2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
33 #include <sys/types.h>
35 #include <urcu/uatomic.h>
39 #include <common/common.h>
40 #include <common/compat/poll.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
48 #include "lttng-sessiond.h"
55 #include "kernel-consumer.h"
59 #include "ust-consumer.h"
63 #include "testpoint.h"
65 #define CONSUMERD_FILE "lttng-consumerd"
68 const char default_home_dir
[] = DEFAULT_HOME_DIR
;
69 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
70 const char default_ust_sock_dir
[] = DEFAULT_UST_SOCK_DIR
;
71 const char default_global_apps_pipe
[] = DEFAULT_GLOBAL_APPS_PIPE
;
74 const char *opt_tracing_group
;
75 static int opt_sig_parent
;
76 static int opt_verbose_consumer
;
77 static int opt_daemon
;
78 static int opt_no_kernel
;
79 static int is_root
; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid
; /* Parent PID for --sig-parent option */
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
87 static struct consumer_data kconsumer_data
= {
88 .type
= LTTNG_CONSUMER_KERNEL
,
89 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
90 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
93 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
94 .lock
= PTHREAD_MUTEX_INITIALIZER
,
95 .cond
= PTHREAD_COND_INITIALIZER
,
96 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
98 static struct consumer_data ustconsumer64_data
= {
99 .type
= LTTNG_CONSUMER64_UST
,
100 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
101 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
104 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
105 .lock
= PTHREAD_MUTEX_INITIALIZER
,
106 .cond
= PTHREAD_COND_INITIALIZER
,
107 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
109 static struct consumer_data ustconsumer32_data
= {
110 .type
= LTTNG_CONSUMER32_UST
,
111 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
112 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
115 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
116 .lock
= PTHREAD_MUTEX_INITIALIZER
,
117 .cond
= PTHREAD_COND_INITIALIZER
,
118 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
121 /* Shared between threads */
122 static int dispatch_thread_exit
;
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path
[PATH_MAX
];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path
[PATH_MAX
];
128 /* global wait shm path for UST */
129 static char wait_shm_path
[PATH_MAX
];
130 /* Global health check unix path */
131 static char health_unix_sock_path
[PATH_MAX
];
133 /* Sockets and FDs */
134 static int client_sock
= -1;
135 static int apps_sock
= -1;
136 int kernel_tracer_fd
= -1;
137 static int kernel_poll_pipe
[2] = { -1, -1 };
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
143 static int thread_quit_pipe
[2] = { -1, -1 };
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
149 static int apps_cmd_pipe
[2] = { -1, -1 };
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread
;
153 static pthread_t reg_apps_thread
;
154 static pthread_t client_thread
;
155 static pthread_t kernel_thread
;
156 static pthread_t dispatch_thread
;
157 static pthread_t health_thread
;
160 * UST registration command queue. This queue is tied with a futex and uses a N
161 * wakers / 1 waiter implemented and detailed in futex.c/.h
163 * The thread_manage_apps and thread_dispatch_ust_registration interact with
164 * this queue and the wait/wake scheme.
166 static struct ust_cmd_queue ust_cmd_queue
;
169 * Pointer initialized before thread creation.
171 * This points to the tracing session list containing the session count and a
172 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
173 * MUST NOT be taken if you call a public function in session.c.
175 * The lock is nested inside the structure: session_list_ptr->lock. Please use
176 * session_lock_list and session_unlock_list for lock acquisition.
178 static struct ltt_session_list
*session_list_ptr
;
180 int ust_consumerd64_fd
= -1;
181 int ust_consumerd32_fd
= -1;
183 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
184 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
185 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
186 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
188 static const char *module_proc_lttng
= "/proc/lttng";
191 * Consumer daemon state which is changed when spawning it, killing it or in
192 * case of a fatal error.
194 enum consumerd_state
{
195 CONSUMER_STARTED
= 1,
196 CONSUMER_STOPPED
= 2,
201 * This consumer daemon state is used to validate if a client command will be
202 * able to reach the consumer. If not, the client is informed. For instance,
203 * doing a "lttng start" when the consumer state is set to ERROR will return an
204 * error to the client.
206 * The following example shows a possible race condition of this scheme:
208 * consumer thread error happens
210 * client cmd checks state -> still OK
211 * consumer thread exit, sets error
212 * client cmd try to talk to consumer
215 * However, since the consumer is a different daemon, we have no way of making
216 * sure the command will reach it safely even with this state flag. This is why
217 * we consider that up to the state validation during command processing, the
218 * command is safe. After that, we can not guarantee the correctness of the
219 * client request vis-a-vis the consumer.
221 static enum consumerd_state ust_consumerd_state
;
222 static enum consumerd_state kernel_consumerd_state
;
224 /* Used for the health monitoring of the session daemon. See health.h */
225 struct health_state health_thread_cmd
;
226 struct health_state health_thread_app_manage
;
227 struct health_state health_thread_app_reg
;
228 struct health_state health_thread_kernel
;
231 * Socket timeout for receiving and sending in seconds.
233 static int app_socket_timeout
;
236 void setup_consumerd_path(void)
238 const char *bin
, *libdir
;
241 * Allow INSTALL_BIN_PATH to be used as a target path for the
242 * native architecture size consumer if CONFIG_CONSUMER*_PATH
243 * has not been defined.
245 #if (CAA_BITS_PER_LONG == 32)
246 if (!consumerd32_bin
[0]) {
247 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
249 if (!consumerd32_libdir
[0]) {
250 consumerd32_libdir
= INSTALL_LIB_PATH
;
252 #elif (CAA_BITS_PER_LONG == 64)
253 if (!consumerd64_bin
[0]) {
254 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
256 if (!consumerd64_libdir
[0]) {
257 consumerd64_libdir
= INSTALL_LIB_PATH
;
260 #error "Unknown bitness"
264 * runtime env. var. overrides the build default.
266 bin
= getenv("LTTNG_CONSUMERD32_BIN");
268 consumerd32_bin
= bin
;
270 bin
= getenv("LTTNG_CONSUMERD64_BIN");
272 consumerd64_bin
= bin
;
274 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
276 consumerd32_libdir
= libdir
;
278 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
280 consumerd64_libdir
= libdir
;
285 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
287 static int create_thread_poll_set(struct lttng_poll_event
*events
,
292 if (events
== NULL
|| size
== 0) {
297 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
303 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
315 * Check if the thread quit pipe was triggered.
317 * Return 1 if it was triggered else 0;
319 static int check_thread_quit_pipe(int fd
, uint32_t events
)
321 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
329 * Return group ID of the tracing group or -1 if not found.
331 static gid_t
allowed_group(void)
335 if (opt_tracing_group
) {
336 grp
= getgrnam(opt_tracing_group
);
338 grp
= getgrnam(default_tracing_group
);
348 * Init thread quit pipe.
350 * Return -1 on error or 0 if all pipes are created.
352 static int init_thread_quit_pipe(void)
356 ret
= pipe(thread_quit_pipe
);
358 PERROR("thread quit pipe");
362 for (i
= 0; i
< 2; i
++) {
363 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
375 * Stop all threads by closing the thread quit pipe.
377 static void stop_threads(void)
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
385 ERR("write error on thread quit pipe");
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
390 futex_nto1_wake(&ust_cmd_queue
.futex
);
396 static void cleanup(void)
400 struct ltt_session
*sess
, *stmp
;
404 /* First thing first, stop all threads */
405 utils_close_pipe(thread_quit_pipe
);
407 DBG("Removing %s directory", rundir
);
408 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
410 ERR("asprintf failed. Something is really wrong!");
413 /* Remove lttng run directory */
416 ERR("Unable to clean %s", rundir
);
421 DBG("Cleaning up all sessions");
423 /* Destroy session list mutex */
424 if (session_list_ptr
!= NULL
) {
425 pthread_mutex_destroy(&session_list_ptr
->lock
);
427 /* Cleanup ALL session */
428 cds_list_for_each_entry_safe(sess
, stmp
,
429 &session_list_ptr
->head
, list
) {
430 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
434 DBG("Closing all UST sockets");
435 ust_app_clean_list();
437 if (is_root
&& !opt_no_kernel
) {
438 DBG2("Closing kernel fd");
439 if (kernel_tracer_fd
>= 0) {
440 ret
= close(kernel_tracer_fd
);
445 DBG("Unloading kernel modules");
446 modprobe_remove_lttng_all();
450 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
451 "Matthew, BEET driven development works!%c[%dm",
452 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
457 * Send data on a unix socket using the liblttsessiondcomm API.
459 * Return lttcomm error code.
461 static int send_unix_sock(int sock
, void *buf
, size_t len
)
463 /* Check valid length */
468 return lttcomm_send_unix_sock(sock
, buf
, len
);
472 * Free memory of a command context structure.
474 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
476 DBG("Clean command context structure");
478 if ((*cmd_ctx
)->llm
) {
479 free((*cmd_ctx
)->llm
);
481 if ((*cmd_ctx
)->lsm
) {
482 free((*cmd_ctx
)->lsm
);
490 * Notify UST applications using the shm mmap futex.
492 static int notify_ust_apps(int active
)
496 DBG("Notifying applications of session daemon state: %d", active
);
498 /* See shm.c for this call implying mmap, shm and futex calls */
499 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
500 if (wait_shm_mmap
== NULL
) {
504 /* Wake waiting process */
505 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
507 /* Apps notified successfully */
515 * Setup the outgoing data buffer for the response (llm) by allocating the
516 * right amount of memory and copying the original information from the lsm
519 * Return total size of the buffer pointed by buf.
521 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
527 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
528 if (cmd_ctx
->llm
== NULL
) {
534 /* Copy common data */
535 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
536 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
538 cmd_ctx
->llm
->data_size
= size
;
539 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
548 * Update the kernel poll set of all channel fd available over all tracing
549 * session. Add the wakeup pipe at the end of the set.
551 static int update_kernel_poll(struct lttng_poll_event
*events
)
554 struct ltt_session
*session
;
555 struct ltt_kernel_channel
*channel
;
557 DBG("Updating kernel poll set");
560 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
561 session_lock(session
);
562 if (session
->kernel_session
== NULL
) {
563 session_unlock(session
);
567 cds_list_for_each_entry(channel
,
568 &session
->kernel_session
->channel_list
.head
, list
) {
569 /* Add channel fd to the kernel poll set */
570 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
572 session_unlock(session
);
575 DBG("Channel fd %d added to kernel set", channel
->fd
);
577 session_unlock(session
);
579 session_unlock_list();
584 session_unlock_list();
589 * Find the channel fd from 'fd' over all tracing session. When found, check
590 * for new channel stream and send those stream fds to the kernel consumer.
592 * Useful for CPU hotplug feature.
594 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
597 struct ltt_session
*session
;
598 struct ltt_kernel_session
*ksess
;
599 struct ltt_kernel_channel
*channel
;
601 DBG("Updating kernel streams for channel fd %d", fd
);
604 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
605 session_lock(session
);
606 if (session
->kernel_session
== NULL
) {
607 session_unlock(session
);
610 ksess
= session
->kernel_session
;
612 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
613 if (channel
->fd
== fd
) {
614 DBG("Channel found, updating kernel streams");
615 ret
= kernel_open_channel_stream(channel
);
621 * Have we already sent fds to the consumer? If yes, it means
622 * that tracing is started so it is safe to send our updated
625 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
626 struct lttng_ht_iter iter
;
627 struct consumer_socket
*socket
;
630 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
631 &iter
.iter
, socket
, node
.node
) {
632 /* Code flow error */
633 assert(socket
->fd
>= 0);
635 pthread_mutex_lock(socket
->lock
);
636 ret
= kernel_consumer_send_channel_stream(socket
,
638 pthread_mutex_unlock(socket
->lock
);
649 session_unlock(session
);
651 session_unlock_list();
655 session_unlock(session
);
656 session_unlock_list();
661 * For each tracing session, update newly registered apps.
663 static void update_ust_app(int app_sock
)
665 struct ltt_session
*sess
, *stmp
;
669 /* For all tracing session(s) */
670 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
672 if (sess
->ust_session
) {
673 ust_app_global_update(sess
->ust_session
, app_sock
);
675 session_unlock(sess
);
678 session_unlock_list();
682 * This thread manage event coming from the kernel.
684 * Features supported in this thread:
687 static void *thread_manage_kernel(void *data
)
689 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
690 uint32_t revents
, nb_fd
;
692 struct lttng_poll_event events
;
694 DBG("[thread] Thread manage kernel started");
696 health_register(HEALTH_TYPE_KERNEL
);
699 * This first step of the while is to clean this structure which could free
700 * non NULL pointers so zero it before the loop.
702 memset(&events
, 0, sizeof(events
));
704 if (testpoint(thread_manage_kernel
)) {
705 goto error_testpoint
;
708 health_code_update(&health_thread_kernel
);
710 if (testpoint(thread_manage_kernel_before_loop
)) {
711 goto error_testpoint
;
715 health_code_update(&health_thread_kernel
);
717 if (update_poll_flag
== 1) {
718 /* Clean events object. We are about to populate it again. */
719 lttng_poll_clean(&events
);
721 ret
= create_thread_poll_set(&events
, 2);
723 goto error_poll_create
;
726 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
731 /* This will add the available kernel channel if any. */
732 ret
= update_kernel_poll(&events
);
736 update_poll_flag
= 0;
739 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
741 /* Poll infinite value of time */
743 health_poll_update(&health_thread_kernel
);
744 ret
= lttng_poll_wait(&events
, -1);
745 health_poll_update(&health_thread_kernel
);
748 * Restart interrupted system call.
750 if (errno
== EINTR
) {
754 } else if (ret
== 0) {
755 /* Should not happen since timeout is infinite */
756 ERR("Return value of poll is 0 with an infinite timeout.\n"
757 "This should not have happened! Continuing...");
763 for (i
= 0; i
< nb_fd
; i
++) {
764 /* Fetch once the poll data */
765 revents
= LTTNG_POLL_GETEV(&events
, i
);
766 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
768 health_code_update(&health_thread_kernel
);
770 /* Thread quit pipe has been closed. Killing thread. */
771 ret
= check_thread_quit_pipe(pollfd
, revents
);
777 /* Check for data on kernel pipe */
778 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
780 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
781 } while (ret
< 0 && errno
== EINTR
);
783 * Ret value is useless here, if this pipe gets any actions an
784 * update is required anyway.
786 update_poll_flag
= 1;
790 * New CPU detected by the kernel. Adding kernel stream to
791 * kernel session and updating the kernel consumer
793 if (revents
& LPOLLIN
) {
794 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
800 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
801 * and unregister kernel stream at this point.
810 lttng_poll_clean(&events
);
813 utils_close_pipe(kernel_poll_pipe
);
814 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
816 health_error(&health_thread_kernel
);
817 ERR("Health error occurred in %s", __func__
);
818 WARN("Kernel thread died unexpectedly. "
819 "Kernel tracing can continue but CPU hotplug is disabled.");
822 DBG("Kernel thread dying");
827 * Signal pthread condition of the consumer data that the thread.
829 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
831 pthread_mutex_lock(&data
->cond_mutex
);
834 * The state is set before signaling. It can be any value, it's the waiter
835 * job to correctly interpret this condition variable associated to the
836 * consumer pthread_cond.
838 * A value of 0 means that the corresponding thread of the consumer data
839 * was not started. 1 indicates that the thread has started and is ready
840 * for action. A negative value means that there was an error during the
843 data
->consumer_thread_is_ready
= state
;
844 (void) pthread_cond_signal(&data
->cond
);
846 pthread_mutex_unlock(&data
->cond_mutex
);
850 * This thread manage the consumer error sent back to the session daemon.
852 static void *thread_manage_consumer(void *data
)
854 int sock
= -1, i
, ret
, pollfd
, err
= -1;
855 uint32_t revents
, nb_fd
;
856 enum lttcomm_return_code code
;
857 struct lttng_poll_event events
;
858 struct consumer_data
*consumer_data
= data
;
860 DBG("[thread] Manage consumer started");
862 health_register(HEALTH_TYPE_CONSUMER
);
865 * Since the consumer thread can be spawned at any moment in time, we init
866 * the health to a poll status (1, which is a valid health over time).
867 * When the thread starts, we update here the health to a "code" path being
868 * an even value so this thread, when reaching a poll wait, does not
869 * trigger an error with an even value.
871 * Here is the use case we avoid.
873 * +1: the first poll update during initialization (main())
874 * +2 * x: multiple code update once in this thread.
875 * +1: poll wait in this thread (being a good health state).
876 * == even number which after the wait period shows as a bad health.
878 * In a nutshell, the following poll update to the health state brings back
879 * the state to an even value meaning a code path.
881 health_poll_update(&consumer_data
->health
);
884 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
885 * Nothing more will be added to this poll set.
887 ret
= create_thread_poll_set(&events
, 2);
893 * The error socket here is already in a listening state which was done
894 * just before spawning this thread to avoid a race between the consumer
895 * daemon exec trying to connect and the listen() call.
897 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
902 health_code_update(&consumer_data
->health
);
904 /* Inifinite blocking call, waiting for transmission */
906 health_poll_update(&consumer_data
->health
);
908 if (testpoint(thread_manage_consumer
)) {
912 ret
= lttng_poll_wait(&events
, -1);
913 health_poll_update(&consumer_data
->health
);
916 * Restart interrupted system call.
918 if (errno
== EINTR
) {
926 for (i
= 0; i
< nb_fd
; i
++) {
927 /* Fetch once the poll data */
928 revents
= LTTNG_POLL_GETEV(&events
, i
);
929 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
931 health_code_update(&consumer_data
->health
);
933 /* Thread quit pipe has been closed. Killing thread. */
934 ret
= check_thread_quit_pipe(pollfd
, revents
);
940 /* Event on the registration socket */
941 if (pollfd
== consumer_data
->err_sock
) {
942 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
943 ERR("consumer err socket poll error");
949 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
955 * Set the CLOEXEC flag. Return code is useless because either way, the
958 (void) utils_set_fd_cloexec(sock
);
960 health_code_update(&consumer_data
->health
);
962 DBG2("Receiving code from consumer err_sock");
964 /* Getting status code from kconsumerd */
965 ret
= lttcomm_recv_unix_sock(sock
, &code
,
966 sizeof(enum lttcomm_return_code
));
971 health_code_update(&consumer_data
->health
);
973 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
974 consumer_data
->cmd_sock
=
975 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
976 if (consumer_data
->cmd_sock
< 0) {
977 /* On error, signal condition and quit. */
978 signal_consumer_condition(consumer_data
, -1);
979 PERROR("consumer connect");
982 signal_consumer_condition(consumer_data
, 1);
983 DBG("Consumer command socket ready");
985 ERR("consumer error when waiting for SOCK_READY : %s",
986 lttcomm_get_readable_code(-code
));
990 /* Remove the kconsumerd error sock since we've established a connexion */
991 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
996 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1001 health_code_update(&consumer_data
->health
);
1003 /* Inifinite blocking call, waiting for transmission */
1005 health_poll_update(&consumer_data
->health
);
1006 ret
= lttng_poll_wait(&events
, -1);
1007 health_poll_update(&consumer_data
->health
);
1010 * Restart interrupted system call.
1012 if (errno
== EINTR
) {
1020 for (i
= 0; i
< nb_fd
; i
++) {
1021 /* Fetch once the poll data */
1022 revents
= LTTNG_POLL_GETEV(&events
, i
);
1023 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1025 health_code_update(&consumer_data
->health
);
1027 /* Thread quit pipe has been closed. Killing thread. */
1028 ret
= check_thread_quit_pipe(pollfd
, revents
);
1034 /* Event on the kconsumerd socket */
1035 if (pollfd
== sock
) {
1036 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1037 ERR("consumer err socket second poll error");
1043 health_code_update(&consumer_data
->health
);
1045 /* Wait for any kconsumerd error */
1046 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1047 sizeof(enum lttcomm_return_code
));
1049 ERR("consumer closed the command socket");
1053 ERR("consumer return code : %s", lttcomm_get_readable_code(-code
));
1057 /* Immediately set the consumerd state to stopped */
1058 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1059 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1060 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1061 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1062 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1064 /* Code flow error... */
1068 if (consumer_data
->err_sock
>= 0) {
1069 ret
= close(consumer_data
->err_sock
);
1074 if (consumer_data
->cmd_sock
>= 0) {
1075 ret
= close(consumer_data
->cmd_sock
);
1087 unlink(consumer_data
->err_unix_sock_path
);
1088 unlink(consumer_data
->cmd_unix_sock_path
);
1089 consumer_data
->pid
= 0;
1091 lttng_poll_clean(&events
);
1094 health_error(&consumer_data
->health
);
1095 ERR("Health error occurred in %s", __func__
);
1097 health_unregister();
1098 DBG("consumer thread cleanup completed");
1104 * This thread manage application communication.
1106 static void *thread_manage_apps(void *data
)
1108 int i
, ret
, pollfd
, err
= -1;
1109 uint32_t revents
, nb_fd
;
1110 struct ust_command ust_cmd
;
1111 struct lttng_poll_event events
;
1113 DBG("[thread] Manage application started");
1115 rcu_register_thread();
1116 rcu_thread_online();
1118 health_register(HEALTH_TYPE_APP_MANAGE
);
1120 if (testpoint(thread_manage_apps
)) {
1121 goto error_testpoint
;
1124 health_code_update(&health_thread_app_manage
);
1126 ret
= create_thread_poll_set(&events
, 2);
1128 goto error_poll_create
;
1131 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1136 if (testpoint(thread_manage_apps_before_loop
)) {
1140 health_code_update(&health_thread_app_manage
);
1143 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1145 /* Inifinite blocking call, waiting for transmission */
1147 health_poll_update(&health_thread_app_manage
);
1148 ret
= lttng_poll_wait(&events
, -1);
1149 health_poll_update(&health_thread_app_manage
);
1152 * Restart interrupted system call.
1154 if (errno
== EINTR
) {
1162 for (i
= 0; i
< nb_fd
; i
++) {
1163 /* Fetch once the poll data */
1164 revents
= LTTNG_POLL_GETEV(&events
, i
);
1165 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1167 health_code_update(&health_thread_app_manage
);
1169 /* Thread quit pipe has been closed. Killing thread. */
1170 ret
= check_thread_quit_pipe(pollfd
, revents
);
1176 /* Inspect the apps cmd pipe */
1177 if (pollfd
== apps_cmd_pipe
[0]) {
1178 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1179 ERR("Apps command pipe error");
1181 } else if (revents
& LPOLLIN
) {
1184 ret
= read(apps_cmd_pipe
[0], &ust_cmd
, sizeof(ust_cmd
));
1185 } while (ret
< 0 && errno
== EINTR
);
1186 if (ret
< 0 || ret
< sizeof(ust_cmd
)) {
1187 PERROR("read apps cmd pipe");
1191 health_code_update(&health_thread_app_manage
);
1193 /* Register applicaton to the session daemon */
1194 ret
= ust_app_register(&ust_cmd
.reg_msg
,
1196 if (ret
== -ENOMEM
) {
1198 } else if (ret
< 0) {
1202 health_code_update(&health_thread_app_manage
);
1205 * Validate UST version compatibility.
1207 ret
= ust_app_validate_version(ust_cmd
.sock
);
1210 * Add channel(s) and event(s) to newly registered apps
1211 * from lttng global UST domain.
1213 update_ust_app(ust_cmd
.sock
);
1216 health_code_update(&health_thread_app_manage
);
1218 ret
= ust_app_register_done(ust_cmd
.sock
);
1221 * If the registration is not possible, we simply
1222 * unregister the apps and continue
1224 ust_app_unregister(ust_cmd
.sock
);
1227 * We only monitor the error events of the socket. This
1228 * thread does not handle any incoming data from UST
1231 ret
= lttng_poll_add(&events
, ust_cmd
.sock
,
1232 LPOLLERR
& LPOLLHUP
& LPOLLRDHUP
);
1237 /* Set socket timeout for both receiving and ending */
1238 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd
.sock
,
1239 app_socket_timeout
);
1240 (void) lttcomm_setsockopt_snd_timeout(ust_cmd
.sock
,
1241 app_socket_timeout
);
1243 DBG("Apps with sock %d added to poll set",
1247 health_code_update(&health_thread_app_manage
);
1253 * At this point, we know that a registered application made
1254 * the event at poll_wait.
1256 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1257 /* Removing from the poll set */
1258 ret
= lttng_poll_del(&events
, pollfd
);
1263 /* Socket closed on remote end. */
1264 ust_app_unregister(pollfd
);
1269 health_code_update(&health_thread_app_manage
);
1275 lttng_poll_clean(&events
);
1278 utils_close_pipe(apps_cmd_pipe
);
1279 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1282 * We don't clean the UST app hash table here since already registered
1283 * applications can still be controlled so let them be until the session
1284 * daemon dies or the applications stop.
1288 health_error(&health_thread_app_manage
);
1289 ERR("Health error occurred in %s", __func__
);
1291 health_unregister();
1292 DBG("Application communication apps thread cleanup complete");
1293 rcu_thread_offline();
1294 rcu_unregister_thread();
1299 * Dispatch request from the registration threads to the application
1300 * communication thread.
1302 static void *thread_dispatch_ust_registration(void *data
)
1305 struct cds_wfq_node
*node
;
1306 struct ust_command
*ust_cmd
= NULL
;
1308 DBG("[thread] Dispatch UST command started");
1310 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1311 /* Atomically prepare the queue futex */
1312 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1315 /* Dequeue command for registration */
1316 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1318 DBG("Woken up but nothing in the UST command queue");
1319 /* Continue thread execution */
1323 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1325 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1326 " gid:%d sock:%d name:%s (version %d.%d)",
1327 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1328 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1329 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1330 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1332 * Inform apps thread of the new application registration. This
1333 * call is blocking so we can be assured that the data will be read
1334 * at some point in time or wait to the end of the world :)
1336 if (apps_cmd_pipe
[1] >= 0) {
1338 ret
= write(apps_cmd_pipe
[1], ust_cmd
,
1339 sizeof(struct ust_command
));
1340 } while (ret
< 0 && errno
== EINTR
);
1341 if (ret
< 0 || ret
!= sizeof(struct ust_command
)) {
1342 PERROR("write apps cmd pipe");
1343 if (errno
== EBADF
) {
1345 * We can't inform the application thread to process
1346 * registration. We will exit or else application
1347 * registration will not occur and tracing will never
1354 /* Application manager thread is not available. */
1355 ret
= close(ust_cmd
->sock
);
1357 PERROR("close ust_cmd sock");
1361 } while (node
!= NULL
);
1363 /* Futex wait on queue. Blocking call on futex() */
1364 futex_nto1_wait(&ust_cmd_queue
.futex
);
1368 DBG("Dispatch thread dying");
1373 * This thread manage application registration.
1375 static void *thread_registration_apps(void *data
)
1377 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1378 uint32_t revents
, nb_fd
;
1379 struct lttng_poll_event events
;
1381 * Get allocated in this thread, enqueued to a global queue, dequeued and
1382 * freed in the manage apps thread.
1384 struct ust_command
*ust_cmd
= NULL
;
1386 DBG("[thread] Manage application registration started");
1388 health_register(HEALTH_TYPE_APP_REG
);
1390 if (testpoint(thread_registration_apps
)) {
1391 goto error_testpoint
;
1394 ret
= lttcomm_listen_unix_sock(apps_sock
);
1400 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1401 * more will be added to this poll set.
1403 ret
= create_thread_poll_set(&events
, 2);
1405 goto error_create_poll
;
1408 /* Add the application registration socket */
1409 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1411 goto error_poll_add
;
1414 /* Notify all applications to register */
1415 ret
= notify_ust_apps(1);
1417 ERR("Failed to notify applications or create the wait shared memory.\n"
1418 "Execution continues but there might be problem for already\n"
1419 "running applications that wishes to register.");
1423 DBG("Accepting application registration");
1425 /* Inifinite blocking call, waiting for transmission */
1427 health_poll_update(&health_thread_app_reg
);
1428 ret
= lttng_poll_wait(&events
, -1);
1429 health_poll_update(&health_thread_app_reg
);
1432 * Restart interrupted system call.
1434 if (errno
== EINTR
) {
1442 for (i
= 0; i
< nb_fd
; i
++) {
1443 health_code_update(&health_thread_app_reg
);
1445 /* Fetch once the poll data */
1446 revents
= LTTNG_POLL_GETEV(&events
, i
);
1447 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1449 /* Thread quit pipe has been closed. Killing thread. */
1450 ret
= check_thread_quit_pipe(pollfd
, revents
);
1456 /* Event on the registration socket */
1457 if (pollfd
== apps_sock
) {
1458 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1459 ERR("Register apps socket poll error");
1461 } else if (revents
& LPOLLIN
) {
1462 sock
= lttcomm_accept_unix_sock(apps_sock
);
1468 * Set the CLOEXEC flag. Return code is useless because
1469 * either way, the show must go on.
1471 (void) utils_set_fd_cloexec(sock
);
1473 /* Create UST registration command for enqueuing */
1474 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1475 if (ust_cmd
== NULL
) {
1476 PERROR("ust command zmalloc");
1481 * Using message-based transmissions to ensure we don't
1482 * have to deal with partially received messages.
1484 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1486 ERR("Exhausted file descriptors allowed for applications.");
1495 health_code_update(&health_thread_app_reg
);
1496 ret
= lttcomm_recv_unix_sock(sock
, &ust_cmd
->reg_msg
,
1497 sizeof(struct ust_register_msg
));
1498 if (ret
< 0 || ret
< sizeof(struct ust_register_msg
)) {
1500 PERROR("lttcomm_recv_unix_sock register apps");
1502 ERR("Wrong size received on apps register");
1509 lttng_fd_put(LTTNG_FD_APPS
, 1);
1513 health_code_update(&health_thread_app_reg
);
1515 ust_cmd
->sock
= sock
;
1518 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1519 " gid:%d sock:%d name:%s (version %d.%d)",
1520 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1521 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1522 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1523 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1526 * Lock free enqueue the registration request. The red pill
1527 * has been taken! This apps will be part of the *system*.
1529 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1532 * Wake the registration queue futex. Implicit memory
1533 * barrier with the exchange in cds_wfq_enqueue.
1535 futex_nto1_wake(&ust_cmd_queue
.futex
);
1544 health_error(&health_thread_app_reg
);
1545 ERR("Health error occurred in %s", __func__
);
1548 /* Notify that the registration thread is gone */
1551 if (apps_sock
>= 0) {
1552 ret
= close(apps_sock
);
1562 lttng_fd_put(LTTNG_FD_APPS
, 1);
1564 unlink(apps_unix_sock_path
);
1567 lttng_poll_clean(&events
);
1571 DBG("UST Registration thread cleanup complete");
1572 health_unregister();
1578 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1579 * exec or it will fails.
1581 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1584 struct timespec timeout
;
1586 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1587 consumer_data
->consumer_thread_is_ready
= 0;
1589 /* Setup pthread condition */
1590 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1593 PERROR("pthread_condattr_init consumer data");
1598 * Set the monotonic clock in order to make sure we DO NOT jump in time
1599 * between the clock_gettime() call and the timedwait call. See bug #324
1600 * for a more details and how we noticed it.
1602 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1605 PERROR("pthread_condattr_setclock consumer data");
1609 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1612 PERROR("pthread_cond_init consumer data");
1616 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1619 PERROR("pthread_create consumer");
1624 /* We are about to wait on a pthread condition */
1625 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1627 /* Get time for sem_timedwait absolute timeout */
1628 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1630 * Set the timeout for the condition timed wait even if the clock gettime
1631 * call fails since we might loop on that call and we want to avoid to
1632 * increment the timeout too many times.
1634 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1637 * The following loop COULD be skipped in some conditions so this is why we
1638 * set ret to 0 in order to make sure at least one round of the loop is
1644 * Loop until the condition is reached or when a timeout is reached. Note
1645 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1646 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1647 * possible. This loop does not take any chances and works with both of
1650 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1651 if (clock_ret
< 0) {
1652 PERROR("clock_gettime spawn consumer");
1653 /* Infinite wait for the consumerd thread to be ready */
1654 ret
= pthread_cond_wait(&consumer_data
->cond
,
1655 &consumer_data
->cond_mutex
);
1657 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1658 &consumer_data
->cond_mutex
, &timeout
);
1662 /* Release the pthread condition */
1663 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1667 if (ret
== ETIMEDOUT
) {
1669 * Call has timed out so we kill the kconsumerd_thread and return
1672 ERR("Condition timed out. The consumer thread was never ready."
1674 ret
= pthread_cancel(consumer_data
->thread
);
1676 PERROR("pthread_cancel consumer thread");
1679 PERROR("pthread_cond_wait failed consumer thread");
1684 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1685 if (consumer_data
->pid
== 0) {
1686 ERR("Consumerd did not start");
1687 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1690 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1699 * Join consumer thread
1701 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1705 /* Consumer pid must be a real one. */
1706 if (consumer_data
->pid
> 0) {
1708 ret
= kill(consumer_data
->pid
, SIGTERM
);
1710 ERR("Error killing consumer daemon");
1713 return pthread_join(consumer_data
->thread
, &status
);
1720 * Fork and exec a consumer daemon (consumerd).
1722 * Return pid if successful else -1.
1724 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1728 const char *consumer_to_use
;
1729 const char *verbosity
;
1732 DBG("Spawning consumerd");
1739 if (opt_verbose_consumer
) {
1740 verbosity
= "--verbose";
1742 verbosity
= "--quiet";
1744 switch (consumer_data
->type
) {
1745 case LTTNG_CONSUMER_KERNEL
:
1747 * Find out which consumerd to execute. We will first try the
1748 * 64-bit path, then the sessiond's installation directory, and
1749 * fallback on the 32-bit one,
1751 DBG3("Looking for a kernel consumer at these locations:");
1752 DBG3(" 1) %s", consumerd64_bin
);
1753 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1754 DBG3(" 3) %s", consumerd32_bin
);
1755 if (stat(consumerd64_bin
, &st
) == 0) {
1756 DBG3("Found location #1");
1757 consumer_to_use
= consumerd64_bin
;
1758 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1759 DBG3("Found location #2");
1760 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1761 } else if (stat(consumerd32_bin
, &st
) == 0) {
1762 DBG3("Found location #3");
1763 consumer_to_use
= consumerd32_bin
;
1765 DBG("Could not find any valid consumerd executable");
1768 DBG("Using kernel consumer at: %s", consumer_to_use
);
1769 execl(consumer_to_use
,
1770 "lttng-consumerd", verbosity
, "-k",
1771 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1772 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1775 case LTTNG_CONSUMER64_UST
:
1777 char *tmpnew
= NULL
;
1779 if (consumerd64_libdir
[0] != '\0') {
1783 tmp
= getenv("LD_LIBRARY_PATH");
1787 tmplen
= strlen("LD_LIBRARY_PATH=")
1788 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1789 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1794 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1795 strcat(tmpnew
, consumerd64_libdir
);
1796 if (tmp
[0] != '\0') {
1797 strcat(tmpnew
, ":");
1798 strcat(tmpnew
, tmp
);
1800 ret
= putenv(tmpnew
);
1806 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1807 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1808 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1809 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1811 if (consumerd64_libdir
[0] != '\0') {
1819 case LTTNG_CONSUMER32_UST
:
1821 char *tmpnew
= NULL
;
1823 if (consumerd32_libdir
[0] != '\0') {
1827 tmp
= getenv("LD_LIBRARY_PATH");
1831 tmplen
= strlen("LD_LIBRARY_PATH=")
1832 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1833 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1838 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1839 strcat(tmpnew
, consumerd32_libdir
);
1840 if (tmp
[0] != '\0') {
1841 strcat(tmpnew
, ":");
1842 strcat(tmpnew
, tmp
);
1844 ret
= putenv(tmpnew
);
1850 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
1851 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
1852 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1853 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1855 if (consumerd32_libdir
[0] != '\0') {
1864 PERROR("unknown consumer type");
1868 PERROR("kernel start consumer exec");
1871 } else if (pid
> 0) {
1874 PERROR("start consumer fork");
1882 * Spawn the consumerd daemon and session daemon thread.
1884 static int start_consumerd(struct consumer_data
*consumer_data
)
1889 * Set the listen() state on the socket since there is a possible race
1890 * between the exec() of the consumer daemon and this call if place in the
1891 * consumer thread. See bug #366 for more details.
1893 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
1898 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1899 if (consumer_data
->pid
!= 0) {
1900 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1904 ret
= spawn_consumerd(consumer_data
);
1906 ERR("Spawning consumerd failed");
1907 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1911 /* Setting up the consumer_data pid */
1912 consumer_data
->pid
= ret
;
1913 DBG2("Consumer pid %d", consumer_data
->pid
);
1914 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1916 DBG2("Spawning consumer control thread");
1917 ret
= spawn_consumer_thread(consumer_data
);
1919 ERR("Fatal error spawning consumer control thread");
1927 /* Cleanup already created socket on error. */
1928 if (consumer_data
->err_sock
>= 0) {
1931 err
= close(consumer_data
->err_sock
);
1933 PERROR("close consumer data error socket");
1940 * Compute health status of each consumer. If one of them is zero (bad
1941 * state), we return 0.
1943 static int check_consumer_health(void)
1947 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
1949 DBG3("Health consumer check %d", ret
);
1955 * Setup necessary data for kernel tracer action.
1957 static int init_kernel_tracer(void)
1961 /* Modprobe lttng kernel modules */
1962 ret
= modprobe_lttng_control();
1967 /* Open debugfs lttng */
1968 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1969 if (kernel_tracer_fd
< 0) {
1970 DBG("Failed to open %s", module_proc_lttng
);
1975 /* Validate kernel version */
1976 ret
= kernel_validate_version(kernel_tracer_fd
);
1981 ret
= modprobe_lttng_data();
1986 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1990 modprobe_remove_lttng_control();
1991 ret
= close(kernel_tracer_fd
);
1995 kernel_tracer_fd
= -1;
1996 return LTTNG_ERR_KERN_VERSION
;
1999 ret
= close(kernel_tracer_fd
);
2005 modprobe_remove_lttng_control();
2008 WARN("No kernel tracer available");
2009 kernel_tracer_fd
= -1;
2011 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2013 return LTTNG_ERR_KERN_NA
;
2019 * Copy consumer output from the tracing session to the domain session. The
2020 * function also applies the right modification on a per domain basis for the
2021 * trace files destination directory.
2023 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2026 const char *dir_name
;
2027 struct consumer_output
*consumer
;
2030 assert(session
->consumer
);
2033 case LTTNG_DOMAIN_KERNEL
:
2034 DBG3("Copying tracing session consumer output in kernel session");
2036 * XXX: We should audit the session creation and what this function
2037 * does "extra" in order to avoid a destroy since this function is used
2038 * in the domain session creation (kernel and ust) only. Same for UST
2041 if (session
->kernel_session
->consumer
) {
2042 consumer_destroy_output(session
->kernel_session
->consumer
);
2044 session
->kernel_session
->consumer
=
2045 consumer_copy_output(session
->consumer
);
2046 /* Ease our life a bit for the next part */
2047 consumer
= session
->kernel_session
->consumer
;
2048 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2050 case LTTNG_DOMAIN_UST
:
2051 DBG3("Copying tracing session consumer output in UST session");
2052 if (session
->ust_session
->consumer
) {
2053 consumer_destroy_output(session
->ust_session
->consumer
);
2055 session
->ust_session
->consumer
=
2056 consumer_copy_output(session
->consumer
);
2057 /* Ease our life a bit for the next part */
2058 consumer
= session
->ust_session
->consumer
;
2059 dir_name
= DEFAULT_UST_TRACE_DIR
;
2062 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2066 /* Append correct directory to subdir */
2067 strncat(consumer
->subdir
, dir_name
,
2068 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2069 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2078 * Create an UST session and add it to the session ust list.
2080 static int create_ust_session(struct ltt_session
*session
,
2081 struct lttng_domain
*domain
)
2084 struct ltt_ust_session
*lus
= NULL
;
2088 assert(session
->consumer
);
2090 switch (domain
->type
) {
2091 case LTTNG_DOMAIN_UST
:
2094 ERR("Unknown UST domain on create session %d", domain
->type
);
2095 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2099 DBG("Creating UST session");
2101 lus
= trace_ust_create_session(session
->path
, session
->id
, domain
);
2103 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2107 lus
->uid
= session
->uid
;
2108 lus
->gid
= session
->gid
;
2109 session
->ust_session
= lus
;
2111 /* Copy session output to the newly created UST session */
2112 ret
= copy_session_consumer(domain
->type
, session
);
2113 if (ret
!= LTTNG_OK
) {
2121 session
->ust_session
= NULL
;
2126 * Create a kernel tracer session then create the default channel.
2128 static int create_kernel_session(struct ltt_session
*session
)
2132 DBG("Creating kernel session");
2134 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2136 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2140 /* Code flow safety */
2141 assert(session
->kernel_session
);
2143 /* Copy session output to the newly created Kernel session */
2144 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2145 if (ret
!= LTTNG_OK
) {
2149 /* Create directory(ies) on local filesystem. */
2150 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2151 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2152 ret
= run_as_mkdir_recursive(
2153 session
->kernel_session
->consumer
->dst
.trace_path
,
2154 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2156 if (ret
!= -EEXIST
) {
2157 ERR("Trace directory creation error");
2163 session
->kernel_session
->uid
= session
->uid
;
2164 session
->kernel_session
->gid
= session
->gid
;
2169 trace_kernel_destroy_session(session
->kernel_session
);
2170 session
->kernel_session
= NULL
;
2175 * Count number of session permitted by uid/gid.
2177 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2180 struct ltt_session
*session
;
2182 DBG("Counting number of available session for UID %d GID %d",
2184 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2186 * Only list the sessions the user can control.
2188 if (!session_access_ok(session
, uid
, gid
)) {
2197 * Process the command requested by the lttng client within the command
2198 * context structure. This function make sure that the return structure (llm)
2199 * is set and ready for transmission before returning.
2201 * Return any error encountered or 0 for success.
2203 * "sock" is only used for special-case var. len data.
2205 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2209 int need_tracing_session
= 1;
2212 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2216 switch (cmd_ctx
->lsm
->cmd_type
) {
2217 case LTTNG_CREATE_SESSION
:
2218 case LTTNG_DESTROY_SESSION
:
2219 case LTTNG_LIST_SESSIONS
:
2220 case LTTNG_LIST_DOMAINS
:
2221 case LTTNG_START_TRACE
:
2222 case LTTNG_STOP_TRACE
:
2223 case LTTNG_DATA_PENDING
:
2230 if (opt_no_kernel
&& need_domain
2231 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2233 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2235 ret
= LTTNG_ERR_KERN_NA
;
2240 /* Deny register consumer if we already have a spawned consumer. */
2241 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2242 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2243 if (kconsumer_data
.pid
> 0) {
2244 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2245 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2248 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2252 * Check for command that don't needs to allocate a returned payload. We do
2253 * this here so we don't have to make the call for no payload at each
2256 switch(cmd_ctx
->lsm
->cmd_type
) {
2257 case LTTNG_LIST_SESSIONS
:
2258 case LTTNG_LIST_TRACEPOINTS
:
2259 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2260 case LTTNG_LIST_DOMAINS
:
2261 case LTTNG_LIST_CHANNELS
:
2262 case LTTNG_LIST_EVENTS
:
2265 /* Setup lttng message with no payload */
2266 ret
= setup_lttng_msg(cmd_ctx
, 0);
2268 /* This label does not try to unlock the session */
2269 goto init_setup_error
;
2273 /* Commands that DO NOT need a session. */
2274 switch (cmd_ctx
->lsm
->cmd_type
) {
2275 case LTTNG_CREATE_SESSION
:
2276 case LTTNG_CALIBRATE
:
2277 case LTTNG_LIST_SESSIONS
:
2278 case LTTNG_LIST_TRACEPOINTS
:
2279 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2280 need_tracing_session
= 0;
2283 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2285 * We keep the session list lock across _all_ commands
2286 * for now, because the per-session lock does not
2287 * handle teardown properly.
2289 session_lock_list();
2290 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2291 if (cmd_ctx
->session
== NULL
) {
2292 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2293 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2295 /* If no session name specified */
2296 ret
= LTTNG_ERR_SELECT_SESS
;
2300 /* Acquire lock for the session */
2301 session_lock(cmd_ctx
->session
);
2311 * Check domain type for specific "pre-action".
2313 switch (cmd_ctx
->lsm
->domain
.type
) {
2314 case LTTNG_DOMAIN_KERNEL
:
2316 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2320 /* Kernel tracer check */
2321 if (kernel_tracer_fd
== -1) {
2322 /* Basically, load kernel tracer modules */
2323 ret
= init_kernel_tracer();
2329 /* Consumer is in an ERROR state. Report back to client */
2330 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2331 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2335 /* Need a session for kernel command */
2336 if (need_tracing_session
) {
2337 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2338 ret
= create_kernel_session(cmd_ctx
->session
);
2340 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2345 /* Start the kernel consumer daemon */
2346 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2347 if (kconsumer_data
.pid
== 0 &&
2348 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2349 cmd_ctx
->session
->start_consumer
) {
2350 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2351 ret
= start_consumerd(&kconsumer_data
);
2353 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2356 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2358 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2362 * The consumer was just spawned so we need to add the socket to
2363 * the consumer output of the session if exist.
2365 ret
= consumer_create_socket(&kconsumer_data
,
2366 cmd_ctx
->session
->kernel_session
->consumer
);
2373 case LTTNG_DOMAIN_UST
:
2375 /* Consumer is in an ERROR state. Report back to client */
2376 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2377 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2381 if (need_tracing_session
) {
2382 /* Create UST session if none exist. */
2383 if (cmd_ctx
->session
->ust_session
== NULL
) {
2384 ret
= create_ust_session(cmd_ctx
->session
,
2385 &cmd_ctx
->lsm
->domain
);
2386 if (ret
!= LTTNG_OK
) {
2391 /* Start the UST consumer daemons */
2393 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2394 if (consumerd64_bin
[0] != '\0' &&
2395 ustconsumer64_data
.pid
== 0 &&
2396 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2397 cmd_ctx
->session
->start_consumer
) {
2398 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2399 ret
= start_consumerd(&ustconsumer64_data
);
2401 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2402 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2406 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2407 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2409 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2413 * Setup socket for consumer 64 bit. No need for atomic access
2414 * since it was set above and can ONLY be set in this thread.
2416 ret
= consumer_create_socket(&ustconsumer64_data
,
2417 cmd_ctx
->session
->ust_session
->consumer
);
2423 if (consumerd32_bin
[0] != '\0' &&
2424 ustconsumer32_data
.pid
== 0 &&
2425 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2426 cmd_ctx
->session
->start_consumer
) {
2427 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2428 ret
= start_consumerd(&ustconsumer32_data
);
2430 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2431 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2435 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2436 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2438 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2442 * Setup socket for consumer 64 bit. No need for atomic access
2443 * since it was set above and can ONLY be set in this thread.
2445 ret
= consumer_create_socket(&ustconsumer32_data
,
2446 cmd_ctx
->session
->ust_session
->consumer
);
2458 /* Validate consumer daemon state when start/stop trace command */
2459 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2460 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2461 switch (cmd_ctx
->lsm
->domain
.type
) {
2462 case LTTNG_DOMAIN_UST
:
2463 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2464 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2468 case LTTNG_DOMAIN_KERNEL
:
2469 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2470 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2478 * Check that the UID or GID match that of the tracing session.
2479 * The root user can interact with all sessions.
2481 if (need_tracing_session
) {
2482 if (!session_access_ok(cmd_ctx
->session
,
2483 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2484 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2485 ret
= LTTNG_ERR_EPERM
;
2490 /* Process by command type */
2491 switch (cmd_ctx
->lsm
->cmd_type
) {
2492 case LTTNG_ADD_CONTEXT
:
2494 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2495 cmd_ctx
->lsm
->u
.context
.channel_name
,
2496 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2499 case LTTNG_DISABLE_CHANNEL
:
2501 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2502 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2505 case LTTNG_DISABLE_EVENT
:
2507 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2508 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2509 cmd_ctx
->lsm
->u
.disable
.name
);
2512 case LTTNG_DISABLE_ALL_EVENT
:
2514 DBG("Disabling all events");
2516 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2517 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2520 case LTTNG_DISABLE_CONSUMER
:
2522 ret
= cmd_disable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2525 case LTTNG_ENABLE_CHANNEL
:
2527 ret
= cmd_enable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2528 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2531 case LTTNG_ENABLE_CONSUMER
:
2534 * XXX: 0 means that this URI should be applied on the session. Should
2535 * be a DOMAIN enuam.
2537 ret
= cmd_enable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2538 if (ret
!= LTTNG_OK
) {
2542 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2543 /* Add the URI for the UST session if a consumer is present. */
2544 if (cmd_ctx
->session
->ust_session
&&
2545 cmd_ctx
->session
->ust_session
->consumer
) {
2546 ret
= cmd_enable_consumer(LTTNG_DOMAIN_UST
, cmd_ctx
->session
);
2547 } else if (cmd_ctx
->session
->kernel_session
&&
2548 cmd_ctx
->session
->kernel_session
->consumer
) {
2549 ret
= cmd_enable_consumer(LTTNG_DOMAIN_KERNEL
,
2555 case LTTNG_ENABLE_EVENT
:
2557 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2558 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2559 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2562 case LTTNG_ENABLE_ALL_EVENT
:
2564 DBG("Enabling all events");
2566 ret
= cmd_enable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2567 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2568 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2571 case LTTNG_LIST_TRACEPOINTS
:
2573 struct lttng_event
*events
;
2576 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2577 if (nb_events
< 0) {
2578 /* Return value is a negative lttng_error_code. */
2584 * Setup lttng message with payload size set to the event list size in
2585 * bytes and then copy list into the llm payload.
2587 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2593 /* Copy event list into message payload */
2594 memcpy(cmd_ctx
->llm
->payload
, events
,
2595 sizeof(struct lttng_event
) * nb_events
);
2602 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2604 struct lttng_event_field
*fields
;
2607 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2609 if (nb_fields
< 0) {
2610 /* Return value is a negative lttng_error_code. */
2616 * Setup lttng message with payload size set to the event list size in
2617 * bytes and then copy list into the llm payload.
2619 ret
= setup_lttng_msg(cmd_ctx
,
2620 sizeof(struct lttng_event_field
) * nb_fields
);
2626 /* Copy event list into message payload */
2627 memcpy(cmd_ctx
->llm
->payload
, fields
,
2628 sizeof(struct lttng_event_field
) * nb_fields
);
2635 case LTTNG_SET_CONSUMER_URI
:
2638 struct lttng_uri
*uris
;
2640 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2641 len
= nb_uri
* sizeof(struct lttng_uri
);
2644 ret
= LTTNG_ERR_INVALID
;
2648 uris
= zmalloc(len
);
2650 ret
= LTTNG_ERR_FATAL
;
2654 /* Receive variable len data */
2655 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2656 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2658 DBG("No URIs received from client... continuing");
2660 ret
= LTTNG_ERR_SESSION_FAIL
;
2665 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2667 if (ret
!= LTTNG_OK
) {
2673 * XXX: 0 means that this URI should be applied on the session. Should
2674 * be a DOMAIN enuam.
2676 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2677 /* Add the URI for the UST session if a consumer is present. */
2678 if (cmd_ctx
->session
->ust_session
&&
2679 cmd_ctx
->session
->ust_session
->consumer
) {
2680 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2682 } else if (cmd_ctx
->session
->kernel_session
&&
2683 cmd_ctx
->session
->kernel_session
->consumer
) {
2684 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2685 cmd_ctx
->session
, nb_uri
, uris
);
2693 case LTTNG_START_TRACE
:
2695 ret
= cmd_start_trace(cmd_ctx
->session
);
2698 case LTTNG_STOP_TRACE
:
2700 ret
= cmd_stop_trace(cmd_ctx
->session
);
2703 case LTTNG_CREATE_SESSION
:
2706 struct lttng_uri
*uris
= NULL
;
2708 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2709 len
= nb_uri
* sizeof(struct lttng_uri
);
2712 uris
= zmalloc(len
);
2714 ret
= LTTNG_ERR_FATAL
;
2718 /* Receive variable len data */
2719 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2720 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2722 DBG("No URIs received from client... continuing");
2724 ret
= LTTNG_ERR_SESSION_FAIL
;
2729 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2730 DBG("Creating session with ONE network URI is a bad call");
2731 ret
= LTTNG_ERR_SESSION_FAIL
;
2737 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2744 case LTTNG_DESTROY_SESSION
:
2746 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2748 /* Set session to NULL so we do not unlock it after free. */
2749 cmd_ctx
->session
= NULL
;
2752 case LTTNG_LIST_DOMAINS
:
2755 struct lttng_domain
*domains
;
2757 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2759 /* Return value is a negative lttng_error_code. */
2764 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2769 /* Copy event list into message payload */
2770 memcpy(cmd_ctx
->llm
->payload
, domains
,
2771 nb_dom
* sizeof(struct lttng_domain
));
2778 case LTTNG_LIST_CHANNELS
:
2781 struct lttng_channel
*channels
;
2783 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2784 cmd_ctx
->session
, &channels
);
2786 /* Return value is a negative lttng_error_code. */
2791 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2796 /* Copy event list into message payload */
2797 memcpy(cmd_ctx
->llm
->payload
, channels
,
2798 nb_chan
* sizeof(struct lttng_channel
));
2805 case LTTNG_LIST_EVENTS
:
2808 struct lttng_event
*events
= NULL
;
2810 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2811 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2813 /* Return value is a negative lttng_error_code. */
2818 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2823 /* Copy event list into message payload */
2824 memcpy(cmd_ctx
->llm
->payload
, events
,
2825 nb_event
* sizeof(struct lttng_event
));
2832 case LTTNG_LIST_SESSIONS
:
2834 unsigned int nr_sessions
;
2836 session_lock_list();
2837 nr_sessions
= lttng_sessions_count(
2838 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2839 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2841 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2843 session_unlock_list();
2847 /* Filled the session array */
2848 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2849 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2850 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2852 session_unlock_list();
2857 case LTTNG_CALIBRATE
:
2859 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
2860 &cmd_ctx
->lsm
->u
.calibrate
);
2863 case LTTNG_REGISTER_CONSUMER
:
2865 struct consumer_data
*cdata
;
2867 switch (cmd_ctx
->lsm
->domain
.type
) {
2868 case LTTNG_DOMAIN_KERNEL
:
2869 cdata
= &kconsumer_data
;
2872 ret
= LTTNG_ERR_UND
;
2876 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2877 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
2880 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
2882 struct lttng_filter_bytecode
*bytecode
;
2884 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
2885 ret
= LTTNG_ERR_FILTER_INVAL
;
2888 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
2889 ret
= LTTNG_ERR_FILTER_INVAL
;
2892 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2894 ret
= LTTNG_ERR_FILTER_NOMEM
;
2897 /* Receive var. len. data */
2898 DBG("Receiving var len data from client ...");
2899 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
2900 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2902 DBG("Nothing recv() from client var len data... continuing");
2904 ret
= LTTNG_ERR_FILTER_INVAL
;
2908 if (bytecode
->len
+ sizeof(*bytecode
)
2909 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
2911 ret
= LTTNG_ERR_FILTER_INVAL
;
2915 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2916 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2917 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
2920 case LTTNG_DATA_PENDING
:
2922 ret
= cmd_data_pending(cmd_ctx
->session
);
2926 ret
= LTTNG_ERR_UND
;
2931 if (cmd_ctx
->llm
== NULL
) {
2932 DBG("Missing llm structure. Allocating one.");
2933 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
2937 /* Set return code */
2938 cmd_ctx
->llm
->ret_code
= ret
;
2940 if (cmd_ctx
->session
) {
2941 session_unlock(cmd_ctx
->session
);
2943 if (need_tracing_session
) {
2944 session_unlock_list();
2951 * Thread managing health check socket.
2953 static void *thread_manage_health(void *data
)
2955 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
2956 uint32_t revents
, nb_fd
;
2957 struct lttng_poll_event events
;
2958 struct lttcomm_health_msg msg
;
2959 struct lttcomm_health_data reply
;
2961 DBG("[thread] Manage health check started");
2963 rcu_register_thread();
2965 /* Create unix socket */
2966 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
2968 ERR("Unable to create health check Unix socket");
2974 * Set the CLOEXEC flag. Return code is useless because either way, the
2977 (void) utils_set_fd_cloexec(sock
);
2979 ret
= lttcomm_listen_unix_sock(sock
);
2985 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2986 * more will be added to this poll set.
2988 ret
= create_thread_poll_set(&events
, 2);
2993 /* Add the application registration socket */
2994 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3000 DBG("Health check ready");
3002 /* Inifinite blocking call, waiting for transmission */
3004 ret
= lttng_poll_wait(&events
, -1);
3007 * Restart interrupted system call.
3009 if (errno
== EINTR
) {
3017 for (i
= 0; i
< nb_fd
; i
++) {
3018 /* Fetch once the poll data */
3019 revents
= LTTNG_POLL_GETEV(&events
, i
);
3020 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3022 /* Thread quit pipe has been closed. Killing thread. */
3023 ret
= check_thread_quit_pipe(pollfd
, revents
);
3029 /* Event on the registration socket */
3030 if (pollfd
== sock
) {
3031 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3032 ERR("Health socket poll error");
3038 new_sock
= lttcomm_accept_unix_sock(sock
);
3044 * Set the CLOEXEC flag. Return code is useless because either way, the
3047 (void) utils_set_fd_cloexec(new_sock
);
3049 DBG("Receiving data from client for health...");
3050 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3052 DBG("Nothing recv() from client... continuing");
3053 ret
= close(new_sock
);
3061 rcu_thread_online();
3063 switch (msg
.component
) {
3064 case LTTNG_HEALTH_CMD
:
3065 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3067 case LTTNG_HEALTH_APP_MANAGE
:
3068 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3070 case LTTNG_HEALTH_APP_REG
:
3071 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3073 case LTTNG_HEALTH_KERNEL
:
3074 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3076 case LTTNG_HEALTH_CONSUMER
:
3077 reply
.ret_code
= check_consumer_health();
3079 case LTTNG_HEALTH_ALL
:
3081 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3082 health_check_state(HEALTH_TYPE_APP_REG
) &&
3083 health_check_state(HEALTH_TYPE_CMD
) &&
3084 health_check_state(HEALTH_TYPE_KERNEL
) &&
3085 check_consumer_health();
3088 reply
.ret_code
= LTTNG_ERR_UND
;
3093 * Flip ret value since 0 is a success and 1 indicates a bad health for
3094 * the client where in the sessiond it is the opposite. Again, this is
3095 * just to make things easier for us poor developer which enjoy a lot
3098 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3099 reply
.ret_code
= !reply
.ret_code
;
3102 DBG2("Health check return value %d", reply
.ret_code
);
3104 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3106 ERR("Failed to send health data back to client");
3109 /* End of transmission */
3110 ret
= close(new_sock
);
3120 ERR("Health error occurred in %s", __func__
);
3122 DBG("Health check thread dying");
3123 unlink(health_unix_sock_path
);
3130 if (new_sock
>= 0) {
3131 ret
= close(new_sock
);
3137 lttng_poll_clean(&events
);
3139 rcu_unregister_thread();
3144 * This thread manage all clients request using the unix client socket for
3147 static void *thread_manage_clients(void *data
)
3149 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3151 uint32_t revents
, nb_fd
;
3152 struct command_ctx
*cmd_ctx
= NULL
;
3153 struct lttng_poll_event events
;
3155 DBG("[thread] Manage client started");
3157 rcu_register_thread();
3159 health_register(HEALTH_TYPE_CMD
);
3161 if (testpoint(thread_manage_clients
)) {
3162 goto error_testpoint
;
3165 health_code_update(&health_thread_cmd
);
3167 ret
= lttcomm_listen_unix_sock(client_sock
);
3173 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3174 * more will be added to this poll set.
3176 ret
= create_thread_poll_set(&events
, 2);
3178 goto error_create_poll
;
3181 /* Add the application registration socket */
3182 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3188 * Notify parent pid that we are ready to accept command for client side.
3190 if (opt_sig_parent
) {
3191 kill(ppid
, SIGUSR1
);
3194 if (testpoint(thread_manage_clients_before_loop
)) {
3198 health_code_update(&health_thread_cmd
);
3201 DBG("Accepting client command ...");
3203 /* Inifinite blocking call, waiting for transmission */
3205 health_poll_update(&health_thread_cmd
);
3206 ret
= lttng_poll_wait(&events
, -1);
3207 health_poll_update(&health_thread_cmd
);
3210 * Restart interrupted system call.
3212 if (errno
== EINTR
) {
3220 for (i
= 0; i
< nb_fd
; i
++) {
3221 /* Fetch once the poll data */
3222 revents
= LTTNG_POLL_GETEV(&events
, i
);
3223 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3225 health_code_update(&health_thread_cmd
);
3227 /* Thread quit pipe has been closed. Killing thread. */
3228 ret
= check_thread_quit_pipe(pollfd
, revents
);
3234 /* Event on the registration socket */
3235 if (pollfd
== client_sock
) {
3236 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3237 ERR("Client socket poll error");
3243 DBG("Wait for client response");
3245 health_code_update(&health_thread_cmd
);
3247 sock
= lttcomm_accept_unix_sock(client_sock
);
3253 * Set the CLOEXEC flag. Return code is useless because either way, the
3256 (void) utils_set_fd_cloexec(sock
);
3258 /* Set socket option for credentials retrieval */
3259 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3264 /* Allocate context command to process the client request */
3265 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3266 if (cmd_ctx
== NULL
) {
3267 PERROR("zmalloc cmd_ctx");
3271 /* Allocate data buffer for reception */
3272 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3273 if (cmd_ctx
->lsm
== NULL
) {
3274 PERROR("zmalloc cmd_ctx->lsm");
3278 cmd_ctx
->llm
= NULL
;
3279 cmd_ctx
->session
= NULL
;
3281 health_code_update(&health_thread_cmd
);
3284 * Data is received from the lttng client. The struct
3285 * lttcomm_session_msg (lsm) contains the command and data request of
3288 DBG("Receiving data from client ...");
3289 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3290 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3292 DBG("Nothing recv() from client... continuing");
3298 clean_command_ctx(&cmd_ctx
);
3302 health_code_update(&health_thread_cmd
);
3304 // TODO: Validate cmd_ctx including sanity check for
3305 // security purpose.
3307 rcu_thread_online();
3309 * This function dispatch the work to the kernel or userspace tracer
3310 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3311 * informations for the client. The command context struct contains
3312 * everything this function may needs.
3314 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3315 rcu_thread_offline();
3325 * TODO: Inform client somehow of the fatal error. At
3326 * this point, ret < 0 means that a zmalloc failed
3327 * (ENOMEM). Error detected but still accept
3328 * command, unless a socket error has been
3331 clean_command_ctx(&cmd_ctx
);
3335 health_code_update(&health_thread_cmd
);
3337 DBG("Sending response (size: %d, retcode: %s)",
3338 cmd_ctx
->lttng_msg_size
,
3339 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3340 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3342 ERR("Failed to send data back to client");
3345 /* End of transmission */
3352 clean_command_ctx(&cmd_ctx
);
3354 health_code_update(&health_thread_cmd
);
3366 lttng_poll_clean(&events
);
3367 clean_command_ctx(&cmd_ctx
);
3372 unlink(client_unix_sock_path
);
3373 if (client_sock
>= 0) {
3374 ret
= close(client_sock
);
3381 health_error(&health_thread_cmd
);
3382 ERR("Health error occurred in %s", __func__
);
3385 health_unregister();
3387 DBG("Client thread dying");
3389 rcu_unregister_thread();
3395 * usage function on stderr
3397 static void usage(void)
3399 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3400 fprintf(stderr
, " -h, --help Display this usage.\n");
3401 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3402 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3403 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3404 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3405 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3406 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3407 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3408 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3409 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3410 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3411 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3412 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3413 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3414 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3415 fprintf(stderr
, " -V, --version Show version number.\n");
3416 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3417 fprintf(stderr
, " -q, --quiet No output at all.\n");
3418 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3419 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3420 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3424 * daemon argument parsing
3426 static int parse_args(int argc
, char **argv
)
3430 static struct option long_options
[] = {
3431 { "client-sock", 1, 0, 'c' },
3432 { "apps-sock", 1, 0, 'a' },
3433 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3434 { "kconsumerd-err-sock", 1, 0, 'E' },
3435 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3436 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3437 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3438 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3439 { "consumerd32-path", 1, 0, 'u' },
3440 { "consumerd32-libdir", 1, 0, 'U' },
3441 { "consumerd64-path", 1, 0, 't' },
3442 { "consumerd64-libdir", 1, 0, 'T' },
3443 { "daemonize", 0, 0, 'd' },
3444 { "sig-parent", 0, 0, 'S' },
3445 { "help", 0, 0, 'h' },
3446 { "group", 1, 0, 'g' },
3447 { "version", 0, 0, 'V' },
3448 { "quiet", 0, 0, 'q' },
3449 { "verbose", 0, 0, 'v' },
3450 { "verbose-consumer", 0, 0, 'Z' },
3451 { "no-kernel", 0, 0, 'N' },
3456 int option_index
= 0;
3457 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3458 long_options
, &option_index
);
3465 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3467 fprintf(stderr
, " with arg %s\n", optarg
);
3471 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3474 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3480 opt_tracing_group
= optarg
;
3486 fprintf(stdout
, "%s\n", VERSION
);
3492 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3495 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3498 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3501 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3504 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3507 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3513 lttng_opt_quiet
= 1;
3516 /* Verbose level can increase using multiple -v */
3517 lttng_opt_verbose
+= 1;
3520 opt_verbose_consumer
+= 1;
3523 consumerd32_bin
= optarg
;
3526 consumerd32_libdir
= optarg
;
3529 consumerd64_bin
= optarg
;
3532 consumerd64_libdir
= optarg
;
3535 /* Unknown option or other error.
3536 * Error is printed by getopt, just return */
3545 * Creates the two needed socket by the daemon.
3546 * apps_sock - The communication socket for all UST apps.
3547 * client_sock - The communication of the cli tool (lttng).
3549 static int init_daemon_socket(void)
3554 old_umask
= umask(0);
3556 /* Create client tool unix socket */
3557 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3558 if (client_sock
< 0) {
3559 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3564 /* Set the cloexec flag */
3565 ret
= utils_set_fd_cloexec(client_sock
);
3567 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3568 "Continuing but note that the consumer daemon will have a "
3569 "reference to this socket on exec()", client_sock
);
3572 /* File permission MUST be 660 */
3573 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3575 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3580 /* Create the application unix socket */
3581 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3582 if (apps_sock
< 0) {
3583 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3588 /* Set the cloexec flag */
3589 ret
= utils_set_fd_cloexec(apps_sock
);
3591 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3592 "Continuing but note that the consumer daemon will have a "
3593 "reference to this socket on exec()", apps_sock
);
3596 /* File permission MUST be 666 */
3597 ret
= chmod(apps_unix_sock_path
,
3598 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3600 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3605 DBG3("Session daemon client socket %d and application socket %d created",
3606 client_sock
, apps_sock
);
3614 * Check if the global socket is available, and if a daemon is answering at the
3615 * other side. If yes, error is returned.
3617 static int check_existing_daemon(void)
3619 /* Is there anybody out there ? */
3620 if (lttng_session_daemon_alive()) {
3628 * Set the tracing group gid onto the client socket.
3630 * Race window between mkdir and chown is OK because we are going from more
3631 * permissive (root.root) to less permissive (root.tracing).
3633 static int set_permissions(char *rundir
)
3638 ret
= allowed_group();
3640 WARN("No tracing group detected");
3647 /* Set lttng run dir */
3648 ret
= chown(rundir
, 0, gid
);
3650 ERR("Unable to set group on %s", rundir
);
3654 /* Ensure tracing group can search the run dir */
3655 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3657 ERR("Unable to set permissions on %s", rundir
);
3661 /* lttng client socket path */
3662 ret
= chown(client_unix_sock_path
, 0, gid
);
3664 ERR("Unable to set group on %s", client_unix_sock_path
);
3668 /* kconsumer error socket path */
3669 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3671 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3675 /* 64-bit ustconsumer error socket path */
3676 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3678 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3682 /* 32-bit ustconsumer compat32 error socket path */
3683 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3685 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3689 DBG("All permissions are set");
3696 * Create the lttng run directory needed for all global sockets and pipe.
3698 static int create_lttng_rundir(const char *rundir
)
3702 DBG3("Creating LTTng run directory: %s", rundir
);
3704 ret
= mkdir(rundir
, S_IRWXU
);
3706 if (errno
!= EEXIST
) {
3707 ERR("Unable to create %s", rundir
);
3719 * Setup sockets and directory needed by the kconsumerd communication with the
3722 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3726 char path
[PATH_MAX
];
3728 switch (consumer_data
->type
) {
3729 case LTTNG_CONSUMER_KERNEL
:
3730 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3732 case LTTNG_CONSUMER64_UST
:
3733 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3735 case LTTNG_CONSUMER32_UST
:
3736 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3739 ERR("Consumer type unknown");
3744 DBG2("Creating consumer directory: %s", path
);
3746 ret
= mkdir(path
, S_IRWXU
);
3748 if (errno
!= EEXIST
) {
3750 ERR("Failed to create %s", path
);
3756 /* Create the kconsumerd error unix socket */
3757 consumer_data
->err_sock
=
3758 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3759 if (consumer_data
->err_sock
< 0) {
3760 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3765 /* File permission MUST be 660 */
3766 ret
= chmod(consumer_data
->err_unix_sock_path
,
3767 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3769 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3779 * Signal handler for the daemon
3781 * Simply stop all worker threads, leaving main() return gracefully after
3782 * joining all threads and calling cleanup().
3784 static void sighandler(int sig
)
3788 DBG("SIGPIPE caught");
3791 DBG("SIGINT caught");
3795 DBG("SIGTERM caught");
3804 * Setup signal handler for :
3805 * SIGINT, SIGTERM, SIGPIPE
3807 static int set_signal_handler(void)
3810 struct sigaction sa
;
3813 if ((ret
= sigemptyset(&sigset
)) < 0) {
3814 PERROR("sigemptyset");
3818 sa
.sa_handler
= sighandler
;
3819 sa
.sa_mask
= sigset
;
3821 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3822 PERROR("sigaction");
3826 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3827 PERROR("sigaction");
3831 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3832 PERROR("sigaction");
3836 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3842 * Set open files limit to unlimited. This daemon can open a large number of
3843 * file descriptors in order to consumer multiple kernel traces.
3845 static void set_ulimit(void)
3850 /* The kernel does not allowed an infinite limit for open files */
3851 lim
.rlim_cur
= 65535;
3852 lim
.rlim_max
= 65535;
3854 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
3856 PERROR("failed to set open files limit");
3863 int main(int argc
, char **argv
)
3867 const char *home_path
, *env_app_timeout
;
3869 init_kernel_workarounds();
3871 rcu_register_thread();
3873 setup_consumerd_path();
3875 /* Parse arguments */
3877 if ((ret
= parse_args(argc
, argv
)) < 0) {
3887 * child: setsid, close FD 0, 1, 2, chdir /
3888 * parent: exit (if fork is successful)
3896 * We are in the child. Make sure all other file
3897 * descriptors are closed, in case we are called with
3898 * more opened file descriptors than the standard ones.
3900 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3905 /* Create thread quit pipe */
3906 if ((ret
= init_thread_quit_pipe()) < 0) {
3910 /* Check if daemon is UID = 0 */
3911 is_root
= !getuid();
3914 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
3916 /* Create global run dir with root access */
3917 ret
= create_lttng_rundir(rundir
);
3922 if (strlen(apps_unix_sock_path
) == 0) {
3923 snprintf(apps_unix_sock_path
, PATH_MAX
,
3924 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
3927 if (strlen(client_unix_sock_path
) == 0) {
3928 snprintf(client_unix_sock_path
, PATH_MAX
,
3929 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
3932 /* Set global SHM for ust */
3933 if (strlen(wait_shm_path
) == 0) {
3934 snprintf(wait_shm_path
, PATH_MAX
,
3935 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
3938 if (strlen(health_unix_sock_path
) == 0) {
3939 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3940 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
3943 /* Setup kernel consumerd path */
3944 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
3945 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
3946 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
3947 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
3949 DBG2("Kernel consumer err path: %s",
3950 kconsumer_data
.err_unix_sock_path
);
3951 DBG2("Kernel consumer cmd path: %s",
3952 kconsumer_data
.cmd_unix_sock_path
);
3954 home_path
= get_home_dir();
3955 if (home_path
== NULL
) {
3956 /* TODO: Add --socket PATH option */
3957 ERR("Can't get HOME directory for sockets creation.");
3963 * Create rundir from home path. This will create something like
3966 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
3972 ret
= create_lttng_rundir(rundir
);
3977 if (strlen(apps_unix_sock_path
) == 0) {
3978 snprintf(apps_unix_sock_path
, PATH_MAX
,
3979 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
3982 /* Set the cli tool unix socket path */
3983 if (strlen(client_unix_sock_path
) == 0) {
3984 snprintf(client_unix_sock_path
, PATH_MAX
,
3985 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
3988 /* Set global SHM for ust */
3989 if (strlen(wait_shm_path
) == 0) {
3990 snprintf(wait_shm_path
, PATH_MAX
,
3991 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, geteuid());
3994 /* Set health check Unix path */
3995 if (strlen(health_unix_sock_path
) == 0) {
3996 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3997 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4001 /* Set consumer initial state */
4002 kernel_consumerd_state
= CONSUMER_STOPPED
;
4003 ust_consumerd_state
= CONSUMER_STOPPED
;
4005 DBG("Client socket path %s", client_unix_sock_path
);
4006 DBG("Application socket path %s", apps_unix_sock_path
);
4007 DBG("LTTng run directory path: %s", rundir
);
4009 /* 32 bits consumerd path setup */
4010 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4011 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4012 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4013 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4015 DBG2("UST consumer 32 bits err path: %s",
4016 ustconsumer32_data
.err_unix_sock_path
);
4017 DBG2("UST consumer 32 bits cmd path: %s",
4018 ustconsumer32_data
.cmd_unix_sock_path
);
4020 /* 64 bits consumerd path setup */
4021 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4022 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4023 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4024 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4026 DBG2("UST consumer 64 bits err path: %s",
4027 ustconsumer64_data
.err_unix_sock_path
);
4028 DBG2("UST consumer 64 bits cmd path: %s",
4029 ustconsumer64_data
.cmd_unix_sock_path
);
4032 * See if daemon already exist.
4034 if ((ret
= check_existing_daemon()) < 0) {
4035 ERR("Already running daemon.\n");
4037 * We do not goto exit because we must not cleanup()
4038 * because a daemon is already running.
4044 * Init UST app hash table. Alloc hash table before this point since
4045 * cleanup() can get called after that point.
4049 /* After this point, we can safely call cleanup() with "goto exit" */
4052 * These actions must be executed as root. We do that *after* setting up
4053 * the sockets path because we MUST make the check for another daemon using
4054 * those paths *before* trying to set the kernel consumer sockets and init
4058 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4063 /* Setup kernel tracer */
4064 if (!opt_no_kernel
) {
4065 init_kernel_tracer();
4068 /* Set ulimit for open files */
4071 /* init lttng_fd tracking must be done after set_ulimit. */
4074 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4079 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4084 if ((ret
= set_signal_handler()) < 0) {
4088 /* Setup the needed unix socket */
4089 if ((ret
= init_daemon_socket()) < 0) {
4093 /* Set credentials to socket */
4094 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4098 /* Get parent pid if -S, --sig-parent is specified. */
4099 if (opt_sig_parent
) {
4103 /* Setup the kernel pipe for waking up the kernel thread */
4104 if (is_root
&& !opt_no_kernel
) {
4105 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4110 /* Setup the thread apps communication pipe. */
4111 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4115 /* Init UST command queue. */
4116 cds_wfq_init(&ust_cmd_queue
.queue
);
4119 * Get session list pointer. This pointer MUST NOT be free(). This list is
4120 * statically declared in session.c
4122 session_list_ptr
= session_get_list();
4124 /* Set up max poll set size */
4125 lttng_poll_set_max_size();
4129 /* Check for the application socket timeout env variable. */
4130 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4131 if (env_app_timeout
) {
4132 app_socket_timeout
= atoi(env_app_timeout
);
4134 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4137 /* Create thread to manage the client socket */
4138 ret
= pthread_create(&health_thread
, NULL
,
4139 thread_manage_health
, (void *) NULL
);
4141 PERROR("pthread_create health");
4145 /* Create thread to manage the client socket */
4146 ret
= pthread_create(&client_thread
, NULL
,
4147 thread_manage_clients
, (void *) NULL
);
4149 PERROR("pthread_create clients");
4153 /* Create thread to dispatch registration */
4154 ret
= pthread_create(&dispatch_thread
, NULL
,
4155 thread_dispatch_ust_registration
, (void *) NULL
);
4157 PERROR("pthread_create dispatch");
4161 /* Create thread to manage application registration. */
4162 ret
= pthread_create(®_apps_thread
, NULL
,
4163 thread_registration_apps
, (void *) NULL
);
4165 PERROR("pthread_create registration");
4169 /* Create thread to manage application socket */
4170 ret
= pthread_create(&apps_thread
, NULL
,
4171 thread_manage_apps
, (void *) NULL
);
4173 PERROR("pthread_create apps");
4177 /* Don't start this thread if kernel tracing is not requested nor root */
4178 if (is_root
&& !opt_no_kernel
) {
4179 /* Create kernel thread to manage kernel event */
4180 ret
= pthread_create(&kernel_thread
, NULL
,
4181 thread_manage_kernel
, (void *) NULL
);
4183 PERROR("pthread_create kernel");
4187 ret
= pthread_join(kernel_thread
, &status
);
4189 PERROR("pthread_join");
4190 goto error
; /* join error, exit without cleanup */
4195 ret
= pthread_join(apps_thread
, &status
);
4197 PERROR("pthread_join");
4198 goto error
; /* join error, exit without cleanup */
4202 ret
= pthread_join(reg_apps_thread
, &status
);
4204 PERROR("pthread_join");
4205 goto error
; /* join error, exit without cleanup */
4209 ret
= pthread_join(dispatch_thread
, &status
);
4211 PERROR("pthread_join");
4212 goto error
; /* join error, exit without cleanup */
4216 ret
= pthread_join(client_thread
, &status
);
4218 PERROR("pthread_join");
4219 goto error
; /* join error, exit without cleanup */
4222 ret
= join_consumer_thread(&kconsumer_data
);
4224 PERROR("join_consumer");
4225 goto error
; /* join error, exit without cleanup */
4228 ret
= join_consumer_thread(&ustconsumer32_data
);
4230 PERROR("join_consumer ust32");
4231 goto error
; /* join error, exit without cleanup */
4234 ret
= join_consumer_thread(&ustconsumer64_data
);
4236 PERROR("join_consumer ust64");
4237 goto error
; /* join error, exit without cleanup */
4241 ret
= pthread_join(health_thread
, &status
);
4243 PERROR("pthread_join health thread");
4244 goto error
; /* join error, exit without cleanup */
4250 * cleanup() is called when no other thread is running.
4252 rcu_thread_online();
4254 rcu_thread_offline();
4255 rcu_unregister_thread();