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
->fd
,
638 pthread_mutex_unlock(socket
->lock
);
647 session_unlock(session
);
649 session_unlock_list();
653 session_unlock(session
);
654 session_unlock_list();
659 * For each tracing session, update newly registered apps.
661 static void update_ust_app(int app_sock
)
663 struct ltt_session
*sess
, *stmp
;
667 /* For all tracing session(s) */
668 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
670 if (sess
->ust_session
) {
671 ust_app_global_update(sess
->ust_session
, app_sock
);
673 session_unlock(sess
);
676 session_unlock_list();
680 * This thread manage event coming from the kernel.
682 * Features supported in this thread:
685 static void *thread_manage_kernel(void *data
)
687 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
688 uint32_t revents
, nb_fd
;
690 struct lttng_poll_event events
;
692 DBG("Thread manage kernel started");
694 testpoint(thread_manage_kernel
);
696 health_code_update(&health_thread_kernel
);
698 testpoint(thread_manage_kernel_before_loop
);
700 ret
= create_thread_poll_set(&events
, 2);
702 goto error_poll_create
;
705 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
711 health_code_update(&health_thread_kernel
);
713 if (update_poll_flag
== 1) {
715 * Reset number of fd in the poll set. Always 2 since there is the thread
716 * quit pipe and the kernel pipe.
720 ret
= update_kernel_poll(&events
);
724 update_poll_flag
= 0;
727 nb_fd
= LTTNG_POLL_GETNB(&events
);
729 DBG("Thread kernel polling on %d fds", nb_fd
);
731 /* Zeroed the poll events */
732 lttng_poll_reset(&events
);
734 /* Poll infinite value of time */
736 health_poll_update(&health_thread_kernel
);
737 ret
= lttng_poll_wait(&events
, -1);
738 health_poll_update(&health_thread_kernel
);
741 * Restart interrupted system call.
743 if (errno
== EINTR
) {
747 } else if (ret
== 0) {
748 /* Should not happen since timeout is infinite */
749 ERR("Return value of poll is 0 with an infinite timeout.\n"
750 "This should not have happened! Continuing...");
754 for (i
= 0; i
< nb_fd
; i
++) {
755 /* Fetch once the poll data */
756 revents
= LTTNG_POLL_GETEV(&events
, i
);
757 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
759 health_code_update(&health_thread_kernel
);
761 /* Thread quit pipe has been closed. Killing thread. */
762 ret
= check_thread_quit_pipe(pollfd
, revents
);
768 /* Check for data on kernel pipe */
769 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
770 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
771 update_poll_flag
= 1;
775 * New CPU detected by the kernel. Adding kernel stream to
776 * kernel session and updating the kernel consumer
778 if (revents
& LPOLLIN
) {
779 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
785 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
786 * and unregister kernel stream at this point.
795 lttng_poll_clean(&events
);
797 utils_close_pipe(kernel_poll_pipe
);
798 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
800 health_error(&health_thread_kernel
);
801 ERR("Health error occurred in %s", __func__
);
802 WARN("Kernel thread died unexpectedly. "
803 "Kernel tracing can continue but CPU hotplug is disabled.");
805 health_exit(&health_thread_kernel
);
806 DBG("Kernel thread dying");
811 * Signal pthread condition of the consumer data that the thread.
813 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
815 pthread_mutex_lock(&data
->cond_mutex
);
818 * The state is set before signaling. It can be any value, it's the waiter
819 * job to correctly interpret this condition variable associated to the
820 * consumer pthread_cond.
822 * A value of 0 means that the corresponding thread of the consumer data
823 * was not started. 1 indicates that the thread has started and is ready
824 * for action. A negative value means that there was an error during the
827 data
->consumer_thread_is_ready
= state
;
828 (void) pthread_cond_signal(&data
->cond
);
830 pthread_mutex_unlock(&data
->cond_mutex
);
834 * This thread manage the consumer error sent back to the session daemon.
836 static void *thread_manage_consumer(void *data
)
838 int sock
= -1, i
, ret
, pollfd
, err
= -1;
839 uint32_t revents
, nb_fd
;
840 enum lttcomm_return_code code
;
841 struct lttng_poll_event events
;
842 struct consumer_data
*consumer_data
= data
;
844 DBG("[thread] Manage consumer started");
847 * Since the consumer thread can be spawned at any moment in time, we init
848 * the health to a poll status (1, which is a valid health over time).
849 * When the thread starts, we update here the health to a "code" path being
850 * an even value so this thread, when reaching a poll wait, does not
851 * trigger an error with an even value.
853 * Here is the use case we avoid.
855 * +1: the first poll update during initialization (main())
856 * +2 * x: multiple code update once in this thread.
857 * +1: poll wait in this thread (being a good health state).
858 * == even number which after the wait period shows as a bad health.
860 * In a nutshell, the following poll update to the health state brings back
861 * the state to an even value meaning a code path.
863 health_poll_update(&consumer_data
->health
);
866 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
867 * Nothing more will be added to this poll set.
869 ret
= create_thread_poll_set(&events
, 2);
875 * The error socket here is already in a listening state which was done
876 * just before spawning this thread to avoid a race between the consumer
877 * daemon exec trying to connect and the listen() call.
879 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
884 nb_fd
= LTTNG_POLL_GETNB(&events
);
886 health_code_update(&consumer_data
->health
);
888 /* Inifinite blocking call, waiting for transmission */
890 health_poll_update(&consumer_data
->health
);
892 testpoint(thread_manage_consumer
);
894 ret
= lttng_poll_wait(&events
, -1);
895 health_poll_update(&consumer_data
->health
);
898 * Restart interrupted system call.
900 if (errno
== EINTR
) {
906 for (i
= 0; i
< nb_fd
; i
++) {
907 /* Fetch once the poll data */
908 revents
= LTTNG_POLL_GETEV(&events
, i
);
909 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
911 health_code_update(&consumer_data
->health
);
913 /* Thread quit pipe has been closed. Killing thread. */
914 ret
= check_thread_quit_pipe(pollfd
, revents
);
920 /* Event on the registration socket */
921 if (pollfd
== consumer_data
->err_sock
) {
922 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
923 ERR("consumer err socket poll error");
929 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
935 * Set the CLOEXEC flag. Return code is useless because either way, the
938 (void) utils_set_fd_cloexec(sock
);
940 health_code_update(&consumer_data
->health
);
942 DBG2("Receiving code from consumer err_sock");
944 /* Getting status code from kconsumerd */
945 ret
= lttcomm_recv_unix_sock(sock
, &code
,
946 sizeof(enum lttcomm_return_code
));
951 health_code_update(&consumer_data
->health
);
953 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
954 consumer_data
->cmd_sock
=
955 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
956 if (consumer_data
->cmd_sock
< 0) {
957 /* On error, signal condition and quit. */
958 signal_consumer_condition(consumer_data
, -1);
959 PERROR("consumer connect");
962 signal_consumer_condition(consumer_data
, 1);
963 DBG("Consumer command socket ready");
965 ERR("consumer error when waiting for SOCK_READY : %s",
966 lttcomm_get_readable_code(-code
));
970 /* Remove the kconsumerd error sock since we've established a connexion */
971 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
976 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
981 health_code_update(&consumer_data
->health
);
983 /* Update number of fd */
984 nb_fd
= LTTNG_POLL_GETNB(&events
);
986 /* Inifinite blocking call, waiting for transmission */
988 health_poll_update(&consumer_data
->health
);
989 ret
= lttng_poll_wait(&events
, -1);
990 health_poll_update(&consumer_data
->health
);
993 * Restart interrupted system call.
995 if (errno
== EINTR
) {
1001 for (i
= 0; i
< nb_fd
; i
++) {
1002 /* Fetch once the poll data */
1003 revents
= LTTNG_POLL_GETEV(&events
, i
);
1004 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1006 health_code_update(&consumer_data
->health
);
1008 /* Thread quit pipe has been closed. Killing thread. */
1009 ret
= check_thread_quit_pipe(pollfd
, revents
);
1015 /* Event on the kconsumerd socket */
1016 if (pollfd
== sock
) {
1017 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1018 ERR("consumer err socket second poll error");
1024 health_code_update(&consumer_data
->health
);
1026 /* Wait for any kconsumerd error */
1027 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1028 sizeof(enum lttcomm_return_code
));
1030 ERR("consumer closed the command socket");
1034 ERR("consumer return code : %s", lttcomm_get_readable_code(-code
));
1038 /* Immediately set the consumerd state to stopped */
1039 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1040 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1041 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1042 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1043 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1045 /* Code flow error... */
1049 if (consumer_data
->err_sock
>= 0) {
1050 ret
= close(consumer_data
->err_sock
);
1055 if (consumer_data
->cmd_sock
>= 0) {
1056 ret
= close(consumer_data
->cmd_sock
);
1068 unlink(consumer_data
->err_unix_sock_path
);
1069 unlink(consumer_data
->cmd_unix_sock_path
);
1070 consumer_data
->pid
= 0;
1072 lttng_poll_clean(&events
);
1075 health_error(&consumer_data
->health
);
1076 ERR("Health error occurred in %s", __func__
);
1078 health_exit(&consumer_data
->health
);
1079 DBG("consumer thread cleanup completed");
1085 * This thread manage application communication.
1087 static void *thread_manage_apps(void *data
)
1089 int i
, ret
, pollfd
, err
= -1;
1090 uint32_t revents
, nb_fd
;
1091 struct ust_command ust_cmd
;
1092 struct lttng_poll_event events
;
1094 DBG("[thread] Manage application started");
1096 testpoint(thread_manage_apps
);
1098 rcu_register_thread();
1099 rcu_thread_online();
1101 health_code_update(&health_thread_app_manage
);
1103 ret
= create_thread_poll_set(&events
, 2);
1105 goto error_poll_create
;
1108 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1113 testpoint(thread_manage_apps_before_loop
);
1115 health_code_update(&health_thread_app_manage
);
1118 /* Zeroed the events structure */
1119 lttng_poll_reset(&events
);
1121 nb_fd
= LTTNG_POLL_GETNB(&events
);
1123 DBG("Apps thread polling on %d fds", nb_fd
);
1125 /* Inifinite blocking call, waiting for transmission */
1127 health_poll_update(&health_thread_app_manage
);
1128 ret
= lttng_poll_wait(&events
, -1);
1129 health_poll_update(&health_thread_app_manage
);
1132 * Restart interrupted system call.
1134 if (errno
== EINTR
) {
1140 for (i
= 0; i
< nb_fd
; i
++) {
1141 /* Fetch once the poll data */
1142 revents
= LTTNG_POLL_GETEV(&events
, i
);
1143 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1145 health_code_update(&health_thread_app_manage
);
1147 /* Thread quit pipe has been closed. Killing thread. */
1148 ret
= check_thread_quit_pipe(pollfd
, revents
);
1154 /* Inspect the apps cmd pipe */
1155 if (pollfd
== apps_cmd_pipe
[0]) {
1156 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1157 ERR("Apps command pipe error");
1159 } else if (revents
& LPOLLIN
) {
1161 ret
= read(apps_cmd_pipe
[0], &ust_cmd
, sizeof(ust_cmd
));
1162 if (ret
< 0 || ret
< sizeof(ust_cmd
)) {
1163 PERROR("read apps cmd pipe");
1167 health_code_update(&health_thread_app_manage
);
1169 /* Register applicaton to the session daemon */
1170 ret
= ust_app_register(&ust_cmd
.reg_msg
,
1172 if (ret
== -ENOMEM
) {
1174 } else if (ret
< 0) {
1178 health_code_update(&health_thread_app_manage
);
1181 * Validate UST version compatibility.
1183 ret
= ust_app_validate_version(ust_cmd
.sock
);
1186 * Add channel(s) and event(s) to newly registered apps
1187 * from lttng global UST domain.
1189 update_ust_app(ust_cmd
.sock
);
1192 health_code_update(&health_thread_app_manage
);
1194 ret
= ust_app_register_done(ust_cmd
.sock
);
1197 * If the registration is not possible, we simply
1198 * unregister the apps and continue
1200 ust_app_unregister(ust_cmd
.sock
);
1203 * We only monitor the error events of the socket. This
1204 * thread does not handle any incoming data from UST
1207 ret
= lttng_poll_add(&events
, ust_cmd
.sock
,
1208 LPOLLERR
& LPOLLHUP
& LPOLLRDHUP
);
1213 /* Set socket timeout for both receiving and ending */
1214 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd
.sock
,
1215 app_socket_timeout
);
1216 (void) lttcomm_setsockopt_snd_timeout(ust_cmd
.sock
,
1217 app_socket_timeout
);
1219 DBG("Apps with sock %d added to poll set",
1223 health_code_update(&health_thread_app_manage
);
1229 * At this point, we know that a registered application made
1230 * the event at poll_wait.
1232 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1233 /* Removing from the poll set */
1234 ret
= lttng_poll_del(&events
, pollfd
);
1239 /* Socket closed on remote end. */
1240 ust_app_unregister(pollfd
);
1245 health_code_update(&health_thread_app_manage
);
1251 lttng_poll_clean(&events
);
1253 utils_close_pipe(apps_cmd_pipe
);
1254 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1257 * We don't clean the UST app hash table here since already registered
1258 * applications can still be controlled so let them be until the session
1259 * daemon dies or the applications stop.
1263 health_error(&health_thread_app_manage
);
1264 ERR("Health error occurred in %s", __func__
);
1266 health_exit(&health_thread_app_manage
);
1267 DBG("Application communication apps thread cleanup complete");
1268 rcu_thread_offline();
1269 rcu_unregister_thread();
1274 * Dispatch request from the registration threads to the application
1275 * communication thread.
1277 static void *thread_dispatch_ust_registration(void *data
)
1280 struct cds_wfq_node
*node
;
1281 struct ust_command
*ust_cmd
= NULL
;
1283 DBG("[thread] Dispatch UST command started");
1285 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1286 /* Atomically prepare the queue futex */
1287 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1290 /* Dequeue command for registration */
1291 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1293 DBG("Woken up but nothing in the UST command queue");
1294 /* Continue thread execution */
1298 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1300 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1301 " gid:%d sock:%d name:%s (version %d.%d)",
1302 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1303 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1304 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1305 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1307 * Inform apps thread of the new application registration. This
1308 * call is blocking so we can be assured that the data will be read
1309 * at some point in time or wait to the end of the world :)
1311 if (apps_cmd_pipe
[1] >= 0) {
1312 ret
= write(apps_cmd_pipe
[1], ust_cmd
,
1313 sizeof(struct ust_command
));
1315 PERROR("write apps cmd pipe");
1316 if (errno
== EBADF
) {
1318 * We can't inform the application thread to process
1319 * registration. We will exit or else application
1320 * registration will not occur and tracing will never
1327 /* Application manager thread is not available. */
1328 ret
= close(ust_cmd
->sock
);
1330 PERROR("close ust_cmd sock");
1334 } while (node
!= NULL
);
1336 /* Futex wait on queue. Blocking call on futex() */
1337 futex_nto1_wait(&ust_cmd_queue
.futex
);
1341 DBG("Dispatch thread dying");
1346 * This thread manage application registration.
1348 static void *thread_registration_apps(void *data
)
1350 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1351 uint32_t revents
, nb_fd
;
1352 struct lttng_poll_event events
;
1354 * Get allocated in this thread, enqueued to a global queue, dequeued and
1355 * freed in the manage apps thread.
1357 struct ust_command
*ust_cmd
= NULL
;
1359 DBG("[thread] Manage application registration started");
1361 testpoint(thread_registration_apps
);
1363 ret
= lttcomm_listen_unix_sock(apps_sock
);
1369 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1370 * more will be added to this poll set.
1372 ret
= create_thread_poll_set(&events
, 2);
1374 goto error_create_poll
;
1377 /* Add the application registration socket */
1378 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1380 goto error_poll_add
;
1383 /* Notify all applications to register */
1384 ret
= notify_ust_apps(1);
1386 ERR("Failed to notify applications or create the wait shared memory.\n"
1387 "Execution continues but there might be problem for already\n"
1388 "running applications that wishes to register.");
1392 DBG("Accepting application registration");
1394 nb_fd
= LTTNG_POLL_GETNB(&events
);
1396 /* Inifinite blocking call, waiting for transmission */
1398 health_poll_update(&health_thread_app_reg
);
1399 ret
= lttng_poll_wait(&events
, -1);
1400 health_poll_update(&health_thread_app_reg
);
1403 * Restart interrupted system call.
1405 if (errno
== EINTR
) {
1411 for (i
= 0; i
< nb_fd
; i
++) {
1412 health_code_update(&health_thread_app_reg
);
1414 /* Fetch once the poll data */
1415 revents
= LTTNG_POLL_GETEV(&events
, i
);
1416 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1418 /* Thread quit pipe has been closed. Killing thread. */
1419 ret
= check_thread_quit_pipe(pollfd
, revents
);
1425 /* Event on the registration socket */
1426 if (pollfd
== apps_sock
) {
1427 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1428 ERR("Register apps socket poll error");
1430 } else if (revents
& LPOLLIN
) {
1431 sock
= lttcomm_accept_unix_sock(apps_sock
);
1437 * Set the CLOEXEC flag. Return code is useless because
1438 * either way, the show must go on.
1440 (void) utils_set_fd_cloexec(sock
);
1442 /* Create UST registration command for enqueuing */
1443 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1444 if (ust_cmd
== NULL
) {
1445 PERROR("ust command zmalloc");
1450 * Using message-based transmissions to ensure we don't
1451 * have to deal with partially received messages.
1453 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1455 ERR("Exhausted file descriptors allowed for applications.");
1464 health_code_update(&health_thread_app_reg
);
1465 ret
= lttcomm_recv_unix_sock(sock
, &ust_cmd
->reg_msg
,
1466 sizeof(struct ust_register_msg
));
1467 if (ret
< 0 || ret
< sizeof(struct ust_register_msg
)) {
1469 PERROR("lttcomm_recv_unix_sock register apps");
1471 ERR("Wrong size received on apps register");
1478 lttng_fd_put(LTTNG_FD_APPS
, 1);
1482 health_code_update(&health_thread_app_reg
);
1484 ust_cmd
->sock
= sock
;
1487 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1488 " gid:%d sock:%d name:%s (version %d.%d)",
1489 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1490 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1491 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1492 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1495 * Lock free enqueue the registration request. The red pill
1496 * has been taken! This apps will be part of the *system*.
1498 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1501 * Wake the registration queue futex. Implicit memory
1502 * barrier with the exchange in cds_wfq_enqueue.
1504 futex_nto1_wake(&ust_cmd_queue
.futex
);
1513 health_error(&health_thread_app_reg
);
1514 ERR("Health error occurred in %s", __func__
);
1516 health_exit(&health_thread_app_reg
);
1518 /* Notify that the registration thread is gone */
1521 if (apps_sock
>= 0) {
1522 ret
= close(apps_sock
);
1532 lttng_fd_put(LTTNG_FD_APPS
, 1);
1534 unlink(apps_unix_sock_path
);
1537 lttng_poll_clean(&events
);
1540 DBG("UST Registration thread cleanup complete");
1546 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1547 * exec or it will fails.
1549 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1552 struct timespec timeout
;
1554 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1555 consumer_data
->consumer_thread_is_ready
= 0;
1557 /* Setup pthread condition */
1558 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1561 PERROR("pthread_condattr_init consumer data");
1566 * Set the monotonic clock in order to make sure we DO NOT jump in time
1567 * between the clock_gettime() call and the timedwait call. See bug #324
1568 * for a more details and how we noticed it.
1570 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1573 PERROR("pthread_condattr_setclock consumer data");
1577 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1580 PERROR("pthread_cond_init consumer data");
1584 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1587 PERROR("pthread_create consumer");
1592 /* We are about to wait on a pthread condition */
1593 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1595 /* Get time for sem_timedwait absolute timeout */
1596 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1598 * Set the timeout for the condition timed wait even if the clock gettime
1599 * call fails since we might loop on that call and we want to avoid to
1600 * increment the timeout too many times.
1602 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1605 * The following loop COULD be skipped in some conditions so this is why we
1606 * set ret to 0 in order to make sure at least one round of the loop is
1612 * Loop until the condition is reached or when a timeout is reached. Note
1613 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1614 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1615 * possible. This loop does not take any chances and works with both of
1618 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1619 if (clock_ret
< 0) {
1620 PERROR("clock_gettime spawn consumer");
1621 /* Infinite wait for the consumerd thread to be ready */
1622 ret
= pthread_cond_wait(&consumer_data
->cond
,
1623 &consumer_data
->cond_mutex
);
1625 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1626 &consumer_data
->cond_mutex
, &timeout
);
1630 /* Release the pthread condition */
1631 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1635 if (ret
== ETIMEDOUT
) {
1637 * Call has timed out so we kill the kconsumerd_thread and return
1640 ERR("Condition timed out. The consumer thread was never ready."
1642 ret
= pthread_cancel(consumer_data
->thread
);
1644 PERROR("pthread_cancel consumer thread");
1647 PERROR("pthread_cond_wait failed consumer thread");
1652 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1653 if (consumer_data
->pid
== 0) {
1654 ERR("Consumerd did not start");
1655 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1658 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1667 * Join consumer thread
1669 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1674 /* Consumer pid must be a real one. */
1675 if (consumer_data
->pid
> 0) {
1676 ret
= kill(consumer_data
->pid
, SIGTERM
);
1678 ERR("Error killing consumer daemon");
1681 return pthread_join(consumer_data
->thread
, &status
);
1688 * Fork and exec a consumer daemon (consumerd).
1690 * Return pid if successful else -1.
1692 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1696 const char *consumer_to_use
;
1697 const char *verbosity
;
1700 DBG("Spawning consumerd");
1707 if (opt_verbose_consumer
) {
1708 verbosity
= "--verbose";
1710 verbosity
= "--quiet";
1712 switch (consumer_data
->type
) {
1713 case LTTNG_CONSUMER_KERNEL
:
1715 * Find out which consumerd to execute. We will first try the
1716 * 64-bit path, then the sessiond's installation directory, and
1717 * fallback on the 32-bit one,
1719 DBG3("Looking for a kernel consumer at these locations:");
1720 DBG3(" 1) %s", consumerd64_bin
);
1721 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1722 DBG3(" 3) %s", consumerd32_bin
);
1723 if (stat(consumerd64_bin
, &st
) == 0) {
1724 DBG3("Found location #1");
1725 consumer_to_use
= consumerd64_bin
;
1726 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1727 DBG3("Found location #2");
1728 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1729 } else if (stat(consumerd32_bin
, &st
) == 0) {
1730 DBG3("Found location #3");
1731 consumer_to_use
= consumerd32_bin
;
1733 DBG("Could not find any valid consumerd executable");
1736 DBG("Using kernel consumer at: %s", consumer_to_use
);
1737 execl(consumer_to_use
,
1738 "lttng-consumerd", verbosity
, "-k",
1739 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1740 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1743 case LTTNG_CONSUMER64_UST
:
1745 char *tmpnew
= NULL
;
1747 if (consumerd64_libdir
[0] != '\0') {
1751 tmp
= getenv("LD_LIBRARY_PATH");
1755 tmplen
= strlen("LD_LIBRARY_PATH=")
1756 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1757 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1762 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1763 strcat(tmpnew
, consumerd64_libdir
);
1764 if (tmp
[0] != '\0') {
1765 strcat(tmpnew
, ":");
1766 strcat(tmpnew
, tmp
);
1768 ret
= putenv(tmpnew
);
1774 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1775 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1776 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1777 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1779 if (consumerd64_libdir
[0] != '\0') {
1787 case LTTNG_CONSUMER32_UST
:
1789 char *tmpnew
= NULL
;
1791 if (consumerd32_libdir
[0] != '\0') {
1795 tmp
= getenv("LD_LIBRARY_PATH");
1799 tmplen
= strlen("LD_LIBRARY_PATH=")
1800 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1801 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1806 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1807 strcat(tmpnew
, consumerd32_libdir
);
1808 if (tmp
[0] != '\0') {
1809 strcat(tmpnew
, ":");
1810 strcat(tmpnew
, tmp
);
1812 ret
= putenv(tmpnew
);
1818 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
1819 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
1820 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1821 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1823 if (consumerd32_libdir
[0] != '\0') {
1832 PERROR("unknown consumer type");
1836 PERROR("kernel start consumer exec");
1839 } else if (pid
> 0) {
1842 PERROR("start consumer fork");
1850 * Spawn the consumerd daemon and session daemon thread.
1852 static int start_consumerd(struct consumer_data
*consumer_data
)
1857 * Set the listen() state on the socket since there is a possible race
1858 * between the exec() of the consumer daemon and this call if place in the
1859 * consumer thread. See bug #366 for more details.
1861 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
1866 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1867 if (consumer_data
->pid
!= 0) {
1868 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1872 ret
= spawn_consumerd(consumer_data
);
1874 ERR("Spawning consumerd failed");
1875 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1879 /* Setting up the consumer_data pid */
1880 consumer_data
->pid
= ret
;
1881 DBG2("Consumer pid %d", consumer_data
->pid
);
1882 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1884 DBG2("Spawning consumer control thread");
1885 ret
= spawn_consumer_thread(consumer_data
);
1887 ERR("Fatal error spawning consumer control thread");
1895 /* Cleanup already created socket on error. */
1896 if (consumer_data
->err_sock
>= 0) {
1897 err
= close(consumer_data
->err_sock
);
1899 PERROR("close consumer data error socket");
1906 * Compute health status of each consumer. If one of them is zero (bad
1907 * state), we return 0.
1909 static int check_consumer_health(void)
1913 ret
= health_check_state(&kconsumer_data
.health
) &&
1914 health_check_state(&ustconsumer32_data
.health
) &&
1915 health_check_state(&ustconsumer64_data
.health
);
1917 DBG3("Health consumer check %d", ret
);
1923 * Setup necessary data for kernel tracer action.
1925 static int init_kernel_tracer(void)
1929 /* Modprobe lttng kernel modules */
1930 ret
= modprobe_lttng_control();
1935 /* Open debugfs lttng */
1936 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1937 if (kernel_tracer_fd
< 0) {
1938 DBG("Failed to open %s", module_proc_lttng
);
1943 /* Validate kernel version */
1944 ret
= kernel_validate_version(kernel_tracer_fd
);
1949 ret
= modprobe_lttng_data();
1954 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1958 modprobe_remove_lttng_control();
1959 ret
= close(kernel_tracer_fd
);
1963 kernel_tracer_fd
= -1;
1964 return LTTNG_ERR_KERN_VERSION
;
1967 ret
= close(kernel_tracer_fd
);
1973 modprobe_remove_lttng_control();
1976 WARN("No kernel tracer available");
1977 kernel_tracer_fd
= -1;
1979 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
1981 return LTTNG_ERR_KERN_NA
;
1987 * Copy consumer output from the tracing session to the domain session. The
1988 * function also applies the right modification on a per domain basis for the
1989 * trace files destination directory.
1991 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
1994 const char *dir_name
;
1995 struct consumer_output
*consumer
;
1998 assert(session
->consumer
);
2001 case LTTNG_DOMAIN_KERNEL
:
2002 DBG3("Copying tracing session consumer output in kernel session");
2004 * XXX: We should audit the session creation and what this function
2005 * does "extra" in order to avoid a destroy since this function is used
2006 * in the domain session creation (kernel and ust) only. Same for UST
2009 if (session
->kernel_session
->consumer
) {
2010 consumer_destroy_output(session
->kernel_session
->consumer
);
2012 session
->kernel_session
->consumer
=
2013 consumer_copy_output(session
->consumer
);
2014 /* Ease our life a bit for the next part */
2015 consumer
= session
->kernel_session
->consumer
;
2016 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2018 case LTTNG_DOMAIN_UST
:
2019 DBG3("Copying tracing session consumer output in UST session");
2020 if (session
->ust_session
->consumer
) {
2021 consumer_destroy_output(session
->ust_session
->consumer
);
2023 session
->ust_session
->consumer
=
2024 consumer_copy_output(session
->consumer
);
2025 /* Ease our life a bit for the next part */
2026 consumer
= session
->ust_session
->consumer
;
2027 dir_name
= DEFAULT_UST_TRACE_DIR
;
2030 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2034 /* Append correct directory to subdir */
2035 strncat(consumer
->subdir
, dir_name
,
2036 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2037 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2046 * Create an UST session and add it to the session ust list.
2048 static int create_ust_session(struct ltt_session
*session
,
2049 struct lttng_domain
*domain
)
2052 struct ltt_ust_session
*lus
= NULL
;
2056 assert(session
->consumer
);
2058 switch (domain
->type
) {
2059 case LTTNG_DOMAIN_UST
:
2062 ERR("Unknown UST domain on create session %d", domain
->type
);
2063 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2067 DBG("Creating UST session");
2069 lus
= trace_ust_create_session(session
->path
, session
->id
, domain
);
2071 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2075 lus
->uid
= session
->uid
;
2076 lus
->gid
= session
->gid
;
2077 session
->ust_session
= lus
;
2079 /* Copy session output to the newly created UST session */
2080 ret
= copy_session_consumer(domain
->type
, session
);
2081 if (ret
!= LTTNG_OK
) {
2089 session
->ust_session
= NULL
;
2094 * Create a kernel tracer session then create the default channel.
2096 static int create_kernel_session(struct ltt_session
*session
)
2100 DBG("Creating kernel session");
2102 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2104 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2108 /* Code flow safety */
2109 assert(session
->kernel_session
);
2111 /* Copy session output to the newly created Kernel session */
2112 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2113 if (ret
!= LTTNG_OK
) {
2117 /* Create directory(ies) on local filesystem. */
2118 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2119 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2120 ret
= run_as_mkdir_recursive(
2121 session
->kernel_session
->consumer
->dst
.trace_path
,
2122 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2124 if (ret
!= -EEXIST
) {
2125 ERR("Trace directory creation error");
2131 session
->kernel_session
->uid
= session
->uid
;
2132 session
->kernel_session
->gid
= session
->gid
;
2137 trace_kernel_destroy_session(session
->kernel_session
);
2138 session
->kernel_session
= NULL
;
2143 * Count number of session permitted by uid/gid.
2145 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2148 struct ltt_session
*session
;
2150 DBG("Counting number of available session for UID %d GID %d",
2152 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2154 * Only list the sessions the user can control.
2156 if (!session_access_ok(session
, uid
, gid
)) {
2165 * Process the command requested by the lttng client within the command
2166 * context structure. This function make sure that the return structure (llm)
2167 * is set and ready for transmission before returning.
2169 * Return any error encountered or 0 for success.
2171 * "sock" is only used for special-case var. len data.
2173 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2177 int need_tracing_session
= 1;
2180 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2184 switch (cmd_ctx
->lsm
->cmd_type
) {
2185 case LTTNG_CREATE_SESSION
:
2186 case LTTNG_DESTROY_SESSION
:
2187 case LTTNG_LIST_SESSIONS
:
2188 case LTTNG_LIST_DOMAINS
:
2189 case LTTNG_START_TRACE
:
2190 case LTTNG_STOP_TRACE
:
2191 case LTTNG_DATA_PENDING
:
2198 if (opt_no_kernel
&& need_domain
2199 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2201 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2203 ret
= LTTNG_ERR_KERN_NA
;
2208 /* Deny register consumer if we already have a spawned consumer. */
2209 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2210 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2211 if (kconsumer_data
.pid
> 0) {
2212 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2213 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2216 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2220 * Check for command that don't needs to allocate a returned payload. We do
2221 * this here so we don't have to make the call for no payload at each
2224 switch(cmd_ctx
->lsm
->cmd_type
) {
2225 case LTTNG_LIST_SESSIONS
:
2226 case LTTNG_LIST_TRACEPOINTS
:
2227 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2228 case LTTNG_LIST_DOMAINS
:
2229 case LTTNG_LIST_CHANNELS
:
2230 case LTTNG_LIST_EVENTS
:
2233 /* Setup lttng message with no payload */
2234 ret
= setup_lttng_msg(cmd_ctx
, 0);
2236 /* This label does not try to unlock the session */
2237 goto init_setup_error
;
2241 /* Commands that DO NOT need a session. */
2242 switch (cmd_ctx
->lsm
->cmd_type
) {
2243 case LTTNG_CREATE_SESSION
:
2244 case LTTNG_CALIBRATE
:
2245 case LTTNG_LIST_SESSIONS
:
2246 case LTTNG_LIST_TRACEPOINTS
:
2247 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2248 need_tracing_session
= 0;
2251 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2253 * We keep the session list lock across _all_ commands
2254 * for now, because the per-session lock does not
2255 * handle teardown properly.
2257 session_lock_list();
2258 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2259 if (cmd_ctx
->session
== NULL
) {
2260 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2261 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2263 /* If no session name specified */
2264 ret
= LTTNG_ERR_SELECT_SESS
;
2268 /* Acquire lock for the session */
2269 session_lock(cmd_ctx
->session
);
2279 * Check domain type for specific "pre-action".
2281 switch (cmd_ctx
->lsm
->domain
.type
) {
2282 case LTTNG_DOMAIN_KERNEL
:
2284 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2288 /* Kernel tracer check */
2289 if (kernel_tracer_fd
== -1) {
2290 /* Basically, load kernel tracer modules */
2291 ret
= init_kernel_tracer();
2297 /* Consumer is in an ERROR state. Report back to client */
2298 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2299 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2303 /* Need a session for kernel command */
2304 if (need_tracing_session
) {
2305 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2306 ret
= create_kernel_session(cmd_ctx
->session
);
2308 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2313 /* Start the kernel consumer daemon */
2314 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2315 if (kconsumer_data
.pid
== 0 &&
2316 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2317 cmd_ctx
->session
->start_consumer
) {
2318 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2319 ret
= start_consumerd(&kconsumer_data
);
2321 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2324 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2326 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2330 * The consumer was just spawned so we need to add the socket to
2331 * the consumer output of the session if exist.
2333 ret
= consumer_create_socket(&kconsumer_data
,
2334 cmd_ctx
->session
->kernel_session
->consumer
);
2341 case LTTNG_DOMAIN_UST
:
2343 /* Consumer is in an ERROR state. Report back to client */
2344 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2345 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2349 if (need_tracing_session
) {
2350 /* Create UST session if none exist. */
2351 if (cmd_ctx
->session
->ust_session
== NULL
) {
2352 ret
= create_ust_session(cmd_ctx
->session
,
2353 &cmd_ctx
->lsm
->domain
);
2354 if (ret
!= LTTNG_OK
) {
2359 /* Start the UST consumer daemons */
2361 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2362 if (consumerd64_bin
[0] != '\0' &&
2363 ustconsumer64_data
.pid
== 0 &&
2364 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2365 cmd_ctx
->session
->start_consumer
) {
2366 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2367 ret
= start_consumerd(&ustconsumer64_data
);
2369 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2370 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2374 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2375 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2377 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2381 * Setup socket for consumer 64 bit. No need for atomic access
2382 * since it was set above and can ONLY be set in this thread.
2384 ret
= consumer_create_socket(&ustconsumer64_data
,
2385 cmd_ctx
->session
->ust_session
->consumer
);
2391 if (consumerd32_bin
[0] != '\0' &&
2392 ustconsumer32_data
.pid
== 0 &&
2393 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2394 cmd_ctx
->session
->start_consumer
) {
2395 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2396 ret
= start_consumerd(&ustconsumer32_data
);
2398 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2399 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2403 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2404 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2406 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2410 * Setup socket for consumer 64 bit. No need for atomic access
2411 * since it was set above and can ONLY be set in this thread.
2413 ret
= consumer_create_socket(&ustconsumer32_data
,
2414 cmd_ctx
->session
->ust_session
->consumer
);
2426 /* Validate consumer daemon state when start/stop trace command */
2427 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2428 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2429 switch (cmd_ctx
->lsm
->domain
.type
) {
2430 case LTTNG_DOMAIN_UST
:
2431 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2432 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2436 case LTTNG_DOMAIN_KERNEL
:
2437 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2438 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2446 * Check that the UID or GID match that of the tracing session.
2447 * The root user can interact with all sessions.
2449 if (need_tracing_session
) {
2450 if (!session_access_ok(cmd_ctx
->session
,
2451 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2452 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2453 ret
= LTTNG_ERR_EPERM
;
2458 /* Process by command type */
2459 switch (cmd_ctx
->lsm
->cmd_type
) {
2460 case LTTNG_ADD_CONTEXT
:
2462 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2463 cmd_ctx
->lsm
->u
.context
.channel_name
,
2464 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2467 case LTTNG_DISABLE_CHANNEL
:
2469 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2470 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2473 case LTTNG_DISABLE_EVENT
:
2475 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2476 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2477 cmd_ctx
->lsm
->u
.disable
.name
);
2480 case LTTNG_DISABLE_ALL_EVENT
:
2482 DBG("Disabling all events");
2484 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2485 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2488 case LTTNG_DISABLE_CONSUMER
:
2490 ret
= cmd_disable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2493 case LTTNG_ENABLE_CHANNEL
:
2495 ret
= cmd_enable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2496 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2499 case LTTNG_ENABLE_CONSUMER
:
2502 * XXX: 0 means that this URI should be applied on the session. Should
2503 * be a DOMAIN enuam.
2505 ret
= cmd_enable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2506 if (ret
!= LTTNG_OK
) {
2510 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2511 /* Add the URI for the UST session if a consumer is present. */
2512 if (cmd_ctx
->session
->ust_session
&&
2513 cmd_ctx
->session
->ust_session
->consumer
) {
2514 ret
= cmd_enable_consumer(LTTNG_DOMAIN_UST
, cmd_ctx
->session
);
2515 } else if (cmd_ctx
->session
->kernel_session
&&
2516 cmd_ctx
->session
->kernel_session
->consumer
) {
2517 ret
= cmd_enable_consumer(LTTNG_DOMAIN_KERNEL
,
2523 case LTTNG_ENABLE_EVENT
:
2525 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2526 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2527 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2530 case LTTNG_ENABLE_ALL_EVENT
:
2532 DBG("Enabling all events");
2534 ret
= cmd_enable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2535 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2536 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2539 case LTTNG_LIST_TRACEPOINTS
:
2541 struct lttng_event
*events
;
2544 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2545 if (nb_events
< 0) {
2546 /* Return value is a negative lttng_error_code. */
2552 * Setup lttng message with payload size set to the event list size in
2553 * bytes and then copy list into the llm payload.
2555 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2561 /* Copy event list into message payload */
2562 memcpy(cmd_ctx
->llm
->payload
, events
,
2563 sizeof(struct lttng_event
) * nb_events
);
2570 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2572 struct lttng_event_field
*fields
;
2575 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2577 if (nb_fields
< 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
,
2588 sizeof(struct lttng_event_field
) * nb_fields
);
2594 /* Copy event list into message payload */
2595 memcpy(cmd_ctx
->llm
->payload
, fields
,
2596 sizeof(struct lttng_event_field
) * nb_fields
);
2603 case LTTNG_SET_CONSUMER_URI
:
2606 struct lttng_uri
*uris
;
2608 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2609 len
= nb_uri
* sizeof(struct lttng_uri
);
2612 ret
= LTTNG_ERR_INVALID
;
2616 uris
= zmalloc(len
);
2618 ret
= LTTNG_ERR_FATAL
;
2622 /* Receive variable len data */
2623 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2624 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2626 DBG("No URIs received from client... continuing");
2628 ret
= LTTNG_ERR_SESSION_FAIL
;
2633 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2635 if (ret
!= LTTNG_OK
) {
2641 * XXX: 0 means that this URI should be applied on the session. Should
2642 * be a DOMAIN enuam.
2644 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2645 /* Add the URI for the UST session if a consumer is present. */
2646 if (cmd_ctx
->session
->ust_session
&&
2647 cmd_ctx
->session
->ust_session
->consumer
) {
2648 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2650 } else if (cmd_ctx
->session
->kernel_session
&&
2651 cmd_ctx
->session
->kernel_session
->consumer
) {
2652 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2653 cmd_ctx
->session
, nb_uri
, uris
);
2661 case LTTNG_START_TRACE
:
2663 ret
= cmd_start_trace(cmd_ctx
->session
);
2666 case LTTNG_STOP_TRACE
:
2668 ret
= cmd_stop_trace(cmd_ctx
->session
);
2671 case LTTNG_CREATE_SESSION
:
2674 struct lttng_uri
*uris
= NULL
;
2676 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2677 len
= nb_uri
* sizeof(struct lttng_uri
);
2680 uris
= zmalloc(len
);
2682 ret
= LTTNG_ERR_FATAL
;
2686 /* Receive variable len data */
2687 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2688 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2690 DBG("No URIs received from client... continuing");
2692 ret
= LTTNG_ERR_SESSION_FAIL
;
2697 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2698 DBG("Creating session with ONE network URI is a bad call");
2699 ret
= LTTNG_ERR_SESSION_FAIL
;
2705 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2712 case LTTNG_DESTROY_SESSION
:
2714 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2716 /* Set session to NULL so we do not unlock it after free. */
2717 cmd_ctx
->session
= NULL
;
2720 case LTTNG_LIST_DOMAINS
:
2723 struct lttng_domain
*domains
;
2725 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2727 /* Return value is a negative lttng_error_code. */
2732 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2737 /* Copy event list into message payload */
2738 memcpy(cmd_ctx
->llm
->payload
, domains
,
2739 nb_dom
* sizeof(struct lttng_domain
));
2746 case LTTNG_LIST_CHANNELS
:
2749 struct lttng_channel
*channels
;
2751 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2752 cmd_ctx
->session
, &channels
);
2754 /* Return value is a negative lttng_error_code. */
2759 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2764 /* Copy event list into message payload */
2765 memcpy(cmd_ctx
->llm
->payload
, channels
,
2766 nb_chan
* sizeof(struct lttng_channel
));
2773 case LTTNG_LIST_EVENTS
:
2776 struct lttng_event
*events
= NULL
;
2778 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2779 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2781 /* Return value is a negative lttng_error_code. */
2786 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2791 /* Copy event list into message payload */
2792 memcpy(cmd_ctx
->llm
->payload
, events
,
2793 nb_event
* sizeof(struct lttng_event
));
2800 case LTTNG_LIST_SESSIONS
:
2802 unsigned int nr_sessions
;
2804 session_lock_list();
2805 nr_sessions
= lttng_sessions_count(
2806 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2807 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2809 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2811 session_unlock_list();
2815 /* Filled the session array */
2816 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2817 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2818 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2820 session_unlock_list();
2825 case LTTNG_CALIBRATE
:
2827 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
2828 &cmd_ctx
->lsm
->u
.calibrate
);
2831 case LTTNG_REGISTER_CONSUMER
:
2833 struct consumer_data
*cdata
;
2835 switch (cmd_ctx
->lsm
->domain
.type
) {
2836 case LTTNG_DOMAIN_KERNEL
:
2837 cdata
= &kconsumer_data
;
2840 ret
= LTTNG_ERR_UND
;
2844 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2845 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
2848 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
2850 struct lttng_filter_bytecode
*bytecode
;
2852 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
2853 ret
= LTTNG_ERR_FILTER_INVAL
;
2856 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
2857 ret
= LTTNG_ERR_FILTER_INVAL
;
2860 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2862 ret
= LTTNG_ERR_FILTER_NOMEM
;
2865 /* Receive var. len. data */
2866 DBG("Receiving var len data from client ...");
2867 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
2868 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2870 DBG("Nothing recv() from client var len data... continuing");
2872 ret
= LTTNG_ERR_FILTER_INVAL
;
2876 if (bytecode
->len
+ sizeof(*bytecode
)
2877 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
2879 ret
= LTTNG_ERR_FILTER_INVAL
;
2883 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2884 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2885 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
2888 case LTTNG_DATA_PENDING
:
2890 ret
= cmd_data_pending(cmd_ctx
->session
);
2894 ret
= LTTNG_ERR_UND
;
2899 if (cmd_ctx
->llm
== NULL
) {
2900 DBG("Missing llm structure. Allocating one.");
2901 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
2905 /* Set return code */
2906 cmd_ctx
->llm
->ret_code
= ret
;
2908 if (cmd_ctx
->session
) {
2909 session_unlock(cmd_ctx
->session
);
2911 if (need_tracing_session
) {
2912 session_unlock_list();
2919 * Thread managing health check socket.
2921 static void *thread_manage_health(void *data
)
2923 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
2924 uint32_t revents
, nb_fd
;
2925 struct lttng_poll_event events
;
2926 struct lttcomm_health_msg msg
;
2927 struct lttcomm_health_data reply
;
2929 DBG("[thread] Manage health check started");
2931 rcu_register_thread();
2933 /* Create unix socket */
2934 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
2936 ERR("Unable to create health check Unix socket");
2942 * Set the CLOEXEC flag. Return code is useless because either way, the
2945 (void) utils_set_fd_cloexec(sock
);
2947 ret
= lttcomm_listen_unix_sock(sock
);
2953 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2954 * more will be added to this poll set.
2956 ret
= create_thread_poll_set(&events
, 2);
2961 /* Add the application registration socket */
2962 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
2968 DBG("Health check ready");
2970 nb_fd
= LTTNG_POLL_GETNB(&events
);
2972 /* Inifinite blocking call, waiting for transmission */
2974 ret
= lttng_poll_wait(&events
, -1);
2977 * Restart interrupted system call.
2979 if (errno
== EINTR
) {
2985 for (i
= 0; i
< nb_fd
; i
++) {
2986 /* Fetch once the poll data */
2987 revents
= LTTNG_POLL_GETEV(&events
, i
);
2988 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2990 /* Thread quit pipe has been closed. Killing thread. */
2991 ret
= check_thread_quit_pipe(pollfd
, revents
);
2997 /* Event on the registration socket */
2998 if (pollfd
== sock
) {
2999 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3000 ERR("Health socket poll error");
3006 new_sock
= lttcomm_accept_unix_sock(sock
);
3012 * Set the CLOEXEC flag. Return code is useless because either way, the
3015 (void) utils_set_fd_cloexec(new_sock
);
3017 DBG("Receiving data from client for health...");
3018 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3020 DBG("Nothing recv() from client... continuing");
3021 ret
= close(new_sock
);
3029 rcu_thread_online();
3031 switch (msg
.component
) {
3032 case LTTNG_HEALTH_CMD
:
3033 reply
.ret_code
= health_check_state(&health_thread_cmd
);
3035 case LTTNG_HEALTH_APP_MANAGE
:
3036 reply
.ret_code
= health_check_state(&health_thread_app_manage
);
3038 case LTTNG_HEALTH_APP_REG
:
3039 reply
.ret_code
= health_check_state(&health_thread_app_reg
);
3041 case LTTNG_HEALTH_KERNEL
:
3042 reply
.ret_code
= health_check_state(&health_thread_kernel
);
3044 case LTTNG_HEALTH_CONSUMER
:
3045 reply
.ret_code
= check_consumer_health();
3047 case LTTNG_HEALTH_ALL
:
3049 health_check_state(&health_thread_app_manage
) &&
3050 health_check_state(&health_thread_app_reg
) &&
3051 health_check_state(&health_thread_cmd
) &&
3052 health_check_state(&health_thread_kernel
) &&
3053 check_consumer_health();
3056 reply
.ret_code
= LTTNG_ERR_UND
;
3061 * Flip ret value since 0 is a success and 1 indicates a bad health for
3062 * the client where in the sessiond it is the opposite. Again, this is
3063 * just to make things easier for us poor developer which enjoy a lot
3066 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3067 reply
.ret_code
= !reply
.ret_code
;
3070 DBG2("Health check return value %d", reply
.ret_code
);
3072 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3074 ERR("Failed to send health data back to client");
3077 /* End of transmission */
3078 ret
= close(new_sock
);
3088 ERR("Health error occurred in %s", __func__
);
3090 DBG("Health check thread dying");
3091 unlink(health_unix_sock_path
);
3098 if (new_sock
>= 0) {
3099 ret
= close(new_sock
);
3105 lttng_poll_clean(&events
);
3107 rcu_unregister_thread();
3112 * This thread manage all clients request using the unix client socket for
3115 static void *thread_manage_clients(void *data
)
3117 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3119 uint32_t revents
, nb_fd
;
3120 struct command_ctx
*cmd_ctx
= NULL
;
3121 struct lttng_poll_event events
;
3123 DBG("[thread] Manage client started");
3125 testpoint(thread_manage_clients
);
3127 rcu_register_thread();
3129 health_code_update(&health_thread_cmd
);
3131 ret
= lttcomm_listen_unix_sock(client_sock
);
3137 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3138 * more will be added to this poll set.
3140 ret
= create_thread_poll_set(&events
, 2);
3142 goto error_create_poll
;
3145 /* Add the application registration socket */
3146 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3152 * Notify parent pid that we are ready to accept command for client side.
3154 if (opt_sig_parent
) {
3155 kill(ppid
, SIGUSR1
);
3158 testpoint(thread_manage_clients_before_loop
);
3160 health_code_update(&health_thread_cmd
);
3163 DBG("Accepting client command ...");
3165 nb_fd
= LTTNG_POLL_GETNB(&events
);
3167 /* Inifinite blocking call, waiting for transmission */
3169 health_poll_update(&health_thread_cmd
);
3170 ret
= lttng_poll_wait(&events
, -1);
3171 health_poll_update(&health_thread_cmd
);
3174 * Restart interrupted system call.
3176 if (errno
== EINTR
) {
3182 for (i
= 0; i
< nb_fd
; i
++) {
3183 /* Fetch once the poll data */
3184 revents
= LTTNG_POLL_GETEV(&events
, i
);
3185 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3187 health_code_update(&health_thread_cmd
);
3189 /* Thread quit pipe has been closed. Killing thread. */
3190 ret
= check_thread_quit_pipe(pollfd
, revents
);
3196 /* Event on the registration socket */
3197 if (pollfd
== client_sock
) {
3198 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3199 ERR("Client socket poll error");
3205 DBG("Wait for client response");
3207 health_code_update(&health_thread_cmd
);
3209 sock
= lttcomm_accept_unix_sock(client_sock
);
3215 * Set the CLOEXEC flag. Return code is useless because either way, the
3218 (void) utils_set_fd_cloexec(sock
);
3220 /* Set socket option for credentials retrieval */
3221 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3226 /* Allocate context command to process the client request */
3227 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3228 if (cmd_ctx
== NULL
) {
3229 PERROR("zmalloc cmd_ctx");
3233 /* Allocate data buffer for reception */
3234 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3235 if (cmd_ctx
->lsm
== NULL
) {
3236 PERROR("zmalloc cmd_ctx->lsm");
3240 cmd_ctx
->llm
= NULL
;
3241 cmd_ctx
->session
= NULL
;
3243 health_code_update(&health_thread_cmd
);
3246 * Data is received from the lttng client. The struct
3247 * lttcomm_session_msg (lsm) contains the command and data request of
3250 DBG("Receiving data from client ...");
3251 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3252 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3254 DBG("Nothing recv() from client... continuing");
3260 clean_command_ctx(&cmd_ctx
);
3264 health_code_update(&health_thread_cmd
);
3266 // TODO: Validate cmd_ctx including sanity check for
3267 // security purpose.
3269 rcu_thread_online();
3271 * This function dispatch the work to the kernel or userspace tracer
3272 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3273 * informations for the client. The command context struct contains
3274 * everything this function may needs.
3276 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3277 rcu_thread_offline();
3287 * TODO: Inform client somehow of the fatal error. At
3288 * this point, ret < 0 means that a zmalloc failed
3289 * (ENOMEM). Error detected but still accept
3290 * command, unless a socket error has been
3293 clean_command_ctx(&cmd_ctx
);
3297 health_code_update(&health_thread_cmd
);
3299 DBG("Sending response (size: %d, retcode: %s)",
3300 cmd_ctx
->lttng_msg_size
,
3301 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3302 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3304 ERR("Failed to send data back to client");
3307 /* End of transmission */
3314 clean_command_ctx(&cmd_ctx
);
3316 health_code_update(&health_thread_cmd
);
3328 lttng_poll_clean(&events
);
3329 clean_command_ctx(&cmd_ctx
);
3333 unlink(client_unix_sock_path
);
3334 if (client_sock
>= 0) {
3335 ret
= close(client_sock
);
3342 health_error(&health_thread_cmd
);
3343 ERR("Health error occurred in %s", __func__
);
3346 health_exit(&health_thread_cmd
);
3348 DBG("Client thread dying");
3350 rcu_unregister_thread();
3356 * usage function on stderr
3358 static void usage(void)
3360 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3361 fprintf(stderr
, " -h, --help Display this usage.\n");
3362 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3363 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3364 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3365 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3366 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3367 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3368 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3369 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3370 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3371 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3372 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3373 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3374 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3375 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3376 fprintf(stderr
, " -V, --version Show version number.\n");
3377 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3378 fprintf(stderr
, " -q, --quiet No output at all.\n");
3379 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3380 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3381 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3385 * daemon argument parsing
3387 static int parse_args(int argc
, char **argv
)
3391 static struct option long_options
[] = {
3392 { "client-sock", 1, 0, 'c' },
3393 { "apps-sock", 1, 0, 'a' },
3394 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3395 { "kconsumerd-err-sock", 1, 0, 'E' },
3396 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3397 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3398 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3399 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3400 { "consumerd32-path", 1, 0, 'u' },
3401 { "consumerd32-libdir", 1, 0, 'U' },
3402 { "consumerd64-path", 1, 0, 't' },
3403 { "consumerd64-libdir", 1, 0, 'T' },
3404 { "daemonize", 0, 0, 'd' },
3405 { "sig-parent", 0, 0, 'S' },
3406 { "help", 0, 0, 'h' },
3407 { "group", 1, 0, 'g' },
3408 { "version", 0, 0, 'V' },
3409 { "quiet", 0, 0, 'q' },
3410 { "verbose", 0, 0, 'v' },
3411 { "verbose-consumer", 0, 0, 'Z' },
3412 { "no-kernel", 0, 0, 'N' },
3417 int option_index
= 0;
3418 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3419 long_options
, &option_index
);
3426 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3428 fprintf(stderr
, " with arg %s\n", optarg
);
3432 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3435 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3441 opt_tracing_group
= optarg
;
3447 fprintf(stdout
, "%s\n", VERSION
);
3453 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3456 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3459 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3462 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3465 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3468 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3474 lttng_opt_quiet
= 1;
3477 /* Verbose level can increase using multiple -v */
3478 lttng_opt_verbose
+= 1;
3481 opt_verbose_consumer
+= 1;
3484 consumerd32_bin
= optarg
;
3487 consumerd32_libdir
= optarg
;
3490 consumerd64_bin
= optarg
;
3493 consumerd64_libdir
= optarg
;
3496 /* Unknown option or other error.
3497 * Error is printed by getopt, just return */
3506 * Creates the two needed socket by the daemon.
3507 * apps_sock - The communication socket for all UST apps.
3508 * client_sock - The communication of the cli tool (lttng).
3510 static int init_daemon_socket(void)
3515 old_umask
= umask(0);
3517 /* Create client tool unix socket */
3518 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3519 if (client_sock
< 0) {
3520 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3525 /* Set the cloexec flag */
3526 ret
= utils_set_fd_cloexec(client_sock
);
3528 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3529 "Continuing but note that the consumer daemon will have a "
3530 "reference to this socket on exec()", client_sock
);
3533 /* File permission MUST be 660 */
3534 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3536 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3541 /* Create the application unix socket */
3542 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3543 if (apps_sock
< 0) {
3544 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3549 /* Set the cloexec flag */
3550 ret
= utils_set_fd_cloexec(apps_sock
);
3552 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3553 "Continuing but note that the consumer daemon will have a "
3554 "reference to this socket on exec()", apps_sock
);
3557 /* File permission MUST be 666 */
3558 ret
= chmod(apps_unix_sock_path
,
3559 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3561 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3566 DBG3("Session daemon client socket %d and application socket %d created",
3567 client_sock
, apps_sock
);
3575 * Check if the global socket is available, and if a daemon is answering at the
3576 * other side. If yes, error is returned.
3578 static int check_existing_daemon(void)
3580 /* Is there anybody out there ? */
3581 if (lttng_session_daemon_alive()) {
3589 * Set the tracing group gid onto the client socket.
3591 * Race window between mkdir and chown is OK because we are going from more
3592 * permissive (root.root) to less permissive (root.tracing).
3594 static int set_permissions(char *rundir
)
3599 ret
= allowed_group();
3601 WARN("No tracing group detected");
3608 /* Set lttng run dir */
3609 ret
= chown(rundir
, 0, gid
);
3611 ERR("Unable to set group on %s", rundir
);
3615 /* Ensure tracing group can search the run dir */
3616 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3618 ERR("Unable to set permissions on %s", rundir
);
3622 /* lttng client socket path */
3623 ret
= chown(client_unix_sock_path
, 0, gid
);
3625 ERR("Unable to set group on %s", client_unix_sock_path
);
3629 /* kconsumer error socket path */
3630 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3632 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3636 /* 64-bit ustconsumer error socket path */
3637 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3639 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3643 /* 32-bit ustconsumer compat32 error socket path */
3644 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3646 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3650 DBG("All permissions are set");
3657 * Create the lttng run directory needed for all global sockets and pipe.
3659 static int create_lttng_rundir(const char *rundir
)
3663 DBG3("Creating LTTng run directory: %s", rundir
);
3665 ret
= mkdir(rundir
, S_IRWXU
);
3667 if (errno
!= EEXIST
) {
3668 ERR("Unable to create %s", rundir
);
3680 * Setup sockets and directory needed by the kconsumerd communication with the
3683 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3687 char path
[PATH_MAX
];
3689 switch (consumer_data
->type
) {
3690 case LTTNG_CONSUMER_KERNEL
:
3691 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3693 case LTTNG_CONSUMER64_UST
:
3694 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3696 case LTTNG_CONSUMER32_UST
:
3697 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3700 ERR("Consumer type unknown");
3705 DBG2("Creating consumer directory: %s", path
);
3707 ret
= mkdir(path
, S_IRWXU
);
3709 if (errno
!= EEXIST
) {
3711 ERR("Failed to create %s", path
);
3717 /* Create the kconsumerd error unix socket */
3718 consumer_data
->err_sock
=
3719 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3720 if (consumer_data
->err_sock
< 0) {
3721 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3726 /* File permission MUST be 660 */
3727 ret
= chmod(consumer_data
->err_unix_sock_path
,
3728 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3730 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3740 * Signal handler for the daemon
3742 * Simply stop all worker threads, leaving main() return gracefully after
3743 * joining all threads and calling cleanup().
3745 static void sighandler(int sig
)
3749 DBG("SIGPIPE caught");
3752 DBG("SIGINT caught");
3756 DBG("SIGTERM caught");
3765 * Setup signal handler for :
3766 * SIGINT, SIGTERM, SIGPIPE
3768 static int set_signal_handler(void)
3771 struct sigaction sa
;
3774 if ((ret
= sigemptyset(&sigset
)) < 0) {
3775 PERROR("sigemptyset");
3779 sa
.sa_handler
= sighandler
;
3780 sa
.sa_mask
= sigset
;
3782 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3783 PERROR("sigaction");
3787 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3788 PERROR("sigaction");
3792 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3793 PERROR("sigaction");
3797 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3803 * Set open files limit to unlimited. This daemon can open a large number of
3804 * file descriptors in order to consumer multiple kernel traces.
3806 static void set_ulimit(void)
3811 /* The kernel does not allowed an infinite limit for open files */
3812 lim
.rlim_cur
= 65535;
3813 lim
.rlim_max
= 65535;
3815 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
3817 PERROR("failed to set open files limit");
3824 int main(int argc
, char **argv
)
3828 const char *home_path
, *env_app_timeout
;
3830 init_kernel_workarounds();
3832 rcu_register_thread();
3834 setup_consumerd_path();
3836 /* Parse arguments */
3838 if ((ret
= parse_args(argc
, argv
) < 0)) {
3848 * child: setsid, close FD 0, 1, 2, chdir /
3849 * parent: exit (if fork is successful)
3857 * We are in the child. Make sure all other file
3858 * descriptors are closed, in case we are called with
3859 * more opened file descriptors than the standard ones.
3861 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3866 /* Create thread quit pipe */
3867 if ((ret
= init_thread_quit_pipe()) < 0) {
3871 /* Check if daemon is UID = 0 */
3872 is_root
= !getuid();
3875 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
3877 /* Create global run dir with root access */
3878 ret
= create_lttng_rundir(rundir
);
3883 if (strlen(apps_unix_sock_path
) == 0) {
3884 snprintf(apps_unix_sock_path
, PATH_MAX
,
3885 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
3888 if (strlen(client_unix_sock_path
) == 0) {
3889 snprintf(client_unix_sock_path
, PATH_MAX
,
3890 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
3893 /* Set global SHM for ust */
3894 if (strlen(wait_shm_path
) == 0) {
3895 snprintf(wait_shm_path
, PATH_MAX
,
3896 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
3899 if (strlen(health_unix_sock_path
) == 0) {
3900 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3901 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
3904 /* Setup kernel consumerd path */
3905 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
3906 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
3907 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
3908 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
3910 DBG2("Kernel consumer err path: %s",
3911 kconsumer_data
.err_unix_sock_path
);
3912 DBG2("Kernel consumer cmd path: %s",
3913 kconsumer_data
.cmd_unix_sock_path
);
3915 home_path
= get_home_dir();
3916 if (home_path
== NULL
) {
3917 /* TODO: Add --socket PATH option */
3918 ERR("Can't get HOME directory for sockets creation.");
3924 * Create rundir from home path. This will create something like
3927 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
3933 ret
= create_lttng_rundir(rundir
);
3938 if (strlen(apps_unix_sock_path
) == 0) {
3939 snprintf(apps_unix_sock_path
, PATH_MAX
,
3940 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
3943 /* Set the cli tool unix socket path */
3944 if (strlen(client_unix_sock_path
) == 0) {
3945 snprintf(client_unix_sock_path
, PATH_MAX
,
3946 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
3949 /* Set global SHM for ust */
3950 if (strlen(wait_shm_path
) == 0) {
3951 snprintf(wait_shm_path
, PATH_MAX
,
3952 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, geteuid());
3955 /* Set health check Unix path */
3956 if (strlen(health_unix_sock_path
) == 0) {
3957 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3958 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
3962 /* Set consumer initial state */
3963 kernel_consumerd_state
= CONSUMER_STOPPED
;
3964 ust_consumerd_state
= CONSUMER_STOPPED
;
3966 DBG("Client socket path %s", client_unix_sock_path
);
3967 DBG("Application socket path %s", apps_unix_sock_path
);
3968 DBG("LTTng run directory path: %s", rundir
);
3970 /* 32 bits consumerd path setup */
3971 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
3972 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
3973 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
3974 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
3976 DBG2("UST consumer 32 bits err path: %s",
3977 ustconsumer32_data
.err_unix_sock_path
);
3978 DBG2("UST consumer 32 bits cmd path: %s",
3979 ustconsumer32_data
.cmd_unix_sock_path
);
3981 /* 64 bits consumerd path setup */
3982 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
3983 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
3984 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
3985 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
3987 DBG2("UST consumer 64 bits err path: %s",
3988 ustconsumer64_data
.err_unix_sock_path
);
3989 DBG2("UST consumer 64 bits cmd path: %s",
3990 ustconsumer64_data
.cmd_unix_sock_path
);
3993 * See if daemon already exist.
3995 if ((ret
= check_existing_daemon()) < 0) {
3996 ERR("Already running daemon.\n");
3998 * We do not goto exit because we must not cleanup()
3999 * because a daemon is already running.
4005 * Init UST app hash table. Alloc hash table before this point since
4006 * cleanup() can get called after that point.
4010 /* After this point, we can safely call cleanup() with "goto exit" */
4013 * These actions must be executed as root. We do that *after* setting up
4014 * the sockets path because we MUST make the check for another daemon using
4015 * those paths *before* trying to set the kernel consumer sockets and init
4019 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4024 /* Setup kernel tracer */
4025 if (!opt_no_kernel
) {
4026 init_kernel_tracer();
4029 /* Set ulimit for open files */
4032 /* init lttng_fd tracking must be done after set_ulimit. */
4035 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4040 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4045 if ((ret
= set_signal_handler()) < 0) {
4049 /* Setup the needed unix socket */
4050 if ((ret
= init_daemon_socket()) < 0) {
4054 /* Set credentials to socket */
4055 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4059 /* Get parent pid if -S, --sig-parent is specified. */
4060 if (opt_sig_parent
) {
4064 /* Setup the kernel pipe for waking up the kernel thread */
4065 if (is_root
&& !opt_no_kernel
) {
4066 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4071 /* Setup the thread apps communication pipe. */
4072 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4076 /* Init UST command queue. */
4077 cds_wfq_init(&ust_cmd_queue
.queue
);
4080 * Get session list pointer. This pointer MUST NOT be free(). This list is
4081 * statically declared in session.c
4083 session_list_ptr
= session_get_list();
4085 /* Set up max poll set size */
4086 lttng_poll_set_max_size();
4090 /* Init all health thread counters. */
4091 health_init(&health_thread_cmd
);
4092 health_init(&health_thread_kernel
);
4093 health_init(&health_thread_app_manage
);
4094 health_init(&health_thread_app_reg
);
4097 * Init health counters of the consumer thread. We do a quick hack here to
4098 * the state of the consumer health is fine even if the thread is not
4099 * started. Once the thread starts, the health state is updated with a poll
4100 * value to set a health code path. This is simply to ease our life and has
4101 * no cost what so ever.
4103 health_init(&kconsumer_data
.health
);
4104 health_poll_update(&kconsumer_data
.health
);
4105 health_init(&ustconsumer32_data
.health
);
4106 health_poll_update(&ustconsumer32_data
.health
);
4107 health_init(&ustconsumer64_data
.health
);
4108 health_poll_update(&ustconsumer64_data
.health
);
4110 /* Check for the application socket timeout env variable. */
4111 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4112 if (env_app_timeout
) {
4113 app_socket_timeout
= atoi(env_app_timeout
);
4115 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4118 /* Create thread to manage the client socket */
4119 ret
= pthread_create(&health_thread
, NULL
,
4120 thread_manage_health
, (void *) NULL
);
4122 PERROR("pthread_create health");
4126 /* Create thread to manage the client socket */
4127 ret
= pthread_create(&client_thread
, NULL
,
4128 thread_manage_clients
, (void *) NULL
);
4130 PERROR("pthread_create clients");
4134 /* Create thread to dispatch registration */
4135 ret
= pthread_create(&dispatch_thread
, NULL
,
4136 thread_dispatch_ust_registration
, (void *) NULL
);
4138 PERROR("pthread_create dispatch");
4142 /* Create thread to manage application registration. */
4143 ret
= pthread_create(®_apps_thread
, NULL
,
4144 thread_registration_apps
, (void *) NULL
);
4146 PERROR("pthread_create registration");
4150 /* Create thread to manage application socket */
4151 ret
= pthread_create(&apps_thread
, NULL
,
4152 thread_manage_apps
, (void *) NULL
);
4154 PERROR("pthread_create apps");
4158 /* Don't start this thread if kernel tracing is not requested nor root */
4159 if (is_root
&& !opt_no_kernel
) {
4160 /* Create kernel thread to manage kernel event */
4161 ret
= pthread_create(&kernel_thread
, NULL
,
4162 thread_manage_kernel
, (void *) NULL
);
4164 PERROR("pthread_create kernel");
4168 ret
= pthread_join(kernel_thread
, &status
);
4170 PERROR("pthread_join");
4171 goto error
; /* join error, exit without cleanup */
4176 ret
= pthread_join(apps_thread
, &status
);
4178 PERROR("pthread_join");
4179 goto error
; /* join error, exit without cleanup */
4183 ret
= pthread_join(reg_apps_thread
, &status
);
4185 PERROR("pthread_join");
4186 goto error
; /* join error, exit without cleanup */
4190 ret
= pthread_join(dispatch_thread
, &status
);
4192 PERROR("pthread_join");
4193 goto error
; /* join error, exit without cleanup */
4197 ret
= pthread_join(client_thread
, &status
);
4199 PERROR("pthread_join");
4200 goto error
; /* join error, exit without cleanup */
4203 ret
= join_consumer_thread(&kconsumer_data
);
4205 PERROR("join_consumer");
4206 goto error
; /* join error, exit without cleanup */
4209 ret
= join_consumer_thread(&ustconsumer32_data
);
4211 PERROR("join_consumer ust32");
4212 goto error
; /* join error, exit without cleanup */
4215 ret
= join_consumer_thread(&ustconsumer64_data
);
4217 PERROR("join_consumer ust64");
4218 goto error
; /* join error, exit without cleanup */
4222 ret
= pthread_join(health_thread
, &status
);
4224 PERROR("pthread_join health thread");
4225 goto error
; /* join error, exit without cleanup */
4231 * cleanup() is called when no other thread is running.
4233 rcu_thread_online();
4235 rcu_thread_offline();
4236 rcu_unregister_thread();