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 void setup_consumerd_path(void)
234 const char *bin
, *libdir
;
237 * Allow INSTALL_BIN_PATH to be used as a target path for the
238 * native architecture size consumer if CONFIG_CONSUMER*_PATH
239 * has not been defined.
241 #if (CAA_BITS_PER_LONG == 32)
242 if (!consumerd32_bin
[0]) {
243 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
245 if (!consumerd32_libdir
[0]) {
246 consumerd32_libdir
= INSTALL_LIB_PATH
;
248 #elif (CAA_BITS_PER_LONG == 64)
249 if (!consumerd64_bin
[0]) {
250 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
252 if (!consumerd64_libdir
[0]) {
253 consumerd64_libdir
= INSTALL_LIB_PATH
;
256 #error "Unknown bitness"
260 * runtime env. var. overrides the build default.
262 bin
= getenv("LTTNG_CONSUMERD32_BIN");
264 consumerd32_bin
= bin
;
266 bin
= getenv("LTTNG_CONSUMERD64_BIN");
268 consumerd64_bin
= bin
;
270 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
272 consumerd32_libdir
= libdir
;
274 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
276 consumerd64_libdir
= libdir
;
281 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
283 static int create_thread_poll_set(struct lttng_poll_event
*events
,
288 if (events
== NULL
|| size
== 0) {
293 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
299 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
311 * Check if the thread quit pipe was triggered.
313 * Return 1 if it was triggered else 0;
315 static int check_thread_quit_pipe(int fd
, uint32_t events
)
317 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
325 * Return group ID of the tracing group or -1 if not found.
327 static gid_t
allowed_group(void)
331 if (opt_tracing_group
) {
332 grp
= getgrnam(opt_tracing_group
);
334 grp
= getgrnam(default_tracing_group
);
344 * Init thread quit pipe.
346 * Return -1 on error or 0 if all pipes are created.
348 static int init_thread_quit_pipe(void)
352 ret
= pipe(thread_quit_pipe
);
354 PERROR("thread quit pipe");
358 for (i
= 0; i
< 2; i
++) {
359 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
371 * Stop all threads by closing the thread quit pipe.
373 static void stop_threads(void)
377 /* Stopping all threads */
378 DBG("Terminating all threads");
379 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
381 ERR("write error on thread quit pipe");
384 /* Dispatch thread */
385 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
386 futex_nto1_wake(&ust_cmd_queue
.futex
);
392 static void cleanup(void)
396 struct ltt_session
*sess
, *stmp
;
400 /* First thing first, stop all threads */
401 utils_close_pipe(thread_quit_pipe
);
403 DBG("Removing %s directory", rundir
);
404 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
406 ERR("asprintf failed. Something is really wrong!");
409 /* Remove lttng run directory */
412 ERR("Unable to clean %s", rundir
);
417 DBG("Cleaning up all sessions");
419 /* Destroy session list mutex */
420 if (session_list_ptr
!= NULL
) {
421 pthread_mutex_destroy(&session_list_ptr
->lock
);
423 /* Cleanup ALL session */
424 cds_list_for_each_entry_safe(sess
, stmp
,
425 &session_list_ptr
->head
, list
) {
426 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
430 DBG("Closing all UST sockets");
431 ust_app_clean_list();
433 if (is_root
&& !opt_no_kernel
) {
434 DBG2("Closing kernel fd");
435 if (kernel_tracer_fd
>= 0) {
436 ret
= close(kernel_tracer_fd
);
441 DBG("Unloading kernel modules");
442 modprobe_remove_lttng_all();
445 utils_close_pipe(kernel_poll_pipe
);
446 utils_close_pipe(apps_cmd_pipe
);
449 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
450 "Matthew, BEET driven development works!%c[%dm",
451 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
456 * Send data on a unix socket using the liblttsessiondcomm API.
458 * Return lttcomm error code.
460 static int send_unix_sock(int sock
, void *buf
, size_t len
)
462 /* Check valid length */
467 return lttcomm_send_unix_sock(sock
, buf
, len
);
471 * Free memory of a command context structure.
473 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
475 DBG("Clean command context structure");
477 if ((*cmd_ctx
)->llm
) {
478 free((*cmd_ctx
)->llm
);
480 if ((*cmd_ctx
)->lsm
) {
481 free((*cmd_ctx
)->lsm
);
489 * Notify UST applications using the shm mmap futex.
491 static int notify_ust_apps(int active
)
495 DBG("Notifying applications of session daemon state: %d", active
);
497 /* See shm.c for this call implying mmap, shm and futex calls */
498 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
499 if (wait_shm_mmap
== NULL
) {
503 /* Wake waiting process */
504 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
506 /* Apps notified successfully */
514 * Setup the outgoing data buffer for the response (llm) by allocating the
515 * right amount of memory and copying the original information from the lsm
518 * Return total size of the buffer pointed by buf.
520 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
526 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
527 if (cmd_ctx
->llm
== NULL
) {
533 /* Copy common data */
534 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
535 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
537 cmd_ctx
->llm
->data_size
= size
;
538 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
547 * Update the kernel poll set of all channel fd available over all tracing
548 * session. Add the wakeup pipe at the end of the set.
550 static int update_kernel_poll(struct lttng_poll_event
*events
)
553 struct ltt_session
*session
;
554 struct ltt_kernel_channel
*channel
;
556 DBG("Updating kernel poll set");
559 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
560 session_lock(session
);
561 if (session
->kernel_session
== NULL
) {
562 session_unlock(session
);
566 cds_list_for_each_entry(channel
,
567 &session
->kernel_session
->channel_list
.head
, list
) {
568 /* Add channel fd to the kernel poll set */
569 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
571 session_unlock(session
);
574 DBG("Channel fd %d added to kernel set", channel
->fd
);
576 session_unlock(session
);
578 session_unlock_list();
583 session_unlock_list();
588 * Find the channel fd from 'fd' over all tracing session. When found, check
589 * for new channel stream and send those stream fds to the kernel consumer.
591 * Useful for CPU hotplug feature.
593 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
596 struct ltt_session
*session
;
597 struct ltt_kernel_session
*ksess
;
598 struct ltt_kernel_channel
*channel
;
600 DBG("Updating kernel streams for channel fd %d", fd
);
603 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
604 session_lock(session
);
605 if (session
->kernel_session
== NULL
) {
606 session_unlock(session
);
609 ksess
= session
->kernel_session
;
611 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
612 if (channel
->fd
== fd
) {
613 DBG("Channel found, updating kernel streams");
614 ret
= kernel_open_channel_stream(channel
);
620 * Have we already sent fds to the consumer? If yes, it means
621 * that tracing is started so it is safe to send our updated
624 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
625 struct lttng_ht_iter iter
;
626 struct consumer_socket
*socket
;
629 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
630 &iter
.iter
, socket
, node
.node
) {
631 /* Code flow error */
632 assert(socket
->fd
>= 0);
634 pthread_mutex_lock(socket
->lock
);
635 ret
= kernel_consumer_send_channel_stream(socket
->fd
,
637 pthread_mutex_unlock(socket
->lock
);
646 session_unlock(session
);
648 session_unlock_list();
652 session_unlock(session
);
653 session_unlock_list();
658 * For each tracing session, update newly registered apps.
660 static void update_ust_app(int app_sock
)
662 struct ltt_session
*sess
, *stmp
;
666 /* For all tracing session(s) */
667 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
669 if (sess
->ust_session
) {
670 ust_app_global_update(sess
->ust_session
, app_sock
);
672 session_unlock(sess
);
675 session_unlock_list();
679 * This thread manage event coming from the kernel.
681 * Features supported in this thread:
684 static void *thread_manage_kernel(void *data
)
686 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
687 uint32_t revents
, nb_fd
;
689 struct lttng_poll_event events
;
691 DBG("Thread manage kernel started");
693 testpoint(thread_manage_kernel
);
695 health_code_update(&health_thread_kernel
);
697 testpoint(thread_manage_kernel_before_loop
);
699 ret
= create_thread_poll_set(&events
, 2);
701 goto error_poll_create
;
704 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
710 health_code_update(&health_thread_kernel
);
712 if (update_poll_flag
== 1) {
714 * Reset number of fd in the poll set. Always 2 since there is the thread
715 * quit pipe and the kernel pipe.
719 ret
= update_kernel_poll(&events
);
723 update_poll_flag
= 0;
726 nb_fd
= LTTNG_POLL_GETNB(&events
);
728 DBG("Thread kernel polling on %d fds", nb_fd
);
730 /* Zeroed the poll events */
731 lttng_poll_reset(&events
);
733 /* Poll infinite value of time */
735 health_poll_update(&health_thread_kernel
);
736 ret
= lttng_poll_wait(&events
, -1);
737 health_poll_update(&health_thread_kernel
);
740 * Restart interrupted system call.
742 if (errno
== EINTR
) {
746 } else if (ret
== 0) {
747 /* Should not happen since timeout is infinite */
748 ERR("Return value of poll is 0 with an infinite timeout.\n"
749 "This should not have happened! Continuing...");
753 for (i
= 0; i
< nb_fd
; i
++) {
754 /* Fetch once the poll data */
755 revents
= LTTNG_POLL_GETEV(&events
, i
);
756 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
758 health_code_update(&health_thread_kernel
);
760 /* Thread quit pipe has been closed. Killing thread. */
761 ret
= check_thread_quit_pipe(pollfd
, revents
);
767 /* Check for data on kernel pipe */
768 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
769 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
770 update_poll_flag
= 1;
774 * New CPU detected by the kernel. Adding kernel stream to
775 * kernel session and updating the kernel consumer
777 if (revents
& LPOLLIN
) {
778 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
784 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
785 * and unregister kernel stream at this point.
794 lttng_poll_clean(&events
);
797 health_error(&health_thread_kernel
);
798 ERR("Health error occurred in %s", __func__
);
800 health_exit(&health_thread_kernel
);
801 DBG("Kernel thread dying");
806 * Signal pthread condition of the consumer data that the thread.
808 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
810 pthread_mutex_lock(&data
->cond_mutex
);
813 * The state is set before signaling. It can be any value, it's the waiter
814 * job to correctly interpret this condition variable associated to the
815 * consumer pthread_cond.
817 * A value of 0 means that the corresponding thread of the consumer data
818 * was not started. 1 indicates that the thread has started and is ready
819 * for action. A negative value means that there was an error during the
822 data
->consumer_thread_is_ready
= state
;
823 (void) pthread_cond_signal(&data
->cond
);
825 pthread_mutex_unlock(&data
->cond_mutex
);
829 * This thread manage the consumer error sent back to the session daemon.
831 static void *thread_manage_consumer(void *data
)
833 int sock
= -1, i
, ret
, pollfd
, err
= -1;
834 uint32_t revents
, nb_fd
;
835 enum lttcomm_return_code code
;
836 struct lttng_poll_event events
;
837 struct consumer_data
*consumer_data
= data
;
839 DBG("[thread] Manage consumer started");
842 * Since the consumer thread can be spawned at any moment in time, we init
843 * the health to a poll status (1, which is a valid health over time).
844 * When the thread starts, we update here the health to a "code" path being
845 * an even value so this thread, when reaching a poll wait, does not
846 * trigger an error with an even value.
848 * Here is the use case we avoid.
850 * +1: the first poll update during initialization (main())
851 * +2 * x: multiple code update once in this thread.
852 * +1: poll wait in this thread (being a good health state).
853 * == even number which after the wait period shows as a bad health.
855 * In a nutshell, the following poll update to the health state brings back
856 * the state to an even value meaning a code path.
858 health_poll_update(&consumer_data
->health
);
860 health_code_update(&consumer_data
->health
);
862 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
868 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
869 * Nothing more will be added to this poll set.
871 ret
= create_thread_poll_set(&events
, 2);
876 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
881 nb_fd
= LTTNG_POLL_GETNB(&events
);
883 health_code_update(&consumer_data
->health
);
885 /* Inifinite blocking call, waiting for transmission */
887 health_poll_update(&consumer_data
->health
);
889 testpoint(thread_manage_consumer
);
891 ret
= lttng_poll_wait(&events
, -1);
892 health_poll_update(&consumer_data
->health
);
895 * Restart interrupted system call.
897 if (errno
== EINTR
) {
903 for (i
= 0; i
< nb_fd
; i
++) {
904 /* Fetch once the poll data */
905 revents
= LTTNG_POLL_GETEV(&events
, i
);
906 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
908 health_code_update(&consumer_data
->health
);
910 /* Thread quit pipe has been closed. Killing thread. */
911 ret
= check_thread_quit_pipe(pollfd
, revents
);
917 /* Event on the registration socket */
918 if (pollfd
== consumer_data
->err_sock
) {
919 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
920 ERR("consumer err socket poll error");
926 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
931 health_code_update(&consumer_data
->health
);
933 DBG2("Receiving code from consumer err_sock");
935 /* Getting status code from kconsumerd */
936 ret
= lttcomm_recv_unix_sock(sock
, &code
,
937 sizeof(enum lttcomm_return_code
));
942 health_code_update(&consumer_data
->health
);
944 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
945 consumer_data
->cmd_sock
=
946 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
947 if (consumer_data
->cmd_sock
< 0) {
948 /* On error, signal condition and quit. */
949 signal_consumer_condition(consumer_data
, -1);
950 PERROR("consumer connect");
953 signal_consumer_condition(consumer_data
, 1);
954 DBG("Consumer command socket ready");
956 ERR("consumer error when waiting for SOCK_READY : %s",
957 lttcomm_get_readable_code(-code
));
961 /* Remove the kconsumerd error sock since we've established a connexion */
962 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
967 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
972 health_code_update(&consumer_data
->health
);
974 /* Update number of fd */
975 nb_fd
= LTTNG_POLL_GETNB(&events
);
977 /* Inifinite blocking call, waiting for transmission */
979 health_poll_update(&consumer_data
->health
);
980 ret
= lttng_poll_wait(&events
, -1);
981 health_poll_update(&consumer_data
->health
);
984 * Restart interrupted system call.
986 if (errno
== EINTR
) {
992 for (i
= 0; i
< nb_fd
; i
++) {
993 /* Fetch once the poll data */
994 revents
= LTTNG_POLL_GETEV(&events
, i
);
995 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
997 health_code_update(&consumer_data
->health
);
999 /* Thread quit pipe has been closed. Killing thread. */
1000 ret
= check_thread_quit_pipe(pollfd
, revents
);
1006 /* Event on the kconsumerd socket */
1007 if (pollfd
== sock
) {
1008 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1009 ERR("consumer err socket second poll error");
1015 health_code_update(&consumer_data
->health
);
1017 /* Wait for any kconsumerd error */
1018 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1019 sizeof(enum lttcomm_return_code
));
1021 ERR("consumer closed the command socket");
1025 ERR("consumer return code : %s", lttcomm_get_readable_code(-code
));
1029 /* Immediately set the consumerd state to stopped */
1030 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1031 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1032 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1033 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1034 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1036 /* Code flow error... */
1040 if (consumer_data
->err_sock
>= 0) {
1041 ret
= close(consumer_data
->err_sock
);
1046 if (consumer_data
->cmd_sock
>= 0) {
1047 ret
= close(consumer_data
->cmd_sock
);
1059 unlink(consumer_data
->err_unix_sock_path
);
1060 unlink(consumer_data
->cmd_unix_sock_path
);
1061 consumer_data
->pid
= 0;
1063 lttng_poll_clean(&events
);
1067 health_error(&consumer_data
->health
);
1068 ERR("Health error occurred in %s", __func__
);
1070 health_exit(&consumer_data
->health
);
1071 DBG("consumer thread cleanup completed");
1077 * This thread manage application communication.
1079 static void *thread_manage_apps(void *data
)
1081 int i
, ret
, pollfd
, err
= -1;
1082 uint32_t revents
, nb_fd
;
1083 struct ust_command ust_cmd
;
1084 struct lttng_poll_event events
;
1086 DBG("[thread] Manage application started");
1088 testpoint(thread_manage_apps
);
1090 rcu_register_thread();
1091 rcu_thread_online();
1093 health_code_update(&health_thread_app_manage
);
1095 ret
= create_thread_poll_set(&events
, 2);
1097 goto error_poll_create
;
1100 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1105 testpoint(thread_manage_apps_before_loop
);
1107 health_code_update(&health_thread_app_manage
);
1110 /* Zeroed the events structure */
1111 lttng_poll_reset(&events
);
1113 nb_fd
= LTTNG_POLL_GETNB(&events
);
1115 DBG("Apps thread polling on %d fds", nb_fd
);
1117 /* Inifinite blocking call, waiting for transmission */
1119 health_poll_update(&health_thread_app_manage
);
1120 ret
= lttng_poll_wait(&events
, -1);
1121 health_poll_update(&health_thread_app_manage
);
1124 * Restart interrupted system call.
1126 if (errno
== EINTR
) {
1132 for (i
= 0; i
< nb_fd
; i
++) {
1133 /* Fetch once the poll data */
1134 revents
= LTTNG_POLL_GETEV(&events
, i
);
1135 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1137 health_code_update(&health_thread_app_manage
);
1139 /* Thread quit pipe has been closed. Killing thread. */
1140 ret
= check_thread_quit_pipe(pollfd
, revents
);
1146 /* Inspect the apps cmd pipe */
1147 if (pollfd
== apps_cmd_pipe
[0]) {
1148 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1149 ERR("Apps command pipe error");
1151 } else if (revents
& LPOLLIN
) {
1153 ret
= read(apps_cmd_pipe
[0], &ust_cmd
, sizeof(ust_cmd
));
1154 if (ret
< 0 || ret
< sizeof(ust_cmd
)) {
1155 PERROR("read apps cmd pipe");
1159 health_code_update(&health_thread_app_manage
);
1161 /* Register applicaton to the session daemon */
1162 ret
= ust_app_register(&ust_cmd
.reg_msg
,
1164 if (ret
== -ENOMEM
) {
1166 } else if (ret
< 0) {
1170 health_code_update(&health_thread_app_manage
);
1173 * Validate UST version compatibility.
1175 ret
= ust_app_validate_version(ust_cmd
.sock
);
1178 * Add channel(s) and event(s) to newly registered apps
1179 * from lttng global UST domain.
1181 update_ust_app(ust_cmd
.sock
);
1184 health_code_update(&health_thread_app_manage
);
1186 ret
= ust_app_register_done(ust_cmd
.sock
);
1189 * If the registration is not possible, we simply
1190 * unregister the apps and continue
1192 ust_app_unregister(ust_cmd
.sock
);
1195 * We only monitor the error events of the socket. This
1196 * thread does not handle any incoming data from UST
1199 ret
= lttng_poll_add(&events
, ust_cmd
.sock
,
1200 LPOLLERR
& LPOLLHUP
& LPOLLRDHUP
);
1205 DBG("Apps with sock %d added to poll set",
1209 health_code_update(&health_thread_app_manage
);
1215 * At this point, we know that a registered application made
1216 * the event at poll_wait.
1218 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1219 /* Removing from the poll set */
1220 ret
= lttng_poll_del(&events
, pollfd
);
1225 /* Socket closed on remote end. */
1226 ust_app_unregister(pollfd
);
1231 health_code_update(&health_thread_app_manage
);
1237 lttng_poll_clean(&events
);
1240 health_error(&health_thread_app_manage
);
1241 ERR("Health error occurred in %s", __func__
);
1243 health_exit(&health_thread_app_manage
);
1244 DBG("Application communication apps thread cleanup complete");
1245 rcu_thread_offline();
1246 rcu_unregister_thread();
1251 * Dispatch request from the registration threads to the application
1252 * communication thread.
1254 static void *thread_dispatch_ust_registration(void *data
)
1257 struct cds_wfq_node
*node
;
1258 struct ust_command
*ust_cmd
= NULL
;
1260 DBG("[thread] Dispatch UST command started");
1262 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1263 /* Atomically prepare the queue futex */
1264 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1267 /* Dequeue command for registration */
1268 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1270 DBG("Woken up but nothing in the UST command queue");
1271 /* Continue thread execution */
1275 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1277 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1278 " gid:%d sock:%d name:%s (version %d.%d)",
1279 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1280 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1281 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1282 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1284 * Inform apps thread of the new application registration. This
1285 * call is blocking so we can be assured that the data will be read
1286 * at some point in time or wait to the end of the world :)
1288 ret
= write(apps_cmd_pipe
[1], ust_cmd
,
1289 sizeof(struct ust_command
));
1291 PERROR("write apps cmd pipe");
1292 if (errno
== EBADF
) {
1294 * We can't inform the application thread to process
1295 * registration. We will exit or else application
1296 * registration will not occur and tracing will never
1303 } while (node
!= NULL
);
1305 /* Futex wait on queue. Blocking call on futex() */
1306 futex_nto1_wait(&ust_cmd_queue
.futex
);
1310 DBG("Dispatch thread dying");
1315 * This thread manage application registration.
1317 static void *thread_registration_apps(void *data
)
1319 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1320 uint32_t revents
, nb_fd
;
1321 struct lttng_poll_event events
;
1323 * Get allocated in this thread, enqueued to a global queue, dequeued and
1324 * freed in the manage apps thread.
1326 struct ust_command
*ust_cmd
= NULL
;
1328 DBG("[thread] Manage application registration started");
1330 testpoint(thread_registration_apps
);
1332 ret
= lttcomm_listen_unix_sock(apps_sock
);
1338 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1339 * more will be added to this poll set.
1341 ret
= create_thread_poll_set(&events
, 2);
1343 goto error_create_poll
;
1346 /* Add the application registration socket */
1347 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1349 goto error_poll_add
;
1352 /* Notify all applications to register */
1353 ret
= notify_ust_apps(1);
1355 ERR("Failed to notify applications or create the wait shared memory.\n"
1356 "Execution continues but there might be problem for already\n"
1357 "running applications that wishes to register.");
1361 DBG("Accepting application registration");
1363 nb_fd
= LTTNG_POLL_GETNB(&events
);
1365 /* Inifinite blocking call, waiting for transmission */
1367 health_poll_update(&health_thread_app_reg
);
1368 ret
= lttng_poll_wait(&events
, -1);
1369 health_poll_update(&health_thread_app_reg
);
1372 * Restart interrupted system call.
1374 if (errno
== EINTR
) {
1380 for (i
= 0; i
< nb_fd
; i
++) {
1381 health_code_update(&health_thread_app_reg
);
1383 /* Fetch once the poll data */
1384 revents
= LTTNG_POLL_GETEV(&events
, i
);
1385 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1387 /* Thread quit pipe has been closed. Killing thread. */
1388 ret
= check_thread_quit_pipe(pollfd
, revents
);
1394 /* Event on the registration socket */
1395 if (pollfd
== apps_sock
) {
1396 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1397 ERR("Register apps socket poll error");
1399 } else if (revents
& LPOLLIN
) {
1400 sock
= lttcomm_accept_unix_sock(apps_sock
);
1405 /* Create UST registration command for enqueuing */
1406 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1407 if (ust_cmd
== NULL
) {
1408 PERROR("ust command zmalloc");
1413 * Using message-based transmissions to ensure we don't
1414 * have to deal with partially received messages.
1416 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1418 ERR("Exhausted file descriptors allowed for applications.");
1427 health_code_update(&health_thread_app_reg
);
1428 ret
= lttcomm_recv_unix_sock(sock
, &ust_cmd
->reg_msg
,
1429 sizeof(struct ust_register_msg
));
1430 if (ret
< 0 || ret
< sizeof(struct ust_register_msg
)) {
1432 PERROR("lttcomm_recv_unix_sock register apps");
1434 ERR("Wrong size received on apps register");
1441 lttng_fd_put(LTTNG_FD_APPS
, 1);
1445 health_code_update(&health_thread_app_reg
);
1447 ust_cmd
->sock
= sock
;
1450 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1451 " gid:%d sock:%d name:%s (version %d.%d)",
1452 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1453 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1454 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1455 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1458 * Lock free enqueue the registration request. The red pill
1459 * has been taken! This apps will be part of the *system*.
1461 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1464 * Wake the registration queue futex. Implicit memory
1465 * barrier with the exchange in cds_wfq_enqueue.
1467 futex_nto1_wake(&ust_cmd_queue
.futex
);
1476 health_error(&health_thread_app_reg
);
1477 ERR("Health error occurred in %s", __func__
);
1479 health_exit(&health_thread_app_reg
);
1481 /* Notify that the registration thread is gone */
1484 if (apps_sock
>= 0) {
1485 ret
= close(apps_sock
);
1495 lttng_fd_put(LTTNG_FD_APPS
, 1);
1497 unlink(apps_unix_sock_path
);
1500 lttng_poll_clean(&events
);
1503 DBG("UST Registration thread cleanup complete");
1509 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1510 * exec or it will fails.
1512 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1515 struct timespec timeout
;
1517 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1518 consumer_data
->consumer_thread_is_ready
= 0;
1520 /* Setup pthread condition */
1521 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1524 PERROR("pthread_condattr_init consumer data");
1529 * Set the monotonic clock in order to make sure we DO NOT jump in time
1530 * between the clock_gettime() call and the timedwait call. See bug #324
1531 * for a more details and how we noticed it.
1533 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1536 PERROR("pthread_condattr_setclock consumer data");
1540 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1543 PERROR("pthread_cond_init consumer data");
1547 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1550 PERROR("pthread_create consumer");
1555 /* We are about to wait on a pthread condition */
1556 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1558 /* Get time for sem_timedwait absolute timeout */
1559 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1561 * Set the timeout for the condition timed wait even if the clock gettime
1562 * call fails since we might loop on that call and we want to avoid to
1563 * increment the timeout too many times.
1565 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1568 * The following loop COULD be skipped in some conditions so this is why we
1569 * set ret to 0 in order to make sure at least one round of the loop is
1575 * Loop until the condition is reached or when a timeout is reached. Note
1576 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1577 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1578 * possible. This loop does not take any chances and works with both of
1581 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1582 if (clock_ret
< 0) {
1583 PERROR("clock_gettime spawn consumer");
1584 /* Infinite wait for the consumerd thread to be ready */
1585 ret
= pthread_cond_wait(&consumer_data
->cond
,
1586 &consumer_data
->cond_mutex
);
1588 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1589 &consumer_data
->cond_mutex
, &timeout
);
1593 /* Release the pthread condition */
1594 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1598 if (ret
== ETIMEDOUT
) {
1600 * Call has timed out so we kill the kconsumerd_thread and return
1603 ERR("Condition timed out. The consumer thread was never ready."
1605 ret
= pthread_cancel(consumer_data
->thread
);
1607 PERROR("pthread_cancel consumer thread");
1610 PERROR("pthread_cond_wait failed consumer thread");
1615 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1616 if (consumer_data
->pid
== 0) {
1617 ERR("Consumerd did not start");
1618 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1621 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1630 * Join consumer thread
1632 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1637 /* Consumer pid must be a real one. */
1638 if (consumer_data
->pid
> 0) {
1639 ret
= kill(consumer_data
->pid
, SIGTERM
);
1641 ERR("Error killing consumer daemon");
1644 return pthread_join(consumer_data
->thread
, &status
);
1651 * Fork and exec a consumer daemon (consumerd).
1653 * Return pid if successful else -1.
1655 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1659 const char *consumer_to_use
;
1660 const char *verbosity
;
1663 DBG("Spawning consumerd");
1670 if (opt_verbose_consumer
) {
1671 verbosity
= "--verbose";
1673 verbosity
= "--quiet";
1675 switch (consumer_data
->type
) {
1676 case LTTNG_CONSUMER_KERNEL
:
1678 * Find out which consumerd to execute. We will first try the
1679 * 64-bit path, then the sessiond's installation directory, and
1680 * fallback on the 32-bit one,
1682 DBG3("Looking for a kernel consumer at these locations:");
1683 DBG3(" 1) %s", consumerd64_bin
);
1684 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1685 DBG3(" 3) %s", consumerd32_bin
);
1686 if (stat(consumerd64_bin
, &st
) == 0) {
1687 DBG3("Found location #1");
1688 consumer_to_use
= consumerd64_bin
;
1689 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1690 DBG3("Found location #2");
1691 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1692 } else if (stat(consumerd32_bin
, &st
) == 0) {
1693 DBG3("Found location #3");
1694 consumer_to_use
= consumerd32_bin
;
1696 DBG("Could not find any valid consumerd executable");
1699 DBG("Using kernel consumer at: %s", consumer_to_use
);
1700 execl(consumer_to_use
,
1701 "lttng-consumerd", verbosity
, "-k",
1702 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1703 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1706 case LTTNG_CONSUMER64_UST
:
1708 char *tmpnew
= NULL
;
1710 if (consumerd64_libdir
[0] != '\0') {
1714 tmp
= getenv("LD_LIBRARY_PATH");
1718 tmplen
= strlen("LD_LIBRARY_PATH=")
1719 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1720 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1725 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1726 strcat(tmpnew
, consumerd64_libdir
);
1727 if (tmp
[0] != '\0') {
1728 strcat(tmpnew
, ":");
1729 strcat(tmpnew
, tmp
);
1731 ret
= putenv(tmpnew
);
1737 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1738 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1739 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1740 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1742 if (consumerd64_libdir
[0] != '\0') {
1750 case LTTNG_CONSUMER32_UST
:
1752 char *tmpnew
= NULL
;
1754 if (consumerd32_libdir
[0] != '\0') {
1758 tmp
= getenv("LD_LIBRARY_PATH");
1762 tmplen
= strlen("LD_LIBRARY_PATH=")
1763 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1764 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1769 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1770 strcat(tmpnew
, consumerd32_libdir
);
1771 if (tmp
[0] != '\0') {
1772 strcat(tmpnew
, ":");
1773 strcat(tmpnew
, tmp
);
1775 ret
= putenv(tmpnew
);
1781 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
1782 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
1783 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1784 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1786 if (consumerd32_libdir
[0] != '\0') {
1795 PERROR("unknown consumer type");
1799 PERROR("kernel start consumer exec");
1802 } else if (pid
> 0) {
1805 PERROR("start consumer fork");
1813 * Spawn the consumerd daemon and session daemon thread.
1815 static int start_consumerd(struct consumer_data
*consumer_data
)
1819 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1820 if (consumer_data
->pid
!= 0) {
1821 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1825 ret
= spawn_consumerd(consumer_data
);
1827 ERR("Spawning consumerd failed");
1828 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1832 /* Setting up the consumer_data pid */
1833 consumer_data
->pid
= ret
;
1834 DBG2("Consumer pid %d", consumer_data
->pid
);
1835 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1837 DBG2("Spawning consumer control thread");
1838 ret
= spawn_consumer_thread(consumer_data
);
1840 ERR("Fatal error spawning consumer control thread");
1852 * Compute health status of each consumer. If one of them is zero (bad
1853 * state), we return 0.
1855 static int check_consumer_health(void)
1859 ret
= health_check_state(&kconsumer_data
.health
) &&
1860 health_check_state(&ustconsumer32_data
.health
) &&
1861 health_check_state(&ustconsumer64_data
.health
);
1863 DBG3("Health consumer check %d", ret
);
1869 * Setup necessary data for kernel tracer action.
1871 static int init_kernel_tracer(void)
1875 /* Modprobe lttng kernel modules */
1876 ret
= modprobe_lttng_control();
1881 /* Open debugfs lttng */
1882 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1883 if (kernel_tracer_fd
< 0) {
1884 DBG("Failed to open %s", module_proc_lttng
);
1889 /* Validate kernel version */
1890 ret
= kernel_validate_version(kernel_tracer_fd
);
1895 ret
= modprobe_lttng_data();
1900 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1904 modprobe_remove_lttng_control();
1905 ret
= close(kernel_tracer_fd
);
1909 kernel_tracer_fd
= -1;
1910 return LTTNG_ERR_KERN_VERSION
;
1913 ret
= close(kernel_tracer_fd
);
1919 modprobe_remove_lttng_control();
1922 WARN("No kernel tracer available");
1923 kernel_tracer_fd
= -1;
1925 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
1927 return LTTNG_ERR_KERN_NA
;
1933 * Copy consumer output from the tracing session to the domain session. The
1934 * function also applies the right modification on a per domain basis for the
1935 * trace files destination directory.
1937 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
1940 const char *dir_name
;
1941 struct consumer_output
*consumer
;
1944 assert(session
->consumer
);
1947 case LTTNG_DOMAIN_KERNEL
:
1948 DBG3("Copying tracing session consumer output in kernel session");
1950 * XXX: We should audit the session creation and what this function
1951 * does "extra" in order to avoid a destroy since this function is used
1952 * in the domain session creation (kernel and ust) only. Same for UST
1955 if (session
->kernel_session
->consumer
) {
1956 consumer_destroy_output(session
->kernel_session
->consumer
);
1958 session
->kernel_session
->consumer
=
1959 consumer_copy_output(session
->consumer
);
1960 /* Ease our life a bit for the next part */
1961 consumer
= session
->kernel_session
->consumer
;
1962 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
1964 case LTTNG_DOMAIN_UST
:
1965 DBG3("Copying tracing session consumer output in UST session");
1966 if (session
->ust_session
->consumer
) {
1967 consumer_destroy_output(session
->ust_session
->consumer
);
1969 session
->ust_session
->consumer
=
1970 consumer_copy_output(session
->consumer
);
1971 /* Ease our life a bit for the next part */
1972 consumer
= session
->ust_session
->consumer
;
1973 dir_name
= DEFAULT_UST_TRACE_DIR
;
1976 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
1980 /* Append correct directory to subdir */
1981 strncat(consumer
->subdir
, dir_name
,
1982 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
1983 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
1992 * Create an UST session and add it to the session ust list.
1994 static int create_ust_session(struct ltt_session
*session
,
1995 struct lttng_domain
*domain
)
1998 struct ltt_ust_session
*lus
= NULL
;
2002 assert(session
->consumer
);
2004 switch (domain
->type
) {
2005 case LTTNG_DOMAIN_UST
:
2008 ERR("Unknown UST domain on create session %d", domain
->type
);
2009 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2013 DBG("Creating UST session");
2015 lus
= trace_ust_create_session(session
->path
, session
->id
, domain
);
2017 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2021 lus
->uid
= session
->uid
;
2022 lus
->gid
= session
->gid
;
2023 session
->ust_session
= lus
;
2025 /* Copy session output to the newly created UST session */
2026 ret
= copy_session_consumer(domain
->type
, session
);
2027 if (ret
!= LTTNG_OK
) {
2035 session
->ust_session
= NULL
;
2040 * Create a kernel tracer session then create the default channel.
2042 static int create_kernel_session(struct ltt_session
*session
)
2046 DBG("Creating kernel session");
2048 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2050 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2054 /* Code flow safety */
2055 assert(session
->kernel_session
);
2057 /* Copy session output to the newly created Kernel session */
2058 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2059 if (ret
!= LTTNG_OK
) {
2063 /* Create directory(ies) on local filesystem. */
2064 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2065 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2066 ret
= run_as_mkdir_recursive(
2067 session
->kernel_session
->consumer
->dst
.trace_path
,
2068 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2070 if (ret
!= -EEXIST
) {
2071 ERR("Trace directory creation error");
2077 session
->kernel_session
->uid
= session
->uid
;
2078 session
->kernel_session
->gid
= session
->gid
;
2083 trace_kernel_destroy_session(session
->kernel_session
);
2084 session
->kernel_session
= NULL
;
2089 * Count number of session permitted by uid/gid.
2091 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2094 struct ltt_session
*session
;
2096 DBG("Counting number of available session for UID %d GID %d",
2098 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2100 * Only list the sessions the user can control.
2102 if (!session_access_ok(session
, uid
, gid
)) {
2111 * Process the command requested by the lttng client within the command
2112 * context structure. This function make sure that the return structure (llm)
2113 * is set and ready for transmission before returning.
2115 * Return any error encountered or 0 for success.
2117 * "sock" is only used for special-case var. len data.
2119 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2123 int need_tracing_session
= 1;
2126 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2130 switch (cmd_ctx
->lsm
->cmd_type
) {
2131 case LTTNG_CREATE_SESSION
:
2132 case LTTNG_DESTROY_SESSION
:
2133 case LTTNG_LIST_SESSIONS
:
2134 case LTTNG_LIST_DOMAINS
:
2135 case LTTNG_START_TRACE
:
2136 case LTTNG_STOP_TRACE
:
2137 case LTTNG_DATA_PENDING
:
2144 if (opt_no_kernel
&& need_domain
2145 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2147 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2149 ret
= LTTNG_ERR_KERN_NA
;
2154 /* Deny register consumer if we already have a spawned consumer. */
2155 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2156 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2157 if (kconsumer_data
.pid
> 0) {
2158 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2159 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2162 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2166 * Check for command that don't needs to allocate a returned payload. We do
2167 * this here so we don't have to make the call for no payload at each
2170 switch(cmd_ctx
->lsm
->cmd_type
) {
2171 case LTTNG_LIST_SESSIONS
:
2172 case LTTNG_LIST_TRACEPOINTS
:
2173 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2174 case LTTNG_LIST_DOMAINS
:
2175 case LTTNG_LIST_CHANNELS
:
2176 case LTTNG_LIST_EVENTS
:
2179 /* Setup lttng message with no payload */
2180 ret
= setup_lttng_msg(cmd_ctx
, 0);
2182 /* This label does not try to unlock the session */
2183 goto init_setup_error
;
2187 /* Commands that DO NOT need a session. */
2188 switch (cmd_ctx
->lsm
->cmd_type
) {
2189 case LTTNG_CREATE_SESSION
:
2190 case LTTNG_CALIBRATE
:
2191 case LTTNG_LIST_SESSIONS
:
2192 case LTTNG_LIST_TRACEPOINTS
:
2193 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2194 need_tracing_session
= 0;
2197 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2199 * We keep the session list lock across _all_ commands
2200 * for now, because the per-session lock does not
2201 * handle teardown properly.
2203 session_lock_list();
2204 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2205 if (cmd_ctx
->session
== NULL
) {
2206 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2207 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2209 /* If no session name specified */
2210 ret
= LTTNG_ERR_SELECT_SESS
;
2214 /* Acquire lock for the session */
2215 session_lock(cmd_ctx
->session
);
2225 * Check domain type for specific "pre-action".
2227 switch (cmd_ctx
->lsm
->domain
.type
) {
2228 case LTTNG_DOMAIN_KERNEL
:
2230 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2234 /* Kernel tracer check */
2235 if (kernel_tracer_fd
== -1) {
2236 /* Basically, load kernel tracer modules */
2237 ret
= init_kernel_tracer();
2243 /* Consumer is in an ERROR state. Report back to client */
2244 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2245 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2249 /* Need a session for kernel command */
2250 if (need_tracing_session
) {
2251 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2252 ret
= create_kernel_session(cmd_ctx
->session
);
2254 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2259 /* Start the kernel consumer daemon */
2260 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2261 if (kconsumer_data
.pid
== 0 &&
2262 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2263 cmd_ctx
->session
->start_consumer
) {
2264 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2265 ret
= start_consumerd(&kconsumer_data
);
2267 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2270 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2272 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2276 * The consumer was just spawned so we need to add the socket to
2277 * the consumer output of the session if exist.
2279 ret
= consumer_create_socket(&kconsumer_data
,
2280 cmd_ctx
->session
->kernel_session
->consumer
);
2287 case LTTNG_DOMAIN_UST
:
2289 /* Consumer is in an ERROR state. Report back to client */
2290 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2291 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2295 if (need_tracing_session
) {
2296 /* Create UST session if none exist. */
2297 if (cmd_ctx
->session
->ust_session
== NULL
) {
2298 ret
= create_ust_session(cmd_ctx
->session
,
2299 &cmd_ctx
->lsm
->domain
);
2300 if (ret
!= LTTNG_OK
) {
2305 /* Start the UST consumer daemons */
2307 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2308 if (consumerd64_bin
[0] != '\0' &&
2309 ustconsumer64_data
.pid
== 0 &&
2310 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2311 cmd_ctx
->session
->start_consumer
) {
2312 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2313 ret
= start_consumerd(&ustconsumer64_data
);
2315 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2316 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2320 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2321 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2323 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2327 * Setup socket for consumer 64 bit. No need for atomic access
2328 * since it was set above and can ONLY be set in this thread.
2330 ret
= consumer_create_socket(&ustconsumer64_data
,
2331 cmd_ctx
->session
->ust_session
->consumer
);
2337 if (consumerd32_bin
[0] != '\0' &&
2338 ustconsumer32_data
.pid
== 0 &&
2339 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2340 cmd_ctx
->session
->start_consumer
) {
2341 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2342 ret
= start_consumerd(&ustconsumer32_data
);
2344 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2345 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2349 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2350 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2352 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2356 * Setup socket for consumer 64 bit. No need for atomic access
2357 * since it was set above and can ONLY be set in this thread.
2359 ret
= consumer_create_socket(&ustconsumer32_data
,
2360 cmd_ctx
->session
->ust_session
->consumer
);
2372 /* Validate consumer daemon state when start/stop trace command */
2373 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2374 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2375 switch (cmd_ctx
->lsm
->domain
.type
) {
2376 case LTTNG_DOMAIN_UST
:
2377 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2378 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2382 case LTTNG_DOMAIN_KERNEL
:
2383 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2384 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2392 * Check that the UID or GID match that of the tracing session.
2393 * The root user can interact with all sessions.
2395 if (need_tracing_session
) {
2396 if (!session_access_ok(cmd_ctx
->session
,
2397 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2398 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2399 ret
= LTTNG_ERR_EPERM
;
2404 /* Process by command type */
2405 switch (cmd_ctx
->lsm
->cmd_type
) {
2406 case LTTNG_ADD_CONTEXT
:
2408 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2409 cmd_ctx
->lsm
->u
.context
.channel_name
,
2410 cmd_ctx
->lsm
->u
.context
.event_name
,
2411 &cmd_ctx
->lsm
->u
.context
.ctx
);
2414 case LTTNG_DISABLE_CHANNEL
:
2416 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2417 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2420 case LTTNG_DISABLE_EVENT
:
2422 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2423 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2424 cmd_ctx
->lsm
->u
.disable
.name
);
2427 case LTTNG_DISABLE_ALL_EVENT
:
2429 DBG("Disabling all events");
2431 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2432 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2435 case LTTNG_DISABLE_CONSUMER
:
2437 ret
= cmd_disable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2440 case LTTNG_ENABLE_CHANNEL
:
2442 ret
= cmd_enable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2443 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2446 case LTTNG_ENABLE_CONSUMER
:
2449 * XXX: 0 means that this URI should be applied on the session. Should
2450 * be a DOMAIN enuam.
2452 ret
= cmd_enable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2453 if (ret
!= LTTNG_OK
) {
2457 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2458 /* Add the URI for the UST session if a consumer is present. */
2459 if (cmd_ctx
->session
->ust_session
&&
2460 cmd_ctx
->session
->ust_session
->consumer
) {
2461 ret
= cmd_enable_consumer(LTTNG_DOMAIN_UST
, cmd_ctx
->session
);
2462 } else if (cmd_ctx
->session
->kernel_session
&&
2463 cmd_ctx
->session
->kernel_session
->consumer
) {
2464 ret
= cmd_enable_consumer(LTTNG_DOMAIN_KERNEL
,
2470 case LTTNG_ENABLE_EVENT
:
2472 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2473 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2474 &cmd_ctx
->lsm
->u
.enable
.event
, kernel_poll_pipe
[1]);
2477 case LTTNG_ENABLE_ALL_EVENT
:
2479 DBG("Enabling all events");
2481 ret
= cmd_enable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2482 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2483 cmd_ctx
->lsm
->u
.enable
.event
.type
, kernel_poll_pipe
[1]);
2486 case LTTNG_LIST_TRACEPOINTS
:
2488 struct lttng_event
*events
;
2491 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2492 if (nb_events
< 0) {
2493 /* Return value is a negative lttng_error_code. */
2499 * Setup lttng message with payload size set to the event list size in
2500 * bytes and then copy list into the llm payload.
2502 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2508 /* Copy event list into message payload */
2509 memcpy(cmd_ctx
->llm
->payload
, events
,
2510 sizeof(struct lttng_event
) * nb_events
);
2517 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2519 struct lttng_event_field
*fields
;
2522 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2524 if (nb_fields
< 0) {
2525 /* Return value is a negative lttng_error_code. */
2531 * Setup lttng message with payload size set to the event list size in
2532 * bytes and then copy list into the llm payload.
2534 ret
= setup_lttng_msg(cmd_ctx
,
2535 sizeof(struct lttng_event_field
) * nb_fields
);
2541 /* Copy event list into message payload */
2542 memcpy(cmd_ctx
->llm
->payload
, fields
,
2543 sizeof(struct lttng_event_field
) * nb_fields
);
2550 case LTTNG_SET_CONSUMER_URI
:
2553 struct lttng_uri
*uris
;
2555 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2556 len
= nb_uri
* sizeof(struct lttng_uri
);
2559 ret
= LTTNG_ERR_INVALID
;
2563 uris
= zmalloc(len
);
2565 ret
= LTTNG_ERR_FATAL
;
2569 /* Receive variable len data */
2570 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2571 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2573 DBG("No URIs received from client... continuing");
2575 ret
= LTTNG_ERR_SESSION_FAIL
;
2580 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2582 if (ret
!= LTTNG_OK
) {
2588 * XXX: 0 means that this URI should be applied on the session. Should
2589 * be a DOMAIN enuam.
2591 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2592 /* Add the URI for the UST session if a consumer is present. */
2593 if (cmd_ctx
->session
->ust_session
&&
2594 cmd_ctx
->session
->ust_session
->consumer
) {
2595 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2597 } else if (cmd_ctx
->session
->kernel_session
&&
2598 cmd_ctx
->session
->kernel_session
->consumer
) {
2599 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2600 cmd_ctx
->session
, nb_uri
, uris
);
2608 case LTTNG_START_TRACE
:
2610 ret
= cmd_start_trace(cmd_ctx
->session
);
2613 case LTTNG_STOP_TRACE
:
2615 ret
= cmd_stop_trace(cmd_ctx
->session
);
2618 case LTTNG_CREATE_SESSION
:
2621 struct lttng_uri
*uris
= NULL
;
2623 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2624 len
= nb_uri
* sizeof(struct lttng_uri
);
2627 uris
= zmalloc(len
);
2629 ret
= LTTNG_ERR_FATAL
;
2633 /* Receive variable len data */
2634 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2635 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2637 DBG("No URIs received from client... continuing");
2639 ret
= LTTNG_ERR_SESSION_FAIL
;
2644 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2645 DBG("Creating session with ONE network URI is a bad call");
2646 ret
= LTTNG_ERR_SESSION_FAIL
;
2652 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2659 case LTTNG_DESTROY_SESSION
:
2661 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2663 /* Set session to NULL so we do not unlock it after free. */
2664 cmd_ctx
->session
= NULL
;
2667 case LTTNG_LIST_DOMAINS
:
2670 struct lttng_domain
*domains
;
2672 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2674 /* Return value is a negative lttng_error_code. */
2679 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2684 /* Copy event list into message payload */
2685 memcpy(cmd_ctx
->llm
->payload
, domains
,
2686 nb_dom
* sizeof(struct lttng_domain
));
2693 case LTTNG_LIST_CHANNELS
:
2696 struct lttng_channel
*channels
;
2698 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2699 cmd_ctx
->session
, &channels
);
2701 /* Return value is a negative lttng_error_code. */
2706 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2711 /* Copy event list into message payload */
2712 memcpy(cmd_ctx
->llm
->payload
, channels
,
2713 nb_chan
* sizeof(struct lttng_channel
));
2720 case LTTNG_LIST_EVENTS
:
2723 struct lttng_event
*events
= NULL
;
2725 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2726 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2728 /* Return value is a negative lttng_error_code. */
2733 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2738 /* Copy event list into message payload */
2739 memcpy(cmd_ctx
->llm
->payload
, events
,
2740 nb_event
* sizeof(struct lttng_event
));
2747 case LTTNG_LIST_SESSIONS
:
2749 unsigned int nr_sessions
;
2751 session_lock_list();
2752 nr_sessions
= lttng_sessions_count(
2753 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2754 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2756 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2758 session_unlock_list();
2762 /* Filled the session array */
2763 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2764 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2765 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2767 session_unlock_list();
2772 case LTTNG_CALIBRATE
:
2774 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
2775 &cmd_ctx
->lsm
->u
.calibrate
);
2778 case LTTNG_REGISTER_CONSUMER
:
2780 struct consumer_data
*cdata
;
2782 switch (cmd_ctx
->lsm
->domain
.type
) {
2783 case LTTNG_DOMAIN_KERNEL
:
2784 cdata
= &kconsumer_data
;
2787 ret
= LTTNG_ERR_UND
;
2791 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2792 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
2795 case LTTNG_SET_FILTER
:
2797 struct lttng_filter_bytecode
*bytecode
;
2799 if (cmd_ctx
->lsm
->u
.filter
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
2800 ret
= LTTNG_ERR_FILTER_INVAL
;
2803 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.filter
.bytecode_len
);
2805 ret
= LTTNG_ERR_FILTER_NOMEM
;
2808 /* Receive var. len. data */
2809 DBG("Receiving var len data from client ...");
2810 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
2811 cmd_ctx
->lsm
->u
.filter
.bytecode_len
);
2813 DBG("Nothing recv() from client var len data... continuing");
2815 ret
= LTTNG_ERR_FILTER_INVAL
;
2819 if (bytecode
->len
+ sizeof(*bytecode
)
2820 != cmd_ctx
->lsm
->u
.filter
.bytecode_len
) {
2822 ret
= LTTNG_ERR_FILTER_INVAL
;
2826 ret
= cmd_set_filter(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2827 cmd_ctx
->lsm
->u
.filter
.channel_name
,
2828 cmd_ctx
->lsm
->u
.filter
.event_name
,
2832 case LTTNG_DATA_PENDING
:
2834 ret
= cmd_data_pending(cmd_ctx
->session
);
2838 ret
= LTTNG_ERR_UND
;
2843 if (cmd_ctx
->llm
== NULL
) {
2844 DBG("Missing llm structure. Allocating one.");
2845 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
2849 /* Set return code */
2850 cmd_ctx
->llm
->ret_code
= ret
;
2852 if (cmd_ctx
->session
) {
2853 session_unlock(cmd_ctx
->session
);
2855 if (need_tracing_session
) {
2856 session_unlock_list();
2863 * Thread managing health check socket.
2865 static void *thread_manage_health(void *data
)
2867 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
2868 uint32_t revents
, nb_fd
;
2869 struct lttng_poll_event events
;
2870 struct lttcomm_health_msg msg
;
2871 struct lttcomm_health_data reply
;
2873 DBG("[thread] Manage health check started");
2875 rcu_register_thread();
2877 /* Create unix socket */
2878 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
2880 ERR("Unable to create health check Unix socket");
2885 ret
= lttcomm_listen_unix_sock(sock
);
2891 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2892 * more will be added to this poll set.
2894 ret
= create_thread_poll_set(&events
, 2);
2899 /* Add the application registration socket */
2900 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
2906 DBG("Health check ready");
2908 nb_fd
= LTTNG_POLL_GETNB(&events
);
2910 /* Inifinite blocking call, waiting for transmission */
2912 ret
= lttng_poll_wait(&events
, -1);
2915 * Restart interrupted system call.
2917 if (errno
== EINTR
) {
2923 for (i
= 0; i
< nb_fd
; i
++) {
2924 /* Fetch once the poll data */
2925 revents
= LTTNG_POLL_GETEV(&events
, i
);
2926 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2928 /* Thread quit pipe has been closed. Killing thread. */
2929 ret
= check_thread_quit_pipe(pollfd
, revents
);
2935 /* Event on the registration socket */
2936 if (pollfd
== sock
) {
2937 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2938 ERR("Health socket poll error");
2944 new_sock
= lttcomm_accept_unix_sock(sock
);
2949 DBG("Receiving data from client for health...");
2950 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
2952 DBG("Nothing recv() from client... continuing");
2953 ret
= close(new_sock
);
2961 rcu_thread_online();
2963 switch (msg
.component
) {
2964 case LTTNG_HEALTH_CMD
:
2965 reply
.ret_code
= health_check_state(&health_thread_cmd
);
2967 case LTTNG_HEALTH_APP_MANAGE
:
2968 reply
.ret_code
= health_check_state(&health_thread_app_manage
);
2970 case LTTNG_HEALTH_APP_REG
:
2971 reply
.ret_code
= health_check_state(&health_thread_app_reg
);
2973 case LTTNG_HEALTH_KERNEL
:
2974 reply
.ret_code
= health_check_state(&health_thread_kernel
);
2976 case LTTNG_HEALTH_CONSUMER
:
2977 reply
.ret_code
= check_consumer_health();
2979 case LTTNG_HEALTH_ALL
:
2981 health_check_state(&health_thread_app_manage
) &&
2982 health_check_state(&health_thread_app_reg
) &&
2983 health_check_state(&health_thread_cmd
) &&
2984 health_check_state(&health_thread_kernel
) &&
2985 check_consumer_health();
2988 reply
.ret_code
= LTTNG_ERR_UND
;
2993 * Flip ret value since 0 is a success and 1 indicates a bad health for
2994 * the client where in the sessiond it is the opposite. Again, this is
2995 * just to make things easier for us poor developer which enjoy a lot
2998 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
2999 reply
.ret_code
= !reply
.ret_code
;
3002 DBG2("Health check return value %d", reply
.ret_code
);
3004 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3006 ERR("Failed to send health data back to client");
3009 /* End of transmission */
3010 ret
= close(new_sock
);
3020 ERR("Health error occurred in %s", __func__
);
3022 DBG("Health check thread dying");
3023 unlink(health_unix_sock_path
);
3030 if (new_sock
>= 0) {
3031 ret
= close(new_sock
);
3037 lttng_poll_clean(&events
);
3039 rcu_unregister_thread();
3044 * This thread manage all clients request using the unix client socket for
3047 static void *thread_manage_clients(void *data
)
3049 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3051 uint32_t revents
, nb_fd
;
3052 struct command_ctx
*cmd_ctx
= NULL
;
3053 struct lttng_poll_event events
;
3055 DBG("[thread] Manage client started");
3057 testpoint(thread_manage_clients
);
3059 rcu_register_thread();
3061 health_code_update(&health_thread_cmd
);
3063 ret
= lttcomm_listen_unix_sock(client_sock
);
3069 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3070 * more will be added to this poll set.
3072 ret
= create_thread_poll_set(&events
, 2);
3077 /* Add the application registration socket */
3078 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3084 * Notify parent pid that we are ready to accept command for client side.
3086 if (opt_sig_parent
) {
3087 kill(ppid
, SIGUSR1
);
3090 testpoint(thread_manage_clients_before_loop
);
3092 health_code_update(&health_thread_cmd
);
3095 DBG("Accepting client command ...");
3097 nb_fd
= LTTNG_POLL_GETNB(&events
);
3099 /* Inifinite blocking call, waiting for transmission */
3101 health_poll_update(&health_thread_cmd
);
3102 ret
= lttng_poll_wait(&events
, -1);
3103 health_poll_update(&health_thread_cmd
);
3106 * Restart interrupted system call.
3108 if (errno
== EINTR
) {
3114 for (i
= 0; i
< nb_fd
; i
++) {
3115 /* Fetch once the poll data */
3116 revents
= LTTNG_POLL_GETEV(&events
, i
);
3117 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3119 health_code_update(&health_thread_cmd
);
3121 /* Thread quit pipe has been closed. Killing thread. */
3122 ret
= check_thread_quit_pipe(pollfd
, revents
);
3128 /* Event on the registration socket */
3129 if (pollfd
== client_sock
) {
3130 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3131 ERR("Client socket poll error");
3137 DBG("Wait for client response");
3139 health_code_update(&health_thread_cmd
);
3141 sock
= lttcomm_accept_unix_sock(client_sock
);
3146 /* Set socket option for credentials retrieval */
3147 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3152 /* Allocate context command to process the client request */
3153 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3154 if (cmd_ctx
== NULL
) {
3155 PERROR("zmalloc cmd_ctx");
3159 /* Allocate data buffer for reception */
3160 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3161 if (cmd_ctx
->lsm
== NULL
) {
3162 PERROR("zmalloc cmd_ctx->lsm");
3166 cmd_ctx
->llm
= NULL
;
3167 cmd_ctx
->session
= NULL
;
3169 health_code_update(&health_thread_cmd
);
3172 * Data is received from the lttng client. The struct
3173 * lttcomm_session_msg (lsm) contains the command and data request of
3176 DBG("Receiving data from client ...");
3177 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3178 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3180 DBG("Nothing recv() from client... continuing");
3186 clean_command_ctx(&cmd_ctx
);
3190 health_code_update(&health_thread_cmd
);
3192 // TODO: Validate cmd_ctx including sanity check for
3193 // security purpose.
3195 rcu_thread_online();
3197 * This function dispatch the work to the kernel or userspace tracer
3198 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3199 * informations for the client. The command context struct contains
3200 * everything this function may needs.
3202 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3203 rcu_thread_offline();
3213 * TODO: Inform client somehow of the fatal error. At
3214 * this point, ret < 0 means that a zmalloc failed
3215 * (ENOMEM). Error detected but still accept
3216 * command, unless a socket error has been
3219 clean_command_ctx(&cmd_ctx
);
3223 health_code_update(&health_thread_cmd
);
3225 DBG("Sending response (size: %d, retcode: %s)",
3226 cmd_ctx
->lttng_msg_size
,
3227 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3228 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3230 ERR("Failed to send data back to client");
3233 /* End of transmission */
3240 clean_command_ctx(&cmd_ctx
);
3242 health_code_update(&health_thread_cmd
);
3248 health_error(&health_thread_cmd
);
3249 ERR("Health error occurred in %s", __func__
);
3251 health_exit(&health_thread_cmd
);
3253 DBG("Client thread dying");
3254 unlink(client_unix_sock_path
);
3255 if (client_sock
>= 0) {
3256 ret
= close(client_sock
);
3268 lttng_poll_clean(&events
);
3269 clean_command_ctx(&cmd_ctx
);
3271 rcu_unregister_thread();
3277 * usage function on stderr
3279 static void usage(void)
3281 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3282 fprintf(stderr
, " -h, --help Display this usage.\n");
3283 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3284 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3285 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3286 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3287 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3288 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3289 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3290 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3291 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3292 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3293 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3294 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3295 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3296 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3297 fprintf(stderr
, " -V, --version Show version number.\n");
3298 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3299 fprintf(stderr
, " -q, --quiet No output at all.\n");
3300 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3301 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3302 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3306 * daemon argument parsing
3308 static int parse_args(int argc
, char **argv
)
3312 static struct option long_options
[] = {
3313 { "client-sock", 1, 0, 'c' },
3314 { "apps-sock", 1, 0, 'a' },
3315 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3316 { "kconsumerd-err-sock", 1, 0, 'E' },
3317 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3318 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3319 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3320 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3321 { "consumerd32-path", 1, 0, 'u' },
3322 { "consumerd32-libdir", 1, 0, 'U' },
3323 { "consumerd64-path", 1, 0, 't' },
3324 { "consumerd64-libdir", 1, 0, 'T' },
3325 { "daemonize", 0, 0, 'd' },
3326 { "sig-parent", 0, 0, 'S' },
3327 { "help", 0, 0, 'h' },
3328 { "group", 1, 0, 'g' },
3329 { "version", 0, 0, 'V' },
3330 { "quiet", 0, 0, 'q' },
3331 { "verbose", 0, 0, 'v' },
3332 { "verbose-consumer", 0, 0, 'Z' },
3333 { "no-kernel", 0, 0, 'N' },
3338 int option_index
= 0;
3339 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3340 long_options
, &option_index
);
3347 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3349 fprintf(stderr
, " with arg %s\n", optarg
);
3353 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3356 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3362 opt_tracing_group
= optarg
;
3368 fprintf(stdout
, "%s\n", VERSION
);
3374 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3377 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3380 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3383 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3386 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3389 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3395 lttng_opt_quiet
= 1;
3398 /* Verbose level can increase using multiple -v */
3399 lttng_opt_verbose
+= 1;
3402 opt_verbose_consumer
+= 1;
3405 consumerd32_bin
= optarg
;
3408 consumerd32_libdir
= optarg
;
3411 consumerd64_bin
= optarg
;
3414 consumerd64_libdir
= optarg
;
3417 /* Unknown option or other error.
3418 * Error is printed by getopt, just return */
3427 * Creates the two needed socket by the daemon.
3428 * apps_sock - The communication socket for all UST apps.
3429 * client_sock - The communication of the cli tool (lttng).
3431 static int init_daemon_socket(void)
3436 old_umask
= umask(0);
3438 /* Create client tool unix socket */
3439 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3440 if (client_sock
< 0) {
3441 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3446 /* File permission MUST be 660 */
3447 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3449 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3454 /* Create the application unix socket */
3455 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3456 if (apps_sock
< 0) {
3457 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3462 /* File permission MUST be 666 */
3463 ret
= chmod(apps_unix_sock_path
,
3464 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3466 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3477 * Check if the global socket is available, and if a daemon is answering at the
3478 * other side. If yes, error is returned.
3480 static int check_existing_daemon(void)
3482 /* Is there anybody out there ? */
3483 if (lttng_session_daemon_alive()) {
3491 * Set the tracing group gid onto the client socket.
3493 * Race window between mkdir and chown is OK because we are going from more
3494 * permissive (root.root) to less permissive (root.tracing).
3496 static int set_permissions(char *rundir
)
3501 ret
= allowed_group();
3503 WARN("No tracing group detected");
3510 /* Set lttng run dir */
3511 ret
= chown(rundir
, 0, gid
);
3513 ERR("Unable to set group on %s", rundir
);
3517 /* Ensure tracing group can search the run dir */
3518 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3520 ERR("Unable to set permissions on %s", rundir
);
3524 /* lttng client socket path */
3525 ret
= chown(client_unix_sock_path
, 0, gid
);
3527 ERR("Unable to set group on %s", client_unix_sock_path
);
3531 /* kconsumer error socket path */
3532 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3534 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3538 /* 64-bit ustconsumer error socket path */
3539 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3541 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3545 /* 32-bit ustconsumer compat32 error socket path */
3546 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3548 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3552 DBG("All permissions are set");
3559 * Create the lttng run directory needed for all global sockets and pipe.
3561 static int create_lttng_rundir(const char *rundir
)
3565 DBG3("Creating LTTng run directory: %s", rundir
);
3567 ret
= mkdir(rundir
, S_IRWXU
);
3569 if (errno
!= EEXIST
) {
3570 ERR("Unable to create %s", rundir
);
3582 * Setup sockets and directory needed by the kconsumerd communication with the
3585 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3589 char path
[PATH_MAX
];
3591 switch (consumer_data
->type
) {
3592 case LTTNG_CONSUMER_KERNEL
:
3593 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3595 case LTTNG_CONSUMER64_UST
:
3596 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3598 case LTTNG_CONSUMER32_UST
:
3599 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3602 ERR("Consumer type unknown");
3607 DBG2("Creating consumer directory: %s", path
);
3609 ret
= mkdir(path
, S_IRWXU
);
3611 if (errno
!= EEXIST
) {
3613 ERR("Failed to create %s", path
);
3619 /* Create the kconsumerd error unix socket */
3620 consumer_data
->err_sock
=
3621 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3622 if (consumer_data
->err_sock
< 0) {
3623 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3628 /* File permission MUST be 660 */
3629 ret
= chmod(consumer_data
->err_unix_sock_path
,
3630 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3632 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3642 * Signal handler for the daemon
3644 * Simply stop all worker threads, leaving main() return gracefully after
3645 * joining all threads and calling cleanup().
3647 static void sighandler(int sig
)
3651 DBG("SIGPIPE caught");
3654 DBG("SIGINT caught");
3658 DBG("SIGTERM caught");
3667 * Setup signal handler for :
3668 * SIGINT, SIGTERM, SIGPIPE
3670 static int set_signal_handler(void)
3673 struct sigaction sa
;
3676 if ((ret
= sigemptyset(&sigset
)) < 0) {
3677 PERROR("sigemptyset");
3681 sa
.sa_handler
= sighandler
;
3682 sa
.sa_mask
= sigset
;
3684 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3685 PERROR("sigaction");
3689 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3690 PERROR("sigaction");
3694 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3695 PERROR("sigaction");
3699 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3705 * Set open files limit to unlimited. This daemon can open a large number of
3706 * file descriptors in order to consumer multiple kernel traces.
3708 static void set_ulimit(void)
3713 /* The kernel does not allowed an infinite limit for open files */
3714 lim
.rlim_cur
= 65535;
3715 lim
.rlim_max
= 65535;
3717 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
3719 PERROR("failed to set open files limit");
3726 int main(int argc
, char **argv
)
3730 const char *home_path
;
3732 init_kernel_workarounds();
3734 rcu_register_thread();
3736 setup_consumerd_path();
3738 /* Parse arguments */
3740 if ((ret
= parse_args(argc
, argv
) < 0)) {
3750 * child: setsid, close FD 0, 1, 2, chdir /
3751 * parent: exit (if fork is successful)
3759 * We are in the child. Make sure all other file
3760 * descriptors are closed, in case we are called with
3761 * more opened file descriptors than the standard ones.
3763 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3768 /* Create thread quit pipe */
3769 if ((ret
= init_thread_quit_pipe()) < 0) {
3773 /* Check if daemon is UID = 0 */
3774 is_root
= !getuid();
3777 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
3779 /* Create global run dir with root access */
3780 ret
= create_lttng_rundir(rundir
);
3785 if (strlen(apps_unix_sock_path
) == 0) {
3786 snprintf(apps_unix_sock_path
, PATH_MAX
,
3787 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
3790 if (strlen(client_unix_sock_path
) == 0) {
3791 snprintf(client_unix_sock_path
, PATH_MAX
,
3792 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
3795 /* Set global SHM for ust */
3796 if (strlen(wait_shm_path
) == 0) {
3797 snprintf(wait_shm_path
, PATH_MAX
,
3798 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
3801 if (strlen(health_unix_sock_path
) == 0) {
3802 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3803 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
3806 /* Setup kernel consumerd path */
3807 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
3808 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
3809 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
3810 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
3812 DBG2("Kernel consumer err path: %s",
3813 kconsumer_data
.err_unix_sock_path
);
3814 DBG2("Kernel consumer cmd path: %s",
3815 kconsumer_data
.cmd_unix_sock_path
);
3817 home_path
= get_home_dir();
3818 if (home_path
== NULL
) {
3819 /* TODO: Add --socket PATH option */
3820 ERR("Can't get HOME directory for sockets creation.");
3826 * Create rundir from home path. This will create something like
3829 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
3835 ret
= create_lttng_rundir(rundir
);
3840 if (strlen(apps_unix_sock_path
) == 0) {
3841 snprintf(apps_unix_sock_path
, PATH_MAX
,
3842 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
3845 /* Set the cli tool unix socket path */
3846 if (strlen(client_unix_sock_path
) == 0) {
3847 snprintf(client_unix_sock_path
, PATH_MAX
,
3848 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
3851 /* Set global SHM for ust */
3852 if (strlen(wait_shm_path
) == 0) {
3853 snprintf(wait_shm_path
, PATH_MAX
,
3854 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, geteuid());
3857 /* Set health check Unix path */
3858 if (strlen(health_unix_sock_path
) == 0) {
3859 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3860 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
3864 /* Set consumer initial state */
3865 kernel_consumerd_state
= CONSUMER_STOPPED
;
3866 ust_consumerd_state
= CONSUMER_STOPPED
;
3868 DBG("Client socket path %s", client_unix_sock_path
);
3869 DBG("Application socket path %s", apps_unix_sock_path
);
3870 DBG("LTTng run directory path: %s", rundir
);
3872 /* 32 bits consumerd path setup */
3873 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
3874 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
3875 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
3876 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
3878 DBG2("UST consumer 32 bits err path: %s",
3879 ustconsumer32_data
.err_unix_sock_path
);
3880 DBG2("UST consumer 32 bits cmd path: %s",
3881 ustconsumer32_data
.cmd_unix_sock_path
);
3883 /* 64 bits consumerd path setup */
3884 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
3885 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
3886 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
3887 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
3889 DBG2("UST consumer 64 bits err path: %s",
3890 ustconsumer64_data
.err_unix_sock_path
);
3891 DBG2("UST consumer 64 bits cmd path: %s",
3892 ustconsumer64_data
.cmd_unix_sock_path
);
3895 * See if daemon already exist.
3897 if ((ret
= check_existing_daemon()) < 0) {
3898 ERR("Already running daemon.\n");
3900 * We do not goto exit because we must not cleanup()
3901 * because a daemon is already running.
3907 * Init UST app hash table. Alloc hash table before this point since
3908 * cleanup() can get called after that point.
3912 /* After this point, we can safely call cleanup() with "goto exit" */
3915 * These actions must be executed as root. We do that *after* setting up
3916 * the sockets path because we MUST make the check for another daemon using
3917 * those paths *before* trying to set the kernel consumer sockets and init
3921 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
3926 /* Setup kernel tracer */
3927 if (!opt_no_kernel
) {
3928 init_kernel_tracer();
3931 /* Set ulimit for open files */
3934 /* init lttng_fd tracking must be done after set_ulimit. */
3937 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
3942 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
3947 if ((ret
= set_signal_handler()) < 0) {
3951 /* Setup the needed unix socket */
3952 if ((ret
= init_daemon_socket()) < 0) {
3956 /* Set credentials to socket */
3957 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
3961 /* Get parent pid if -S, --sig-parent is specified. */
3962 if (opt_sig_parent
) {
3966 /* Setup the kernel pipe for waking up the kernel thread */
3967 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
3971 /* Setup the thread apps communication pipe. */
3972 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
3976 /* Init UST command queue. */
3977 cds_wfq_init(&ust_cmd_queue
.queue
);
3980 * Get session list pointer. This pointer MUST NOT be free(). This list is
3981 * statically declared in session.c
3983 session_list_ptr
= session_get_list();
3985 /* Set up max poll set size */
3986 lttng_poll_set_max_size();
3990 /* Init all health thread counters. */
3991 health_init(&health_thread_cmd
);
3992 health_init(&health_thread_kernel
);
3993 health_init(&health_thread_app_manage
);
3994 health_init(&health_thread_app_reg
);
3997 * Init health counters of the consumer thread. We do a quick hack here to
3998 * the state of the consumer health is fine even if the thread is not
3999 * started. Once the thread starts, the health state is updated with a poll
4000 * value to set a health code path. This is simply to ease our life and has
4001 * no cost what so ever.
4003 health_init(&kconsumer_data
.health
);
4004 health_poll_update(&kconsumer_data
.health
);
4005 health_init(&ustconsumer32_data
.health
);
4006 health_poll_update(&ustconsumer32_data
.health
);
4007 health_init(&ustconsumer64_data
.health
);
4008 health_poll_update(&ustconsumer64_data
.health
);
4010 /* Create thread to manage the client socket */
4011 ret
= pthread_create(&health_thread
, NULL
,
4012 thread_manage_health
, (void *) NULL
);
4014 PERROR("pthread_create health");
4018 /* Create thread to manage the client socket */
4019 ret
= pthread_create(&client_thread
, NULL
,
4020 thread_manage_clients
, (void *) NULL
);
4022 PERROR("pthread_create clients");
4026 /* Create thread to dispatch registration */
4027 ret
= pthread_create(&dispatch_thread
, NULL
,
4028 thread_dispatch_ust_registration
, (void *) NULL
);
4030 PERROR("pthread_create dispatch");
4034 /* Create thread to manage application registration. */
4035 ret
= pthread_create(®_apps_thread
, NULL
,
4036 thread_registration_apps
, (void *) NULL
);
4038 PERROR("pthread_create registration");
4042 /* Create thread to manage application socket */
4043 ret
= pthread_create(&apps_thread
, NULL
,
4044 thread_manage_apps
, (void *) NULL
);
4046 PERROR("pthread_create apps");
4050 /* Create kernel thread to manage kernel event */
4051 ret
= pthread_create(&kernel_thread
, NULL
,
4052 thread_manage_kernel
, (void *) NULL
);
4054 PERROR("pthread_create kernel");
4058 ret
= pthread_join(kernel_thread
, &status
);
4060 PERROR("pthread_join");
4061 goto error
; /* join error, exit without cleanup */
4065 ret
= pthread_join(apps_thread
, &status
);
4067 PERROR("pthread_join");
4068 goto error
; /* join error, exit without cleanup */
4072 ret
= pthread_join(reg_apps_thread
, &status
);
4074 PERROR("pthread_join");
4075 goto error
; /* join error, exit without cleanup */
4079 ret
= pthread_join(dispatch_thread
, &status
);
4081 PERROR("pthread_join");
4082 goto error
; /* join error, exit without cleanup */
4086 ret
= pthread_join(client_thread
, &status
);
4088 PERROR("pthread_join");
4089 goto error
; /* join error, exit without cleanup */
4092 ret
= join_consumer_thread(&kconsumer_data
);
4094 PERROR("join_consumer");
4095 goto error
; /* join error, exit without cleanup */
4098 ret
= join_consumer_thread(&ustconsumer32_data
);
4100 PERROR("join_consumer ust32");
4101 goto error
; /* join error, exit without cleanup */
4104 ret
= join_consumer_thread(&ustconsumer64_data
);
4106 PERROR("join_consumer ust64");
4107 goto error
; /* join error, exit without cleanup */
4111 ret
= pthread_join(health_thread
, &status
);
4113 PERROR("pthread_join health thread");
4114 goto error
; /* join error, exit without cleanup */
4120 * cleanup() is called when no other thread is running.
4122 rcu_thread_online();
4124 rcu_thread_offline();
4125 rcu_unregister_thread();