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/socket.h>
41 #include <common/defaults.h>
42 #include <common/kernel-consumer/kernel-consumer.h>
43 #include <common/futex.h>
44 #include <common/relayd/relayd.h>
45 #include <common/utils.h>
47 #include "lttng-sessiond.h"
54 #include "kernel-consumer.h"
58 #include "ust-consumer.h"
62 #include "testpoint.h"
63 #include "ust-thread.h"
65 #define CONSUMERD_FILE "lttng-consumerd"
68 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
71 const char *opt_tracing_group
;
72 static const char *opt_pidfile
;
73 static int opt_sig_parent
;
74 static int opt_verbose_consumer
;
75 static int opt_daemon
;
76 static int opt_no_kernel
;
77 static int is_root
; /* Set to 1 if the daemon is running as root */
78 static pid_t ppid
; /* Parent PID for --sig-parent option */
82 * Consumer daemon specific control data. Every value not initialized here is
83 * set to 0 by the static definition.
85 static struct consumer_data kconsumer_data
= {
86 .type
= LTTNG_CONSUMER_KERNEL
,
87 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
88 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
91 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
92 .lock
= PTHREAD_MUTEX_INITIALIZER
,
93 .cond
= PTHREAD_COND_INITIALIZER
,
94 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
96 static struct consumer_data ustconsumer64_data
= {
97 .type
= LTTNG_CONSUMER64_UST
,
98 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
99 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
102 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
103 .lock
= PTHREAD_MUTEX_INITIALIZER
,
104 .cond
= PTHREAD_COND_INITIALIZER
,
105 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 static struct consumer_data ustconsumer32_data
= {
108 .type
= LTTNG_CONSUMER32_UST
,
109 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
110 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
113 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
114 .lock
= PTHREAD_MUTEX_INITIALIZER
,
115 .cond
= PTHREAD_COND_INITIALIZER
,
116 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 /* Shared between threads */
120 static int dispatch_thread_exit
;
122 /* Global application Unix socket path */
123 static char apps_unix_sock_path
[PATH_MAX
];
124 /* Global client Unix socket path */
125 static char client_unix_sock_path
[PATH_MAX
];
126 /* global wait shm path for UST */
127 static char wait_shm_path
[PATH_MAX
];
128 /* Global health check unix path */
129 static char health_unix_sock_path
[PATH_MAX
];
131 /* Sockets and FDs */
132 static int client_sock
= -1;
133 static int apps_sock
= -1;
134 int kernel_tracer_fd
= -1;
135 static int kernel_poll_pipe
[2] = { -1, -1 };
138 * Quit pipe for all threads. This permits a single cancellation point
139 * for all threads when receiving an event on the pipe.
141 static int thread_quit_pipe
[2] = { -1, -1 };
144 * This pipe is used to inform the thread managing application communication
145 * that a command is queued and ready to be processed.
147 static int apps_cmd_pipe
[2] = { -1, -1 };
149 int apps_cmd_notify_pipe
[2] = { -1, -1 };
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread
;
153 static pthread_t apps_notify_thread
;
154 static pthread_t reg_apps_thread
;
155 static pthread_t client_thread
;
156 static pthread_t kernel_thread
;
157 static pthread_t dispatch_thread
;
158 static pthread_t health_thread
;
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
167 static struct ust_cmd_queue ust_cmd_queue
;
170 * Pointer initialized before thread creation.
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
177 * session_lock_list and session_unlock_list for lock acquisition.
179 static struct ltt_session_list
*session_list_ptr
;
181 int ust_consumerd64_fd
= -1;
182 int ust_consumerd32_fd
= -1;
184 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
185 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
186 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
187 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
189 static const char *module_proc_lttng
= "/proc/lttng";
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
195 enum consumerd_state
{
196 CONSUMER_STARTED
= 1,
197 CONSUMER_STOPPED
= 2,
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
207 * The following example shows a possible race condition of this scheme:
209 * consumer thread error happens
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
222 static enum consumerd_state ust_consumerd_state
;
223 static enum consumerd_state kernel_consumerd_state
;
226 * Socket timeout for receiving and sending in seconds.
228 static int app_socket_timeout
;
231 void setup_consumerd_path(void)
233 const char *bin
, *libdir
;
236 * Allow INSTALL_BIN_PATH to be used as a target path for the
237 * native architecture size consumer if CONFIG_CONSUMER*_PATH
238 * has not been defined.
240 #if (CAA_BITS_PER_LONG == 32)
241 if (!consumerd32_bin
[0]) {
242 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
244 if (!consumerd32_libdir
[0]) {
245 consumerd32_libdir
= INSTALL_LIB_PATH
;
247 #elif (CAA_BITS_PER_LONG == 64)
248 if (!consumerd64_bin
[0]) {
249 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
251 if (!consumerd64_libdir
[0]) {
252 consumerd64_libdir
= INSTALL_LIB_PATH
;
255 #error "Unknown bitness"
259 * runtime env. var. overrides the build default.
261 bin
= getenv("LTTNG_CONSUMERD32_BIN");
263 consumerd32_bin
= bin
;
265 bin
= getenv("LTTNG_CONSUMERD64_BIN");
267 consumerd64_bin
= bin
;
269 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
271 consumerd32_libdir
= libdir
;
273 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
275 consumerd64_libdir
= libdir
;
280 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
282 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
288 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
294 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
306 * Check if the thread quit pipe was triggered.
308 * Return 1 if it was triggered else 0;
310 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
312 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
320 * Return group ID of the tracing group or -1 if not found.
322 static gid_t
allowed_group(void)
326 if (opt_tracing_group
) {
327 grp
= getgrnam(opt_tracing_group
);
329 grp
= getgrnam(default_tracing_group
);
339 * Init thread quit pipe.
341 * Return -1 on error or 0 if all pipes are created.
343 static int init_thread_quit_pipe(void)
347 ret
= pipe(thread_quit_pipe
);
349 PERROR("thread quit pipe");
353 for (i
= 0; i
< 2; i
++) {
354 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
366 * Stop all threads by closing the thread quit pipe.
368 static void stop_threads(void)
372 /* Stopping all threads */
373 DBG("Terminating all threads");
374 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
376 ERR("write error on thread quit pipe");
379 /* Dispatch thread */
380 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
381 futex_nto1_wake(&ust_cmd_queue
.futex
);
387 static void cleanup(void)
391 struct ltt_session
*sess
, *stmp
;
395 /* First thing first, stop all threads */
396 utils_close_pipe(thread_quit_pipe
);
399 * If opt_pidfile is undefined, the default file will be wiped when
400 * removing the rundir.
403 ret
= remove(opt_pidfile
);
405 PERROR("remove pidfile %s", opt_pidfile
);
409 DBG("Removing %s directory", rundir
);
410 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
412 ERR("asprintf failed. Something is really wrong!");
415 /* Remove lttng run directory */
418 ERR("Unable to clean %s", rundir
);
423 DBG("Cleaning up all sessions");
425 /* Destroy session list mutex */
426 if (session_list_ptr
!= NULL
) {
427 pthread_mutex_destroy(&session_list_ptr
->lock
);
429 /* Cleanup ALL session */
430 cds_list_for_each_entry_safe(sess
, stmp
,
431 &session_list_ptr
->head
, list
) {
432 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
436 DBG("Closing all UST sockets");
437 ust_app_clean_list();
439 if (is_root
&& !opt_no_kernel
) {
440 DBG2("Closing kernel fd");
441 if (kernel_tracer_fd
>= 0) {
442 ret
= close(kernel_tracer_fd
);
447 DBG("Unloading kernel modules");
448 modprobe_remove_lttng_all();
452 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
453 "Matthew, BEET driven development works!%c[%dm",
454 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
459 * Send data on a unix socket using the liblttsessiondcomm API.
461 * Return lttcomm error code.
463 static int send_unix_sock(int sock
, void *buf
, size_t len
)
465 /* Check valid length */
470 return lttcomm_send_unix_sock(sock
, buf
, len
);
474 * Free memory of a command context structure.
476 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
478 DBG("Clean command context structure");
480 if ((*cmd_ctx
)->llm
) {
481 free((*cmd_ctx
)->llm
);
483 if ((*cmd_ctx
)->lsm
) {
484 free((*cmd_ctx
)->lsm
);
492 * Notify UST applications using the shm mmap futex.
494 static int notify_ust_apps(int active
)
498 DBG("Notifying applications of session daemon state: %d", active
);
500 /* See shm.c for this call implying mmap, shm and futex calls */
501 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
502 if (wait_shm_mmap
== NULL
) {
506 /* Wake waiting process */
507 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
509 /* Apps notified successfully */
517 * Setup the outgoing data buffer for the response (llm) by allocating the
518 * right amount of memory and copying the original information from the lsm
521 * Return total size of the buffer pointed by buf.
523 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
529 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
530 if (cmd_ctx
->llm
== NULL
) {
536 /* Copy common data */
537 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
538 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
540 cmd_ctx
->llm
->data_size
= size
;
541 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
550 * Update the kernel poll set of all channel fd available over all tracing
551 * session. Add the wakeup pipe at the end of the set.
553 static int update_kernel_poll(struct lttng_poll_event
*events
)
556 struct ltt_session
*session
;
557 struct ltt_kernel_channel
*channel
;
559 DBG("Updating kernel poll set");
562 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
563 session_lock(session
);
564 if (session
->kernel_session
== NULL
) {
565 session_unlock(session
);
569 cds_list_for_each_entry(channel
,
570 &session
->kernel_session
->channel_list
.head
, list
) {
571 /* Add channel fd to the kernel poll set */
572 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
574 session_unlock(session
);
577 DBG("Channel fd %d added to kernel set", channel
->fd
);
579 session_unlock(session
);
581 session_unlock_list();
586 session_unlock_list();
591 * Find the channel fd from 'fd' over all tracing session. When found, check
592 * for new channel stream and send those stream fds to the kernel consumer.
594 * Useful for CPU hotplug feature.
596 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
599 struct ltt_session
*session
;
600 struct ltt_kernel_session
*ksess
;
601 struct ltt_kernel_channel
*channel
;
603 DBG("Updating kernel streams for channel fd %d", fd
);
606 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
607 session_lock(session
);
608 if (session
->kernel_session
== NULL
) {
609 session_unlock(session
);
612 ksess
= session
->kernel_session
;
614 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
615 if (channel
->fd
== fd
) {
616 DBG("Channel found, updating kernel streams");
617 ret
= kernel_open_channel_stream(channel
);
623 * Have we already sent fds to the consumer? If yes, it means
624 * that tracing is started so it is safe to send our updated
627 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
628 struct lttng_ht_iter iter
;
629 struct consumer_socket
*socket
;
632 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
633 &iter
.iter
, socket
, node
.node
) {
634 /* Code flow error */
635 assert(socket
->fd
>= 0);
637 pthread_mutex_lock(socket
->lock
);
638 ret
= kernel_consumer_send_channel_stream(socket
,
640 pthread_mutex_unlock(socket
->lock
);
651 session_unlock(session
);
653 session_unlock_list();
657 session_unlock(session
);
658 session_unlock_list();
663 * For each tracing session, update newly registered apps. The session list
664 * lock MUST be acquired before calling this.
666 static void update_ust_app(int app_sock
)
668 struct ltt_session
*sess
, *stmp
;
670 /* For all tracing session(s) */
671 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
673 if (sess
->ust_session
) {
674 ust_app_global_update(sess
->ust_session
, app_sock
);
676 session_unlock(sess
);
681 * This thread manage event coming from the kernel.
683 * Features supported in this thread:
686 static void *thread_manage_kernel(void *data
)
688 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
689 uint32_t revents
, nb_fd
;
691 struct lttng_poll_event events
;
693 DBG("[thread] Thread manage kernel started");
695 health_register(HEALTH_TYPE_KERNEL
);
698 * This first step of the while is to clean this structure which could free
699 * non NULL pointers so zero it before the loop.
701 memset(&events
, 0, sizeof(events
));
703 if (testpoint(thread_manage_kernel
)) {
704 goto error_testpoint
;
707 health_code_update();
709 if (testpoint(thread_manage_kernel_before_loop
)) {
710 goto error_testpoint
;
714 health_code_update();
716 if (update_poll_flag
== 1) {
717 /* Clean events object. We are about to populate it again. */
718 lttng_poll_clean(&events
);
720 ret
= sessiond_set_thread_pollset(&events
, 2);
722 goto error_poll_create
;
725 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
730 /* This will add the available kernel channel if any. */
731 ret
= update_kernel_poll(&events
);
735 update_poll_flag
= 0;
738 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
740 /* Poll infinite value of time */
743 ret
= lttng_poll_wait(&events
, -1);
747 * Restart interrupted system call.
749 if (errno
== EINTR
) {
753 } else if (ret
== 0) {
754 /* Should not happen since timeout is infinite */
755 ERR("Return value of poll is 0 with an infinite timeout.\n"
756 "This should not have happened! Continuing...");
762 for (i
= 0; i
< nb_fd
; i
++) {
763 /* Fetch once the poll data */
764 revents
= LTTNG_POLL_GETEV(&events
, i
);
765 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
767 health_code_update();
769 /* Thread quit pipe has been closed. Killing thread. */
770 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
776 /* Check for data on kernel pipe */
777 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
779 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
780 } while (ret
< 0 && errno
== EINTR
);
782 * Ret value is useless here, if this pipe gets any actions an
783 * update is required anyway.
785 update_poll_flag
= 1;
789 * New CPU detected by the kernel. Adding kernel stream to
790 * kernel session and updating the kernel consumer
792 if (revents
& LPOLLIN
) {
793 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
799 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
800 * and unregister kernel stream at this point.
809 lttng_poll_clean(&events
);
812 utils_close_pipe(kernel_poll_pipe
);
813 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
816 ERR("Health error occurred in %s", __func__
);
817 WARN("Kernel thread died unexpectedly. "
818 "Kernel tracing can continue but CPU hotplug is disabled.");
821 DBG("Kernel thread dying");
826 * Signal pthread condition of the consumer data that the thread.
828 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
830 pthread_mutex_lock(&data
->cond_mutex
);
833 * The state is set before signaling. It can be any value, it's the waiter
834 * job to correctly interpret this condition variable associated to the
835 * consumer pthread_cond.
837 * A value of 0 means that the corresponding thread of the consumer data
838 * was not started. 1 indicates that the thread has started and is ready
839 * for action. A negative value means that there was an error during the
842 data
->consumer_thread_is_ready
= state
;
843 (void) pthread_cond_signal(&data
->cond
);
845 pthread_mutex_unlock(&data
->cond_mutex
);
849 * This thread manage the consumer error sent back to the session daemon.
851 static void *thread_manage_consumer(void *data
)
853 int sock
= -1, i
, ret
, pollfd
, err
= -1;
854 uint32_t revents
, nb_fd
;
855 enum lttcomm_return_code code
;
856 struct lttng_poll_event events
;
857 struct consumer_data
*consumer_data
= data
;
859 DBG("[thread] Manage consumer started");
861 health_register(HEALTH_TYPE_CONSUMER
);
863 health_code_update();
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
= sessiond_set_thread_pollset(&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 health_code_update();
886 /* Inifinite blocking call, waiting for transmission */
890 if (testpoint(thread_manage_consumer
)) {
894 ret
= lttng_poll_wait(&events
, -1);
898 * Restart interrupted system call.
900 if (errno
== EINTR
) {
908 for (i
= 0; i
< nb_fd
; i
++) {
909 /* Fetch once the poll data */
910 revents
= LTTNG_POLL_GETEV(&events
, i
);
911 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
913 health_code_update();
915 /* Thread quit pipe has been closed. Killing thread. */
916 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
922 /* Event on the registration socket */
923 if (pollfd
== consumer_data
->err_sock
) {
924 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
925 ERR("consumer err socket poll error");
931 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
937 * Set the CLOEXEC flag. Return code is useless because either way, the
940 (void) utils_set_fd_cloexec(sock
);
942 health_code_update();
944 DBG2("Receiving code from consumer err_sock");
946 /* Getting status code from kconsumerd */
947 ret
= lttcomm_recv_unix_sock(sock
, &code
,
948 sizeof(enum lttcomm_return_code
));
953 health_code_update();
955 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
956 consumer_data
->cmd_sock
=
957 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
958 if (consumer_data
->cmd_sock
< 0) {
959 /* On error, signal condition and quit. */
960 signal_consumer_condition(consumer_data
, -1);
961 PERROR("consumer connect");
964 signal_consumer_condition(consumer_data
, 1);
965 DBG("Consumer command socket ready");
967 ERR("consumer error when waiting for SOCK_READY : %s",
968 lttcomm_get_readable_code(-code
));
972 /* Remove the kconsumerd error sock since we've established a connexion */
973 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
978 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
983 health_code_update();
985 /* Inifinite blocking call, waiting for transmission */
988 ret
= lttng_poll_wait(&events
, -1);
992 * Restart interrupted system call.
994 if (errno
== EINTR
) {
1002 for (i
= 0; i
< nb_fd
; i
++) {
1003 /* Fetch once the poll data */
1004 revents
= LTTNG_POLL_GETEV(&events
, i
);
1005 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1007 health_code_update();
1009 /* Thread quit pipe has been closed. Killing thread. */
1010 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1016 /* Event on the kconsumerd socket */
1017 if (pollfd
== sock
) {
1018 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1019 ERR("consumer err socket second poll error");
1025 health_code_update();
1027 /* Wait for any kconsumerd error */
1028 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1029 sizeof(enum lttcomm_return_code
));
1031 ERR("consumer closed the command socket");
1035 ERR("consumer return code : %s", lttcomm_get_readable_code(-code
));
1039 /* Immediately set the consumerd state to stopped */
1040 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1041 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1042 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1043 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1044 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1046 /* Code flow error... */
1050 if (consumer_data
->err_sock
>= 0) {
1051 ret
= close(consumer_data
->err_sock
);
1056 if (consumer_data
->cmd_sock
>= 0) {
1057 ret
= close(consumer_data
->cmd_sock
);
1069 unlink(consumer_data
->err_unix_sock_path
);
1070 unlink(consumer_data
->cmd_unix_sock_path
);
1071 consumer_data
->pid
= 0;
1073 lttng_poll_clean(&events
);
1077 ERR("Health error occurred in %s", __func__
);
1079 health_unregister();
1080 DBG("consumer thread cleanup completed");
1086 * This thread manage application communication.
1088 static void *thread_manage_apps(void *data
)
1090 int i
, ret
, pollfd
, err
= -1;
1091 uint32_t revents
, nb_fd
;
1092 struct lttng_poll_event events
;
1094 DBG("[thread] Manage application started");
1096 rcu_register_thread();
1097 rcu_thread_online();
1099 health_register(HEALTH_TYPE_APP_MANAGE
);
1101 if (testpoint(thread_manage_apps
)) {
1102 goto error_testpoint
;
1105 health_code_update();
1107 ret
= sessiond_set_thread_pollset(&events
, 2);
1109 goto error_poll_create
;
1112 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1117 if (testpoint(thread_manage_apps_before_loop
)) {
1121 health_code_update();
1124 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1126 /* Inifinite blocking call, waiting for transmission */
1128 health_poll_entry();
1129 ret
= lttng_poll_wait(&events
, -1);
1133 * Restart interrupted system call.
1135 if (errno
== EINTR
) {
1143 for (i
= 0; i
< nb_fd
; i
++) {
1144 /* Fetch once the poll data */
1145 revents
= LTTNG_POLL_GETEV(&events
, i
);
1146 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1148 health_code_update();
1150 /* Thread quit pipe has been closed. Killing thread. */
1151 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1157 /* Inspect the apps cmd pipe */
1158 if (pollfd
== apps_cmd_pipe
[0]) {
1159 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1160 ERR("Apps command pipe error");
1162 } else if (revents
& LPOLLIN
) {
1167 ret
= read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1168 } while (ret
< 0 && errno
== EINTR
);
1169 if (ret
< 0 || ret
< sizeof(sock
)) {
1170 PERROR("read apps cmd pipe");
1174 health_code_update();
1177 * We only monitor the error events of the socket. This
1178 * thread does not handle any incoming data from UST
1181 ret
= lttng_poll_add(&events
, sock
,
1182 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1187 /* Set socket timeout for both receiving and ending */
1188 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1189 app_socket_timeout
);
1190 (void) lttcomm_setsockopt_snd_timeout(sock
,
1191 app_socket_timeout
);
1193 DBG("Apps with sock %d added to poll set", sock
);
1195 health_code_update();
1201 * At this point, we know that a registered application made
1202 * the event at poll_wait.
1204 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1205 /* Removing from the poll set */
1206 ret
= lttng_poll_del(&events
, pollfd
);
1211 /* Socket closed on remote end. */
1212 ust_app_unregister(pollfd
);
1217 health_code_update();
1223 lttng_poll_clean(&events
);
1226 utils_close_pipe(apps_cmd_pipe
);
1227 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1230 * We don't clean the UST app hash table here since already registered
1231 * applications can still be controlled so let them be until the session
1232 * daemon dies or the applications stop.
1237 ERR("Health error occurred in %s", __func__
);
1239 health_unregister();
1240 DBG("Application communication apps thread cleanup complete");
1241 rcu_thread_offline();
1242 rcu_unregister_thread();
1247 * Send a socket to a thread This is called from the dispatch UST registration
1248 * thread once all sockets are set for the application.
1250 * On success, return 0 else a negative value being the errno message of the
1253 static int send_socket_to_thread(int fd
, int sock
)
1257 /* Sockets MUST be set or else this should not have been called. */
1262 ret
= write(fd
, &sock
, sizeof(sock
));
1263 } while (ret
< 0 && errno
== EINTR
);
1264 if (ret
< 0 || ret
!= sizeof(sock
)) {
1265 PERROR("write apps pipe %d", fd
);
1272 /* All good. Don't send back the write positive ret value. */
1279 * Dispatch request from the registration threads to the application
1280 * communication thread.
1282 static void *thread_dispatch_ust_registration(void *data
)
1285 struct cds_wfq_node
*node
;
1286 struct ust_command
*ust_cmd
= NULL
;
1288 struct ust_app
*app
;
1289 struct cds_list_head head
;
1290 } *wait_node
= NULL
, *tmp_wait_node
;
1292 CDS_LIST_HEAD(wait_queue
);
1294 DBG("[thread] Dispatch UST command started");
1296 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1297 /* Atomically prepare the queue futex */
1298 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1301 struct ust_app
*app
= NULL
;
1303 /* Dequeue command for registration */
1304 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1306 DBG("Woken up but nothing in the UST command queue");
1307 /* Continue thread execution */
1311 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1313 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1314 " gid:%d sock:%d name:%s (version %d.%d)",
1315 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1316 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1317 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1318 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1320 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1321 wait_node
= zmalloc(sizeof(*wait_node
));
1323 PERROR("zmalloc wait_node dispatch");
1326 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1328 /* Create application object if socket is CMD. */
1329 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1331 if (!wait_node
->app
) {
1332 ret
= close(ust_cmd
->sock
);
1334 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1336 lttng_fd_put(1, LTTNG_FD_APPS
);
1341 * Add application to the wait queue so we can set the notify
1342 * socket before putting this object in the global ht.
1344 cds_list_add(&wait_node
->head
, &wait_queue
);
1347 * We have to continue here since we don't have the notify
1348 * socket and the application MUST be added to the hash table
1349 * only at that moment.
1354 * Look for the application in the local wait queue and set the
1355 * notify socket if found.
1357 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1358 &wait_queue
, head
) {
1359 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1360 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1361 cds_list_del(&wait_node
->head
);
1362 app
= wait_node
->app
;
1364 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1372 * @session_lock_list
1374 * Lock the global session list so from the register up to the
1375 * registration done message, no thread can see the application
1376 * and change its state.
1378 session_lock_list();
1382 * Add application to the global hash table. This needs to be
1383 * done before the update to the UST registry can locate the
1388 /* Set app version. This call will print an error if needed. */
1389 (void) ust_app_version(app
);
1391 /* Send notify socket through the notify pipe. */
1392 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1396 session_unlock_list();
1397 /* No notify thread, stop the UST tracing. */
1402 * Update newly registered application with the tracing
1403 * registry info already enabled information.
1405 update_ust_app(app
->sock
);
1408 * Don't care about return value. Let the manage apps threads
1409 * handle app unregistration upon socket close.
1411 (void) ust_app_register_done(app
->sock
);
1414 * Even if the application socket has been closed, send the app
1415 * to the thread and unregistration will take place at that
1418 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1421 session_unlock_list();
1422 /* No apps. thread, stop the UST tracing. */
1427 session_unlock_list();
1429 /* Application manager threads are not available. */
1430 ret
= close(ust_cmd
->sock
);
1432 PERROR("close ust_cmd sock");
1434 lttng_fd_put(1, LTTNG_FD_APPS
);
1437 } while (node
!= NULL
);
1439 /* Futex wait on queue. Blocking call on futex() */
1440 futex_nto1_wait(&ust_cmd_queue
.futex
);
1444 /* Clean up wait queue. */
1445 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1446 &wait_queue
, head
) {
1447 cds_list_del(&wait_node
->head
);
1451 DBG("Dispatch thread dying");
1456 * This thread manage application registration.
1458 static void *thread_registration_apps(void *data
)
1460 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1461 uint32_t revents
, nb_fd
;
1462 struct lttng_poll_event events
;
1464 * Get allocated in this thread, enqueued to a global queue, dequeued and
1465 * freed in the manage apps thread.
1467 struct ust_command
*ust_cmd
= NULL
;
1469 DBG("[thread] Manage application registration started");
1471 health_register(HEALTH_TYPE_APP_REG
);
1473 if (testpoint(thread_registration_apps
)) {
1474 goto error_testpoint
;
1477 ret
= lttcomm_listen_unix_sock(apps_sock
);
1483 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1484 * more will be added to this poll set.
1486 ret
= sessiond_set_thread_pollset(&events
, 2);
1488 goto error_create_poll
;
1491 /* Add the application registration socket */
1492 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1494 goto error_poll_add
;
1497 /* Notify all applications to register */
1498 ret
= notify_ust_apps(1);
1500 ERR("Failed to notify applications or create the wait shared memory.\n"
1501 "Execution continues but there might be problem for already\n"
1502 "running applications that wishes to register.");
1506 DBG("Accepting application registration");
1508 /* Inifinite blocking call, waiting for transmission */
1510 health_poll_entry();
1511 ret
= lttng_poll_wait(&events
, -1);
1515 * Restart interrupted system call.
1517 if (errno
== EINTR
) {
1525 for (i
= 0; i
< nb_fd
; i
++) {
1526 health_code_update();
1528 /* Fetch once the poll data */
1529 revents
= LTTNG_POLL_GETEV(&events
, i
);
1530 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1532 /* Thread quit pipe has been closed. Killing thread. */
1533 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1539 /* Event on the registration socket */
1540 if (pollfd
== apps_sock
) {
1541 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1542 ERR("Register apps socket poll error");
1544 } else if (revents
& LPOLLIN
) {
1545 sock
= lttcomm_accept_unix_sock(apps_sock
);
1551 * Set the CLOEXEC flag. Return code is useless because
1552 * either way, the show must go on.
1554 (void) utils_set_fd_cloexec(sock
);
1556 /* Create UST registration command for enqueuing */
1557 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1558 if (ust_cmd
== NULL
) {
1559 PERROR("ust command zmalloc");
1564 * Using message-based transmissions to ensure we don't
1565 * have to deal with partially received messages.
1567 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1569 ERR("Exhausted file descriptors allowed for applications.");
1579 health_code_update();
1580 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1583 /* Close socket of the application. */
1588 lttng_fd_put(LTTNG_FD_APPS
, 1);
1592 health_code_update();
1594 ust_cmd
->sock
= sock
;
1597 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1598 " gid:%d sock:%d name:%s (version %d.%d)",
1599 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1600 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1601 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1602 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1605 * Lock free enqueue the registration request. The red pill
1606 * has been taken! This apps will be part of the *system*.
1608 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1611 * Wake the registration queue futex. Implicit memory
1612 * barrier with the exchange in cds_wfq_enqueue.
1614 futex_nto1_wake(&ust_cmd_queue
.futex
);
1624 ERR("Health error occurred in %s", __func__
);
1627 /* Notify that the registration thread is gone */
1630 if (apps_sock
>= 0) {
1631 ret
= close(apps_sock
);
1641 lttng_fd_put(LTTNG_FD_APPS
, 1);
1643 unlink(apps_unix_sock_path
);
1646 lttng_poll_clean(&events
);
1650 DBG("UST Registration thread cleanup complete");
1651 health_unregister();
1657 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1658 * exec or it will fails.
1660 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1663 struct timespec timeout
;
1665 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1666 consumer_data
->consumer_thread_is_ready
= 0;
1668 /* Setup pthread condition */
1669 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1672 PERROR("pthread_condattr_init consumer data");
1677 * Set the monotonic clock in order to make sure we DO NOT jump in time
1678 * between the clock_gettime() call and the timedwait call. See bug #324
1679 * for a more details and how we noticed it.
1681 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1684 PERROR("pthread_condattr_setclock consumer data");
1688 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1691 PERROR("pthread_cond_init consumer data");
1695 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1698 PERROR("pthread_create consumer");
1703 /* We are about to wait on a pthread condition */
1704 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1706 /* Get time for sem_timedwait absolute timeout */
1707 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1709 * Set the timeout for the condition timed wait even if the clock gettime
1710 * call fails since we might loop on that call and we want to avoid to
1711 * increment the timeout too many times.
1713 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1716 * The following loop COULD be skipped in some conditions so this is why we
1717 * set ret to 0 in order to make sure at least one round of the loop is
1723 * Loop until the condition is reached or when a timeout is reached. Note
1724 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1725 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1726 * possible. This loop does not take any chances and works with both of
1729 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1730 if (clock_ret
< 0) {
1731 PERROR("clock_gettime spawn consumer");
1732 /* Infinite wait for the consumerd thread to be ready */
1733 ret
= pthread_cond_wait(&consumer_data
->cond
,
1734 &consumer_data
->cond_mutex
);
1736 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1737 &consumer_data
->cond_mutex
, &timeout
);
1741 /* Release the pthread condition */
1742 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1746 if (ret
== ETIMEDOUT
) {
1748 * Call has timed out so we kill the kconsumerd_thread and return
1751 ERR("Condition timed out. The consumer thread was never ready."
1753 ret
= pthread_cancel(consumer_data
->thread
);
1755 PERROR("pthread_cancel consumer thread");
1758 PERROR("pthread_cond_wait failed consumer thread");
1763 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1764 if (consumer_data
->pid
== 0) {
1765 ERR("Consumerd did not start");
1766 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1769 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1778 * Join consumer thread
1780 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1784 /* Consumer pid must be a real one. */
1785 if (consumer_data
->pid
> 0) {
1787 ret
= kill(consumer_data
->pid
, SIGTERM
);
1789 ERR("Error killing consumer daemon");
1792 return pthread_join(consumer_data
->thread
, &status
);
1799 * Fork and exec a consumer daemon (consumerd).
1801 * Return pid if successful else -1.
1803 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1807 const char *consumer_to_use
;
1808 const char *verbosity
;
1811 DBG("Spawning consumerd");
1818 if (opt_verbose_consumer
) {
1819 verbosity
= "--verbose";
1821 verbosity
= "--quiet";
1823 switch (consumer_data
->type
) {
1824 case LTTNG_CONSUMER_KERNEL
:
1826 * Find out which consumerd to execute. We will first try the
1827 * 64-bit path, then the sessiond's installation directory, and
1828 * fallback on the 32-bit one,
1830 DBG3("Looking for a kernel consumer at these locations:");
1831 DBG3(" 1) %s", consumerd64_bin
);
1832 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1833 DBG3(" 3) %s", consumerd32_bin
);
1834 if (stat(consumerd64_bin
, &st
) == 0) {
1835 DBG3("Found location #1");
1836 consumer_to_use
= consumerd64_bin
;
1837 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1838 DBG3("Found location #2");
1839 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1840 } else if (stat(consumerd32_bin
, &st
) == 0) {
1841 DBG3("Found location #3");
1842 consumer_to_use
= consumerd32_bin
;
1844 DBG("Could not find any valid consumerd executable");
1847 DBG("Using kernel consumer at: %s", consumer_to_use
);
1848 execl(consumer_to_use
,
1849 "lttng-consumerd", verbosity
, "-k",
1850 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1851 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1854 case LTTNG_CONSUMER64_UST
:
1856 char *tmpnew
= NULL
;
1858 if (consumerd64_libdir
[0] != '\0') {
1862 tmp
= getenv("LD_LIBRARY_PATH");
1866 tmplen
= strlen("LD_LIBRARY_PATH=")
1867 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1868 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1873 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1874 strcat(tmpnew
, consumerd64_libdir
);
1875 if (tmp
[0] != '\0') {
1876 strcat(tmpnew
, ":");
1877 strcat(tmpnew
, tmp
);
1879 ret
= putenv(tmpnew
);
1885 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1886 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1887 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1888 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1890 if (consumerd64_libdir
[0] != '\0') {
1898 case LTTNG_CONSUMER32_UST
:
1900 char *tmpnew
= NULL
;
1902 if (consumerd32_libdir
[0] != '\0') {
1906 tmp
= getenv("LD_LIBRARY_PATH");
1910 tmplen
= strlen("LD_LIBRARY_PATH=")
1911 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1912 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1917 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1918 strcat(tmpnew
, consumerd32_libdir
);
1919 if (tmp
[0] != '\0') {
1920 strcat(tmpnew
, ":");
1921 strcat(tmpnew
, tmp
);
1923 ret
= putenv(tmpnew
);
1929 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
1930 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
1931 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1932 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1934 if (consumerd32_libdir
[0] != '\0') {
1943 PERROR("unknown consumer type");
1947 PERROR("kernel start consumer exec");
1950 } else if (pid
> 0) {
1953 PERROR("start consumer fork");
1961 * Spawn the consumerd daemon and session daemon thread.
1963 static int start_consumerd(struct consumer_data
*consumer_data
)
1968 * Set the listen() state on the socket since there is a possible race
1969 * between the exec() of the consumer daemon and this call if place in the
1970 * consumer thread. See bug #366 for more details.
1972 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
1977 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1978 if (consumer_data
->pid
!= 0) {
1979 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1983 ret
= spawn_consumerd(consumer_data
);
1985 ERR("Spawning consumerd failed");
1986 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1990 /* Setting up the consumer_data pid */
1991 consumer_data
->pid
= ret
;
1992 DBG2("Consumer pid %d", consumer_data
->pid
);
1993 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1995 DBG2("Spawning consumer control thread");
1996 ret
= spawn_consumer_thread(consumer_data
);
1998 ERR("Fatal error spawning consumer control thread");
2006 /* Cleanup already created socket on error. */
2007 if (consumer_data
->err_sock
>= 0) {
2010 err
= close(consumer_data
->err_sock
);
2012 PERROR("close consumer data error socket");
2019 * Compute health status of each consumer. If one of them is zero (bad
2020 * state), we return 0.
2022 static int check_consumer_health(void)
2026 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2028 DBG3("Health consumer check %d", ret
);
2034 * Setup necessary data for kernel tracer action.
2036 static int init_kernel_tracer(void)
2040 /* Modprobe lttng kernel modules */
2041 ret
= modprobe_lttng_control();
2046 /* Open debugfs lttng */
2047 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2048 if (kernel_tracer_fd
< 0) {
2049 DBG("Failed to open %s", module_proc_lttng
);
2054 /* Validate kernel version */
2055 ret
= kernel_validate_version(kernel_tracer_fd
);
2060 ret
= modprobe_lttng_data();
2065 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2069 modprobe_remove_lttng_control();
2070 ret
= close(kernel_tracer_fd
);
2074 kernel_tracer_fd
= -1;
2075 return LTTNG_ERR_KERN_VERSION
;
2078 ret
= close(kernel_tracer_fd
);
2084 modprobe_remove_lttng_control();
2087 WARN("No kernel tracer available");
2088 kernel_tracer_fd
= -1;
2090 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2092 return LTTNG_ERR_KERN_NA
;
2098 * Copy consumer output from the tracing session to the domain session. The
2099 * function also applies the right modification on a per domain basis for the
2100 * trace files destination directory.
2102 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2105 const char *dir_name
;
2106 struct consumer_output
*consumer
;
2109 assert(session
->consumer
);
2112 case LTTNG_DOMAIN_KERNEL
:
2113 DBG3("Copying tracing session consumer output in kernel session");
2115 * XXX: We should audit the session creation and what this function
2116 * does "extra" in order to avoid a destroy since this function is used
2117 * in the domain session creation (kernel and ust) only. Same for UST
2120 if (session
->kernel_session
->consumer
) {
2121 consumer_destroy_output(session
->kernel_session
->consumer
);
2123 session
->kernel_session
->consumer
=
2124 consumer_copy_output(session
->consumer
);
2125 /* Ease our life a bit for the next part */
2126 consumer
= session
->kernel_session
->consumer
;
2127 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2129 case LTTNG_DOMAIN_UST
:
2130 DBG3("Copying tracing session consumer output in UST session");
2131 if (session
->ust_session
->consumer
) {
2132 consumer_destroy_output(session
->ust_session
->consumer
);
2134 session
->ust_session
->consumer
=
2135 consumer_copy_output(session
->consumer
);
2136 /* Ease our life a bit for the next part */
2137 consumer
= session
->ust_session
->consumer
;
2138 dir_name
= DEFAULT_UST_TRACE_DIR
;
2141 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2145 /* Append correct directory to subdir */
2146 strncat(consumer
->subdir
, dir_name
,
2147 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2148 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2157 * Create an UST session and add it to the session ust list.
2159 static int create_ust_session(struct ltt_session
*session
,
2160 struct lttng_domain
*domain
)
2163 struct ltt_ust_session
*lus
= NULL
;
2167 assert(session
->consumer
);
2169 switch (domain
->type
) {
2170 case LTTNG_DOMAIN_UST
:
2173 ERR("Unknown UST domain on create session %d", domain
->type
);
2174 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2178 DBG("Creating UST session");
2180 lus
= trace_ust_create_session(session
->path
, session
->id
);
2182 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2186 lus
->uid
= session
->uid
;
2187 lus
->gid
= session
->gid
;
2188 session
->ust_session
= lus
;
2190 /* Copy session output to the newly created UST session */
2191 ret
= copy_session_consumer(domain
->type
, session
);
2192 if (ret
!= LTTNG_OK
) {
2200 session
->ust_session
= NULL
;
2205 * Create a kernel tracer session then create the default channel.
2207 static int create_kernel_session(struct ltt_session
*session
)
2211 DBG("Creating kernel session");
2213 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2215 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2219 /* Code flow safety */
2220 assert(session
->kernel_session
);
2222 /* Copy session output to the newly created Kernel session */
2223 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2224 if (ret
!= LTTNG_OK
) {
2228 /* Create directory(ies) on local filesystem. */
2229 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2230 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2231 ret
= run_as_mkdir_recursive(
2232 session
->kernel_session
->consumer
->dst
.trace_path
,
2233 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2235 if (ret
!= -EEXIST
) {
2236 ERR("Trace directory creation error");
2242 session
->kernel_session
->uid
= session
->uid
;
2243 session
->kernel_session
->gid
= session
->gid
;
2248 trace_kernel_destroy_session(session
->kernel_session
);
2249 session
->kernel_session
= NULL
;
2254 * Count number of session permitted by uid/gid.
2256 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2259 struct ltt_session
*session
;
2261 DBG("Counting number of available session for UID %d GID %d",
2263 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2265 * Only list the sessions the user can control.
2267 if (!session_access_ok(session
, uid
, gid
)) {
2276 * Process the command requested by the lttng client within the command
2277 * context structure. This function make sure that the return structure (llm)
2278 * is set and ready for transmission before returning.
2280 * Return any error encountered or 0 for success.
2282 * "sock" is only used for special-case var. len data.
2284 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2288 int need_tracing_session
= 1;
2291 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2295 switch (cmd_ctx
->lsm
->cmd_type
) {
2296 case LTTNG_CREATE_SESSION
:
2297 case LTTNG_DESTROY_SESSION
:
2298 case LTTNG_LIST_SESSIONS
:
2299 case LTTNG_LIST_DOMAINS
:
2300 case LTTNG_START_TRACE
:
2301 case LTTNG_STOP_TRACE
:
2302 case LTTNG_DATA_PENDING
:
2309 if (opt_no_kernel
&& need_domain
2310 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2312 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2314 ret
= LTTNG_ERR_KERN_NA
;
2319 /* Deny register consumer if we already have a spawned consumer. */
2320 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2321 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2322 if (kconsumer_data
.pid
> 0) {
2323 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2324 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2327 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2331 * Check for command that don't needs to allocate a returned payload. We do
2332 * this here so we don't have to make the call for no payload at each
2335 switch(cmd_ctx
->lsm
->cmd_type
) {
2336 case LTTNG_LIST_SESSIONS
:
2337 case LTTNG_LIST_TRACEPOINTS
:
2338 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2339 case LTTNG_LIST_DOMAINS
:
2340 case LTTNG_LIST_CHANNELS
:
2341 case LTTNG_LIST_EVENTS
:
2344 /* Setup lttng message with no payload */
2345 ret
= setup_lttng_msg(cmd_ctx
, 0);
2347 /* This label does not try to unlock the session */
2348 goto init_setup_error
;
2352 /* Commands that DO NOT need a session. */
2353 switch (cmd_ctx
->lsm
->cmd_type
) {
2354 case LTTNG_CREATE_SESSION
:
2355 case LTTNG_CALIBRATE
:
2356 case LTTNG_LIST_SESSIONS
:
2357 case LTTNG_LIST_TRACEPOINTS
:
2358 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2359 need_tracing_session
= 0;
2362 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2364 * We keep the session list lock across _all_ commands
2365 * for now, because the per-session lock does not
2366 * handle teardown properly.
2368 session_lock_list();
2369 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2370 if (cmd_ctx
->session
== NULL
) {
2371 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2372 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2374 /* If no session name specified */
2375 ret
= LTTNG_ERR_SELECT_SESS
;
2379 /* Acquire lock for the session */
2380 session_lock(cmd_ctx
->session
);
2390 * Check domain type for specific "pre-action".
2392 switch (cmd_ctx
->lsm
->domain
.type
) {
2393 case LTTNG_DOMAIN_KERNEL
:
2395 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2399 /* Kernel tracer check */
2400 if (kernel_tracer_fd
== -1) {
2401 /* Basically, load kernel tracer modules */
2402 ret
= init_kernel_tracer();
2408 /* Consumer is in an ERROR state. Report back to client */
2409 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2410 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2414 /* Need a session for kernel command */
2415 if (need_tracing_session
) {
2416 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2417 ret
= create_kernel_session(cmd_ctx
->session
);
2419 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2424 /* Start the kernel consumer daemon */
2425 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2426 if (kconsumer_data
.pid
== 0 &&
2427 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2428 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2429 ret
= start_consumerd(&kconsumer_data
);
2431 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2434 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2436 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2440 * The consumer was just spawned so we need to add the socket to
2441 * the consumer output of the session if exist.
2443 ret
= consumer_create_socket(&kconsumer_data
,
2444 cmd_ctx
->session
->kernel_session
->consumer
);
2451 case LTTNG_DOMAIN_UST
:
2453 /* Consumer is in an ERROR state. Report back to client */
2454 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2455 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2459 if (need_tracing_session
) {
2460 /* Create UST session if none exist. */
2461 if (cmd_ctx
->session
->ust_session
== NULL
) {
2462 ret
= create_ust_session(cmd_ctx
->session
,
2463 &cmd_ctx
->lsm
->domain
);
2464 if (ret
!= LTTNG_OK
) {
2469 /* Start the UST consumer daemons */
2471 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2472 if (consumerd64_bin
[0] != '\0' &&
2473 ustconsumer64_data
.pid
== 0 &&
2474 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2475 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2476 ret
= start_consumerd(&ustconsumer64_data
);
2478 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2479 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2483 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2484 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2486 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2490 * Setup socket for consumer 64 bit. No need for atomic access
2491 * since it was set above and can ONLY be set in this thread.
2493 ret
= consumer_create_socket(&ustconsumer64_data
,
2494 cmd_ctx
->session
->ust_session
->consumer
);
2500 if (consumerd32_bin
[0] != '\0' &&
2501 ustconsumer32_data
.pid
== 0 &&
2502 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2503 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2504 ret
= start_consumerd(&ustconsumer32_data
);
2506 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2507 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2511 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2512 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2514 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2518 * Setup socket for consumer 64 bit. No need for atomic access
2519 * since it was set above and can ONLY be set in this thread.
2521 ret
= consumer_create_socket(&ustconsumer32_data
,
2522 cmd_ctx
->session
->ust_session
->consumer
);
2534 /* Validate consumer daemon state when start/stop trace command */
2535 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2536 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2537 switch (cmd_ctx
->lsm
->domain
.type
) {
2538 case LTTNG_DOMAIN_UST
:
2539 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2540 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2544 case LTTNG_DOMAIN_KERNEL
:
2545 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2546 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2554 * Check that the UID or GID match that of the tracing session.
2555 * The root user can interact with all sessions.
2557 if (need_tracing_session
) {
2558 if (!session_access_ok(cmd_ctx
->session
,
2559 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2560 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2561 ret
= LTTNG_ERR_EPERM
;
2567 * Send relayd information to consumer as soon as we have a domain and a
2570 if (cmd_ctx
->session
&& need_domain
) {
2572 * Setup relayd if not done yet. If the relayd information was already
2573 * sent to the consumer, this call will gracefully return.
2575 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2576 if (ret
!= LTTNG_OK
) {
2581 /* Process by command type */
2582 switch (cmd_ctx
->lsm
->cmd_type
) {
2583 case LTTNG_ADD_CONTEXT
:
2585 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2586 cmd_ctx
->lsm
->u
.context
.channel_name
,
2587 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2590 case LTTNG_DISABLE_CHANNEL
:
2592 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2593 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2596 case LTTNG_DISABLE_EVENT
:
2598 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2599 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2600 cmd_ctx
->lsm
->u
.disable
.name
);
2603 case LTTNG_DISABLE_ALL_EVENT
:
2605 DBG("Disabling all events");
2607 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2608 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2611 case LTTNG_ENABLE_CHANNEL
:
2613 ret
= cmd_enable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2614 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2617 case LTTNG_ENABLE_EVENT
:
2619 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2620 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2621 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2624 case LTTNG_ENABLE_ALL_EVENT
:
2626 DBG("Enabling all events");
2628 ret
= cmd_enable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2629 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2630 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2633 case LTTNG_LIST_TRACEPOINTS
:
2635 struct lttng_event
*events
;
2638 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2639 if (nb_events
< 0) {
2640 /* Return value is a negative lttng_error_code. */
2646 * Setup lttng message with payload size set to the event list size in
2647 * bytes and then copy list into the llm payload.
2649 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2655 /* Copy event list into message payload */
2656 memcpy(cmd_ctx
->llm
->payload
, events
,
2657 sizeof(struct lttng_event
) * nb_events
);
2664 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2666 struct lttng_event_field
*fields
;
2669 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2671 if (nb_fields
< 0) {
2672 /* Return value is a negative lttng_error_code. */
2678 * Setup lttng message with payload size set to the event list size in
2679 * bytes and then copy list into the llm payload.
2681 ret
= setup_lttng_msg(cmd_ctx
,
2682 sizeof(struct lttng_event_field
) * nb_fields
);
2688 /* Copy event list into message payload */
2689 memcpy(cmd_ctx
->llm
->payload
, fields
,
2690 sizeof(struct lttng_event_field
) * nb_fields
);
2697 case LTTNG_SET_CONSUMER_URI
:
2700 struct lttng_uri
*uris
;
2702 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2703 len
= nb_uri
* sizeof(struct lttng_uri
);
2706 ret
= LTTNG_ERR_INVALID
;
2710 uris
= zmalloc(len
);
2712 ret
= LTTNG_ERR_FATAL
;
2716 /* Receive variable len data */
2717 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2718 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2720 DBG("No URIs received from client... continuing");
2722 ret
= LTTNG_ERR_SESSION_FAIL
;
2727 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2729 if (ret
!= LTTNG_OK
) {
2735 * XXX: 0 means that this URI should be applied on the session. Should
2736 * be a DOMAIN enuam.
2738 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2739 /* Add the URI for the UST session if a consumer is present. */
2740 if (cmd_ctx
->session
->ust_session
&&
2741 cmd_ctx
->session
->ust_session
->consumer
) {
2742 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2744 } else if (cmd_ctx
->session
->kernel_session
&&
2745 cmd_ctx
->session
->kernel_session
->consumer
) {
2746 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2747 cmd_ctx
->session
, nb_uri
, uris
);
2755 case LTTNG_START_TRACE
:
2757 ret
= cmd_start_trace(cmd_ctx
->session
);
2760 case LTTNG_STOP_TRACE
:
2762 ret
= cmd_stop_trace(cmd_ctx
->session
);
2765 case LTTNG_CREATE_SESSION
:
2768 struct lttng_uri
*uris
= NULL
;
2770 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2771 len
= nb_uri
* sizeof(struct lttng_uri
);
2774 uris
= zmalloc(len
);
2776 ret
= LTTNG_ERR_FATAL
;
2780 /* Receive variable len data */
2781 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2782 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2784 DBG("No URIs received from client... continuing");
2786 ret
= LTTNG_ERR_SESSION_FAIL
;
2791 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2792 DBG("Creating session with ONE network URI is a bad call");
2793 ret
= LTTNG_ERR_SESSION_FAIL
;
2799 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2806 case LTTNG_DESTROY_SESSION
:
2808 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2810 /* Set session to NULL so we do not unlock it after free. */
2811 cmd_ctx
->session
= NULL
;
2814 case LTTNG_LIST_DOMAINS
:
2817 struct lttng_domain
*domains
;
2819 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2821 /* Return value is a negative lttng_error_code. */
2826 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2831 /* Copy event list into message payload */
2832 memcpy(cmd_ctx
->llm
->payload
, domains
,
2833 nb_dom
* sizeof(struct lttng_domain
));
2840 case LTTNG_LIST_CHANNELS
:
2843 struct lttng_channel
*channels
;
2845 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2846 cmd_ctx
->session
, &channels
);
2848 /* Return value is a negative lttng_error_code. */
2853 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2858 /* Copy event list into message payload */
2859 memcpy(cmd_ctx
->llm
->payload
, channels
,
2860 nb_chan
* sizeof(struct lttng_channel
));
2867 case LTTNG_LIST_EVENTS
:
2870 struct lttng_event
*events
= NULL
;
2872 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2873 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2875 /* Return value is a negative lttng_error_code. */
2880 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2885 /* Copy event list into message payload */
2886 memcpy(cmd_ctx
->llm
->payload
, events
,
2887 nb_event
* sizeof(struct lttng_event
));
2894 case LTTNG_LIST_SESSIONS
:
2896 unsigned int nr_sessions
;
2898 session_lock_list();
2899 nr_sessions
= lttng_sessions_count(
2900 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2901 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2903 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2905 session_unlock_list();
2909 /* Filled the session array */
2910 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2911 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2912 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2914 session_unlock_list();
2919 case LTTNG_CALIBRATE
:
2921 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
2922 &cmd_ctx
->lsm
->u
.calibrate
);
2925 case LTTNG_REGISTER_CONSUMER
:
2927 struct consumer_data
*cdata
;
2929 switch (cmd_ctx
->lsm
->domain
.type
) {
2930 case LTTNG_DOMAIN_KERNEL
:
2931 cdata
= &kconsumer_data
;
2934 ret
= LTTNG_ERR_UND
;
2938 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2939 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
2942 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
2944 struct lttng_filter_bytecode
*bytecode
;
2946 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
2947 ret
= LTTNG_ERR_FILTER_INVAL
;
2950 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
2951 ret
= LTTNG_ERR_FILTER_INVAL
;
2954 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2956 ret
= LTTNG_ERR_FILTER_NOMEM
;
2959 /* Receive var. len. data */
2960 DBG("Receiving var len data from client ...");
2961 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
2962 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2964 DBG("Nothing recv() from client var len data... continuing");
2966 ret
= LTTNG_ERR_FILTER_INVAL
;
2970 if (bytecode
->len
+ sizeof(*bytecode
)
2971 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
2973 ret
= LTTNG_ERR_FILTER_INVAL
;
2977 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2978 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2979 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
2982 case LTTNG_DATA_PENDING
:
2984 ret
= cmd_data_pending(cmd_ctx
->session
);
2988 ret
= LTTNG_ERR_UND
;
2993 if (cmd_ctx
->llm
== NULL
) {
2994 DBG("Missing llm structure. Allocating one.");
2995 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
2999 /* Set return code */
3000 cmd_ctx
->llm
->ret_code
= ret
;
3002 if (cmd_ctx
->session
) {
3003 session_unlock(cmd_ctx
->session
);
3005 if (need_tracing_session
) {
3006 session_unlock_list();
3013 * Thread managing health check socket.
3015 static void *thread_manage_health(void *data
)
3017 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3018 uint32_t revents
, nb_fd
;
3019 struct lttng_poll_event events
;
3020 struct lttcomm_health_msg msg
;
3021 struct lttcomm_health_data reply
;
3023 DBG("[thread] Manage health check started");
3025 rcu_register_thread();
3027 /* Create unix socket */
3028 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3030 ERR("Unable to create health check Unix socket");
3036 * Set the CLOEXEC flag. Return code is useless because either way, the
3039 (void) utils_set_fd_cloexec(sock
);
3041 ret
= lttcomm_listen_unix_sock(sock
);
3047 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3048 * more will be added to this poll set.
3050 ret
= sessiond_set_thread_pollset(&events
, 2);
3055 /* Add the application registration socket */
3056 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3062 DBG("Health check ready");
3064 /* Inifinite blocking call, waiting for transmission */
3066 ret
= lttng_poll_wait(&events
, -1);
3069 * Restart interrupted system call.
3071 if (errno
== EINTR
) {
3079 for (i
= 0; i
< nb_fd
; i
++) {
3080 /* Fetch once the poll data */
3081 revents
= LTTNG_POLL_GETEV(&events
, i
);
3082 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3084 /* Thread quit pipe has been closed. Killing thread. */
3085 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3091 /* Event on the registration socket */
3092 if (pollfd
== sock
) {
3093 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3094 ERR("Health socket poll error");
3100 new_sock
= lttcomm_accept_unix_sock(sock
);
3106 * Set the CLOEXEC flag. Return code is useless because either way, the
3109 (void) utils_set_fd_cloexec(new_sock
);
3111 DBG("Receiving data from client for health...");
3112 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3114 DBG("Nothing recv() from client... continuing");
3115 ret
= close(new_sock
);
3123 rcu_thread_online();
3125 switch (msg
.component
) {
3126 case LTTNG_HEALTH_CMD
:
3127 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3129 case LTTNG_HEALTH_APP_MANAGE
:
3130 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3132 case LTTNG_HEALTH_APP_REG
:
3133 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3135 case LTTNG_HEALTH_KERNEL
:
3136 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3138 case LTTNG_HEALTH_CONSUMER
:
3139 reply
.ret_code
= check_consumer_health();
3141 case LTTNG_HEALTH_ALL
:
3143 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3144 health_check_state(HEALTH_TYPE_APP_REG
) &&
3145 health_check_state(HEALTH_TYPE_CMD
) &&
3146 health_check_state(HEALTH_TYPE_KERNEL
) &&
3147 check_consumer_health();
3150 reply
.ret_code
= LTTNG_ERR_UND
;
3155 * Flip ret value since 0 is a success and 1 indicates a bad health for
3156 * the client where in the sessiond it is the opposite. Again, this is
3157 * just to make things easier for us poor developer which enjoy a lot
3160 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3161 reply
.ret_code
= !reply
.ret_code
;
3164 DBG2("Health check return value %d", reply
.ret_code
);
3166 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3168 ERR("Failed to send health data back to client");
3171 /* End of transmission */
3172 ret
= close(new_sock
);
3182 ERR("Health error occurred in %s", __func__
);
3184 DBG("Health check thread dying");
3185 unlink(health_unix_sock_path
);
3192 if (new_sock
>= 0) {
3193 ret
= close(new_sock
);
3199 lttng_poll_clean(&events
);
3201 rcu_unregister_thread();
3206 * This thread manage all clients request using the unix client socket for
3209 static void *thread_manage_clients(void *data
)
3211 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3213 uint32_t revents
, nb_fd
;
3214 struct command_ctx
*cmd_ctx
= NULL
;
3215 struct lttng_poll_event events
;
3217 DBG("[thread] Manage client started");
3219 rcu_register_thread();
3221 health_register(HEALTH_TYPE_CMD
);
3223 if (testpoint(thread_manage_clients
)) {
3224 goto error_testpoint
;
3227 health_code_update();
3229 ret
= lttcomm_listen_unix_sock(client_sock
);
3235 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3236 * more will be added to this poll set.
3238 ret
= sessiond_set_thread_pollset(&events
, 2);
3240 goto error_create_poll
;
3243 /* Add the application registration socket */
3244 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3250 * Notify parent pid that we are ready to accept command for client side.
3252 if (opt_sig_parent
) {
3253 kill(ppid
, SIGUSR1
);
3256 if (testpoint(thread_manage_clients_before_loop
)) {
3260 health_code_update();
3263 DBG("Accepting client command ...");
3265 /* Inifinite blocking call, waiting for transmission */
3267 health_poll_entry();
3268 ret
= lttng_poll_wait(&events
, -1);
3272 * Restart interrupted system call.
3274 if (errno
== EINTR
) {
3282 for (i
= 0; i
< nb_fd
; i
++) {
3283 /* Fetch once the poll data */
3284 revents
= LTTNG_POLL_GETEV(&events
, i
);
3285 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3287 health_code_update();
3289 /* Thread quit pipe has been closed. Killing thread. */
3290 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3296 /* Event on the registration socket */
3297 if (pollfd
== client_sock
) {
3298 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3299 ERR("Client socket poll error");
3305 DBG("Wait for client response");
3307 health_code_update();
3309 sock
= lttcomm_accept_unix_sock(client_sock
);
3315 * Set the CLOEXEC flag. Return code is useless because either way, the
3318 (void) utils_set_fd_cloexec(sock
);
3320 /* Set socket option for credentials retrieval */
3321 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3326 /* Allocate context command to process the client request */
3327 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3328 if (cmd_ctx
== NULL
) {
3329 PERROR("zmalloc cmd_ctx");
3333 /* Allocate data buffer for reception */
3334 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3335 if (cmd_ctx
->lsm
== NULL
) {
3336 PERROR("zmalloc cmd_ctx->lsm");
3340 cmd_ctx
->llm
= NULL
;
3341 cmd_ctx
->session
= NULL
;
3343 health_code_update();
3346 * Data is received from the lttng client. The struct
3347 * lttcomm_session_msg (lsm) contains the command and data request of
3350 DBG("Receiving data from client ...");
3351 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3352 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3354 DBG("Nothing recv() from client... continuing");
3360 clean_command_ctx(&cmd_ctx
);
3364 health_code_update();
3366 // TODO: Validate cmd_ctx including sanity check for
3367 // security purpose.
3369 rcu_thread_online();
3371 * This function dispatch the work to the kernel or userspace tracer
3372 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3373 * informations for the client. The command context struct contains
3374 * everything this function may needs.
3376 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3377 rcu_thread_offline();
3387 * TODO: Inform client somehow of the fatal error. At
3388 * this point, ret < 0 means that a zmalloc failed
3389 * (ENOMEM). Error detected but still accept
3390 * command, unless a socket error has been
3393 clean_command_ctx(&cmd_ctx
);
3397 health_code_update();
3399 DBG("Sending response (size: %d, retcode: %s)",
3400 cmd_ctx
->lttng_msg_size
,
3401 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3402 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3404 ERR("Failed to send data back to client");
3407 /* End of transmission */
3414 clean_command_ctx(&cmd_ctx
);
3416 health_code_update();
3428 lttng_poll_clean(&events
);
3429 clean_command_ctx(&cmd_ctx
);
3434 unlink(client_unix_sock_path
);
3435 if (client_sock
>= 0) {
3436 ret
= close(client_sock
);
3444 ERR("Health error occurred in %s", __func__
);
3447 health_unregister();
3449 DBG("Client thread dying");
3451 rcu_unregister_thread();
3457 * usage function on stderr
3459 static void usage(void)
3461 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3462 fprintf(stderr
, " -h, --help Display this usage.\n");
3463 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3464 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3465 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3466 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3467 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3468 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3469 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3470 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3471 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3472 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3473 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3474 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3475 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3476 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3477 fprintf(stderr
, " -V, --version Show version number.\n");
3478 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3479 fprintf(stderr
, " -q, --quiet No output at all.\n");
3480 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3481 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3482 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3483 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3487 * daemon argument parsing
3489 static int parse_args(int argc
, char **argv
)
3493 static struct option long_options
[] = {
3494 { "client-sock", 1, 0, 'c' },
3495 { "apps-sock", 1, 0, 'a' },
3496 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3497 { "kconsumerd-err-sock", 1, 0, 'E' },
3498 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3499 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3500 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3501 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3502 { "consumerd32-path", 1, 0, 'u' },
3503 { "consumerd32-libdir", 1, 0, 'U' },
3504 { "consumerd64-path", 1, 0, 't' },
3505 { "consumerd64-libdir", 1, 0, 'T' },
3506 { "daemonize", 0, 0, 'd' },
3507 { "sig-parent", 0, 0, 'S' },
3508 { "help", 0, 0, 'h' },
3509 { "group", 1, 0, 'g' },
3510 { "version", 0, 0, 'V' },
3511 { "quiet", 0, 0, 'q' },
3512 { "verbose", 0, 0, 'v' },
3513 { "verbose-consumer", 0, 0, 'Z' },
3514 { "no-kernel", 0, 0, 'N' },
3515 { "pidfile", 1, 0, 'p' },
3520 int option_index
= 0;
3521 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3522 long_options
, &option_index
);
3529 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3531 fprintf(stderr
, " with arg %s\n", optarg
);
3535 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3538 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3544 opt_tracing_group
= optarg
;
3550 fprintf(stdout
, "%s\n", VERSION
);
3556 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3559 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3562 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3565 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3568 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3571 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3577 lttng_opt_quiet
= 1;
3580 /* Verbose level can increase using multiple -v */
3581 lttng_opt_verbose
+= 1;
3584 opt_verbose_consumer
+= 1;
3587 consumerd32_bin
= optarg
;
3590 consumerd32_libdir
= optarg
;
3593 consumerd64_bin
= optarg
;
3596 consumerd64_libdir
= optarg
;
3599 opt_pidfile
= optarg
;
3602 /* Unknown option or other error.
3603 * Error is printed by getopt, just return */
3612 * Creates the two needed socket by the daemon.
3613 * apps_sock - The communication socket for all UST apps.
3614 * client_sock - The communication of the cli tool (lttng).
3616 static int init_daemon_socket(void)
3621 old_umask
= umask(0);
3623 /* Create client tool unix socket */
3624 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3625 if (client_sock
< 0) {
3626 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3631 /* Set the cloexec flag */
3632 ret
= utils_set_fd_cloexec(client_sock
);
3634 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3635 "Continuing but note that the consumer daemon will have a "
3636 "reference to this socket on exec()", client_sock
);
3639 /* File permission MUST be 660 */
3640 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3642 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3647 /* Create the application unix socket */
3648 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3649 if (apps_sock
< 0) {
3650 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3655 /* Set the cloexec flag */
3656 ret
= utils_set_fd_cloexec(apps_sock
);
3658 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3659 "Continuing but note that the consumer daemon will have a "
3660 "reference to this socket on exec()", apps_sock
);
3663 /* File permission MUST be 666 */
3664 ret
= chmod(apps_unix_sock_path
,
3665 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3667 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3672 DBG3("Session daemon client socket %d and application socket %d created",
3673 client_sock
, apps_sock
);
3681 * Check if the global socket is available, and if a daemon is answering at the
3682 * other side. If yes, error is returned.
3684 static int check_existing_daemon(void)
3686 /* Is there anybody out there ? */
3687 if (lttng_session_daemon_alive()) {
3695 * Set the tracing group gid onto the client socket.
3697 * Race window between mkdir and chown is OK because we are going from more
3698 * permissive (root.root) to less permissive (root.tracing).
3700 static int set_permissions(char *rundir
)
3705 ret
= allowed_group();
3707 WARN("No tracing group detected");
3714 /* Set lttng run dir */
3715 ret
= chown(rundir
, 0, gid
);
3717 ERR("Unable to set group on %s", rundir
);
3721 /* Ensure tracing group can search the run dir */
3722 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3724 ERR("Unable to set permissions on %s", rundir
);
3728 /* lttng client socket path */
3729 ret
= chown(client_unix_sock_path
, 0, gid
);
3731 ERR("Unable to set group on %s", client_unix_sock_path
);
3735 /* kconsumer error socket path */
3736 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3738 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3742 /* 64-bit ustconsumer error socket path */
3743 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3745 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3749 /* 32-bit ustconsumer compat32 error socket path */
3750 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3752 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3756 DBG("All permissions are set");
3763 * Create the lttng run directory needed for all global sockets and pipe.
3765 static int create_lttng_rundir(const char *rundir
)
3769 DBG3("Creating LTTng run directory: %s", rundir
);
3771 ret
= mkdir(rundir
, S_IRWXU
);
3773 if (errno
!= EEXIST
) {
3774 ERR("Unable to create %s", rundir
);
3786 * Setup sockets and directory needed by the kconsumerd communication with the
3789 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3793 char path
[PATH_MAX
];
3795 switch (consumer_data
->type
) {
3796 case LTTNG_CONSUMER_KERNEL
:
3797 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3799 case LTTNG_CONSUMER64_UST
:
3800 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3802 case LTTNG_CONSUMER32_UST
:
3803 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3806 ERR("Consumer type unknown");
3811 DBG2("Creating consumer directory: %s", path
);
3813 ret
= mkdir(path
, S_IRWXU
);
3815 if (errno
!= EEXIST
) {
3817 ERR("Failed to create %s", path
);
3823 /* Create the kconsumerd error unix socket */
3824 consumer_data
->err_sock
=
3825 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3826 if (consumer_data
->err_sock
< 0) {
3827 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3832 /* File permission MUST be 660 */
3833 ret
= chmod(consumer_data
->err_unix_sock_path
,
3834 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3836 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3846 * Signal handler for the daemon
3848 * Simply stop all worker threads, leaving main() return gracefully after
3849 * joining all threads and calling cleanup().
3851 static void sighandler(int sig
)
3855 DBG("SIGPIPE caught");
3858 DBG("SIGINT caught");
3862 DBG("SIGTERM caught");
3871 * Setup signal handler for :
3872 * SIGINT, SIGTERM, SIGPIPE
3874 static int set_signal_handler(void)
3877 struct sigaction sa
;
3880 if ((ret
= sigemptyset(&sigset
)) < 0) {
3881 PERROR("sigemptyset");
3885 sa
.sa_handler
= sighandler
;
3886 sa
.sa_mask
= sigset
;
3888 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3889 PERROR("sigaction");
3893 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3894 PERROR("sigaction");
3898 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3899 PERROR("sigaction");
3903 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3909 * Set open files limit to unlimited. This daemon can open a large number of
3910 * file descriptors in order to consumer multiple kernel traces.
3912 static void set_ulimit(void)
3917 /* The kernel does not allowed an infinite limit for open files */
3918 lim
.rlim_cur
= 65535;
3919 lim
.rlim_max
= 65535;
3921 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
3923 PERROR("failed to set open files limit");
3928 * Write pidfile using the rundir and opt_pidfile.
3930 static void write_pidfile(void)
3933 char pidfile_path
[PATH_MAX
];
3938 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
3940 /* Build pidfile path from rundir and opt_pidfile. */
3941 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
3942 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
3944 PERROR("snprintf pidfile path");
3950 * Create pid file in rundir. Return value is of no importance. The
3951 * execution will continue even though we are not able to write the file.
3953 (void) utils_create_pid_file(getpid(), pidfile_path
);
3962 int main(int argc
, char **argv
)
3966 const char *home_path
, *env_app_timeout
;
3968 init_kernel_workarounds();
3970 rcu_register_thread();
3972 setup_consumerd_path();
3974 /* Parse arguments */
3976 if ((ret
= parse_args(argc
, argv
)) < 0) {
3986 * child: setsid, close FD 0, 1, 2, chdir /
3987 * parent: exit (if fork is successful)
3995 * We are in the child. Make sure all other file
3996 * descriptors are closed, in case we are called with
3997 * more opened file descriptors than the standard ones.
3999 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4004 /* Create thread quit pipe */
4005 if ((ret
= init_thread_quit_pipe()) < 0) {
4009 /* Check if daemon is UID = 0 */
4010 is_root
= !getuid();
4013 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4015 /* Create global run dir with root access */
4016 ret
= create_lttng_rundir(rundir
);
4021 if (strlen(apps_unix_sock_path
) == 0) {
4022 snprintf(apps_unix_sock_path
, PATH_MAX
,
4023 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4026 if (strlen(client_unix_sock_path
) == 0) {
4027 snprintf(client_unix_sock_path
, PATH_MAX
,
4028 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4031 /* Set global SHM for ust */
4032 if (strlen(wait_shm_path
) == 0) {
4033 snprintf(wait_shm_path
, PATH_MAX
,
4034 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4037 if (strlen(health_unix_sock_path
) == 0) {
4038 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4039 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4042 /* Setup kernel consumerd path */
4043 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4044 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4045 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4046 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4048 DBG2("Kernel consumer err path: %s",
4049 kconsumer_data
.err_unix_sock_path
);
4050 DBG2("Kernel consumer cmd path: %s",
4051 kconsumer_data
.cmd_unix_sock_path
);
4053 home_path
= get_home_dir();
4054 if (home_path
== NULL
) {
4055 /* TODO: Add --socket PATH option */
4056 ERR("Can't get HOME directory for sockets creation.");
4062 * Create rundir from home path. This will create something like
4065 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4071 ret
= create_lttng_rundir(rundir
);
4076 if (strlen(apps_unix_sock_path
) == 0) {
4077 snprintf(apps_unix_sock_path
, PATH_MAX
,
4078 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4081 /* Set the cli tool unix socket path */
4082 if (strlen(client_unix_sock_path
) == 0) {
4083 snprintf(client_unix_sock_path
, PATH_MAX
,
4084 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4087 /* Set global SHM for ust */
4088 if (strlen(wait_shm_path
) == 0) {
4089 snprintf(wait_shm_path
, PATH_MAX
,
4090 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4093 /* Set health check Unix path */
4094 if (strlen(health_unix_sock_path
) == 0) {
4095 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4096 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4100 /* Set consumer initial state */
4101 kernel_consumerd_state
= CONSUMER_STOPPED
;
4102 ust_consumerd_state
= CONSUMER_STOPPED
;
4104 DBG("Client socket path %s", client_unix_sock_path
);
4105 DBG("Application socket path %s", apps_unix_sock_path
);
4106 DBG("Application wait path %s", wait_shm_path
);
4107 DBG("LTTng run directory path: %s", rundir
);
4109 /* 32 bits consumerd path setup */
4110 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4111 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4112 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4113 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4115 DBG2("UST consumer 32 bits err path: %s",
4116 ustconsumer32_data
.err_unix_sock_path
);
4117 DBG2("UST consumer 32 bits cmd path: %s",
4118 ustconsumer32_data
.cmd_unix_sock_path
);
4120 /* 64 bits consumerd path setup */
4121 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4122 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4123 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4124 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4126 DBG2("UST consumer 64 bits err path: %s",
4127 ustconsumer64_data
.err_unix_sock_path
);
4128 DBG2("UST consumer 64 bits cmd path: %s",
4129 ustconsumer64_data
.cmd_unix_sock_path
);
4132 * See if daemon already exist.
4134 if ((ret
= check_existing_daemon()) < 0) {
4135 ERR("Already running daemon.\n");
4137 * We do not goto exit because we must not cleanup()
4138 * because a daemon is already running.
4144 * Init UST app hash table. Alloc hash table before this point since
4145 * cleanup() can get called after that point.
4149 /* After this point, we can safely call cleanup() with "goto exit" */
4152 * These actions must be executed as root. We do that *after* setting up
4153 * the sockets path because we MUST make the check for another daemon using
4154 * those paths *before* trying to set the kernel consumer sockets and init
4158 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4163 /* Setup kernel tracer */
4164 if (!opt_no_kernel
) {
4165 init_kernel_tracer();
4168 /* Set ulimit for open files */
4171 /* init lttng_fd tracking must be done after set_ulimit. */
4174 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4179 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4184 if ((ret
= set_signal_handler()) < 0) {
4188 /* Setup the needed unix socket */
4189 if ((ret
= init_daemon_socket()) < 0) {
4193 /* Set credentials to socket */
4194 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4198 /* Get parent pid if -S, --sig-parent is specified. */
4199 if (opt_sig_parent
) {
4203 /* Setup the kernel pipe for waking up the kernel thread */
4204 if (is_root
&& !opt_no_kernel
) {
4205 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4210 /* Setup the thread apps communication pipe. */
4211 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4215 /* Setup the thread apps notify communication pipe. */
4216 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4220 /* Init UST command queue. */
4221 cds_wfq_init(&ust_cmd_queue
.queue
);
4224 * Get session list pointer. This pointer MUST NOT be free(). This list is
4225 * statically declared in session.c
4227 session_list_ptr
= session_get_list();
4229 /* Set up max poll set size */
4230 lttng_poll_set_max_size();
4234 /* Check for the application socket timeout env variable. */
4235 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4236 if (env_app_timeout
) {
4237 app_socket_timeout
= atoi(env_app_timeout
);
4239 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4244 /* Create thread to manage the client socket */
4245 ret
= pthread_create(&health_thread
, NULL
,
4246 thread_manage_health
, (void *) NULL
);
4248 PERROR("pthread_create health");
4252 /* Create thread to manage the client socket */
4253 ret
= pthread_create(&client_thread
, NULL
,
4254 thread_manage_clients
, (void *) NULL
);
4256 PERROR("pthread_create clients");
4260 /* Create thread to dispatch registration */
4261 ret
= pthread_create(&dispatch_thread
, NULL
,
4262 thread_dispatch_ust_registration
, (void *) NULL
);
4264 PERROR("pthread_create dispatch");
4268 /* Create thread to manage application registration. */
4269 ret
= pthread_create(®_apps_thread
, NULL
,
4270 thread_registration_apps
, (void *) NULL
);
4272 PERROR("pthread_create registration");
4276 /* Create thread to manage application socket */
4277 ret
= pthread_create(&apps_thread
, NULL
,
4278 thread_manage_apps
, (void *) NULL
);
4280 PERROR("pthread_create apps");
4284 /* Create thread to manage application notify socket */
4285 ret
= pthread_create(&apps_notify_thread
, NULL
,
4286 ust_thread_manage_notify
, (void *) NULL
);
4288 PERROR("pthread_create apps");
4292 /* Don't start this thread if kernel tracing is not requested nor root */
4293 if (is_root
&& !opt_no_kernel
) {
4294 /* Create kernel thread to manage kernel event */
4295 ret
= pthread_create(&kernel_thread
, NULL
,
4296 thread_manage_kernel
, (void *) NULL
);
4298 PERROR("pthread_create kernel");
4302 ret
= pthread_join(kernel_thread
, &status
);
4304 PERROR("pthread_join");
4305 goto error
; /* join error, exit without cleanup */
4310 ret
= pthread_join(apps_thread
, &status
);
4312 PERROR("pthread_join");
4313 goto error
; /* join error, exit without cleanup */
4317 ret
= pthread_join(reg_apps_thread
, &status
);
4319 PERROR("pthread_join");
4320 goto error
; /* join error, exit without cleanup */
4324 ret
= pthread_join(dispatch_thread
, &status
);
4326 PERROR("pthread_join");
4327 goto error
; /* join error, exit without cleanup */
4331 ret
= pthread_join(client_thread
, &status
);
4333 PERROR("pthread_join");
4334 goto error
; /* join error, exit without cleanup */
4337 ret
= join_consumer_thread(&kconsumer_data
);
4339 PERROR("join_consumer");
4340 goto error
; /* join error, exit without cleanup */
4343 ret
= join_consumer_thread(&ustconsumer32_data
);
4345 PERROR("join_consumer ust32");
4346 goto error
; /* join error, exit without cleanup */
4349 ret
= join_consumer_thread(&ustconsumer64_data
);
4351 PERROR("join_consumer ust64");
4352 goto error
; /* join error, exit without cleanup */
4356 ret
= pthread_join(health_thread
, &status
);
4358 PERROR("pthread_join health thread");
4359 goto error
; /* join error, exit without cleanup */
4365 * cleanup() is called when no other thread is running.
4367 rcu_thread_online();
4369 rcu_thread_offline();
4370 rcu_unregister_thread();