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.
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
34 #include <sys/types.h>
36 #include <urcu/uatomic.h>
40 #include <common/common.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
48 #include "lttng-sessiond.h"
49 #include "buffer-registry.h"
56 #include "kernel-consumer.h"
60 #include "ust-consumer.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
67 #define CONSUMERD_FILE "lttng-consumerd"
70 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
73 const char *opt_tracing_group
;
74 static const char *opt_pidfile
;
75 static int opt_sig_parent
;
76 static int opt_verbose_consumer
;
77 static int opt_daemon
;
78 static int opt_no_kernel
;
79 static int is_root
; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid
; /* Parent PID for --sig-parent option */
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
87 static struct consumer_data kconsumer_data
= {
88 .type
= LTTNG_CONSUMER_KERNEL
,
89 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
90 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
93 .metadata_sock
.fd
= -1,
94 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
95 .lock
= PTHREAD_MUTEX_INITIALIZER
,
96 .cond
= PTHREAD_COND_INITIALIZER
,
97 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
99 static struct consumer_data ustconsumer64_data
= {
100 .type
= LTTNG_CONSUMER64_UST
,
101 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
102 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
105 .metadata_sock
.fd
= -1,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer32_data
= {
112 .type
= LTTNG_CONSUMER32_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
117 .metadata_sock
.fd
= -1,
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
;
167 * UST registration command queue. This queue is tied with a futex and uses a N
168 * wakers / 1 waiter implemented and detailed in futex.c/.h
170 * The thread_manage_apps and thread_dispatch_ust_registration interact with
171 * this queue and the wait/wake scheme.
173 static struct ust_cmd_queue ust_cmd_queue
;
176 * Pointer initialized before thread creation.
178 * This points to the tracing session list containing the session count and a
179 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
180 * MUST NOT be taken if you call a public function in session.c.
182 * The lock is nested inside the structure: session_list_ptr->lock. Please use
183 * session_lock_list and session_unlock_list for lock acquisition.
185 static struct ltt_session_list
*session_list_ptr
;
187 int ust_consumerd64_fd
= -1;
188 int ust_consumerd32_fd
= -1;
190 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
191 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
192 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
193 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
195 static const char *module_proc_lttng
= "/proc/lttng";
198 * Consumer daemon state which is changed when spawning it, killing it or in
199 * case of a fatal error.
201 enum consumerd_state
{
202 CONSUMER_STARTED
= 1,
203 CONSUMER_STOPPED
= 2,
208 * This consumer daemon state is used to validate if a client command will be
209 * able to reach the consumer. If not, the client is informed. For instance,
210 * doing a "lttng start" when the consumer state is set to ERROR will return an
211 * error to the client.
213 * The following example shows a possible race condition of this scheme:
215 * consumer thread error happens
217 * client cmd checks state -> still OK
218 * consumer thread exit, sets error
219 * client cmd try to talk to consumer
222 * However, since the consumer is a different daemon, we have no way of making
223 * sure the command will reach it safely even with this state flag. This is why
224 * we consider that up to the state validation during command processing, the
225 * command is safe. After that, we can not guarantee the correctness of the
226 * client request vis-a-vis the consumer.
228 static enum consumerd_state ust_consumerd_state
;
229 static enum consumerd_state kernel_consumerd_state
;
232 * Socket timeout for receiving and sending in seconds.
234 static int app_socket_timeout
;
236 /* Set in main() with the current page size. */
240 void setup_consumerd_path(void)
242 const char *bin
, *libdir
;
245 * Allow INSTALL_BIN_PATH to be used as a target path for the
246 * native architecture size consumer if CONFIG_CONSUMER*_PATH
247 * has not been defined.
249 #if (CAA_BITS_PER_LONG == 32)
250 if (!consumerd32_bin
[0]) {
251 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
253 if (!consumerd32_libdir
[0]) {
254 consumerd32_libdir
= INSTALL_LIB_PATH
;
256 #elif (CAA_BITS_PER_LONG == 64)
257 if (!consumerd64_bin
[0]) {
258 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
260 if (!consumerd64_libdir
[0]) {
261 consumerd64_libdir
= INSTALL_LIB_PATH
;
264 #error "Unknown bitness"
268 * runtime env. var. overrides the build default.
270 bin
= getenv("LTTNG_CONSUMERD32_BIN");
272 consumerd32_bin
= bin
;
274 bin
= getenv("LTTNG_CONSUMERD64_BIN");
276 consumerd64_bin
= bin
;
278 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
280 consumerd32_libdir
= libdir
;
282 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
284 consumerd64_libdir
= libdir
;
289 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
291 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
297 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
303 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
315 * Check if the thread quit pipe was triggered.
317 * Return 1 if it was triggered else 0;
319 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
321 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
329 * Return group ID of the tracing group or -1 if not found.
331 static gid_t
allowed_group(void)
335 if (opt_tracing_group
) {
336 grp
= getgrnam(opt_tracing_group
);
338 grp
= getgrnam(default_tracing_group
);
348 * Init thread quit pipe.
350 * Return -1 on error or 0 if all pipes are created.
352 static int init_thread_quit_pipe(void)
356 ret
= pipe(thread_quit_pipe
);
358 PERROR("thread quit pipe");
362 for (i
= 0; i
< 2; i
++) {
363 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
375 * Stop all threads by closing the thread quit pipe.
377 static void stop_threads(void)
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
385 ERR("write error on thread quit pipe");
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
390 futex_nto1_wake(&ust_cmd_queue
.futex
);
394 * Close every consumer sockets.
396 static void close_consumer_sockets(void)
400 if (kconsumer_data
.err_sock
>= 0) {
401 ret
= close(kconsumer_data
.err_sock
);
403 PERROR("kernel consumer err_sock close");
406 if (ustconsumer32_data
.err_sock
>= 0) {
407 ret
= close(ustconsumer32_data
.err_sock
);
409 PERROR("UST consumerd32 err_sock close");
412 if (ustconsumer64_data
.err_sock
>= 0) {
413 ret
= close(ustconsumer64_data
.err_sock
);
415 PERROR("UST consumerd64 err_sock close");
418 if (kconsumer_data
.cmd_sock
>= 0) {
419 ret
= close(kconsumer_data
.cmd_sock
);
421 PERROR("kernel consumer cmd_sock close");
424 if (ustconsumer32_data
.cmd_sock
>= 0) {
425 ret
= close(ustconsumer32_data
.cmd_sock
);
427 PERROR("UST consumerd32 cmd_sock close");
430 if (ustconsumer64_data
.cmd_sock
>= 0) {
431 ret
= close(ustconsumer64_data
.cmd_sock
);
433 PERROR("UST consumerd64 cmd_sock close");
441 static void cleanup(void)
445 struct ltt_session
*sess
, *stmp
;
450 * Close the thread quit pipe. It has already done its job,
451 * since we are now called.
453 utils_close_pipe(thread_quit_pipe
);
456 * If opt_pidfile is undefined, the default file will be wiped when
457 * removing the rundir.
460 ret
= remove(opt_pidfile
);
462 PERROR("remove pidfile %s", opt_pidfile
);
466 DBG("Removing %s directory", rundir
);
467 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
469 ERR("asprintf failed. Something is really wrong!");
472 /* Remove lttng run directory */
475 ERR("Unable to clean %s", rundir
);
480 DBG("Cleaning up all sessions");
482 /* Destroy session list mutex */
483 if (session_list_ptr
!= NULL
) {
484 pthread_mutex_destroy(&session_list_ptr
->lock
);
486 /* Cleanup ALL session */
487 cds_list_for_each_entry_safe(sess
, stmp
,
488 &session_list_ptr
->head
, list
) {
489 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
493 DBG("Closing all UST sockets");
494 ust_app_clean_list();
495 buffer_reg_destroy_registries();
497 if (is_root
&& !opt_no_kernel
) {
498 DBG2("Closing kernel fd");
499 if (kernel_tracer_fd
>= 0) {
500 ret
= close(kernel_tracer_fd
);
505 DBG("Unloading kernel modules");
506 modprobe_remove_lttng_all();
509 close_consumer_sockets();
512 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
513 "Matthew, BEET driven development works!%c[%dm",
514 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
519 * Send data on a unix socket using the liblttsessiondcomm API.
521 * Return lttcomm error code.
523 static int send_unix_sock(int sock
, void *buf
, size_t len
)
525 /* Check valid length */
530 return lttcomm_send_unix_sock(sock
, buf
, len
);
534 * Free memory of a command context structure.
536 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
538 DBG("Clean command context structure");
540 if ((*cmd_ctx
)->llm
) {
541 free((*cmd_ctx
)->llm
);
543 if ((*cmd_ctx
)->lsm
) {
544 free((*cmd_ctx
)->lsm
);
552 * Notify UST applications using the shm mmap futex.
554 static int notify_ust_apps(int active
)
558 DBG("Notifying applications of session daemon state: %d", active
);
560 /* See shm.c for this call implying mmap, shm and futex calls */
561 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
562 if (wait_shm_mmap
== NULL
) {
566 /* Wake waiting process */
567 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
569 /* Apps notified successfully */
577 * Setup the outgoing data buffer for the response (llm) by allocating the
578 * right amount of memory and copying the original information from the lsm
581 * Return total size of the buffer pointed by buf.
583 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
589 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
590 if (cmd_ctx
->llm
== NULL
) {
596 /* Copy common data */
597 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
598 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
600 cmd_ctx
->llm
->data_size
= size
;
601 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
610 * Update the kernel poll set of all channel fd available over all tracing
611 * session. Add the wakeup pipe at the end of the set.
613 static int update_kernel_poll(struct lttng_poll_event
*events
)
616 struct ltt_session
*session
;
617 struct ltt_kernel_channel
*channel
;
619 DBG("Updating kernel poll set");
622 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
623 session_lock(session
);
624 if (session
->kernel_session
== NULL
) {
625 session_unlock(session
);
629 cds_list_for_each_entry(channel
,
630 &session
->kernel_session
->channel_list
.head
, list
) {
631 /* Add channel fd to the kernel poll set */
632 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
634 session_unlock(session
);
637 DBG("Channel fd %d added to kernel set", channel
->fd
);
639 session_unlock(session
);
641 session_unlock_list();
646 session_unlock_list();
651 * Find the channel fd from 'fd' over all tracing session. When found, check
652 * for new channel stream and send those stream fds to the kernel consumer.
654 * Useful for CPU hotplug feature.
656 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
659 struct ltt_session
*session
;
660 struct ltt_kernel_session
*ksess
;
661 struct ltt_kernel_channel
*channel
;
663 DBG("Updating kernel streams for channel fd %d", fd
);
666 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
667 session_lock(session
);
668 if (session
->kernel_session
== NULL
) {
669 session_unlock(session
);
672 ksess
= session
->kernel_session
;
674 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
675 if (channel
->fd
== fd
) {
676 DBG("Channel found, updating kernel streams");
677 ret
= kernel_open_channel_stream(channel
);
681 /* Update the stream global counter */
682 ksess
->stream_count_global
+= ret
;
685 * Have we already sent fds to the consumer? If yes, it means
686 * that tracing is started so it is safe to send our updated
689 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
690 struct lttng_ht_iter iter
;
691 struct consumer_socket
*socket
;
694 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
695 &iter
.iter
, socket
, node
.node
) {
696 /* Code flow error */
697 assert(socket
->fd
>= 0);
699 pthread_mutex_lock(socket
->lock
);
700 ret
= kernel_consumer_send_channel_stream(socket
,
702 session
->output_traces
? 1 : 0);
703 pthread_mutex_unlock(socket
->lock
);
714 session_unlock(session
);
716 session_unlock_list();
720 session_unlock(session
);
721 session_unlock_list();
726 * For each tracing session, update newly registered apps. The session list
727 * lock MUST be acquired before calling this.
729 static void update_ust_app(int app_sock
)
731 struct ltt_session
*sess
, *stmp
;
733 /* For all tracing session(s) */
734 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
736 if (sess
->ust_session
) {
737 ust_app_global_update(sess
->ust_session
, app_sock
);
739 session_unlock(sess
);
744 * This thread manage event coming from the kernel.
746 * Features supported in this thread:
749 static void *thread_manage_kernel(void *data
)
751 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
752 uint32_t revents
, nb_fd
;
754 struct lttng_poll_event events
;
756 DBG("[thread] Thread manage kernel started");
758 health_register(HEALTH_TYPE_KERNEL
);
761 * This first step of the while is to clean this structure which could free
762 * non NULL pointers so initialize it before the loop.
764 lttng_poll_init(&events
);
766 if (testpoint(thread_manage_kernel
)) {
767 goto error_testpoint
;
770 health_code_update();
772 if (testpoint(thread_manage_kernel_before_loop
)) {
773 goto error_testpoint
;
777 health_code_update();
779 if (update_poll_flag
== 1) {
780 /* Clean events object. We are about to populate it again. */
781 lttng_poll_clean(&events
);
783 ret
= sessiond_set_thread_pollset(&events
, 2);
785 goto error_poll_create
;
788 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
793 /* This will add the available kernel channel if any. */
794 ret
= update_kernel_poll(&events
);
798 update_poll_flag
= 0;
801 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
803 /* Poll infinite value of time */
806 ret
= lttng_poll_wait(&events
, -1);
810 * Restart interrupted system call.
812 if (errno
== EINTR
) {
816 } else if (ret
== 0) {
817 /* Should not happen since timeout is infinite */
818 ERR("Return value of poll is 0 with an infinite timeout.\n"
819 "This should not have happened! Continuing...");
825 for (i
= 0; i
< nb_fd
; i
++) {
826 /* Fetch once the poll data */
827 revents
= LTTNG_POLL_GETEV(&events
, i
);
828 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
830 health_code_update();
832 /* Thread quit pipe has been closed. Killing thread. */
833 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
839 /* Check for data on kernel pipe */
840 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
842 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
843 } while (ret
< 0 && errno
== EINTR
);
845 * Ret value is useless here, if this pipe gets any actions an
846 * update is required anyway.
848 update_poll_flag
= 1;
852 * New CPU detected by the kernel. Adding kernel stream to
853 * kernel session and updating the kernel consumer
855 if (revents
& LPOLLIN
) {
856 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
862 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
863 * and unregister kernel stream at this point.
872 lttng_poll_clean(&events
);
875 utils_close_pipe(kernel_poll_pipe
);
876 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
879 ERR("Health error occurred in %s", __func__
);
880 WARN("Kernel thread died unexpectedly. "
881 "Kernel tracing can continue but CPU hotplug is disabled.");
884 DBG("Kernel thread dying");
889 * Signal pthread condition of the consumer data that the thread.
891 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
893 pthread_mutex_lock(&data
->cond_mutex
);
896 * The state is set before signaling. It can be any value, it's the waiter
897 * job to correctly interpret this condition variable associated to the
898 * consumer pthread_cond.
900 * A value of 0 means that the corresponding thread of the consumer data
901 * was not started. 1 indicates that the thread has started and is ready
902 * for action. A negative value means that there was an error during the
905 data
->consumer_thread_is_ready
= state
;
906 (void) pthread_cond_signal(&data
->cond
);
908 pthread_mutex_unlock(&data
->cond_mutex
);
912 * This thread manage the consumer error sent back to the session daemon.
914 static void *thread_manage_consumer(void *data
)
916 int sock
= -1, i
, ret
, pollfd
, err
= -1;
917 uint32_t revents
, nb_fd
;
918 enum lttcomm_return_code code
;
919 struct lttng_poll_event events
;
920 struct consumer_data
*consumer_data
= data
;
922 DBG("[thread] Manage consumer started");
924 health_register(HEALTH_TYPE_CONSUMER
);
926 health_code_update();
929 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
930 * metadata_sock. Nothing more will be added to this poll set.
932 ret
= sessiond_set_thread_pollset(&events
, 3);
938 * The error socket here is already in a listening state which was done
939 * just before spawning this thread to avoid a race between the consumer
940 * daemon exec trying to connect and the listen() call.
942 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
947 health_code_update();
949 /* Infinite blocking call, waiting for transmission */
953 if (testpoint(thread_manage_consumer
)) {
957 ret
= lttng_poll_wait(&events
, -1);
961 * Restart interrupted system call.
963 if (errno
== EINTR
) {
971 for (i
= 0; i
< nb_fd
; i
++) {
972 /* Fetch once the poll data */
973 revents
= LTTNG_POLL_GETEV(&events
, i
);
974 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
976 health_code_update();
978 /* Thread quit pipe has been closed. Killing thread. */
979 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
985 /* Event on the registration socket */
986 if (pollfd
== consumer_data
->err_sock
) {
987 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
988 ERR("consumer err socket poll error");
994 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1000 * Set the CLOEXEC flag. Return code is useless because either way, the
1003 (void) utils_set_fd_cloexec(sock
);
1005 health_code_update();
1007 DBG2("Receiving code from consumer err_sock");
1009 /* Getting status code from kconsumerd */
1010 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1011 sizeof(enum lttcomm_return_code
));
1016 health_code_update();
1018 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1019 /* Connect both socket, command and metadata. */
1020 consumer_data
->cmd_sock
=
1021 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1022 consumer_data
->metadata_sock
.fd
=
1023 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1024 if (consumer_data
->cmd_sock
< 0 ||
1025 consumer_data
->metadata_sock
.fd
< 0) {
1026 PERROR("consumer connect cmd socket");
1027 /* On error, signal condition and quit. */
1028 signal_consumer_condition(consumer_data
, -1);
1031 /* Create metadata socket lock. */
1032 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1033 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1034 PERROR("zmalloc pthread mutex");
1038 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1040 signal_consumer_condition(consumer_data
, 1);
1041 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1042 DBG("Consumer metadata socket ready (fd: %d)",
1043 consumer_data
->metadata_sock
.fd
);
1045 ERR("consumer error when waiting for SOCK_READY : %s",
1046 lttcomm_get_readable_code(-code
));
1050 /* Remove the consumerd error sock since we've established a connexion */
1051 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1056 /* Add new accepted error socket. */
1057 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1062 /* Add metadata socket that is successfully connected. */
1063 ret
= lttng_poll_add(&events
, consumer_data
->metadata_sock
.fd
,
1064 LPOLLIN
| LPOLLRDHUP
);
1069 health_code_update();
1071 /* Infinite blocking call, waiting for transmission */
1074 health_poll_entry();
1075 ret
= lttng_poll_wait(&events
, -1);
1079 * Restart interrupted system call.
1081 if (errno
== EINTR
) {
1089 for (i
= 0; i
< nb_fd
; i
++) {
1090 /* Fetch once the poll data */
1091 revents
= LTTNG_POLL_GETEV(&events
, i
);
1092 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1094 health_code_update();
1096 /* Thread quit pipe has been closed. Killing thread. */
1097 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1103 if (pollfd
== sock
) {
1104 /* Event on the consumerd socket */
1105 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1106 ERR("consumer err socket second poll error");
1109 health_code_update();
1110 /* Wait for any kconsumerd error */
1111 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1112 sizeof(enum lttcomm_return_code
));
1114 ERR("consumer closed the command socket");
1118 ERR("consumer return code : %s",
1119 lttcomm_get_readable_code(-code
));
1122 } else if (pollfd
== consumer_data
->metadata_sock
.fd
) {
1123 /* UST metadata requests */
1124 ret
= ust_consumer_metadata_request(
1125 &consumer_data
->metadata_sock
);
1127 ERR("Handling metadata request");
1132 ERR("Unknown pollfd");
1136 health_code_update();
1141 /* Immediately set the consumerd state to stopped */
1142 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1143 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1144 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1145 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1146 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1148 /* Code flow error... */
1152 if (consumer_data
->err_sock
>= 0) {
1153 ret
= close(consumer_data
->err_sock
);
1157 consumer_data
->err_sock
= -1;
1159 if (consumer_data
->cmd_sock
>= 0) {
1160 ret
= close(consumer_data
->cmd_sock
);
1164 consumer_data
->cmd_sock
= -1;
1166 if (consumer_data
->metadata_sock
.fd
>= 0) {
1167 ret
= close(consumer_data
->metadata_sock
.fd
);
1172 /* Cleanup metadata socket mutex. */
1173 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1174 free(consumer_data
->metadata_sock
.lock
);
1183 unlink(consumer_data
->err_unix_sock_path
);
1184 unlink(consumer_data
->cmd_unix_sock_path
);
1185 consumer_data
->pid
= 0;
1187 lttng_poll_clean(&events
);
1191 ERR("Health error occurred in %s", __func__
);
1193 health_unregister();
1194 DBG("consumer thread cleanup completed");
1200 * This thread manage application communication.
1202 static void *thread_manage_apps(void *data
)
1204 int i
, ret
, pollfd
, err
= -1;
1205 uint32_t revents
, nb_fd
;
1206 struct lttng_poll_event events
;
1208 DBG("[thread] Manage application started");
1210 rcu_register_thread();
1211 rcu_thread_online();
1213 health_register(HEALTH_TYPE_APP_MANAGE
);
1215 if (testpoint(thread_manage_apps
)) {
1216 goto error_testpoint
;
1219 health_code_update();
1221 ret
= sessiond_set_thread_pollset(&events
, 2);
1223 goto error_poll_create
;
1226 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1231 if (testpoint(thread_manage_apps_before_loop
)) {
1235 health_code_update();
1238 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1240 /* Inifinite blocking call, waiting for transmission */
1242 health_poll_entry();
1243 ret
= lttng_poll_wait(&events
, -1);
1247 * Restart interrupted system call.
1249 if (errno
== EINTR
) {
1257 for (i
= 0; i
< nb_fd
; i
++) {
1258 /* Fetch once the poll data */
1259 revents
= LTTNG_POLL_GETEV(&events
, i
);
1260 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1262 health_code_update();
1264 /* Thread quit pipe has been closed. Killing thread. */
1265 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1271 /* Inspect the apps cmd pipe */
1272 if (pollfd
== apps_cmd_pipe
[0]) {
1273 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1274 ERR("Apps command pipe error");
1276 } else if (revents
& LPOLLIN
) {
1281 ret
= read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1282 } while (ret
< 0 && errno
== EINTR
);
1283 if (ret
< 0 || ret
< sizeof(sock
)) {
1284 PERROR("read apps cmd pipe");
1288 health_code_update();
1291 * We only monitor the error events of the socket. This
1292 * thread does not handle any incoming data from UST
1295 ret
= lttng_poll_add(&events
, sock
,
1296 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1302 * Set socket timeout for both receiving and ending.
1303 * app_socket_timeout is in seconds, whereas
1304 * lttcomm_setsockopt_rcv_timeout and
1305 * lttcomm_setsockopt_snd_timeout expect msec as
1308 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1309 app_socket_timeout
* 1000);
1310 (void) lttcomm_setsockopt_snd_timeout(sock
,
1311 app_socket_timeout
* 1000);
1313 DBG("Apps with sock %d added to poll set", sock
);
1315 health_code_update();
1321 * At this point, we know that a registered application made
1322 * the event at poll_wait.
1324 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1325 /* Removing from the poll set */
1326 ret
= lttng_poll_del(&events
, pollfd
);
1331 /* Socket closed on remote end. */
1332 ust_app_unregister(pollfd
);
1337 health_code_update();
1343 lttng_poll_clean(&events
);
1346 utils_close_pipe(apps_cmd_pipe
);
1347 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1350 * We don't clean the UST app hash table here since already registered
1351 * applications can still be controlled so let them be until the session
1352 * daemon dies or the applications stop.
1357 ERR("Health error occurred in %s", __func__
);
1359 health_unregister();
1360 DBG("Application communication apps thread cleanup complete");
1361 rcu_thread_offline();
1362 rcu_unregister_thread();
1367 * Send a socket to a thread This is called from the dispatch UST registration
1368 * thread once all sockets are set for the application.
1370 * On success, return 0 else a negative value being the errno message of the
1373 static int send_socket_to_thread(int fd
, int sock
)
1377 /* Sockets MUST be set or else this should not have been called. */
1382 ret
= write(fd
, &sock
, sizeof(sock
));
1383 } while (ret
< 0 && errno
== EINTR
);
1384 if (ret
< 0 || ret
!= sizeof(sock
)) {
1385 PERROR("write apps pipe %d", fd
);
1392 /* All good. Don't send back the write positive ret value. */
1399 * Sanitize the wait queue of the dispatch registration thread meaning removing
1400 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1401 * notify socket is never received.
1403 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1405 int ret
, nb_fd
= 0, i
;
1406 unsigned int fd_added
= 0;
1407 struct lttng_poll_event events
;
1408 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1412 lttng_poll_init(&events
);
1414 /* Just skip everything for an empty queue. */
1415 if (!wait_queue
->count
) {
1419 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1424 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1425 &wait_queue
->head
, head
) {
1426 assert(wait_node
->app
);
1427 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1428 LPOLLHUP
| LPOLLERR
);
1441 * Poll but don't block so we can quickly identify the faulty events and
1442 * clean them afterwards from the wait queue.
1444 ret
= lttng_poll_wait(&events
, 0);
1450 for (i
= 0; i
< nb_fd
; i
++) {
1451 /* Get faulty FD. */
1452 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1453 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1455 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1456 &wait_queue
->head
, head
) {
1457 if (pollfd
== wait_node
->app
->sock
&&
1458 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1459 cds_list_del(&wait_node
->head
);
1460 wait_queue
->count
--;
1461 ust_app_destroy(wait_node
->app
);
1469 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1473 lttng_poll_clean(&events
);
1477 lttng_poll_clean(&events
);
1479 ERR("Unable to sanitize wait queue");
1484 * Dispatch request from the registration threads to the application
1485 * communication thread.
1487 static void *thread_dispatch_ust_registration(void *data
)
1490 struct cds_wfq_node
*node
;
1491 struct ust_command
*ust_cmd
= NULL
;
1492 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1493 struct ust_reg_wait_queue wait_queue
= {
1497 health_register(HEALTH_TYPE_APP_REG_DISPATCH
);
1499 health_code_update();
1501 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1503 DBG("[thread] Dispatch UST command started");
1505 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1506 health_code_update();
1508 /* Atomically prepare the queue futex */
1509 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1512 struct ust_app
*app
= NULL
;
1516 * Make sure we don't have node(s) that have hung up before receiving
1517 * the notify socket. This is to clean the list in order to avoid
1518 * memory leaks from notify socket that are never seen.
1520 sanitize_wait_queue(&wait_queue
);
1522 health_code_update();
1523 /* Dequeue command for registration */
1524 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1526 DBG("Woken up but nothing in the UST command queue");
1527 /* Continue thread execution */
1531 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1533 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1534 " gid:%d sock:%d name:%s (version %d.%d)",
1535 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1536 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1537 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1538 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1540 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1541 wait_node
= zmalloc(sizeof(*wait_node
));
1543 PERROR("zmalloc wait_node dispatch");
1544 ret
= close(ust_cmd
->sock
);
1546 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1548 lttng_fd_put(1, LTTNG_FD_APPS
);
1552 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1554 /* Create application object if socket is CMD. */
1555 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1557 if (!wait_node
->app
) {
1558 ret
= close(ust_cmd
->sock
);
1560 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1562 lttng_fd_put(1, LTTNG_FD_APPS
);
1568 * Add application to the wait queue so we can set the notify
1569 * socket before putting this object in the global ht.
1571 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1576 * We have to continue here since we don't have the notify
1577 * socket and the application MUST be added to the hash table
1578 * only at that moment.
1583 * Look for the application in the local wait queue and set the
1584 * notify socket if found.
1586 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1587 &wait_queue
.head
, head
) {
1588 health_code_update();
1589 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1590 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1591 cds_list_del(&wait_node
->head
);
1593 app
= wait_node
->app
;
1595 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1601 * With no application at this stage the received socket is
1602 * basically useless so close it before we free the cmd data
1603 * structure for good.
1606 ret
= close(ust_cmd
->sock
);
1608 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1610 lttng_fd_put(1, LTTNG_FD_APPS
);
1617 * @session_lock_list
1619 * Lock the global session list so from the register up to the
1620 * registration done message, no thread can see the application
1621 * and change its state.
1623 session_lock_list();
1627 * Add application to the global hash table. This needs to be
1628 * done before the update to the UST registry can locate the
1633 /* Set app version. This call will print an error if needed. */
1634 (void) ust_app_version(app
);
1636 /* Send notify socket through the notify pipe. */
1637 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1641 session_unlock_list();
1642 /* No notify thread, stop the UST tracing. */
1647 * Update newly registered application with the tracing
1648 * registry info already enabled information.
1650 update_ust_app(app
->sock
);
1653 * Don't care about return value. Let the manage apps threads
1654 * handle app unregistration upon socket close.
1656 (void) ust_app_register_done(app
->sock
);
1659 * Even if the application socket has been closed, send the app
1660 * to the thread and unregistration will take place at that
1663 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1666 session_unlock_list();
1667 /* No apps. thread, stop the UST tracing. */
1672 session_unlock_list();
1674 } while (node
!= NULL
);
1676 health_poll_entry();
1677 /* Futex wait on queue. Blocking call on futex() */
1678 futex_nto1_wait(&ust_cmd_queue
.futex
);
1681 /* Normal exit, no error */
1685 /* Clean up wait queue. */
1686 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1687 &wait_queue
.head
, head
) {
1688 cds_list_del(&wait_node
->head
);
1693 DBG("Dispatch thread dying");
1696 ERR("Health error occurred in %s", __func__
);
1698 health_unregister();
1703 * This thread manage application registration.
1705 static void *thread_registration_apps(void *data
)
1707 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1708 uint32_t revents
, nb_fd
;
1709 struct lttng_poll_event events
;
1711 * Get allocated in this thread, enqueued to a global queue, dequeued and
1712 * freed in the manage apps thread.
1714 struct ust_command
*ust_cmd
= NULL
;
1716 DBG("[thread] Manage application registration started");
1718 health_register(HEALTH_TYPE_APP_REG
);
1720 if (testpoint(thread_registration_apps
)) {
1721 goto error_testpoint
;
1724 ret
= lttcomm_listen_unix_sock(apps_sock
);
1730 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1731 * more will be added to this poll set.
1733 ret
= sessiond_set_thread_pollset(&events
, 2);
1735 goto error_create_poll
;
1738 /* Add the application registration socket */
1739 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1741 goto error_poll_add
;
1744 /* Notify all applications to register */
1745 ret
= notify_ust_apps(1);
1747 ERR("Failed to notify applications or create the wait shared memory.\n"
1748 "Execution continues but there might be problem for already\n"
1749 "running applications that wishes to register.");
1753 DBG("Accepting application registration");
1755 /* Inifinite blocking call, waiting for transmission */
1757 health_poll_entry();
1758 ret
= lttng_poll_wait(&events
, -1);
1762 * Restart interrupted system call.
1764 if (errno
== EINTR
) {
1772 for (i
= 0; i
< nb_fd
; i
++) {
1773 health_code_update();
1775 /* Fetch once the poll data */
1776 revents
= LTTNG_POLL_GETEV(&events
, i
);
1777 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1779 /* Thread quit pipe has been closed. Killing thread. */
1780 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1786 /* Event on the registration socket */
1787 if (pollfd
== apps_sock
) {
1788 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1789 ERR("Register apps socket poll error");
1791 } else if (revents
& LPOLLIN
) {
1792 sock
= lttcomm_accept_unix_sock(apps_sock
);
1798 * Set the CLOEXEC flag. Return code is useless because
1799 * either way, the show must go on.
1801 (void) utils_set_fd_cloexec(sock
);
1803 /* Create UST registration command for enqueuing */
1804 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1805 if (ust_cmd
== NULL
) {
1806 PERROR("ust command zmalloc");
1811 * Using message-based transmissions to ensure we don't
1812 * have to deal with partially received messages.
1814 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1816 ERR("Exhausted file descriptors allowed for applications.");
1826 health_code_update();
1827 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1830 /* Close socket of the application. */
1835 lttng_fd_put(LTTNG_FD_APPS
, 1);
1839 health_code_update();
1841 ust_cmd
->sock
= sock
;
1844 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1845 " gid:%d sock:%d name:%s (version %d.%d)",
1846 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1847 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1848 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1849 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1852 * Lock free enqueue the registration request. The red pill
1853 * has been taken! This apps will be part of the *system*.
1855 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1858 * Wake the registration queue futex. Implicit memory
1859 * barrier with the exchange in cds_wfq_enqueue.
1861 futex_nto1_wake(&ust_cmd_queue
.futex
);
1871 ERR("Health error occurred in %s", __func__
);
1874 /* Notify that the registration thread is gone */
1877 if (apps_sock
>= 0) {
1878 ret
= close(apps_sock
);
1888 lttng_fd_put(LTTNG_FD_APPS
, 1);
1890 unlink(apps_unix_sock_path
);
1893 lttng_poll_clean(&events
);
1897 DBG("UST Registration thread cleanup complete");
1898 health_unregister();
1904 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1905 * exec or it will fails.
1907 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1910 struct timespec timeout
;
1912 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1913 consumer_data
->consumer_thread_is_ready
= 0;
1915 /* Setup pthread condition */
1916 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1919 PERROR("pthread_condattr_init consumer data");
1924 * Set the monotonic clock in order to make sure we DO NOT jump in time
1925 * between the clock_gettime() call and the timedwait call. See bug #324
1926 * for a more details and how we noticed it.
1928 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1931 PERROR("pthread_condattr_setclock consumer data");
1935 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1938 PERROR("pthread_cond_init consumer data");
1942 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1945 PERROR("pthread_create consumer");
1950 /* We are about to wait on a pthread condition */
1951 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1953 /* Get time for sem_timedwait absolute timeout */
1954 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1956 * Set the timeout for the condition timed wait even if the clock gettime
1957 * call fails since we might loop on that call and we want to avoid to
1958 * increment the timeout too many times.
1960 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1963 * The following loop COULD be skipped in some conditions so this is why we
1964 * set ret to 0 in order to make sure at least one round of the loop is
1970 * Loop until the condition is reached or when a timeout is reached. Note
1971 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1972 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1973 * possible. This loop does not take any chances and works with both of
1976 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1977 if (clock_ret
< 0) {
1978 PERROR("clock_gettime spawn consumer");
1979 /* Infinite wait for the consumerd thread to be ready */
1980 ret
= pthread_cond_wait(&consumer_data
->cond
,
1981 &consumer_data
->cond_mutex
);
1983 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1984 &consumer_data
->cond_mutex
, &timeout
);
1988 /* Release the pthread condition */
1989 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1993 if (ret
== ETIMEDOUT
) {
1995 * Call has timed out so we kill the kconsumerd_thread and return
1998 ERR("Condition timed out. The consumer thread was never ready."
2000 ret
= pthread_cancel(consumer_data
->thread
);
2002 PERROR("pthread_cancel consumer thread");
2005 PERROR("pthread_cond_wait failed consumer thread");
2010 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2011 if (consumer_data
->pid
== 0) {
2012 ERR("Consumerd did not start");
2013 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2016 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2025 * Join consumer thread
2027 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2031 /* Consumer pid must be a real one. */
2032 if (consumer_data
->pid
> 0) {
2034 ret
= kill(consumer_data
->pid
, SIGTERM
);
2036 ERR("Error killing consumer daemon");
2039 return pthread_join(consumer_data
->thread
, &status
);
2046 * Fork and exec a consumer daemon (consumerd).
2048 * Return pid if successful else -1.
2050 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2054 const char *consumer_to_use
;
2055 const char *verbosity
;
2058 DBG("Spawning consumerd");
2065 if (opt_verbose_consumer
) {
2066 verbosity
= "--verbose";
2068 verbosity
= "--quiet";
2070 switch (consumer_data
->type
) {
2071 case LTTNG_CONSUMER_KERNEL
:
2073 * Find out which consumerd to execute. We will first try the
2074 * 64-bit path, then the sessiond's installation directory, and
2075 * fallback on the 32-bit one,
2077 DBG3("Looking for a kernel consumer at these locations:");
2078 DBG3(" 1) %s", consumerd64_bin
);
2079 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2080 DBG3(" 3) %s", consumerd32_bin
);
2081 if (stat(consumerd64_bin
, &st
) == 0) {
2082 DBG3("Found location #1");
2083 consumer_to_use
= consumerd64_bin
;
2084 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2085 DBG3("Found location #2");
2086 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2087 } else if (stat(consumerd32_bin
, &st
) == 0) {
2088 DBG3("Found location #3");
2089 consumer_to_use
= consumerd32_bin
;
2091 DBG("Could not find any valid consumerd executable");
2094 DBG("Using kernel consumer at: %s", consumer_to_use
);
2095 execl(consumer_to_use
,
2096 "lttng-consumerd", verbosity
, "-k",
2097 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2098 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2101 case LTTNG_CONSUMER64_UST
:
2103 char *tmpnew
= NULL
;
2105 if (consumerd64_libdir
[0] != '\0') {
2109 tmp
= getenv("LD_LIBRARY_PATH");
2113 tmplen
= strlen("LD_LIBRARY_PATH=")
2114 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2115 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2120 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2121 strcat(tmpnew
, consumerd64_libdir
);
2122 if (tmp
[0] != '\0') {
2123 strcat(tmpnew
, ":");
2124 strcat(tmpnew
, tmp
);
2126 ret
= putenv(tmpnew
);
2133 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2134 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2135 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2136 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2138 if (consumerd64_libdir
[0] != '\0') {
2146 case LTTNG_CONSUMER32_UST
:
2148 char *tmpnew
= NULL
;
2150 if (consumerd32_libdir
[0] != '\0') {
2154 tmp
= getenv("LD_LIBRARY_PATH");
2158 tmplen
= strlen("LD_LIBRARY_PATH=")
2159 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2160 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2165 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2166 strcat(tmpnew
, consumerd32_libdir
);
2167 if (tmp
[0] != '\0') {
2168 strcat(tmpnew
, ":");
2169 strcat(tmpnew
, tmp
);
2171 ret
= putenv(tmpnew
);
2178 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2179 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2180 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2181 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2183 if (consumerd32_libdir
[0] != '\0') {
2192 PERROR("unknown consumer type");
2196 PERROR("kernel start consumer exec");
2199 } else if (pid
> 0) {
2202 PERROR("start consumer fork");
2210 * Spawn the consumerd daemon and session daemon thread.
2212 static int start_consumerd(struct consumer_data
*consumer_data
)
2217 * Set the listen() state on the socket since there is a possible race
2218 * between the exec() of the consumer daemon and this call if place in the
2219 * consumer thread. See bug #366 for more details.
2221 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2226 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2227 if (consumer_data
->pid
!= 0) {
2228 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2232 ret
= spawn_consumerd(consumer_data
);
2234 ERR("Spawning consumerd failed");
2235 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2239 /* Setting up the consumer_data pid */
2240 consumer_data
->pid
= ret
;
2241 DBG2("Consumer pid %d", consumer_data
->pid
);
2242 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2244 DBG2("Spawning consumer control thread");
2245 ret
= spawn_consumer_thread(consumer_data
);
2247 ERR("Fatal error spawning consumer control thread");
2255 /* Cleanup already created sockets on error. */
2256 if (consumer_data
->err_sock
>= 0) {
2259 err
= close(consumer_data
->err_sock
);
2261 PERROR("close consumer data error socket");
2268 * Compute health status of each consumer. If one of them is zero (bad
2269 * state), we return 0.
2271 static int check_consumer_health(void)
2275 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2277 DBG3("Health consumer check %d", ret
);
2283 * Setup necessary data for kernel tracer action.
2285 static int init_kernel_tracer(void)
2289 /* Modprobe lttng kernel modules */
2290 ret
= modprobe_lttng_control();
2295 /* Open debugfs lttng */
2296 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2297 if (kernel_tracer_fd
< 0) {
2298 DBG("Failed to open %s", module_proc_lttng
);
2303 /* Validate kernel version */
2304 ret
= kernel_validate_version(kernel_tracer_fd
);
2309 ret
= modprobe_lttng_data();
2314 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2318 modprobe_remove_lttng_control();
2319 ret
= close(kernel_tracer_fd
);
2323 kernel_tracer_fd
= -1;
2324 return LTTNG_ERR_KERN_VERSION
;
2327 ret
= close(kernel_tracer_fd
);
2333 modprobe_remove_lttng_control();
2336 WARN("No kernel tracer available");
2337 kernel_tracer_fd
= -1;
2339 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2341 return LTTNG_ERR_KERN_NA
;
2347 * Copy consumer output from the tracing session to the domain session. The
2348 * function also applies the right modification on a per domain basis for the
2349 * trace files destination directory.
2351 * Should *NOT* be called with RCU read-side lock held.
2353 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2356 const char *dir_name
;
2357 struct consumer_output
*consumer
;
2360 assert(session
->consumer
);
2363 case LTTNG_DOMAIN_KERNEL
:
2364 DBG3("Copying tracing session consumer output in kernel session");
2366 * XXX: We should audit the session creation and what this function
2367 * does "extra" in order to avoid a destroy since this function is used
2368 * in the domain session creation (kernel and ust) only. Same for UST
2371 if (session
->kernel_session
->consumer
) {
2372 consumer_destroy_output(session
->kernel_session
->consumer
);
2374 session
->kernel_session
->consumer
=
2375 consumer_copy_output(session
->consumer
);
2376 /* Ease our life a bit for the next part */
2377 consumer
= session
->kernel_session
->consumer
;
2378 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2380 case LTTNG_DOMAIN_UST
:
2381 DBG3("Copying tracing session consumer output in UST session");
2382 if (session
->ust_session
->consumer
) {
2383 consumer_destroy_output(session
->ust_session
->consumer
);
2385 session
->ust_session
->consumer
=
2386 consumer_copy_output(session
->consumer
);
2387 /* Ease our life a bit for the next part */
2388 consumer
= session
->ust_session
->consumer
;
2389 dir_name
= DEFAULT_UST_TRACE_DIR
;
2392 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2396 /* Append correct directory to subdir */
2397 strncat(consumer
->subdir
, dir_name
,
2398 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2399 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2408 * Create an UST session and add it to the session ust list.
2410 * Should *NOT* be called with RCU read-side lock held.
2412 static int create_ust_session(struct ltt_session
*session
,
2413 struct lttng_domain
*domain
)
2416 struct ltt_ust_session
*lus
= NULL
;
2420 assert(session
->consumer
);
2422 switch (domain
->type
) {
2423 case LTTNG_DOMAIN_UST
:
2426 ERR("Unknown UST domain on create session %d", domain
->type
);
2427 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2431 DBG("Creating UST session");
2433 lus
= trace_ust_create_session(session
->id
);
2435 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2439 lus
->uid
= session
->uid
;
2440 lus
->gid
= session
->gid
;
2441 lus
->output_traces
= session
->output_traces
;
2442 lus
->snapshot_mode
= session
->snapshot_mode
;
2443 session
->ust_session
= lus
;
2445 /* Copy session output to the newly created UST session */
2446 ret
= copy_session_consumer(domain
->type
, session
);
2447 if (ret
!= LTTNG_OK
) {
2455 session
->ust_session
= NULL
;
2460 * Create a kernel tracer session then create the default channel.
2462 static int create_kernel_session(struct ltt_session
*session
)
2466 DBG("Creating kernel session");
2468 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2470 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2474 /* Code flow safety */
2475 assert(session
->kernel_session
);
2477 /* Copy session output to the newly created Kernel session */
2478 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2479 if (ret
!= LTTNG_OK
) {
2483 /* Create directory(ies) on local filesystem. */
2484 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2485 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2486 ret
= run_as_mkdir_recursive(
2487 session
->kernel_session
->consumer
->dst
.trace_path
,
2488 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2490 if (ret
!= -EEXIST
) {
2491 ERR("Trace directory creation error");
2497 session
->kernel_session
->uid
= session
->uid
;
2498 session
->kernel_session
->gid
= session
->gid
;
2499 session
->kernel_session
->output_traces
= session
->output_traces
;
2500 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2505 trace_kernel_destroy_session(session
->kernel_session
);
2506 session
->kernel_session
= NULL
;
2511 * Count number of session permitted by uid/gid.
2513 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2516 struct ltt_session
*session
;
2518 DBG("Counting number of available session for UID %d GID %d",
2520 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2522 * Only list the sessions the user can control.
2524 if (!session_access_ok(session
, uid
, gid
)) {
2533 * Process the command requested by the lttng client within the command
2534 * context structure. This function make sure that the return structure (llm)
2535 * is set and ready for transmission before returning.
2537 * Return any error encountered or 0 for success.
2539 * "sock" is only used for special-case var. len data.
2541 * Should *NOT* be called with RCU read-side lock held.
2543 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2547 int need_tracing_session
= 1;
2550 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2554 switch (cmd_ctx
->lsm
->cmd_type
) {
2555 case LTTNG_CREATE_SESSION
:
2556 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2557 case LTTNG_DESTROY_SESSION
:
2558 case LTTNG_LIST_SESSIONS
:
2559 case LTTNG_LIST_DOMAINS
:
2560 case LTTNG_START_TRACE
:
2561 case LTTNG_STOP_TRACE
:
2562 case LTTNG_DATA_PENDING
:
2563 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2564 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2565 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2566 case LTTNG_SNAPSHOT_RECORD
:
2573 if (opt_no_kernel
&& need_domain
2574 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2576 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2578 ret
= LTTNG_ERR_KERN_NA
;
2583 /* Deny register consumer if we already have a spawned consumer. */
2584 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2585 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2586 if (kconsumer_data
.pid
> 0) {
2587 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2588 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2591 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2595 * Check for command that don't needs to allocate a returned payload. We do
2596 * this here so we don't have to make the call for no payload at each
2599 switch(cmd_ctx
->lsm
->cmd_type
) {
2600 case LTTNG_LIST_SESSIONS
:
2601 case LTTNG_LIST_TRACEPOINTS
:
2602 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2603 case LTTNG_LIST_DOMAINS
:
2604 case LTTNG_LIST_CHANNELS
:
2605 case LTTNG_LIST_EVENTS
:
2608 /* Setup lttng message with no payload */
2609 ret
= setup_lttng_msg(cmd_ctx
, 0);
2611 /* This label does not try to unlock the session */
2612 goto init_setup_error
;
2616 /* Commands that DO NOT need a session. */
2617 switch (cmd_ctx
->lsm
->cmd_type
) {
2618 case LTTNG_CREATE_SESSION
:
2619 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2620 case LTTNG_CALIBRATE
:
2621 case LTTNG_LIST_SESSIONS
:
2622 case LTTNG_LIST_TRACEPOINTS
:
2623 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2624 need_tracing_session
= 0;
2627 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2629 * We keep the session list lock across _all_ commands
2630 * for now, because the per-session lock does not
2631 * handle teardown properly.
2633 session_lock_list();
2634 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2635 if (cmd_ctx
->session
== NULL
) {
2636 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2639 /* Acquire lock for the session */
2640 session_lock(cmd_ctx
->session
);
2650 * Check domain type for specific "pre-action".
2652 switch (cmd_ctx
->lsm
->domain
.type
) {
2653 case LTTNG_DOMAIN_KERNEL
:
2655 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2659 /* Kernel tracer check */
2660 if (kernel_tracer_fd
== -1) {
2661 /* Basically, load kernel tracer modules */
2662 ret
= init_kernel_tracer();
2668 /* Consumer is in an ERROR state. Report back to client */
2669 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2670 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2674 /* Need a session for kernel command */
2675 if (need_tracing_session
) {
2676 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2677 ret
= create_kernel_session(cmd_ctx
->session
);
2679 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2684 /* Start the kernel consumer daemon */
2685 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2686 if (kconsumer_data
.pid
== 0 &&
2687 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2688 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2689 ret
= start_consumerd(&kconsumer_data
);
2691 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2694 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2696 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2700 * The consumer was just spawned so we need to add the socket to
2701 * the consumer output of the session if exist.
2703 ret
= consumer_create_socket(&kconsumer_data
,
2704 cmd_ctx
->session
->kernel_session
->consumer
);
2711 case LTTNG_DOMAIN_UST
:
2713 /* Consumer is in an ERROR state. Report back to client */
2714 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2715 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2719 if (need_tracing_session
) {
2720 /* Create UST session if none exist. */
2721 if (cmd_ctx
->session
->ust_session
== NULL
) {
2722 ret
= create_ust_session(cmd_ctx
->session
,
2723 &cmd_ctx
->lsm
->domain
);
2724 if (ret
!= LTTNG_OK
) {
2729 /* Start the UST consumer daemons */
2731 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2732 if (consumerd64_bin
[0] != '\0' &&
2733 ustconsumer64_data
.pid
== 0 &&
2734 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2735 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2736 ret
= start_consumerd(&ustconsumer64_data
);
2738 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2739 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2743 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2744 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2746 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2750 * Setup socket for consumer 64 bit. No need for atomic access
2751 * since it was set above and can ONLY be set in this thread.
2753 ret
= consumer_create_socket(&ustconsumer64_data
,
2754 cmd_ctx
->session
->ust_session
->consumer
);
2760 if (consumerd32_bin
[0] != '\0' &&
2761 ustconsumer32_data
.pid
== 0 &&
2762 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2763 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2764 ret
= start_consumerd(&ustconsumer32_data
);
2766 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2767 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2771 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2772 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2774 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2778 * Setup socket for consumer 64 bit. No need for atomic access
2779 * since it was set above and can ONLY be set in this thread.
2781 ret
= consumer_create_socket(&ustconsumer32_data
,
2782 cmd_ctx
->session
->ust_session
->consumer
);
2794 /* Validate consumer daemon state when start/stop trace command */
2795 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2796 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2797 switch (cmd_ctx
->lsm
->domain
.type
) {
2798 case LTTNG_DOMAIN_UST
:
2799 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2800 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2804 case LTTNG_DOMAIN_KERNEL
:
2805 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2806 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2814 * Check that the UID or GID match that of the tracing session.
2815 * The root user can interact with all sessions.
2817 if (need_tracing_session
) {
2818 if (!session_access_ok(cmd_ctx
->session
,
2819 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2820 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2821 ret
= LTTNG_ERR_EPERM
;
2827 * Send relayd information to consumer as soon as we have a domain and a
2830 if (cmd_ctx
->session
&& need_domain
) {
2832 * Setup relayd if not done yet. If the relayd information was already
2833 * sent to the consumer, this call will gracefully return.
2835 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2836 if (ret
!= LTTNG_OK
) {
2841 /* Process by command type */
2842 switch (cmd_ctx
->lsm
->cmd_type
) {
2843 case LTTNG_ADD_CONTEXT
:
2845 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2846 cmd_ctx
->lsm
->u
.context
.channel_name
,
2847 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2850 case LTTNG_DISABLE_CHANNEL
:
2852 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2853 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2856 case LTTNG_DISABLE_EVENT
:
2858 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2859 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2860 cmd_ctx
->lsm
->u
.disable
.name
);
2863 case LTTNG_DISABLE_ALL_EVENT
:
2865 DBG("Disabling all events");
2867 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2868 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2871 case LTTNG_ENABLE_CHANNEL
:
2873 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2874 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2877 case LTTNG_ENABLE_EVENT
:
2879 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2880 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2881 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2884 case LTTNG_ENABLE_ALL_EVENT
:
2886 DBG("Enabling all events");
2888 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2889 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2890 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2893 case LTTNG_LIST_TRACEPOINTS
:
2895 struct lttng_event
*events
;
2898 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2899 if (nb_events
< 0) {
2900 /* Return value is a negative lttng_error_code. */
2906 * Setup lttng message with payload size set to the event list size in
2907 * bytes and then copy list into the llm payload.
2909 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2915 /* Copy event list into message payload */
2916 memcpy(cmd_ctx
->llm
->payload
, events
,
2917 sizeof(struct lttng_event
) * nb_events
);
2924 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2926 struct lttng_event_field
*fields
;
2929 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2931 if (nb_fields
< 0) {
2932 /* Return value is a negative lttng_error_code. */
2938 * Setup lttng message with payload size set to the event list size in
2939 * bytes and then copy list into the llm payload.
2941 ret
= setup_lttng_msg(cmd_ctx
,
2942 sizeof(struct lttng_event_field
) * nb_fields
);
2948 /* Copy event list into message payload */
2949 memcpy(cmd_ctx
->llm
->payload
, fields
,
2950 sizeof(struct lttng_event_field
) * nb_fields
);
2957 case LTTNG_SET_CONSUMER_URI
:
2960 struct lttng_uri
*uris
;
2962 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2963 len
= nb_uri
* sizeof(struct lttng_uri
);
2966 ret
= LTTNG_ERR_INVALID
;
2970 uris
= zmalloc(len
);
2972 ret
= LTTNG_ERR_FATAL
;
2976 /* Receive variable len data */
2977 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2978 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2980 DBG("No URIs received from client... continuing");
2982 ret
= LTTNG_ERR_SESSION_FAIL
;
2987 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2989 if (ret
!= LTTNG_OK
) {
2995 * XXX: 0 means that this URI should be applied on the session. Should
2996 * be a DOMAIN enuam.
2998 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2999 /* Add the URI for the UST session if a consumer is present. */
3000 if (cmd_ctx
->session
->ust_session
&&
3001 cmd_ctx
->session
->ust_session
->consumer
) {
3002 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3004 } else if (cmd_ctx
->session
->kernel_session
&&
3005 cmd_ctx
->session
->kernel_session
->consumer
) {
3006 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3007 cmd_ctx
->session
, nb_uri
, uris
);
3015 case LTTNG_START_TRACE
:
3017 ret
= cmd_start_trace(cmd_ctx
->session
);
3020 case LTTNG_STOP_TRACE
:
3022 ret
= cmd_stop_trace(cmd_ctx
->session
);
3025 case LTTNG_CREATE_SESSION
:
3028 struct lttng_uri
*uris
= NULL
;
3030 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3031 len
= nb_uri
* sizeof(struct lttng_uri
);
3034 uris
= zmalloc(len
);
3036 ret
= LTTNG_ERR_FATAL
;
3040 /* Receive variable len data */
3041 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3042 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3044 DBG("No URIs received from client... continuing");
3046 ret
= LTTNG_ERR_SESSION_FAIL
;
3051 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3052 DBG("Creating session with ONE network URI is a bad call");
3053 ret
= LTTNG_ERR_SESSION_FAIL
;
3059 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3066 case LTTNG_DESTROY_SESSION
:
3068 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3070 /* Set session to NULL so we do not unlock it after free. */
3071 cmd_ctx
->session
= NULL
;
3074 case LTTNG_LIST_DOMAINS
:
3077 struct lttng_domain
*domains
;
3079 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3081 /* Return value is a negative lttng_error_code. */
3086 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3092 /* Copy event list into message payload */
3093 memcpy(cmd_ctx
->llm
->payload
, domains
,
3094 nb_dom
* sizeof(struct lttng_domain
));
3101 case LTTNG_LIST_CHANNELS
:
3104 struct lttng_channel
*channels
;
3106 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3107 cmd_ctx
->session
, &channels
);
3109 /* Return value is a negative lttng_error_code. */
3114 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3120 /* Copy event list into message payload */
3121 memcpy(cmd_ctx
->llm
->payload
, channels
,
3122 nb_chan
* sizeof(struct lttng_channel
));
3129 case LTTNG_LIST_EVENTS
:
3132 struct lttng_event
*events
= NULL
;
3134 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3135 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3137 /* Return value is a negative lttng_error_code. */
3142 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3148 /* Copy event list into message payload */
3149 memcpy(cmd_ctx
->llm
->payload
, events
,
3150 nb_event
* sizeof(struct lttng_event
));
3157 case LTTNG_LIST_SESSIONS
:
3159 unsigned int nr_sessions
;
3161 session_lock_list();
3162 nr_sessions
= lttng_sessions_count(
3163 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3164 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3166 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3168 session_unlock_list();
3172 /* Filled the session array */
3173 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3174 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3175 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3177 session_unlock_list();
3182 case LTTNG_CALIBRATE
:
3184 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3185 &cmd_ctx
->lsm
->u
.calibrate
);
3188 case LTTNG_REGISTER_CONSUMER
:
3190 struct consumer_data
*cdata
;
3192 switch (cmd_ctx
->lsm
->domain
.type
) {
3193 case LTTNG_DOMAIN_KERNEL
:
3194 cdata
= &kconsumer_data
;
3197 ret
= LTTNG_ERR_UND
;
3201 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3202 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3205 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3207 struct lttng_filter_bytecode
*bytecode
;
3209 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3210 ret
= LTTNG_ERR_FILTER_INVAL
;
3213 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3214 ret
= LTTNG_ERR_FILTER_INVAL
;
3217 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3219 ret
= LTTNG_ERR_FILTER_NOMEM
;
3222 /* Receive var. len. data */
3223 DBG("Receiving var len data from client ...");
3224 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3225 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3227 DBG("Nothing recv() from client var len data... continuing");
3229 ret
= LTTNG_ERR_FILTER_INVAL
;
3233 if (bytecode
->len
+ sizeof(*bytecode
)
3234 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3236 ret
= LTTNG_ERR_FILTER_INVAL
;
3240 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3241 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3242 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3245 case LTTNG_DATA_PENDING
:
3247 ret
= cmd_data_pending(cmd_ctx
->session
);
3250 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3252 struct lttcomm_lttng_output_id reply
;
3254 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3255 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3256 if (ret
!= LTTNG_OK
) {
3260 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3265 /* Copy output list into message payload */
3266 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3270 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3272 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3273 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3276 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3279 struct lttng_snapshot_output
*outputs
= NULL
;
3281 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3282 if (nb_output
< 0) {
3287 ret
= setup_lttng_msg(cmd_ctx
,
3288 nb_output
* sizeof(struct lttng_snapshot_output
));
3295 /* Copy output list into message payload */
3296 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3297 nb_output
* sizeof(struct lttng_snapshot_output
));
3304 case LTTNG_SNAPSHOT_RECORD
:
3306 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3307 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3308 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3311 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3314 struct lttng_uri
*uris
= NULL
;
3316 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3317 len
= nb_uri
* sizeof(struct lttng_uri
);
3320 uris
= zmalloc(len
);
3322 ret
= LTTNG_ERR_FATAL
;
3326 /* Receive variable len data */
3327 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3328 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3330 DBG("No URIs received from client... continuing");
3332 ret
= LTTNG_ERR_SESSION_FAIL
;
3337 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3338 DBG("Creating session with ONE network URI is a bad call");
3339 ret
= LTTNG_ERR_SESSION_FAIL
;
3345 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3346 nb_uri
, &cmd_ctx
->creds
);
3351 ret
= LTTNG_ERR_UND
;
3356 if (cmd_ctx
->llm
== NULL
) {
3357 DBG("Missing llm structure. Allocating one.");
3358 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3362 /* Set return code */
3363 cmd_ctx
->llm
->ret_code
= ret
;
3365 if (cmd_ctx
->session
) {
3366 session_unlock(cmd_ctx
->session
);
3368 if (need_tracing_session
) {
3369 session_unlock_list();
3376 * Thread managing health check socket.
3378 static void *thread_manage_health(void *data
)
3380 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3381 uint32_t revents
, nb_fd
;
3382 struct lttng_poll_event events
;
3383 struct lttcomm_health_msg msg
;
3384 struct lttcomm_health_data reply
;
3386 DBG("[thread] Manage health check started");
3388 rcu_register_thread();
3390 /* We might hit an error path before this is created. */
3391 lttng_poll_init(&events
);
3393 /* Create unix socket */
3394 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3396 ERR("Unable to create health check Unix socket");
3402 * Set the CLOEXEC flag. Return code is useless because either way, the
3405 (void) utils_set_fd_cloexec(sock
);
3407 ret
= lttcomm_listen_unix_sock(sock
);
3413 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3414 * more will be added to this poll set.
3416 ret
= sessiond_set_thread_pollset(&events
, 2);
3421 /* Add the application registration socket */
3422 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3428 DBG("Health check ready");
3430 /* Inifinite blocking call, waiting for transmission */
3432 ret
= lttng_poll_wait(&events
, -1);
3435 * Restart interrupted system call.
3437 if (errno
== EINTR
) {
3445 for (i
= 0; i
< nb_fd
; i
++) {
3446 /* Fetch once the poll data */
3447 revents
= LTTNG_POLL_GETEV(&events
, i
);
3448 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3450 /* Thread quit pipe has been closed. Killing thread. */
3451 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3457 /* Event on the registration socket */
3458 if (pollfd
== sock
) {
3459 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3460 ERR("Health socket poll error");
3466 new_sock
= lttcomm_accept_unix_sock(sock
);
3472 * Set the CLOEXEC flag. Return code is useless because either way, the
3475 (void) utils_set_fd_cloexec(new_sock
);
3477 DBG("Receiving data from client for health...");
3478 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3480 DBG("Nothing recv() from client... continuing");
3481 ret
= close(new_sock
);
3489 rcu_thread_online();
3491 switch (msg
.component
) {
3492 case LTTNG_HEALTH_CMD
:
3493 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3495 case LTTNG_HEALTH_APP_MANAGE
:
3496 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3498 case LTTNG_HEALTH_APP_REG
:
3499 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3501 case LTTNG_HEALTH_KERNEL
:
3502 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3504 case LTTNG_HEALTH_CONSUMER
:
3505 reply
.ret_code
= check_consumer_health();
3507 case LTTNG_HEALTH_HT_CLEANUP
:
3508 reply
.ret_code
= health_check_state(HEALTH_TYPE_HT_CLEANUP
);
3510 case LTTNG_HEALTH_APP_MANAGE_NOTIFY
:
3511 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3513 case LTTNG_HEALTH_APP_REG_DISPATCH
:
3514 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3516 case LTTNG_HEALTH_ALL
:
3518 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3519 health_check_state(HEALTH_TYPE_APP_REG
) &&
3520 health_check_state(HEALTH_TYPE_CMD
) &&
3521 health_check_state(HEALTH_TYPE_KERNEL
) &&
3522 check_consumer_health() &&
3523 health_check_state(HEALTH_TYPE_HT_CLEANUP
) &&
3524 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
) &&
3525 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3528 reply
.ret_code
= LTTNG_ERR_UND
;
3533 * Flip ret value since 0 is a success and 1 indicates a bad health for
3534 * the client where in the sessiond it is the opposite. Again, this is
3535 * just to make things easier for us poor developer which enjoy a lot
3538 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3539 reply
.ret_code
= !reply
.ret_code
;
3542 DBG2("Health check return value %d", reply
.ret_code
);
3544 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3546 ERR("Failed to send health data back to client");
3549 /* End of transmission */
3550 ret
= close(new_sock
);
3560 ERR("Health error occurred in %s", __func__
);
3562 DBG("Health check thread dying");
3563 unlink(health_unix_sock_path
);
3571 lttng_poll_clean(&events
);
3573 rcu_unregister_thread();
3578 * This thread manage all clients request using the unix client socket for
3581 static void *thread_manage_clients(void *data
)
3583 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3585 uint32_t revents
, nb_fd
;
3586 struct command_ctx
*cmd_ctx
= NULL
;
3587 struct lttng_poll_event events
;
3589 DBG("[thread] Manage client started");
3591 rcu_register_thread();
3593 health_register(HEALTH_TYPE_CMD
);
3595 if (testpoint(thread_manage_clients
)) {
3596 goto error_testpoint
;
3599 health_code_update();
3601 ret
= lttcomm_listen_unix_sock(client_sock
);
3607 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3608 * more will be added to this poll set.
3610 ret
= sessiond_set_thread_pollset(&events
, 2);
3612 goto error_create_poll
;
3615 /* Add the application registration socket */
3616 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3622 * Notify parent pid that we are ready to accept command for client side.
3624 if (opt_sig_parent
) {
3625 kill(ppid
, SIGUSR1
);
3628 if (testpoint(thread_manage_clients_before_loop
)) {
3632 health_code_update();
3635 DBG("Accepting client command ...");
3637 /* Inifinite blocking call, waiting for transmission */
3639 health_poll_entry();
3640 ret
= lttng_poll_wait(&events
, -1);
3644 * Restart interrupted system call.
3646 if (errno
== EINTR
) {
3654 for (i
= 0; i
< nb_fd
; i
++) {
3655 /* Fetch once the poll data */
3656 revents
= LTTNG_POLL_GETEV(&events
, i
);
3657 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3659 health_code_update();
3661 /* Thread quit pipe has been closed. Killing thread. */
3662 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3668 /* Event on the registration socket */
3669 if (pollfd
== client_sock
) {
3670 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3671 ERR("Client socket poll error");
3677 DBG("Wait for client response");
3679 health_code_update();
3681 sock
= lttcomm_accept_unix_sock(client_sock
);
3687 * Set the CLOEXEC flag. Return code is useless because either way, the
3690 (void) utils_set_fd_cloexec(sock
);
3692 /* Set socket option for credentials retrieval */
3693 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3698 /* Allocate context command to process the client request */
3699 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3700 if (cmd_ctx
== NULL
) {
3701 PERROR("zmalloc cmd_ctx");
3705 /* Allocate data buffer for reception */
3706 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3707 if (cmd_ctx
->lsm
== NULL
) {
3708 PERROR("zmalloc cmd_ctx->lsm");
3712 cmd_ctx
->llm
= NULL
;
3713 cmd_ctx
->session
= NULL
;
3715 health_code_update();
3718 * Data is received from the lttng client. The struct
3719 * lttcomm_session_msg (lsm) contains the command and data request of
3722 DBG("Receiving data from client ...");
3723 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3724 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3726 DBG("Nothing recv() from client... continuing");
3732 clean_command_ctx(&cmd_ctx
);
3736 health_code_update();
3738 // TODO: Validate cmd_ctx including sanity check for
3739 // security purpose.
3741 rcu_thread_online();
3743 * This function dispatch the work to the kernel or userspace tracer
3744 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3745 * informations for the client. The command context struct contains
3746 * everything this function may needs.
3748 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3749 rcu_thread_offline();
3757 * TODO: Inform client somehow of the fatal error. At
3758 * this point, ret < 0 means that a zmalloc failed
3759 * (ENOMEM). Error detected but still accept
3760 * command, unless a socket error has been
3763 clean_command_ctx(&cmd_ctx
);
3767 health_code_update();
3769 DBG("Sending response (size: %d, retcode: %s)",
3770 cmd_ctx
->lttng_msg_size
,
3771 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3772 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3774 ERR("Failed to send data back to client");
3777 /* End of transmission */
3784 clean_command_ctx(&cmd_ctx
);
3786 health_code_update();
3798 lttng_poll_clean(&events
);
3799 clean_command_ctx(&cmd_ctx
);
3804 unlink(client_unix_sock_path
);
3805 if (client_sock
>= 0) {
3806 ret
= close(client_sock
);
3814 ERR("Health error occurred in %s", __func__
);
3817 health_unregister();
3819 DBG("Client thread dying");
3821 rcu_unregister_thread();
3827 * usage function on stderr
3829 static void usage(void)
3831 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3832 fprintf(stderr
, " -h, --help Display this usage.\n");
3833 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3834 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3835 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3836 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3837 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3838 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3839 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3840 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3841 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3842 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3843 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3844 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3845 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3846 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3847 fprintf(stderr
, " -V, --version Show version number.\n");
3848 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3849 fprintf(stderr
, " -q, --quiet No output at all.\n");
3850 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3851 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3852 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3853 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3857 * daemon argument parsing
3859 static int parse_args(int argc
, char **argv
)
3863 static struct option long_options
[] = {
3864 { "client-sock", 1, 0, 'c' },
3865 { "apps-sock", 1, 0, 'a' },
3866 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3867 { "kconsumerd-err-sock", 1, 0, 'E' },
3868 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3869 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3870 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3871 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3872 { "consumerd32-path", 1, 0, 'u' },
3873 { "consumerd32-libdir", 1, 0, 'U' },
3874 { "consumerd64-path", 1, 0, 't' },
3875 { "consumerd64-libdir", 1, 0, 'T' },
3876 { "daemonize", 0, 0, 'd' },
3877 { "sig-parent", 0, 0, 'S' },
3878 { "help", 0, 0, 'h' },
3879 { "group", 1, 0, 'g' },
3880 { "version", 0, 0, 'V' },
3881 { "quiet", 0, 0, 'q' },
3882 { "verbose", 0, 0, 'v' },
3883 { "verbose-consumer", 0, 0, 'Z' },
3884 { "no-kernel", 0, 0, 'N' },
3885 { "pidfile", 1, 0, 'p' },
3890 int option_index
= 0;
3891 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3892 long_options
, &option_index
);
3899 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3901 fprintf(stderr
, " with arg %s\n", optarg
);
3905 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3908 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3914 opt_tracing_group
= optarg
;
3920 fprintf(stdout
, "%s\n", VERSION
);
3926 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3929 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3932 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3935 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3938 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3941 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3947 lttng_opt_quiet
= 1;
3950 /* Verbose level can increase using multiple -v */
3951 lttng_opt_verbose
+= 1;
3954 opt_verbose_consumer
+= 1;
3957 consumerd32_bin
= optarg
;
3960 consumerd32_libdir
= optarg
;
3963 consumerd64_bin
= optarg
;
3966 consumerd64_libdir
= optarg
;
3969 opt_pidfile
= optarg
;
3972 /* Unknown option or other error.
3973 * Error is printed by getopt, just return */
3982 * Creates the two needed socket by the daemon.
3983 * apps_sock - The communication socket for all UST apps.
3984 * client_sock - The communication of the cli tool (lttng).
3986 static int init_daemon_socket(void)
3991 old_umask
= umask(0);
3993 /* Create client tool unix socket */
3994 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3995 if (client_sock
< 0) {
3996 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4001 /* Set the cloexec flag */
4002 ret
= utils_set_fd_cloexec(client_sock
);
4004 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4005 "Continuing but note that the consumer daemon will have a "
4006 "reference to this socket on exec()", client_sock
);
4009 /* File permission MUST be 660 */
4010 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4012 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4017 /* Create the application unix socket */
4018 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4019 if (apps_sock
< 0) {
4020 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4025 /* Set the cloexec flag */
4026 ret
= utils_set_fd_cloexec(apps_sock
);
4028 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4029 "Continuing but note that the consumer daemon will have a "
4030 "reference to this socket on exec()", apps_sock
);
4033 /* File permission MUST be 666 */
4034 ret
= chmod(apps_unix_sock_path
,
4035 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4037 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4042 DBG3("Session daemon client socket %d and application socket %d created",
4043 client_sock
, apps_sock
);
4051 * Check if the global socket is available, and if a daemon is answering at the
4052 * other side. If yes, error is returned.
4054 static int check_existing_daemon(void)
4056 /* Is there anybody out there ? */
4057 if (lttng_session_daemon_alive()) {
4065 * Set the tracing group gid onto the client socket.
4067 * Race window between mkdir and chown is OK because we are going from more
4068 * permissive (root.root) to less permissive (root.tracing).
4070 static int set_permissions(char *rundir
)
4075 ret
= allowed_group();
4077 WARN("No tracing group detected");
4078 /* Setting gid to 0 if no tracing group is found */
4084 /* Set lttng run dir */
4085 ret
= chown(rundir
, 0, gid
);
4087 ERR("Unable to set group on %s", rundir
);
4091 /* Ensure all applications and tracing group can search the run dir */
4092 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
4094 ERR("Unable to set permissions on %s", rundir
);
4098 /* lttng client socket path */
4099 ret
= chown(client_unix_sock_path
, 0, gid
);
4101 ERR("Unable to set group on %s", client_unix_sock_path
);
4105 /* kconsumer error socket path */
4106 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
4108 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4112 /* 64-bit ustconsumer error socket path */
4113 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
4115 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4119 /* 32-bit ustconsumer compat32 error socket path */
4120 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
4122 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4126 DBG("All permissions are set");
4132 * Create the lttng run directory needed for all global sockets and pipe.
4134 static int create_lttng_rundir(const char *rundir
)
4138 DBG3("Creating LTTng run directory: %s", rundir
);
4140 ret
= mkdir(rundir
, S_IRWXU
);
4142 if (errno
!= EEXIST
) {
4143 ERR("Unable to create %s", rundir
);
4155 * Setup sockets and directory needed by the kconsumerd communication with the
4158 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4162 char path
[PATH_MAX
];
4164 switch (consumer_data
->type
) {
4165 case LTTNG_CONSUMER_KERNEL
:
4166 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4168 case LTTNG_CONSUMER64_UST
:
4169 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4171 case LTTNG_CONSUMER32_UST
:
4172 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4175 ERR("Consumer type unknown");
4180 DBG2("Creating consumer directory: %s", path
);
4182 ret
= mkdir(path
, S_IRWXU
);
4184 if (errno
!= EEXIST
) {
4186 ERR("Failed to create %s", path
);
4192 /* Create the kconsumerd error unix socket */
4193 consumer_data
->err_sock
=
4194 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4195 if (consumer_data
->err_sock
< 0) {
4196 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4202 * Set the CLOEXEC flag. Return code is useless because either way, the
4205 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4207 PERROR("utils_set_fd_cloexec");
4208 /* continue anyway */
4211 /* File permission MUST be 660 */
4212 ret
= chmod(consumer_data
->err_unix_sock_path
,
4213 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4215 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4225 * Signal handler for the daemon
4227 * Simply stop all worker threads, leaving main() return gracefully after
4228 * joining all threads and calling cleanup().
4230 static void sighandler(int sig
)
4234 DBG("SIGPIPE caught");
4237 DBG("SIGINT caught");
4241 DBG("SIGTERM caught");
4250 * Setup signal handler for :
4251 * SIGINT, SIGTERM, SIGPIPE
4253 static int set_signal_handler(void)
4256 struct sigaction sa
;
4259 if ((ret
= sigemptyset(&sigset
)) < 0) {
4260 PERROR("sigemptyset");
4264 sa
.sa_handler
= sighandler
;
4265 sa
.sa_mask
= sigset
;
4267 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4268 PERROR("sigaction");
4272 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4273 PERROR("sigaction");
4277 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4278 PERROR("sigaction");
4282 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4288 * Set open files limit to unlimited. This daemon can open a large number of
4289 * file descriptors in order to consumer multiple kernel traces.
4291 static void set_ulimit(void)
4296 /* The kernel does not allowed an infinite limit for open files */
4297 lim
.rlim_cur
= 65535;
4298 lim
.rlim_max
= 65535;
4300 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4302 PERROR("failed to set open files limit");
4307 * Write pidfile using the rundir and opt_pidfile.
4309 static void write_pidfile(void)
4312 char pidfile_path
[PATH_MAX
];
4317 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4319 /* Build pidfile path from rundir and opt_pidfile. */
4320 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4321 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4323 PERROR("snprintf pidfile path");
4329 * Create pid file in rundir. Return value is of no importance. The
4330 * execution will continue even though we are not able to write the file.
4332 (void) utils_create_pid_file(getpid(), pidfile_path
);
4341 int main(int argc
, char **argv
)
4345 const char *home_path
, *env_app_timeout
;
4347 init_kernel_workarounds();
4349 rcu_register_thread();
4351 setup_consumerd_path();
4353 page_size
= sysconf(_SC_PAGESIZE
);
4354 if (page_size
< 0) {
4355 PERROR("sysconf _SC_PAGESIZE");
4356 page_size
= LONG_MAX
;
4357 WARN("Fallback page size to %ld", page_size
);
4360 /* Parse arguments */
4362 if ((ret
= parse_args(argc
, argv
)) < 0) {
4372 * child: setsid, close FD 0, 1, 2, chdir /
4373 * parent: exit (if fork is successful)
4381 * We are in the child. Make sure all other file
4382 * descriptors are closed, in case we are called with
4383 * more opened file descriptors than the standard ones.
4385 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4390 /* Create thread quit pipe */
4391 if ((ret
= init_thread_quit_pipe()) < 0) {
4395 /* Check if daemon is UID = 0 */
4396 is_root
= !getuid();
4399 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4401 /* Create global run dir with root access */
4402 ret
= create_lttng_rundir(rundir
);
4407 if (strlen(apps_unix_sock_path
) == 0) {
4408 snprintf(apps_unix_sock_path
, PATH_MAX
,
4409 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4412 if (strlen(client_unix_sock_path
) == 0) {
4413 snprintf(client_unix_sock_path
, PATH_MAX
,
4414 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4417 /* Set global SHM for ust */
4418 if (strlen(wait_shm_path
) == 0) {
4419 snprintf(wait_shm_path
, PATH_MAX
,
4420 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4423 if (strlen(health_unix_sock_path
) == 0) {
4424 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4425 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4428 /* Setup kernel consumerd path */
4429 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4430 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4431 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4432 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4434 DBG2("Kernel consumer err path: %s",
4435 kconsumer_data
.err_unix_sock_path
);
4436 DBG2("Kernel consumer cmd path: %s",
4437 kconsumer_data
.cmd_unix_sock_path
);
4439 home_path
= utils_get_home_dir();
4440 if (home_path
== NULL
) {
4441 /* TODO: Add --socket PATH option */
4442 ERR("Can't get HOME directory for sockets creation.");
4448 * Create rundir from home path. This will create something like
4451 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4457 ret
= create_lttng_rundir(rundir
);
4462 if (strlen(apps_unix_sock_path
) == 0) {
4463 snprintf(apps_unix_sock_path
, PATH_MAX
,
4464 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4467 /* Set the cli tool unix socket path */
4468 if (strlen(client_unix_sock_path
) == 0) {
4469 snprintf(client_unix_sock_path
, PATH_MAX
,
4470 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4473 /* Set global SHM for ust */
4474 if (strlen(wait_shm_path
) == 0) {
4475 snprintf(wait_shm_path
, PATH_MAX
,
4476 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4479 /* Set health check Unix path */
4480 if (strlen(health_unix_sock_path
) == 0) {
4481 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4482 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4486 /* Set consumer initial state */
4487 kernel_consumerd_state
= CONSUMER_STOPPED
;
4488 ust_consumerd_state
= CONSUMER_STOPPED
;
4490 DBG("Client socket path %s", client_unix_sock_path
);
4491 DBG("Application socket path %s", apps_unix_sock_path
);
4492 DBG("Application wait path %s", wait_shm_path
);
4493 DBG("LTTng run directory path: %s", rundir
);
4495 /* 32 bits consumerd path setup */
4496 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4497 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4498 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4499 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4501 DBG2("UST consumer 32 bits err path: %s",
4502 ustconsumer32_data
.err_unix_sock_path
);
4503 DBG2("UST consumer 32 bits cmd path: %s",
4504 ustconsumer32_data
.cmd_unix_sock_path
);
4506 /* 64 bits consumerd path setup */
4507 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4508 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4509 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4510 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4512 DBG2("UST consumer 64 bits err path: %s",
4513 ustconsumer64_data
.err_unix_sock_path
);
4514 DBG2("UST consumer 64 bits cmd path: %s",
4515 ustconsumer64_data
.cmd_unix_sock_path
);
4518 * See if daemon already exist.
4520 if ((ret
= check_existing_daemon()) < 0) {
4521 ERR("Already running daemon.\n");
4523 * We do not goto exit because we must not cleanup()
4524 * because a daemon is already running.
4530 * Init UST app hash table. Alloc hash table before this point since
4531 * cleanup() can get called after that point.
4535 /* After this point, we can safely call cleanup() with "goto exit" */
4538 * These actions must be executed as root. We do that *after* setting up
4539 * the sockets path because we MUST make the check for another daemon using
4540 * those paths *before* trying to set the kernel consumer sockets and init
4544 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4549 /* Setup kernel tracer */
4550 if (!opt_no_kernel
) {
4551 init_kernel_tracer();
4554 /* Set ulimit for open files */
4557 /* init lttng_fd tracking must be done after set_ulimit. */
4560 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4565 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4570 if ((ret
= set_signal_handler()) < 0) {
4574 /* Setup the needed unix socket */
4575 if ((ret
= init_daemon_socket()) < 0) {
4579 /* Set credentials to socket */
4580 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4584 /* Get parent pid if -S, --sig-parent is specified. */
4585 if (opt_sig_parent
) {
4589 /* Setup the kernel pipe for waking up the kernel thread */
4590 if (is_root
&& !opt_no_kernel
) {
4591 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4596 /* Setup the thread ht_cleanup communication pipe. */
4597 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4601 /* Setup the thread apps communication pipe. */
4602 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4606 /* Setup the thread apps notify communication pipe. */
4607 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4611 /* Initialize global buffer per UID and PID registry. */
4612 buffer_reg_init_uid_registry();
4613 buffer_reg_init_pid_registry();
4615 /* Init UST command queue. */
4616 cds_wfq_init(&ust_cmd_queue
.queue
);
4619 * Get session list pointer. This pointer MUST NOT be free(). This list is
4620 * statically declared in session.c
4622 session_list_ptr
= session_get_list();
4624 /* Set up max poll set size */
4625 lttng_poll_set_max_size();
4629 /* Check for the application socket timeout env variable. */
4630 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4631 if (env_app_timeout
) {
4632 app_socket_timeout
= atoi(env_app_timeout
);
4634 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4639 /* Initialize communication library */
4642 /* Create thread to manage the client socket */
4643 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4644 thread_ht_cleanup
, (void *) NULL
);
4646 PERROR("pthread_create ht_cleanup");
4647 goto exit_ht_cleanup
;
4650 /* Create thread to manage the client socket */
4651 ret
= pthread_create(&health_thread
, NULL
,
4652 thread_manage_health
, (void *) NULL
);
4654 PERROR("pthread_create health");
4658 /* Create thread to manage the client socket */
4659 ret
= pthread_create(&client_thread
, NULL
,
4660 thread_manage_clients
, (void *) NULL
);
4662 PERROR("pthread_create clients");
4666 /* Create thread to dispatch registration */
4667 ret
= pthread_create(&dispatch_thread
, NULL
,
4668 thread_dispatch_ust_registration
, (void *) NULL
);
4670 PERROR("pthread_create dispatch");
4674 /* Create thread to manage application registration. */
4675 ret
= pthread_create(®_apps_thread
, NULL
,
4676 thread_registration_apps
, (void *) NULL
);
4678 PERROR("pthread_create registration");
4682 /* Create thread to manage application socket */
4683 ret
= pthread_create(&apps_thread
, NULL
,
4684 thread_manage_apps
, (void *) NULL
);
4686 PERROR("pthread_create apps");
4690 /* Create thread to manage application notify socket */
4691 ret
= pthread_create(&apps_notify_thread
, NULL
,
4692 ust_thread_manage_notify
, (void *) NULL
);
4694 PERROR("pthread_create apps");
4698 /* Don't start this thread if kernel tracing is not requested nor root */
4699 if (is_root
&& !opt_no_kernel
) {
4700 /* Create kernel thread to manage kernel event */
4701 ret
= pthread_create(&kernel_thread
, NULL
,
4702 thread_manage_kernel
, (void *) NULL
);
4704 PERROR("pthread_create kernel");
4708 ret
= pthread_join(kernel_thread
, &status
);
4710 PERROR("pthread_join");
4711 goto error
; /* join error, exit without cleanup */
4716 ret
= pthread_join(apps_thread
, &status
);
4718 PERROR("pthread_join");
4719 goto error
; /* join error, exit without cleanup */
4723 ret
= pthread_join(reg_apps_thread
, &status
);
4725 PERROR("pthread_join");
4726 goto error
; /* join error, exit without cleanup */
4730 ret
= pthread_join(dispatch_thread
, &status
);
4732 PERROR("pthread_join");
4733 goto error
; /* join error, exit without cleanup */
4737 ret
= pthread_join(client_thread
, &status
);
4739 PERROR("pthread_join");
4740 goto error
; /* join error, exit without cleanup */
4743 ret
= join_consumer_thread(&kconsumer_data
);
4745 PERROR("join_consumer");
4746 goto error
; /* join error, exit without cleanup */
4749 ret
= join_consumer_thread(&ustconsumer32_data
);
4751 PERROR("join_consumer ust32");
4752 goto error
; /* join error, exit without cleanup */
4755 ret
= join_consumer_thread(&ustconsumer64_data
);
4757 PERROR("join_consumer ust64");
4758 goto error
; /* join error, exit without cleanup */
4762 ret
= pthread_join(health_thread
, &status
);
4764 PERROR("pthread_join health thread");
4765 goto error
; /* join error, exit without cleanup */
4769 ret
= pthread_join(ht_cleanup_thread
, &status
);
4771 PERROR("pthread_join ht cleanup thread");
4772 goto error
; /* join error, exit without cleanup */
4777 * cleanup() is called when no other thread is running.
4779 rcu_thread_online();
4781 rcu_thread_offline();
4782 rcu_unregister_thread();