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"
63 #include "health-sessiond.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66 #include "jul-thread.h"
68 #define CONSUMERD_FILE "lttng-consumerd"
71 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
72 static const char *opt_pidfile
;
73 static int opt_sig_parent
;
74 static int opt_verbose_consumer
;
75 static int opt_daemon
;
76 static int opt_no_kernel
;
77 static int is_root
; /* Set to 1 if the daemon is running as root */
78 static pid_t ppid
; /* Parent PID for --sig-parent option */
82 * Consumer daemon specific control data. Every value not initialized here is
83 * set to 0 by the static definition.
85 static struct consumer_data kconsumer_data
= {
86 .type
= LTTNG_CONSUMER_KERNEL
,
87 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
88 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
91 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
92 .lock
= PTHREAD_MUTEX_INITIALIZER
,
93 .cond
= PTHREAD_COND_INITIALIZER
,
94 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
96 static struct consumer_data ustconsumer64_data
= {
97 .type
= LTTNG_CONSUMER64_UST
,
98 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
99 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
102 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
103 .lock
= PTHREAD_MUTEX_INITIALIZER
,
104 .cond
= PTHREAD_COND_INITIALIZER
,
105 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 static struct consumer_data ustconsumer32_data
= {
108 .type
= LTTNG_CONSUMER32_UST
,
109 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
110 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
113 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
114 .lock
= PTHREAD_MUTEX_INITIALIZER
,
115 .cond
= PTHREAD_COND_INITIALIZER
,
116 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 /* Shared between threads */
120 static int dispatch_thread_exit
;
122 /* Global application Unix socket path */
123 static char apps_unix_sock_path
[PATH_MAX
];
124 /* Global client Unix socket path */
125 static char client_unix_sock_path
[PATH_MAX
];
126 /* global wait shm path for UST */
127 static char wait_shm_path
[PATH_MAX
];
128 /* Global health check unix path */
129 static char health_unix_sock_path
[PATH_MAX
];
131 /* Sockets and FDs */
132 static int client_sock
= -1;
133 static int apps_sock
= -1;
134 int kernel_tracer_fd
= -1;
135 static int kernel_poll_pipe
[2] = { -1, -1 };
138 * Quit pipe for all threads. This permits a single cancellation point
139 * for all threads when receiving an event on the pipe.
141 static int thread_quit_pipe
[2] = { -1, -1 };
144 * This pipe is used to inform the thread managing application communication
145 * that a command is queued and ready to be processed.
147 static int apps_cmd_pipe
[2] = { -1, -1 };
149 int apps_cmd_notify_pipe
[2] = { -1, -1 };
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread
;
153 static pthread_t apps_notify_thread
;
154 static pthread_t reg_apps_thread
;
155 static pthread_t client_thread
;
156 static pthread_t kernel_thread
;
157 static pthread_t dispatch_thread
;
158 static pthread_t health_thread
;
159 static pthread_t ht_cleanup_thread
;
160 static pthread_t jul_reg_thread
;
163 * UST registration command queue. This queue is tied with a futex and uses a N
164 * wakers / 1 waiter implemented and detailed in futex.c/.h
166 * The thread_manage_apps and thread_dispatch_ust_registration interact with
167 * this queue and the wait/wake scheme.
169 static struct ust_cmd_queue ust_cmd_queue
;
172 * Pointer initialized before thread creation.
174 * This points to the tracing session list containing the session count and a
175 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
176 * MUST NOT be taken if you call a public function in session.c.
178 * The lock is nested inside the structure: session_list_ptr->lock. Please use
179 * session_lock_list and session_unlock_list for lock acquisition.
181 static struct ltt_session_list
*session_list_ptr
;
183 int ust_consumerd64_fd
= -1;
184 int ust_consumerd32_fd
= -1;
186 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
187 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
188 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
189 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
191 static const char *module_proc_lttng
= "/proc/lttng";
194 * Consumer daemon state which is changed when spawning it, killing it or in
195 * case of a fatal error.
197 enum consumerd_state
{
198 CONSUMER_STARTED
= 1,
199 CONSUMER_STOPPED
= 2,
204 * This consumer daemon state is used to validate if a client command will be
205 * able to reach the consumer. If not, the client is informed. For instance,
206 * doing a "lttng start" when the consumer state is set to ERROR will return an
207 * error to the client.
209 * The following example shows a possible race condition of this scheme:
211 * consumer thread error happens
213 * client cmd checks state -> still OK
214 * consumer thread exit, sets error
215 * client cmd try to talk to consumer
218 * However, since the consumer is a different daemon, we have no way of making
219 * sure the command will reach it safely even with this state flag. This is why
220 * we consider that up to the state validation during command processing, the
221 * command is safe. After that, we can not guarantee the correctness of the
222 * client request vis-a-vis the consumer.
224 static enum consumerd_state ust_consumerd_state
;
225 static enum consumerd_state kernel_consumerd_state
;
228 * Socket timeout for receiving and sending in seconds.
230 static int app_socket_timeout
;
232 /* Set in main() with the current page size. */
235 /* Application health monitoring */
236 struct health_app
*health_sessiond
;
238 /* JUL TCP port for registration. Used by the JUL thread. */
239 unsigned int jul_tcp_port
= DEFAULT_JUL_TCP_PORT
;
242 void setup_consumerd_path(void)
244 const char *bin
, *libdir
;
247 * Allow INSTALL_BIN_PATH to be used as a target path for the
248 * native architecture size consumer if CONFIG_CONSUMER*_PATH
249 * has not been defined.
251 #if (CAA_BITS_PER_LONG == 32)
252 if (!consumerd32_bin
[0]) {
253 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
255 if (!consumerd32_libdir
[0]) {
256 consumerd32_libdir
= INSTALL_LIB_PATH
;
258 #elif (CAA_BITS_PER_LONG == 64)
259 if (!consumerd64_bin
[0]) {
260 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
262 if (!consumerd64_libdir
[0]) {
263 consumerd64_libdir
= INSTALL_LIB_PATH
;
266 #error "Unknown bitness"
270 * runtime env. var. overrides the build default.
272 bin
= getenv("LTTNG_CONSUMERD32_BIN");
274 consumerd32_bin
= bin
;
276 bin
= getenv("LTTNG_CONSUMERD64_BIN");
278 consumerd64_bin
= bin
;
280 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
282 consumerd32_libdir
= libdir
;
284 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
286 consumerd64_libdir
= libdir
;
291 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
293 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
299 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
305 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
317 * Check if the thread quit pipe was triggered.
319 * Return 1 if it was triggered else 0;
321 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
323 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
331 * Init thread quit pipe.
333 * Return -1 on error or 0 if all pipes are created.
335 static int init_thread_quit_pipe(void)
339 ret
= pipe(thread_quit_pipe
);
341 PERROR("thread quit pipe");
345 for (i
= 0; i
< 2; i
++) {
346 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
358 * Stop all threads by closing the thread quit pipe.
360 static void stop_threads(void)
364 /* Stopping all threads */
365 DBG("Terminating all threads");
366 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
368 ERR("write error on thread quit pipe");
371 /* Dispatch thread */
372 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
373 futex_nto1_wake(&ust_cmd_queue
.futex
);
377 * Close every consumer sockets.
379 static void close_consumer_sockets(void)
383 if (kconsumer_data
.err_sock
>= 0) {
384 ret
= close(kconsumer_data
.err_sock
);
386 PERROR("kernel consumer err_sock close");
389 if (ustconsumer32_data
.err_sock
>= 0) {
390 ret
= close(ustconsumer32_data
.err_sock
);
392 PERROR("UST consumerd32 err_sock close");
395 if (ustconsumer64_data
.err_sock
>= 0) {
396 ret
= close(ustconsumer64_data
.err_sock
);
398 PERROR("UST consumerd64 err_sock close");
401 if (kconsumer_data
.cmd_sock
>= 0) {
402 ret
= close(kconsumer_data
.cmd_sock
);
404 PERROR("kernel consumer cmd_sock close");
407 if (ustconsumer32_data
.cmd_sock
>= 0) {
408 ret
= close(ustconsumer32_data
.cmd_sock
);
410 PERROR("UST consumerd32 cmd_sock close");
413 if (ustconsumer64_data
.cmd_sock
>= 0) {
414 ret
= close(ustconsumer64_data
.cmd_sock
);
416 PERROR("UST consumerd64 cmd_sock close");
424 static void cleanup(void)
427 struct ltt_session
*sess
, *stmp
;
433 * Close the thread quit pipe. It has already done its job,
434 * since we are now called.
436 utils_close_pipe(thread_quit_pipe
);
439 * If opt_pidfile is undefined, the default file will be wiped when
440 * removing the rundir.
443 ret
= remove(opt_pidfile
);
445 PERROR("remove pidfile %s", opt_pidfile
);
449 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
452 snprintf(path
, PATH_MAX
,
454 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
455 DBG("Removing %s", path
);
458 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
459 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
);
460 DBG("Removing %s", path
);
464 snprintf(path
, PATH_MAX
,
465 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
467 DBG("Removing %s", path
);
470 snprintf(path
, PATH_MAX
,
471 DEFAULT_KCONSUMERD_PATH
,
473 DBG("Removing directory %s", path
);
476 /* ust consumerd 32 */
477 snprintf(path
, PATH_MAX
,
478 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
480 DBG("Removing %s", path
);
483 snprintf(path
, PATH_MAX
,
484 DEFAULT_USTCONSUMERD32_PATH
,
486 DBG("Removing directory %s", path
);
489 /* ust consumerd 64 */
490 snprintf(path
, PATH_MAX
,
491 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
493 DBG("Removing %s", path
);
496 snprintf(path
, PATH_MAX
,
497 DEFAULT_USTCONSUMERD64_PATH
,
499 DBG("Removing directory %s", path
);
502 (void) rmdir(rundir
);
505 DBG("Cleaning up all sessions");
507 /* Destroy session list mutex */
508 if (session_list_ptr
!= NULL
) {
509 pthread_mutex_destroy(&session_list_ptr
->lock
);
511 /* Cleanup ALL session */
512 cds_list_for_each_entry_safe(sess
, stmp
,
513 &session_list_ptr
->head
, list
) {
514 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
518 DBG("Closing all UST sockets");
519 ust_app_clean_list();
520 buffer_reg_destroy_registries();
522 if (is_root
&& !opt_no_kernel
) {
523 DBG2("Closing kernel fd");
524 if (kernel_tracer_fd
>= 0) {
525 ret
= close(kernel_tracer_fd
);
530 DBG("Unloading kernel modules");
531 modprobe_remove_lttng_all();
534 close_consumer_sockets();
537 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
538 "Matthew, BEET driven development works!%c[%dm",
539 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
544 * Send data on a unix socket using the liblttsessiondcomm API.
546 * Return lttcomm error code.
548 static int send_unix_sock(int sock
, void *buf
, size_t len
)
550 /* Check valid length */
555 return lttcomm_send_unix_sock(sock
, buf
, len
);
559 * Free memory of a command context structure.
561 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
563 DBG("Clean command context structure");
565 if ((*cmd_ctx
)->llm
) {
566 free((*cmd_ctx
)->llm
);
568 if ((*cmd_ctx
)->lsm
) {
569 free((*cmd_ctx
)->lsm
);
577 * Notify UST applications using the shm mmap futex.
579 static int notify_ust_apps(int active
)
583 DBG("Notifying applications of session daemon state: %d", active
);
585 /* See shm.c for this call implying mmap, shm and futex calls */
586 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
587 if (wait_shm_mmap
== NULL
) {
591 /* Wake waiting process */
592 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
594 /* Apps notified successfully */
602 * Setup the outgoing data buffer for the response (llm) by allocating the
603 * right amount of memory and copying the original information from the lsm
606 * Return total size of the buffer pointed by buf.
608 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
614 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
615 if (cmd_ctx
->llm
== NULL
) {
621 /* Copy common data */
622 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
623 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
625 cmd_ctx
->llm
->data_size
= size
;
626 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
635 * Update the kernel poll set of all channel fd available over all tracing
636 * session. Add the wakeup pipe at the end of the set.
638 static int update_kernel_poll(struct lttng_poll_event
*events
)
641 struct ltt_session
*session
;
642 struct ltt_kernel_channel
*channel
;
644 DBG("Updating kernel poll set");
647 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
648 session_lock(session
);
649 if (session
->kernel_session
== NULL
) {
650 session_unlock(session
);
654 cds_list_for_each_entry(channel
,
655 &session
->kernel_session
->channel_list
.head
, list
) {
656 /* Add channel fd to the kernel poll set */
657 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
659 session_unlock(session
);
662 DBG("Channel fd %d added to kernel set", channel
->fd
);
664 session_unlock(session
);
666 session_unlock_list();
671 session_unlock_list();
676 * Find the channel fd from 'fd' over all tracing session. When found, check
677 * for new channel stream and send those stream fds to the kernel consumer.
679 * Useful for CPU hotplug feature.
681 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
684 struct ltt_session
*session
;
685 struct ltt_kernel_session
*ksess
;
686 struct ltt_kernel_channel
*channel
;
688 DBG("Updating kernel streams for channel fd %d", fd
);
691 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
692 session_lock(session
);
693 if (session
->kernel_session
== NULL
) {
694 session_unlock(session
);
697 ksess
= session
->kernel_session
;
699 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
700 if (channel
->fd
== fd
) {
701 DBG("Channel found, updating kernel streams");
702 ret
= kernel_open_channel_stream(channel
);
706 /* Update the stream global counter */
707 ksess
->stream_count_global
+= ret
;
710 * Have we already sent fds to the consumer? If yes, it means
711 * that tracing is started so it is safe to send our updated
714 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
715 struct lttng_ht_iter iter
;
716 struct consumer_socket
*socket
;
719 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
720 &iter
.iter
, socket
, node
.node
) {
721 pthread_mutex_lock(socket
->lock
);
722 ret
= kernel_consumer_send_channel_stream(socket
,
724 session
->output_traces
? 1 : 0);
725 pthread_mutex_unlock(socket
->lock
);
736 session_unlock(session
);
738 session_unlock_list();
742 session_unlock(session
);
743 session_unlock_list();
748 * For each tracing session, update newly registered apps. The session list
749 * lock MUST be acquired before calling this.
751 static void update_ust_app(int app_sock
)
753 struct ltt_session
*sess
, *stmp
;
755 /* Consumer is in an ERROR state. Stop any application update. */
756 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
757 /* Stop the update process since the consumer is dead. */
761 /* For all tracing session(s) */
762 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
764 if (sess
->ust_session
) {
765 ust_app_global_update(sess
->ust_session
, app_sock
);
767 session_unlock(sess
);
772 * This thread manage event coming from the kernel.
774 * Features supported in this thread:
777 static void *thread_manage_kernel(void *data
)
779 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
780 uint32_t revents
, nb_fd
;
782 struct lttng_poll_event events
;
784 DBG("[thread] Thread manage kernel started");
786 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
789 * This first step of the while is to clean this structure which could free
790 * non NULL pointers so initialize it before the loop.
792 lttng_poll_init(&events
);
794 if (testpoint(thread_manage_kernel
)) {
795 goto error_testpoint
;
798 health_code_update();
800 if (testpoint(thread_manage_kernel_before_loop
)) {
801 goto error_testpoint
;
805 health_code_update();
807 if (update_poll_flag
== 1) {
808 /* Clean events object. We are about to populate it again. */
809 lttng_poll_clean(&events
);
811 ret
= sessiond_set_thread_pollset(&events
, 2);
813 goto error_poll_create
;
816 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
821 /* This will add the available kernel channel if any. */
822 ret
= update_kernel_poll(&events
);
826 update_poll_flag
= 0;
829 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
831 /* Poll infinite value of time */
834 ret
= lttng_poll_wait(&events
, -1);
838 * Restart interrupted system call.
840 if (errno
== EINTR
) {
844 } else if (ret
== 0) {
845 /* Should not happen since timeout is infinite */
846 ERR("Return value of poll is 0 with an infinite timeout.\n"
847 "This should not have happened! Continuing...");
853 for (i
= 0; i
< nb_fd
; i
++) {
854 /* Fetch once the poll data */
855 revents
= LTTNG_POLL_GETEV(&events
, i
);
856 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
858 health_code_update();
860 /* Thread quit pipe has been closed. Killing thread. */
861 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
867 /* Check for data on kernel pipe */
868 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
870 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
871 } while (ret
< 0 && errno
== EINTR
);
873 * Ret value is useless here, if this pipe gets any actions an
874 * update is required anyway.
876 update_poll_flag
= 1;
880 * New CPU detected by the kernel. Adding kernel stream to
881 * kernel session and updating the kernel consumer
883 if (revents
& LPOLLIN
) {
884 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
890 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
891 * and unregister kernel stream at this point.
900 lttng_poll_clean(&events
);
903 utils_close_pipe(kernel_poll_pipe
);
904 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
907 ERR("Health error occurred in %s", __func__
);
908 WARN("Kernel thread died unexpectedly. "
909 "Kernel tracing can continue but CPU hotplug is disabled.");
911 health_unregister(health_sessiond
);
912 DBG("Kernel thread dying");
917 * Signal pthread condition of the consumer data that the thread.
919 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
921 pthread_mutex_lock(&data
->cond_mutex
);
924 * The state is set before signaling. It can be any value, it's the waiter
925 * job to correctly interpret this condition variable associated to the
926 * consumer pthread_cond.
928 * A value of 0 means that the corresponding thread of the consumer data
929 * was not started. 1 indicates that the thread has started and is ready
930 * for action. A negative value means that there was an error during the
933 data
->consumer_thread_is_ready
= state
;
934 (void) pthread_cond_signal(&data
->cond
);
936 pthread_mutex_unlock(&data
->cond_mutex
);
940 * This thread manage the consumer error sent back to the session daemon.
942 static void *thread_manage_consumer(void *data
)
944 int sock
= -1, i
, ret
, pollfd
, err
= -1;
945 uint32_t revents
, nb_fd
;
946 enum lttcomm_return_code code
;
947 struct lttng_poll_event events
;
948 struct consumer_data
*consumer_data
= data
;
950 DBG("[thread] Manage consumer started");
952 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
954 health_code_update();
957 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
958 * metadata_sock. Nothing more will be added to this poll set.
960 ret
= sessiond_set_thread_pollset(&events
, 3);
966 * The error socket here is already in a listening state which was done
967 * just before spawning this thread to avoid a race between the consumer
968 * daemon exec trying to connect and the listen() call.
970 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
975 health_code_update();
977 /* Infinite blocking call, waiting for transmission */
981 if (testpoint(thread_manage_consumer
)) {
985 ret
= lttng_poll_wait(&events
, -1);
989 * Restart interrupted system call.
991 if (errno
== EINTR
) {
999 for (i
= 0; i
< nb_fd
; i
++) {
1000 /* Fetch once the poll data */
1001 revents
= LTTNG_POLL_GETEV(&events
, i
);
1002 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1004 health_code_update();
1006 /* Thread quit pipe has been closed. Killing thread. */
1007 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1013 /* Event on the registration socket */
1014 if (pollfd
== consumer_data
->err_sock
) {
1015 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1016 ERR("consumer err socket poll error");
1022 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1028 * Set the CLOEXEC flag. Return code is useless because either way, the
1031 (void) utils_set_fd_cloexec(sock
);
1033 health_code_update();
1035 DBG2("Receiving code from consumer err_sock");
1037 /* Getting status code from kconsumerd */
1038 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1039 sizeof(enum lttcomm_return_code
));
1044 health_code_update();
1046 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1047 /* Connect both socket, command and metadata. */
1048 consumer_data
->cmd_sock
=
1049 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1050 consumer_data
->metadata_fd
=
1051 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1052 if (consumer_data
->cmd_sock
< 0
1053 || consumer_data
->metadata_fd
< 0) {
1054 PERROR("consumer connect cmd socket");
1055 /* On error, signal condition and quit. */
1056 signal_consumer_condition(consumer_data
, -1);
1059 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1060 /* Create metadata socket lock. */
1061 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1062 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1063 PERROR("zmalloc pthread mutex");
1067 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1069 signal_consumer_condition(consumer_data
, 1);
1070 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1071 DBG("Consumer metadata socket ready (fd: %d)",
1072 consumer_data
->metadata_fd
);
1074 ERR("consumer error when waiting for SOCK_READY : %s",
1075 lttcomm_get_readable_code(-code
));
1079 /* Remove the consumerd error sock since we've established a connexion */
1080 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1085 /* Add new accepted error socket. */
1086 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1091 /* Add metadata socket that is successfully connected. */
1092 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1093 LPOLLIN
| LPOLLRDHUP
);
1098 health_code_update();
1100 /* Infinite blocking call, waiting for transmission */
1103 health_poll_entry();
1104 ret
= lttng_poll_wait(&events
, -1);
1108 * Restart interrupted system call.
1110 if (errno
== EINTR
) {
1118 for (i
= 0; i
< nb_fd
; i
++) {
1119 /* Fetch once the poll data */
1120 revents
= LTTNG_POLL_GETEV(&events
, i
);
1121 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1123 health_code_update();
1125 /* Thread quit pipe has been closed. Killing thread. */
1126 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1132 if (pollfd
== sock
) {
1133 /* Event on the consumerd socket */
1134 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1135 ERR("consumer err socket second poll error");
1138 health_code_update();
1139 /* Wait for any kconsumerd error */
1140 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1141 sizeof(enum lttcomm_return_code
));
1143 ERR("consumer closed the command socket");
1147 ERR("consumer return code : %s",
1148 lttcomm_get_readable_code(-code
));
1151 } else if (pollfd
== consumer_data
->metadata_fd
) {
1152 /* UST metadata requests */
1153 ret
= ust_consumer_metadata_request(
1154 &consumer_data
->metadata_sock
);
1156 ERR("Handling metadata request");
1161 ERR("Unknown pollfd");
1165 health_code_update();
1171 * We lock here because we are about to close the sockets and some other
1172 * thread might be using them so get exclusive access which will abort all
1173 * other consumer command by other threads.
1175 pthread_mutex_lock(&consumer_data
->lock
);
1177 /* Immediately set the consumerd state to stopped */
1178 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1179 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1180 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1181 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1182 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1184 /* Code flow error... */
1188 if (consumer_data
->err_sock
>= 0) {
1189 ret
= close(consumer_data
->err_sock
);
1193 consumer_data
->err_sock
= -1;
1195 if (consumer_data
->cmd_sock
>= 0) {
1196 ret
= close(consumer_data
->cmd_sock
);
1200 consumer_data
->cmd_sock
= -1;
1202 if (*consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1203 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1216 unlink(consumer_data
->err_unix_sock_path
);
1217 unlink(consumer_data
->cmd_unix_sock_path
);
1218 consumer_data
->pid
= 0;
1219 pthread_mutex_unlock(&consumer_data
->lock
);
1221 /* Cleanup metadata socket mutex. */
1222 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1223 free(consumer_data
->metadata_sock
.lock
);
1225 lttng_poll_clean(&events
);
1229 ERR("Health error occurred in %s", __func__
);
1231 health_unregister(health_sessiond
);
1232 DBG("consumer thread cleanup completed");
1238 * This thread manage application communication.
1240 static void *thread_manage_apps(void *data
)
1242 int i
, ret
, pollfd
, err
= -1;
1243 uint32_t revents
, nb_fd
;
1244 struct lttng_poll_event events
;
1246 DBG("[thread] Manage application started");
1248 rcu_register_thread();
1249 rcu_thread_online();
1251 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1253 if (testpoint(thread_manage_apps
)) {
1254 goto error_testpoint
;
1257 health_code_update();
1259 ret
= sessiond_set_thread_pollset(&events
, 2);
1261 goto error_poll_create
;
1264 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1269 if (testpoint(thread_manage_apps_before_loop
)) {
1273 health_code_update();
1276 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1278 /* Inifinite blocking call, waiting for transmission */
1280 health_poll_entry();
1281 ret
= lttng_poll_wait(&events
, -1);
1285 * Restart interrupted system call.
1287 if (errno
== EINTR
) {
1295 for (i
= 0; i
< nb_fd
; i
++) {
1296 /* Fetch once the poll data */
1297 revents
= LTTNG_POLL_GETEV(&events
, i
);
1298 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1300 health_code_update();
1302 /* Thread quit pipe has been closed. Killing thread. */
1303 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1309 /* Inspect the apps cmd pipe */
1310 if (pollfd
== apps_cmd_pipe
[0]) {
1311 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1312 ERR("Apps command pipe error");
1314 } else if (revents
& LPOLLIN
) {
1319 ret
= read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1320 } while (ret
< 0 && errno
== EINTR
);
1321 if (ret
< 0 || ret
< sizeof(sock
)) {
1322 PERROR("read apps cmd pipe");
1326 health_code_update();
1329 * We only monitor the error events of the socket. This
1330 * thread does not handle any incoming data from UST
1333 ret
= lttng_poll_add(&events
, sock
,
1334 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1339 DBG("Apps with sock %d added to poll set", sock
);
1341 health_code_update();
1347 * At this point, we know that a registered application made
1348 * the event at poll_wait.
1350 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1351 /* Removing from the poll set */
1352 ret
= lttng_poll_del(&events
, pollfd
);
1357 /* Socket closed on remote end. */
1358 ust_app_unregister(pollfd
);
1363 health_code_update();
1369 lttng_poll_clean(&events
);
1372 utils_close_pipe(apps_cmd_pipe
);
1373 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1376 * We don't clean the UST app hash table here since already registered
1377 * applications can still be controlled so let them be until the session
1378 * daemon dies or the applications stop.
1383 ERR("Health error occurred in %s", __func__
);
1385 health_unregister(health_sessiond
);
1386 DBG("Application communication apps thread cleanup complete");
1387 rcu_thread_offline();
1388 rcu_unregister_thread();
1393 * Send a socket to a thread This is called from the dispatch UST registration
1394 * thread once all sockets are set for the application.
1396 * The sock value can be invalid, we don't really care, the thread will handle
1397 * it and make the necessary cleanup if so.
1399 * On success, return 0 else a negative value being the errno message of the
1402 static int send_socket_to_thread(int fd
, int sock
)
1407 * It's possible that the FD is set as invalid with -1 concurrently just
1408 * before calling this function being a shutdown state of the thread.
1416 ret
= write(fd
, &sock
, sizeof(sock
));
1417 } while (ret
< 0 && errno
== EINTR
);
1418 if (ret
< 0 || ret
!= sizeof(sock
)) {
1419 PERROR("write apps pipe %d", fd
);
1426 /* All good. Don't send back the write positive ret value. */
1433 * Sanitize the wait queue of the dispatch registration thread meaning removing
1434 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1435 * notify socket is never received.
1437 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1439 int ret
, nb_fd
= 0, i
;
1440 unsigned int fd_added
= 0;
1441 struct lttng_poll_event events
;
1442 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1446 lttng_poll_init(&events
);
1448 /* Just skip everything for an empty queue. */
1449 if (!wait_queue
->count
) {
1453 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1458 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1459 &wait_queue
->head
, head
) {
1460 assert(wait_node
->app
);
1461 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1462 LPOLLHUP
| LPOLLERR
);
1475 * Poll but don't block so we can quickly identify the faulty events and
1476 * clean them afterwards from the wait queue.
1478 ret
= lttng_poll_wait(&events
, 0);
1484 for (i
= 0; i
< nb_fd
; i
++) {
1485 /* Get faulty FD. */
1486 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1487 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1489 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1490 &wait_queue
->head
, head
) {
1491 if (pollfd
== wait_node
->app
->sock
&&
1492 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1493 cds_list_del(&wait_node
->head
);
1494 wait_queue
->count
--;
1495 ust_app_destroy(wait_node
->app
);
1503 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1507 lttng_poll_clean(&events
);
1511 lttng_poll_clean(&events
);
1513 ERR("Unable to sanitize wait queue");
1518 * Dispatch request from the registration threads to the application
1519 * communication thread.
1521 static void *thread_dispatch_ust_registration(void *data
)
1524 struct cds_wfq_node
*node
;
1525 struct ust_command
*ust_cmd
= NULL
;
1526 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1527 struct ust_reg_wait_queue wait_queue
= {
1531 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1533 health_code_update();
1535 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1537 DBG("[thread] Dispatch UST command started");
1539 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1540 health_code_update();
1542 /* Atomically prepare the queue futex */
1543 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1546 struct ust_app
*app
= NULL
;
1550 * Make sure we don't have node(s) that have hung up before receiving
1551 * the notify socket. This is to clean the list in order to avoid
1552 * memory leaks from notify socket that are never seen.
1554 sanitize_wait_queue(&wait_queue
);
1556 health_code_update();
1557 /* Dequeue command for registration */
1558 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1560 DBG("Woken up but nothing in the UST command queue");
1561 /* Continue thread execution */
1565 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1567 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1568 " gid:%d sock:%d name:%s (version %d.%d)",
1569 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1570 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1571 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1572 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1574 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1575 wait_node
= zmalloc(sizeof(*wait_node
));
1577 PERROR("zmalloc wait_node dispatch");
1578 ret
= close(ust_cmd
->sock
);
1580 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1582 lttng_fd_put(LTTNG_FD_APPS
, 1);
1586 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1588 /* Create application object if socket is CMD. */
1589 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1591 if (!wait_node
->app
) {
1592 ret
= close(ust_cmd
->sock
);
1594 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1596 lttng_fd_put(LTTNG_FD_APPS
, 1);
1602 * Add application to the wait queue so we can set the notify
1603 * socket before putting this object in the global ht.
1605 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1610 * We have to continue here since we don't have the notify
1611 * socket and the application MUST be added to the hash table
1612 * only at that moment.
1617 * Look for the application in the local wait queue and set the
1618 * notify socket if found.
1620 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1621 &wait_queue
.head
, head
) {
1622 health_code_update();
1623 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1624 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1625 cds_list_del(&wait_node
->head
);
1627 app
= wait_node
->app
;
1629 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1635 * With no application at this stage the received socket is
1636 * basically useless so close it before we free the cmd data
1637 * structure for good.
1640 ret
= close(ust_cmd
->sock
);
1642 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1644 lttng_fd_put(LTTNG_FD_APPS
, 1);
1651 * @session_lock_list
1653 * Lock the global session list so from the register up to the
1654 * registration done message, no thread can see the application
1655 * and change its state.
1657 session_lock_list();
1661 * Add application to the global hash table. This needs to be
1662 * done before the update to the UST registry can locate the
1667 /* Set app version. This call will print an error if needed. */
1668 (void) ust_app_version(app
);
1670 /* Send notify socket through the notify pipe. */
1671 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1675 session_unlock_list();
1677 * No notify thread, stop the UST tracing. However, this is
1678 * not an internal error of the this thread thus setting
1679 * the health error code to a normal exit.
1686 * Update newly registered application with the tracing
1687 * registry info already enabled information.
1689 update_ust_app(app
->sock
);
1692 * Don't care about return value. Let the manage apps threads
1693 * handle app unregistration upon socket close.
1695 (void) ust_app_register_done(app
->sock
);
1698 * Even if the application socket has been closed, send the app
1699 * to the thread and unregistration will take place at that
1702 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1705 session_unlock_list();
1707 * No apps. thread, stop the UST tracing. However, this is
1708 * not an internal error of the this thread thus setting
1709 * the health error code to a normal exit.
1716 session_unlock_list();
1718 } while (node
!= NULL
);
1720 health_poll_entry();
1721 /* Futex wait on queue. Blocking call on futex() */
1722 futex_nto1_wait(&ust_cmd_queue
.futex
);
1725 /* Normal exit, no error */
1729 /* Clean up wait queue. */
1730 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1731 &wait_queue
.head
, head
) {
1732 cds_list_del(&wait_node
->head
);
1737 DBG("Dispatch thread dying");
1740 ERR("Health error occurred in %s", __func__
);
1742 health_unregister(health_sessiond
);
1747 * This thread manage application registration.
1749 static void *thread_registration_apps(void *data
)
1751 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1752 uint32_t revents
, nb_fd
;
1753 struct lttng_poll_event events
;
1755 * Get allocated in this thread, enqueued to a global queue, dequeued and
1756 * freed in the manage apps thread.
1758 struct ust_command
*ust_cmd
= NULL
;
1760 DBG("[thread] Manage application registration started");
1762 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1764 if (testpoint(thread_registration_apps
)) {
1765 goto error_testpoint
;
1768 ret
= lttcomm_listen_unix_sock(apps_sock
);
1774 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1775 * more will be added to this poll set.
1777 ret
= sessiond_set_thread_pollset(&events
, 2);
1779 goto error_create_poll
;
1782 /* Add the application registration socket */
1783 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1785 goto error_poll_add
;
1788 /* Notify all applications to register */
1789 ret
= notify_ust_apps(1);
1791 ERR("Failed to notify applications or create the wait shared memory.\n"
1792 "Execution continues but there might be problem for already\n"
1793 "running applications that wishes to register.");
1797 DBG("Accepting application registration");
1799 /* Inifinite blocking call, waiting for transmission */
1801 health_poll_entry();
1802 ret
= lttng_poll_wait(&events
, -1);
1806 * Restart interrupted system call.
1808 if (errno
== EINTR
) {
1816 for (i
= 0; i
< nb_fd
; i
++) {
1817 health_code_update();
1819 /* Fetch once the poll data */
1820 revents
= LTTNG_POLL_GETEV(&events
, i
);
1821 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1823 /* Thread quit pipe has been closed. Killing thread. */
1824 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1830 /* Event on the registration socket */
1831 if (pollfd
== apps_sock
) {
1832 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1833 ERR("Register apps socket poll error");
1835 } else if (revents
& LPOLLIN
) {
1836 sock
= lttcomm_accept_unix_sock(apps_sock
);
1842 * Set socket timeout for both receiving and ending.
1843 * app_socket_timeout is in seconds, whereas
1844 * lttcomm_setsockopt_rcv_timeout and
1845 * lttcomm_setsockopt_snd_timeout expect msec as
1848 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1849 app_socket_timeout
* 1000);
1850 (void) lttcomm_setsockopt_snd_timeout(sock
,
1851 app_socket_timeout
* 1000);
1854 * Set the CLOEXEC flag. Return code is useless because
1855 * either way, the show must go on.
1857 (void) utils_set_fd_cloexec(sock
);
1859 /* Create UST registration command for enqueuing */
1860 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1861 if (ust_cmd
== NULL
) {
1862 PERROR("ust command zmalloc");
1867 * Using message-based transmissions to ensure we don't
1868 * have to deal with partially received messages.
1870 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1872 ERR("Exhausted file descriptors allowed for applications.");
1882 health_code_update();
1883 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1886 /* Close socket of the application. */
1891 lttng_fd_put(LTTNG_FD_APPS
, 1);
1895 health_code_update();
1897 ust_cmd
->sock
= sock
;
1900 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1901 " gid:%d sock:%d name:%s (version %d.%d)",
1902 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1903 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1904 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1905 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1908 * Lock free enqueue the registration request. The red pill
1909 * has been taken! This apps will be part of the *system*.
1911 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1914 * Wake the registration queue futex. Implicit memory
1915 * barrier with the exchange in cds_wfq_enqueue.
1917 futex_nto1_wake(&ust_cmd_queue
.futex
);
1927 ERR("Health error occurred in %s", __func__
);
1930 /* Notify that the registration thread is gone */
1933 if (apps_sock
>= 0) {
1934 ret
= close(apps_sock
);
1944 lttng_fd_put(LTTNG_FD_APPS
, 1);
1946 unlink(apps_unix_sock_path
);
1949 lttng_poll_clean(&events
);
1953 DBG("UST Registration thread cleanup complete");
1954 health_unregister(health_sessiond
);
1960 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1961 * exec or it will fails.
1963 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1966 struct timespec timeout
;
1968 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1969 consumer_data
->consumer_thread_is_ready
= 0;
1971 /* Setup pthread condition */
1972 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1975 PERROR("pthread_condattr_init consumer data");
1980 * Set the monotonic clock in order to make sure we DO NOT jump in time
1981 * between the clock_gettime() call and the timedwait call. See bug #324
1982 * for a more details and how we noticed it.
1984 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1987 PERROR("pthread_condattr_setclock consumer data");
1991 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1994 PERROR("pthread_cond_init consumer data");
1998 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2001 PERROR("pthread_create consumer");
2006 /* We are about to wait on a pthread condition */
2007 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2009 /* Get time for sem_timedwait absolute timeout */
2010 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2012 * Set the timeout for the condition timed wait even if the clock gettime
2013 * call fails since we might loop on that call and we want to avoid to
2014 * increment the timeout too many times.
2016 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2019 * The following loop COULD be skipped in some conditions so this is why we
2020 * set ret to 0 in order to make sure at least one round of the loop is
2026 * Loop until the condition is reached or when a timeout is reached. Note
2027 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2028 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2029 * possible. This loop does not take any chances and works with both of
2032 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2033 if (clock_ret
< 0) {
2034 PERROR("clock_gettime spawn consumer");
2035 /* Infinite wait for the consumerd thread to be ready */
2036 ret
= pthread_cond_wait(&consumer_data
->cond
,
2037 &consumer_data
->cond_mutex
);
2039 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2040 &consumer_data
->cond_mutex
, &timeout
);
2044 /* Release the pthread condition */
2045 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2049 if (ret
== ETIMEDOUT
) {
2051 * Call has timed out so we kill the kconsumerd_thread and return
2054 ERR("Condition timed out. The consumer thread was never ready."
2056 ret
= pthread_cancel(consumer_data
->thread
);
2058 PERROR("pthread_cancel consumer thread");
2061 PERROR("pthread_cond_wait failed consumer thread");
2066 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2067 if (consumer_data
->pid
== 0) {
2068 ERR("Consumerd did not start");
2069 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2072 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2081 * Join consumer thread
2083 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2087 /* Consumer pid must be a real one. */
2088 if (consumer_data
->pid
> 0) {
2090 ret
= kill(consumer_data
->pid
, SIGTERM
);
2092 ERR("Error killing consumer daemon");
2095 return pthread_join(consumer_data
->thread
, &status
);
2102 * Fork and exec a consumer daemon (consumerd).
2104 * Return pid if successful else -1.
2106 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2110 const char *consumer_to_use
;
2111 const char *verbosity
;
2114 DBG("Spawning consumerd");
2121 if (opt_verbose_consumer
) {
2122 verbosity
= "--verbose";
2124 verbosity
= "--quiet";
2126 switch (consumer_data
->type
) {
2127 case LTTNG_CONSUMER_KERNEL
:
2129 * Find out which consumerd to execute. We will first try the
2130 * 64-bit path, then the sessiond's installation directory, and
2131 * fallback on the 32-bit one,
2133 DBG3("Looking for a kernel consumer at these locations:");
2134 DBG3(" 1) %s", consumerd64_bin
);
2135 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2136 DBG3(" 3) %s", consumerd32_bin
);
2137 if (stat(consumerd64_bin
, &st
) == 0) {
2138 DBG3("Found location #1");
2139 consumer_to_use
= consumerd64_bin
;
2140 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2141 DBG3("Found location #2");
2142 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2143 } else if (stat(consumerd32_bin
, &st
) == 0) {
2144 DBG3("Found location #3");
2145 consumer_to_use
= consumerd32_bin
;
2147 DBG("Could not find any valid consumerd executable");
2150 DBG("Using kernel consumer at: %s", consumer_to_use
);
2151 execl(consumer_to_use
,
2152 "lttng-consumerd", verbosity
, "-k",
2153 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2154 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2155 "--group", tracing_group_name
,
2158 case LTTNG_CONSUMER64_UST
:
2160 char *tmpnew
= NULL
;
2162 if (consumerd64_libdir
[0] != '\0') {
2166 tmp
= getenv("LD_LIBRARY_PATH");
2170 tmplen
= strlen("LD_LIBRARY_PATH=")
2171 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2172 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2177 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2178 strcat(tmpnew
, consumerd64_libdir
);
2179 if (tmp
[0] != '\0') {
2180 strcat(tmpnew
, ":");
2181 strcat(tmpnew
, tmp
);
2183 ret
= putenv(tmpnew
);
2190 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2191 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2192 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2193 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2194 "--group", tracing_group_name
,
2196 if (consumerd64_libdir
[0] != '\0') {
2204 case LTTNG_CONSUMER32_UST
:
2206 char *tmpnew
= NULL
;
2208 if (consumerd32_libdir
[0] != '\0') {
2212 tmp
= getenv("LD_LIBRARY_PATH");
2216 tmplen
= strlen("LD_LIBRARY_PATH=")
2217 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2218 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2223 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2224 strcat(tmpnew
, consumerd32_libdir
);
2225 if (tmp
[0] != '\0') {
2226 strcat(tmpnew
, ":");
2227 strcat(tmpnew
, tmp
);
2229 ret
= putenv(tmpnew
);
2236 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2237 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2238 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2239 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2240 "--group", tracing_group_name
,
2242 if (consumerd32_libdir
[0] != '\0') {
2251 PERROR("unknown consumer type");
2255 PERROR("kernel start consumer exec");
2258 } else if (pid
> 0) {
2261 PERROR("start consumer fork");
2269 * Spawn the consumerd daemon and session daemon thread.
2271 static int start_consumerd(struct consumer_data
*consumer_data
)
2276 * Set the listen() state on the socket since there is a possible race
2277 * between the exec() of the consumer daemon and this call if place in the
2278 * consumer thread. See bug #366 for more details.
2280 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2285 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2286 if (consumer_data
->pid
!= 0) {
2287 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2291 ret
= spawn_consumerd(consumer_data
);
2293 ERR("Spawning consumerd failed");
2294 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2298 /* Setting up the consumer_data pid */
2299 consumer_data
->pid
= ret
;
2300 DBG2("Consumer pid %d", consumer_data
->pid
);
2301 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2303 DBG2("Spawning consumer control thread");
2304 ret
= spawn_consumer_thread(consumer_data
);
2306 ERR("Fatal error spawning consumer control thread");
2314 /* Cleanup already created sockets on error. */
2315 if (consumer_data
->err_sock
>= 0) {
2318 err
= close(consumer_data
->err_sock
);
2320 PERROR("close consumer data error socket");
2327 * Setup necessary data for kernel tracer action.
2329 static int init_kernel_tracer(void)
2333 /* Modprobe lttng kernel modules */
2334 ret
= modprobe_lttng_control();
2339 /* Open debugfs lttng */
2340 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2341 if (kernel_tracer_fd
< 0) {
2342 DBG("Failed to open %s", module_proc_lttng
);
2347 /* Validate kernel version */
2348 ret
= kernel_validate_version(kernel_tracer_fd
);
2353 ret
= modprobe_lttng_data();
2358 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2362 modprobe_remove_lttng_control();
2363 ret
= close(kernel_tracer_fd
);
2367 kernel_tracer_fd
= -1;
2368 return LTTNG_ERR_KERN_VERSION
;
2371 ret
= close(kernel_tracer_fd
);
2377 modprobe_remove_lttng_control();
2380 WARN("No kernel tracer available");
2381 kernel_tracer_fd
= -1;
2383 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2385 return LTTNG_ERR_KERN_NA
;
2391 * Copy consumer output from the tracing session to the domain session. The
2392 * function also applies the right modification on a per domain basis for the
2393 * trace files destination directory.
2395 * Should *NOT* be called with RCU read-side lock held.
2397 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2400 const char *dir_name
;
2401 struct consumer_output
*consumer
;
2404 assert(session
->consumer
);
2407 case LTTNG_DOMAIN_KERNEL
:
2408 DBG3("Copying tracing session consumer output in kernel session");
2410 * XXX: We should audit the session creation and what this function
2411 * does "extra" in order to avoid a destroy since this function is used
2412 * in the domain session creation (kernel and ust) only. Same for UST
2415 if (session
->kernel_session
->consumer
) {
2416 consumer_destroy_output(session
->kernel_session
->consumer
);
2418 session
->kernel_session
->consumer
=
2419 consumer_copy_output(session
->consumer
);
2420 /* Ease our life a bit for the next part */
2421 consumer
= session
->kernel_session
->consumer
;
2422 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2424 case LTTNG_DOMAIN_JUL
:
2425 case LTTNG_DOMAIN_UST
:
2426 DBG3("Copying tracing session consumer output in UST session");
2427 if (session
->ust_session
->consumer
) {
2428 consumer_destroy_output(session
->ust_session
->consumer
);
2430 session
->ust_session
->consumer
=
2431 consumer_copy_output(session
->consumer
);
2432 /* Ease our life a bit for the next part */
2433 consumer
= session
->ust_session
->consumer
;
2434 dir_name
= DEFAULT_UST_TRACE_DIR
;
2437 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2441 /* Append correct directory to subdir */
2442 strncat(consumer
->subdir
, dir_name
,
2443 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2444 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2453 * Create an UST session and add it to the session ust list.
2455 * Should *NOT* be called with RCU read-side lock held.
2457 static int create_ust_session(struct ltt_session
*session
,
2458 struct lttng_domain
*domain
)
2461 struct ltt_ust_session
*lus
= NULL
;
2465 assert(session
->consumer
);
2467 switch (domain
->type
) {
2468 case LTTNG_DOMAIN_JUL
:
2469 case LTTNG_DOMAIN_UST
:
2472 ERR("Unknown UST domain on create session %d", domain
->type
);
2473 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2477 DBG("Creating UST session");
2479 lus
= trace_ust_create_session(session
->id
);
2481 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2485 lus
->uid
= session
->uid
;
2486 lus
->gid
= session
->gid
;
2487 lus
->output_traces
= session
->output_traces
;
2488 lus
->snapshot_mode
= session
->snapshot_mode
;
2489 lus
->live_timer_interval
= session
->live_timer
;
2490 session
->ust_session
= lus
;
2492 /* Copy session output to the newly created UST session */
2493 ret
= copy_session_consumer(domain
->type
, session
);
2494 if (ret
!= LTTNG_OK
) {
2502 session
->ust_session
= NULL
;
2507 * Create a kernel tracer session then create the default channel.
2509 static int create_kernel_session(struct ltt_session
*session
)
2513 DBG("Creating kernel session");
2515 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2517 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2521 /* Code flow safety */
2522 assert(session
->kernel_session
);
2524 /* Copy session output to the newly created Kernel session */
2525 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2526 if (ret
!= LTTNG_OK
) {
2530 /* Create directory(ies) on local filesystem. */
2531 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2532 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2533 ret
= run_as_mkdir_recursive(
2534 session
->kernel_session
->consumer
->dst
.trace_path
,
2535 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2537 if (ret
!= -EEXIST
) {
2538 ERR("Trace directory creation error");
2544 session
->kernel_session
->uid
= session
->uid
;
2545 session
->kernel_session
->gid
= session
->gid
;
2546 session
->kernel_session
->output_traces
= session
->output_traces
;
2547 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2552 trace_kernel_destroy_session(session
->kernel_session
);
2553 session
->kernel_session
= NULL
;
2558 * Count number of session permitted by uid/gid.
2560 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2563 struct ltt_session
*session
;
2565 DBG("Counting number of available session for UID %d GID %d",
2567 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2569 * Only list the sessions the user can control.
2571 if (!session_access_ok(session
, uid
, gid
)) {
2580 * Process the command requested by the lttng client within the command
2581 * context structure. This function make sure that the return structure (llm)
2582 * is set and ready for transmission before returning.
2584 * Return any error encountered or 0 for success.
2586 * "sock" is only used for special-case var. len data.
2588 * Should *NOT* be called with RCU read-side lock held.
2590 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2594 int need_tracing_session
= 1;
2597 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2601 switch (cmd_ctx
->lsm
->cmd_type
) {
2602 case LTTNG_CREATE_SESSION
:
2603 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2604 case LTTNG_CREATE_SESSION_LIVE
:
2605 case LTTNG_DESTROY_SESSION
:
2606 case LTTNG_LIST_SESSIONS
:
2607 case LTTNG_LIST_DOMAINS
:
2608 case LTTNG_START_TRACE
:
2609 case LTTNG_STOP_TRACE
:
2610 case LTTNG_DATA_PENDING
:
2611 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2612 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2613 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2614 case LTTNG_SNAPSHOT_RECORD
:
2621 if (opt_no_kernel
&& need_domain
2622 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2624 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2626 ret
= LTTNG_ERR_KERN_NA
;
2631 /* Deny register consumer if we already have a spawned consumer. */
2632 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2633 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2634 if (kconsumer_data
.pid
> 0) {
2635 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2636 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2639 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2643 * Check for command that don't needs to allocate a returned payload. We do
2644 * this here so we don't have to make the call for no payload at each
2647 switch(cmd_ctx
->lsm
->cmd_type
) {
2648 case LTTNG_LIST_SESSIONS
:
2649 case LTTNG_LIST_TRACEPOINTS
:
2650 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2651 case LTTNG_LIST_DOMAINS
:
2652 case LTTNG_LIST_CHANNELS
:
2653 case LTTNG_LIST_EVENTS
:
2656 /* Setup lttng message with no payload */
2657 ret
= setup_lttng_msg(cmd_ctx
, 0);
2659 /* This label does not try to unlock the session */
2660 goto init_setup_error
;
2664 /* Commands that DO NOT need a session. */
2665 switch (cmd_ctx
->lsm
->cmd_type
) {
2666 case LTTNG_CREATE_SESSION
:
2667 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2668 case LTTNG_CREATE_SESSION_LIVE
:
2669 case LTTNG_CALIBRATE
:
2670 case LTTNG_LIST_SESSIONS
:
2671 case LTTNG_LIST_TRACEPOINTS
:
2672 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2673 need_tracing_session
= 0;
2676 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2678 * We keep the session list lock across _all_ commands
2679 * for now, because the per-session lock does not
2680 * handle teardown properly.
2682 session_lock_list();
2683 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2684 if (cmd_ctx
->session
== NULL
) {
2685 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2688 /* Acquire lock for the session */
2689 session_lock(cmd_ctx
->session
);
2699 * Check domain type for specific "pre-action".
2701 switch (cmd_ctx
->lsm
->domain
.type
) {
2702 case LTTNG_DOMAIN_KERNEL
:
2704 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2708 /* Kernel tracer check */
2709 if (kernel_tracer_fd
== -1) {
2710 /* Basically, load kernel tracer modules */
2711 ret
= init_kernel_tracer();
2717 /* Consumer is in an ERROR state. Report back to client */
2718 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2719 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2723 /* Need a session for kernel command */
2724 if (need_tracing_session
) {
2725 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2726 ret
= create_kernel_session(cmd_ctx
->session
);
2728 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2733 /* Start the kernel consumer daemon */
2734 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2735 if (kconsumer_data
.pid
== 0 &&
2736 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2737 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2738 ret
= start_consumerd(&kconsumer_data
);
2740 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2743 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2745 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2749 * The consumer was just spawned so we need to add the socket to
2750 * the consumer output of the session if exist.
2752 ret
= consumer_create_socket(&kconsumer_data
,
2753 cmd_ctx
->session
->kernel_session
->consumer
);
2760 case LTTNG_DOMAIN_JUL
:
2761 case LTTNG_DOMAIN_UST
:
2763 if (!ust_app_supported()) {
2764 ret
= LTTNG_ERR_NO_UST
;
2767 /* Consumer is in an ERROR state. Report back to client */
2768 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2769 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2773 if (need_tracing_session
) {
2774 /* Create UST session if none exist. */
2775 if (cmd_ctx
->session
->ust_session
== NULL
) {
2776 ret
= create_ust_session(cmd_ctx
->session
,
2777 &cmd_ctx
->lsm
->domain
);
2778 if (ret
!= LTTNG_OK
) {
2783 /* Start the UST consumer daemons */
2785 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2786 if (consumerd64_bin
[0] != '\0' &&
2787 ustconsumer64_data
.pid
== 0 &&
2788 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2789 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2790 ret
= start_consumerd(&ustconsumer64_data
);
2792 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2793 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2797 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2798 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2800 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2804 * Setup socket for consumer 64 bit. No need for atomic access
2805 * since it was set above and can ONLY be set in this thread.
2807 ret
= consumer_create_socket(&ustconsumer64_data
,
2808 cmd_ctx
->session
->ust_session
->consumer
);
2814 if (consumerd32_bin
[0] != '\0' &&
2815 ustconsumer32_data
.pid
== 0 &&
2816 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2817 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2818 ret
= start_consumerd(&ustconsumer32_data
);
2820 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2821 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2825 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2826 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2828 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2832 * Setup socket for consumer 64 bit. No need for atomic access
2833 * since it was set above and can ONLY be set in this thread.
2835 ret
= consumer_create_socket(&ustconsumer32_data
,
2836 cmd_ctx
->session
->ust_session
->consumer
);
2848 /* Validate consumer daemon state when start/stop trace command */
2849 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2850 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2851 switch (cmd_ctx
->lsm
->domain
.type
) {
2852 case LTTNG_DOMAIN_JUL
:
2853 case LTTNG_DOMAIN_UST
:
2854 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2855 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2859 case LTTNG_DOMAIN_KERNEL
:
2860 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2861 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2869 * Check that the UID or GID match that of the tracing session.
2870 * The root user can interact with all sessions.
2872 if (need_tracing_session
) {
2873 if (!session_access_ok(cmd_ctx
->session
,
2874 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2875 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2876 ret
= LTTNG_ERR_EPERM
;
2882 * Send relayd information to consumer as soon as we have a domain and a
2885 if (cmd_ctx
->session
&& need_domain
) {
2887 * Setup relayd if not done yet. If the relayd information was already
2888 * sent to the consumer, this call will gracefully return.
2890 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2891 if (ret
!= LTTNG_OK
) {
2896 /* Process by command type */
2897 switch (cmd_ctx
->lsm
->cmd_type
) {
2898 case LTTNG_ADD_CONTEXT
:
2900 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2901 cmd_ctx
->lsm
->u
.context
.channel_name
,
2902 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2905 case LTTNG_DISABLE_CHANNEL
:
2907 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2908 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2911 case LTTNG_DISABLE_EVENT
:
2913 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2914 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2915 cmd_ctx
->lsm
->u
.disable
.name
);
2918 case LTTNG_DISABLE_ALL_EVENT
:
2920 DBG("Disabling all events");
2922 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2923 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2926 case LTTNG_ENABLE_CHANNEL
:
2928 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2929 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2932 case LTTNG_ENABLE_EVENT
:
2934 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2935 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2936 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2939 case LTTNG_ENABLE_ALL_EVENT
:
2941 DBG("Enabling all events");
2943 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2944 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2945 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2948 case LTTNG_LIST_TRACEPOINTS
:
2950 struct lttng_event
*events
;
2953 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2954 if (nb_events
< 0) {
2955 /* Return value is a negative lttng_error_code. */
2961 * Setup lttng message with payload size set to the event list size in
2962 * bytes and then copy list into the llm payload.
2964 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2970 /* Copy event list into message payload */
2971 memcpy(cmd_ctx
->llm
->payload
, events
,
2972 sizeof(struct lttng_event
) * nb_events
);
2979 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2981 struct lttng_event_field
*fields
;
2984 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2986 if (nb_fields
< 0) {
2987 /* Return value is a negative lttng_error_code. */
2993 * Setup lttng message with payload size set to the event list size in
2994 * bytes and then copy list into the llm payload.
2996 ret
= setup_lttng_msg(cmd_ctx
,
2997 sizeof(struct lttng_event_field
) * nb_fields
);
3003 /* Copy event list into message payload */
3004 memcpy(cmd_ctx
->llm
->payload
, fields
,
3005 sizeof(struct lttng_event_field
) * nb_fields
);
3012 case LTTNG_SET_CONSUMER_URI
:
3015 struct lttng_uri
*uris
;
3017 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3018 len
= nb_uri
* sizeof(struct lttng_uri
);
3021 ret
= LTTNG_ERR_INVALID
;
3025 uris
= zmalloc(len
);
3027 ret
= LTTNG_ERR_FATAL
;
3031 /* Receive variable len data */
3032 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3033 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3035 DBG("No URIs received from client... continuing");
3037 ret
= LTTNG_ERR_SESSION_FAIL
;
3042 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3044 if (ret
!= LTTNG_OK
) {
3050 * XXX: 0 means that this URI should be applied on the session. Should
3051 * be a DOMAIN enuam.
3053 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3054 /* Add the URI for the UST session if a consumer is present. */
3055 if (cmd_ctx
->session
->ust_session
&&
3056 cmd_ctx
->session
->ust_session
->consumer
) {
3057 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3059 } else if (cmd_ctx
->session
->kernel_session
&&
3060 cmd_ctx
->session
->kernel_session
->consumer
) {
3061 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3062 cmd_ctx
->session
, nb_uri
, uris
);
3070 case LTTNG_START_TRACE
:
3072 ret
= cmd_start_trace(cmd_ctx
->session
);
3075 case LTTNG_STOP_TRACE
:
3077 ret
= cmd_stop_trace(cmd_ctx
->session
);
3080 case LTTNG_CREATE_SESSION
:
3083 struct lttng_uri
*uris
= NULL
;
3085 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3086 len
= nb_uri
* sizeof(struct lttng_uri
);
3089 uris
= zmalloc(len
);
3091 ret
= LTTNG_ERR_FATAL
;
3095 /* Receive variable len data */
3096 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3097 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3099 DBG("No URIs received from client... continuing");
3101 ret
= LTTNG_ERR_SESSION_FAIL
;
3106 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3107 DBG("Creating session with ONE network URI is a bad call");
3108 ret
= LTTNG_ERR_SESSION_FAIL
;
3114 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3115 &cmd_ctx
->creds
, 0);
3121 case LTTNG_DESTROY_SESSION
:
3123 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3125 /* Set session to NULL so we do not unlock it after free. */
3126 cmd_ctx
->session
= NULL
;
3129 case LTTNG_LIST_DOMAINS
:
3132 struct lttng_domain
*domains
;
3134 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3136 /* Return value is a negative lttng_error_code. */
3141 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3147 /* Copy event list into message payload */
3148 memcpy(cmd_ctx
->llm
->payload
, domains
,
3149 nb_dom
* sizeof(struct lttng_domain
));
3156 case LTTNG_LIST_CHANNELS
:
3159 struct lttng_channel
*channels
;
3161 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3162 cmd_ctx
->session
, &channels
);
3164 /* Return value is a negative lttng_error_code. */
3169 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3175 /* Copy event list into message payload */
3176 memcpy(cmd_ctx
->llm
->payload
, channels
,
3177 nb_chan
* sizeof(struct lttng_channel
));
3184 case LTTNG_LIST_EVENTS
:
3187 struct lttng_event
*events
= NULL
;
3189 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3190 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3192 /* Return value is a negative lttng_error_code. */
3197 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3203 /* Copy event list into message payload */
3204 memcpy(cmd_ctx
->llm
->payload
, events
,
3205 nb_event
* sizeof(struct lttng_event
));
3212 case LTTNG_LIST_SESSIONS
:
3214 unsigned int nr_sessions
;
3216 session_lock_list();
3217 nr_sessions
= lttng_sessions_count(
3218 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3219 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3221 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3223 session_unlock_list();
3227 /* Filled the session array */
3228 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3229 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3230 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3232 session_unlock_list();
3237 case LTTNG_CALIBRATE
:
3239 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3240 &cmd_ctx
->lsm
->u
.calibrate
);
3243 case LTTNG_REGISTER_CONSUMER
:
3245 struct consumer_data
*cdata
;
3247 switch (cmd_ctx
->lsm
->domain
.type
) {
3248 case LTTNG_DOMAIN_KERNEL
:
3249 cdata
= &kconsumer_data
;
3252 ret
= LTTNG_ERR_UND
;
3256 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3257 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3260 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3262 struct lttng_filter_bytecode
*bytecode
;
3264 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3265 ret
= LTTNG_ERR_FILTER_INVAL
;
3268 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3269 ret
= LTTNG_ERR_FILTER_INVAL
;
3272 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3274 ret
= LTTNG_ERR_FILTER_NOMEM
;
3277 /* Receive var. len. data */
3278 DBG("Receiving var len data from client ...");
3279 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3280 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3282 DBG("Nothing recv() from client var len data... continuing");
3284 ret
= LTTNG_ERR_FILTER_INVAL
;
3288 if (bytecode
->len
+ sizeof(*bytecode
)
3289 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3291 ret
= LTTNG_ERR_FILTER_INVAL
;
3295 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3296 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3297 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3300 case LTTNG_DATA_PENDING
:
3302 ret
= cmd_data_pending(cmd_ctx
->session
);
3305 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3307 struct lttcomm_lttng_output_id reply
;
3309 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3310 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3311 if (ret
!= LTTNG_OK
) {
3315 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3320 /* Copy output list into message payload */
3321 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3325 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3327 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3328 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3331 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3334 struct lttng_snapshot_output
*outputs
= NULL
;
3336 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3337 if (nb_output
< 0) {
3342 ret
= setup_lttng_msg(cmd_ctx
,
3343 nb_output
* sizeof(struct lttng_snapshot_output
));
3350 /* Copy output list into message payload */
3351 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3352 nb_output
* sizeof(struct lttng_snapshot_output
));
3359 case LTTNG_SNAPSHOT_RECORD
:
3361 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3362 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3363 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3366 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3369 struct lttng_uri
*uris
= NULL
;
3371 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3372 len
= nb_uri
* sizeof(struct lttng_uri
);
3375 uris
= zmalloc(len
);
3377 ret
= LTTNG_ERR_FATAL
;
3381 /* Receive variable len data */
3382 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3383 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3385 DBG("No URIs received from client... continuing");
3387 ret
= LTTNG_ERR_SESSION_FAIL
;
3392 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3393 DBG("Creating session with ONE network URI is a bad call");
3394 ret
= LTTNG_ERR_SESSION_FAIL
;
3400 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3401 nb_uri
, &cmd_ctx
->creds
);
3405 case LTTNG_CREATE_SESSION_LIVE
:
3408 struct lttng_uri
*uris
= NULL
;
3410 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3411 len
= nb_uri
* sizeof(struct lttng_uri
);
3414 uris
= zmalloc(len
);
3416 ret
= LTTNG_ERR_FATAL
;
3420 /* Receive variable len data */
3421 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3422 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3424 DBG("No URIs received from client... continuing");
3426 ret
= LTTNG_ERR_SESSION_FAIL
;
3431 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3432 DBG("Creating session with ONE network URI is a bad call");
3433 ret
= LTTNG_ERR_SESSION_FAIL
;
3439 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3440 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3445 ret
= LTTNG_ERR_UND
;
3450 if (cmd_ctx
->llm
== NULL
) {
3451 DBG("Missing llm structure. Allocating one.");
3452 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3456 /* Set return code */
3457 cmd_ctx
->llm
->ret_code
= ret
;
3459 if (cmd_ctx
->session
) {
3460 session_unlock(cmd_ctx
->session
);
3462 if (need_tracing_session
) {
3463 session_unlock_list();
3470 * Thread managing health check socket.
3472 static void *thread_manage_health(void *data
)
3474 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3475 uint32_t revents
, nb_fd
;
3476 struct lttng_poll_event events
;
3477 struct health_comm_msg msg
;
3478 struct health_comm_reply reply
;
3480 DBG("[thread] Manage health check started");
3482 rcu_register_thread();
3484 /* We might hit an error path before this is created. */
3485 lttng_poll_init(&events
);
3487 /* Create unix socket */
3488 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3490 ERR("Unable to create health check Unix socket");
3496 /* lttng health client socket path permissions */
3497 ret
= chown(health_unix_sock_path
, 0,
3498 utils_get_group_id(tracing_group_name
));
3500 ERR("Unable to set group on %s", health_unix_sock_path
);
3506 ret
= chmod(health_unix_sock_path
,
3507 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3509 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3517 * Set the CLOEXEC flag. Return code is useless because either way, the
3520 (void) utils_set_fd_cloexec(sock
);
3522 ret
= lttcomm_listen_unix_sock(sock
);
3528 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3529 * more will be added to this poll set.
3531 ret
= sessiond_set_thread_pollset(&events
, 2);
3536 /* Add the application registration socket */
3537 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3543 DBG("Health check ready");
3545 /* Inifinite blocking call, waiting for transmission */
3547 ret
= lttng_poll_wait(&events
, -1);
3550 * Restart interrupted system call.
3552 if (errno
== EINTR
) {
3560 for (i
= 0; i
< nb_fd
; i
++) {
3561 /* Fetch once the poll data */
3562 revents
= LTTNG_POLL_GETEV(&events
, i
);
3563 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3565 /* Thread quit pipe has been closed. Killing thread. */
3566 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3572 /* Event on the registration socket */
3573 if (pollfd
== sock
) {
3574 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3575 ERR("Health socket poll error");
3581 new_sock
= lttcomm_accept_unix_sock(sock
);
3587 * Set the CLOEXEC flag. Return code is useless because either way, the
3590 (void) utils_set_fd_cloexec(new_sock
);
3592 DBG("Receiving data from client for health...");
3593 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3595 DBG("Nothing recv() from client... continuing");
3596 ret
= close(new_sock
);
3604 rcu_thread_online();
3607 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3609 * health_check_state returns 0 if health is
3612 if (!health_check_state(health_sessiond
, i
)) {
3613 reply
.ret_code
|= 1ULL << i
;
3617 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3619 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3621 ERR("Failed to send health data back to client");
3624 /* End of transmission */
3625 ret
= close(new_sock
);
3635 ERR("Health error occurred in %s", __func__
);
3637 DBG("Health check thread dying");
3638 unlink(health_unix_sock_path
);
3646 lttng_poll_clean(&events
);
3648 rcu_unregister_thread();
3653 * This thread manage all clients request using the unix client socket for
3656 static void *thread_manage_clients(void *data
)
3658 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3660 uint32_t revents
, nb_fd
;
3661 struct command_ctx
*cmd_ctx
= NULL
;
3662 struct lttng_poll_event events
;
3664 DBG("[thread] Manage client started");
3666 rcu_register_thread();
3668 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3670 if (testpoint(thread_manage_clients
)) {
3671 goto error_testpoint
;
3674 health_code_update();
3676 ret
= lttcomm_listen_unix_sock(client_sock
);
3682 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3683 * more will be added to this poll set.
3685 ret
= sessiond_set_thread_pollset(&events
, 2);
3687 goto error_create_poll
;
3690 /* Add the application registration socket */
3691 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3697 * Notify parent pid that we are ready to accept command for client side.
3699 if (opt_sig_parent
) {
3700 kill(ppid
, SIGUSR1
);
3703 if (testpoint(thread_manage_clients_before_loop
)) {
3707 health_code_update();
3710 DBG("Accepting client command ...");
3712 /* Inifinite blocking call, waiting for transmission */
3714 health_poll_entry();
3715 ret
= lttng_poll_wait(&events
, -1);
3719 * Restart interrupted system call.
3721 if (errno
== EINTR
) {
3729 for (i
= 0; i
< nb_fd
; i
++) {
3730 /* Fetch once the poll data */
3731 revents
= LTTNG_POLL_GETEV(&events
, i
);
3732 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3734 health_code_update();
3736 /* Thread quit pipe has been closed. Killing thread. */
3737 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3743 /* Event on the registration socket */
3744 if (pollfd
== client_sock
) {
3745 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3746 ERR("Client socket poll error");
3752 DBG("Wait for client response");
3754 health_code_update();
3756 sock
= lttcomm_accept_unix_sock(client_sock
);
3762 * Set the CLOEXEC flag. Return code is useless because either way, the
3765 (void) utils_set_fd_cloexec(sock
);
3767 /* Set socket option for credentials retrieval */
3768 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3773 /* Allocate context command to process the client request */
3774 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3775 if (cmd_ctx
== NULL
) {
3776 PERROR("zmalloc cmd_ctx");
3780 /* Allocate data buffer for reception */
3781 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3782 if (cmd_ctx
->lsm
== NULL
) {
3783 PERROR("zmalloc cmd_ctx->lsm");
3787 cmd_ctx
->llm
= NULL
;
3788 cmd_ctx
->session
= NULL
;
3790 health_code_update();
3793 * Data is received from the lttng client. The struct
3794 * lttcomm_session_msg (lsm) contains the command and data request of
3797 DBG("Receiving data from client ...");
3798 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3799 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3801 DBG("Nothing recv() from client... continuing");
3807 clean_command_ctx(&cmd_ctx
);
3811 health_code_update();
3813 // TODO: Validate cmd_ctx including sanity check for
3814 // security purpose.
3816 rcu_thread_online();
3818 * This function dispatch the work to the kernel or userspace tracer
3819 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3820 * informations for the client. The command context struct contains
3821 * everything this function may needs.
3823 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3824 rcu_thread_offline();
3832 * TODO: Inform client somehow of the fatal error. At
3833 * this point, ret < 0 means that a zmalloc failed
3834 * (ENOMEM). Error detected but still accept
3835 * command, unless a socket error has been
3838 clean_command_ctx(&cmd_ctx
);
3842 health_code_update();
3844 DBG("Sending response (size: %d, retcode: %s)",
3845 cmd_ctx
->lttng_msg_size
,
3846 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3847 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3849 ERR("Failed to send data back to client");
3852 /* End of transmission */
3859 clean_command_ctx(&cmd_ctx
);
3861 health_code_update();
3873 lttng_poll_clean(&events
);
3874 clean_command_ctx(&cmd_ctx
);
3879 unlink(client_unix_sock_path
);
3880 if (client_sock
>= 0) {
3881 ret
= close(client_sock
);
3889 ERR("Health error occurred in %s", __func__
);
3892 health_unregister(health_sessiond
);
3894 DBG("Client thread dying");
3896 rcu_unregister_thread();
3902 * usage function on stderr
3904 static void usage(void)
3906 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3907 fprintf(stderr
, " -h, --help Display this usage.\n");
3908 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3909 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3910 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3911 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3912 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3913 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3914 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3915 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3916 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3917 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3918 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3919 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3920 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3921 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3922 fprintf(stderr
, " -V, --version Show version number.\n");
3923 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3924 fprintf(stderr
, " -q, --quiet No output at all.\n");
3925 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3926 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3927 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3928 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3929 fprintf(stderr
, " --jul-tcp-port JUL application registration TCP port\n");
3933 * daemon argument parsing
3935 static int parse_args(int argc
, char **argv
)
3939 static struct option long_options
[] = {
3940 { "client-sock", 1, 0, 'c' },
3941 { "apps-sock", 1, 0, 'a' },
3942 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3943 { "kconsumerd-err-sock", 1, 0, 'E' },
3944 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3945 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3946 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3947 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3948 { "consumerd32-path", 1, 0, 'u' },
3949 { "consumerd32-libdir", 1, 0, 'U' },
3950 { "consumerd64-path", 1, 0, 't' },
3951 { "consumerd64-libdir", 1, 0, 'T' },
3952 { "daemonize", 0, 0, 'd' },
3953 { "sig-parent", 0, 0, 'S' },
3954 { "help", 0, 0, 'h' },
3955 { "group", 1, 0, 'g' },
3956 { "version", 0, 0, 'V' },
3957 { "quiet", 0, 0, 'q' },
3958 { "verbose", 0, 0, 'v' },
3959 { "verbose-consumer", 0, 0, 'Z' },
3960 { "no-kernel", 0, 0, 'N' },
3961 { "pidfile", 1, 0, 'p' },
3962 { "jul-tcp-port", 1, 0, 'J' },
3967 int option_index
= 0;
3968 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:",
3969 long_options
, &option_index
);
3976 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3978 fprintf(stderr
, " with arg %s\n", optarg
);
3982 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3985 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3991 tracing_group_name
= optarg
;
3997 fprintf(stdout
, "%s\n", VERSION
);
4003 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4006 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4009 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4012 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4015 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4018 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4024 lttng_opt_quiet
= 1;
4027 /* Verbose level can increase using multiple -v */
4028 lttng_opt_verbose
+= 1;
4031 opt_verbose_consumer
+= 1;
4034 consumerd32_bin
= optarg
;
4037 consumerd32_libdir
= optarg
;
4040 consumerd64_bin
= optarg
;
4043 consumerd64_libdir
= optarg
;
4046 opt_pidfile
= optarg
;
4048 case 'J': /* JUL TCP port. */
4053 v
= strtoul(optarg
, NULL
, 0);
4054 if (errno
!= 0 || !isdigit(optarg
[0])) {
4055 ERR("Wrong value in --jul-tcp-port parameter: %s", optarg
);
4058 if (v
== 0 || v
>= 65535) {
4059 ERR("Port overflow in --jul-tcp-port parameter: %s", optarg
);
4062 jul_tcp_port
= (uint32_t) v
;
4063 DBG3("JUL TCP port set to non default: %u", jul_tcp_port
);
4067 /* Unknown option or other error.
4068 * Error is printed by getopt, just return */
4077 * Creates the two needed socket by the daemon.
4078 * apps_sock - The communication socket for all UST apps.
4079 * client_sock - The communication of the cli tool (lttng).
4081 static int init_daemon_socket(void)
4086 old_umask
= umask(0);
4088 /* Create client tool unix socket */
4089 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4090 if (client_sock
< 0) {
4091 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4096 /* Set the cloexec flag */
4097 ret
= utils_set_fd_cloexec(client_sock
);
4099 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4100 "Continuing but note that the consumer daemon will have a "
4101 "reference to this socket on exec()", client_sock
);
4104 /* File permission MUST be 660 */
4105 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4107 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4112 /* Create the application unix socket */
4113 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4114 if (apps_sock
< 0) {
4115 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4120 /* Set the cloexec flag */
4121 ret
= utils_set_fd_cloexec(apps_sock
);
4123 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4124 "Continuing but note that the consumer daemon will have a "
4125 "reference to this socket on exec()", apps_sock
);
4128 /* File permission MUST be 666 */
4129 ret
= chmod(apps_unix_sock_path
,
4130 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4132 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4137 DBG3("Session daemon client socket %d and application socket %d created",
4138 client_sock
, apps_sock
);
4146 * Check if the global socket is available, and if a daemon is answering at the
4147 * other side. If yes, error is returned.
4149 static int check_existing_daemon(void)
4151 /* Is there anybody out there ? */
4152 if (lttng_session_daemon_alive()) {
4160 * Set the tracing group gid onto the client socket.
4162 * Race window between mkdir and chown is OK because we are going from more
4163 * permissive (root.root) to less permissive (root.tracing).
4165 static int set_permissions(char *rundir
)
4170 gid
= utils_get_group_id(tracing_group_name
);
4172 /* Set lttng run dir */
4173 ret
= chown(rundir
, 0, gid
);
4175 ERR("Unable to set group on %s", rundir
);
4180 * Ensure all applications and tracing group can search the run
4181 * dir. Allow everyone to read the directory, since it does not
4182 * buy us anything to hide its content.
4184 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4186 ERR("Unable to set permissions on %s", rundir
);
4190 /* lttng client socket path */
4191 ret
= chown(client_unix_sock_path
, 0, gid
);
4193 ERR("Unable to set group on %s", client_unix_sock_path
);
4197 /* kconsumer error socket path */
4198 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4200 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4204 /* 64-bit ustconsumer error socket path */
4205 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4207 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4211 /* 32-bit ustconsumer compat32 error socket path */
4212 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4214 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4218 DBG("All permissions are set");
4224 * Create the lttng run directory needed for all global sockets and pipe.
4226 static int create_lttng_rundir(const char *rundir
)
4230 DBG3("Creating LTTng run directory: %s", rundir
);
4232 ret
= mkdir(rundir
, S_IRWXU
);
4234 if (errno
!= EEXIST
) {
4235 ERR("Unable to create %s", rundir
);
4247 * Setup sockets and directory needed by the kconsumerd communication with the
4250 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4254 char path
[PATH_MAX
];
4256 switch (consumer_data
->type
) {
4257 case LTTNG_CONSUMER_KERNEL
:
4258 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4260 case LTTNG_CONSUMER64_UST
:
4261 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4263 case LTTNG_CONSUMER32_UST
:
4264 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4267 ERR("Consumer type unknown");
4272 DBG2("Creating consumer directory: %s", path
);
4274 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4276 if (errno
!= EEXIST
) {
4278 ERR("Failed to create %s", path
);
4284 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4286 ERR("Unable to set group on %s", path
);
4292 /* Create the kconsumerd error unix socket */
4293 consumer_data
->err_sock
=
4294 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4295 if (consumer_data
->err_sock
< 0) {
4296 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4302 * Set the CLOEXEC flag. Return code is useless because either way, the
4305 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4307 PERROR("utils_set_fd_cloexec");
4308 /* continue anyway */
4311 /* File permission MUST be 660 */
4312 ret
= chmod(consumer_data
->err_unix_sock_path
,
4313 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4315 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4325 * Signal handler for the daemon
4327 * Simply stop all worker threads, leaving main() return gracefully after
4328 * joining all threads and calling cleanup().
4330 static void sighandler(int sig
)
4334 DBG("SIGPIPE caught");
4337 DBG("SIGINT caught");
4341 DBG("SIGTERM caught");
4350 * Setup signal handler for :
4351 * SIGINT, SIGTERM, SIGPIPE
4353 static int set_signal_handler(void)
4356 struct sigaction sa
;
4359 if ((ret
= sigemptyset(&sigset
)) < 0) {
4360 PERROR("sigemptyset");
4364 sa
.sa_handler
= sighandler
;
4365 sa
.sa_mask
= sigset
;
4367 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4368 PERROR("sigaction");
4372 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4373 PERROR("sigaction");
4377 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4378 PERROR("sigaction");
4382 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4388 * Set open files limit to unlimited. This daemon can open a large number of
4389 * file descriptors in order to consumer multiple kernel traces.
4391 static void set_ulimit(void)
4396 /* The kernel does not allowed an infinite limit for open files */
4397 lim
.rlim_cur
= 65535;
4398 lim
.rlim_max
= 65535;
4400 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4402 PERROR("failed to set open files limit");
4407 * Write pidfile using the rundir and opt_pidfile.
4409 static void write_pidfile(void)
4412 char pidfile_path
[PATH_MAX
];
4417 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4419 /* Build pidfile path from rundir and opt_pidfile. */
4420 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4421 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4423 PERROR("snprintf pidfile path");
4429 * Create pid file in rundir. Return value is of no importance. The
4430 * execution will continue even though we are not able to write the file.
4432 (void) utils_create_pid_file(getpid(), pidfile_path
);
4439 * Write JUL TCP port using the rundir.
4441 static void write_julport(void)
4444 char path
[PATH_MAX
];
4448 ret
= snprintf(path
, sizeof(path
), "%s/"
4449 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
, rundir
);
4451 PERROR("snprintf julport path");
4456 * Create TCP JUL port file in rundir. Return value is of no importance.
4457 * The execution will continue even though we are not able to write the
4460 (void) utils_create_pid_file(jul_tcp_port
, path
);
4469 int main(int argc
, char **argv
)
4473 const char *home_path
, *env_app_timeout
;
4475 init_kernel_workarounds();
4477 rcu_register_thread();
4479 setup_consumerd_path();
4481 page_size
= sysconf(_SC_PAGESIZE
);
4482 if (page_size
< 0) {
4483 PERROR("sysconf _SC_PAGESIZE");
4484 page_size
= LONG_MAX
;
4485 WARN("Fallback page size to %ld", page_size
);
4488 /* Parse arguments */
4490 if ((ret
= parse_args(argc
, argv
)) < 0) {
4500 * child: setsid, close FD 0, 1, 2, chdir /
4501 * parent: exit (if fork is successful)
4509 * We are in the child. Make sure all other file
4510 * descriptors are closed, in case we are called with
4511 * more opened file descriptors than the standard ones.
4513 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4518 /* Create thread quit pipe */
4519 if ((ret
= init_thread_quit_pipe()) < 0) {
4523 /* Check if daemon is UID = 0 */
4524 is_root
= !getuid();
4527 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4529 /* Create global run dir with root access */
4530 ret
= create_lttng_rundir(rundir
);
4535 if (strlen(apps_unix_sock_path
) == 0) {
4536 snprintf(apps_unix_sock_path
, PATH_MAX
,
4537 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4540 if (strlen(client_unix_sock_path
) == 0) {
4541 snprintf(client_unix_sock_path
, PATH_MAX
,
4542 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4545 /* Set global SHM for ust */
4546 if (strlen(wait_shm_path
) == 0) {
4547 snprintf(wait_shm_path
, PATH_MAX
,
4548 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4551 if (strlen(health_unix_sock_path
) == 0) {
4552 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4553 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4556 /* Setup kernel consumerd path */
4557 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4558 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4559 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4560 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4562 DBG2("Kernel consumer err path: %s",
4563 kconsumer_data
.err_unix_sock_path
);
4564 DBG2("Kernel consumer cmd path: %s",
4565 kconsumer_data
.cmd_unix_sock_path
);
4567 home_path
= utils_get_home_dir();
4568 if (home_path
== NULL
) {
4569 /* TODO: Add --socket PATH option */
4570 ERR("Can't get HOME directory for sockets creation.");
4576 * Create rundir from home path. This will create something like
4579 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4585 ret
= create_lttng_rundir(rundir
);
4590 if (strlen(apps_unix_sock_path
) == 0) {
4591 snprintf(apps_unix_sock_path
, PATH_MAX
,
4592 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4595 /* Set the cli tool unix socket path */
4596 if (strlen(client_unix_sock_path
) == 0) {
4597 snprintf(client_unix_sock_path
, PATH_MAX
,
4598 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4601 /* Set global SHM for ust */
4602 if (strlen(wait_shm_path
) == 0) {
4603 snprintf(wait_shm_path
, PATH_MAX
,
4604 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4607 /* Set health check Unix path */
4608 if (strlen(health_unix_sock_path
) == 0) {
4609 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4610 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4614 /* Set consumer initial state */
4615 kernel_consumerd_state
= CONSUMER_STOPPED
;
4616 ust_consumerd_state
= CONSUMER_STOPPED
;
4618 DBG("Client socket path %s", client_unix_sock_path
);
4619 DBG("Application socket path %s", apps_unix_sock_path
);
4620 DBG("Application wait path %s", wait_shm_path
);
4621 DBG("LTTng run directory path: %s", rundir
);
4623 /* 32 bits consumerd path setup */
4624 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4625 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4626 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4627 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4629 DBG2("UST consumer 32 bits err path: %s",
4630 ustconsumer32_data
.err_unix_sock_path
);
4631 DBG2("UST consumer 32 bits cmd path: %s",
4632 ustconsumer32_data
.cmd_unix_sock_path
);
4634 /* 64 bits consumerd path setup */
4635 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4636 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4637 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4638 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4640 DBG2("UST consumer 64 bits err path: %s",
4641 ustconsumer64_data
.err_unix_sock_path
);
4642 DBG2("UST consumer 64 bits cmd path: %s",
4643 ustconsumer64_data
.cmd_unix_sock_path
);
4646 * See if daemon already exist.
4648 if ((ret
= check_existing_daemon()) < 0) {
4649 ERR("Already running daemon.\n");
4651 * We do not goto exit because we must not cleanup()
4652 * because a daemon is already running.
4658 * Init UST app hash table. Alloc hash table before this point since
4659 * cleanup() can get called after that point.
4663 /* Initialize JUL domain subsystem. */
4664 if ((ret
= jul_init()) < 0) {
4665 /* ENOMEM at this point. */
4669 /* After this point, we can safely call cleanup() with "goto exit" */
4672 * These actions must be executed as root. We do that *after* setting up
4673 * the sockets path because we MUST make the check for another daemon using
4674 * those paths *before* trying to set the kernel consumer sockets and init
4678 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4683 /* Setup kernel tracer */
4684 if (!opt_no_kernel
) {
4685 init_kernel_tracer();
4688 /* Set ulimit for open files */
4691 /* init lttng_fd tracking must be done after set_ulimit. */
4694 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4699 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4704 if ((ret
= set_signal_handler()) < 0) {
4708 /* Setup the needed unix socket */
4709 if ((ret
= init_daemon_socket()) < 0) {
4713 /* Set credentials to socket */
4714 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4718 /* Get parent pid if -S, --sig-parent is specified. */
4719 if (opt_sig_parent
) {
4723 /* Setup the kernel pipe for waking up the kernel thread */
4724 if (is_root
&& !opt_no_kernel
) {
4725 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4730 /* Setup the thread ht_cleanup communication pipe. */
4731 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4735 /* Setup the thread apps communication pipe. */
4736 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4740 /* Setup the thread apps notify communication pipe. */
4741 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4745 /* Initialize global buffer per UID and PID registry. */
4746 buffer_reg_init_uid_registry();
4747 buffer_reg_init_pid_registry();
4749 /* Init UST command queue. */
4750 cds_wfq_init(&ust_cmd_queue
.queue
);
4753 * Get session list pointer. This pointer MUST NOT be free(). This list is
4754 * statically declared in session.c
4756 session_list_ptr
= session_get_list();
4758 /* Set up max poll set size */
4759 lttng_poll_set_max_size();
4763 /* Check for the application socket timeout env variable. */
4764 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4765 if (env_app_timeout
) {
4766 app_socket_timeout
= atoi(env_app_timeout
);
4768 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4774 /* Initialize communication library */
4776 /* This is to get the TCP timeout value. */
4777 lttcomm_inet_init();
4780 * Initialize the health check subsystem. This call should set the
4781 * appropriate time values.
4783 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
4784 if (!health_sessiond
) {
4785 PERROR("health_app_create error");
4786 goto exit_health_sessiond_cleanup
;
4789 /* Create thread to manage the client socket */
4790 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4791 thread_ht_cleanup
, (void *) NULL
);
4793 PERROR("pthread_create ht_cleanup");
4794 goto exit_ht_cleanup
;
4797 /* Create thread to manage the client socket */
4798 ret
= pthread_create(&health_thread
, NULL
,
4799 thread_manage_health
, (void *) NULL
);
4801 PERROR("pthread_create health");
4805 /* Create thread to manage the client socket */
4806 ret
= pthread_create(&client_thread
, NULL
,
4807 thread_manage_clients
, (void *) NULL
);
4809 PERROR("pthread_create clients");
4813 /* Create thread to dispatch registration */
4814 ret
= pthread_create(&dispatch_thread
, NULL
,
4815 thread_dispatch_ust_registration
, (void *) NULL
);
4817 PERROR("pthread_create dispatch");
4821 /* Create thread to manage application registration. */
4822 ret
= pthread_create(®_apps_thread
, NULL
,
4823 thread_registration_apps
, (void *) NULL
);
4825 PERROR("pthread_create registration");
4829 /* Create thread to manage application socket */
4830 ret
= pthread_create(&apps_thread
, NULL
,
4831 thread_manage_apps
, (void *) NULL
);
4833 PERROR("pthread_create apps");
4837 /* Create thread to manage application notify socket */
4838 ret
= pthread_create(&apps_notify_thread
, NULL
,
4839 ust_thread_manage_notify
, (void *) NULL
);
4841 PERROR("pthread_create apps");
4842 goto exit_apps_notify
;
4845 /* Create JUL registration thread. */
4846 ret
= pthread_create(&jul_reg_thread
, NULL
,
4847 jul_thread_manage_registration
, (void *) NULL
);
4849 PERROR("pthread_create apps");
4853 /* Don't start this thread if kernel tracing is not requested nor root */
4854 if (is_root
&& !opt_no_kernel
) {
4855 /* Create kernel thread to manage kernel event */
4856 ret
= pthread_create(&kernel_thread
, NULL
,
4857 thread_manage_kernel
, (void *) NULL
);
4859 PERROR("pthread_create kernel");
4863 ret
= pthread_join(kernel_thread
, &status
);
4865 PERROR("pthread_join");
4866 goto error
; /* join error, exit without cleanup */
4871 ret
= pthread_join(jul_reg_thread
, &status
);
4873 PERROR("pthread_join JUL");
4874 goto error
; /* join error, exit without cleanup */
4878 ret
= pthread_join(apps_notify_thread
, &status
);
4880 PERROR("pthread_join apps notify");
4881 goto error
; /* join error, exit without cleanup */
4885 ret
= pthread_join(apps_thread
, &status
);
4887 PERROR("pthread_join apps");
4888 goto error
; /* join error, exit without cleanup */
4893 ret
= pthread_join(reg_apps_thread
, &status
);
4895 PERROR("pthread_join");
4896 goto error
; /* join error, exit without cleanup */
4900 ret
= pthread_join(dispatch_thread
, &status
);
4902 PERROR("pthread_join");
4903 goto error
; /* join error, exit without cleanup */
4907 ret
= pthread_join(client_thread
, &status
);
4909 PERROR("pthread_join");
4910 goto error
; /* join error, exit without cleanup */
4913 ret
= join_consumer_thread(&kconsumer_data
);
4915 PERROR("join_consumer");
4916 goto error
; /* join error, exit without cleanup */
4919 ret
= join_consumer_thread(&ustconsumer32_data
);
4921 PERROR("join_consumer ust32");
4922 goto error
; /* join error, exit without cleanup */
4925 ret
= join_consumer_thread(&ustconsumer64_data
);
4927 PERROR("join_consumer ust64");
4928 goto error
; /* join error, exit without cleanup */
4932 ret
= pthread_join(health_thread
, &status
);
4934 PERROR("pthread_join health thread");
4935 goto error
; /* join error, exit without cleanup */
4939 ret
= pthread_join(ht_cleanup_thread
, &status
);
4941 PERROR("pthread_join ht cleanup thread");
4942 goto error
; /* join error, exit without cleanup */
4945 health_app_destroy(health_sessiond
);
4946 exit_health_sessiond_cleanup
:
4949 * cleanup() is called when no other thread is running.
4951 rcu_thread_online();
4953 rcu_thread_offline();
4954 rcu_unregister_thread();