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
);
416 DBG("Cleaning up all sessions");
418 /* Destroy session list mutex */
419 if (session_list_ptr
!= NULL
) {
420 pthread_mutex_destroy(&session_list_ptr
->lock
);
422 /* Cleanup ALL session */
423 cds_list_for_each_entry_safe(sess
, stmp
,
424 &session_list_ptr
->head
, list
) {
425 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
429 DBG("Closing all UST sockets");
430 ust_app_clean_list();
432 if (is_root
&& !opt_no_kernel
) {
433 DBG2("Closing kernel fd");
434 if (kernel_tracer_fd
>= 0) {
435 ret
= close(kernel_tracer_fd
);
440 DBG("Unloading kernel modules");
441 modprobe_remove_lttng_all();
444 utils_close_pipe(kernel_poll_pipe
);
445 utils_close_pipe(apps_cmd_pipe
);
448 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
449 "Matthew, BEET driven development works!%c[%dm",
450 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
455 * Send data on a unix socket using the liblttsessiondcomm API.
457 * Return lttcomm error code.
459 static int send_unix_sock(int sock
, void *buf
, size_t len
)
461 /* Check valid length */
466 return lttcomm_send_unix_sock(sock
, buf
, len
);
470 * Free memory of a command context structure.
472 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
474 DBG("Clean command context structure");
476 if ((*cmd_ctx
)->llm
) {
477 free((*cmd_ctx
)->llm
);
479 if ((*cmd_ctx
)->lsm
) {
480 free((*cmd_ctx
)->lsm
);
488 * Notify UST applications using the shm mmap futex.
490 static int notify_ust_apps(int active
)
494 DBG("Notifying applications of session daemon state: %d", active
);
496 /* See shm.c for this call implying mmap, shm and futex calls */
497 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
498 if (wait_shm_mmap
== NULL
) {
502 /* Wake waiting process */
503 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
505 /* Apps notified successfully */
513 * Setup the outgoing data buffer for the response (llm) by allocating the
514 * right amount of memory and copying the original information from the lsm
517 * Return total size of the buffer pointed by buf.
519 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
525 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
526 if (cmd_ctx
->llm
== NULL
) {
532 /* Copy common data */
533 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
534 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
536 cmd_ctx
->llm
->data_size
= size
;
537 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
546 * Update the kernel poll set of all channel fd available over all tracing
547 * session. Add the wakeup pipe at the end of the set.
549 static int update_kernel_poll(struct lttng_poll_event
*events
)
552 struct ltt_session
*session
;
553 struct ltt_kernel_channel
*channel
;
555 DBG("Updating kernel poll set");
558 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
559 session_lock(session
);
560 if (session
->kernel_session
== NULL
) {
561 session_unlock(session
);
565 cds_list_for_each_entry(channel
,
566 &session
->kernel_session
->channel_list
.head
, list
) {
567 /* Add channel fd to the kernel poll set */
568 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
570 session_unlock(session
);
573 DBG("Channel fd %d added to kernel set", channel
->fd
);
575 session_unlock(session
);
577 session_unlock_list();
582 session_unlock_list();
587 * Find the channel fd from 'fd' over all tracing session. When found, check
588 * for new channel stream and send those stream fds to the kernel consumer.
590 * Useful for CPU hotplug feature.
592 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
595 struct ltt_session
*session
;
596 struct ltt_kernel_session
*ksess
;
597 struct ltt_kernel_channel
*channel
;
599 DBG("Updating kernel streams for channel fd %d", fd
);
602 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
603 session_lock(session
);
604 if (session
->kernel_session
== NULL
) {
605 session_unlock(session
);
608 ksess
= session
->kernel_session
;
610 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
611 if (channel
->fd
== fd
) {
612 DBG("Channel found, updating kernel streams");
613 ret
= kernel_open_channel_stream(channel
);
619 * Have we already sent fds to the consumer? If yes, it means
620 * that tracing is started so it is safe to send our updated
623 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
624 struct lttng_ht_iter iter
;
625 struct consumer_socket
*socket
;
628 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
629 &iter
.iter
, socket
, node
.node
) {
630 /* Code flow error */
631 assert(socket
->fd
>= 0);
633 pthread_mutex_lock(socket
->lock
);
634 ret
= kernel_consumer_send_channel_stream(socket
->fd
,
636 pthread_mutex_unlock(socket
->lock
);
645 session_unlock(session
);
647 session_unlock_list();
651 session_unlock(session
);
652 session_unlock_list();
657 * For each tracing session, update newly registered apps.
659 static void update_ust_app(int app_sock
)
661 struct ltt_session
*sess
, *stmp
;
665 /* For all tracing session(s) */
666 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
668 if (sess
->ust_session
) {
669 ust_app_global_update(sess
->ust_session
, app_sock
);
671 session_unlock(sess
);
674 session_unlock_list();
678 * This thread manage event coming from the kernel.
680 * Features supported in this thread:
683 static void *thread_manage_kernel(void *data
)
685 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
686 uint32_t revents
, nb_fd
;
688 struct lttng_poll_event events
;
690 DBG("Thread manage kernel started");
692 testpoint(thread_manage_kernel
);
694 health_code_update(&health_thread_kernel
);
696 testpoint(thread_manage_kernel_before_loop
);
698 ret
= create_thread_poll_set(&events
, 2);
700 goto error_poll_create
;
703 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
709 health_code_update(&health_thread_kernel
);
711 if (update_poll_flag
== 1) {
713 * Reset number of fd in the poll set. Always 2 since there is the thread
714 * quit pipe and the kernel pipe.
718 ret
= update_kernel_poll(&events
);
722 update_poll_flag
= 0;
725 nb_fd
= LTTNG_POLL_GETNB(&events
);
727 DBG("Thread kernel polling on %d fds", nb_fd
);
729 /* Zeroed the poll events */
730 lttng_poll_reset(&events
);
732 /* Poll infinite value of time */
734 health_poll_update(&health_thread_kernel
);
735 ret
= lttng_poll_wait(&events
, -1);
736 health_poll_update(&health_thread_kernel
);
739 * Restart interrupted system call.
741 if (errno
== EINTR
) {
745 } else if (ret
== 0) {
746 /* Should not happen since timeout is infinite */
747 ERR("Return value of poll is 0 with an infinite timeout.\n"
748 "This should not have happened! Continuing...");
752 for (i
= 0; i
< nb_fd
; i
++) {
753 /* Fetch once the poll data */
754 revents
= LTTNG_POLL_GETEV(&events
, i
);
755 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
757 health_code_update(&health_thread_kernel
);
759 /* Thread quit pipe has been closed. Killing thread. */
760 ret
= check_thread_quit_pipe(pollfd
, revents
);
766 /* Check for data on kernel pipe */
767 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
768 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
769 update_poll_flag
= 1;
773 * New CPU detected by the kernel. Adding kernel stream to
774 * kernel session and updating the kernel consumer
776 if (revents
& LPOLLIN
) {
777 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
783 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
784 * and unregister kernel stream at this point.
793 lttng_poll_clean(&events
);
796 health_error(&health_thread_kernel
);
797 ERR("Health error occurred in %s", __func__
);
799 health_exit(&health_thread_kernel
);
800 DBG("Kernel thread dying");
805 * Signal pthread condition of the consumer data that the thread.
807 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
809 pthread_mutex_lock(&data
->cond_mutex
);
812 * The state is set before signaling. It can be any value, it's the waiter
813 * job to correctly interpret this condition variable associated to the
814 * consumer pthread_cond.
816 * A value of 0 means that the corresponding thread of the consumer data
817 * was not started. 1 indicates that the thread has started and is ready
818 * for action. A negative value means that there was an error during the
821 data
->consumer_thread_is_ready
= state
;
822 (void) pthread_cond_signal(&data
->cond
);
824 pthread_mutex_unlock(&data
->cond_mutex
);
828 * This thread manage the consumer error sent back to the session daemon.
830 static void *thread_manage_consumer(void *data
)
832 int sock
= -1, i
, ret
, pollfd
, err
= -1;
833 uint32_t revents
, nb_fd
;
834 enum lttcomm_return_code code
;
835 struct lttng_poll_event events
;
836 struct consumer_data
*consumer_data
= data
;
838 DBG("[thread] Manage consumer started");
840 health_code_update(&consumer_data
->health
);
842 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
848 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
849 * Nothing more will be added to this poll set.
851 ret
= create_thread_poll_set(&events
, 2);
856 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
861 nb_fd
= LTTNG_POLL_GETNB(&events
);
863 health_code_update(&consumer_data
->health
);
865 /* Inifinite blocking call, waiting for transmission */
867 health_poll_update(&consumer_data
->health
);
869 testpoint(thread_manage_consumer
);
871 ret
= lttng_poll_wait(&events
, -1);
872 health_poll_update(&consumer_data
->health
);
875 * Restart interrupted system call.
877 if (errno
== EINTR
) {
883 for (i
= 0; i
< nb_fd
; i
++) {
884 /* Fetch once the poll data */
885 revents
= LTTNG_POLL_GETEV(&events
, i
);
886 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
888 health_code_update(&consumer_data
->health
);
890 /* Thread quit pipe has been closed. Killing thread. */
891 ret
= check_thread_quit_pipe(pollfd
, revents
);
897 /* Event on the registration socket */
898 if (pollfd
== consumer_data
->err_sock
) {
899 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
900 ERR("consumer err socket poll error");
906 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
911 health_code_update(&consumer_data
->health
);
913 DBG2("Receiving code from consumer err_sock");
915 /* Getting status code from kconsumerd */
916 ret
= lttcomm_recv_unix_sock(sock
, &code
,
917 sizeof(enum lttcomm_return_code
));
922 health_code_update(&consumer_data
->health
);
924 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
925 consumer_data
->cmd_sock
=
926 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
927 if (consumer_data
->cmd_sock
< 0) {
928 /* On error, signal condition and quit. */
929 signal_consumer_condition(consumer_data
, -1);
930 PERROR("consumer connect");
933 signal_consumer_condition(consumer_data
, 1);
934 DBG("Consumer command socket ready");
936 ERR("consumer error when waiting for SOCK_READY : %s",
937 lttcomm_get_readable_code(-code
));
941 /* Remove the kconsumerd error sock since we've established a connexion */
942 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
947 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
952 health_code_update(&consumer_data
->health
);
954 /* Update number of fd */
955 nb_fd
= LTTNG_POLL_GETNB(&events
);
957 /* Inifinite blocking call, waiting for transmission */
959 health_poll_update(&consumer_data
->health
);
960 ret
= lttng_poll_wait(&events
, -1);
961 health_poll_update(&consumer_data
->health
);
964 * Restart interrupted system call.
966 if (errno
== EINTR
) {
972 for (i
= 0; i
< nb_fd
; i
++) {
973 /* Fetch once the poll data */
974 revents
= LTTNG_POLL_GETEV(&events
, i
);
975 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
977 health_code_update(&consumer_data
->health
);
979 /* Thread quit pipe has been closed. Killing thread. */
980 ret
= check_thread_quit_pipe(pollfd
, revents
);
986 /* Event on the kconsumerd socket */
987 if (pollfd
== sock
) {
988 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
989 ERR("consumer err socket second poll error");
995 health_code_update(&consumer_data
->health
);
997 /* Wait for any kconsumerd error */
998 ret
= lttcomm_recv_unix_sock(sock
, &code
,
999 sizeof(enum lttcomm_return_code
));
1001 ERR("consumer closed the command socket");
1005 ERR("consumer return code : %s", lttcomm_get_readable_code(-code
));
1009 /* Immediately set the consumerd state to stopped */
1010 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1011 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1012 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1013 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1014 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1016 /* Code flow error... */
1020 if (consumer_data
->err_sock
>= 0) {
1021 ret
= close(consumer_data
->err_sock
);
1026 if (consumer_data
->cmd_sock
>= 0) {
1027 ret
= close(consumer_data
->cmd_sock
);
1039 unlink(consumer_data
->err_unix_sock_path
);
1040 unlink(consumer_data
->cmd_unix_sock_path
);
1041 consumer_data
->pid
= 0;
1043 lttng_poll_clean(&events
);
1047 health_error(&consumer_data
->health
);
1048 ERR("Health error occurred in %s", __func__
);
1050 health_exit(&consumer_data
->health
);
1051 DBG("consumer thread cleanup completed");
1057 * This thread manage application communication.
1059 static void *thread_manage_apps(void *data
)
1061 int i
, ret
, pollfd
, err
= -1;
1062 uint32_t revents
, nb_fd
;
1063 struct ust_command ust_cmd
;
1064 struct lttng_poll_event events
;
1066 DBG("[thread] Manage application started");
1068 testpoint(thread_manage_apps
);
1070 rcu_register_thread();
1071 rcu_thread_online();
1073 health_code_update(&health_thread_app_manage
);
1075 ret
= create_thread_poll_set(&events
, 2);
1077 goto error_poll_create
;
1080 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1085 testpoint(thread_manage_apps_before_loop
);
1087 health_code_update(&health_thread_app_manage
);
1090 /* Zeroed the events structure */
1091 lttng_poll_reset(&events
);
1093 nb_fd
= LTTNG_POLL_GETNB(&events
);
1095 DBG("Apps thread polling on %d fds", nb_fd
);
1097 /* Inifinite blocking call, waiting for transmission */
1099 health_poll_update(&health_thread_app_manage
);
1100 ret
= lttng_poll_wait(&events
, -1);
1101 health_poll_update(&health_thread_app_manage
);
1104 * Restart interrupted system call.
1106 if (errno
== EINTR
) {
1112 for (i
= 0; i
< nb_fd
; i
++) {
1113 /* Fetch once the poll data */
1114 revents
= LTTNG_POLL_GETEV(&events
, i
);
1115 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1117 health_code_update(&health_thread_app_manage
);
1119 /* Thread quit pipe has been closed. Killing thread. */
1120 ret
= check_thread_quit_pipe(pollfd
, revents
);
1126 /* Inspect the apps cmd pipe */
1127 if (pollfd
== apps_cmd_pipe
[0]) {
1128 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1129 ERR("Apps command pipe error");
1131 } else if (revents
& LPOLLIN
) {
1133 ret
= read(apps_cmd_pipe
[0], &ust_cmd
, sizeof(ust_cmd
));
1134 if (ret
< 0 || ret
< sizeof(ust_cmd
)) {
1135 PERROR("read apps cmd pipe");
1139 health_code_update(&health_thread_app_manage
);
1141 /* Register applicaton to the session daemon */
1142 ret
= ust_app_register(&ust_cmd
.reg_msg
,
1144 if (ret
== -ENOMEM
) {
1146 } else if (ret
< 0) {
1150 health_code_update(&health_thread_app_manage
);
1153 * Validate UST version compatibility.
1155 ret
= ust_app_validate_version(ust_cmd
.sock
);
1158 * Add channel(s) and event(s) to newly registered apps
1159 * from lttng global UST domain.
1161 update_ust_app(ust_cmd
.sock
);
1164 health_code_update(&health_thread_app_manage
);
1166 ret
= ust_app_register_done(ust_cmd
.sock
);
1169 * If the registration is not possible, we simply
1170 * unregister the apps and continue
1172 ust_app_unregister(ust_cmd
.sock
);
1175 * We just need here to monitor the close of the UST
1176 * socket and poll set monitor those by default.
1177 * Listen on POLLIN (even if we never expect any
1178 * data) to ensure that hangup wakes us.
1180 ret
= lttng_poll_add(&events
, ust_cmd
.sock
, LPOLLIN
);
1185 DBG("Apps with sock %d added to poll set",
1189 health_code_update(&health_thread_app_manage
);
1195 * At this point, we know that a registered application made
1196 * the event at poll_wait.
1198 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1199 /* Removing from the poll set */
1200 ret
= lttng_poll_del(&events
, pollfd
);
1205 /* Socket closed on remote end. */
1206 ust_app_unregister(pollfd
);
1211 health_code_update(&health_thread_app_manage
);
1217 lttng_poll_clean(&events
);
1220 health_error(&health_thread_app_manage
);
1221 ERR("Health error occurred in %s", __func__
);
1223 health_exit(&health_thread_app_manage
);
1224 DBG("Application communication apps thread cleanup complete");
1225 rcu_thread_offline();
1226 rcu_unregister_thread();
1231 * Dispatch request from the registration threads to the application
1232 * communication thread.
1234 static void *thread_dispatch_ust_registration(void *data
)
1237 struct cds_wfq_node
*node
;
1238 struct ust_command
*ust_cmd
= NULL
;
1240 DBG("[thread] Dispatch UST command started");
1242 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1243 /* Atomically prepare the queue futex */
1244 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1247 /* Dequeue command for registration */
1248 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1250 DBG("Woken up but nothing in the UST command queue");
1251 /* Continue thread execution */
1255 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1257 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1258 " gid:%d sock:%d name:%s (version %d.%d)",
1259 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1260 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1261 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1262 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1264 * Inform apps thread of the new application registration. This
1265 * call is blocking so we can be assured that the data will be read
1266 * at some point in time or wait to the end of the world :)
1268 ret
= write(apps_cmd_pipe
[1], ust_cmd
,
1269 sizeof(struct ust_command
));
1271 PERROR("write apps cmd pipe");
1272 if (errno
== EBADF
) {
1274 * We can't inform the application thread to process
1275 * registration. We will exit or else application
1276 * registration will not occur and tracing will never
1283 } while (node
!= NULL
);
1285 /* Futex wait on queue. Blocking call on futex() */
1286 futex_nto1_wait(&ust_cmd_queue
.futex
);
1290 DBG("Dispatch thread dying");
1295 * This thread manage application registration.
1297 static void *thread_registration_apps(void *data
)
1299 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1300 uint32_t revents
, nb_fd
;
1301 struct lttng_poll_event events
;
1303 * Get allocated in this thread, enqueued to a global queue, dequeued and
1304 * freed in the manage apps thread.
1306 struct ust_command
*ust_cmd
= NULL
;
1308 DBG("[thread] Manage application registration started");
1310 testpoint(thread_registration_apps
);
1312 ret
= lttcomm_listen_unix_sock(apps_sock
);
1318 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1319 * more will be added to this poll set.
1321 ret
= create_thread_poll_set(&events
, 2);
1323 goto error_create_poll
;
1326 /* Add the application registration socket */
1327 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1329 goto error_poll_add
;
1332 /* Notify all applications to register */
1333 ret
= notify_ust_apps(1);
1335 ERR("Failed to notify applications or create the wait shared memory.\n"
1336 "Execution continues but there might be problem for already\n"
1337 "running applications that wishes to register.");
1341 DBG("Accepting application registration");
1343 nb_fd
= LTTNG_POLL_GETNB(&events
);
1345 /* Inifinite blocking call, waiting for transmission */
1347 health_poll_update(&health_thread_app_reg
);
1348 ret
= lttng_poll_wait(&events
, -1);
1349 health_poll_update(&health_thread_app_reg
);
1352 * Restart interrupted system call.
1354 if (errno
== EINTR
) {
1360 for (i
= 0; i
< nb_fd
; i
++) {
1361 health_code_update(&health_thread_app_reg
);
1363 /* Fetch once the poll data */
1364 revents
= LTTNG_POLL_GETEV(&events
, i
);
1365 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1367 /* Thread quit pipe has been closed. Killing thread. */
1368 ret
= check_thread_quit_pipe(pollfd
, revents
);
1374 /* Event on the registration socket */
1375 if (pollfd
== apps_sock
) {
1376 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1377 ERR("Register apps socket poll error");
1379 } else if (revents
& LPOLLIN
) {
1380 sock
= lttcomm_accept_unix_sock(apps_sock
);
1385 /* Create UST registration command for enqueuing */
1386 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1387 if (ust_cmd
== NULL
) {
1388 PERROR("ust command zmalloc");
1393 * Using message-based transmissions to ensure we don't
1394 * have to deal with partially received messages.
1396 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1398 ERR("Exhausted file descriptors allowed for applications.");
1407 health_code_update(&health_thread_app_reg
);
1408 ret
= lttcomm_recv_unix_sock(sock
, &ust_cmd
->reg_msg
,
1409 sizeof(struct ust_register_msg
));
1410 if (ret
< 0 || ret
< sizeof(struct ust_register_msg
)) {
1412 PERROR("lttcomm_recv_unix_sock register apps");
1414 ERR("Wrong size received on apps register");
1421 lttng_fd_put(LTTNG_FD_APPS
, 1);
1425 health_code_update(&health_thread_app_reg
);
1427 ust_cmd
->sock
= sock
;
1430 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1431 " gid:%d sock:%d name:%s (version %d.%d)",
1432 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1433 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1434 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1435 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1438 * Lock free enqueue the registration request. The red pill
1439 * has been taken! This apps will be part of the *system*.
1441 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1444 * Wake the registration queue futex. Implicit memory
1445 * barrier with the exchange in cds_wfq_enqueue.
1447 futex_nto1_wake(&ust_cmd_queue
.futex
);
1456 health_error(&health_thread_app_reg
);
1457 ERR("Health error occurred in %s", __func__
);
1459 health_exit(&health_thread_app_reg
);
1461 /* Notify that the registration thread is gone */
1464 if (apps_sock
>= 0) {
1465 ret
= close(apps_sock
);
1475 lttng_fd_put(LTTNG_FD_APPS
, 1);
1477 unlink(apps_unix_sock_path
);
1480 lttng_poll_clean(&events
);
1483 DBG("UST Registration thread cleanup complete");
1489 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1490 * exec or it will fails.
1492 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1495 struct timespec timeout
;
1497 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1498 consumer_data
->consumer_thread_is_ready
= 0;
1500 /* Setup pthread condition */
1501 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1504 PERROR("pthread_condattr_init consumer data");
1509 * Set the monotonic clock in order to make sure we DO NOT jump in time
1510 * between the clock_gettime() call and the timedwait call. See bug #324
1511 * for a more details and how we noticed it.
1513 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1516 PERROR("pthread_condattr_setclock consumer data");
1520 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1523 PERROR("pthread_cond_init consumer data");
1527 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1530 PERROR("pthread_create consumer");
1535 /* We are about to wait on a pthread condition */
1536 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1538 /* Get time for sem_timedwait absolute timeout */
1539 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1541 * Set the timeout for the condition timed wait even if the clock gettime
1542 * call fails since we might loop on that call and we want to avoid to
1543 * increment the timeout too many times.
1545 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1548 * The following loop COULD be skipped in some conditions so this is why we
1549 * set ret to 0 in order to make sure at least one round of the loop is
1555 * Loop until the condition is reached or when a timeout is reached. Note
1556 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1557 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1558 * possible. This loop does not take any chances and works with both of
1561 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1562 if (clock_ret
< 0) {
1563 PERROR("clock_gettime spawn consumer");
1564 /* Infinite wait for the consumerd thread to be ready */
1565 ret
= pthread_cond_wait(&consumer_data
->cond
,
1566 &consumer_data
->cond_mutex
);
1568 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1569 &consumer_data
->cond_mutex
, &timeout
);
1573 /* Release the pthread condition */
1574 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1578 if (ret
== ETIMEDOUT
) {
1580 * Call has timed out so we kill the kconsumerd_thread and return
1583 ERR("Condition timed out. The consumer thread was never ready."
1585 ret
= pthread_cancel(consumer_data
->thread
);
1587 PERROR("pthread_cancel consumer thread");
1590 PERROR("pthread_cond_wait failed consumer thread");
1595 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1596 if (consumer_data
->pid
== 0) {
1597 ERR("Consumerd did not start");
1598 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1601 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1610 * Join consumer thread
1612 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1617 /* Consumer pid must be a real one. */
1618 if (consumer_data
->pid
> 0) {
1619 ret
= kill(consumer_data
->pid
, SIGTERM
);
1621 ERR("Error killing consumer daemon");
1624 return pthread_join(consumer_data
->thread
, &status
);
1631 * Fork and exec a consumer daemon (consumerd).
1633 * Return pid if successful else -1.
1635 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1639 const char *consumer_to_use
;
1640 const char *verbosity
;
1643 DBG("Spawning consumerd");
1650 if (opt_verbose_consumer
) {
1651 verbosity
= "--verbose";
1653 verbosity
= "--quiet";
1655 switch (consumer_data
->type
) {
1656 case LTTNG_CONSUMER_KERNEL
:
1658 * Find out which consumerd to execute. We will first try the
1659 * 64-bit path, then the sessiond's installation directory, and
1660 * fallback on the 32-bit one,
1662 DBG3("Looking for a kernel consumer at these locations:");
1663 DBG3(" 1) %s", consumerd64_bin
);
1664 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1665 DBG3(" 3) %s", consumerd32_bin
);
1666 if (stat(consumerd64_bin
, &st
) == 0) {
1667 DBG3("Found location #1");
1668 consumer_to_use
= consumerd64_bin
;
1669 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1670 DBG3("Found location #2");
1671 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1672 } else if (stat(consumerd32_bin
, &st
) == 0) {
1673 DBG3("Found location #3");
1674 consumer_to_use
= consumerd32_bin
;
1676 DBG("Could not find any valid consumerd executable");
1679 DBG("Using kernel consumer at: %s", consumer_to_use
);
1680 execl(consumer_to_use
,
1681 "lttng-consumerd", verbosity
, "-k",
1682 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1683 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1686 case LTTNG_CONSUMER64_UST
:
1688 char *tmpnew
= NULL
;
1690 if (consumerd64_libdir
[0] != '\0') {
1694 tmp
= getenv("LD_LIBRARY_PATH");
1698 tmplen
= strlen("LD_LIBRARY_PATH=")
1699 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1700 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1705 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1706 strcat(tmpnew
, consumerd64_libdir
);
1707 if (tmp
[0] != '\0') {
1708 strcat(tmpnew
, ":");
1709 strcat(tmpnew
, tmp
);
1711 ret
= putenv(tmpnew
);
1717 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1718 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1719 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1720 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1722 if (consumerd64_libdir
[0] != '\0') {
1730 case LTTNG_CONSUMER32_UST
:
1732 char *tmpnew
= NULL
;
1734 if (consumerd32_libdir
[0] != '\0') {
1738 tmp
= getenv("LD_LIBRARY_PATH");
1742 tmplen
= strlen("LD_LIBRARY_PATH=")
1743 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1744 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1749 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1750 strcat(tmpnew
, consumerd32_libdir
);
1751 if (tmp
[0] != '\0') {
1752 strcat(tmpnew
, ":");
1753 strcat(tmpnew
, tmp
);
1755 ret
= putenv(tmpnew
);
1761 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
1762 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
1763 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1764 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1766 if (consumerd32_libdir
[0] != '\0') {
1775 PERROR("unknown consumer type");
1779 PERROR("kernel start consumer exec");
1782 } else if (pid
> 0) {
1785 PERROR("start consumer fork");
1793 * Spawn the consumerd daemon and session daemon thread.
1795 static int start_consumerd(struct consumer_data
*consumer_data
)
1799 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1800 if (consumer_data
->pid
!= 0) {
1801 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1805 ret
= spawn_consumerd(consumer_data
);
1807 ERR("Spawning consumerd failed");
1808 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1812 /* Setting up the consumer_data pid */
1813 consumer_data
->pid
= ret
;
1814 DBG2("Consumer pid %d", consumer_data
->pid
);
1815 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1817 DBG2("Spawning consumer control thread");
1818 ret
= spawn_consumer_thread(consumer_data
);
1820 ERR("Fatal error spawning consumer control thread");
1832 * Compute health status of each consumer. If one of them is zero (bad
1833 * state), we return 0.
1835 static int check_consumer_health(void)
1839 ret
= health_check_state(&kconsumer_data
.health
) &&
1840 health_check_state(&ustconsumer32_data
.health
) &&
1841 health_check_state(&ustconsumer64_data
.health
);
1843 DBG3("Health consumer check %d", ret
);
1849 * Setup necessary data for kernel tracer action.
1851 static int init_kernel_tracer(void)
1855 /* Modprobe lttng kernel modules */
1856 ret
= modprobe_lttng_control();
1861 /* Open debugfs lttng */
1862 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1863 if (kernel_tracer_fd
< 0) {
1864 DBG("Failed to open %s", module_proc_lttng
);
1869 /* Validate kernel version */
1870 ret
= kernel_validate_version(kernel_tracer_fd
);
1875 ret
= modprobe_lttng_data();
1880 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1884 modprobe_remove_lttng_control();
1885 ret
= close(kernel_tracer_fd
);
1889 kernel_tracer_fd
= -1;
1890 return LTTNG_ERR_KERN_VERSION
;
1893 ret
= close(kernel_tracer_fd
);
1899 modprobe_remove_lttng_control();
1902 WARN("No kernel tracer available");
1903 kernel_tracer_fd
= -1;
1905 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
1907 return LTTNG_ERR_KERN_NA
;
1913 * Copy consumer output from the tracing session to the domain session. The
1914 * function also applies the right modification on a per domain basis for the
1915 * trace files destination directory.
1917 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
1920 const char *dir_name
;
1921 struct consumer_output
*consumer
;
1924 assert(session
->consumer
);
1927 case LTTNG_DOMAIN_KERNEL
:
1928 DBG3("Copying tracing session consumer output in kernel session");
1930 * XXX: We should audit the session creation and what this function
1931 * does "extra" in order to avoid a destroy since this function is used
1932 * in the domain session creation (kernel and ust) only. Same for UST
1935 if (session
->kernel_session
->consumer
) {
1936 consumer_destroy_output(session
->kernel_session
->consumer
);
1938 session
->kernel_session
->consumer
=
1939 consumer_copy_output(session
->consumer
);
1940 /* Ease our life a bit for the next part */
1941 consumer
= session
->kernel_session
->consumer
;
1942 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
1944 case LTTNG_DOMAIN_UST
:
1945 DBG3("Copying tracing session consumer output in UST session");
1946 if (session
->ust_session
->consumer
) {
1947 consumer_destroy_output(session
->ust_session
->consumer
);
1949 session
->ust_session
->consumer
=
1950 consumer_copy_output(session
->consumer
);
1951 /* Ease our life a bit for the next part */
1952 consumer
= session
->ust_session
->consumer
;
1953 dir_name
= DEFAULT_UST_TRACE_DIR
;
1956 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
1960 /* Append correct directory to subdir */
1961 strncat(consumer
->subdir
, dir_name
,
1962 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
1963 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
1972 * Create an UST session and add it to the session ust list.
1974 static int create_ust_session(struct ltt_session
*session
,
1975 struct lttng_domain
*domain
)
1978 struct ltt_ust_session
*lus
= NULL
;
1982 assert(session
->consumer
);
1984 switch (domain
->type
) {
1985 case LTTNG_DOMAIN_UST
:
1988 ERR("Unknown UST domain on create session %d", domain
->type
);
1989 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
1993 DBG("Creating UST session");
1995 lus
= trace_ust_create_session(session
->path
, session
->id
, domain
);
1997 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2001 lus
->uid
= session
->uid
;
2002 lus
->gid
= session
->gid
;
2003 session
->ust_session
= lus
;
2005 /* Copy session output to the newly created UST session */
2006 ret
= copy_session_consumer(domain
->type
, session
);
2007 if (ret
!= LTTNG_OK
) {
2015 session
->ust_session
= NULL
;
2020 * Create a kernel tracer session then create the default channel.
2022 static int create_kernel_session(struct ltt_session
*session
)
2026 DBG("Creating kernel session");
2028 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2030 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2034 /* Code flow safety */
2035 assert(session
->kernel_session
);
2037 /* Copy session output to the newly created Kernel session */
2038 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2039 if (ret
!= LTTNG_OK
) {
2043 /* Create directory(ies) on local filesystem. */
2044 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2045 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2046 ret
= run_as_mkdir_recursive(
2047 session
->kernel_session
->consumer
->dst
.trace_path
,
2048 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2050 if (ret
!= -EEXIST
) {
2051 ERR("Trace directory creation error");
2057 session
->kernel_session
->uid
= session
->uid
;
2058 session
->kernel_session
->gid
= session
->gid
;
2063 trace_kernel_destroy_session(session
->kernel_session
);
2064 session
->kernel_session
= NULL
;
2069 * Count number of session permitted by uid/gid.
2071 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2074 struct ltt_session
*session
;
2076 DBG("Counting number of available session for UID %d GID %d",
2078 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2080 * Only list the sessions the user can control.
2082 if (!session_access_ok(session
, uid
, gid
)) {
2091 * Process the command requested by the lttng client within the command
2092 * context structure. This function make sure that the return structure (llm)
2093 * is set and ready for transmission before returning.
2095 * Return any error encountered or 0 for success.
2097 * "sock" is only used for special-case var. len data.
2099 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2103 int need_tracing_session
= 1;
2106 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2110 switch (cmd_ctx
->lsm
->cmd_type
) {
2111 case LTTNG_CREATE_SESSION
:
2112 case LTTNG_DESTROY_SESSION
:
2113 case LTTNG_LIST_SESSIONS
:
2114 case LTTNG_LIST_DOMAINS
:
2115 case LTTNG_START_TRACE
:
2116 case LTTNG_STOP_TRACE
:
2117 case LTTNG_DATA_AVAILABLE
:
2124 if (opt_no_kernel
&& need_domain
2125 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2127 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2129 ret
= LTTNG_ERR_KERN_NA
;
2134 /* Deny register consumer if we already have a spawned consumer. */
2135 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2136 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2137 if (kconsumer_data
.pid
> 0) {
2138 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2139 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2142 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2146 * Check for command that don't needs to allocate a returned payload. We do
2147 * this here so we don't have to make the call for no payload at each
2150 switch(cmd_ctx
->lsm
->cmd_type
) {
2151 case LTTNG_LIST_SESSIONS
:
2152 case LTTNG_LIST_TRACEPOINTS
:
2153 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2154 case LTTNG_LIST_DOMAINS
:
2155 case LTTNG_LIST_CHANNELS
:
2156 case LTTNG_LIST_EVENTS
:
2159 /* Setup lttng message with no payload */
2160 ret
= setup_lttng_msg(cmd_ctx
, 0);
2162 /* This label does not try to unlock the session */
2163 goto init_setup_error
;
2167 /* Commands that DO NOT need a session. */
2168 switch (cmd_ctx
->lsm
->cmd_type
) {
2169 case LTTNG_CREATE_SESSION
:
2170 case LTTNG_CALIBRATE
:
2171 case LTTNG_LIST_SESSIONS
:
2172 case LTTNG_LIST_TRACEPOINTS
:
2173 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2174 need_tracing_session
= 0;
2177 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2179 * We keep the session list lock across _all_ commands
2180 * for now, because the per-session lock does not
2181 * handle teardown properly.
2183 session_lock_list();
2184 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2185 if (cmd_ctx
->session
== NULL
) {
2186 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2187 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2189 /* If no session name specified */
2190 ret
= LTTNG_ERR_SELECT_SESS
;
2194 /* Acquire lock for the session */
2195 session_lock(cmd_ctx
->session
);
2205 * Check domain type for specific "pre-action".
2207 switch (cmd_ctx
->lsm
->domain
.type
) {
2208 case LTTNG_DOMAIN_KERNEL
:
2210 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2214 /* Kernel tracer check */
2215 if (kernel_tracer_fd
== -1) {
2216 /* Basically, load kernel tracer modules */
2217 ret
= init_kernel_tracer();
2223 /* Consumer is in an ERROR state. Report back to client */
2224 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2225 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2229 /* Need a session for kernel command */
2230 if (need_tracing_session
) {
2231 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2232 ret
= create_kernel_session(cmd_ctx
->session
);
2234 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2239 /* Start the kernel consumer daemon */
2240 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2241 if (kconsumer_data
.pid
== 0 &&
2242 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2243 cmd_ctx
->session
->start_consumer
) {
2244 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2245 ret
= start_consumerd(&kconsumer_data
);
2247 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2250 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2252 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2256 * The consumer was just spawned so we need to add the socket to
2257 * the consumer output of the session if exist.
2259 ret
= consumer_create_socket(&kconsumer_data
,
2260 cmd_ctx
->session
->kernel_session
->consumer
);
2267 case LTTNG_DOMAIN_UST
:
2269 /* Consumer is in an ERROR state. Report back to client */
2270 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2271 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2275 if (need_tracing_session
) {
2276 /* Create UST session if none exist. */
2277 if (cmd_ctx
->session
->ust_session
== NULL
) {
2278 ret
= create_ust_session(cmd_ctx
->session
,
2279 &cmd_ctx
->lsm
->domain
);
2280 if (ret
!= LTTNG_OK
) {
2285 /* Start the UST consumer daemons */
2287 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2288 if (consumerd64_bin
[0] != '\0' &&
2289 ustconsumer64_data
.pid
== 0 &&
2290 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2291 cmd_ctx
->session
->start_consumer
) {
2292 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2293 ret
= start_consumerd(&ustconsumer64_data
);
2295 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2296 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2300 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2301 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2303 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2307 * Setup socket for consumer 64 bit. No need for atomic access
2308 * since it was set above and can ONLY be set in this thread.
2310 ret
= consumer_create_socket(&ustconsumer64_data
,
2311 cmd_ctx
->session
->ust_session
->consumer
);
2317 if (consumerd32_bin
[0] != '\0' &&
2318 ustconsumer32_data
.pid
== 0 &&
2319 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2320 cmd_ctx
->session
->start_consumer
) {
2321 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2322 ret
= start_consumerd(&ustconsumer32_data
);
2324 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2325 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2329 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2330 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2332 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2336 * Setup socket for consumer 64 bit. No need for atomic access
2337 * since it was set above and can ONLY be set in this thread.
2339 ret
= consumer_create_socket(&ustconsumer32_data
,
2340 cmd_ctx
->session
->ust_session
->consumer
);
2352 /* Validate consumer daemon state when start/stop trace command */
2353 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2354 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2355 switch (cmd_ctx
->lsm
->domain
.type
) {
2356 case LTTNG_DOMAIN_UST
:
2357 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2358 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2362 case LTTNG_DOMAIN_KERNEL
:
2363 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2364 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2372 * Check that the UID or GID match that of the tracing session.
2373 * The root user can interact with all sessions.
2375 if (need_tracing_session
) {
2376 if (!session_access_ok(cmd_ctx
->session
,
2377 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2378 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2379 ret
= LTTNG_ERR_EPERM
;
2384 /* Process by command type */
2385 switch (cmd_ctx
->lsm
->cmd_type
) {
2386 case LTTNG_ADD_CONTEXT
:
2388 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2389 cmd_ctx
->lsm
->u
.context
.channel_name
,
2390 cmd_ctx
->lsm
->u
.context
.event_name
,
2391 &cmd_ctx
->lsm
->u
.context
.ctx
);
2394 case LTTNG_DISABLE_CHANNEL
:
2396 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2397 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2400 case LTTNG_DISABLE_EVENT
:
2402 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2403 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2404 cmd_ctx
->lsm
->u
.disable
.name
);
2407 case LTTNG_DISABLE_ALL_EVENT
:
2409 DBG("Disabling all events");
2411 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2412 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2415 case LTTNG_DISABLE_CONSUMER
:
2417 ret
= cmd_disable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2420 case LTTNG_ENABLE_CHANNEL
:
2422 ret
= cmd_enable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2423 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2426 case LTTNG_ENABLE_CONSUMER
:
2429 * XXX: 0 means that this URI should be applied on the session. Should
2430 * be a DOMAIN enuam.
2432 ret
= cmd_enable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2433 if (ret
!= LTTNG_OK
) {
2437 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2438 /* Add the URI for the UST session if a consumer is present. */
2439 if (cmd_ctx
->session
->ust_session
&&
2440 cmd_ctx
->session
->ust_session
->consumer
) {
2441 ret
= cmd_enable_consumer(LTTNG_DOMAIN_UST
, cmd_ctx
->session
);
2442 } else if (cmd_ctx
->session
->kernel_session
&&
2443 cmd_ctx
->session
->kernel_session
->consumer
) {
2444 ret
= cmd_enable_consumer(LTTNG_DOMAIN_KERNEL
,
2450 case LTTNG_ENABLE_EVENT
:
2452 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2453 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2454 &cmd_ctx
->lsm
->u
.enable
.event
, kernel_poll_pipe
[1]);
2457 case LTTNG_ENABLE_ALL_EVENT
:
2459 DBG("Enabling all events");
2461 ret
= cmd_enable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2462 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2463 cmd_ctx
->lsm
->u
.enable
.event
.type
, kernel_poll_pipe
[1]);
2466 case LTTNG_LIST_TRACEPOINTS
:
2468 struct lttng_event
*events
;
2471 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2472 if (nb_events
< 0) {
2473 /* Return value is a negative lttng_error_code. */
2479 * Setup lttng message with payload size set to the event list size in
2480 * bytes and then copy list into the llm payload.
2482 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2488 /* Copy event list into message payload */
2489 memcpy(cmd_ctx
->llm
->payload
, events
,
2490 sizeof(struct lttng_event
) * nb_events
);
2497 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2499 struct lttng_event_field
*fields
;
2502 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2504 if (nb_fields
< 0) {
2505 /* Return value is a negative lttng_error_code. */
2511 * Setup lttng message with payload size set to the event list size in
2512 * bytes and then copy list into the llm payload.
2514 ret
= setup_lttng_msg(cmd_ctx
,
2515 sizeof(struct lttng_event_field
) * nb_fields
);
2521 /* Copy event list into message payload */
2522 memcpy(cmd_ctx
->llm
->payload
, fields
,
2523 sizeof(struct lttng_event_field
) * nb_fields
);
2530 case LTTNG_SET_CONSUMER_URI
:
2533 struct lttng_uri
*uris
;
2535 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2536 len
= nb_uri
* sizeof(struct lttng_uri
);
2539 ret
= LTTNG_ERR_INVALID
;
2543 uris
= zmalloc(len
);
2545 ret
= LTTNG_ERR_FATAL
;
2549 /* Receive variable len data */
2550 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2551 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2553 DBG("No URIs received from client... continuing");
2555 ret
= LTTNG_ERR_SESSION_FAIL
;
2559 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2561 if (ret
!= LTTNG_OK
) {
2566 * XXX: 0 means that this URI should be applied on the session. Should
2567 * be a DOMAIN enuam.
2569 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2570 /* Add the URI for the UST session if a consumer is present. */
2571 if (cmd_ctx
->session
->ust_session
&&
2572 cmd_ctx
->session
->ust_session
->consumer
) {
2573 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2575 } else if (cmd_ctx
->session
->kernel_session
&&
2576 cmd_ctx
->session
->kernel_session
->consumer
) {
2577 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2578 cmd_ctx
->session
, nb_uri
, uris
);
2584 case LTTNG_START_TRACE
:
2586 ret
= cmd_start_trace(cmd_ctx
->session
);
2589 case LTTNG_STOP_TRACE
:
2591 ret
= cmd_stop_trace(cmd_ctx
->session
);
2594 case LTTNG_CREATE_SESSION
:
2597 struct lttng_uri
*uris
= NULL
;
2599 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2600 len
= nb_uri
* sizeof(struct lttng_uri
);
2603 uris
= zmalloc(len
);
2605 ret
= LTTNG_ERR_FATAL
;
2609 /* Receive variable len data */
2610 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2611 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2613 DBG("No URIs received from client... continuing");
2615 ret
= LTTNG_ERR_SESSION_FAIL
;
2619 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2620 DBG("Creating session with ONE network URI is a bad call");
2621 ret
= LTTNG_ERR_SESSION_FAIL
;
2626 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2631 case LTTNG_DESTROY_SESSION
:
2633 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2635 /* Set session to NULL so we do not unlock it after free. */
2636 cmd_ctx
->session
= NULL
;
2639 case LTTNG_LIST_DOMAINS
:
2642 struct lttng_domain
*domains
;
2644 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2646 /* Return value is a negative lttng_error_code. */
2651 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2656 /* Copy event list into message payload */
2657 memcpy(cmd_ctx
->llm
->payload
, domains
,
2658 nb_dom
* sizeof(struct lttng_domain
));
2665 case LTTNG_LIST_CHANNELS
:
2668 struct lttng_channel
*channels
;
2670 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2671 cmd_ctx
->session
, &channels
);
2673 /* Return value is a negative lttng_error_code. */
2678 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2683 /* Copy event list into message payload */
2684 memcpy(cmd_ctx
->llm
->payload
, channels
,
2685 nb_chan
* sizeof(struct lttng_channel
));
2692 case LTTNG_LIST_EVENTS
:
2695 struct lttng_event
*events
= NULL
;
2697 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2698 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2700 /* Return value is a negative lttng_error_code. */
2705 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2710 /* Copy event list into message payload */
2711 memcpy(cmd_ctx
->llm
->payload
, events
,
2712 nb_event
* sizeof(struct lttng_event
));
2719 case LTTNG_LIST_SESSIONS
:
2721 unsigned int nr_sessions
;
2723 session_lock_list();
2724 nr_sessions
= lttng_sessions_count(
2725 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2726 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2728 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2730 session_unlock_list();
2734 /* Filled the session array */
2735 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2736 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2737 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2739 session_unlock_list();
2744 case LTTNG_CALIBRATE
:
2746 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
2747 &cmd_ctx
->lsm
->u
.calibrate
);
2750 case LTTNG_REGISTER_CONSUMER
:
2752 struct consumer_data
*cdata
;
2754 switch (cmd_ctx
->lsm
->domain
.type
) {
2755 case LTTNG_DOMAIN_KERNEL
:
2756 cdata
= &kconsumer_data
;
2759 ret
= LTTNG_ERR_UND
;
2763 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2764 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
2767 case LTTNG_SET_FILTER
:
2769 struct lttng_filter_bytecode
*bytecode
;
2771 if (cmd_ctx
->lsm
->u
.filter
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
2772 ret
= LTTNG_ERR_FILTER_INVAL
;
2775 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.filter
.bytecode_len
);
2777 ret
= LTTNG_ERR_FILTER_NOMEM
;
2780 /* Receive var. len. data */
2781 DBG("Receiving var len data from client ...");
2782 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
2783 cmd_ctx
->lsm
->u
.filter
.bytecode_len
);
2785 DBG("Nothing recv() from client var len data... continuing");
2787 ret
= LTTNG_ERR_FILTER_INVAL
;
2791 if (bytecode
->len
+ sizeof(*bytecode
)
2792 != cmd_ctx
->lsm
->u
.filter
.bytecode_len
) {
2794 ret
= LTTNG_ERR_FILTER_INVAL
;
2798 ret
= cmd_set_filter(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2799 cmd_ctx
->lsm
->u
.filter
.channel_name
,
2800 cmd_ctx
->lsm
->u
.filter
.event_name
,
2804 case LTTNG_DATA_AVAILABLE
:
2806 ret
= cmd_data_available(cmd_ctx
->session
);
2810 ret
= LTTNG_ERR_UND
;
2815 if (cmd_ctx
->llm
== NULL
) {
2816 DBG("Missing llm structure. Allocating one.");
2817 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
2821 /* Set return code */
2822 cmd_ctx
->llm
->ret_code
= ret
;
2824 if (cmd_ctx
->session
) {
2825 session_unlock(cmd_ctx
->session
);
2827 if (need_tracing_session
) {
2828 session_unlock_list();
2835 * Thread managing health check socket.
2837 static void *thread_manage_health(void *data
)
2839 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
2840 uint32_t revents
, nb_fd
;
2841 struct lttng_poll_event events
;
2842 struct lttcomm_health_msg msg
;
2843 struct lttcomm_health_data reply
;
2845 DBG("[thread] Manage health check started");
2847 rcu_register_thread();
2849 /* Create unix socket */
2850 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
2852 ERR("Unable to create health check Unix socket");
2857 ret
= lttcomm_listen_unix_sock(sock
);
2863 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2864 * more will be added to this poll set.
2866 ret
= create_thread_poll_set(&events
, 2);
2871 /* Add the application registration socket */
2872 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
2878 DBG("Health check ready");
2880 nb_fd
= LTTNG_POLL_GETNB(&events
);
2882 /* Inifinite blocking call, waiting for transmission */
2884 ret
= lttng_poll_wait(&events
, -1);
2887 * Restart interrupted system call.
2889 if (errno
== EINTR
) {
2895 for (i
= 0; i
< nb_fd
; i
++) {
2896 /* Fetch once the poll data */
2897 revents
= LTTNG_POLL_GETEV(&events
, i
);
2898 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2900 /* Thread quit pipe has been closed. Killing thread. */
2901 ret
= check_thread_quit_pipe(pollfd
, revents
);
2907 /* Event on the registration socket */
2908 if (pollfd
== sock
) {
2909 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2910 ERR("Health socket poll error");
2916 new_sock
= lttcomm_accept_unix_sock(sock
);
2921 DBG("Receiving data from client for health...");
2922 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
2924 DBG("Nothing recv() from client... continuing");
2925 ret
= close(new_sock
);
2933 rcu_thread_online();
2935 switch (msg
.component
) {
2936 case LTTNG_HEALTH_CMD
:
2937 reply
.ret_code
= health_check_state(&health_thread_cmd
);
2939 case LTTNG_HEALTH_APP_MANAGE
:
2940 reply
.ret_code
= health_check_state(&health_thread_app_manage
);
2942 case LTTNG_HEALTH_APP_REG
:
2943 reply
.ret_code
= health_check_state(&health_thread_app_reg
);
2945 case LTTNG_HEALTH_KERNEL
:
2946 reply
.ret_code
= health_check_state(&health_thread_kernel
);
2948 case LTTNG_HEALTH_CONSUMER
:
2949 reply
.ret_code
= check_consumer_health();
2951 case LTTNG_HEALTH_ALL
:
2953 health_check_state(&health_thread_app_manage
) &&
2954 health_check_state(&health_thread_app_reg
) &&
2955 health_check_state(&health_thread_cmd
) &&
2956 health_check_state(&health_thread_kernel
) &&
2957 check_consumer_health();
2960 reply
.ret_code
= LTTNG_ERR_UND
;
2965 * Flip ret value since 0 is a success and 1 indicates a bad health for
2966 * the client where in the sessiond it is the opposite. Again, this is
2967 * just to make things easier for us poor developer which enjoy a lot
2970 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
2971 reply
.ret_code
= !reply
.ret_code
;
2974 DBG2("Health check return value %d", reply
.ret_code
);
2976 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
2978 ERR("Failed to send health data back to client");
2981 /* End of transmission */
2982 ret
= close(new_sock
);
2992 ERR("Health error occurred in %s", __func__
);
2994 DBG("Health check thread dying");
2995 unlink(health_unix_sock_path
);
3002 if (new_sock
>= 0) {
3003 ret
= close(new_sock
);
3009 lttng_poll_clean(&events
);
3011 rcu_unregister_thread();
3016 * This thread manage all clients request using the unix client socket for
3019 static void *thread_manage_clients(void *data
)
3021 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3023 uint32_t revents
, nb_fd
;
3024 struct command_ctx
*cmd_ctx
= NULL
;
3025 struct lttng_poll_event events
;
3027 DBG("[thread] Manage client started");
3029 testpoint(thread_manage_clients
);
3031 rcu_register_thread();
3033 health_code_update(&health_thread_cmd
);
3035 ret
= lttcomm_listen_unix_sock(client_sock
);
3041 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3042 * more will be added to this poll set.
3044 ret
= create_thread_poll_set(&events
, 2);
3049 /* Add the application registration socket */
3050 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3056 * Notify parent pid that we are ready to accept command for client side.
3058 if (opt_sig_parent
) {
3059 kill(ppid
, SIGUSR1
);
3062 testpoint(thread_manage_clients_before_loop
);
3064 health_code_update(&health_thread_cmd
);
3067 DBG("Accepting client command ...");
3069 nb_fd
= LTTNG_POLL_GETNB(&events
);
3071 /* Inifinite blocking call, waiting for transmission */
3073 health_poll_update(&health_thread_cmd
);
3074 ret
= lttng_poll_wait(&events
, -1);
3075 health_poll_update(&health_thread_cmd
);
3078 * Restart interrupted system call.
3080 if (errno
== EINTR
) {
3086 for (i
= 0; i
< nb_fd
; i
++) {
3087 /* Fetch once the poll data */
3088 revents
= LTTNG_POLL_GETEV(&events
, i
);
3089 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3091 health_code_update(&health_thread_cmd
);
3093 /* Thread quit pipe has been closed. Killing thread. */
3094 ret
= check_thread_quit_pipe(pollfd
, revents
);
3100 /* Event on the registration socket */
3101 if (pollfd
== client_sock
) {
3102 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3103 ERR("Client socket poll error");
3109 DBG("Wait for client response");
3111 health_code_update(&health_thread_cmd
);
3113 sock
= lttcomm_accept_unix_sock(client_sock
);
3118 /* Set socket option for credentials retrieval */
3119 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3124 /* Allocate context command to process the client request */
3125 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3126 if (cmd_ctx
== NULL
) {
3127 PERROR("zmalloc cmd_ctx");
3131 /* Allocate data buffer for reception */
3132 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3133 if (cmd_ctx
->lsm
== NULL
) {
3134 PERROR("zmalloc cmd_ctx->lsm");
3138 cmd_ctx
->llm
= NULL
;
3139 cmd_ctx
->session
= NULL
;
3141 health_code_update(&health_thread_cmd
);
3144 * Data is received from the lttng client. The struct
3145 * lttcomm_session_msg (lsm) contains the command and data request of
3148 DBG("Receiving data from client ...");
3149 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3150 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3152 DBG("Nothing recv() from client... continuing");
3158 clean_command_ctx(&cmd_ctx
);
3162 health_code_update(&health_thread_cmd
);
3164 // TODO: Validate cmd_ctx including sanity check for
3165 // security purpose.
3167 rcu_thread_online();
3169 * This function dispatch the work to the kernel or userspace tracer
3170 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3171 * informations for the client. The command context struct contains
3172 * everything this function may needs.
3174 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3175 rcu_thread_offline();
3185 * TODO: Inform client somehow of the fatal error. At
3186 * this point, ret < 0 means that a zmalloc failed
3187 * (ENOMEM). Error detected but still accept
3188 * command, unless a socket error has been
3191 clean_command_ctx(&cmd_ctx
);
3195 health_code_update(&health_thread_cmd
);
3197 DBG("Sending response (size: %d, retcode: %s)",
3198 cmd_ctx
->lttng_msg_size
,
3199 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3200 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3202 ERR("Failed to send data back to client");
3205 /* End of transmission */
3212 clean_command_ctx(&cmd_ctx
);
3214 health_code_update(&health_thread_cmd
);
3220 health_error(&health_thread_cmd
);
3221 ERR("Health error occurred in %s", __func__
);
3223 health_exit(&health_thread_cmd
);
3225 DBG("Client thread dying");
3226 unlink(client_unix_sock_path
);
3227 if (client_sock
>= 0) {
3228 ret
= close(client_sock
);
3240 lttng_poll_clean(&events
);
3241 clean_command_ctx(&cmd_ctx
);
3243 rcu_unregister_thread();
3249 * usage function on stderr
3251 static void usage(void)
3253 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3254 fprintf(stderr
, " -h, --help Display this usage.\n");
3255 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3256 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3257 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3258 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3259 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3260 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3261 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3262 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3263 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3264 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3265 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3266 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3267 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3268 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3269 fprintf(stderr
, " -V, --version Show version number.\n");
3270 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3271 fprintf(stderr
, " -q, --quiet No output at all.\n");
3272 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3273 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3274 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3278 * daemon argument parsing
3280 static int parse_args(int argc
, char **argv
)
3284 static struct option long_options
[] = {
3285 { "client-sock", 1, 0, 'c' },
3286 { "apps-sock", 1, 0, 'a' },
3287 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3288 { "kconsumerd-err-sock", 1, 0, 'E' },
3289 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3290 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3291 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3292 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3293 { "consumerd32-path", 1, 0, 'u' },
3294 { "consumerd32-libdir", 1, 0, 'U' },
3295 { "consumerd64-path", 1, 0, 't' },
3296 { "consumerd64-libdir", 1, 0, 'T' },
3297 { "daemonize", 0, 0, 'd' },
3298 { "sig-parent", 0, 0, 'S' },
3299 { "help", 0, 0, 'h' },
3300 { "group", 1, 0, 'g' },
3301 { "version", 0, 0, 'V' },
3302 { "quiet", 0, 0, 'q' },
3303 { "verbose", 0, 0, 'v' },
3304 { "verbose-consumer", 0, 0, 'Z' },
3305 { "no-kernel", 0, 0, 'N' },
3310 int option_index
= 0;
3311 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3312 long_options
, &option_index
);
3319 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3321 fprintf(stderr
, " with arg %s\n", optarg
);
3325 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3328 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3334 opt_tracing_group
= optarg
;
3340 fprintf(stdout
, "%s\n", VERSION
);
3346 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3349 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3352 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3355 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3358 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3361 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3367 lttng_opt_quiet
= 1;
3370 /* Verbose level can increase using multiple -v */
3371 lttng_opt_verbose
+= 1;
3374 opt_verbose_consumer
+= 1;
3377 consumerd32_bin
= optarg
;
3380 consumerd32_libdir
= optarg
;
3383 consumerd64_bin
= optarg
;
3386 consumerd64_libdir
= optarg
;
3389 /* Unknown option or other error.
3390 * Error is printed by getopt, just return */
3399 * Creates the two needed socket by the daemon.
3400 * apps_sock - The communication socket for all UST apps.
3401 * client_sock - The communication of the cli tool (lttng).
3403 static int init_daemon_socket(void)
3408 old_umask
= umask(0);
3410 /* Create client tool unix socket */
3411 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3412 if (client_sock
< 0) {
3413 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3418 /* File permission MUST be 660 */
3419 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3421 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3426 /* Create the application unix socket */
3427 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3428 if (apps_sock
< 0) {
3429 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3434 /* File permission MUST be 666 */
3435 ret
= chmod(apps_unix_sock_path
,
3436 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3438 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3449 * Check if the global socket is available, and if a daemon is answering at the
3450 * other side. If yes, error is returned.
3452 static int check_existing_daemon(void)
3454 /* Is there anybody out there ? */
3455 if (lttng_session_daemon_alive()) {
3463 * Set the tracing group gid onto the client socket.
3465 * Race window between mkdir and chown is OK because we are going from more
3466 * permissive (root.root) to less permissive (root.tracing).
3468 static int set_permissions(char *rundir
)
3473 ret
= allowed_group();
3475 WARN("No tracing group detected");
3482 /* Set lttng run dir */
3483 ret
= chown(rundir
, 0, gid
);
3485 ERR("Unable to set group on %s", rundir
);
3489 /* Ensure tracing group can search the run dir */
3490 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3492 ERR("Unable to set permissions on %s", rundir
);
3496 /* lttng client socket path */
3497 ret
= chown(client_unix_sock_path
, 0, gid
);
3499 ERR("Unable to set group on %s", client_unix_sock_path
);
3503 /* kconsumer error socket path */
3504 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3506 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3510 /* 64-bit ustconsumer error socket path */
3511 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3513 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3517 /* 32-bit ustconsumer compat32 error socket path */
3518 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3520 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3524 DBG("All permissions are set");
3531 * Create the lttng run directory needed for all global sockets and pipe.
3533 static int create_lttng_rundir(const char *rundir
)
3537 DBG3("Creating LTTng run directory: %s", rundir
);
3539 ret
= mkdir(rundir
, S_IRWXU
);
3541 if (errno
!= EEXIST
) {
3542 ERR("Unable to create %s", rundir
);
3554 * Setup sockets and directory needed by the kconsumerd communication with the
3557 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3561 char path
[PATH_MAX
];
3563 switch (consumer_data
->type
) {
3564 case LTTNG_CONSUMER_KERNEL
:
3565 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3567 case LTTNG_CONSUMER64_UST
:
3568 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3570 case LTTNG_CONSUMER32_UST
:
3571 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3574 ERR("Consumer type unknown");
3579 DBG2("Creating consumer directory: %s", path
);
3581 ret
= mkdir(path
, S_IRWXU
);
3583 if (errno
!= EEXIST
) {
3585 ERR("Failed to create %s", path
);
3591 /* Create the kconsumerd error unix socket */
3592 consumer_data
->err_sock
=
3593 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3594 if (consumer_data
->err_sock
< 0) {
3595 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3600 /* File permission MUST be 660 */
3601 ret
= chmod(consumer_data
->err_unix_sock_path
,
3602 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3604 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3614 * Signal handler for the daemon
3616 * Simply stop all worker threads, leaving main() return gracefully after
3617 * joining all threads and calling cleanup().
3619 static void sighandler(int sig
)
3623 DBG("SIGPIPE caught");
3626 DBG("SIGINT caught");
3630 DBG("SIGTERM caught");
3639 * Setup signal handler for :
3640 * SIGINT, SIGTERM, SIGPIPE
3642 static int set_signal_handler(void)
3645 struct sigaction sa
;
3648 if ((ret
= sigemptyset(&sigset
)) < 0) {
3649 PERROR("sigemptyset");
3653 sa
.sa_handler
= sighandler
;
3654 sa
.sa_mask
= sigset
;
3656 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3657 PERROR("sigaction");
3661 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3662 PERROR("sigaction");
3666 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3667 PERROR("sigaction");
3671 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3677 * Set open files limit to unlimited. This daemon can open a large number of
3678 * file descriptors in order to consumer multiple kernel traces.
3680 static void set_ulimit(void)
3685 /* The kernel does not allowed an infinite limit for open files */
3686 lim
.rlim_cur
= 65535;
3687 lim
.rlim_max
= 65535;
3689 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
3691 PERROR("failed to set open files limit");
3698 int main(int argc
, char **argv
)
3702 const char *home_path
;
3704 init_kernel_workarounds();
3706 rcu_register_thread();
3708 setup_consumerd_path();
3710 /* Parse arguments */
3712 if ((ret
= parse_args(argc
, argv
) < 0)) {
3722 * child: setsid, close FD 0, 1, 2, chdir /
3723 * parent: exit (if fork is successful)
3731 * We are in the child. Make sure all other file
3732 * descriptors are closed, in case we are called with
3733 * more opened file descriptors than the standard ones.
3735 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3740 /* Create thread quit pipe */
3741 if ((ret
= init_thread_quit_pipe()) < 0) {
3745 /* Check if daemon is UID = 0 */
3746 is_root
= !getuid();
3749 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
3751 /* Create global run dir with root access */
3752 ret
= create_lttng_rundir(rundir
);
3757 if (strlen(apps_unix_sock_path
) == 0) {
3758 snprintf(apps_unix_sock_path
, PATH_MAX
,
3759 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
3762 if (strlen(client_unix_sock_path
) == 0) {
3763 snprintf(client_unix_sock_path
, PATH_MAX
,
3764 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
3767 /* Set global SHM for ust */
3768 if (strlen(wait_shm_path
) == 0) {
3769 snprintf(wait_shm_path
, PATH_MAX
,
3770 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
3773 if (strlen(health_unix_sock_path
) == 0) {
3774 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3775 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
3778 /* Setup kernel consumerd path */
3779 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
3780 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
3781 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
3782 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
3784 DBG2("Kernel consumer err path: %s",
3785 kconsumer_data
.err_unix_sock_path
);
3786 DBG2("Kernel consumer cmd path: %s",
3787 kconsumer_data
.cmd_unix_sock_path
);
3789 home_path
= get_home_dir();
3790 if (home_path
== NULL
) {
3791 /* TODO: Add --socket PATH option */
3792 ERR("Can't get HOME directory for sockets creation.");
3798 * Create rundir from home path. This will create something like
3801 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
3807 ret
= create_lttng_rundir(rundir
);
3812 if (strlen(apps_unix_sock_path
) == 0) {
3813 snprintf(apps_unix_sock_path
, PATH_MAX
,
3814 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
3817 /* Set the cli tool unix socket path */
3818 if (strlen(client_unix_sock_path
) == 0) {
3819 snprintf(client_unix_sock_path
, PATH_MAX
,
3820 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
3823 /* Set global SHM for ust */
3824 if (strlen(wait_shm_path
) == 0) {
3825 snprintf(wait_shm_path
, PATH_MAX
,
3826 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, geteuid());
3829 /* Set health check Unix path */
3830 if (strlen(health_unix_sock_path
) == 0) {
3831 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3832 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
3836 /* Set consumer initial state */
3837 kernel_consumerd_state
= CONSUMER_STOPPED
;
3838 ust_consumerd_state
= CONSUMER_STOPPED
;
3840 DBG("Client socket path %s", client_unix_sock_path
);
3841 DBG("Application socket path %s", apps_unix_sock_path
);
3842 DBG("LTTng run directory path: %s", rundir
);
3844 /* 32 bits consumerd path setup */
3845 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
3846 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
3847 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
3848 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
3850 DBG2("UST consumer 32 bits err path: %s",
3851 ustconsumer32_data
.err_unix_sock_path
);
3852 DBG2("UST consumer 32 bits cmd path: %s",
3853 ustconsumer32_data
.cmd_unix_sock_path
);
3855 /* 64 bits consumerd path setup */
3856 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
3857 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
3858 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
3859 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
3861 DBG2("UST consumer 64 bits err path: %s",
3862 ustconsumer64_data
.err_unix_sock_path
);
3863 DBG2("UST consumer 64 bits cmd path: %s",
3864 ustconsumer64_data
.cmd_unix_sock_path
);
3867 * See if daemon already exist.
3869 if ((ret
= check_existing_daemon()) < 0) {
3870 ERR("Already running daemon.\n");
3872 * We do not goto exit because we must not cleanup()
3873 * because a daemon is already running.
3879 * Init UST app hash table. Alloc hash table before this point since
3880 * cleanup() can get called after that point.
3884 /* After this point, we can safely call cleanup() with "goto exit" */
3887 * These actions must be executed as root. We do that *after* setting up
3888 * the sockets path because we MUST make the check for another daemon using
3889 * those paths *before* trying to set the kernel consumer sockets and init
3893 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
3898 /* Setup kernel tracer */
3899 if (!opt_no_kernel
) {
3900 init_kernel_tracer();
3903 /* Set ulimit for open files */
3906 /* init lttng_fd tracking must be done after set_ulimit. */
3909 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
3914 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
3919 if ((ret
= set_signal_handler()) < 0) {
3923 /* Setup the needed unix socket */
3924 if ((ret
= init_daemon_socket()) < 0) {
3928 /* Set credentials to socket */
3929 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
3933 /* Get parent pid if -S, --sig-parent is specified. */
3934 if (opt_sig_parent
) {
3938 /* Setup the kernel pipe for waking up the kernel thread */
3939 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
3943 /* Setup the thread apps communication pipe. */
3944 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
3948 /* Init UST command queue. */
3949 cds_wfq_init(&ust_cmd_queue
.queue
);
3952 * Get session list pointer. This pointer MUST NOT be free(). This list is
3953 * statically declared in session.c
3955 session_list_ptr
= session_get_list();
3957 /* Set up max poll set size */
3958 lttng_poll_set_max_size();
3962 /* Init all health thread counters. */
3963 health_init(&health_thread_cmd
);
3964 health_init(&health_thread_kernel
);
3965 health_init(&health_thread_app_manage
);
3966 health_init(&health_thread_app_reg
);
3969 * Init health counters of the consumer thread. We do a quick hack here to
3970 * the state of the consumer health is fine even if the thread is not
3971 * started. This is simply to ease our life and has no cost what so ever.
3973 health_init(&kconsumer_data
.health
);
3974 health_poll_update(&kconsumer_data
.health
);
3975 health_init(&ustconsumer32_data
.health
);
3976 health_poll_update(&ustconsumer32_data
.health
);
3977 health_init(&ustconsumer64_data
.health
);
3978 health_poll_update(&ustconsumer64_data
.health
);
3980 /* Create thread to manage the client socket */
3981 ret
= pthread_create(&health_thread
, NULL
,
3982 thread_manage_health
, (void *) NULL
);
3984 PERROR("pthread_create health");
3988 /* Create thread to manage the client socket */
3989 ret
= pthread_create(&client_thread
, NULL
,
3990 thread_manage_clients
, (void *) NULL
);
3992 PERROR("pthread_create clients");
3996 /* Create thread to dispatch registration */
3997 ret
= pthread_create(&dispatch_thread
, NULL
,
3998 thread_dispatch_ust_registration
, (void *) NULL
);
4000 PERROR("pthread_create dispatch");
4004 /* Create thread to manage application registration. */
4005 ret
= pthread_create(®_apps_thread
, NULL
,
4006 thread_registration_apps
, (void *) NULL
);
4008 PERROR("pthread_create registration");
4012 /* Create thread to manage application socket */
4013 ret
= pthread_create(&apps_thread
, NULL
,
4014 thread_manage_apps
, (void *) NULL
);
4016 PERROR("pthread_create apps");
4020 /* Create kernel thread to manage kernel event */
4021 ret
= pthread_create(&kernel_thread
, NULL
,
4022 thread_manage_kernel
, (void *) NULL
);
4024 PERROR("pthread_create kernel");
4028 ret
= pthread_join(kernel_thread
, &status
);
4030 PERROR("pthread_join");
4031 goto error
; /* join error, exit without cleanup */
4035 ret
= pthread_join(apps_thread
, &status
);
4037 PERROR("pthread_join");
4038 goto error
; /* join error, exit without cleanup */
4042 ret
= pthread_join(reg_apps_thread
, &status
);
4044 PERROR("pthread_join");
4045 goto error
; /* join error, exit without cleanup */
4049 ret
= pthread_join(dispatch_thread
, &status
);
4051 PERROR("pthread_join");
4052 goto error
; /* join error, exit without cleanup */
4056 ret
= pthread_join(client_thread
, &status
);
4058 PERROR("pthread_join");
4059 goto error
; /* join error, exit without cleanup */
4062 ret
= join_consumer_thread(&kconsumer_data
);
4064 PERROR("join_consumer");
4065 goto error
; /* join error, exit without cleanup */
4072 * cleanup() is called when no other thread is running.
4074 rcu_thread_online();
4076 rcu_thread_offline();
4077 rcu_unregister_thread();