2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
34 #include <sys/types.h>
36 #include <urcu/uatomic.h>
40 #include <common/common.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
48 #include "lttng-sessiond.h"
49 #include "buffer-registry.h"
56 #include "kernel-consumer.h"
60 #include "ust-consumer.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
67 #define CONSUMERD_FILE "lttng-consumerd"
70 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
73 const char *opt_tracing_group
;
74 static const char *opt_pidfile
;
75 static int opt_sig_parent
;
76 static int opt_verbose_consumer
;
77 static int opt_daemon
;
78 static int opt_no_kernel
;
79 static int is_root
; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid
; /* Parent PID for --sig-parent option */
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
87 static struct consumer_data kconsumer_data
= {
88 .type
= LTTNG_CONSUMER_KERNEL
,
89 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
90 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
93 .metadata_sock
.fd
= -1,
94 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
95 .lock
= PTHREAD_MUTEX_INITIALIZER
,
96 .cond
= PTHREAD_COND_INITIALIZER
,
97 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
99 static struct consumer_data ustconsumer64_data
= {
100 .type
= LTTNG_CONSUMER64_UST
,
101 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
102 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
105 .metadata_sock
.fd
= -1,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer32_data
= {
112 .type
= LTTNG_CONSUMER32_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
117 .metadata_sock
.fd
= -1,
118 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 .lock
= PTHREAD_MUTEX_INITIALIZER
,
120 .cond
= PTHREAD_COND_INITIALIZER
,
121 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
124 /* Shared between threads */
125 static int dispatch_thread_exit
;
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path
[PATH_MAX
];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path
[PATH_MAX
];
131 /* global wait shm path for UST */
132 static char wait_shm_path
[PATH_MAX
];
133 /* Global health check unix path */
134 static char health_unix_sock_path
[PATH_MAX
];
136 /* Sockets and FDs */
137 static int client_sock
= -1;
138 static int apps_sock
= -1;
139 int kernel_tracer_fd
= -1;
140 static int kernel_poll_pipe
[2] = { -1, -1 };
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
146 static int thread_quit_pipe
[2] = { -1, -1 };
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
152 static int apps_cmd_pipe
[2] = { -1, -1 };
154 int apps_cmd_notify_pipe
[2] = { -1, -1 };
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread
;
158 static pthread_t apps_notify_thread
;
159 static pthread_t reg_apps_thread
;
160 static pthread_t client_thread
;
161 static pthread_t kernel_thread
;
162 static pthread_t dispatch_thread
;
163 static pthread_t health_thread
;
164 static pthread_t ht_cleanup_thread
;
167 * UST registration command queue. This queue is tied with a futex and uses a N
168 * wakers / 1 waiter implemented and detailed in futex.c/.h
170 * The thread_manage_apps and thread_dispatch_ust_registration interact with
171 * this queue and the wait/wake scheme.
173 static struct ust_cmd_queue ust_cmd_queue
;
176 * Pointer initialized before thread creation.
178 * This points to the tracing session list containing the session count and a
179 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
180 * MUST NOT be taken if you call a public function in session.c.
182 * The lock is nested inside the structure: session_list_ptr->lock. Please use
183 * session_lock_list and session_unlock_list for lock acquisition.
185 static struct ltt_session_list
*session_list_ptr
;
187 int ust_consumerd64_fd
= -1;
188 int ust_consumerd32_fd
= -1;
190 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
191 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
192 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
193 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
195 static const char *module_proc_lttng
= "/proc/lttng";
198 * Consumer daemon state which is changed when spawning it, killing it or in
199 * case of a fatal error.
201 enum consumerd_state
{
202 CONSUMER_STARTED
= 1,
203 CONSUMER_STOPPED
= 2,
208 * This consumer daemon state is used to validate if a client command will be
209 * able to reach the consumer. If not, the client is informed. For instance,
210 * doing a "lttng start" when the consumer state is set to ERROR will return an
211 * error to the client.
213 * The following example shows a possible race condition of this scheme:
215 * consumer thread error happens
217 * client cmd checks state -> still OK
218 * consumer thread exit, sets error
219 * client cmd try to talk to consumer
222 * However, since the consumer is a different daemon, we have no way of making
223 * sure the command will reach it safely even with this state flag. This is why
224 * we consider that up to the state validation during command processing, the
225 * command is safe. After that, we can not guarantee the correctness of the
226 * client request vis-a-vis the consumer.
228 static enum consumerd_state ust_consumerd_state
;
229 static enum consumerd_state kernel_consumerd_state
;
232 * Socket timeout for receiving and sending in seconds.
234 static int app_socket_timeout
;
236 /* Set in main() with the current page size. */
240 void setup_consumerd_path(void)
242 const char *bin
, *libdir
;
245 * Allow INSTALL_BIN_PATH to be used as a target path for the
246 * native architecture size consumer if CONFIG_CONSUMER*_PATH
247 * has not been defined.
249 #if (CAA_BITS_PER_LONG == 32)
250 if (!consumerd32_bin
[0]) {
251 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
253 if (!consumerd32_libdir
[0]) {
254 consumerd32_libdir
= INSTALL_LIB_PATH
;
256 #elif (CAA_BITS_PER_LONG == 64)
257 if (!consumerd64_bin
[0]) {
258 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
260 if (!consumerd64_libdir
[0]) {
261 consumerd64_libdir
= INSTALL_LIB_PATH
;
264 #error "Unknown bitness"
268 * runtime env. var. overrides the build default.
270 bin
= getenv("LTTNG_CONSUMERD32_BIN");
272 consumerd32_bin
= bin
;
274 bin
= getenv("LTTNG_CONSUMERD64_BIN");
276 consumerd64_bin
= bin
;
278 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
280 consumerd32_libdir
= libdir
;
282 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
284 consumerd64_libdir
= libdir
;
289 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
291 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
297 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
303 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
315 * Check if the thread quit pipe was triggered.
317 * Return 1 if it was triggered else 0;
319 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
321 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
329 * Return group ID of the tracing group or -1 if not found.
331 static gid_t
allowed_group(void)
335 if (opt_tracing_group
) {
336 grp
= getgrnam(opt_tracing_group
);
338 grp
= getgrnam(default_tracing_group
);
348 * Init thread quit pipe.
350 * Return -1 on error or 0 if all pipes are created.
352 static int init_thread_quit_pipe(void)
356 ret
= pipe(thread_quit_pipe
);
358 PERROR("thread quit pipe");
362 for (i
= 0; i
< 2; i
++) {
363 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
375 * Stop all threads by closing the thread quit pipe.
377 static void stop_threads(void)
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
385 ERR("write error on thread quit pipe");
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
390 futex_nto1_wake(&ust_cmd_queue
.futex
);
396 static void cleanup(void)
400 struct ltt_session
*sess
, *stmp
;
404 /* First thing first, stop all threads */
405 utils_close_pipe(thread_quit_pipe
);
408 * If opt_pidfile is undefined, the default file will be wiped when
409 * removing the rundir.
412 ret
= remove(opt_pidfile
);
414 PERROR("remove pidfile %s", opt_pidfile
);
418 DBG("Removing %s directory", rundir
);
419 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
421 ERR("asprintf failed. Something is really wrong!");
424 /* Remove lttng run directory */
427 ERR("Unable to clean %s", rundir
);
432 DBG("Cleaning up all sessions");
434 /* Destroy session list mutex */
435 if (session_list_ptr
!= NULL
) {
436 pthread_mutex_destroy(&session_list_ptr
->lock
);
438 /* Cleanup ALL session */
439 cds_list_for_each_entry_safe(sess
, stmp
,
440 &session_list_ptr
->head
, list
) {
441 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
445 DBG("Closing all UST sockets");
446 ust_app_clean_list();
447 buffer_reg_destroy_registries();
449 if (is_root
&& !opt_no_kernel
) {
450 DBG2("Closing kernel fd");
451 if (kernel_tracer_fd
>= 0) {
452 ret
= close(kernel_tracer_fd
);
457 DBG("Unloading kernel modules");
458 modprobe_remove_lttng_all();
462 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
463 "Matthew, BEET driven development works!%c[%dm",
464 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
469 * Send data on a unix socket using the liblttsessiondcomm API.
471 * Return lttcomm error code.
473 static int send_unix_sock(int sock
, void *buf
, size_t len
)
475 /* Check valid length */
480 return lttcomm_send_unix_sock(sock
, buf
, len
);
484 * Free memory of a command context structure.
486 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
488 DBG("Clean command context structure");
490 if ((*cmd_ctx
)->llm
) {
491 free((*cmd_ctx
)->llm
);
493 if ((*cmd_ctx
)->lsm
) {
494 free((*cmd_ctx
)->lsm
);
502 * Notify UST applications using the shm mmap futex.
504 static int notify_ust_apps(int active
)
508 DBG("Notifying applications of session daemon state: %d", active
);
510 /* See shm.c for this call implying mmap, shm and futex calls */
511 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
512 if (wait_shm_mmap
== NULL
) {
516 /* Wake waiting process */
517 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
519 /* Apps notified successfully */
527 * Setup the outgoing data buffer for the response (llm) by allocating the
528 * right amount of memory and copying the original information from the lsm
531 * Return total size of the buffer pointed by buf.
533 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
539 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
540 if (cmd_ctx
->llm
== NULL
) {
546 /* Copy common data */
547 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
548 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
550 cmd_ctx
->llm
->data_size
= size
;
551 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
560 * Update the kernel poll set of all channel fd available over all tracing
561 * session. Add the wakeup pipe at the end of the set.
563 static int update_kernel_poll(struct lttng_poll_event
*events
)
566 struct ltt_session
*session
;
567 struct ltt_kernel_channel
*channel
;
569 DBG("Updating kernel poll set");
572 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
573 session_lock(session
);
574 if (session
->kernel_session
== NULL
) {
575 session_unlock(session
);
579 cds_list_for_each_entry(channel
,
580 &session
->kernel_session
->channel_list
.head
, list
) {
581 /* Add channel fd to the kernel poll set */
582 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
584 session_unlock(session
);
587 DBG("Channel fd %d added to kernel set", channel
->fd
);
589 session_unlock(session
);
591 session_unlock_list();
596 session_unlock_list();
601 * Find the channel fd from 'fd' over all tracing session. When found, check
602 * for new channel stream and send those stream fds to the kernel consumer.
604 * Useful for CPU hotplug feature.
606 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
609 struct ltt_session
*session
;
610 struct ltt_kernel_session
*ksess
;
611 struct ltt_kernel_channel
*channel
;
613 DBG("Updating kernel streams for channel fd %d", fd
);
616 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
617 session_lock(session
);
618 if (session
->kernel_session
== NULL
) {
619 session_unlock(session
);
622 ksess
= session
->kernel_session
;
624 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
625 if (channel
->fd
== fd
) {
626 DBG("Channel found, updating kernel streams");
627 ret
= kernel_open_channel_stream(channel
);
633 * Have we already sent fds to the consumer? If yes, it means
634 * that tracing is started so it is safe to send our updated
637 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
638 struct lttng_ht_iter iter
;
639 struct consumer_socket
*socket
;
642 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
643 &iter
.iter
, socket
, node
.node
) {
644 /* Code flow error */
645 assert(socket
->fd
>= 0);
647 pthread_mutex_lock(socket
->lock
);
648 ret
= kernel_consumer_send_channel_stream(socket
,
650 pthread_mutex_unlock(socket
->lock
);
661 session_unlock(session
);
663 session_unlock_list();
667 session_unlock(session
);
668 session_unlock_list();
673 * For each tracing session, update newly registered apps. The session list
674 * lock MUST be acquired before calling this.
676 static void update_ust_app(int app_sock
)
678 struct ltt_session
*sess
, *stmp
;
680 /* For all tracing session(s) */
681 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
683 if (sess
->ust_session
) {
684 ust_app_global_update(sess
->ust_session
, app_sock
);
686 session_unlock(sess
);
691 * This thread manage event coming from the kernel.
693 * Features supported in this thread:
696 static void *thread_manage_kernel(void *data
)
698 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
699 uint32_t revents
, nb_fd
;
701 struct lttng_poll_event events
;
703 DBG("[thread] Thread manage kernel started");
705 health_register(HEALTH_TYPE_KERNEL
);
708 * This first step of the while is to clean this structure which could free
709 * non NULL pointers so initialize it before the loop.
711 lttng_poll_init(&events
);
713 if (testpoint(thread_manage_kernel
)) {
714 goto error_testpoint
;
717 health_code_update();
719 if (testpoint(thread_manage_kernel_before_loop
)) {
720 goto error_testpoint
;
724 health_code_update();
726 if (update_poll_flag
== 1) {
727 /* Clean events object. We are about to populate it again. */
728 lttng_poll_clean(&events
);
730 ret
= sessiond_set_thread_pollset(&events
, 2);
732 goto error_poll_create
;
735 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
740 /* This will add the available kernel channel if any. */
741 ret
= update_kernel_poll(&events
);
745 update_poll_flag
= 0;
748 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
750 /* Poll infinite value of time */
753 ret
= lttng_poll_wait(&events
, -1);
757 * Restart interrupted system call.
759 if (errno
== EINTR
) {
763 } else if (ret
== 0) {
764 /* Should not happen since timeout is infinite */
765 ERR("Return value of poll is 0 with an infinite timeout.\n"
766 "This should not have happened! Continuing...");
772 for (i
= 0; i
< nb_fd
; i
++) {
773 /* Fetch once the poll data */
774 revents
= LTTNG_POLL_GETEV(&events
, i
);
775 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
777 health_code_update();
779 /* Thread quit pipe has been closed. Killing thread. */
780 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
786 /* Check for data on kernel pipe */
787 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
789 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
790 } while (ret
< 0 && errno
== EINTR
);
792 * Ret value is useless here, if this pipe gets any actions an
793 * update is required anyway.
795 update_poll_flag
= 1;
799 * New CPU detected by the kernel. Adding kernel stream to
800 * kernel session and updating the kernel consumer
802 if (revents
& LPOLLIN
) {
803 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
809 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
810 * and unregister kernel stream at this point.
819 lttng_poll_clean(&events
);
822 utils_close_pipe(kernel_poll_pipe
);
823 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
826 ERR("Health error occurred in %s", __func__
);
827 WARN("Kernel thread died unexpectedly. "
828 "Kernel tracing can continue but CPU hotplug is disabled.");
831 DBG("Kernel thread dying");
836 * Signal pthread condition of the consumer data that the thread.
838 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
840 pthread_mutex_lock(&data
->cond_mutex
);
843 * The state is set before signaling. It can be any value, it's the waiter
844 * job to correctly interpret this condition variable associated to the
845 * consumer pthread_cond.
847 * A value of 0 means that the corresponding thread of the consumer data
848 * was not started. 1 indicates that the thread has started and is ready
849 * for action. A negative value means that there was an error during the
852 data
->consumer_thread_is_ready
= state
;
853 (void) pthread_cond_signal(&data
->cond
);
855 pthread_mutex_unlock(&data
->cond_mutex
);
859 * This thread manage the consumer error sent back to the session daemon.
861 static void *thread_manage_consumer(void *data
)
863 int sock
= -1, i
, ret
, pollfd
, err
= -1;
864 uint32_t revents
, nb_fd
;
865 enum lttcomm_return_code code
;
866 struct lttng_poll_event events
;
867 struct consumer_data
*consumer_data
= data
;
869 DBG("[thread] Manage consumer started");
871 health_register(HEALTH_TYPE_CONSUMER
);
873 health_code_update();
876 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
877 * metadata_sock. Nothing more will be added to this poll set.
879 ret
= sessiond_set_thread_pollset(&events
, 3);
885 * The error socket here is already in a listening state which was done
886 * just before spawning this thread to avoid a race between the consumer
887 * daemon exec trying to connect and the listen() call.
889 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
894 health_code_update();
896 /* Infinite blocking call, waiting for transmission */
900 if (testpoint(thread_manage_consumer
)) {
904 ret
= lttng_poll_wait(&events
, -1);
908 * Restart interrupted system call.
910 if (errno
== EINTR
) {
918 for (i
= 0; i
< nb_fd
; i
++) {
919 /* Fetch once the poll data */
920 revents
= LTTNG_POLL_GETEV(&events
, i
);
921 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
923 health_code_update();
925 /* Thread quit pipe has been closed. Killing thread. */
926 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
932 /* Event on the registration socket */
933 if (pollfd
== consumer_data
->err_sock
) {
934 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
935 ERR("consumer err socket poll error");
941 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
947 * Set the CLOEXEC flag. Return code is useless because either way, the
950 (void) utils_set_fd_cloexec(sock
);
952 health_code_update();
954 DBG2("Receiving code from consumer err_sock");
956 /* Getting status code from kconsumerd */
957 ret
= lttcomm_recv_unix_sock(sock
, &code
,
958 sizeof(enum lttcomm_return_code
));
963 health_code_update();
965 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
966 /* Connect both socket, command and metadata. */
967 consumer_data
->cmd_sock
=
968 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
969 consumer_data
->metadata_sock
.fd
=
970 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
971 if (consumer_data
->cmd_sock
< 0 ||
972 consumer_data
->metadata_sock
.fd
< 0) {
973 PERROR("consumer connect cmd socket");
974 /* On error, signal condition and quit. */
975 signal_consumer_condition(consumer_data
, -1);
978 /* Create metadata socket lock. */
979 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
980 if (consumer_data
->metadata_sock
.lock
== NULL
) {
981 PERROR("zmalloc pthread mutex");
985 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
987 signal_consumer_condition(consumer_data
, 1);
988 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
989 DBG("Consumer metadata socket ready (fd: %d)",
990 consumer_data
->metadata_sock
.fd
);
992 ERR("consumer error when waiting for SOCK_READY : %s",
993 lttcomm_get_readable_code(-code
));
997 /* Remove the consumerd error sock since we've established a connexion */
998 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1003 /* Add new accepted error socket. */
1004 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1009 /* Add metadata socket that is successfully connected. */
1010 ret
= lttng_poll_add(&events
, consumer_data
->metadata_sock
.fd
,
1011 LPOLLIN
| LPOLLRDHUP
);
1016 health_code_update();
1018 /* Infinite blocking call, waiting for transmission */
1021 health_poll_entry();
1022 ret
= lttng_poll_wait(&events
, -1);
1026 * Restart interrupted system call.
1028 if (errno
== EINTR
) {
1036 for (i
= 0; i
< nb_fd
; i
++) {
1037 /* Fetch once the poll data */
1038 revents
= LTTNG_POLL_GETEV(&events
, i
);
1039 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1041 health_code_update();
1043 /* Thread quit pipe has been closed. Killing thread. */
1044 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1050 if (pollfd
== sock
) {
1051 /* Event on the consumerd socket */
1052 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1053 ERR("consumer err socket second poll error");
1056 health_code_update();
1057 /* Wait for any kconsumerd error */
1058 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1059 sizeof(enum lttcomm_return_code
));
1061 ERR("consumer closed the command socket");
1065 ERR("consumer return code : %s",
1066 lttcomm_get_readable_code(-code
));
1069 } else if (pollfd
== consumer_data
->metadata_sock
.fd
) {
1070 /* UST metadata requests */
1071 ret
= ust_consumer_metadata_request(
1072 &consumer_data
->metadata_sock
);
1074 ERR("Handling metadata request");
1079 ERR("Unknown pollfd");
1083 health_code_update();
1088 /* Immediately set the consumerd state to stopped */
1089 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1090 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1091 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1092 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1093 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1095 /* Code flow error... */
1099 if (consumer_data
->err_sock
>= 0) {
1100 ret
= close(consumer_data
->err_sock
);
1105 if (consumer_data
->cmd_sock
>= 0) {
1106 ret
= close(consumer_data
->cmd_sock
);
1111 if (consumer_data
->metadata_sock
.fd
>= 0) {
1112 ret
= close(consumer_data
->metadata_sock
.fd
);
1117 /* Cleanup metadata socket mutex. */
1118 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1119 free(consumer_data
->metadata_sock
.lock
);
1128 unlink(consumer_data
->err_unix_sock_path
);
1129 unlink(consumer_data
->cmd_unix_sock_path
);
1130 consumer_data
->pid
= 0;
1132 lttng_poll_clean(&events
);
1136 ERR("Health error occurred in %s", __func__
);
1138 health_unregister();
1139 DBG("consumer thread cleanup completed");
1145 * This thread manage application communication.
1147 static void *thread_manage_apps(void *data
)
1149 int i
, ret
, pollfd
, err
= -1;
1150 uint32_t revents
, nb_fd
;
1151 struct lttng_poll_event events
;
1153 DBG("[thread] Manage application started");
1155 rcu_register_thread();
1156 rcu_thread_online();
1158 health_register(HEALTH_TYPE_APP_MANAGE
);
1160 if (testpoint(thread_manage_apps
)) {
1161 goto error_testpoint
;
1164 health_code_update();
1166 ret
= sessiond_set_thread_pollset(&events
, 2);
1168 goto error_poll_create
;
1171 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1176 if (testpoint(thread_manage_apps_before_loop
)) {
1180 health_code_update();
1183 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1185 /* Inifinite blocking call, waiting for transmission */
1187 health_poll_entry();
1188 ret
= lttng_poll_wait(&events
, -1);
1192 * Restart interrupted system call.
1194 if (errno
== EINTR
) {
1202 for (i
= 0; i
< nb_fd
; i
++) {
1203 /* Fetch once the poll data */
1204 revents
= LTTNG_POLL_GETEV(&events
, i
);
1205 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1207 health_code_update();
1209 /* Thread quit pipe has been closed. Killing thread. */
1210 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1216 /* Inspect the apps cmd pipe */
1217 if (pollfd
== apps_cmd_pipe
[0]) {
1218 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1219 ERR("Apps command pipe error");
1221 } else if (revents
& LPOLLIN
) {
1226 ret
= read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1227 } while (ret
< 0 && errno
== EINTR
);
1228 if (ret
< 0 || ret
< sizeof(sock
)) {
1229 PERROR("read apps cmd pipe");
1233 health_code_update();
1236 * We only monitor the error events of the socket. This
1237 * thread does not handle any incoming data from UST
1240 ret
= lttng_poll_add(&events
, sock
,
1241 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1246 /* Set socket timeout for both receiving and ending */
1247 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1248 app_socket_timeout
);
1249 (void) lttcomm_setsockopt_snd_timeout(sock
,
1250 app_socket_timeout
);
1252 DBG("Apps with sock %d added to poll set", sock
);
1254 health_code_update();
1260 * At this point, we know that a registered application made
1261 * the event at poll_wait.
1263 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1264 /* Removing from the poll set */
1265 ret
= lttng_poll_del(&events
, pollfd
);
1270 /* Socket closed on remote end. */
1271 ust_app_unregister(pollfd
);
1276 health_code_update();
1282 lttng_poll_clean(&events
);
1285 utils_close_pipe(apps_cmd_pipe
);
1286 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1289 * We don't clean the UST app hash table here since already registered
1290 * applications can still be controlled so let them be until the session
1291 * daemon dies or the applications stop.
1296 ERR("Health error occurred in %s", __func__
);
1298 health_unregister();
1299 DBG("Application communication apps thread cleanup complete");
1300 rcu_thread_offline();
1301 rcu_unregister_thread();
1306 * Send a socket to a thread This is called from the dispatch UST registration
1307 * thread once all sockets are set for the application.
1309 * On success, return 0 else a negative value being the errno message of the
1312 static int send_socket_to_thread(int fd
, int sock
)
1316 /* Sockets MUST be set or else this should not have been called. */
1321 ret
= write(fd
, &sock
, sizeof(sock
));
1322 } while (ret
< 0 && errno
== EINTR
);
1323 if (ret
< 0 || ret
!= sizeof(sock
)) {
1324 PERROR("write apps pipe %d", fd
);
1331 /* All good. Don't send back the write positive ret value. */
1338 * Sanitize the wait queue of the dispatch registration thread meaning removing
1339 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1340 * notify socket is never received.
1342 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1344 int ret
, nb_fd
= 0, i
;
1345 unsigned int fd_added
= 0;
1346 struct lttng_poll_event events
;
1347 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1351 lttng_poll_init(&events
);
1353 /* Just skip everything for an empty queue. */
1354 if (!wait_queue
->count
) {
1358 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1363 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1364 &wait_queue
->head
, head
) {
1365 assert(wait_node
->app
);
1366 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1367 LPOLLHUP
| LPOLLERR
);
1380 * Poll but don't block so we can quickly identify the faulty events and
1381 * clean them afterwards from the wait queue.
1383 ret
= lttng_poll_wait(&events
, 0);
1389 for (i
= 0; i
< nb_fd
; i
++) {
1390 /* Get faulty FD. */
1391 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1392 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1394 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1395 &wait_queue
->head
, head
) {
1396 if (pollfd
== wait_node
->app
->sock
&&
1397 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1398 cds_list_del(&wait_node
->head
);
1399 wait_queue
->count
--;
1400 ust_app_destroy(wait_node
->app
);
1408 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1412 lttng_poll_clean(&events
);
1416 lttng_poll_clean(&events
);
1418 ERR("Unable to sanitize wait queue");
1423 * Dispatch request from the registration threads to the application
1424 * communication thread.
1426 static void *thread_dispatch_ust_registration(void *data
)
1429 struct cds_wfq_node
*node
;
1430 struct ust_command
*ust_cmd
= NULL
;
1431 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1432 struct ust_reg_wait_queue wait_queue
= {
1436 health_register(HEALTH_TYPE_APP_REG_DISPATCH
);
1438 health_code_update();
1440 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1442 DBG("[thread] Dispatch UST command started");
1444 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1445 health_code_update();
1447 /* Atomically prepare the queue futex */
1448 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1451 struct ust_app
*app
= NULL
;
1455 * Make sure we don't have node(s) that have hung up before receiving
1456 * the notify socket. This is to clean the list in order to avoid
1457 * memory leaks from notify socket that are never seen.
1459 sanitize_wait_queue(&wait_queue
);
1461 health_code_update();
1462 /* Dequeue command for registration */
1463 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1465 DBG("Woken up but nothing in the UST command queue");
1466 /* Continue thread execution */
1470 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1472 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1473 " gid:%d sock:%d name:%s (version %d.%d)",
1474 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1475 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1476 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1477 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1479 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1480 wait_node
= zmalloc(sizeof(*wait_node
));
1482 PERROR("zmalloc wait_node dispatch");
1483 ret
= close(ust_cmd
->sock
);
1485 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1487 lttng_fd_put(1, LTTNG_FD_APPS
);
1491 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1493 /* Create application object if socket is CMD. */
1494 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1496 if (!wait_node
->app
) {
1497 ret
= close(ust_cmd
->sock
);
1499 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1501 lttng_fd_put(1, LTTNG_FD_APPS
);
1507 * Add application to the wait queue so we can set the notify
1508 * socket before putting this object in the global ht.
1510 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1515 * We have to continue here since we don't have the notify
1516 * socket and the application MUST be added to the hash table
1517 * only at that moment.
1522 * Look for the application in the local wait queue and set the
1523 * notify socket if found.
1525 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1526 &wait_queue
.head
, head
) {
1527 health_code_update();
1528 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1529 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1530 cds_list_del(&wait_node
->head
);
1532 app
= wait_node
->app
;
1534 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1540 * With no application at this stage the received socket is
1541 * basically useless so close it before we free the cmd data
1542 * structure for good.
1545 ret
= close(ust_cmd
->sock
);
1547 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1549 lttng_fd_put(1, LTTNG_FD_APPS
);
1556 * @session_lock_list
1558 * Lock the global session list so from the register up to the
1559 * registration done message, no thread can see the application
1560 * and change its state.
1562 session_lock_list();
1566 * Add application to the global hash table. This needs to be
1567 * done before the update to the UST registry can locate the
1572 /* Set app version. This call will print an error if needed. */
1573 (void) ust_app_version(app
);
1575 /* Send notify socket through the notify pipe. */
1576 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1580 session_unlock_list();
1581 /* No notify thread, stop the UST tracing. */
1586 * Update newly registered application with the tracing
1587 * registry info already enabled information.
1589 update_ust_app(app
->sock
);
1592 * Don't care about return value. Let the manage apps threads
1593 * handle app unregistration upon socket close.
1595 (void) ust_app_register_done(app
->sock
);
1598 * Even if the application socket has been closed, send the app
1599 * to the thread and unregistration will take place at that
1602 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1605 session_unlock_list();
1606 /* No apps. thread, stop the UST tracing. */
1611 session_unlock_list();
1613 } while (node
!= NULL
);
1615 health_poll_entry();
1616 /* Futex wait on queue. Blocking call on futex() */
1617 futex_nto1_wait(&ust_cmd_queue
.futex
);
1620 /* Normal exit, no error */
1624 /* Clean up wait queue. */
1625 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1626 &wait_queue
.head
, head
) {
1627 cds_list_del(&wait_node
->head
);
1632 DBG("Dispatch thread dying");
1635 ERR("Health error occurred in %s", __func__
);
1637 health_unregister();
1642 * This thread manage application registration.
1644 static void *thread_registration_apps(void *data
)
1646 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1647 uint32_t revents
, nb_fd
;
1648 struct lttng_poll_event events
;
1650 * Get allocated in this thread, enqueued to a global queue, dequeued and
1651 * freed in the manage apps thread.
1653 struct ust_command
*ust_cmd
= NULL
;
1655 DBG("[thread] Manage application registration started");
1657 health_register(HEALTH_TYPE_APP_REG
);
1659 if (testpoint(thread_registration_apps
)) {
1660 goto error_testpoint
;
1663 ret
= lttcomm_listen_unix_sock(apps_sock
);
1669 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1670 * more will be added to this poll set.
1672 ret
= sessiond_set_thread_pollset(&events
, 2);
1674 goto error_create_poll
;
1677 /* Add the application registration socket */
1678 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1680 goto error_poll_add
;
1683 /* Notify all applications to register */
1684 ret
= notify_ust_apps(1);
1686 ERR("Failed to notify applications or create the wait shared memory.\n"
1687 "Execution continues but there might be problem for already\n"
1688 "running applications that wishes to register.");
1692 DBG("Accepting application registration");
1694 /* Inifinite blocking call, waiting for transmission */
1696 health_poll_entry();
1697 ret
= lttng_poll_wait(&events
, -1);
1701 * Restart interrupted system call.
1703 if (errno
== EINTR
) {
1711 for (i
= 0; i
< nb_fd
; i
++) {
1712 health_code_update();
1714 /* Fetch once the poll data */
1715 revents
= LTTNG_POLL_GETEV(&events
, i
);
1716 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1718 /* Thread quit pipe has been closed. Killing thread. */
1719 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1725 /* Event on the registration socket */
1726 if (pollfd
== apps_sock
) {
1727 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1728 ERR("Register apps socket poll error");
1730 } else if (revents
& LPOLLIN
) {
1731 sock
= lttcomm_accept_unix_sock(apps_sock
);
1737 * Set the CLOEXEC flag. Return code is useless because
1738 * either way, the show must go on.
1740 (void) utils_set_fd_cloexec(sock
);
1742 /* Create UST registration command for enqueuing */
1743 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1744 if (ust_cmd
== NULL
) {
1745 PERROR("ust command zmalloc");
1750 * Using message-based transmissions to ensure we don't
1751 * have to deal with partially received messages.
1753 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1755 ERR("Exhausted file descriptors allowed for applications.");
1765 health_code_update();
1766 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1769 /* Close socket of the application. */
1774 lttng_fd_put(LTTNG_FD_APPS
, 1);
1778 health_code_update();
1780 ust_cmd
->sock
= sock
;
1783 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1784 " gid:%d sock:%d name:%s (version %d.%d)",
1785 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1786 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1787 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1788 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1791 * Lock free enqueue the registration request. The red pill
1792 * has been taken! This apps will be part of the *system*.
1794 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1797 * Wake the registration queue futex. Implicit memory
1798 * barrier with the exchange in cds_wfq_enqueue.
1800 futex_nto1_wake(&ust_cmd_queue
.futex
);
1810 ERR("Health error occurred in %s", __func__
);
1813 /* Notify that the registration thread is gone */
1816 if (apps_sock
>= 0) {
1817 ret
= close(apps_sock
);
1827 lttng_fd_put(LTTNG_FD_APPS
, 1);
1829 unlink(apps_unix_sock_path
);
1832 lttng_poll_clean(&events
);
1836 DBG("UST Registration thread cleanup complete");
1837 health_unregister();
1843 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1844 * exec or it will fails.
1846 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1849 struct timespec timeout
;
1851 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1852 consumer_data
->consumer_thread_is_ready
= 0;
1854 /* Setup pthread condition */
1855 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1858 PERROR("pthread_condattr_init consumer data");
1863 * Set the monotonic clock in order to make sure we DO NOT jump in time
1864 * between the clock_gettime() call and the timedwait call. See bug #324
1865 * for a more details and how we noticed it.
1867 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1870 PERROR("pthread_condattr_setclock consumer data");
1874 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1877 PERROR("pthread_cond_init consumer data");
1881 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1884 PERROR("pthread_create consumer");
1889 /* We are about to wait on a pthread condition */
1890 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1892 /* Get time for sem_timedwait absolute timeout */
1893 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1895 * Set the timeout for the condition timed wait even if the clock gettime
1896 * call fails since we might loop on that call and we want to avoid to
1897 * increment the timeout too many times.
1899 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1902 * The following loop COULD be skipped in some conditions so this is why we
1903 * set ret to 0 in order to make sure at least one round of the loop is
1909 * Loop until the condition is reached or when a timeout is reached. Note
1910 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1911 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1912 * possible. This loop does not take any chances and works with both of
1915 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1916 if (clock_ret
< 0) {
1917 PERROR("clock_gettime spawn consumer");
1918 /* Infinite wait for the consumerd thread to be ready */
1919 ret
= pthread_cond_wait(&consumer_data
->cond
,
1920 &consumer_data
->cond_mutex
);
1922 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1923 &consumer_data
->cond_mutex
, &timeout
);
1927 /* Release the pthread condition */
1928 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1932 if (ret
== ETIMEDOUT
) {
1934 * Call has timed out so we kill the kconsumerd_thread and return
1937 ERR("Condition timed out. The consumer thread was never ready."
1939 ret
= pthread_cancel(consumer_data
->thread
);
1941 PERROR("pthread_cancel consumer thread");
1944 PERROR("pthread_cond_wait failed consumer thread");
1949 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1950 if (consumer_data
->pid
== 0) {
1951 ERR("Consumerd did not start");
1952 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1955 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1964 * Join consumer thread
1966 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1970 /* Consumer pid must be a real one. */
1971 if (consumer_data
->pid
> 0) {
1973 ret
= kill(consumer_data
->pid
, SIGTERM
);
1975 ERR("Error killing consumer daemon");
1978 return pthread_join(consumer_data
->thread
, &status
);
1985 * Fork and exec a consumer daemon (consumerd).
1987 * Return pid if successful else -1.
1989 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1993 const char *consumer_to_use
;
1994 const char *verbosity
;
1997 DBG("Spawning consumerd");
2004 if (opt_verbose_consumer
) {
2005 verbosity
= "--verbose";
2007 verbosity
= "--quiet";
2009 switch (consumer_data
->type
) {
2010 case LTTNG_CONSUMER_KERNEL
:
2012 * Find out which consumerd to execute. We will first try the
2013 * 64-bit path, then the sessiond's installation directory, and
2014 * fallback on the 32-bit one,
2016 DBG3("Looking for a kernel consumer at these locations:");
2017 DBG3(" 1) %s", consumerd64_bin
);
2018 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2019 DBG3(" 3) %s", consumerd32_bin
);
2020 if (stat(consumerd64_bin
, &st
) == 0) {
2021 DBG3("Found location #1");
2022 consumer_to_use
= consumerd64_bin
;
2023 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2024 DBG3("Found location #2");
2025 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2026 } else if (stat(consumerd32_bin
, &st
) == 0) {
2027 DBG3("Found location #3");
2028 consumer_to_use
= consumerd32_bin
;
2030 DBG("Could not find any valid consumerd executable");
2033 DBG("Using kernel consumer at: %s", consumer_to_use
);
2034 execl(consumer_to_use
,
2035 "lttng-consumerd", verbosity
, "-k",
2036 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2037 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2040 case LTTNG_CONSUMER64_UST
:
2042 char *tmpnew
= NULL
;
2044 if (consumerd64_libdir
[0] != '\0') {
2048 tmp
= getenv("LD_LIBRARY_PATH");
2052 tmplen
= strlen("LD_LIBRARY_PATH=")
2053 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2054 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2059 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2060 strcat(tmpnew
, consumerd64_libdir
);
2061 if (tmp
[0] != '\0') {
2062 strcat(tmpnew
, ":");
2063 strcat(tmpnew
, tmp
);
2065 ret
= putenv(tmpnew
);
2072 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2073 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2074 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2075 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2077 if (consumerd64_libdir
[0] != '\0') {
2085 case LTTNG_CONSUMER32_UST
:
2087 char *tmpnew
= NULL
;
2089 if (consumerd32_libdir
[0] != '\0') {
2093 tmp
= getenv("LD_LIBRARY_PATH");
2097 tmplen
= strlen("LD_LIBRARY_PATH=")
2098 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2099 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2104 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2105 strcat(tmpnew
, consumerd32_libdir
);
2106 if (tmp
[0] != '\0') {
2107 strcat(tmpnew
, ":");
2108 strcat(tmpnew
, tmp
);
2110 ret
= putenv(tmpnew
);
2117 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2118 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2119 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2120 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2122 if (consumerd32_libdir
[0] != '\0') {
2131 PERROR("unknown consumer type");
2135 PERROR("kernel start consumer exec");
2138 } else if (pid
> 0) {
2141 PERROR("start consumer fork");
2149 * Spawn the consumerd daemon and session daemon thread.
2151 static int start_consumerd(struct consumer_data
*consumer_data
)
2156 * Set the listen() state on the socket since there is a possible race
2157 * between the exec() of the consumer daemon and this call if place in the
2158 * consumer thread. See bug #366 for more details.
2160 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2165 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2166 if (consumer_data
->pid
!= 0) {
2167 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2171 ret
= spawn_consumerd(consumer_data
);
2173 ERR("Spawning consumerd failed");
2174 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2178 /* Setting up the consumer_data pid */
2179 consumer_data
->pid
= ret
;
2180 DBG2("Consumer pid %d", consumer_data
->pid
);
2181 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2183 DBG2("Spawning consumer control thread");
2184 ret
= spawn_consumer_thread(consumer_data
);
2186 ERR("Fatal error spawning consumer control thread");
2194 /* Cleanup already created sockets on error. */
2195 if (consumer_data
->err_sock
>= 0) {
2198 err
= close(consumer_data
->err_sock
);
2200 PERROR("close consumer data error socket");
2207 * Compute health status of each consumer. If one of them is zero (bad
2208 * state), we return 0.
2210 static int check_consumer_health(void)
2214 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2216 DBG3("Health consumer check %d", ret
);
2222 * Setup necessary data for kernel tracer action.
2224 static int init_kernel_tracer(void)
2228 /* Modprobe lttng kernel modules */
2229 ret
= modprobe_lttng_control();
2234 /* Open debugfs lttng */
2235 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2236 if (kernel_tracer_fd
< 0) {
2237 DBG("Failed to open %s", module_proc_lttng
);
2242 /* Validate kernel version */
2243 ret
= kernel_validate_version(kernel_tracer_fd
);
2248 ret
= modprobe_lttng_data();
2253 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2257 modprobe_remove_lttng_control();
2258 ret
= close(kernel_tracer_fd
);
2262 kernel_tracer_fd
= -1;
2263 return LTTNG_ERR_KERN_VERSION
;
2266 ret
= close(kernel_tracer_fd
);
2272 modprobe_remove_lttng_control();
2275 WARN("No kernel tracer available");
2276 kernel_tracer_fd
= -1;
2278 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2280 return LTTNG_ERR_KERN_NA
;
2286 * Copy consumer output from the tracing session to the domain session. The
2287 * function also applies the right modification on a per domain basis for the
2288 * trace files destination directory.
2290 * Should *NOT* be called with RCU read-side lock held.
2292 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2295 const char *dir_name
;
2296 struct consumer_output
*consumer
;
2299 assert(session
->consumer
);
2302 case LTTNG_DOMAIN_KERNEL
:
2303 DBG3("Copying tracing session consumer output in kernel session");
2305 * XXX: We should audit the session creation and what this function
2306 * does "extra" in order to avoid a destroy since this function is used
2307 * in the domain session creation (kernel and ust) only. Same for UST
2310 if (session
->kernel_session
->consumer
) {
2311 consumer_destroy_output(session
->kernel_session
->consumer
);
2313 session
->kernel_session
->consumer
=
2314 consumer_copy_output(session
->consumer
);
2315 /* Ease our life a bit for the next part */
2316 consumer
= session
->kernel_session
->consumer
;
2317 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2319 case LTTNG_DOMAIN_UST
:
2320 DBG3("Copying tracing session consumer output in UST session");
2321 if (session
->ust_session
->consumer
) {
2322 consumer_destroy_output(session
->ust_session
->consumer
);
2324 session
->ust_session
->consumer
=
2325 consumer_copy_output(session
->consumer
);
2326 /* Ease our life a bit for the next part */
2327 consumer
= session
->ust_session
->consumer
;
2328 dir_name
= DEFAULT_UST_TRACE_DIR
;
2331 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2335 /* Append correct directory to subdir */
2336 strncat(consumer
->subdir
, dir_name
,
2337 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2338 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2347 * Create an UST session and add it to the session ust list.
2349 * Should *NOT* be called with RCU read-side lock held.
2351 static int create_ust_session(struct ltt_session
*session
,
2352 struct lttng_domain
*domain
)
2355 struct ltt_ust_session
*lus
= NULL
;
2359 assert(session
->consumer
);
2361 switch (domain
->type
) {
2362 case LTTNG_DOMAIN_UST
:
2365 ERR("Unknown UST domain on create session %d", domain
->type
);
2366 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2370 DBG("Creating UST session");
2372 lus
= trace_ust_create_session(session
->id
);
2374 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2378 lus
->uid
= session
->uid
;
2379 lus
->gid
= session
->gid
;
2380 session
->ust_session
= lus
;
2382 /* Copy session output to the newly created UST session */
2383 ret
= copy_session_consumer(domain
->type
, session
);
2384 if (ret
!= LTTNG_OK
) {
2392 session
->ust_session
= NULL
;
2397 * Create a kernel tracer session then create the default channel.
2399 static int create_kernel_session(struct ltt_session
*session
)
2403 DBG("Creating kernel session");
2405 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2407 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2411 /* Code flow safety */
2412 assert(session
->kernel_session
);
2414 /* Copy session output to the newly created Kernel session */
2415 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2416 if (ret
!= LTTNG_OK
) {
2420 /* Create directory(ies) on local filesystem. */
2421 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2422 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2423 ret
= run_as_mkdir_recursive(
2424 session
->kernel_session
->consumer
->dst
.trace_path
,
2425 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2427 if (ret
!= -EEXIST
) {
2428 ERR("Trace directory creation error");
2434 session
->kernel_session
->uid
= session
->uid
;
2435 session
->kernel_session
->gid
= session
->gid
;
2440 trace_kernel_destroy_session(session
->kernel_session
);
2441 session
->kernel_session
= NULL
;
2446 * Count number of session permitted by uid/gid.
2448 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2451 struct ltt_session
*session
;
2453 DBG("Counting number of available session for UID %d GID %d",
2455 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2457 * Only list the sessions the user can control.
2459 if (!session_access_ok(session
, uid
, gid
)) {
2468 * Process the command requested by the lttng client within the command
2469 * context structure. This function make sure that the return structure (llm)
2470 * is set and ready for transmission before returning.
2472 * Return any error encountered or 0 for success.
2474 * "sock" is only used for special-case var. len data.
2476 * Should *NOT* be called with RCU read-side lock held.
2478 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2482 int need_tracing_session
= 1;
2485 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2489 switch (cmd_ctx
->lsm
->cmd_type
) {
2490 case LTTNG_CREATE_SESSION
:
2491 case LTTNG_DESTROY_SESSION
:
2492 case LTTNG_LIST_SESSIONS
:
2493 case LTTNG_LIST_DOMAINS
:
2494 case LTTNG_START_TRACE
:
2495 case LTTNG_STOP_TRACE
:
2496 case LTTNG_DATA_PENDING
:
2503 if (opt_no_kernel
&& need_domain
2504 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2506 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2508 ret
= LTTNG_ERR_KERN_NA
;
2513 /* Deny register consumer if we already have a spawned consumer. */
2514 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2515 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2516 if (kconsumer_data
.pid
> 0) {
2517 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2518 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2521 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2525 * Check for command that don't needs to allocate a returned payload. We do
2526 * this here so we don't have to make the call for no payload at each
2529 switch(cmd_ctx
->lsm
->cmd_type
) {
2530 case LTTNG_LIST_SESSIONS
:
2531 case LTTNG_LIST_TRACEPOINTS
:
2532 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2533 case LTTNG_LIST_DOMAINS
:
2534 case LTTNG_LIST_CHANNELS
:
2535 case LTTNG_LIST_EVENTS
:
2538 /* Setup lttng message with no payload */
2539 ret
= setup_lttng_msg(cmd_ctx
, 0);
2541 /* This label does not try to unlock the session */
2542 goto init_setup_error
;
2546 /* Commands that DO NOT need a session. */
2547 switch (cmd_ctx
->lsm
->cmd_type
) {
2548 case LTTNG_CREATE_SESSION
:
2549 case LTTNG_CALIBRATE
:
2550 case LTTNG_LIST_SESSIONS
:
2551 case LTTNG_LIST_TRACEPOINTS
:
2552 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2553 need_tracing_session
= 0;
2556 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2558 * We keep the session list lock across _all_ commands
2559 * for now, because the per-session lock does not
2560 * handle teardown properly.
2562 session_lock_list();
2563 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2564 if (cmd_ctx
->session
== NULL
) {
2565 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2568 /* Acquire lock for the session */
2569 session_lock(cmd_ctx
->session
);
2579 * Check domain type for specific "pre-action".
2581 switch (cmd_ctx
->lsm
->domain
.type
) {
2582 case LTTNG_DOMAIN_KERNEL
:
2584 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2588 /* Kernel tracer check */
2589 if (kernel_tracer_fd
== -1) {
2590 /* Basically, load kernel tracer modules */
2591 ret
= init_kernel_tracer();
2597 /* Consumer is in an ERROR state. Report back to client */
2598 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2599 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2603 /* Need a session for kernel command */
2604 if (need_tracing_session
) {
2605 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2606 ret
= create_kernel_session(cmd_ctx
->session
);
2608 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2613 /* Start the kernel consumer daemon */
2614 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2615 if (kconsumer_data
.pid
== 0 &&
2616 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2617 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2618 ret
= start_consumerd(&kconsumer_data
);
2620 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2623 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2625 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2629 * The consumer was just spawned so we need to add the socket to
2630 * the consumer output of the session if exist.
2632 ret
= consumer_create_socket(&kconsumer_data
,
2633 cmd_ctx
->session
->kernel_session
->consumer
);
2640 case LTTNG_DOMAIN_UST
:
2642 /* Consumer is in an ERROR state. Report back to client */
2643 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2644 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2648 if (need_tracing_session
) {
2649 /* Create UST session if none exist. */
2650 if (cmd_ctx
->session
->ust_session
== NULL
) {
2651 ret
= create_ust_session(cmd_ctx
->session
,
2652 &cmd_ctx
->lsm
->domain
);
2653 if (ret
!= LTTNG_OK
) {
2658 /* Start the UST consumer daemons */
2660 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2661 if (consumerd64_bin
[0] != '\0' &&
2662 ustconsumer64_data
.pid
== 0 &&
2663 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2664 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2665 ret
= start_consumerd(&ustconsumer64_data
);
2667 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2668 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2672 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2673 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2675 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2679 * Setup socket for consumer 64 bit. No need for atomic access
2680 * since it was set above and can ONLY be set in this thread.
2682 ret
= consumer_create_socket(&ustconsumer64_data
,
2683 cmd_ctx
->session
->ust_session
->consumer
);
2689 if (consumerd32_bin
[0] != '\0' &&
2690 ustconsumer32_data
.pid
== 0 &&
2691 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2692 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2693 ret
= start_consumerd(&ustconsumer32_data
);
2695 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2696 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2700 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2701 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2703 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2707 * Setup socket for consumer 64 bit. No need for atomic access
2708 * since it was set above and can ONLY be set in this thread.
2710 ret
= consumer_create_socket(&ustconsumer32_data
,
2711 cmd_ctx
->session
->ust_session
->consumer
);
2723 /* Validate consumer daemon state when start/stop trace command */
2724 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2725 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2726 switch (cmd_ctx
->lsm
->domain
.type
) {
2727 case LTTNG_DOMAIN_UST
:
2728 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2729 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2733 case LTTNG_DOMAIN_KERNEL
:
2734 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2735 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2743 * Check that the UID or GID match that of the tracing session.
2744 * The root user can interact with all sessions.
2746 if (need_tracing_session
) {
2747 if (!session_access_ok(cmd_ctx
->session
,
2748 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2749 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2750 ret
= LTTNG_ERR_EPERM
;
2756 * Send relayd information to consumer as soon as we have a domain and a
2759 if (cmd_ctx
->session
&& need_domain
) {
2761 * Setup relayd if not done yet. If the relayd information was already
2762 * sent to the consumer, this call will gracefully return.
2764 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2765 if (ret
!= LTTNG_OK
) {
2770 /* Process by command type */
2771 switch (cmd_ctx
->lsm
->cmd_type
) {
2772 case LTTNG_ADD_CONTEXT
:
2774 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2775 cmd_ctx
->lsm
->u
.context
.channel_name
,
2776 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2779 case LTTNG_DISABLE_CHANNEL
:
2781 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2782 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2785 case LTTNG_DISABLE_EVENT
:
2787 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2788 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2789 cmd_ctx
->lsm
->u
.disable
.name
);
2792 case LTTNG_DISABLE_ALL_EVENT
:
2794 DBG("Disabling all events");
2796 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2797 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2800 case LTTNG_ENABLE_CHANNEL
:
2802 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2803 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2806 case LTTNG_ENABLE_EVENT
:
2808 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2809 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2810 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2813 case LTTNG_ENABLE_ALL_EVENT
:
2815 DBG("Enabling all events");
2817 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2818 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2819 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2822 case LTTNG_LIST_TRACEPOINTS
:
2824 struct lttng_event
*events
;
2827 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2828 if (nb_events
< 0) {
2829 /* Return value is a negative lttng_error_code. */
2835 * Setup lttng message with payload size set to the event list size in
2836 * bytes and then copy list into the llm payload.
2838 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2844 /* Copy event list into message payload */
2845 memcpy(cmd_ctx
->llm
->payload
, events
,
2846 sizeof(struct lttng_event
) * nb_events
);
2853 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2855 struct lttng_event_field
*fields
;
2858 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2860 if (nb_fields
< 0) {
2861 /* Return value is a negative lttng_error_code. */
2867 * Setup lttng message with payload size set to the event list size in
2868 * bytes and then copy list into the llm payload.
2870 ret
= setup_lttng_msg(cmd_ctx
,
2871 sizeof(struct lttng_event_field
) * nb_fields
);
2877 /* Copy event list into message payload */
2878 memcpy(cmd_ctx
->llm
->payload
, fields
,
2879 sizeof(struct lttng_event_field
) * nb_fields
);
2886 case LTTNG_SET_CONSUMER_URI
:
2889 struct lttng_uri
*uris
;
2891 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2892 len
= nb_uri
* sizeof(struct lttng_uri
);
2895 ret
= LTTNG_ERR_INVALID
;
2899 uris
= zmalloc(len
);
2901 ret
= LTTNG_ERR_FATAL
;
2905 /* Receive variable len data */
2906 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2907 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2909 DBG("No URIs received from client... continuing");
2911 ret
= LTTNG_ERR_SESSION_FAIL
;
2916 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2918 if (ret
!= LTTNG_OK
) {
2924 * XXX: 0 means that this URI should be applied on the session. Should
2925 * be a DOMAIN enuam.
2927 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2928 /* Add the URI for the UST session if a consumer is present. */
2929 if (cmd_ctx
->session
->ust_session
&&
2930 cmd_ctx
->session
->ust_session
->consumer
) {
2931 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2933 } else if (cmd_ctx
->session
->kernel_session
&&
2934 cmd_ctx
->session
->kernel_session
->consumer
) {
2935 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2936 cmd_ctx
->session
, nb_uri
, uris
);
2944 case LTTNG_START_TRACE
:
2946 ret
= cmd_start_trace(cmd_ctx
->session
);
2949 case LTTNG_STOP_TRACE
:
2951 ret
= cmd_stop_trace(cmd_ctx
->session
);
2954 case LTTNG_CREATE_SESSION
:
2957 struct lttng_uri
*uris
= NULL
;
2959 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2960 len
= nb_uri
* sizeof(struct lttng_uri
);
2963 uris
= zmalloc(len
);
2965 ret
= LTTNG_ERR_FATAL
;
2969 /* Receive variable len data */
2970 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2971 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2973 DBG("No URIs received from client... continuing");
2975 ret
= LTTNG_ERR_SESSION_FAIL
;
2980 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2981 DBG("Creating session with ONE network URI is a bad call");
2982 ret
= LTTNG_ERR_SESSION_FAIL
;
2988 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2995 case LTTNG_DESTROY_SESSION
:
2997 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2999 /* Set session to NULL so we do not unlock it after free. */
3000 cmd_ctx
->session
= NULL
;
3003 case LTTNG_LIST_DOMAINS
:
3006 struct lttng_domain
*domains
;
3008 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3010 /* Return value is a negative lttng_error_code. */
3015 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3021 /* Copy event list into message payload */
3022 memcpy(cmd_ctx
->llm
->payload
, domains
,
3023 nb_dom
* sizeof(struct lttng_domain
));
3030 case LTTNG_LIST_CHANNELS
:
3033 struct lttng_channel
*channels
;
3035 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3036 cmd_ctx
->session
, &channels
);
3038 /* Return value is a negative lttng_error_code. */
3043 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3049 /* Copy event list into message payload */
3050 memcpy(cmd_ctx
->llm
->payload
, channels
,
3051 nb_chan
* sizeof(struct lttng_channel
));
3058 case LTTNG_LIST_EVENTS
:
3061 struct lttng_event
*events
= NULL
;
3063 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3064 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3066 /* Return value is a negative lttng_error_code. */
3071 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3077 /* Copy event list into message payload */
3078 memcpy(cmd_ctx
->llm
->payload
, events
,
3079 nb_event
* sizeof(struct lttng_event
));
3086 case LTTNG_LIST_SESSIONS
:
3088 unsigned int nr_sessions
;
3090 session_lock_list();
3091 nr_sessions
= lttng_sessions_count(
3092 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3093 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3095 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3097 session_unlock_list();
3101 /* Filled the session array */
3102 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3103 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3104 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3106 session_unlock_list();
3111 case LTTNG_CALIBRATE
:
3113 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3114 &cmd_ctx
->lsm
->u
.calibrate
);
3117 case LTTNG_REGISTER_CONSUMER
:
3119 struct consumer_data
*cdata
;
3121 switch (cmd_ctx
->lsm
->domain
.type
) {
3122 case LTTNG_DOMAIN_KERNEL
:
3123 cdata
= &kconsumer_data
;
3126 ret
= LTTNG_ERR_UND
;
3130 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3131 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3134 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3136 struct lttng_filter_bytecode
*bytecode
;
3138 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3139 ret
= LTTNG_ERR_FILTER_INVAL
;
3142 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3143 ret
= LTTNG_ERR_FILTER_INVAL
;
3146 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3148 ret
= LTTNG_ERR_FILTER_NOMEM
;
3151 /* Receive var. len. data */
3152 DBG("Receiving var len data from client ...");
3153 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3154 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3156 DBG("Nothing recv() from client var len data... continuing");
3158 ret
= LTTNG_ERR_FILTER_INVAL
;
3162 if (bytecode
->len
+ sizeof(*bytecode
)
3163 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3165 ret
= LTTNG_ERR_FILTER_INVAL
;
3169 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3170 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3171 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3174 case LTTNG_DATA_PENDING
:
3176 ret
= cmd_data_pending(cmd_ctx
->session
);
3180 ret
= LTTNG_ERR_UND
;
3185 if (cmd_ctx
->llm
== NULL
) {
3186 DBG("Missing llm structure. Allocating one.");
3187 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3191 /* Set return code */
3192 cmd_ctx
->llm
->ret_code
= ret
;
3194 if (cmd_ctx
->session
) {
3195 session_unlock(cmd_ctx
->session
);
3197 if (need_tracing_session
) {
3198 session_unlock_list();
3205 * Thread managing health check socket.
3207 static void *thread_manage_health(void *data
)
3209 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3210 uint32_t revents
, nb_fd
;
3211 struct lttng_poll_event events
;
3212 struct lttcomm_health_msg msg
;
3213 struct lttcomm_health_data reply
;
3215 DBG("[thread] Manage health check started");
3217 rcu_register_thread();
3219 /* We might hit an error path before this is created. */
3220 lttng_poll_init(&events
);
3222 /* Create unix socket */
3223 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3225 ERR("Unable to create health check Unix socket");
3231 * Set the CLOEXEC flag. Return code is useless because either way, the
3234 (void) utils_set_fd_cloexec(sock
);
3236 ret
= lttcomm_listen_unix_sock(sock
);
3242 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3243 * more will be added to this poll set.
3245 ret
= sessiond_set_thread_pollset(&events
, 2);
3250 /* Add the application registration socket */
3251 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3257 DBG("Health check ready");
3259 /* Inifinite blocking call, waiting for transmission */
3261 ret
= lttng_poll_wait(&events
, -1);
3264 * Restart interrupted system call.
3266 if (errno
== EINTR
) {
3274 for (i
= 0; i
< nb_fd
; i
++) {
3275 /* Fetch once the poll data */
3276 revents
= LTTNG_POLL_GETEV(&events
, i
);
3277 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3279 /* Thread quit pipe has been closed. Killing thread. */
3280 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3286 /* Event on the registration socket */
3287 if (pollfd
== sock
) {
3288 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3289 ERR("Health socket poll error");
3295 new_sock
= lttcomm_accept_unix_sock(sock
);
3301 * Set the CLOEXEC flag. Return code is useless because either way, the
3304 (void) utils_set_fd_cloexec(new_sock
);
3306 DBG("Receiving data from client for health...");
3307 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3309 DBG("Nothing recv() from client... continuing");
3310 ret
= close(new_sock
);
3318 rcu_thread_online();
3320 switch (msg
.component
) {
3321 case LTTNG_HEALTH_CMD
:
3322 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3324 case LTTNG_HEALTH_APP_MANAGE
:
3325 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3327 case LTTNG_HEALTH_APP_REG
:
3328 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3330 case LTTNG_HEALTH_KERNEL
:
3331 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3333 case LTTNG_HEALTH_CONSUMER
:
3334 reply
.ret_code
= check_consumer_health();
3336 case LTTNG_HEALTH_HT_CLEANUP
:
3337 reply
.ret_code
= health_check_state(HEALTH_TYPE_HT_CLEANUP
);
3339 case LTTNG_HEALTH_APP_MANAGE_NOTIFY
:
3340 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3342 case LTTNG_HEALTH_APP_REG_DISPATCH
:
3343 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3345 case LTTNG_HEALTH_ALL
:
3347 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3348 health_check_state(HEALTH_TYPE_APP_REG
) &&
3349 health_check_state(HEALTH_TYPE_CMD
) &&
3350 health_check_state(HEALTH_TYPE_KERNEL
) &&
3351 check_consumer_health() &&
3352 health_check_state(HEALTH_TYPE_HT_CLEANUP
) &&
3353 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
) &&
3354 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3357 reply
.ret_code
= LTTNG_ERR_UND
;
3362 * Flip ret value since 0 is a success and 1 indicates a bad health for
3363 * the client where in the sessiond it is the opposite. Again, this is
3364 * just to make things easier for us poor developer which enjoy a lot
3367 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3368 reply
.ret_code
= !reply
.ret_code
;
3371 DBG2("Health check return value %d", reply
.ret_code
);
3373 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3375 ERR("Failed to send health data back to client");
3378 /* End of transmission */
3379 ret
= close(new_sock
);
3389 ERR("Health error occurred in %s", __func__
);
3391 DBG("Health check thread dying");
3392 unlink(health_unix_sock_path
);
3400 lttng_poll_clean(&events
);
3402 rcu_unregister_thread();
3407 * This thread manage all clients request using the unix client socket for
3410 static void *thread_manage_clients(void *data
)
3412 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3414 uint32_t revents
, nb_fd
;
3415 struct command_ctx
*cmd_ctx
= NULL
;
3416 struct lttng_poll_event events
;
3418 DBG("[thread] Manage client started");
3420 rcu_register_thread();
3422 health_register(HEALTH_TYPE_CMD
);
3424 if (testpoint(thread_manage_clients
)) {
3425 goto error_testpoint
;
3428 health_code_update();
3430 ret
= lttcomm_listen_unix_sock(client_sock
);
3436 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3437 * more will be added to this poll set.
3439 ret
= sessiond_set_thread_pollset(&events
, 2);
3441 goto error_create_poll
;
3444 /* Add the application registration socket */
3445 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3451 * Notify parent pid that we are ready to accept command for client side.
3453 if (opt_sig_parent
) {
3454 kill(ppid
, SIGUSR1
);
3457 if (testpoint(thread_manage_clients_before_loop
)) {
3461 health_code_update();
3464 DBG("Accepting client command ...");
3466 /* Inifinite blocking call, waiting for transmission */
3468 health_poll_entry();
3469 ret
= lttng_poll_wait(&events
, -1);
3473 * Restart interrupted system call.
3475 if (errno
== EINTR
) {
3483 for (i
= 0; i
< nb_fd
; i
++) {
3484 /* Fetch once the poll data */
3485 revents
= LTTNG_POLL_GETEV(&events
, i
);
3486 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3488 health_code_update();
3490 /* Thread quit pipe has been closed. Killing thread. */
3491 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3497 /* Event on the registration socket */
3498 if (pollfd
== client_sock
) {
3499 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3500 ERR("Client socket poll error");
3506 DBG("Wait for client response");
3508 health_code_update();
3510 sock
= lttcomm_accept_unix_sock(client_sock
);
3516 * Set the CLOEXEC flag. Return code is useless because either way, the
3519 (void) utils_set_fd_cloexec(sock
);
3521 /* Set socket option for credentials retrieval */
3522 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3527 /* Allocate context command to process the client request */
3528 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3529 if (cmd_ctx
== NULL
) {
3530 PERROR("zmalloc cmd_ctx");
3534 /* Allocate data buffer for reception */
3535 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3536 if (cmd_ctx
->lsm
== NULL
) {
3537 PERROR("zmalloc cmd_ctx->lsm");
3541 cmd_ctx
->llm
= NULL
;
3542 cmd_ctx
->session
= NULL
;
3544 health_code_update();
3547 * Data is received from the lttng client. The struct
3548 * lttcomm_session_msg (lsm) contains the command and data request of
3551 DBG("Receiving data from client ...");
3552 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3553 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3555 DBG("Nothing recv() from client... continuing");
3561 clean_command_ctx(&cmd_ctx
);
3565 health_code_update();
3567 // TODO: Validate cmd_ctx including sanity check for
3568 // security purpose.
3570 rcu_thread_online();
3572 * This function dispatch the work to the kernel or userspace tracer
3573 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3574 * informations for the client. The command context struct contains
3575 * everything this function may needs.
3577 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3578 rcu_thread_offline();
3586 * TODO: Inform client somehow of the fatal error. At
3587 * this point, ret < 0 means that a zmalloc failed
3588 * (ENOMEM). Error detected but still accept
3589 * command, unless a socket error has been
3592 clean_command_ctx(&cmd_ctx
);
3596 health_code_update();
3598 DBG("Sending response (size: %d, retcode: %s)",
3599 cmd_ctx
->lttng_msg_size
,
3600 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3601 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3603 ERR("Failed to send data back to client");
3606 /* End of transmission */
3613 clean_command_ctx(&cmd_ctx
);
3615 health_code_update();
3627 lttng_poll_clean(&events
);
3628 clean_command_ctx(&cmd_ctx
);
3633 unlink(client_unix_sock_path
);
3634 if (client_sock
>= 0) {
3635 ret
= close(client_sock
);
3643 ERR("Health error occurred in %s", __func__
);
3646 health_unregister();
3648 DBG("Client thread dying");
3650 rcu_unregister_thread();
3656 * usage function on stderr
3658 static void usage(void)
3660 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3661 fprintf(stderr
, " -h, --help Display this usage.\n");
3662 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3663 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3664 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3665 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3666 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3667 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3668 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3669 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3670 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3671 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3672 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3673 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3674 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3675 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3676 fprintf(stderr
, " -V, --version Show version number.\n");
3677 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3678 fprintf(stderr
, " -q, --quiet No output at all.\n");
3679 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3680 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3681 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3682 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3686 * daemon argument parsing
3688 static int parse_args(int argc
, char **argv
)
3692 static struct option long_options
[] = {
3693 { "client-sock", 1, 0, 'c' },
3694 { "apps-sock", 1, 0, 'a' },
3695 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3696 { "kconsumerd-err-sock", 1, 0, 'E' },
3697 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3698 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3699 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3700 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3701 { "consumerd32-path", 1, 0, 'u' },
3702 { "consumerd32-libdir", 1, 0, 'U' },
3703 { "consumerd64-path", 1, 0, 't' },
3704 { "consumerd64-libdir", 1, 0, 'T' },
3705 { "daemonize", 0, 0, 'd' },
3706 { "sig-parent", 0, 0, 'S' },
3707 { "help", 0, 0, 'h' },
3708 { "group", 1, 0, 'g' },
3709 { "version", 0, 0, 'V' },
3710 { "quiet", 0, 0, 'q' },
3711 { "verbose", 0, 0, 'v' },
3712 { "verbose-consumer", 0, 0, 'Z' },
3713 { "no-kernel", 0, 0, 'N' },
3714 { "pidfile", 1, 0, 'p' },
3719 int option_index
= 0;
3720 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3721 long_options
, &option_index
);
3728 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3730 fprintf(stderr
, " with arg %s\n", optarg
);
3734 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3737 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3743 opt_tracing_group
= optarg
;
3749 fprintf(stdout
, "%s\n", VERSION
);
3755 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3758 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3761 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3764 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3767 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3770 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3776 lttng_opt_quiet
= 1;
3779 /* Verbose level can increase using multiple -v */
3780 lttng_opt_verbose
+= 1;
3783 opt_verbose_consumer
+= 1;
3786 consumerd32_bin
= optarg
;
3789 consumerd32_libdir
= optarg
;
3792 consumerd64_bin
= optarg
;
3795 consumerd64_libdir
= optarg
;
3798 opt_pidfile
= optarg
;
3801 /* Unknown option or other error.
3802 * Error is printed by getopt, just return */
3811 * Creates the two needed socket by the daemon.
3812 * apps_sock - The communication socket for all UST apps.
3813 * client_sock - The communication of the cli tool (lttng).
3815 static int init_daemon_socket(void)
3820 old_umask
= umask(0);
3822 /* Create client tool unix socket */
3823 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3824 if (client_sock
< 0) {
3825 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3830 /* Set the cloexec flag */
3831 ret
= utils_set_fd_cloexec(client_sock
);
3833 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3834 "Continuing but note that the consumer daemon will have a "
3835 "reference to this socket on exec()", client_sock
);
3838 /* File permission MUST be 660 */
3839 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3841 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3846 /* Create the application unix socket */
3847 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3848 if (apps_sock
< 0) {
3849 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3854 /* Set the cloexec flag */
3855 ret
= utils_set_fd_cloexec(apps_sock
);
3857 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3858 "Continuing but note that the consumer daemon will have a "
3859 "reference to this socket on exec()", apps_sock
);
3862 /* File permission MUST be 666 */
3863 ret
= chmod(apps_unix_sock_path
,
3864 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3866 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3871 DBG3("Session daemon client socket %d and application socket %d created",
3872 client_sock
, apps_sock
);
3880 * Check if the global socket is available, and if a daemon is answering at the
3881 * other side. If yes, error is returned.
3883 static int check_existing_daemon(void)
3885 /* Is there anybody out there ? */
3886 if (lttng_session_daemon_alive()) {
3894 * Set the tracing group gid onto the client socket.
3896 * Race window between mkdir and chown is OK because we are going from more
3897 * permissive (root.root) to less permissive (root.tracing).
3899 static int set_permissions(char *rundir
)
3904 ret
= allowed_group();
3906 WARN("No tracing group detected");
3913 /* Set lttng run dir */
3914 ret
= chown(rundir
, 0, gid
);
3916 ERR("Unable to set group on %s", rundir
);
3920 /* Ensure tracing group can search the run dir */
3921 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3923 ERR("Unable to set permissions on %s", rundir
);
3927 /* lttng client socket path */
3928 ret
= chown(client_unix_sock_path
, 0, gid
);
3930 ERR("Unable to set group on %s", client_unix_sock_path
);
3934 /* kconsumer error socket path */
3935 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3937 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3941 /* 64-bit ustconsumer error socket path */
3942 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3944 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3948 /* 32-bit ustconsumer compat32 error socket path */
3949 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3951 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3955 DBG("All permissions are set");
3962 * Create the lttng run directory needed for all global sockets and pipe.
3964 static int create_lttng_rundir(const char *rundir
)
3968 DBG3("Creating LTTng run directory: %s", rundir
);
3970 ret
= mkdir(rundir
, S_IRWXU
);
3972 if (errno
!= EEXIST
) {
3973 ERR("Unable to create %s", rundir
);
3985 * Setup sockets and directory needed by the kconsumerd communication with the
3988 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3992 char path
[PATH_MAX
];
3994 switch (consumer_data
->type
) {
3995 case LTTNG_CONSUMER_KERNEL
:
3996 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3998 case LTTNG_CONSUMER64_UST
:
3999 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4001 case LTTNG_CONSUMER32_UST
:
4002 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4005 ERR("Consumer type unknown");
4010 DBG2("Creating consumer directory: %s", path
);
4012 ret
= mkdir(path
, S_IRWXU
);
4014 if (errno
!= EEXIST
) {
4016 ERR("Failed to create %s", path
);
4022 /* Create the kconsumerd error unix socket */
4023 consumer_data
->err_sock
=
4024 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4025 if (consumer_data
->err_sock
< 0) {
4026 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4031 /* File permission MUST be 660 */
4032 ret
= chmod(consumer_data
->err_unix_sock_path
,
4033 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4035 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4045 * Signal handler for the daemon
4047 * Simply stop all worker threads, leaving main() return gracefully after
4048 * joining all threads and calling cleanup().
4050 static void sighandler(int sig
)
4054 DBG("SIGPIPE caught");
4057 DBG("SIGINT caught");
4061 DBG("SIGTERM caught");
4070 * Setup signal handler for :
4071 * SIGINT, SIGTERM, SIGPIPE
4073 static int set_signal_handler(void)
4076 struct sigaction sa
;
4079 if ((ret
= sigemptyset(&sigset
)) < 0) {
4080 PERROR("sigemptyset");
4084 sa
.sa_handler
= sighandler
;
4085 sa
.sa_mask
= sigset
;
4087 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4088 PERROR("sigaction");
4092 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4093 PERROR("sigaction");
4097 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4098 PERROR("sigaction");
4102 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4108 * Set open files limit to unlimited. This daemon can open a large number of
4109 * file descriptors in order to consumer multiple kernel traces.
4111 static void set_ulimit(void)
4116 /* The kernel does not allowed an infinite limit for open files */
4117 lim
.rlim_cur
= 65535;
4118 lim
.rlim_max
= 65535;
4120 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4122 PERROR("failed to set open files limit");
4127 * Write pidfile using the rundir and opt_pidfile.
4129 static void write_pidfile(void)
4132 char pidfile_path
[PATH_MAX
];
4137 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4139 /* Build pidfile path from rundir and opt_pidfile. */
4140 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4141 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4143 PERROR("snprintf pidfile path");
4149 * Create pid file in rundir. Return value is of no importance. The
4150 * execution will continue even though we are not able to write the file.
4152 (void) utils_create_pid_file(getpid(), pidfile_path
);
4161 int main(int argc
, char **argv
)
4165 const char *home_path
, *env_app_timeout
;
4167 init_kernel_workarounds();
4169 rcu_register_thread();
4171 setup_consumerd_path();
4173 page_size
= sysconf(_SC_PAGESIZE
);
4174 if (page_size
< 0) {
4175 PERROR("sysconf _SC_PAGESIZE");
4176 page_size
= LONG_MAX
;
4177 WARN("Fallback page size to %ld", page_size
);
4180 /* Parse arguments */
4182 if ((ret
= parse_args(argc
, argv
)) < 0) {
4192 * child: setsid, close FD 0, 1, 2, chdir /
4193 * parent: exit (if fork is successful)
4201 * We are in the child. Make sure all other file
4202 * descriptors are closed, in case we are called with
4203 * more opened file descriptors than the standard ones.
4205 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4210 /* Create thread quit pipe */
4211 if ((ret
= init_thread_quit_pipe()) < 0) {
4215 /* Check if daemon is UID = 0 */
4216 is_root
= !getuid();
4219 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4221 /* Create global run dir with root access */
4222 ret
= create_lttng_rundir(rundir
);
4227 if (strlen(apps_unix_sock_path
) == 0) {
4228 snprintf(apps_unix_sock_path
, PATH_MAX
,
4229 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4232 if (strlen(client_unix_sock_path
) == 0) {
4233 snprintf(client_unix_sock_path
, PATH_MAX
,
4234 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4237 /* Set global SHM for ust */
4238 if (strlen(wait_shm_path
) == 0) {
4239 snprintf(wait_shm_path
, PATH_MAX
,
4240 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4243 if (strlen(health_unix_sock_path
) == 0) {
4244 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4245 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4248 /* Setup kernel consumerd path */
4249 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4250 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4251 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4252 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4254 DBG2("Kernel consumer err path: %s",
4255 kconsumer_data
.err_unix_sock_path
);
4256 DBG2("Kernel consumer cmd path: %s",
4257 kconsumer_data
.cmd_unix_sock_path
);
4259 home_path
= get_home_dir();
4260 if (home_path
== NULL
) {
4261 /* TODO: Add --socket PATH option */
4262 ERR("Can't get HOME directory for sockets creation.");
4268 * Create rundir from home path. This will create something like
4271 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4277 ret
= create_lttng_rundir(rundir
);
4282 if (strlen(apps_unix_sock_path
) == 0) {
4283 snprintf(apps_unix_sock_path
, PATH_MAX
,
4284 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4287 /* Set the cli tool unix socket path */
4288 if (strlen(client_unix_sock_path
) == 0) {
4289 snprintf(client_unix_sock_path
, PATH_MAX
,
4290 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4293 /* Set global SHM for ust */
4294 if (strlen(wait_shm_path
) == 0) {
4295 snprintf(wait_shm_path
, PATH_MAX
,
4296 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4299 /* Set health check Unix path */
4300 if (strlen(health_unix_sock_path
) == 0) {
4301 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4302 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4306 /* Set consumer initial state */
4307 kernel_consumerd_state
= CONSUMER_STOPPED
;
4308 ust_consumerd_state
= CONSUMER_STOPPED
;
4310 DBG("Client socket path %s", client_unix_sock_path
);
4311 DBG("Application socket path %s", apps_unix_sock_path
);
4312 DBG("Application wait path %s", wait_shm_path
);
4313 DBG("LTTng run directory path: %s", rundir
);
4315 /* 32 bits consumerd path setup */
4316 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4317 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4318 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4319 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4321 DBG2("UST consumer 32 bits err path: %s",
4322 ustconsumer32_data
.err_unix_sock_path
);
4323 DBG2("UST consumer 32 bits cmd path: %s",
4324 ustconsumer32_data
.cmd_unix_sock_path
);
4326 /* 64 bits consumerd path setup */
4327 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4328 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4329 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4330 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4332 DBG2("UST consumer 64 bits err path: %s",
4333 ustconsumer64_data
.err_unix_sock_path
);
4334 DBG2("UST consumer 64 bits cmd path: %s",
4335 ustconsumer64_data
.cmd_unix_sock_path
);
4338 * See if daemon already exist.
4340 if ((ret
= check_existing_daemon()) < 0) {
4341 ERR("Already running daemon.\n");
4343 * We do not goto exit because we must not cleanup()
4344 * because a daemon is already running.
4350 * Init UST app hash table. Alloc hash table before this point since
4351 * cleanup() can get called after that point.
4355 /* After this point, we can safely call cleanup() with "goto exit" */
4358 * These actions must be executed as root. We do that *after* setting up
4359 * the sockets path because we MUST make the check for another daemon using
4360 * those paths *before* trying to set the kernel consumer sockets and init
4364 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4369 /* Setup kernel tracer */
4370 if (!opt_no_kernel
) {
4371 init_kernel_tracer();
4374 /* Set ulimit for open files */
4377 /* init lttng_fd tracking must be done after set_ulimit. */
4380 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4385 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4390 if ((ret
= set_signal_handler()) < 0) {
4394 /* Setup the needed unix socket */
4395 if ((ret
= init_daemon_socket()) < 0) {
4399 /* Set credentials to socket */
4400 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4404 /* Get parent pid if -S, --sig-parent is specified. */
4405 if (opt_sig_parent
) {
4409 /* Setup the kernel pipe for waking up the kernel thread */
4410 if (is_root
&& !opt_no_kernel
) {
4411 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4416 /* Setup the thread ht_cleanup communication pipe. */
4417 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4421 /* Setup the thread apps communication pipe. */
4422 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4426 /* Setup the thread apps notify communication pipe. */
4427 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4431 /* Initialize global buffer per UID and PID registry. */
4432 buffer_reg_init_uid_registry();
4433 buffer_reg_init_pid_registry();
4435 /* Init UST command queue. */
4436 cds_wfq_init(&ust_cmd_queue
.queue
);
4439 * Get session list pointer. This pointer MUST NOT be free(). This list is
4440 * statically declared in session.c
4442 session_list_ptr
= session_get_list();
4444 /* Set up max poll set size */
4445 lttng_poll_set_max_size();
4449 /* Check for the application socket timeout env variable. */
4450 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4451 if (env_app_timeout
) {
4452 app_socket_timeout
= atoi(env_app_timeout
);
4454 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4459 /* Create thread to manage the client socket */
4460 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4461 thread_ht_cleanup
, (void *) NULL
);
4463 PERROR("pthread_create ht_cleanup");
4464 goto exit_ht_cleanup
;
4467 /* Create thread to manage the client socket */
4468 ret
= pthread_create(&health_thread
, NULL
,
4469 thread_manage_health
, (void *) NULL
);
4471 PERROR("pthread_create health");
4475 /* Create thread to manage the client socket */
4476 ret
= pthread_create(&client_thread
, NULL
,
4477 thread_manage_clients
, (void *) NULL
);
4479 PERROR("pthread_create clients");
4483 /* Create thread to dispatch registration */
4484 ret
= pthread_create(&dispatch_thread
, NULL
,
4485 thread_dispatch_ust_registration
, (void *) NULL
);
4487 PERROR("pthread_create dispatch");
4491 /* Create thread to manage application registration. */
4492 ret
= pthread_create(®_apps_thread
, NULL
,
4493 thread_registration_apps
, (void *) NULL
);
4495 PERROR("pthread_create registration");
4499 /* Create thread to manage application socket */
4500 ret
= pthread_create(&apps_thread
, NULL
,
4501 thread_manage_apps
, (void *) NULL
);
4503 PERROR("pthread_create apps");
4507 /* Create thread to manage application notify socket */
4508 ret
= pthread_create(&apps_notify_thread
, NULL
,
4509 ust_thread_manage_notify
, (void *) NULL
);
4511 PERROR("pthread_create apps");
4515 /* Don't start this thread if kernel tracing is not requested nor root */
4516 if (is_root
&& !opt_no_kernel
) {
4517 /* Create kernel thread to manage kernel event */
4518 ret
= pthread_create(&kernel_thread
, NULL
,
4519 thread_manage_kernel
, (void *) NULL
);
4521 PERROR("pthread_create kernel");
4525 ret
= pthread_join(kernel_thread
, &status
);
4527 PERROR("pthread_join");
4528 goto error
; /* join error, exit without cleanup */
4533 ret
= pthread_join(apps_thread
, &status
);
4535 PERROR("pthread_join");
4536 goto error
; /* join error, exit without cleanup */
4540 ret
= pthread_join(reg_apps_thread
, &status
);
4542 PERROR("pthread_join");
4543 goto error
; /* join error, exit without cleanup */
4547 ret
= pthread_join(dispatch_thread
, &status
);
4549 PERROR("pthread_join");
4550 goto error
; /* join error, exit without cleanup */
4554 ret
= pthread_join(client_thread
, &status
);
4556 PERROR("pthread_join");
4557 goto error
; /* join error, exit without cleanup */
4560 ret
= join_consumer_thread(&kconsumer_data
);
4562 PERROR("join_consumer");
4563 goto error
; /* join error, exit without cleanup */
4566 ret
= join_consumer_thread(&ustconsumer32_data
);
4568 PERROR("join_consumer ust32");
4569 goto error
; /* join error, exit without cleanup */
4572 ret
= join_consumer_thread(&ustconsumer64_data
);
4574 PERROR("join_consumer ust64");
4575 goto error
; /* join error, exit without cleanup */
4579 ret
= pthread_join(health_thread
, &status
);
4581 PERROR("pthread_join health thread");
4582 goto error
; /* join error, exit without cleanup */
4586 ret
= pthread_join(ht_cleanup_thread
, &status
);
4588 PERROR("pthread_join ht cleanup thread");
4589 goto error
; /* join error, exit without cleanup */
4594 * cleanup() is called when no other thread is running.
4596 rcu_thread_online();
4598 rcu_thread_offline();
4599 rcu_unregister_thread();