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.
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
35 #include <sys/types.h>
37 #include <urcu/uatomic.h>
41 #include <common/common.h>
42 #include <common/compat/socket.h>
43 #include <common/defaults.h>
44 #include <common/kernel-consumer/kernel-consumer.h>
45 #include <common/futex.h>
46 #include <common/relayd/relayd.h>
47 #include <common/utils.h>
49 #include "lttng-sessiond.h"
50 #include "buffer-registry.h"
57 #include "kernel-consumer.h"
61 #include "ust-consumer.h"
64 #include "health-sessiond.h"
65 #include "testpoint.h"
66 #include "ust-thread.h"
67 #include "jul-thread.h"
69 #define CONSUMERD_FILE "lttng-consumerd"
72 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
73 static const char *opt_pidfile
;
74 static int opt_sig_parent
;
75 static int opt_verbose_consumer
;
76 static int opt_daemon
;
77 static int opt_no_kernel
;
78 static int is_root
; /* Set to 1 if the daemon is running as root */
79 static pid_t ppid
; /* Parent PID for --sig-parent option */
80 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
83 /* Set to 1 when a SIGUSR1 signal is received. */
84 static int recv_child_signal
;
87 * Consumer daemon specific control data. Every value not initialized here is
88 * set to 0 by the static definition.
90 static struct consumer_data kconsumer_data
= {
91 .type
= LTTNG_CONSUMER_KERNEL
,
92 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
93 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
96 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
97 .lock
= PTHREAD_MUTEX_INITIALIZER
,
98 .cond
= PTHREAD_COND_INITIALIZER
,
99 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
101 static struct consumer_data ustconsumer64_data
= {
102 .type
= LTTNG_CONSUMER64_UST
,
103 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
104 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
107 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
108 .lock
= PTHREAD_MUTEX_INITIALIZER
,
109 .cond
= PTHREAD_COND_INITIALIZER
,
110 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
112 static struct consumer_data ustconsumer32_data
= {
113 .type
= LTTNG_CONSUMER32_UST
,
114 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
115 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
118 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 .lock
= PTHREAD_MUTEX_INITIALIZER
,
120 .cond
= PTHREAD_COND_INITIALIZER
,
121 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
124 /* Shared between threads */
125 static int dispatch_thread_exit
;
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path
[PATH_MAX
];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path
[PATH_MAX
];
131 /* global wait shm path for UST */
132 static char wait_shm_path
[PATH_MAX
];
133 /* Global health check unix path */
134 static char health_unix_sock_path
[PATH_MAX
];
136 /* Sockets and FDs */
137 static int client_sock
= -1;
138 static int apps_sock
= -1;
139 int kernel_tracer_fd
= -1;
140 static int kernel_poll_pipe
[2] = { -1, -1 };
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
146 static int thread_quit_pipe
[2] = { -1, -1 };
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
152 static int apps_cmd_pipe
[2] = { -1, -1 };
154 int apps_cmd_notify_pipe
[2] = { -1, -1 };
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread
;
158 static pthread_t apps_notify_thread
;
159 static pthread_t reg_apps_thread
;
160 static pthread_t client_thread
;
161 static pthread_t kernel_thread
;
162 static pthread_t dispatch_thread
;
163 static pthread_t health_thread
;
164 static pthread_t ht_cleanup_thread
;
165 static pthread_t jul_reg_thread
;
168 * UST registration command queue. This queue is tied with a futex and uses a N
169 * wakers / 1 waiter implemented and detailed in futex.c/.h
171 * The thread_manage_apps and thread_dispatch_ust_registration interact with
172 * this queue and the wait/wake scheme.
174 static struct ust_cmd_queue ust_cmd_queue
;
177 * Pointer initialized before thread creation.
179 * This points to the tracing session list containing the session count and a
180 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
181 * MUST NOT be taken if you call a public function in session.c.
183 * The lock is nested inside the structure: session_list_ptr->lock. Please use
184 * session_lock_list and session_unlock_list for lock acquisition.
186 static struct ltt_session_list
*session_list_ptr
;
188 int ust_consumerd64_fd
= -1;
189 int ust_consumerd32_fd
= -1;
191 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
192 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
193 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
194 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
196 static const char *module_proc_lttng
= "/proc/lttng";
199 * Consumer daemon state which is changed when spawning it, killing it or in
200 * case of a fatal error.
202 enum consumerd_state
{
203 CONSUMER_STARTED
= 1,
204 CONSUMER_STOPPED
= 2,
209 * This consumer daemon state is used to validate if a client command will be
210 * able to reach the consumer. If not, the client is informed. For instance,
211 * doing a "lttng start" when the consumer state is set to ERROR will return an
212 * error to the client.
214 * The following example shows a possible race condition of this scheme:
216 * consumer thread error happens
218 * client cmd checks state -> still OK
219 * consumer thread exit, sets error
220 * client cmd try to talk to consumer
223 * However, since the consumer is a different daemon, we have no way of making
224 * sure the command will reach it safely even with this state flag. This is why
225 * we consider that up to the state validation during command processing, the
226 * command is safe. After that, we can not guarantee the correctness of the
227 * client request vis-a-vis the consumer.
229 static enum consumerd_state ust_consumerd_state
;
230 static enum consumerd_state kernel_consumerd_state
;
233 * Socket timeout for receiving and sending in seconds.
235 static int app_socket_timeout
;
237 /* Set in main() with the current page size. */
240 /* Application health monitoring */
241 struct health_app
*health_sessiond
;
243 /* JUL TCP port for registration. Used by the JUL thread. */
244 unsigned int jul_tcp_port
= DEFAULT_JUL_TCP_PORT
;
247 void setup_consumerd_path(void)
249 const char *bin
, *libdir
;
252 * Allow INSTALL_BIN_PATH to be used as a target path for the
253 * native architecture size consumer if CONFIG_CONSUMER*_PATH
254 * has not been defined.
256 #if (CAA_BITS_PER_LONG == 32)
257 if (!consumerd32_bin
[0]) {
258 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
260 if (!consumerd32_libdir
[0]) {
261 consumerd32_libdir
= INSTALL_LIB_PATH
;
263 #elif (CAA_BITS_PER_LONG == 64)
264 if (!consumerd64_bin
[0]) {
265 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
267 if (!consumerd64_libdir
[0]) {
268 consumerd64_libdir
= INSTALL_LIB_PATH
;
271 #error "Unknown bitness"
275 * runtime env. var. overrides the build default.
277 bin
= getenv("LTTNG_CONSUMERD32_BIN");
279 consumerd32_bin
= bin
;
281 bin
= getenv("LTTNG_CONSUMERD64_BIN");
283 consumerd64_bin
= bin
;
285 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
287 consumerd32_libdir
= libdir
;
289 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
291 consumerd64_libdir
= libdir
;
296 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
298 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
304 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
310 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
322 * Check if the thread quit pipe was triggered.
324 * Return 1 if it was triggered else 0;
326 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
328 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
336 * Init thread quit pipe.
338 * Return -1 on error or 0 if all pipes are created.
340 static int init_thread_quit_pipe(void)
344 ret
= pipe(thread_quit_pipe
);
346 PERROR("thread quit pipe");
350 for (i
= 0; i
< 2; i
++) {
351 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
363 * Stop all threads by closing the thread quit pipe.
365 static void stop_threads(void)
369 /* Stopping all threads */
370 DBG("Terminating all threads");
371 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
373 ERR("write error on thread quit pipe");
376 /* Dispatch thread */
377 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
378 futex_nto1_wake(&ust_cmd_queue
.futex
);
382 * Close every consumer sockets.
384 static void close_consumer_sockets(void)
388 if (kconsumer_data
.err_sock
>= 0) {
389 ret
= close(kconsumer_data
.err_sock
);
391 PERROR("kernel consumer err_sock close");
394 if (ustconsumer32_data
.err_sock
>= 0) {
395 ret
= close(ustconsumer32_data
.err_sock
);
397 PERROR("UST consumerd32 err_sock close");
400 if (ustconsumer64_data
.err_sock
>= 0) {
401 ret
= close(ustconsumer64_data
.err_sock
);
403 PERROR("UST consumerd64 err_sock close");
406 if (kconsumer_data
.cmd_sock
>= 0) {
407 ret
= close(kconsumer_data
.cmd_sock
);
409 PERROR("kernel consumer cmd_sock close");
412 if (ustconsumer32_data
.cmd_sock
>= 0) {
413 ret
= close(ustconsumer32_data
.cmd_sock
);
415 PERROR("UST consumerd32 cmd_sock close");
418 if (ustconsumer64_data
.cmd_sock
>= 0) {
419 ret
= close(ustconsumer64_data
.cmd_sock
);
421 PERROR("UST consumerd64 cmd_sock close");
429 static void cleanup(void)
432 struct ltt_session
*sess
, *stmp
;
438 * Close the thread quit pipe. It has already done its job,
439 * since we are now called.
441 utils_close_pipe(thread_quit_pipe
);
444 * If opt_pidfile is undefined, the default file will be wiped when
445 * removing the rundir.
448 ret
= remove(opt_pidfile
);
450 PERROR("remove pidfile %s", opt_pidfile
);
454 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
457 snprintf(path
, PATH_MAX
,
459 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
460 DBG("Removing %s", path
);
463 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
464 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
);
465 DBG("Removing %s", path
);
469 snprintf(path
, PATH_MAX
,
470 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
472 DBG("Removing %s", path
);
475 snprintf(path
, PATH_MAX
,
476 DEFAULT_KCONSUMERD_PATH
,
478 DBG("Removing directory %s", path
);
481 /* ust consumerd 32 */
482 snprintf(path
, PATH_MAX
,
483 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
485 DBG("Removing %s", path
);
488 snprintf(path
, PATH_MAX
,
489 DEFAULT_USTCONSUMERD32_PATH
,
491 DBG("Removing directory %s", path
);
494 /* ust consumerd 64 */
495 snprintf(path
, PATH_MAX
,
496 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
498 DBG("Removing %s", path
);
501 snprintf(path
, PATH_MAX
,
502 DEFAULT_USTCONSUMERD64_PATH
,
504 DBG("Removing directory %s", path
);
508 * We do NOT rmdir rundir because there are other processes
509 * using it, for instance lttng-relayd, which can start in
510 * parallel with this teardown.
515 DBG("Cleaning up all sessions");
517 /* Destroy session list mutex */
518 if (session_list_ptr
!= NULL
) {
519 pthread_mutex_destroy(&session_list_ptr
->lock
);
521 /* Cleanup ALL session */
522 cds_list_for_each_entry_safe(sess
, stmp
,
523 &session_list_ptr
->head
, list
) {
524 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
528 DBG("Closing all UST sockets");
529 ust_app_clean_list();
530 buffer_reg_destroy_registries();
532 if (is_root
&& !opt_no_kernel
) {
533 DBG2("Closing kernel fd");
534 if (kernel_tracer_fd
>= 0) {
535 ret
= close(kernel_tracer_fd
);
540 DBG("Unloading kernel modules");
541 modprobe_remove_lttng_all();
544 close_consumer_sockets();
547 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
548 "Matthew, BEET driven development works!%c[%dm",
549 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
554 * Send data on a unix socket using the liblttsessiondcomm API.
556 * Return lttcomm error code.
558 static int send_unix_sock(int sock
, void *buf
, size_t len
)
560 /* Check valid length */
565 return lttcomm_send_unix_sock(sock
, buf
, len
);
569 * Free memory of a command context structure.
571 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
573 DBG("Clean command context structure");
575 if ((*cmd_ctx
)->llm
) {
576 free((*cmd_ctx
)->llm
);
578 if ((*cmd_ctx
)->lsm
) {
579 free((*cmd_ctx
)->lsm
);
587 * Notify UST applications using the shm mmap futex.
589 static int notify_ust_apps(int active
)
593 DBG("Notifying applications of session daemon state: %d", active
);
595 /* See shm.c for this call implying mmap, shm and futex calls */
596 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
597 if (wait_shm_mmap
== NULL
) {
601 /* Wake waiting process */
602 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
604 /* Apps notified successfully */
612 * Setup the outgoing data buffer for the response (llm) by allocating the
613 * right amount of memory and copying the original information from the lsm
616 * Return total size of the buffer pointed by buf.
618 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
624 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
625 if (cmd_ctx
->llm
== NULL
) {
631 /* Copy common data */
632 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
633 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
635 cmd_ctx
->llm
->data_size
= size
;
636 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
645 * Update the kernel poll set of all channel fd available over all tracing
646 * session. Add the wakeup pipe at the end of the set.
648 static int update_kernel_poll(struct lttng_poll_event
*events
)
651 struct ltt_session
*session
;
652 struct ltt_kernel_channel
*channel
;
654 DBG("Updating kernel poll set");
657 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
658 session_lock(session
);
659 if (session
->kernel_session
== NULL
) {
660 session_unlock(session
);
664 cds_list_for_each_entry(channel
,
665 &session
->kernel_session
->channel_list
.head
, list
) {
666 /* Add channel fd to the kernel poll set */
667 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
669 session_unlock(session
);
672 DBG("Channel fd %d added to kernel set", channel
->fd
);
674 session_unlock(session
);
676 session_unlock_list();
681 session_unlock_list();
686 * Find the channel fd from 'fd' over all tracing session. When found, check
687 * for new channel stream and send those stream fds to the kernel consumer.
689 * Useful for CPU hotplug feature.
691 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
694 struct ltt_session
*session
;
695 struct ltt_kernel_session
*ksess
;
696 struct ltt_kernel_channel
*channel
;
698 DBG("Updating kernel streams for channel fd %d", fd
);
701 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
702 session_lock(session
);
703 if (session
->kernel_session
== NULL
) {
704 session_unlock(session
);
707 ksess
= session
->kernel_session
;
709 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
710 if (channel
->fd
== fd
) {
711 DBG("Channel found, updating kernel streams");
712 ret
= kernel_open_channel_stream(channel
);
716 /* Update the stream global counter */
717 ksess
->stream_count_global
+= ret
;
720 * Have we already sent fds to the consumer? If yes, it means
721 * that tracing is started so it is safe to send our updated
724 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
725 struct lttng_ht_iter iter
;
726 struct consumer_socket
*socket
;
729 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
730 &iter
.iter
, socket
, node
.node
) {
731 pthread_mutex_lock(socket
->lock
);
732 ret
= kernel_consumer_send_channel_stream(socket
,
734 session
->output_traces
? 1 : 0);
735 pthread_mutex_unlock(socket
->lock
);
746 session_unlock(session
);
748 session_unlock_list();
752 session_unlock(session
);
753 session_unlock_list();
758 * For each tracing session, update newly registered apps. The session list
759 * lock MUST be acquired before calling this.
761 static void update_ust_app(int app_sock
)
763 struct ltt_session
*sess
, *stmp
;
765 /* Consumer is in an ERROR state. Stop any application update. */
766 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
767 /* Stop the update process since the consumer is dead. */
771 /* For all tracing session(s) */
772 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
774 if (sess
->ust_session
) {
775 ust_app_global_update(sess
->ust_session
, app_sock
);
777 session_unlock(sess
);
782 * This thread manage event coming from the kernel.
784 * Features supported in this thread:
787 static void *thread_manage_kernel(void *data
)
789 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
790 uint32_t revents
, nb_fd
;
792 struct lttng_poll_event events
;
794 DBG("[thread] Thread manage kernel started");
796 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
799 * This first step of the while is to clean this structure which could free
800 * non NULL pointers so initialize it before the loop.
802 lttng_poll_init(&events
);
804 if (testpoint(thread_manage_kernel
)) {
805 goto error_testpoint
;
808 health_code_update();
810 if (testpoint(thread_manage_kernel_before_loop
)) {
811 goto error_testpoint
;
815 health_code_update();
817 if (update_poll_flag
== 1) {
818 /* Clean events object. We are about to populate it again. */
819 lttng_poll_clean(&events
);
821 ret
= sessiond_set_thread_pollset(&events
, 2);
823 goto error_poll_create
;
826 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
831 /* This will add the available kernel channel if any. */
832 ret
= update_kernel_poll(&events
);
836 update_poll_flag
= 0;
839 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
841 /* Poll infinite value of time */
844 ret
= lttng_poll_wait(&events
, -1);
848 * Restart interrupted system call.
850 if (errno
== EINTR
) {
854 } else if (ret
== 0) {
855 /* Should not happen since timeout is infinite */
856 ERR("Return value of poll is 0 with an infinite timeout.\n"
857 "This should not have happened! Continuing...");
863 for (i
= 0; i
< nb_fd
; i
++) {
864 /* Fetch once the poll data */
865 revents
= LTTNG_POLL_GETEV(&events
, i
);
866 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
868 health_code_update();
870 /* Thread quit pipe has been closed. Killing thread. */
871 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
877 /* Check for data on kernel pipe */
878 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
879 (void) lttng_read(kernel_poll_pipe
[0],
882 * Ret value is useless here, if this pipe gets any actions an
883 * update is required anyway.
885 update_poll_flag
= 1;
889 * New CPU detected by the kernel. Adding kernel stream to
890 * kernel session and updating the kernel consumer
892 if (revents
& LPOLLIN
) {
893 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
899 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
900 * and unregister kernel stream at this point.
909 lttng_poll_clean(&events
);
912 utils_close_pipe(kernel_poll_pipe
);
913 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
916 ERR("Health error occurred in %s", __func__
);
917 WARN("Kernel thread died unexpectedly. "
918 "Kernel tracing can continue but CPU hotplug is disabled.");
920 health_unregister(health_sessiond
);
921 DBG("Kernel thread dying");
926 * Signal pthread condition of the consumer data that the thread.
928 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
930 pthread_mutex_lock(&data
->cond_mutex
);
933 * The state is set before signaling. It can be any value, it's the waiter
934 * job to correctly interpret this condition variable associated to the
935 * consumer pthread_cond.
937 * A value of 0 means that the corresponding thread of the consumer data
938 * was not started. 1 indicates that the thread has started and is ready
939 * for action. A negative value means that there was an error during the
942 data
->consumer_thread_is_ready
= state
;
943 (void) pthread_cond_signal(&data
->cond
);
945 pthread_mutex_unlock(&data
->cond_mutex
);
949 * This thread manage the consumer error sent back to the session daemon.
951 static void *thread_manage_consumer(void *data
)
953 int sock
= -1, i
, ret
, pollfd
, err
= -1;
954 uint32_t revents
, nb_fd
;
955 enum lttcomm_return_code code
;
956 struct lttng_poll_event events
;
957 struct consumer_data
*consumer_data
= data
;
959 DBG("[thread] Manage consumer started");
961 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
963 health_code_update();
966 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
967 * metadata_sock. Nothing more will be added to this poll set.
969 ret
= sessiond_set_thread_pollset(&events
, 3);
975 * The error socket here is already in a listening state which was done
976 * just before spawning this thread to avoid a race between the consumer
977 * daemon exec trying to connect and the listen() call.
979 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
984 health_code_update();
986 /* Infinite blocking call, waiting for transmission */
990 if (testpoint(thread_manage_consumer
)) {
994 ret
= lttng_poll_wait(&events
, -1);
998 * Restart interrupted system call.
1000 if (errno
== EINTR
) {
1008 for (i
= 0; i
< nb_fd
; i
++) {
1009 /* Fetch once the poll data */
1010 revents
= LTTNG_POLL_GETEV(&events
, i
);
1011 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1013 health_code_update();
1015 /* Thread quit pipe has been closed. Killing thread. */
1016 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1022 /* Event on the registration socket */
1023 if (pollfd
== consumer_data
->err_sock
) {
1024 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1025 ERR("consumer err socket poll error");
1031 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1037 * Set the CLOEXEC flag. Return code is useless because either way, the
1040 (void) utils_set_fd_cloexec(sock
);
1042 health_code_update();
1044 DBG2("Receiving code from consumer err_sock");
1046 /* Getting status code from kconsumerd */
1047 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1048 sizeof(enum lttcomm_return_code
));
1053 health_code_update();
1055 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1056 /* Connect both socket, command and metadata. */
1057 consumer_data
->cmd_sock
=
1058 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1059 consumer_data
->metadata_fd
=
1060 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1061 if (consumer_data
->cmd_sock
< 0
1062 || consumer_data
->metadata_fd
< 0) {
1063 PERROR("consumer connect cmd socket");
1064 /* On error, signal condition and quit. */
1065 signal_consumer_condition(consumer_data
, -1);
1068 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1069 /* Create metadata socket lock. */
1070 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1071 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1072 PERROR("zmalloc pthread mutex");
1076 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1078 signal_consumer_condition(consumer_data
, 1);
1079 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1080 DBG("Consumer metadata socket ready (fd: %d)",
1081 consumer_data
->metadata_fd
);
1083 ERR("consumer error when waiting for SOCK_READY : %s",
1084 lttcomm_get_readable_code(-code
));
1088 /* Remove the consumerd error sock since we've established a connexion */
1089 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1094 /* Add new accepted error socket. */
1095 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1100 /* Add metadata socket that is successfully connected. */
1101 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1102 LPOLLIN
| LPOLLRDHUP
);
1107 health_code_update();
1109 /* Infinite blocking call, waiting for transmission */
1112 health_poll_entry();
1113 ret
= lttng_poll_wait(&events
, -1);
1117 * Restart interrupted system call.
1119 if (errno
== EINTR
) {
1127 for (i
= 0; i
< nb_fd
; i
++) {
1128 /* Fetch once the poll data */
1129 revents
= LTTNG_POLL_GETEV(&events
, i
);
1130 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1132 health_code_update();
1134 /* Thread quit pipe has been closed. Killing thread. */
1135 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1141 if (pollfd
== sock
) {
1142 /* Event on the consumerd socket */
1143 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1144 ERR("consumer err socket second poll error");
1147 health_code_update();
1148 /* Wait for any kconsumerd error */
1149 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1150 sizeof(enum lttcomm_return_code
));
1152 ERR("consumer closed the command socket");
1156 ERR("consumer return code : %s",
1157 lttcomm_get_readable_code(-code
));
1160 } else if (pollfd
== consumer_data
->metadata_fd
) {
1161 /* UST metadata requests */
1162 ret
= ust_consumer_metadata_request(
1163 &consumer_data
->metadata_sock
);
1165 ERR("Handling metadata request");
1170 ERR("Unknown pollfd");
1174 health_code_update();
1180 * We lock here because we are about to close the sockets and some other
1181 * thread might be using them so get exclusive access which will abort all
1182 * other consumer command by other threads.
1184 pthread_mutex_lock(&consumer_data
->lock
);
1186 /* Immediately set the consumerd state to stopped */
1187 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1188 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1189 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1190 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1191 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1193 /* Code flow error... */
1197 if (consumer_data
->err_sock
>= 0) {
1198 ret
= close(consumer_data
->err_sock
);
1202 consumer_data
->err_sock
= -1;
1204 if (consumer_data
->cmd_sock
>= 0) {
1205 ret
= close(consumer_data
->cmd_sock
);
1209 consumer_data
->cmd_sock
= -1;
1211 if (*consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1212 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1225 unlink(consumer_data
->err_unix_sock_path
);
1226 unlink(consumer_data
->cmd_unix_sock_path
);
1227 consumer_data
->pid
= 0;
1228 pthread_mutex_unlock(&consumer_data
->lock
);
1230 /* Cleanup metadata socket mutex. */
1231 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1232 free(consumer_data
->metadata_sock
.lock
);
1234 lttng_poll_clean(&events
);
1238 ERR("Health error occurred in %s", __func__
);
1240 health_unregister(health_sessiond
);
1241 DBG("consumer thread cleanup completed");
1247 * This thread manage application communication.
1249 static void *thread_manage_apps(void *data
)
1251 int i
, ret
, pollfd
, err
= -1;
1253 uint32_t revents
, nb_fd
;
1254 struct lttng_poll_event events
;
1256 DBG("[thread] Manage application started");
1258 rcu_register_thread();
1259 rcu_thread_online();
1261 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1263 if (testpoint(thread_manage_apps
)) {
1264 goto error_testpoint
;
1267 health_code_update();
1269 ret
= sessiond_set_thread_pollset(&events
, 2);
1271 goto error_poll_create
;
1274 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1279 if (testpoint(thread_manage_apps_before_loop
)) {
1283 health_code_update();
1286 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1288 /* Inifinite blocking call, waiting for transmission */
1290 health_poll_entry();
1291 ret
= lttng_poll_wait(&events
, -1);
1295 * Restart interrupted system call.
1297 if (errno
== EINTR
) {
1305 for (i
= 0; i
< nb_fd
; i
++) {
1306 /* Fetch once the poll data */
1307 revents
= LTTNG_POLL_GETEV(&events
, i
);
1308 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1310 health_code_update();
1312 /* Thread quit pipe has been closed. Killing thread. */
1313 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1319 /* Inspect the apps cmd pipe */
1320 if (pollfd
== apps_cmd_pipe
[0]) {
1321 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1322 ERR("Apps command pipe error");
1324 } else if (revents
& LPOLLIN
) {
1328 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1329 if (size_ret
< sizeof(sock
)) {
1330 PERROR("read apps cmd pipe");
1334 health_code_update();
1337 * We only monitor the error events of the socket. This
1338 * thread does not handle any incoming data from UST
1341 ret
= lttng_poll_add(&events
, sock
,
1342 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1347 DBG("Apps with sock %d added to poll set", sock
);
1349 health_code_update();
1355 * At this point, we know that a registered application made
1356 * the event at poll_wait.
1358 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1359 /* Removing from the poll set */
1360 ret
= lttng_poll_del(&events
, pollfd
);
1365 /* Socket closed on remote end. */
1366 ust_app_unregister(pollfd
);
1371 health_code_update();
1377 lttng_poll_clean(&events
);
1380 utils_close_pipe(apps_cmd_pipe
);
1381 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1384 * We don't clean the UST app hash table here since already registered
1385 * applications can still be controlled so let them be until the session
1386 * daemon dies or the applications stop.
1391 ERR("Health error occurred in %s", __func__
);
1393 health_unregister(health_sessiond
);
1394 DBG("Application communication apps thread cleanup complete");
1395 rcu_thread_offline();
1396 rcu_unregister_thread();
1401 * Send a socket to a thread This is called from the dispatch UST registration
1402 * thread once all sockets are set for the application.
1404 * The sock value can be invalid, we don't really care, the thread will handle
1405 * it and make the necessary cleanup if so.
1407 * On success, return 0 else a negative value being the errno message of the
1410 static int send_socket_to_thread(int fd
, int sock
)
1415 * It's possible that the FD is set as invalid with -1 concurrently just
1416 * before calling this function being a shutdown state of the thread.
1423 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1424 if (ret
< sizeof(sock
)) {
1425 PERROR("write apps pipe %d", fd
);
1432 /* All good. Don't send back the write positive ret value. */
1439 * Sanitize the wait queue of the dispatch registration thread meaning removing
1440 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1441 * notify socket is never received.
1443 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1445 int ret
, nb_fd
= 0, i
;
1446 unsigned int fd_added
= 0;
1447 struct lttng_poll_event events
;
1448 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1452 lttng_poll_init(&events
);
1454 /* Just skip everything for an empty queue. */
1455 if (!wait_queue
->count
) {
1459 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1464 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1465 &wait_queue
->head
, head
) {
1466 assert(wait_node
->app
);
1467 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1468 LPOLLHUP
| LPOLLERR
);
1481 * Poll but don't block so we can quickly identify the faulty events and
1482 * clean them afterwards from the wait queue.
1484 ret
= lttng_poll_wait(&events
, 0);
1490 for (i
= 0; i
< nb_fd
; i
++) {
1491 /* Get faulty FD. */
1492 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1493 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1495 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1496 &wait_queue
->head
, head
) {
1497 if (pollfd
== wait_node
->app
->sock
&&
1498 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1499 cds_list_del(&wait_node
->head
);
1500 wait_queue
->count
--;
1501 ust_app_destroy(wait_node
->app
);
1509 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1513 lttng_poll_clean(&events
);
1517 lttng_poll_clean(&events
);
1519 ERR("Unable to sanitize wait queue");
1524 * Dispatch request from the registration threads to the application
1525 * communication thread.
1527 static void *thread_dispatch_ust_registration(void *data
)
1530 struct cds_wfq_node
*node
;
1531 struct ust_command
*ust_cmd
= NULL
;
1532 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1533 struct ust_reg_wait_queue wait_queue
= {
1537 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1539 health_code_update();
1541 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1543 DBG("[thread] Dispatch UST command started");
1545 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1546 health_code_update();
1548 /* Atomically prepare the queue futex */
1549 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1552 struct ust_app
*app
= NULL
;
1556 * Make sure we don't have node(s) that have hung up before receiving
1557 * the notify socket. This is to clean the list in order to avoid
1558 * memory leaks from notify socket that are never seen.
1560 sanitize_wait_queue(&wait_queue
);
1562 health_code_update();
1563 /* Dequeue command for registration */
1564 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1566 DBG("Woken up but nothing in the UST command queue");
1567 /* Continue thread execution */
1571 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1573 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1574 " gid:%d sock:%d name:%s (version %d.%d)",
1575 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1576 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1577 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1578 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1580 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1581 wait_node
= zmalloc(sizeof(*wait_node
));
1583 PERROR("zmalloc wait_node dispatch");
1584 ret
= close(ust_cmd
->sock
);
1586 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1588 lttng_fd_put(LTTNG_FD_APPS
, 1);
1592 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1594 /* Create application object if socket is CMD. */
1595 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1597 if (!wait_node
->app
) {
1598 ret
= close(ust_cmd
->sock
);
1600 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1602 lttng_fd_put(LTTNG_FD_APPS
, 1);
1608 * Add application to the wait queue so we can set the notify
1609 * socket before putting this object in the global ht.
1611 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1616 * We have to continue here since we don't have the notify
1617 * socket and the application MUST be added to the hash table
1618 * only at that moment.
1623 * Look for the application in the local wait queue and set the
1624 * notify socket if found.
1626 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1627 &wait_queue
.head
, head
) {
1628 health_code_update();
1629 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1630 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1631 cds_list_del(&wait_node
->head
);
1633 app
= wait_node
->app
;
1635 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1641 * With no application at this stage the received socket is
1642 * basically useless so close it before we free the cmd data
1643 * structure for good.
1646 ret
= close(ust_cmd
->sock
);
1648 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1650 lttng_fd_put(LTTNG_FD_APPS
, 1);
1657 * @session_lock_list
1659 * Lock the global session list so from the register up to the
1660 * registration done message, no thread can see the application
1661 * and change its state.
1663 session_lock_list();
1667 * Add application to the global hash table. This needs to be
1668 * done before the update to the UST registry can locate the
1673 /* Set app version. This call will print an error if needed. */
1674 (void) ust_app_version(app
);
1676 /* Send notify socket through the notify pipe. */
1677 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1681 session_unlock_list();
1683 * No notify thread, stop the UST tracing. However, this is
1684 * not an internal error of the this thread thus setting
1685 * the health error code to a normal exit.
1692 * Update newly registered application with the tracing
1693 * registry info already enabled information.
1695 update_ust_app(app
->sock
);
1698 * Don't care about return value. Let the manage apps threads
1699 * handle app unregistration upon socket close.
1701 (void) ust_app_register_done(app
->sock
);
1704 * Even if the application socket has been closed, send the app
1705 * to the thread and unregistration will take place at that
1708 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1711 session_unlock_list();
1713 * No apps. thread, stop the UST tracing. However, this is
1714 * not an internal error of the this thread thus setting
1715 * the health error code to a normal exit.
1722 session_unlock_list();
1724 } while (node
!= NULL
);
1726 health_poll_entry();
1727 /* Futex wait on queue. Blocking call on futex() */
1728 futex_nto1_wait(&ust_cmd_queue
.futex
);
1731 /* Normal exit, no error */
1735 /* Clean up wait queue. */
1736 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1737 &wait_queue
.head
, head
) {
1738 cds_list_del(&wait_node
->head
);
1743 DBG("Dispatch thread dying");
1746 ERR("Health error occurred in %s", __func__
);
1748 health_unregister(health_sessiond
);
1753 * This thread manage application registration.
1755 static void *thread_registration_apps(void *data
)
1757 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1758 uint32_t revents
, nb_fd
;
1759 struct lttng_poll_event events
;
1761 * Get allocated in this thread, enqueued to a global queue, dequeued and
1762 * freed in the manage apps thread.
1764 struct ust_command
*ust_cmd
= NULL
;
1766 DBG("[thread] Manage application registration started");
1768 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1770 if (testpoint(thread_registration_apps
)) {
1771 goto error_testpoint
;
1774 ret
= lttcomm_listen_unix_sock(apps_sock
);
1780 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1781 * more will be added to this poll set.
1783 ret
= sessiond_set_thread_pollset(&events
, 2);
1785 goto error_create_poll
;
1788 /* Add the application registration socket */
1789 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1791 goto error_poll_add
;
1794 /* Notify all applications to register */
1795 ret
= notify_ust_apps(1);
1797 ERR("Failed to notify applications or create the wait shared memory.\n"
1798 "Execution continues but there might be problem for already\n"
1799 "running applications that wishes to register.");
1803 DBG("Accepting application registration");
1805 /* Inifinite blocking call, waiting for transmission */
1807 health_poll_entry();
1808 ret
= lttng_poll_wait(&events
, -1);
1812 * Restart interrupted system call.
1814 if (errno
== EINTR
) {
1822 for (i
= 0; i
< nb_fd
; i
++) {
1823 health_code_update();
1825 /* Fetch once the poll data */
1826 revents
= LTTNG_POLL_GETEV(&events
, i
);
1827 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1829 /* Thread quit pipe has been closed. Killing thread. */
1830 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1836 /* Event on the registration socket */
1837 if (pollfd
== apps_sock
) {
1838 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1839 ERR("Register apps socket poll error");
1841 } else if (revents
& LPOLLIN
) {
1842 sock
= lttcomm_accept_unix_sock(apps_sock
);
1848 * Set socket timeout for both receiving and ending.
1849 * app_socket_timeout is in seconds, whereas
1850 * lttcomm_setsockopt_rcv_timeout and
1851 * lttcomm_setsockopt_snd_timeout expect msec as
1854 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1855 app_socket_timeout
* 1000);
1856 (void) lttcomm_setsockopt_snd_timeout(sock
,
1857 app_socket_timeout
* 1000);
1860 * Set the CLOEXEC flag. Return code is useless because
1861 * either way, the show must go on.
1863 (void) utils_set_fd_cloexec(sock
);
1865 /* Create UST registration command for enqueuing */
1866 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1867 if (ust_cmd
== NULL
) {
1868 PERROR("ust command zmalloc");
1873 * Using message-based transmissions to ensure we don't
1874 * have to deal with partially received messages.
1876 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1878 ERR("Exhausted file descriptors allowed for applications.");
1888 health_code_update();
1889 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1892 /* Close socket of the application. */
1897 lttng_fd_put(LTTNG_FD_APPS
, 1);
1901 health_code_update();
1903 ust_cmd
->sock
= sock
;
1906 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1907 " gid:%d sock:%d name:%s (version %d.%d)",
1908 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1909 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1910 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1911 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1914 * Lock free enqueue the registration request. The red pill
1915 * has been taken! This apps will be part of the *system*.
1917 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1920 * Wake the registration queue futex. Implicit memory
1921 * barrier with the exchange in cds_wfq_enqueue.
1923 futex_nto1_wake(&ust_cmd_queue
.futex
);
1933 ERR("Health error occurred in %s", __func__
);
1936 /* Notify that the registration thread is gone */
1939 if (apps_sock
>= 0) {
1940 ret
= close(apps_sock
);
1950 lttng_fd_put(LTTNG_FD_APPS
, 1);
1952 unlink(apps_unix_sock_path
);
1955 lttng_poll_clean(&events
);
1959 DBG("UST Registration thread cleanup complete");
1960 health_unregister(health_sessiond
);
1966 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1967 * exec or it will fails.
1969 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1972 struct timespec timeout
;
1974 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1975 consumer_data
->consumer_thread_is_ready
= 0;
1977 /* Setup pthread condition */
1978 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1981 PERROR("pthread_condattr_init consumer data");
1986 * Set the monotonic clock in order to make sure we DO NOT jump in time
1987 * between the clock_gettime() call and the timedwait call. See bug #324
1988 * for a more details and how we noticed it.
1990 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1993 PERROR("pthread_condattr_setclock consumer data");
1997 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2000 PERROR("pthread_cond_init consumer data");
2004 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2007 PERROR("pthread_create consumer");
2012 /* We are about to wait on a pthread condition */
2013 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2015 /* Get time for sem_timedwait absolute timeout */
2016 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2018 * Set the timeout for the condition timed wait even if the clock gettime
2019 * call fails since we might loop on that call and we want to avoid to
2020 * increment the timeout too many times.
2022 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2025 * The following loop COULD be skipped in some conditions so this is why we
2026 * set ret to 0 in order to make sure at least one round of the loop is
2032 * Loop until the condition is reached or when a timeout is reached. Note
2033 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2034 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2035 * possible. This loop does not take any chances and works with both of
2038 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2039 if (clock_ret
< 0) {
2040 PERROR("clock_gettime spawn consumer");
2041 /* Infinite wait for the consumerd thread to be ready */
2042 ret
= pthread_cond_wait(&consumer_data
->cond
,
2043 &consumer_data
->cond_mutex
);
2045 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2046 &consumer_data
->cond_mutex
, &timeout
);
2050 /* Release the pthread condition */
2051 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2055 if (ret
== ETIMEDOUT
) {
2057 * Call has timed out so we kill the kconsumerd_thread and return
2060 ERR("Condition timed out. The consumer thread was never ready."
2062 ret
= pthread_cancel(consumer_data
->thread
);
2064 PERROR("pthread_cancel consumer thread");
2067 PERROR("pthread_cond_wait failed consumer thread");
2072 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2073 if (consumer_data
->pid
== 0) {
2074 ERR("Consumerd did not start");
2075 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2078 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2087 * Join consumer thread
2089 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2093 /* Consumer pid must be a real one. */
2094 if (consumer_data
->pid
> 0) {
2096 ret
= kill(consumer_data
->pid
, SIGTERM
);
2098 ERR("Error killing consumer daemon");
2101 return pthread_join(consumer_data
->thread
, &status
);
2108 * Fork and exec a consumer daemon (consumerd).
2110 * Return pid if successful else -1.
2112 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2116 const char *consumer_to_use
;
2117 const char *verbosity
;
2120 DBG("Spawning consumerd");
2127 if (opt_verbose_consumer
) {
2128 verbosity
= "--verbose";
2130 verbosity
= "--quiet";
2132 switch (consumer_data
->type
) {
2133 case LTTNG_CONSUMER_KERNEL
:
2135 * Find out which consumerd to execute. We will first try the
2136 * 64-bit path, then the sessiond's installation directory, and
2137 * fallback on the 32-bit one,
2139 DBG3("Looking for a kernel consumer at these locations:");
2140 DBG3(" 1) %s", consumerd64_bin
);
2141 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2142 DBG3(" 3) %s", consumerd32_bin
);
2143 if (stat(consumerd64_bin
, &st
) == 0) {
2144 DBG3("Found location #1");
2145 consumer_to_use
= consumerd64_bin
;
2146 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2147 DBG3("Found location #2");
2148 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2149 } else if (stat(consumerd32_bin
, &st
) == 0) {
2150 DBG3("Found location #3");
2151 consumer_to_use
= consumerd32_bin
;
2153 DBG("Could not find any valid consumerd executable");
2156 DBG("Using kernel consumer at: %s", consumer_to_use
);
2157 execl(consumer_to_use
,
2158 "lttng-consumerd", verbosity
, "-k",
2159 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2160 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2161 "--group", tracing_group_name
,
2164 case LTTNG_CONSUMER64_UST
:
2166 char *tmpnew
= NULL
;
2168 if (consumerd64_libdir
[0] != '\0') {
2172 tmp
= getenv("LD_LIBRARY_PATH");
2176 tmplen
= strlen("LD_LIBRARY_PATH=")
2177 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2178 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2183 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2184 strcat(tmpnew
, consumerd64_libdir
);
2185 if (tmp
[0] != '\0') {
2186 strcat(tmpnew
, ":");
2187 strcat(tmpnew
, tmp
);
2189 ret
= putenv(tmpnew
);
2196 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2197 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2198 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2199 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2200 "--group", tracing_group_name
,
2202 if (consumerd64_libdir
[0] != '\0') {
2210 case LTTNG_CONSUMER32_UST
:
2212 char *tmpnew
= NULL
;
2214 if (consumerd32_libdir
[0] != '\0') {
2218 tmp
= getenv("LD_LIBRARY_PATH");
2222 tmplen
= strlen("LD_LIBRARY_PATH=")
2223 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2224 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2229 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2230 strcat(tmpnew
, consumerd32_libdir
);
2231 if (tmp
[0] != '\0') {
2232 strcat(tmpnew
, ":");
2233 strcat(tmpnew
, tmp
);
2235 ret
= putenv(tmpnew
);
2242 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2243 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2244 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2245 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2246 "--group", tracing_group_name
,
2248 if (consumerd32_libdir
[0] != '\0') {
2257 PERROR("unknown consumer type");
2261 PERROR("kernel start consumer exec");
2264 } else if (pid
> 0) {
2267 PERROR("start consumer fork");
2275 * Spawn the consumerd daemon and session daemon thread.
2277 static int start_consumerd(struct consumer_data
*consumer_data
)
2282 * Set the listen() state on the socket since there is a possible race
2283 * between the exec() of the consumer daemon and this call if place in the
2284 * consumer thread. See bug #366 for more details.
2286 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2291 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2292 if (consumer_data
->pid
!= 0) {
2293 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2297 ret
= spawn_consumerd(consumer_data
);
2299 ERR("Spawning consumerd failed");
2300 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2304 /* Setting up the consumer_data pid */
2305 consumer_data
->pid
= ret
;
2306 DBG2("Consumer pid %d", consumer_data
->pid
);
2307 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2309 DBG2("Spawning consumer control thread");
2310 ret
= spawn_consumer_thread(consumer_data
);
2312 ERR("Fatal error spawning consumer control thread");
2320 /* Cleanup already created sockets on error. */
2321 if (consumer_data
->err_sock
>= 0) {
2324 err
= close(consumer_data
->err_sock
);
2326 PERROR("close consumer data error socket");
2333 * Setup necessary data for kernel tracer action.
2335 static int init_kernel_tracer(void)
2339 /* Modprobe lttng kernel modules */
2340 ret
= modprobe_lttng_control();
2345 /* Open debugfs lttng */
2346 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2347 if (kernel_tracer_fd
< 0) {
2348 DBG("Failed to open %s", module_proc_lttng
);
2353 /* Validate kernel version */
2354 ret
= kernel_validate_version(kernel_tracer_fd
);
2359 ret
= modprobe_lttng_data();
2364 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2368 modprobe_remove_lttng_control();
2369 ret
= close(kernel_tracer_fd
);
2373 kernel_tracer_fd
= -1;
2374 return LTTNG_ERR_KERN_VERSION
;
2377 ret
= close(kernel_tracer_fd
);
2383 modprobe_remove_lttng_control();
2386 WARN("No kernel tracer available");
2387 kernel_tracer_fd
= -1;
2389 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2391 return LTTNG_ERR_KERN_NA
;
2397 * Copy consumer output from the tracing session to the domain session. The
2398 * function also applies the right modification on a per domain basis for the
2399 * trace files destination directory.
2401 * Should *NOT* be called with RCU read-side lock held.
2403 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2406 const char *dir_name
;
2407 struct consumer_output
*consumer
;
2410 assert(session
->consumer
);
2413 case LTTNG_DOMAIN_KERNEL
:
2414 DBG3("Copying tracing session consumer output in kernel session");
2416 * XXX: We should audit the session creation and what this function
2417 * does "extra" in order to avoid a destroy since this function is used
2418 * in the domain session creation (kernel and ust) only. Same for UST
2421 if (session
->kernel_session
->consumer
) {
2422 consumer_destroy_output(session
->kernel_session
->consumer
);
2424 session
->kernel_session
->consumer
=
2425 consumer_copy_output(session
->consumer
);
2426 /* Ease our life a bit for the next part */
2427 consumer
= session
->kernel_session
->consumer
;
2428 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2430 case LTTNG_DOMAIN_JUL
:
2431 case LTTNG_DOMAIN_UST
:
2432 DBG3("Copying tracing session consumer output in UST session");
2433 if (session
->ust_session
->consumer
) {
2434 consumer_destroy_output(session
->ust_session
->consumer
);
2436 session
->ust_session
->consumer
=
2437 consumer_copy_output(session
->consumer
);
2438 /* Ease our life a bit for the next part */
2439 consumer
= session
->ust_session
->consumer
;
2440 dir_name
= DEFAULT_UST_TRACE_DIR
;
2443 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2447 /* Append correct directory to subdir */
2448 strncat(consumer
->subdir
, dir_name
,
2449 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2450 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2459 * Create an UST session and add it to the session ust list.
2461 * Should *NOT* be called with RCU read-side lock held.
2463 static int create_ust_session(struct ltt_session
*session
,
2464 struct lttng_domain
*domain
)
2467 struct ltt_ust_session
*lus
= NULL
;
2471 assert(session
->consumer
);
2473 switch (domain
->type
) {
2474 case LTTNG_DOMAIN_JUL
:
2475 case LTTNG_DOMAIN_UST
:
2478 ERR("Unknown UST domain on create session %d", domain
->type
);
2479 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2483 DBG("Creating UST session");
2485 lus
= trace_ust_create_session(session
->id
);
2487 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2491 lus
->uid
= session
->uid
;
2492 lus
->gid
= session
->gid
;
2493 lus
->output_traces
= session
->output_traces
;
2494 lus
->snapshot_mode
= session
->snapshot_mode
;
2495 lus
->live_timer_interval
= session
->live_timer
;
2496 session
->ust_session
= lus
;
2498 /* Copy session output to the newly created UST session */
2499 ret
= copy_session_consumer(domain
->type
, session
);
2500 if (ret
!= LTTNG_OK
) {
2508 session
->ust_session
= NULL
;
2513 * Create a kernel tracer session then create the default channel.
2515 static int create_kernel_session(struct ltt_session
*session
)
2519 DBG("Creating kernel session");
2521 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2523 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2527 /* Code flow safety */
2528 assert(session
->kernel_session
);
2530 /* Copy session output to the newly created Kernel session */
2531 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2532 if (ret
!= LTTNG_OK
) {
2536 /* Create directory(ies) on local filesystem. */
2537 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2538 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2539 ret
= run_as_mkdir_recursive(
2540 session
->kernel_session
->consumer
->dst
.trace_path
,
2541 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2543 if (ret
!= -EEXIST
) {
2544 ERR("Trace directory creation error");
2550 session
->kernel_session
->uid
= session
->uid
;
2551 session
->kernel_session
->gid
= session
->gid
;
2552 session
->kernel_session
->output_traces
= session
->output_traces
;
2553 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2558 trace_kernel_destroy_session(session
->kernel_session
);
2559 session
->kernel_session
= NULL
;
2564 * Count number of session permitted by uid/gid.
2566 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2569 struct ltt_session
*session
;
2571 DBG("Counting number of available session for UID %d GID %d",
2573 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2575 * Only list the sessions the user can control.
2577 if (!session_access_ok(session
, uid
, gid
)) {
2586 * Process the command requested by the lttng client within the command
2587 * context structure. This function make sure that the return structure (llm)
2588 * is set and ready for transmission before returning.
2590 * Return any error encountered or 0 for success.
2592 * "sock" is only used for special-case var. len data.
2594 * Should *NOT* be called with RCU read-side lock held.
2596 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2600 int need_tracing_session
= 1;
2603 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2607 switch (cmd_ctx
->lsm
->cmd_type
) {
2608 case LTTNG_CREATE_SESSION
:
2609 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2610 case LTTNG_CREATE_SESSION_LIVE
:
2611 case LTTNG_DESTROY_SESSION
:
2612 case LTTNG_LIST_SESSIONS
:
2613 case LTTNG_LIST_DOMAINS
:
2614 case LTTNG_START_TRACE
:
2615 case LTTNG_STOP_TRACE
:
2616 case LTTNG_DATA_PENDING
:
2617 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2618 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2619 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2620 case LTTNG_SNAPSHOT_RECORD
:
2627 if (opt_no_kernel
&& need_domain
2628 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2630 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2632 ret
= LTTNG_ERR_KERN_NA
;
2637 /* Deny register consumer if we already have a spawned consumer. */
2638 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2639 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2640 if (kconsumer_data
.pid
> 0) {
2641 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2642 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2645 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2649 * Check for command that don't needs to allocate a returned payload. We do
2650 * this here so we don't have to make the call for no payload at each
2653 switch(cmd_ctx
->lsm
->cmd_type
) {
2654 case LTTNG_LIST_SESSIONS
:
2655 case LTTNG_LIST_TRACEPOINTS
:
2656 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2657 case LTTNG_LIST_DOMAINS
:
2658 case LTTNG_LIST_CHANNELS
:
2659 case LTTNG_LIST_EVENTS
:
2662 /* Setup lttng message with no payload */
2663 ret
= setup_lttng_msg(cmd_ctx
, 0);
2665 /* This label does not try to unlock the session */
2666 goto init_setup_error
;
2670 /* Commands that DO NOT need a session. */
2671 switch (cmd_ctx
->lsm
->cmd_type
) {
2672 case LTTNG_CREATE_SESSION
:
2673 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2674 case LTTNG_CREATE_SESSION_LIVE
:
2675 case LTTNG_CALIBRATE
:
2676 case LTTNG_LIST_SESSIONS
:
2677 case LTTNG_LIST_TRACEPOINTS
:
2678 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2679 need_tracing_session
= 0;
2682 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2684 * We keep the session list lock across _all_ commands
2685 * for now, because the per-session lock does not
2686 * handle teardown properly.
2688 session_lock_list();
2689 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2690 if (cmd_ctx
->session
== NULL
) {
2691 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2694 /* Acquire lock for the session */
2695 session_lock(cmd_ctx
->session
);
2705 * Check domain type for specific "pre-action".
2707 switch (cmd_ctx
->lsm
->domain
.type
) {
2708 case LTTNG_DOMAIN_KERNEL
:
2710 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2714 /* Kernel tracer check */
2715 if (kernel_tracer_fd
== -1) {
2716 /* Basically, load kernel tracer modules */
2717 ret
= init_kernel_tracer();
2723 /* Consumer is in an ERROR state. Report back to client */
2724 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2725 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2729 /* Need a session for kernel command */
2730 if (need_tracing_session
) {
2731 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2732 ret
= create_kernel_session(cmd_ctx
->session
);
2734 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2739 /* Start the kernel consumer daemon */
2740 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2741 if (kconsumer_data
.pid
== 0 &&
2742 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2743 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2744 ret
= start_consumerd(&kconsumer_data
);
2746 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2749 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2751 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2755 * The consumer was just spawned so we need to add the socket to
2756 * the consumer output of the session if exist.
2758 ret
= consumer_create_socket(&kconsumer_data
,
2759 cmd_ctx
->session
->kernel_session
->consumer
);
2766 case LTTNG_DOMAIN_JUL
:
2767 case LTTNG_DOMAIN_UST
:
2769 if (!ust_app_supported()) {
2770 ret
= LTTNG_ERR_NO_UST
;
2773 /* Consumer is in an ERROR state. Report back to client */
2774 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2775 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2779 if (need_tracing_session
) {
2780 /* Create UST session if none exist. */
2781 if (cmd_ctx
->session
->ust_session
== NULL
) {
2782 ret
= create_ust_session(cmd_ctx
->session
,
2783 &cmd_ctx
->lsm
->domain
);
2784 if (ret
!= LTTNG_OK
) {
2789 /* Start the UST consumer daemons */
2791 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2792 if (consumerd64_bin
[0] != '\0' &&
2793 ustconsumer64_data
.pid
== 0 &&
2794 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2795 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2796 ret
= start_consumerd(&ustconsumer64_data
);
2798 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2799 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2803 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2804 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2806 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2810 * Setup socket for consumer 64 bit. No need for atomic access
2811 * since it was set above and can ONLY be set in this thread.
2813 ret
= consumer_create_socket(&ustconsumer64_data
,
2814 cmd_ctx
->session
->ust_session
->consumer
);
2820 if (consumerd32_bin
[0] != '\0' &&
2821 ustconsumer32_data
.pid
== 0 &&
2822 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2823 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2824 ret
= start_consumerd(&ustconsumer32_data
);
2826 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2827 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2831 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2832 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2834 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2838 * Setup socket for consumer 64 bit. No need for atomic access
2839 * since it was set above and can ONLY be set in this thread.
2841 ret
= consumer_create_socket(&ustconsumer32_data
,
2842 cmd_ctx
->session
->ust_session
->consumer
);
2854 /* Validate consumer daemon state when start/stop trace command */
2855 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2856 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2857 switch (cmd_ctx
->lsm
->domain
.type
) {
2858 case LTTNG_DOMAIN_JUL
:
2859 case LTTNG_DOMAIN_UST
:
2860 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2861 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2865 case LTTNG_DOMAIN_KERNEL
:
2866 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2867 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2875 * Check that the UID or GID match that of the tracing session.
2876 * The root user can interact with all sessions.
2878 if (need_tracing_session
) {
2879 if (!session_access_ok(cmd_ctx
->session
,
2880 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2881 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2882 ret
= LTTNG_ERR_EPERM
;
2888 * Send relayd information to consumer as soon as we have a domain and a
2891 if (cmd_ctx
->session
&& need_domain
) {
2893 * Setup relayd if not done yet. If the relayd information was already
2894 * sent to the consumer, this call will gracefully return.
2896 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2897 if (ret
!= LTTNG_OK
) {
2902 /* Process by command type */
2903 switch (cmd_ctx
->lsm
->cmd_type
) {
2904 case LTTNG_ADD_CONTEXT
:
2906 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2907 cmd_ctx
->lsm
->u
.context
.channel_name
,
2908 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2911 case LTTNG_DISABLE_CHANNEL
:
2913 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2914 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2917 case LTTNG_DISABLE_EVENT
:
2919 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2920 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2921 cmd_ctx
->lsm
->u
.disable
.name
);
2924 case LTTNG_DISABLE_ALL_EVENT
:
2926 DBG("Disabling all events");
2928 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2929 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2932 case LTTNG_ENABLE_CHANNEL
:
2934 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2935 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2938 case LTTNG_ENABLE_EVENT
:
2939 case LTTNG_ENABLE_EVENT_WITH_EXCLUSION
:
2940 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
2942 struct lttng_event_exclusion
*exclusion
= NULL
;
2943 struct lttng_filter_bytecode
*bytecode
= NULL
;
2945 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_ENABLE_EVENT
||
2946 (cmd_ctx
->lsm
->u
.enable
.exclusion_count
== 0 && cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0)) {
2947 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2948 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2949 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, NULL
, kernel_poll_pipe
[1]);
2951 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
!= 0) {
2952 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
2953 cmd_ctx
->lsm
->u
.enable
.exclusion_count
* LTTNG_SYMBOL_NAME_LEN
);
2955 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
2958 DBG("Receiving var len data from client ...");
2959 exclusion
->count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
2960 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
2961 cmd_ctx
->lsm
->u
.enable
.exclusion_count
* LTTNG_SYMBOL_NAME_LEN
);
2963 DBG("Nothing recv() from client var len data... continuing");
2965 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
2969 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
!= 0) {
2970 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2974 ret
= LTTNG_ERR_FILTER_NOMEM
;
2977 /* Receive var. len. data */
2978 DBG("Receiving var len data from client ...");
2979 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
2980 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2982 DBG("Nothing recv() from client car len data... continuing");
2986 ret
= LTTNG_ERR_FILTER_INVAL
;
2990 if (bytecode
->len
+ sizeof(*bytecode
)
2991 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
2995 ret
= LTTNG_ERR_FILTER_INVAL
;
3000 ret
= cmd_enable_event(cmd_ctx
->session
,
3001 &cmd_ctx
->lsm
->domain
,
3002 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3003 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
,
3005 kernel_poll_pipe
[1]);
3009 case LTTNG_ENABLE_ALL_EVENT
:
3011 DBG("Enabling all events");
3013 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3014 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3015 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
3018 case LTTNG_LIST_TRACEPOINTS
:
3020 struct lttng_event
*events
;
3023 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3024 if (nb_events
< 0) {
3025 /* Return value is a negative lttng_error_code. */
3031 * Setup lttng message with payload size set to the event list size in
3032 * bytes and then copy list into the llm payload.
3034 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3040 /* Copy event list into message payload */
3041 memcpy(cmd_ctx
->llm
->payload
, events
,
3042 sizeof(struct lttng_event
) * nb_events
);
3049 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3051 struct lttng_event_field
*fields
;
3054 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3056 if (nb_fields
< 0) {
3057 /* Return value is a negative lttng_error_code. */
3063 * Setup lttng message with payload size set to the event list size in
3064 * bytes and then copy list into the llm payload.
3066 ret
= setup_lttng_msg(cmd_ctx
,
3067 sizeof(struct lttng_event_field
) * nb_fields
);
3073 /* Copy event list into message payload */
3074 memcpy(cmd_ctx
->llm
->payload
, fields
,
3075 sizeof(struct lttng_event_field
) * nb_fields
);
3082 case LTTNG_SET_CONSUMER_URI
:
3085 struct lttng_uri
*uris
;
3087 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3088 len
= nb_uri
* sizeof(struct lttng_uri
);
3091 ret
= LTTNG_ERR_INVALID
;
3095 uris
= zmalloc(len
);
3097 ret
= LTTNG_ERR_FATAL
;
3101 /* Receive variable len data */
3102 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3103 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3105 DBG("No URIs received from client... continuing");
3107 ret
= LTTNG_ERR_SESSION_FAIL
;
3112 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3114 if (ret
!= LTTNG_OK
) {
3120 * XXX: 0 means that this URI should be applied on the session. Should
3121 * be a DOMAIN enuam.
3123 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3124 /* Add the URI for the UST session if a consumer is present. */
3125 if (cmd_ctx
->session
->ust_session
&&
3126 cmd_ctx
->session
->ust_session
->consumer
) {
3127 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3129 } else if (cmd_ctx
->session
->kernel_session
&&
3130 cmd_ctx
->session
->kernel_session
->consumer
) {
3131 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3132 cmd_ctx
->session
, nb_uri
, uris
);
3140 case LTTNG_START_TRACE
:
3142 ret
= cmd_start_trace(cmd_ctx
->session
);
3145 case LTTNG_STOP_TRACE
:
3147 ret
= cmd_stop_trace(cmd_ctx
->session
);
3150 case LTTNG_CREATE_SESSION
:
3153 struct lttng_uri
*uris
= NULL
;
3155 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3156 len
= nb_uri
* sizeof(struct lttng_uri
);
3159 uris
= zmalloc(len
);
3161 ret
= LTTNG_ERR_FATAL
;
3165 /* Receive variable len data */
3166 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3167 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3169 DBG("No URIs received from client... continuing");
3171 ret
= LTTNG_ERR_SESSION_FAIL
;
3176 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3177 DBG("Creating session with ONE network URI is a bad call");
3178 ret
= LTTNG_ERR_SESSION_FAIL
;
3184 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3185 &cmd_ctx
->creds
, 0);
3191 case LTTNG_DESTROY_SESSION
:
3193 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3195 /* Set session to NULL so we do not unlock it after free. */
3196 cmd_ctx
->session
= NULL
;
3199 case LTTNG_LIST_DOMAINS
:
3202 struct lttng_domain
*domains
;
3204 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3206 /* Return value is a negative lttng_error_code. */
3211 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3217 /* Copy event list into message payload */
3218 memcpy(cmd_ctx
->llm
->payload
, domains
,
3219 nb_dom
* sizeof(struct lttng_domain
));
3226 case LTTNG_LIST_CHANNELS
:
3229 struct lttng_channel
*channels
;
3231 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3232 cmd_ctx
->session
, &channels
);
3234 /* Return value is a negative lttng_error_code. */
3239 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3245 /* Copy event list into message payload */
3246 memcpy(cmd_ctx
->llm
->payload
, channels
,
3247 nb_chan
* sizeof(struct lttng_channel
));
3254 case LTTNG_LIST_EVENTS
:
3257 struct lttng_event
*events
= NULL
;
3259 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3260 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3262 /* Return value is a negative lttng_error_code. */
3267 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3273 /* Copy event list into message payload */
3274 memcpy(cmd_ctx
->llm
->payload
, events
,
3275 nb_event
* sizeof(struct lttng_event
));
3282 case LTTNG_LIST_SESSIONS
:
3284 unsigned int nr_sessions
;
3286 session_lock_list();
3287 nr_sessions
= lttng_sessions_count(
3288 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3289 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3291 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3293 session_unlock_list();
3297 /* Filled the session array */
3298 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3299 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3300 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3302 session_unlock_list();
3307 case LTTNG_CALIBRATE
:
3309 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3310 &cmd_ctx
->lsm
->u
.calibrate
);
3313 case LTTNG_REGISTER_CONSUMER
:
3315 struct consumer_data
*cdata
;
3317 switch (cmd_ctx
->lsm
->domain
.type
) {
3318 case LTTNG_DOMAIN_KERNEL
:
3319 cdata
= &kconsumer_data
;
3322 ret
= LTTNG_ERR_UND
;
3326 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3327 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3330 case LTTNG_DATA_PENDING
:
3332 ret
= cmd_data_pending(cmd_ctx
->session
);
3335 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3337 struct lttcomm_lttng_output_id reply
;
3339 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3340 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3341 if (ret
!= LTTNG_OK
) {
3345 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3350 /* Copy output list into message payload */
3351 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3355 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3357 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3358 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3361 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3364 struct lttng_snapshot_output
*outputs
= NULL
;
3366 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3367 if (nb_output
< 0) {
3372 ret
= setup_lttng_msg(cmd_ctx
,
3373 nb_output
* sizeof(struct lttng_snapshot_output
));
3380 /* Copy output list into message payload */
3381 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3382 nb_output
* sizeof(struct lttng_snapshot_output
));
3389 case LTTNG_SNAPSHOT_RECORD
:
3391 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3392 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3393 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3396 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3399 struct lttng_uri
*uris
= NULL
;
3401 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3402 len
= nb_uri
* sizeof(struct lttng_uri
);
3405 uris
= zmalloc(len
);
3407 ret
= LTTNG_ERR_FATAL
;
3411 /* Receive variable len data */
3412 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3413 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3415 DBG("No URIs received from client... continuing");
3417 ret
= LTTNG_ERR_SESSION_FAIL
;
3422 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3423 DBG("Creating session with ONE network URI is a bad call");
3424 ret
= LTTNG_ERR_SESSION_FAIL
;
3430 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3431 nb_uri
, &cmd_ctx
->creds
);
3435 case LTTNG_CREATE_SESSION_LIVE
:
3438 struct lttng_uri
*uris
= NULL
;
3440 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3441 len
= nb_uri
* sizeof(struct lttng_uri
);
3444 uris
= zmalloc(len
);
3446 ret
= LTTNG_ERR_FATAL
;
3450 /* Receive variable len data */
3451 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3452 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3454 DBG("No URIs received from client... continuing");
3456 ret
= LTTNG_ERR_SESSION_FAIL
;
3461 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3462 DBG("Creating session with ONE network URI is a bad call");
3463 ret
= LTTNG_ERR_SESSION_FAIL
;
3469 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3470 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3475 ret
= LTTNG_ERR_UND
;
3480 if (cmd_ctx
->llm
== NULL
) {
3481 DBG("Missing llm structure. Allocating one.");
3482 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3486 /* Set return code */
3487 cmd_ctx
->llm
->ret_code
= ret
;
3489 if (cmd_ctx
->session
) {
3490 session_unlock(cmd_ctx
->session
);
3492 if (need_tracing_session
) {
3493 session_unlock_list();
3500 * Thread managing health check socket.
3502 static void *thread_manage_health(void *data
)
3504 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3505 uint32_t revents
, nb_fd
;
3506 struct lttng_poll_event events
;
3507 struct health_comm_msg msg
;
3508 struct health_comm_reply reply
;
3510 DBG("[thread] Manage health check started");
3512 rcu_register_thread();
3514 /* We might hit an error path before this is created. */
3515 lttng_poll_init(&events
);
3517 /* Create unix socket */
3518 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3520 ERR("Unable to create health check Unix socket");
3526 /* lttng health client socket path permissions */
3527 ret
= chown(health_unix_sock_path
, 0,
3528 utils_get_group_id(tracing_group_name
));
3530 ERR("Unable to set group on %s", health_unix_sock_path
);
3536 ret
= chmod(health_unix_sock_path
,
3537 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3539 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3547 * Set the CLOEXEC flag. Return code is useless because either way, the
3550 (void) utils_set_fd_cloexec(sock
);
3552 ret
= lttcomm_listen_unix_sock(sock
);
3558 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3559 * more will be added to this poll set.
3561 ret
= sessiond_set_thread_pollset(&events
, 2);
3566 /* Add the application registration socket */
3567 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3573 DBG("Health check ready");
3575 /* Inifinite blocking call, waiting for transmission */
3577 ret
= lttng_poll_wait(&events
, -1);
3580 * Restart interrupted system call.
3582 if (errno
== EINTR
) {
3590 for (i
= 0; i
< nb_fd
; i
++) {
3591 /* Fetch once the poll data */
3592 revents
= LTTNG_POLL_GETEV(&events
, i
);
3593 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3595 /* Thread quit pipe has been closed. Killing thread. */
3596 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3602 /* Event on the registration socket */
3603 if (pollfd
== sock
) {
3604 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3605 ERR("Health socket poll error");
3611 new_sock
= lttcomm_accept_unix_sock(sock
);
3617 * Set the CLOEXEC flag. Return code is useless because either way, the
3620 (void) utils_set_fd_cloexec(new_sock
);
3622 DBG("Receiving data from client for health...");
3623 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3625 DBG("Nothing recv() from client... continuing");
3626 ret
= close(new_sock
);
3634 rcu_thread_online();
3637 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3639 * health_check_state returns 0 if health is
3642 if (!health_check_state(health_sessiond
, i
)) {
3643 reply
.ret_code
|= 1ULL << i
;
3647 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3649 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3651 ERR("Failed to send health data back to client");
3654 /* End of transmission */
3655 ret
= close(new_sock
);
3665 ERR("Health error occurred in %s", __func__
);
3667 DBG("Health check thread dying");
3668 unlink(health_unix_sock_path
);
3676 lttng_poll_clean(&events
);
3678 rcu_unregister_thread();
3683 * This thread manage all clients request using the unix client socket for
3686 static void *thread_manage_clients(void *data
)
3688 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3690 uint32_t revents
, nb_fd
;
3691 struct command_ctx
*cmd_ctx
= NULL
;
3692 struct lttng_poll_event events
;
3694 DBG("[thread] Manage client started");
3696 rcu_register_thread();
3698 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3700 if (testpoint(thread_manage_clients
)) {
3701 goto error_testpoint
;
3704 health_code_update();
3706 ret
= lttcomm_listen_unix_sock(client_sock
);
3712 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3713 * more will be added to this poll set.
3715 ret
= sessiond_set_thread_pollset(&events
, 2);
3717 goto error_create_poll
;
3720 /* Add the application registration socket */
3721 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3727 * Notify parent pid that we are ready to accept command for client side.
3728 * This ppid is the one from the external process that spawned us.
3730 if (opt_sig_parent
) {
3731 kill(ppid
, SIGUSR1
);
3734 /* Notify the parent of the fork() process that we are ready. */
3736 kill(child_ppid
, SIGUSR1
);
3739 if (testpoint(thread_manage_clients_before_loop
)) {
3743 health_code_update();
3746 DBG("Accepting client command ...");
3748 /* Inifinite blocking call, waiting for transmission */
3750 health_poll_entry();
3751 ret
= lttng_poll_wait(&events
, -1);
3755 * Restart interrupted system call.
3757 if (errno
== EINTR
) {
3765 for (i
= 0; i
< nb_fd
; i
++) {
3766 /* Fetch once the poll data */
3767 revents
= LTTNG_POLL_GETEV(&events
, i
);
3768 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3770 health_code_update();
3772 /* Thread quit pipe has been closed. Killing thread. */
3773 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3779 /* Event on the registration socket */
3780 if (pollfd
== client_sock
) {
3781 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3782 ERR("Client socket poll error");
3788 DBG("Wait for client response");
3790 health_code_update();
3792 sock
= lttcomm_accept_unix_sock(client_sock
);
3798 * Set the CLOEXEC flag. Return code is useless because either way, the
3801 (void) utils_set_fd_cloexec(sock
);
3803 /* Set socket option for credentials retrieval */
3804 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3809 /* Allocate context command to process the client request */
3810 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3811 if (cmd_ctx
== NULL
) {
3812 PERROR("zmalloc cmd_ctx");
3816 /* Allocate data buffer for reception */
3817 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3818 if (cmd_ctx
->lsm
== NULL
) {
3819 PERROR("zmalloc cmd_ctx->lsm");
3823 cmd_ctx
->llm
= NULL
;
3824 cmd_ctx
->session
= NULL
;
3826 health_code_update();
3829 * Data is received from the lttng client. The struct
3830 * lttcomm_session_msg (lsm) contains the command and data request of
3833 DBG("Receiving data from client ...");
3834 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3835 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3837 DBG("Nothing recv() from client... continuing");
3843 clean_command_ctx(&cmd_ctx
);
3847 health_code_update();
3849 // TODO: Validate cmd_ctx including sanity check for
3850 // security purpose.
3852 rcu_thread_online();
3854 * This function dispatch the work to the kernel or userspace tracer
3855 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3856 * informations for the client. The command context struct contains
3857 * everything this function may needs.
3859 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3860 rcu_thread_offline();
3868 * TODO: Inform client somehow of the fatal error. At
3869 * this point, ret < 0 means that a zmalloc failed
3870 * (ENOMEM). Error detected but still accept
3871 * command, unless a socket error has been
3874 clean_command_ctx(&cmd_ctx
);
3878 health_code_update();
3880 DBG("Sending response (size: %d, retcode: %s)",
3881 cmd_ctx
->lttng_msg_size
,
3882 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3883 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3885 ERR("Failed to send data back to client");
3888 /* End of transmission */
3895 clean_command_ctx(&cmd_ctx
);
3897 health_code_update();
3909 lttng_poll_clean(&events
);
3910 clean_command_ctx(&cmd_ctx
);
3915 unlink(client_unix_sock_path
);
3916 if (client_sock
>= 0) {
3917 ret
= close(client_sock
);
3925 ERR("Health error occurred in %s", __func__
);
3928 health_unregister(health_sessiond
);
3930 DBG("Client thread dying");
3932 rcu_unregister_thread();
3938 * usage function on stderr
3940 static void usage(void)
3942 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3943 fprintf(stderr
, " -h, --help Display this usage.\n");
3944 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3945 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3946 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3947 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3948 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3949 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3950 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3951 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3952 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3953 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3954 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3955 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3956 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3957 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3958 fprintf(stderr
, " -V, --version Show version number.\n");
3959 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
3960 fprintf(stderr
, " -q, --quiet No output at all.\n");
3961 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3962 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3963 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3964 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3965 fprintf(stderr
, " --jul-tcp-port JUL application registration TCP port\n");
3969 * daemon argument parsing
3971 static int parse_args(int argc
, char **argv
)
3975 static struct option long_options
[] = {
3976 { "client-sock", 1, 0, 'c' },
3977 { "apps-sock", 1, 0, 'a' },
3978 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3979 { "kconsumerd-err-sock", 1, 0, 'E' },
3980 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3981 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3982 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3983 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3984 { "consumerd32-path", 1, 0, 'u' },
3985 { "consumerd32-libdir", 1, 0, 'U' },
3986 { "consumerd64-path", 1, 0, 't' },
3987 { "consumerd64-libdir", 1, 0, 'T' },
3988 { "daemonize", 0, 0, 'd' },
3989 { "sig-parent", 0, 0, 'S' },
3990 { "help", 0, 0, 'h' },
3991 { "group", 1, 0, 'g' },
3992 { "version", 0, 0, 'V' },
3993 { "quiet", 0, 0, 'q' },
3994 { "verbose", 0, 0, 'v' },
3995 { "verbose-consumer", 0, 0, 'Z' },
3996 { "no-kernel", 0, 0, 'N' },
3997 { "pidfile", 1, 0, 'p' },
3998 { "jul-tcp-port", 1, 0, 'J' },
4003 int option_index
= 0;
4004 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:",
4005 long_options
, &option_index
);
4012 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
4014 fprintf(stderr
, " with arg %s\n", optarg
);
4018 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4021 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4027 tracing_group_name
= optarg
;
4033 fprintf(stdout
, "%s\n", VERSION
);
4039 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4042 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4045 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4048 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4051 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4054 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4060 lttng_opt_quiet
= 1;
4063 /* Verbose level can increase using multiple -v */
4064 lttng_opt_verbose
+= 1;
4067 opt_verbose_consumer
+= 1;
4070 consumerd32_bin
= optarg
;
4073 consumerd32_libdir
= optarg
;
4076 consumerd64_bin
= optarg
;
4079 consumerd64_libdir
= optarg
;
4082 opt_pidfile
= optarg
;
4084 case 'J': /* JUL TCP port. */
4089 v
= strtoul(optarg
, NULL
, 0);
4090 if (errno
!= 0 || !isdigit(optarg
[0])) {
4091 ERR("Wrong value in --jul-tcp-port parameter: %s", optarg
);
4094 if (v
== 0 || v
>= 65535) {
4095 ERR("Port overflow in --jul-tcp-port parameter: %s", optarg
);
4098 jul_tcp_port
= (uint32_t) v
;
4099 DBG3("JUL TCP port set to non default: %u", jul_tcp_port
);
4103 /* Unknown option or other error.
4104 * Error is printed by getopt, just return */
4113 * Creates the two needed socket by the daemon.
4114 * apps_sock - The communication socket for all UST apps.
4115 * client_sock - The communication of the cli tool (lttng).
4117 static int init_daemon_socket(void)
4122 old_umask
= umask(0);
4124 /* Create client tool unix socket */
4125 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4126 if (client_sock
< 0) {
4127 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4132 /* Set the cloexec flag */
4133 ret
= utils_set_fd_cloexec(client_sock
);
4135 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4136 "Continuing but note that the consumer daemon will have a "
4137 "reference to this socket on exec()", client_sock
);
4140 /* File permission MUST be 660 */
4141 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4143 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4148 /* Create the application unix socket */
4149 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4150 if (apps_sock
< 0) {
4151 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4156 /* Set the cloexec flag */
4157 ret
= utils_set_fd_cloexec(apps_sock
);
4159 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4160 "Continuing but note that the consumer daemon will have a "
4161 "reference to this socket on exec()", apps_sock
);
4164 /* File permission MUST be 666 */
4165 ret
= chmod(apps_unix_sock_path
,
4166 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4168 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4173 DBG3("Session daemon client socket %d and application socket %d created",
4174 client_sock
, apps_sock
);
4182 * Check if the global socket is available, and if a daemon is answering at the
4183 * other side. If yes, error is returned.
4185 static int check_existing_daemon(void)
4187 /* Is there anybody out there ? */
4188 if (lttng_session_daemon_alive()) {
4196 * Set the tracing group gid onto the client socket.
4198 * Race window between mkdir and chown is OK because we are going from more
4199 * permissive (root.root) to less permissive (root.tracing).
4201 static int set_permissions(char *rundir
)
4206 gid
= utils_get_group_id(tracing_group_name
);
4208 /* Set lttng run dir */
4209 ret
= chown(rundir
, 0, gid
);
4211 ERR("Unable to set group on %s", rundir
);
4216 * Ensure all applications and tracing group can search the run
4217 * dir. Allow everyone to read the directory, since it does not
4218 * buy us anything to hide its content.
4220 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4222 ERR("Unable to set permissions on %s", rundir
);
4226 /* lttng client socket path */
4227 ret
= chown(client_unix_sock_path
, 0, gid
);
4229 ERR("Unable to set group on %s", client_unix_sock_path
);
4233 /* kconsumer error socket path */
4234 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4236 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4240 /* 64-bit ustconsumer error socket path */
4241 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4243 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4247 /* 32-bit ustconsumer compat32 error socket path */
4248 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4250 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4254 DBG("All permissions are set");
4260 * Create the lttng run directory needed for all global sockets and pipe.
4262 static int create_lttng_rundir(const char *rundir
)
4266 DBG3("Creating LTTng run directory: %s", rundir
);
4268 ret
= mkdir(rundir
, S_IRWXU
);
4270 if (errno
!= EEXIST
) {
4271 ERR("Unable to create %s", rundir
);
4283 * Setup sockets and directory needed by the kconsumerd communication with the
4286 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4290 char path
[PATH_MAX
];
4292 switch (consumer_data
->type
) {
4293 case LTTNG_CONSUMER_KERNEL
:
4294 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4296 case LTTNG_CONSUMER64_UST
:
4297 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4299 case LTTNG_CONSUMER32_UST
:
4300 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4303 ERR("Consumer type unknown");
4308 DBG2("Creating consumer directory: %s", path
);
4310 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4312 if (errno
!= EEXIST
) {
4314 ERR("Failed to create %s", path
);
4320 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4322 ERR("Unable to set group on %s", path
);
4328 /* Create the kconsumerd error unix socket */
4329 consumer_data
->err_sock
=
4330 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4331 if (consumer_data
->err_sock
< 0) {
4332 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4338 * Set the CLOEXEC flag. Return code is useless because either way, the
4341 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4343 PERROR("utils_set_fd_cloexec");
4344 /* continue anyway */
4347 /* File permission MUST be 660 */
4348 ret
= chmod(consumer_data
->err_unix_sock_path
,
4349 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4351 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4361 * Signal handler for the daemon
4363 * Simply stop all worker threads, leaving main() return gracefully after
4364 * joining all threads and calling cleanup().
4366 static void sighandler(int sig
)
4370 DBG("SIGPIPE caught");
4373 DBG("SIGINT caught");
4377 DBG("SIGTERM caught");
4381 CMM_STORE_SHARED(recv_child_signal
, 1);
4389 * Setup signal handler for :
4390 * SIGINT, SIGTERM, SIGPIPE
4392 static int set_signal_handler(void)
4395 struct sigaction sa
;
4398 if ((ret
= sigemptyset(&sigset
)) < 0) {
4399 PERROR("sigemptyset");
4403 sa
.sa_handler
= sighandler
;
4404 sa
.sa_mask
= sigset
;
4406 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4407 PERROR("sigaction");
4411 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4412 PERROR("sigaction");
4416 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4417 PERROR("sigaction");
4421 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4422 PERROR("sigaction");
4426 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4432 * Set open files limit to unlimited. This daemon can open a large number of
4433 * file descriptors in order to consumer multiple kernel traces.
4435 static void set_ulimit(void)
4440 /* The kernel does not allowed an infinite limit for open files */
4441 lim
.rlim_cur
= 65535;
4442 lim
.rlim_max
= 65535;
4444 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4446 PERROR("failed to set open files limit");
4451 * Write pidfile using the rundir and opt_pidfile.
4453 static void write_pidfile(void)
4456 char pidfile_path
[PATH_MAX
];
4461 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4463 /* Build pidfile path from rundir and opt_pidfile. */
4464 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4465 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4467 PERROR("snprintf pidfile path");
4473 * Create pid file in rundir. Return value is of no importance. The
4474 * execution will continue even though we are not able to write the file.
4476 (void) utils_create_pid_file(getpid(), pidfile_path
);
4483 * Write JUL TCP port using the rundir.
4485 static void write_julport(void)
4488 char path
[PATH_MAX
];
4492 ret
= snprintf(path
, sizeof(path
), "%s/"
4493 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
, rundir
);
4495 PERROR("snprintf julport path");
4500 * Create TCP JUL port file in rundir. Return value is of no importance.
4501 * The execution will continue even though we are not able to write the
4504 (void) utils_create_pid_file(jul_tcp_port
, path
);
4511 * Daemonize this process by forking and making the parent wait for the child
4512 * to signal it indicating readiness. Once received, the parent successfully
4515 * The child process undergoes the same action that daemon(3) does meaning
4516 * setsid, chdir, and dup /dev/null into 0, 1 and 2.
4518 * Return 0 on success else -1 on error.
4520 static int daemonize(void)
4525 /* Get parent pid of this process. */
4526 child_ppid
= getppid();
4532 } else if (pid
== 0) {
4539 * Get the newly created parent pid so we can signal that process when
4540 * we are ready to operate.
4542 child_ppid
= getppid();
4550 /* Try to change directory to /. If we can't well at least notify. */
4556 fd
= open(_PATH_DEVNULL
, O_RDWR
, 0);
4558 PERROR("open %s", _PATH_DEVNULL
);
4559 /* Let 0, 1 and 2 open since we can't bind them to /dev/null. */
4561 (void) dup2(fd
, STDIN_FILENO
);
4562 (void) dup2(fd
, STDOUT_FILENO
);
4563 (void) dup2(fd
, STDERR_FILENO
);
4576 * Waiting for child to notify this parent that it can exit. Note that
4577 * sleep() is interrupted before the 1 second delay as soon as the
4578 * signal is received, so it will not cause visible delay for the
4581 while (!CMM_LOAD_SHARED(recv_child_signal
)) {
4586 * From this point on, the parent can exit and the child is now an
4587 * operationnal session daemon ready to serve clients and applications.
4602 int main(int argc
, char **argv
)
4606 const char *home_path
, *env_app_timeout
;
4608 init_kernel_workarounds();
4610 rcu_register_thread();
4612 if ((ret
= set_signal_handler()) < 0) {
4616 setup_consumerd_path();
4618 page_size
= sysconf(_SC_PAGESIZE
);
4619 if (page_size
< 0) {
4620 PERROR("sysconf _SC_PAGESIZE");
4621 page_size
= LONG_MAX
;
4622 WARN("Fallback page size to %ld", page_size
);
4625 /* Parse arguments */
4627 if ((ret
= parse_args(argc
, argv
)) < 0) {
4641 * We are in the child. Make sure all other file descriptors are
4642 * closed, in case we are called with more opened file descriptors than
4643 * the standard ones.
4645 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4650 /* Create thread quit pipe */
4651 if ((ret
= init_thread_quit_pipe()) < 0) {
4655 /* Check if daemon is UID = 0 */
4656 is_root
= !getuid();
4659 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4661 /* Create global run dir with root access */
4662 ret
= create_lttng_rundir(rundir
);
4667 if (strlen(apps_unix_sock_path
) == 0) {
4668 snprintf(apps_unix_sock_path
, PATH_MAX
,
4669 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4672 if (strlen(client_unix_sock_path
) == 0) {
4673 snprintf(client_unix_sock_path
, PATH_MAX
,
4674 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4677 /* Set global SHM for ust */
4678 if (strlen(wait_shm_path
) == 0) {
4679 snprintf(wait_shm_path
, PATH_MAX
,
4680 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4683 if (strlen(health_unix_sock_path
) == 0) {
4684 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4685 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4688 /* Setup kernel consumerd path */
4689 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4690 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4691 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4692 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4694 DBG2("Kernel consumer err path: %s",
4695 kconsumer_data
.err_unix_sock_path
);
4696 DBG2("Kernel consumer cmd path: %s",
4697 kconsumer_data
.cmd_unix_sock_path
);
4699 home_path
= utils_get_home_dir();
4700 if (home_path
== NULL
) {
4701 /* TODO: Add --socket PATH option */
4702 ERR("Can't get HOME directory for sockets creation.");
4708 * Create rundir from home path. This will create something like
4711 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4717 ret
= create_lttng_rundir(rundir
);
4722 if (strlen(apps_unix_sock_path
) == 0) {
4723 snprintf(apps_unix_sock_path
, PATH_MAX
,
4724 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4727 /* Set the cli tool unix socket path */
4728 if (strlen(client_unix_sock_path
) == 0) {
4729 snprintf(client_unix_sock_path
, PATH_MAX
,
4730 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4733 /* Set global SHM for ust */
4734 if (strlen(wait_shm_path
) == 0) {
4735 snprintf(wait_shm_path
, PATH_MAX
,
4736 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4739 /* Set health check Unix path */
4740 if (strlen(health_unix_sock_path
) == 0) {
4741 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4742 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4746 /* Set consumer initial state */
4747 kernel_consumerd_state
= CONSUMER_STOPPED
;
4748 ust_consumerd_state
= CONSUMER_STOPPED
;
4750 DBG("Client socket path %s", client_unix_sock_path
);
4751 DBG("Application socket path %s", apps_unix_sock_path
);
4752 DBG("Application wait path %s", wait_shm_path
);
4753 DBG("LTTng run directory path: %s", rundir
);
4755 /* 32 bits consumerd path setup */
4756 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4757 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4758 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4759 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4761 DBG2("UST consumer 32 bits err path: %s",
4762 ustconsumer32_data
.err_unix_sock_path
);
4763 DBG2("UST consumer 32 bits cmd path: %s",
4764 ustconsumer32_data
.cmd_unix_sock_path
);
4766 /* 64 bits consumerd path setup */
4767 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4768 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4769 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4770 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4772 DBG2("UST consumer 64 bits err path: %s",
4773 ustconsumer64_data
.err_unix_sock_path
);
4774 DBG2("UST consumer 64 bits cmd path: %s",
4775 ustconsumer64_data
.cmd_unix_sock_path
);
4778 * See if daemon already exist.
4780 if ((ret
= check_existing_daemon()) < 0) {
4781 ERR("Already running daemon.\n");
4783 * We do not goto exit because we must not cleanup()
4784 * because a daemon is already running.
4790 * Init UST app hash table. Alloc hash table before this point since
4791 * cleanup() can get called after that point.
4795 /* Initialize JUL domain subsystem. */
4796 if ((ret
= jul_init()) < 0) {
4797 /* ENOMEM at this point. */
4801 /* After this point, we can safely call cleanup() with "goto exit" */
4804 * These actions must be executed as root. We do that *after* setting up
4805 * the sockets path because we MUST make the check for another daemon using
4806 * those paths *before* trying to set the kernel consumer sockets and init
4810 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4815 /* Setup kernel tracer */
4816 if (!opt_no_kernel
) {
4817 init_kernel_tracer();
4820 /* Set ulimit for open files */
4823 /* init lttng_fd tracking must be done after set_ulimit. */
4826 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4831 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4836 /* Setup the needed unix socket */
4837 if ((ret
= init_daemon_socket()) < 0) {
4841 /* Set credentials to socket */
4842 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4846 /* Get parent pid if -S, --sig-parent is specified. */
4847 if (opt_sig_parent
) {
4851 /* Setup the kernel pipe for waking up the kernel thread */
4852 if (is_root
&& !opt_no_kernel
) {
4853 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4858 /* Setup the thread ht_cleanup communication pipe. */
4859 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4863 /* Setup the thread apps communication pipe. */
4864 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4868 /* Setup the thread apps notify communication pipe. */
4869 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4873 /* Initialize global buffer per UID and PID registry. */
4874 buffer_reg_init_uid_registry();
4875 buffer_reg_init_pid_registry();
4877 /* Init UST command queue. */
4878 cds_wfq_init(&ust_cmd_queue
.queue
);
4881 * Get session list pointer. This pointer MUST NOT be free(). This list is
4882 * statically declared in session.c
4884 session_list_ptr
= session_get_list();
4886 /* Set up max poll set size */
4887 lttng_poll_set_max_size();
4891 /* Check for the application socket timeout env variable. */
4892 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4893 if (env_app_timeout
) {
4894 app_socket_timeout
= atoi(env_app_timeout
);
4896 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4902 /* Initialize communication library */
4904 /* This is to get the TCP timeout value. */
4905 lttcomm_inet_init();
4908 * Initialize the health check subsystem. This call should set the
4909 * appropriate time values.
4911 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
4912 if (!health_sessiond
) {
4913 PERROR("health_app_create error");
4914 goto exit_health_sessiond_cleanup
;
4917 /* Create thread to clean up RCU hash tables */
4918 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4919 thread_ht_cleanup
, (void *) NULL
);
4921 PERROR("pthread_create ht_cleanup");
4922 goto exit_ht_cleanup
;
4925 /* Create health-check thread */
4926 ret
= pthread_create(&health_thread
, NULL
,
4927 thread_manage_health
, (void *) NULL
);
4929 PERROR("pthread_create health");
4933 /* Create thread to manage the client socket */
4934 ret
= pthread_create(&client_thread
, NULL
,
4935 thread_manage_clients
, (void *) NULL
);
4937 PERROR("pthread_create clients");
4941 /* Create thread to dispatch registration */
4942 ret
= pthread_create(&dispatch_thread
, NULL
,
4943 thread_dispatch_ust_registration
, (void *) NULL
);
4945 PERROR("pthread_create dispatch");
4949 /* Create thread to manage application registration. */
4950 ret
= pthread_create(®_apps_thread
, NULL
,
4951 thread_registration_apps
, (void *) NULL
);
4953 PERROR("pthread_create registration");
4957 /* Create thread to manage application socket */
4958 ret
= pthread_create(&apps_thread
, NULL
,
4959 thread_manage_apps
, (void *) NULL
);
4961 PERROR("pthread_create apps");
4965 /* Create thread to manage application notify socket */
4966 ret
= pthread_create(&apps_notify_thread
, NULL
,
4967 ust_thread_manage_notify
, (void *) NULL
);
4969 PERROR("pthread_create apps");
4970 goto exit_apps_notify
;
4973 /* Create JUL registration thread. */
4974 ret
= pthread_create(&jul_reg_thread
, NULL
,
4975 jul_thread_manage_registration
, (void *) NULL
);
4977 PERROR("pthread_create apps");
4981 /* Don't start this thread if kernel tracing is not requested nor root */
4982 if (is_root
&& !opt_no_kernel
) {
4983 /* Create kernel thread to manage kernel event */
4984 ret
= pthread_create(&kernel_thread
, NULL
,
4985 thread_manage_kernel
, (void *) NULL
);
4987 PERROR("pthread_create kernel");
4991 ret
= pthread_join(kernel_thread
, &status
);
4993 PERROR("pthread_join");
4994 goto error
; /* join error, exit without cleanup */
4999 ret
= pthread_join(jul_reg_thread
, &status
);
5001 PERROR("pthread_join JUL");
5002 goto error
; /* join error, exit without cleanup */
5006 ret
= pthread_join(apps_notify_thread
, &status
);
5008 PERROR("pthread_join apps notify");
5009 goto error
; /* join error, exit without cleanup */
5013 ret
= pthread_join(apps_thread
, &status
);
5015 PERROR("pthread_join apps");
5016 goto error
; /* join error, exit without cleanup */
5021 ret
= pthread_join(reg_apps_thread
, &status
);
5023 PERROR("pthread_join");
5024 goto error
; /* join error, exit without cleanup */
5028 ret
= pthread_join(dispatch_thread
, &status
);
5030 PERROR("pthread_join");
5031 goto error
; /* join error, exit without cleanup */
5035 ret
= pthread_join(client_thread
, &status
);
5037 PERROR("pthread_join");
5038 goto error
; /* join error, exit without cleanup */
5041 ret
= join_consumer_thread(&kconsumer_data
);
5043 PERROR("join_consumer");
5044 goto error
; /* join error, exit without cleanup */
5047 ret
= join_consumer_thread(&ustconsumer32_data
);
5049 PERROR("join_consumer ust32");
5050 goto error
; /* join error, exit without cleanup */
5053 ret
= join_consumer_thread(&ustconsumer64_data
);
5055 PERROR("join_consumer ust64");
5056 goto error
; /* join error, exit without cleanup */
5060 ret
= pthread_join(health_thread
, &status
);
5062 PERROR("pthread_join health thread");
5063 goto error
; /* join error, exit without cleanup */
5067 ret
= pthread_join(ht_cleanup_thread
, &status
);
5069 PERROR("pthread_join ht cleanup thread");
5070 goto error
; /* join error, exit without cleanup */
5073 health_app_destroy(health_sessiond
);
5074 exit_health_sessiond_cleanup
:
5077 * cleanup() is called when no other thread is running.
5079 rcu_thread_online();
5081 rcu_thread_offline();
5082 rcu_unregister_thread();