptors in order to consumer multiple kernel traces.
3804 static void set_ulimit(void)
3809 /* The kernel does not allowed an infinite limit for open files */
3810 lim
.rlim_cur
= 65535;
3811 lim
.rlim_max
= 65535;
3813 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
3815 PERROR("failed to set open files limit");
3822 int main(int argc
, char **argv
)
3826 const char *home_path
, *env_app_timeout
;
3828 init_kernel_workarounds();
3830 rcu_register_thread();
3832 setup_consumerd_path();
3834 /* Parse arguments */
3836 if ((ret
= parse_args(argc
, argv
) < 0)) {
3846 * child: setsid, close FD 0, 1, 2, chdir /
3847 * parent: exit (if fork is successful)
3855 * We are in the child. Make sure all other file
3856 * descriptors are closed, in case we are called with
3857 * more opened file descriptors than the standard ones.
3859 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3864 /* Create thread quit pipe */
3865 if ((ret
= init_thread_quit_pipe()) < 0) {
3869 /* Check if daemon is UID = 0 */
3870 is_root
= !getuid();
3873 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
3875 /* Create global run dir with root access */
3876 ret
= create_lttng_rundir(rundir
);
3881 if (strlen(apps_unix_sock_path
) == 0) {
3882 snprintf(apps_unix_sock_path
, PATH_MAX
,
3883 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
3886 if (strlen(client_unix_sock_path
) == 0) {
3887 snprintf(client_unix_sock_path
, PATH_MAX
,
3888 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
3891 /* Set global SHM for ust */
3892 if (strlen(wait_shm_path
) == 0) {
3893 snprintf(wait_shm_path
, PATH_MAX
,
3894 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
3897 if (strlen(health_unix_sock_path
) == 0) {
3898 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3899 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
3902 /* Setup kernel consumerd path */
3903 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
3904 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
3905 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
3906 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
3908 DBG2("Kernel consumer err path: %s",
3909 kconsumer_data
.err_unix_sock_path
);
3910 DBG2("Kernel consumer cmd path: %s",
3911 kconsumer_data
.cmd_unix_sock_path
);
3913 home_path
= get_home_dir();
3914 if (home_path
== NULL
) {
3915 /* TODO: Add --socket PATH option */
3916 ERR("Can't get HOME directory for sockets creation.");
3922 * Create rundir from home path. This will create something like
3925 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
3931 ret
= create_lttng_rundir(rundir
);
3936 if (strlen(apps_unix_sock_path
) == 0) {
3937 snprintf(apps_unix_sock_path
, PATH_MAX
,
3938 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
3941 /* Set the cli tool unix socket path */
3942 if (strlen(client_unix_sock_path
) == 0) {
3943 snprintf(client_unix_sock_path
, PATH_MAX
,
3944 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
3947 /* Set global SHM for ust */
3948 if (strlen(wait_shm_path
) == 0) {
3949 snprintf(wait_shm_path
, PATH_MAX
,
3950 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, geteuid());
3953 /* Set health check Unix path */
3954 if (strlen(health_unix_sock_path
) == 0) {
3955 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3956 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
3960 /* Set consumer initial state */
3961 kernel_consumerd_state
= CONSUMER_STOPPED
;
3962 ust_consumerd_state
= CONSUMER_STOPPED
;
3964 DBG("Client socket path %s", client_unix_sock_path
);
3965 DBG("Application socket path %s", apps_unix_sock_path
);
3966 DBG("LTTng run directory path: %s", rundir
);
3968 /* 32 bits consumerd path setup */
3969 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
3970 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
3971 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
3972 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
3974 DBG2("UST consumer 32 bits err path: %s",
3975 ustconsumer32_data
.err_unix_sock_path
);
3976 DBG2("UST consumer 32 bits cmd path: %s",
3977 ustconsumer32_data
.cmd_unix_sock_path
);
3979 /* 64 bits consumerd path setup */
3980 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
3981 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
3982 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
3983 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
3985 DBG2("UST consumer 64 bits err path: %s",
3986 ustconsumer64_data
.err_unix_sock_path
);
3987 DBG2("UST consumer 64 bits cmd path: %s",
3988 ustconsumer64_data
.cmd_unix_sock_path
);
3991 * See if daemon already exist.
3993 if ((ret
= check_existing_daemon()) < 0) {
3994 ERR("Already running daemon.\n");
3996 * We do not goto exit because we must not cleanup()
3997 * because a daemon is already running.
4003 * Init UST app hash table. Alloc hash table before this point since
4004 * cleanup() can get called after that point.
4008 /* After this point, we can safely call cleanup() with "goto exit" */
4011 * These actions must be executed as root. We do that *after* setting up
4012 * the sockets path because we MUST make the check for another daemon using
4013 * those paths *before* trying to set the kernel consumer sockets and init
4017 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4022 /* Setup kernel tracer */
4023 if (!opt_no_kernel
) {
4024 init_kernel_tracer();
4027 /* Set ulimit for open files */
4030 /* init lttng_fd tracking must be done after set_ulimit. */
4033 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4038 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4043 if ((ret
= set_signal_handler()) < 0) {
4047 /* Setup the needed unix socket */
4048 if ((ret
= init_daemon_socket()) < 0) {
4052 /* Set credentials to socket */
4053 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4057 /* Get parent pid if -S, --sig-parent is specified. */
4058 if (opt_sig_parent
) {
4062 /* Setup the kernel pipe for waking up the kernel thread */
4063 if (is_root
&& !opt_no_kernel
) {
4064 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4069 /* Setup the thread apps communication pipe. */
4070 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4074 /* Init UST command queue. */
4075 cds_wfq_init(&ust_cmd_queue
.queue
);
4078 * Get session list pointer. This pointer MUST NOT be free(). This list is
4079 * statically declared in session.c
4081 session_list_ptr
= session_get_list();
4083 /* Set up max poll set size */
4084 lttng_poll_set_max_size();
4088 /* Init all health thread counters. */
4089 health_init(&health_thread_cmd
);
4090 health_init(&health_thread_kernel
);
4091 health_init(&health_thread_app_manage
);
4092 health_init(&health_thread_app_reg
);
4095 * Init health counters of the consumer thread. We do a quick hack here to
4096 * the state of the consumer health is fine even if the thread is not
4097 * started. Once the thread starts, the health state is updated with a poll
4098 * value to set a health code path. This is simply to ease our life and has
4099 * no cost what so ever.
4101 health_init(&kconsumer_data
.health
);
4102 health_poll_update(&kconsumer_data
.health
);
4103 health_init(&ustconsumer32_data
.health
);
4104 health_poll_update(&ustconsumer32_data
.health
);
4105 health_init(&ustconsumer64_data
.health
);
4106 health_poll_update(&ustconsumer64_data
.health
);
4108 /* Check for the application socket timeout env variable. */
4109 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4110 if (env_app_timeout
) {
4111 app_socket_timeout
= atoi(env_app_timeout
);
4113 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4116 /* Create thread to manage the client socket */
4117 ret
= pthread_create(&health_thread
, NULL
,
4118 thread_manage_health
, (void *) NULL
);
4120 PERROR("pthread_create health");
4124 /* Create thread to manage the client socket */
4125 ret
= pthread_create(&client_thread
, NULL
,
4126 thread_manage_clients
, (void *) NULL
);
4128 PERROR("pthread_create clients");
4132 /* Create thread to dispatch registration */
4133 ret
= pthread_create(&dispatch_thread
, NULL
,
4134 thread_dispatch_ust_registration
, (void *) NULL
);
4136 PERROR("pthread_create dispatch");
4140 /* Create thread to manage application registration. */
4141 ret
= pthread_create(®_apps_thread
, NULL
,
4142 thread_registration_apps
, (void *) NULL
);
4144 PERROR("pthread_create registration");
4148 /* Create thread to manage application socket */
4149 ret
= pthread_create(&apps_thread
, NULL
,
4150 thread_manage_apps
, (void *) NULL
);
4152 PERROR("pthread_create apps");
4156 /* Don't start this thread if kernel tracing is not requested nor root */
4157 if (is_root
&& !opt_no_kernel
) {
4158 /* Create kernel thread to manage kernel event */
4159 ret
= pthread_create(&kernel_thread
, NULL
,
4160 thread_manage_kernel
, (void *) NULL
);
4162 PERROR("pthread_create kernel");
4166 ret
= pthread_join(kernel_thread
, &status
);
4168 PERROR("pthread_join");
4169 goto error
; /* join error, exit without cleanup */
4174 ret
= pthread_join(apps_thread
, &status
);
4176 PERROR("pthread_join");
4177 goto error
; /* join error, exit without cleanup */
4181 ret
= pthread_join(reg_apps_thread
, &status
);
4183 PERROR("pthread_join");
4184 goto error
; /* join error, exit without cleanup */
4188 ret
= pthread_join(dispatch_thread
, &status
);
4190 PERROR("pthread_join");
4191 goto error
; /* join error, exit without cleanup */
4195 ret
= pthread_join(client_thread
, &status
);
4197 PERROR("pthread_join");
4198 goto error
; /* join error, exit without cleanup */
4201 ret
= join_consumer_thread(&kconsumer_data
);
4203 PERROR("join_consumer");
4204 goto error
; /* join error, exit without cleanup */
4207 ret
= join_consumer_thread(&ustconsumer32_data
);
4209 PERROR("join_consumer ust32");
4210 goto error
; /* join error, exit without cleanup */
4213 ret
= join_consumer_thread(&ustconsumer64_data
);
4215 PERROR("join_consumer ust64");
4216 goto error
; /* join error, exit without cleanup */
4220 ret
= pthread_join(health_thread
, &status
);
4222 PERROR("pthread_join health thread");
4223 goto error
; /* join error, exit without cleanup */
4229 * cleanup() is called when no other thread is running.
4231 rcu_thread_online();
4233 rcu_thread_offline();
4234 rcu_unregister_thread();