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 * Dispatch request from the registration threads to the application
1339 * communication thread.
1341 static void *thread_dispatch_ust_registration(void *data
)
1344 struct cds_wfq_node
*node
;
1345 struct ust_command
*ust_cmd
= NULL
;
1347 struct ust_app
*app
;
1348 struct cds_list_head head
;
1349 } *wait_node
= NULL
, *tmp_wait_node
;
1351 CDS_LIST_HEAD(wait_queue
);
1353 DBG("[thread] Dispatch UST command started");
1355 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1356 /* Atomically prepare the queue futex */
1357 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1360 struct ust_app
*app
= NULL
;
1363 /* Dequeue command for registration */
1364 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1366 DBG("Woken up but nothing in the UST command queue");
1367 /* Continue thread execution */
1371 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1373 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1374 " gid:%d sock:%d name:%s (version %d.%d)",
1375 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1376 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1377 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1378 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1380 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1381 wait_node
= zmalloc(sizeof(*wait_node
));
1383 PERROR("zmalloc wait_node dispatch");
1384 ret
= close(ust_cmd
->sock
);
1386 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1388 lttng_fd_put(1, LTTNG_FD_APPS
);
1392 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1394 /* Create application object if socket is CMD. */
1395 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1397 if (!wait_node
->app
) {
1398 ret
= close(ust_cmd
->sock
);
1400 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1402 lttng_fd_put(1, LTTNG_FD_APPS
);
1408 * Add application to the wait queue so we can set the notify
1409 * socket before putting this object in the global ht.
1411 cds_list_add(&wait_node
->head
, &wait_queue
);
1415 * We have to continue here since we don't have the notify
1416 * socket and the application MUST be added to the hash table
1417 * only at that moment.
1422 * Look for the application in the local wait queue and set the
1423 * notify socket if found.
1425 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1426 &wait_queue
, head
) {
1427 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1428 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1429 cds_list_del(&wait_node
->head
);
1430 app
= wait_node
->app
;
1432 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1438 * With no application at this stage the received socket is
1439 * basically useless so close it before we free the cmd data
1440 * structure for good.
1443 ret
= close(ust_cmd
->sock
);
1445 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1447 lttng_fd_put(1, LTTNG_FD_APPS
);
1454 * @session_lock_list
1456 * Lock the global session list so from the register up to the
1457 * registration done message, no thread can see the application
1458 * and change its state.
1460 session_lock_list();
1464 * Add application to the global hash table. This needs to be
1465 * done before the update to the UST registry can locate the
1470 /* Set app version. This call will print an error if needed. */
1471 (void) ust_app_version(app
);
1473 /* Send notify socket through the notify pipe. */
1474 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1478 session_unlock_list();
1479 /* No notify thread, stop the UST tracing. */
1484 * Update newly registered application with the tracing
1485 * registry info already enabled information.
1487 update_ust_app(app
->sock
);
1490 * Don't care about return value. Let the manage apps threads
1491 * handle app unregistration upon socket close.
1493 (void) ust_app_register_done(app
->sock
);
1496 * Even if the application socket has been closed, send the app
1497 * to the thread and unregistration will take place at that
1500 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1503 session_unlock_list();
1504 /* No apps. thread, stop the UST tracing. */
1509 session_unlock_list();
1511 } while (node
!= NULL
);
1513 /* Futex wait on queue. Blocking call on futex() */
1514 futex_nto1_wait(&ust_cmd_queue
.futex
);
1518 /* Clean up wait queue. */
1519 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1520 &wait_queue
, head
) {
1521 cds_list_del(&wait_node
->head
);
1525 DBG("Dispatch thread dying");
1530 * This thread manage application registration.
1532 static void *thread_registration_apps(void *data
)
1534 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1535 uint32_t revents
, nb_fd
;
1536 struct lttng_poll_event events
;
1538 * Get allocated in this thread, enqueued to a global queue, dequeued and
1539 * freed in the manage apps thread.
1541 struct ust_command
*ust_cmd
= NULL
;
1543 DBG("[thread] Manage application registration started");
1545 health_register(HEALTH_TYPE_APP_REG
);
1547 if (testpoint(thread_registration_apps
)) {
1548 goto error_testpoint
;
1551 ret
= lttcomm_listen_unix_sock(apps_sock
);
1557 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1558 * more will be added to this poll set.
1560 ret
= sessiond_set_thread_pollset(&events
, 2);
1562 goto error_create_poll
;
1565 /* Add the application registration socket */
1566 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1568 goto error_poll_add
;
1571 /* Notify all applications to register */
1572 ret
= notify_ust_apps(1);
1574 ERR("Failed to notify applications or create the wait shared memory.\n"
1575 "Execution continues but there might be problem for already\n"
1576 "running applications that wishes to register.");
1580 DBG("Accepting application registration");
1582 /* Inifinite blocking call, waiting for transmission */
1584 health_poll_entry();
1585 ret
= lttng_poll_wait(&events
, -1);
1589 * Restart interrupted system call.
1591 if (errno
== EINTR
) {
1599 for (i
= 0; i
< nb_fd
; i
++) {
1600 health_code_update();
1602 /* Fetch once the poll data */
1603 revents
= LTTNG_POLL_GETEV(&events
, i
);
1604 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1606 /* Thread quit pipe has been closed. Killing thread. */
1607 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1613 /* Event on the registration socket */
1614 if (pollfd
== apps_sock
) {
1615 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1616 ERR("Register apps socket poll error");
1618 } else if (revents
& LPOLLIN
) {
1619 sock
= lttcomm_accept_unix_sock(apps_sock
);
1625 * Set the CLOEXEC flag. Return code is useless because
1626 * either way, the show must go on.
1628 (void) utils_set_fd_cloexec(sock
);
1630 /* Create UST registration command for enqueuing */
1631 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1632 if (ust_cmd
== NULL
) {
1633 PERROR("ust command zmalloc");
1638 * Using message-based transmissions to ensure we don't
1639 * have to deal with partially received messages.
1641 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1643 ERR("Exhausted file descriptors allowed for applications.");
1653 health_code_update();
1654 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1657 /* Close socket of the application. */
1662 lttng_fd_put(LTTNG_FD_APPS
, 1);
1666 health_code_update();
1668 ust_cmd
->sock
= sock
;
1671 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1672 " gid:%d sock:%d name:%s (version %d.%d)",
1673 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1674 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1675 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1676 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1679 * Lock free enqueue the registration request. The red pill
1680 * has been taken! This apps will be part of the *system*.
1682 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1685 * Wake the registration queue futex. Implicit memory
1686 * barrier with the exchange in cds_wfq_enqueue.
1688 futex_nto1_wake(&ust_cmd_queue
.futex
);
1698 ERR("Health error occurred in %s", __func__
);
1701 /* Notify that the registration thread is gone */
1704 if (apps_sock
>= 0) {
1705 ret
= close(apps_sock
);
1715 lttng_fd_put(LTTNG_FD_APPS
, 1);
1717 unlink(apps_unix_sock_path
);
1720 lttng_poll_clean(&events
);
1724 DBG("UST Registration thread cleanup complete");
1725 health_unregister();
1731 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1732 * exec or it will fails.
1734 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1737 struct timespec timeout
;
1739 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1740 consumer_data
->consumer_thread_is_ready
= 0;
1742 /* Setup pthread condition */
1743 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1746 PERROR("pthread_condattr_init consumer data");
1751 * Set the monotonic clock in order to make sure we DO NOT jump in time
1752 * between the clock_gettime() call and the timedwait call. See bug #324
1753 * for a more details and how we noticed it.
1755 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1758 PERROR("pthread_condattr_setclock consumer data");
1762 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1765 PERROR("pthread_cond_init consumer data");
1769 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1772 PERROR("pthread_create consumer");
1777 /* We are about to wait on a pthread condition */
1778 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1780 /* Get time for sem_timedwait absolute timeout */
1781 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1783 * Set the timeout for the condition timed wait even if the clock gettime
1784 * call fails since we might loop on that call and we want to avoid to
1785 * increment the timeout too many times.
1787 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1790 * The following loop COULD be skipped in some conditions so this is why we
1791 * set ret to 0 in order to make sure at least one round of the loop is
1797 * Loop until the condition is reached or when a timeout is reached. Note
1798 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1799 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1800 * possible. This loop does not take any chances and works with both of
1803 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1804 if (clock_ret
< 0) {
1805 PERROR("clock_gettime spawn consumer");
1806 /* Infinite wait for the consumerd thread to be ready */
1807 ret
= pthread_cond_wait(&consumer_data
->cond
,
1808 &consumer_data
->cond_mutex
);
1810 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1811 &consumer_data
->cond_mutex
, &timeout
);
1815 /* Release the pthread condition */
1816 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1820 if (ret
== ETIMEDOUT
) {
1822 * Call has timed out so we kill the kconsumerd_thread and return
1825 ERR("Condition timed out. The consumer thread was never ready."
1827 ret
= pthread_cancel(consumer_data
->thread
);
1829 PERROR("pthread_cancel consumer thread");
1832 PERROR("pthread_cond_wait failed consumer thread");
1837 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1838 if (consumer_data
->pid
== 0) {
1839 ERR("Consumerd did not start");
1840 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1843 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1852 * Join consumer thread
1854 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1858 /* Consumer pid must be a real one. */
1859 if (consumer_data
->pid
> 0) {
1861 ret
= kill(consumer_data
->pid
, SIGTERM
);
1863 ERR("Error killing consumer daemon");
1866 return pthread_join(consumer_data
->thread
, &status
);
1873 * Fork and exec a consumer daemon (consumerd).
1875 * Return pid if successful else -1.
1877 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1881 const char *consumer_to_use
;
1882 const char *verbosity
;
1885 DBG("Spawning consumerd");
1892 if (opt_verbose_consumer
) {
1893 verbosity
= "--verbose";
1895 verbosity
= "--quiet";
1897 switch (consumer_data
->type
) {
1898 case LTTNG_CONSUMER_KERNEL
:
1900 * Find out which consumerd to execute. We will first try the
1901 * 64-bit path, then the sessiond's installation directory, and
1902 * fallback on the 32-bit one,
1904 DBG3("Looking for a kernel consumer at these locations:");
1905 DBG3(" 1) %s", consumerd64_bin
);
1906 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1907 DBG3(" 3) %s", consumerd32_bin
);
1908 if (stat(consumerd64_bin
, &st
) == 0) {
1909 DBG3("Found location #1");
1910 consumer_to_use
= consumerd64_bin
;
1911 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1912 DBG3("Found location #2");
1913 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1914 } else if (stat(consumerd32_bin
, &st
) == 0) {
1915 DBG3("Found location #3");
1916 consumer_to_use
= consumerd32_bin
;
1918 DBG("Could not find any valid consumerd executable");
1921 DBG("Using kernel consumer at: %s", consumer_to_use
);
1922 execl(consumer_to_use
,
1923 "lttng-consumerd", verbosity
, "-k",
1924 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1925 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1928 case LTTNG_CONSUMER64_UST
:
1930 char *tmpnew
= NULL
;
1932 if (consumerd64_libdir
[0] != '\0') {
1936 tmp
= getenv("LD_LIBRARY_PATH");
1940 tmplen
= strlen("LD_LIBRARY_PATH=")
1941 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1942 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1947 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1948 strcat(tmpnew
, consumerd64_libdir
);
1949 if (tmp
[0] != '\0') {
1950 strcat(tmpnew
, ":");
1951 strcat(tmpnew
, tmp
);
1953 ret
= putenv(tmpnew
);
1960 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1961 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1962 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1963 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1965 if (consumerd64_libdir
[0] != '\0') {
1973 case LTTNG_CONSUMER32_UST
:
1975 char *tmpnew
= NULL
;
1977 if (consumerd32_libdir
[0] != '\0') {
1981 tmp
= getenv("LD_LIBRARY_PATH");
1985 tmplen
= strlen("LD_LIBRARY_PATH=")
1986 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1987 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1992 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1993 strcat(tmpnew
, consumerd32_libdir
);
1994 if (tmp
[0] != '\0') {
1995 strcat(tmpnew
, ":");
1996 strcat(tmpnew
, tmp
);
1998 ret
= putenv(tmpnew
);
2005 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2006 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2007 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2008 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2010 if (consumerd32_libdir
[0] != '\0') {
2019 PERROR("unknown consumer type");
2023 PERROR("kernel start consumer exec");
2026 } else if (pid
> 0) {
2029 PERROR("start consumer fork");
2037 * Spawn the consumerd daemon and session daemon thread.
2039 static int start_consumerd(struct consumer_data
*consumer_data
)
2044 * Set the listen() state on the socket since there is a possible race
2045 * between the exec() of the consumer daemon and this call if place in the
2046 * consumer thread. See bug #366 for more details.
2048 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2053 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2054 if (consumer_data
->pid
!= 0) {
2055 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2059 ret
= spawn_consumerd(consumer_data
);
2061 ERR("Spawning consumerd failed");
2062 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2066 /* Setting up the consumer_data pid */
2067 consumer_data
->pid
= ret
;
2068 DBG2("Consumer pid %d", consumer_data
->pid
);
2069 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2071 DBG2("Spawning consumer control thread");
2072 ret
= spawn_consumer_thread(consumer_data
);
2074 ERR("Fatal error spawning consumer control thread");
2082 /* Cleanup already created sockets on error. */
2083 if (consumer_data
->err_sock
>= 0) {
2086 err
= close(consumer_data
->err_sock
);
2088 PERROR("close consumer data error socket");
2095 * Compute health status of each consumer. If one of them is zero (bad
2096 * state), we return 0.
2098 static int check_consumer_health(void)
2102 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2104 DBG3("Health consumer check %d", ret
);
2110 * Setup necessary data for kernel tracer action.
2112 static int init_kernel_tracer(void)
2116 /* Modprobe lttng kernel modules */
2117 ret
= modprobe_lttng_control();
2122 /* Open debugfs lttng */
2123 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2124 if (kernel_tracer_fd
< 0) {
2125 DBG("Failed to open %s", module_proc_lttng
);
2130 /* Validate kernel version */
2131 ret
= kernel_validate_version(kernel_tracer_fd
);
2136 ret
= modprobe_lttng_data();
2141 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2145 modprobe_remove_lttng_control();
2146 ret
= close(kernel_tracer_fd
);
2150 kernel_tracer_fd
= -1;
2151 return LTTNG_ERR_KERN_VERSION
;
2154 ret
= close(kernel_tracer_fd
);
2160 modprobe_remove_lttng_control();
2163 WARN("No kernel tracer available");
2164 kernel_tracer_fd
= -1;
2166 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2168 return LTTNG_ERR_KERN_NA
;
2174 * Copy consumer output from the tracing session to the domain session. The
2175 * function also applies the right modification on a per domain basis for the
2176 * trace files destination directory.
2178 * Should *NOT* be called with RCU read-side lock held.
2180 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2183 const char *dir_name
;
2184 struct consumer_output
*consumer
;
2187 assert(session
->consumer
);
2190 case LTTNG_DOMAIN_KERNEL
:
2191 DBG3("Copying tracing session consumer output in kernel session");
2193 * XXX: We should audit the session creation and what this function
2194 * does "extra" in order to avoid a destroy since this function is used
2195 * in the domain session creation (kernel and ust) only. Same for UST
2198 if (session
->kernel_session
->consumer
) {
2199 consumer_destroy_output(session
->kernel_session
->consumer
);
2201 session
->kernel_session
->consumer
=
2202 consumer_copy_output(session
->consumer
);
2203 /* Ease our life a bit for the next part */
2204 consumer
= session
->kernel_session
->consumer
;
2205 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2207 case LTTNG_DOMAIN_UST
:
2208 DBG3("Copying tracing session consumer output in UST session");
2209 if (session
->ust_session
->consumer
) {
2210 consumer_destroy_output(session
->ust_session
->consumer
);
2212 session
->ust_session
->consumer
=
2213 consumer_copy_output(session
->consumer
);
2214 /* Ease our life a bit for the next part */
2215 consumer
= session
->ust_session
->consumer
;
2216 dir_name
= DEFAULT_UST_TRACE_DIR
;
2219 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2223 /* Append correct directory to subdir */
2224 strncat(consumer
->subdir
, dir_name
,
2225 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2226 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2235 * Create an UST session and add it to the session ust list.
2237 * Should *NOT* be called with RCU read-side lock held.
2239 static int create_ust_session(struct ltt_session
*session
,
2240 struct lttng_domain
*domain
)
2243 struct ltt_ust_session
*lus
= NULL
;
2247 assert(session
->consumer
);
2249 switch (domain
->type
) {
2250 case LTTNG_DOMAIN_UST
:
2253 ERR("Unknown UST domain on create session %d", domain
->type
);
2254 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2258 DBG("Creating UST session");
2260 lus
= trace_ust_create_session(session
->id
);
2262 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2266 lus
->uid
= session
->uid
;
2267 lus
->gid
= session
->gid
;
2268 session
->ust_session
= lus
;
2270 /* Copy session output to the newly created UST session */
2271 ret
= copy_session_consumer(domain
->type
, session
);
2272 if (ret
!= LTTNG_OK
) {
2280 session
->ust_session
= NULL
;
2285 * Create a kernel tracer session then create the default channel.
2287 static int create_kernel_session(struct ltt_session
*session
)
2291 DBG("Creating kernel session");
2293 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2295 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2299 /* Code flow safety */
2300 assert(session
->kernel_session
);
2302 /* Copy session output to the newly created Kernel session */
2303 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2304 if (ret
!= LTTNG_OK
) {
2308 /* Create directory(ies) on local filesystem. */
2309 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2310 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2311 ret
= run_as_mkdir_recursive(
2312 session
->kernel_session
->consumer
->dst
.trace_path
,
2313 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2315 if (ret
!= -EEXIST
) {
2316 ERR("Trace directory creation error");
2322 session
->kernel_session
->uid
= session
->uid
;
2323 session
->kernel_session
->gid
= session
->gid
;
2328 trace_kernel_destroy_session(session
->kernel_session
);
2329 session
->kernel_session
= NULL
;
2334 * Count number of session permitted by uid/gid.
2336 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2339 struct ltt_session
*session
;
2341 DBG("Counting number of available session for UID %d GID %d",
2343 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2345 * Only list the sessions the user can control.
2347 if (!session_access_ok(session
, uid
, gid
)) {
2356 * Process the command requested by the lttng client within the command
2357 * context structure. This function make sure that the return structure (llm)
2358 * is set and ready for transmission before returning.
2360 * Return any error encountered or 0 for success.
2362 * "sock" is only used for special-case var. len data.
2364 * Should *NOT* be called with RCU read-side lock held.
2366 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2370 int need_tracing_session
= 1;
2373 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2377 switch (cmd_ctx
->lsm
->cmd_type
) {
2378 case LTTNG_CREATE_SESSION
:
2379 case LTTNG_DESTROY_SESSION
:
2380 case LTTNG_LIST_SESSIONS
:
2381 case LTTNG_LIST_DOMAINS
:
2382 case LTTNG_START_TRACE
:
2383 case LTTNG_STOP_TRACE
:
2384 case LTTNG_DATA_PENDING
:
2391 if (opt_no_kernel
&& need_domain
2392 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2394 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2396 ret
= LTTNG_ERR_KERN_NA
;
2401 /* Deny register consumer if we already have a spawned consumer. */
2402 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2403 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2404 if (kconsumer_data
.pid
> 0) {
2405 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2406 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2409 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2413 * Check for command that don't needs to allocate a returned payload. We do
2414 * this here so we don't have to make the call for no payload at each
2417 switch(cmd_ctx
->lsm
->cmd_type
) {
2418 case LTTNG_LIST_SESSIONS
:
2419 case LTTNG_LIST_TRACEPOINTS
:
2420 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2421 case LTTNG_LIST_DOMAINS
:
2422 case LTTNG_LIST_CHANNELS
:
2423 case LTTNG_LIST_EVENTS
:
2426 /* Setup lttng message with no payload */
2427 ret
= setup_lttng_msg(cmd_ctx
, 0);
2429 /* This label does not try to unlock the session */
2430 goto init_setup_error
;
2434 /* Commands that DO NOT need a session. */
2435 switch (cmd_ctx
->lsm
->cmd_type
) {
2436 case LTTNG_CREATE_SESSION
:
2437 case LTTNG_CALIBRATE
:
2438 case LTTNG_LIST_SESSIONS
:
2439 case LTTNG_LIST_TRACEPOINTS
:
2440 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2441 need_tracing_session
= 0;
2444 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2446 * We keep the session list lock across _all_ commands
2447 * for now, because the per-session lock does not
2448 * handle teardown properly.
2450 session_lock_list();
2451 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2452 if (cmd_ctx
->session
== NULL
) {
2453 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2454 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2456 /* If no session name specified */
2457 ret
= LTTNG_ERR_SELECT_SESS
;
2461 /* Acquire lock for the session */
2462 session_lock(cmd_ctx
->session
);
2472 * Check domain type for specific "pre-action".
2474 switch (cmd_ctx
->lsm
->domain
.type
) {
2475 case LTTNG_DOMAIN_KERNEL
:
2477 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2481 /* Kernel tracer check */
2482 if (kernel_tracer_fd
== -1) {
2483 /* Basically, load kernel tracer modules */
2484 ret
= init_kernel_tracer();
2490 /* Consumer is in an ERROR state. Report back to client */
2491 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2492 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2496 /* Need a session for kernel command */
2497 if (need_tracing_session
) {
2498 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2499 ret
= create_kernel_session(cmd_ctx
->session
);
2501 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2506 /* Start the kernel consumer daemon */
2507 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2508 if (kconsumer_data
.pid
== 0 &&
2509 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2510 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2511 ret
= start_consumerd(&kconsumer_data
);
2513 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2516 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2518 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2522 * The consumer was just spawned so we need to add the socket to
2523 * the consumer output of the session if exist.
2525 ret
= consumer_create_socket(&kconsumer_data
,
2526 cmd_ctx
->session
->kernel_session
->consumer
);
2533 case LTTNG_DOMAIN_UST
:
2535 /* Consumer is in an ERROR state. Report back to client */
2536 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2537 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2541 if (need_tracing_session
) {
2542 /* Create UST session if none exist. */
2543 if (cmd_ctx
->session
->ust_session
== NULL
) {
2544 ret
= create_ust_session(cmd_ctx
->session
,
2545 &cmd_ctx
->lsm
->domain
);
2546 if (ret
!= LTTNG_OK
) {
2551 /* Start the UST consumer daemons */
2553 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2554 if (consumerd64_bin
[0] != '\0' &&
2555 ustconsumer64_data
.pid
== 0 &&
2556 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2557 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2558 ret
= start_consumerd(&ustconsumer64_data
);
2560 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2561 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2565 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2566 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2568 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2572 * Setup socket for consumer 64 bit. No need for atomic access
2573 * since it was set above and can ONLY be set in this thread.
2575 ret
= consumer_create_socket(&ustconsumer64_data
,
2576 cmd_ctx
->session
->ust_session
->consumer
);
2582 if (consumerd32_bin
[0] != '\0' &&
2583 ustconsumer32_data
.pid
== 0 &&
2584 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2585 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2586 ret
= start_consumerd(&ustconsumer32_data
);
2588 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2589 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2593 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2594 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2596 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2600 * Setup socket for consumer 64 bit. No need for atomic access
2601 * since it was set above and can ONLY be set in this thread.
2603 ret
= consumer_create_socket(&ustconsumer32_data
,
2604 cmd_ctx
->session
->ust_session
->consumer
);
2616 /* Validate consumer daemon state when start/stop trace command */
2617 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2618 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2619 switch (cmd_ctx
->lsm
->domain
.type
) {
2620 case LTTNG_DOMAIN_UST
:
2621 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2622 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2626 case LTTNG_DOMAIN_KERNEL
:
2627 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2628 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2636 * Check that the UID or GID match that of the tracing session.
2637 * The root user can interact with all sessions.
2639 if (need_tracing_session
) {
2640 if (!session_access_ok(cmd_ctx
->session
,
2641 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2642 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2643 ret
= LTTNG_ERR_EPERM
;
2649 * Send relayd information to consumer as soon as we have a domain and a
2652 if (cmd_ctx
->session
&& need_domain
) {
2654 * Setup relayd if not done yet. If the relayd information was already
2655 * sent to the consumer, this call will gracefully return.
2657 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2658 if (ret
!= LTTNG_OK
) {
2663 /* Process by command type */
2664 switch (cmd_ctx
->lsm
->cmd_type
) {
2665 case LTTNG_ADD_CONTEXT
:
2667 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2668 cmd_ctx
->lsm
->u
.context
.channel_name
,
2669 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2672 case LTTNG_DISABLE_CHANNEL
:
2674 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2675 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2678 case LTTNG_DISABLE_EVENT
:
2680 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2681 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2682 cmd_ctx
->lsm
->u
.disable
.name
);
2685 case LTTNG_DISABLE_ALL_EVENT
:
2687 DBG("Disabling all events");
2689 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2690 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2693 case LTTNG_ENABLE_CHANNEL
:
2695 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2696 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2699 case LTTNG_ENABLE_EVENT
:
2701 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2702 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2703 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2706 case LTTNG_ENABLE_ALL_EVENT
:
2708 DBG("Enabling all events");
2710 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2711 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2712 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2715 case LTTNG_LIST_TRACEPOINTS
:
2717 struct lttng_event
*events
;
2720 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2721 if (nb_events
< 0) {
2722 /* Return value is a negative lttng_error_code. */
2728 * Setup lttng message with payload size set to the event list size in
2729 * bytes and then copy list into the llm payload.
2731 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2737 /* Copy event list into message payload */
2738 memcpy(cmd_ctx
->llm
->payload
, events
,
2739 sizeof(struct lttng_event
) * nb_events
);
2746 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2748 struct lttng_event_field
*fields
;
2751 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2753 if (nb_fields
< 0) {
2754 /* Return value is a negative lttng_error_code. */
2760 * Setup lttng message with payload size set to the event list size in
2761 * bytes and then copy list into the llm payload.
2763 ret
= setup_lttng_msg(cmd_ctx
,
2764 sizeof(struct lttng_event_field
) * nb_fields
);
2770 /* Copy event list into message payload */
2771 memcpy(cmd_ctx
->llm
->payload
, fields
,
2772 sizeof(struct lttng_event_field
) * nb_fields
);
2779 case LTTNG_SET_CONSUMER_URI
:
2782 struct lttng_uri
*uris
;
2784 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2785 len
= nb_uri
* sizeof(struct lttng_uri
);
2788 ret
= LTTNG_ERR_INVALID
;
2792 uris
= zmalloc(len
);
2794 ret
= LTTNG_ERR_FATAL
;
2798 /* Receive variable len data */
2799 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2800 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2802 DBG("No URIs received from client... continuing");
2804 ret
= LTTNG_ERR_SESSION_FAIL
;
2809 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2811 if (ret
!= LTTNG_OK
) {
2817 * XXX: 0 means that this URI should be applied on the session. Should
2818 * be a DOMAIN enuam.
2820 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2821 /* Add the URI for the UST session if a consumer is present. */
2822 if (cmd_ctx
->session
->ust_session
&&
2823 cmd_ctx
->session
->ust_session
->consumer
) {
2824 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2826 } else if (cmd_ctx
->session
->kernel_session
&&
2827 cmd_ctx
->session
->kernel_session
->consumer
) {
2828 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2829 cmd_ctx
->session
, nb_uri
, uris
);
2837 case LTTNG_START_TRACE
:
2839 ret
= cmd_start_trace(cmd_ctx
->session
);
2842 case LTTNG_STOP_TRACE
:
2844 ret
= cmd_stop_trace(cmd_ctx
->session
);
2847 case LTTNG_CREATE_SESSION
:
2850 struct lttng_uri
*uris
= NULL
;
2852 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2853 len
= nb_uri
* sizeof(struct lttng_uri
);
2856 uris
= zmalloc(len
);
2858 ret
= LTTNG_ERR_FATAL
;
2862 /* Receive variable len data */
2863 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2864 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2866 DBG("No URIs received from client... continuing");
2868 ret
= LTTNG_ERR_SESSION_FAIL
;
2873 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2874 DBG("Creating session with ONE network URI is a bad call");
2875 ret
= LTTNG_ERR_SESSION_FAIL
;
2881 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2888 case LTTNG_DESTROY_SESSION
:
2890 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2892 /* Set session to NULL so we do not unlock it after free. */
2893 cmd_ctx
->session
= NULL
;
2896 case LTTNG_LIST_DOMAINS
:
2899 struct lttng_domain
*domains
;
2901 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2903 /* Return value is a negative lttng_error_code. */
2908 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2914 /* Copy event list into message payload */
2915 memcpy(cmd_ctx
->llm
->payload
, domains
,
2916 nb_dom
* sizeof(struct lttng_domain
));
2923 case LTTNG_LIST_CHANNELS
:
2926 struct lttng_channel
*channels
;
2928 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2929 cmd_ctx
->session
, &channels
);
2931 /* Return value is a negative lttng_error_code. */
2936 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2942 /* Copy event list into message payload */
2943 memcpy(cmd_ctx
->llm
->payload
, channels
,
2944 nb_chan
* sizeof(struct lttng_channel
));
2951 case LTTNG_LIST_EVENTS
:
2954 struct lttng_event
*events
= NULL
;
2956 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2957 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2959 /* Return value is a negative lttng_error_code. */
2964 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2970 /* Copy event list into message payload */
2971 memcpy(cmd_ctx
->llm
->payload
, events
,
2972 nb_event
* sizeof(struct lttng_event
));
2979 case LTTNG_LIST_SESSIONS
:
2981 unsigned int nr_sessions
;
2983 session_lock_list();
2984 nr_sessions
= lttng_sessions_count(
2985 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2986 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2988 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2990 session_unlock_list();
2994 /* Filled the session array */
2995 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2996 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2997 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2999 session_unlock_list();
3004 case LTTNG_CALIBRATE
:
3006 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3007 &cmd_ctx
->lsm
->u
.calibrate
);
3010 case LTTNG_REGISTER_CONSUMER
:
3012 struct consumer_data
*cdata
;
3014 switch (cmd_ctx
->lsm
->domain
.type
) {
3015 case LTTNG_DOMAIN_KERNEL
:
3016 cdata
= &kconsumer_data
;
3019 ret
= LTTNG_ERR_UND
;
3023 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3024 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3027 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3029 struct lttng_filter_bytecode
*bytecode
;
3031 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3032 ret
= LTTNG_ERR_FILTER_INVAL
;
3035 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3036 ret
= LTTNG_ERR_FILTER_INVAL
;
3039 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3041 ret
= LTTNG_ERR_FILTER_NOMEM
;
3044 /* Receive var. len. data */
3045 DBG("Receiving var len data from client ...");
3046 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3047 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3049 DBG("Nothing recv() from client var len data... continuing");
3051 ret
= LTTNG_ERR_FILTER_INVAL
;
3055 if (bytecode
->len
+ sizeof(*bytecode
)
3056 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3058 ret
= LTTNG_ERR_FILTER_INVAL
;
3062 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3063 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3064 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3067 case LTTNG_DATA_PENDING
:
3069 ret
= cmd_data_pending(cmd_ctx
->session
);
3073 ret
= LTTNG_ERR_UND
;
3078 if (cmd_ctx
->llm
== NULL
) {
3079 DBG("Missing llm structure. Allocating one.");
3080 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3084 /* Set return code */
3085 cmd_ctx
->llm
->ret_code
= ret
;
3087 if (cmd_ctx
->session
) {
3088 session_unlock(cmd_ctx
->session
);
3090 if (need_tracing_session
) {
3091 session_unlock_list();
3098 * Thread managing health check socket.
3100 static void *thread_manage_health(void *data
)
3102 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3103 uint32_t revents
, nb_fd
;
3104 struct lttng_poll_event events
;
3105 struct lttcomm_health_msg msg
;
3106 struct lttcomm_health_data reply
;
3108 DBG("[thread] Manage health check started");
3110 rcu_register_thread();
3112 /* We might hit an error path before this is created. */
3113 lttng_poll_init(&events
);
3115 /* Create unix socket */
3116 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3118 ERR("Unable to create health check Unix socket");
3124 * Set the CLOEXEC flag. Return code is useless because either way, the
3127 (void) utils_set_fd_cloexec(sock
);
3129 ret
= lttcomm_listen_unix_sock(sock
);
3135 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3136 * more will be added to this poll set.
3138 ret
= sessiond_set_thread_pollset(&events
, 2);
3143 /* Add the application registration socket */
3144 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3150 DBG("Health check ready");
3152 /* Inifinite blocking call, waiting for transmission */
3154 ret
= lttng_poll_wait(&events
, -1);
3157 * Restart interrupted system call.
3159 if (errno
== EINTR
) {
3167 for (i
= 0; i
< nb_fd
; i
++) {
3168 /* Fetch once the poll data */
3169 revents
= LTTNG_POLL_GETEV(&events
, i
);
3170 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3172 /* Thread quit pipe has been closed. Killing thread. */
3173 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3179 /* Event on the registration socket */
3180 if (pollfd
== sock
) {
3181 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3182 ERR("Health socket poll error");
3188 new_sock
= lttcomm_accept_unix_sock(sock
);
3194 * Set the CLOEXEC flag. Return code is useless because either way, the
3197 (void) utils_set_fd_cloexec(new_sock
);
3199 DBG("Receiving data from client for health...");
3200 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3202 DBG("Nothing recv() from client... continuing");
3203 ret
= close(new_sock
);
3211 rcu_thread_online();
3213 switch (msg
.component
) {
3214 case LTTNG_HEALTH_CMD
:
3215 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3217 case LTTNG_HEALTH_APP_MANAGE
:
3218 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3220 case LTTNG_HEALTH_APP_REG
:
3221 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3223 case LTTNG_HEALTH_KERNEL
:
3224 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3226 case LTTNG_HEALTH_CONSUMER
:
3227 reply
.ret_code
= check_consumer_health();
3229 case LTTNG_HEALTH_HT_CLEANUP
:
3230 reply
.ret_code
= health_check_state(HEALTH_TYPE_HT_CLEANUP
);
3232 case LTTNG_HEALTH_APP_MANAGE_NOTIFY
:
3233 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3235 case LTTNG_HEALTH_ALL
:
3237 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3238 health_check_state(HEALTH_TYPE_APP_REG
) &&
3239 health_check_state(HEALTH_TYPE_CMD
) &&
3240 health_check_state(HEALTH_TYPE_KERNEL
) &&
3241 check_consumer_health() &&
3242 health_check_state(HEALTH_TYPE_HT_CLEANUP
) &&
3243 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3246 reply
.ret_code
= LTTNG_ERR_UND
;
3251 * Flip ret value since 0 is a success and 1 indicates a bad health for
3252 * the client where in the sessiond it is the opposite. Again, this is
3253 * just to make things easier for us poor developer which enjoy a lot
3256 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3257 reply
.ret_code
= !reply
.ret_code
;
3260 DBG2("Health check return value %d", reply
.ret_code
);
3262 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3264 ERR("Failed to send health data back to client");
3267 /* End of transmission */
3268 ret
= close(new_sock
);
3278 ERR("Health error occurred in %s", __func__
);
3280 DBG("Health check thread dying");
3281 unlink(health_unix_sock_path
);
3289 lttng_poll_clean(&events
);
3291 rcu_unregister_thread();
3296 * This thread manage all clients request using the unix client socket for
3299 static void *thread_manage_clients(void *data
)
3301 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3303 uint32_t revents
, nb_fd
;
3304 struct command_ctx
*cmd_ctx
= NULL
;
3305 struct lttng_poll_event events
;
3307 DBG("[thread] Manage client started");
3309 rcu_register_thread();
3311 health_register(HEALTH_TYPE_CMD
);
3313 if (testpoint(thread_manage_clients
)) {
3314 goto error_testpoint
;
3317 health_code_update();
3319 ret
= lttcomm_listen_unix_sock(client_sock
);
3325 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3326 * more will be added to this poll set.
3328 ret
= sessiond_set_thread_pollset(&events
, 2);
3330 goto error_create_poll
;
3333 /* Add the application registration socket */
3334 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3340 * Notify parent pid that we are ready to accept command for client side.
3342 if (opt_sig_parent
) {
3343 kill(ppid
, SIGUSR1
);
3346 if (testpoint(thread_manage_clients_before_loop
)) {
3350 health_code_update();
3353 DBG("Accepting client command ...");
3355 /* Inifinite blocking call, waiting for transmission */
3357 health_poll_entry();
3358 ret
= lttng_poll_wait(&events
, -1);
3362 * Restart interrupted system call.
3364 if (errno
== EINTR
) {
3372 for (i
= 0; i
< nb_fd
; i
++) {
3373 /* Fetch once the poll data */
3374 revents
= LTTNG_POLL_GETEV(&events
, i
);
3375 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3377 health_code_update();
3379 /* Thread quit pipe has been closed. Killing thread. */
3380 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3386 /* Event on the registration socket */
3387 if (pollfd
== client_sock
) {
3388 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3389 ERR("Client socket poll error");
3395 DBG("Wait for client response");
3397 health_code_update();
3399 sock
= lttcomm_accept_unix_sock(client_sock
);
3405 * Set the CLOEXEC flag. Return code is useless because either way, the
3408 (void) utils_set_fd_cloexec(sock
);
3410 /* Set socket option for credentials retrieval */
3411 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3416 /* Allocate context command to process the client request */
3417 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3418 if (cmd_ctx
== NULL
) {
3419 PERROR("zmalloc cmd_ctx");
3423 /* Allocate data buffer for reception */
3424 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3425 if (cmd_ctx
->lsm
== NULL
) {
3426 PERROR("zmalloc cmd_ctx->lsm");
3430 cmd_ctx
->llm
= NULL
;
3431 cmd_ctx
->session
= NULL
;
3433 health_code_update();
3436 * Data is received from the lttng client. The struct
3437 * lttcomm_session_msg (lsm) contains the command and data request of
3440 DBG("Receiving data from client ...");
3441 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3442 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3444 DBG("Nothing recv() from client... continuing");
3450 clean_command_ctx(&cmd_ctx
);
3454 health_code_update();
3456 // TODO: Validate cmd_ctx including sanity check for
3457 // security purpose.
3459 rcu_thread_online();
3461 * This function dispatch the work to the kernel or userspace tracer
3462 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3463 * informations for the client. The command context struct contains
3464 * everything this function may needs.
3466 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3467 rcu_thread_offline();
3475 * TODO: Inform client somehow of the fatal error. At
3476 * this point, ret < 0 means that a zmalloc failed
3477 * (ENOMEM). Error detected but still accept
3478 * command, unless a socket error has been
3481 clean_command_ctx(&cmd_ctx
);
3485 health_code_update();
3487 DBG("Sending response (size: %d, retcode: %s)",
3488 cmd_ctx
->lttng_msg_size
,
3489 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3490 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3492 ERR("Failed to send data back to client");
3495 /* End of transmission */
3502 clean_command_ctx(&cmd_ctx
);
3504 health_code_update();
3516 lttng_poll_clean(&events
);
3517 clean_command_ctx(&cmd_ctx
);
3522 unlink(client_unix_sock_path
);
3523 if (client_sock
>= 0) {
3524 ret
= close(client_sock
);
3532 ERR("Health error occurred in %s", __func__
);
3535 health_unregister();
3537 DBG("Client thread dying");
3539 rcu_unregister_thread();
3545 * usage function on stderr
3547 static void usage(void)
3549 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3550 fprintf(stderr
, " -h, --help Display this usage.\n");
3551 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3552 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3553 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3554 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3555 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3556 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3557 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3558 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3559 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3560 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3561 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3562 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3563 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3564 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3565 fprintf(stderr
, " -V, --version Show version number.\n");
3566 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3567 fprintf(stderr
, " -q, --quiet No output at all.\n");
3568 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3569 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3570 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3571 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3575 * daemon argument parsing
3577 static int parse_args(int argc
, char **argv
)
3581 static struct option long_options
[] = {
3582 { "client-sock", 1, 0, 'c' },
3583 { "apps-sock", 1, 0, 'a' },
3584 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3585 { "kconsumerd-err-sock", 1, 0, 'E' },
3586 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3587 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3588 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3589 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3590 { "consumerd32-path", 1, 0, 'u' },
3591 { "consumerd32-libdir", 1, 0, 'U' },
3592 { "consumerd64-path", 1, 0, 't' },
3593 { "consumerd64-libdir", 1, 0, 'T' },
3594 { "daemonize", 0, 0, 'd' },
3595 { "sig-parent", 0, 0, 'S' },
3596 { "help", 0, 0, 'h' },
3597 { "group", 1, 0, 'g' },
3598 { "version", 0, 0, 'V' },
3599 { "quiet", 0, 0, 'q' },
3600 { "verbose", 0, 0, 'v' },
3601 { "verbose-consumer", 0, 0, 'Z' },
3602 { "no-kernel", 0, 0, 'N' },
3603 { "pidfile", 1, 0, 'p' },
3608 int option_index
= 0;
3609 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3610 long_options
, &option_index
);
3617 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3619 fprintf(stderr
, " with arg %s\n", optarg
);
3623 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3626 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3632 opt_tracing_group
= optarg
;
3638 fprintf(stdout
, "%s\n", VERSION
);
3644 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3647 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3650 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3653 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3656 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3659 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3665 lttng_opt_quiet
= 1;
3668 /* Verbose level can increase using multiple -v */
3669 lttng_opt_verbose
+= 1;
3672 opt_verbose_consumer
+= 1;
3675 consumerd32_bin
= optarg
;
3678 consumerd32_libdir
= optarg
;
3681 consumerd64_bin
= optarg
;
3684 consumerd64_libdir
= optarg
;
3687 opt_pidfile
= optarg
;
3690 /* Unknown option or other error.
3691 * Error is printed by getopt, just return */
3700 * Creates the two needed socket by the daemon.
3701 * apps_sock - The communication socket for all UST apps.
3702 * client_sock - The communication of the cli tool (lttng).
3704 static int init_daemon_socket(void)
3709 old_umask
= umask(0);
3711 /* Create client tool unix socket */
3712 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3713 if (client_sock
< 0) {
3714 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3719 /* Set the cloexec flag */
3720 ret
= utils_set_fd_cloexec(client_sock
);
3722 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3723 "Continuing but note that the consumer daemon will have a "
3724 "reference to this socket on exec()", client_sock
);
3727 /* File permission MUST be 660 */
3728 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3730 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3735 /* Create the application unix socket */
3736 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3737 if (apps_sock
< 0) {
3738 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3743 /* Set the cloexec flag */
3744 ret
= utils_set_fd_cloexec(apps_sock
);
3746 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3747 "Continuing but note that the consumer daemon will have a "
3748 "reference to this socket on exec()", apps_sock
);
3751 /* File permission MUST be 666 */
3752 ret
= chmod(apps_unix_sock_path
,
3753 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3755 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3760 DBG3("Session daemon client socket %d and application socket %d created",
3761 client_sock
, apps_sock
);
3769 * Check if the global socket is available, and if a daemon is answering at the
3770 * other side. If yes, error is returned.
3772 static int check_existing_daemon(void)
3774 /* Is there anybody out there ? */
3775 if (lttng_session_daemon_alive()) {
3783 * Set the tracing group gid onto the client socket.
3785 * Race window between mkdir and chown is OK because we are going from more
3786 * permissive (root.root) to less permissive (root.tracing).
3788 static int set_permissions(char *rundir
)
3793 ret
= allowed_group();
3795 WARN("No tracing group detected");
3802 /* Set lttng run dir */
3803 ret
= chown(rundir
, 0, gid
);
3805 ERR("Unable to set group on %s", rundir
);
3809 /* Ensure tracing group can search the run dir */
3810 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3812 ERR("Unable to set permissions on %s", rundir
);
3816 /* lttng client socket path */
3817 ret
= chown(client_unix_sock_path
, 0, gid
);
3819 ERR("Unable to set group on %s", client_unix_sock_path
);
3823 /* kconsumer error socket path */
3824 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3826 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3830 /* 64-bit ustconsumer error socket path */
3831 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3833 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3837 /* 32-bit ustconsumer compat32 error socket path */
3838 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3840 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3844 DBG("All permissions are set");
3851 * Create the lttng run directory needed for all global sockets and pipe.
3853 static int create_lttng_rundir(const char *rundir
)
3857 DBG3("Creating LTTng run directory: %s", rundir
);
3859 ret
= mkdir(rundir
, S_IRWXU
);
3861 if (errno
!= EEXIST
) {
3862 ERR("Unable to create %s", rundir
);
3874 * Setup sockets and directory needed by the kconsumerd communication with the
3877 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3881 char path
[PATH_MAX
];
3883 switch (consumer_data
->type
) {
3884 case LTTNG_CONSUMER_KERNEL
:
3885 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3887 case LTTNG_CONSUMER64_UST
:
3888 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3890 case LTTNG_CONSUMER32_UST
:
3891 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3894 ERR("Consumer type unknown");
3899 DBG2("Creating consumer directory: %s", path
);
3901 ret
= mkdir(path
, S_IRWXU
);
3903 if (errno
!= EEXIST
) {
3905 ERR("Failed to create %s", path
);
3911 /* Create the kconsumerd error unix socket */
3912 consumer_data
->err_sock
=
3913 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3914 if (consumer_data
->err_sock
< 0) {
3915 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3920 /* File permission MUST be 660 */
3921 ret
= chmod(consumer_data
->err_unix_sock_path
,
3922 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3924 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3934 * Signal handler for the daemon
3936 * Simply stop all worker threads, leaving main() return gracefully after
3937 * joining all threads and calling cleanup().
3939 static void sighandler(int sig
)
3943 DBG("SIGPIPE caught");
3946 DBG("SIGINT caught");
3950 DBG("SIGTERM caught");
3959 * Setup signal handler for :
3960 * SIGINT, SIGTERM, SIGPIPE
3962 static int set_signal_handler(void)
3965 struct sigaction sa
;
3968 if ((ret
= sigemptyset(&sigset
)) < 0) {
3969 PERROR("sigemptyset");
3973 sa
.sa_handler
= sighandler
;
3974 sa
.sa_mask
= sigset
;
3976 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3977 PERROR("sigaction");
3981 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3982 PERROR("sigaction");
3986 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3987 PERROR("sigaction");
3991 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3997 * Set open files limit to unlimited. This daemon can open a large number of
3998 * file descriptors in order to consumer multiple kernel traces.
4000 static void set_ulimit(void)
4005 /* The kernel does not allowed an infinite limit for open files */
4006 lim
.rlim_cur
= 65535;
4007 lim
.rlim_max
= 65535;
4009 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4011 PERROR("failed to set open files limit");
4016 * Write pidfile using the rundir and opt_pidfile.
4018 static void write_pidfile(void)
4021 char pidfile_path
[PATH_MAX
];
4026 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4028 /* Build pidfile path from rundir and opt_pidfile. */
4029 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4030 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4032 PERROR("snprintf pidfile path");
4038 * Create pid file in rundir. Return value is of no importance. The
4039 * execution will continue even though we are not able to write the file.
4041 (void) utils_create_pid_file(getpid(), pidfile_path
);
4050 int main(int argc
, char **argv
)
4054 const char *home_path
, *env_app_timeout
;
4056 init_kernel_workarounds();
4058 rcu_register_thread();
4060 setup_consumerd_path();
4062 page_size
= sysconf(_SC_PAGESIZE
);
4063 if (page_size
< 0) {
4064 PERROR("sysconf _SC_PAGESIZE");
4065 page_size
= LONG_MAX
;
4066 WARN("Fallback page size to %ld", page_size
);
4069 /* Parse arguments */
4071 if ((ret
= parse_args(argc
, argv
)) < 0) {
4081 * child: setsid, close FD 0, 1, 2, chdir /
4082 * parent: exit (if fork is successful)
4090 * We are in the child. Make sure all other file
4091 * descriptors are closed, in case we are called with
4092 * more opened file descriptors than the standard ones.
4094 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4099 /* Create thread quit pipe */
4100 if ((ret
= init_thread_quit_pipe()) < 0) {
4104 /* Check if daemon is UID = 0 */
4105 is_root
= !getuid();
4108 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4110 /* Create global run dir with root access */
4111 ret
= create_lttng_rundir(rundir
);
4116 if (strlen(apps_unix_sock_path
) == 0) {
4117 snprintf(apps_unix_sock_path
, PATH_MAX
,
4118 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4121 if (strlen(client_unix_sock_path
) == 0) {
4122 snprintf(client_unix_sock_path
, PATH_MAX
,
4123 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4126 /* Set global SHM for ust */
4127 if (strlen(wait_shm_path
) == 0) {
4128 snprintf(wait_shm_path
, PATH_MAX
,
4129 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4132 if (strlen(health_unix_sock_path
) == 0) {
4133 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4134 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4137 /* Setup kernel consumerd path */
4138 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4139 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4140 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4141 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4143 DBG2("Kernel consumer err path: %s",
4144 kconsumer_data
.err_unix_sock_path
);
4145 DBG2("Kernel consumer cmd path: %s",
4146 kconsumer_data
.cmd_unix_sock_path
);
4148 home_path
= get_home_dir();
4149 if (home_path
== NULL
) {
4150 /* TODO: Add --socket PATH option */
4151 ERR("Can't get HOME directory for sockets creation.");
4157 * Create rundir from home path. This will create something like
4160 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4166 ret
= create_lttng_rundir(rundir
);
4171 if (strlen(apps_unix_sock_path
) == 0) {
4172 snprintf(apps_unix_sock_path
, PATH_MAX
,
4173 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4176 /* Set the cli tool unix socket path */
4177 if (strlen(client_unix_sock_path
) == 0) {
4178 snprintf(client_unix_sock_path
, PATH_MAX
,
4179 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4182 /* Set global SHM for ust */
4183 if (strlen(wait_shm_path
) == 0) {
4184 snprintf(wait_shm_path
, PATH_MAX
,
4185 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4188 /* Set health check Unix path */
4189 if (strlen(health_unix_sock_path
) == 0) {
4190 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4191 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4195 /* Set consumer initial state */
4196 kernel_consumerd_state
= CONSUMER_STOPPED
;
4197 ust_consumerd_state
= CONSUMER_STOPPED
;
4199 DBG("Client socket path %s", client_unix_sock_path
);
4200 DBG("Application socket path %s", apps_unix_sock_path
);
4201 DBG("Application wait path %s", wait_shm_path
);
4202 DBG("LTTng run directory path: %s", rundir
);
4204 /* 32 bits consumerd path setup */
4205 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4206 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4207 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4208 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4210 DBG2("UST consumer 32 bits err path: %s",
4211 ustconsumer32_data
.err_unix_sock_path
);
4212 DBG2("UST consumer 32 bits cmd path: %s",
4213 ustconsumer32_data
.cmd_unix_sock_path
);
4215 /* 64 bits consumerd path setup */
4216 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4217 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4218 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4219 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4221 DBG2("UST consumer 64 bits err path: %s",
4222 ustconsumer64_data
.err_unix_sock_path
);
4223 DBG2("UST consumer 64 bits cmd path: %s",
4224 ustconsumer64_data
.cmd_unix_sock_path
);
4227 * See if daemon already exist.
4229 if ((ret
= check_existing_daemon()) < 0) {
4230 ERR("Already running daemon.\n");
4232 * We do not goto exit because we must not cleanup()
4233 * because a daemon is already running.
4239 * Init UST app hash table. Alloc hash table before this point since
4240 * cleanup() can get called after that point.
4244 /* After this point, we can safely call cleanup() with "goto exit" */
4247 * These actions must be executed as root. We do that *after* setting up
4248 * the sockets path because we MUST make the check for another daemon using
4249 * those paths *before* trying to set the kernel consumer sockets and init
4253 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4258 /* Setup kernel tracer */
4259 if (!opt_no_kernel
) {
4260 init_kernel_tracer();
4263 /* Set ulimit for open files */
4266 /* init lttng_fd tracking must be done after set_ulimit. */
4269 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4274 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4279 if ((ret
= set_signal_handler()) < 0) {
4283 /* Setup the needed unix socket */
4284 if ((ret
= init_daemon_socket()) < 0) {
4288 /* Set credentials to socket */
4289 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4293 /* Get parent pid if -S, --sig-parent is specified. */
4294 if (opt_sig_parent
) {
4298 /* Setup the kernel pipe for waking up the kernel thread */
4299 if (is_root
&& !opt_no_kernel
) {
4300 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4305 /* Setup the thread ht_cleanup communication pipe. */
4306 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4310 /* Setup the thread apps communication pipe. */
4311 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4315 /* Setup the thread apps notify communication pipe. */
4316 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4320 /* Initialize global buffer per UID and PID registry. */
4321 buffer_reg_init_uid_registry();
4322 buffer_reg_init_pid_registry();
4324 /* Init UST command queue. */
4325 cds_wfq_init(&ust_cmd_queue
.queue
);
4328 * Get session list pointer. This pointer MUST NOT be free(). This list is
4329 * statically declared in session.c
4331 session_list_ptr
= session_get_list();
4333 /* Set up max poll set size */
4334 lttng_poll_set_max_size();
4338 /* Check for the application socket timeout env variable. */
4339 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4340 if (env_app_timeout
) {
4341 app_socket_timeout
= atoi(env_app_timeout
);
4343 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4348 /* Create thread to manage the client socket */
4349 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4350 thread_ht_cleanup
, (void *) NULL
);
4352 PERROR("pthread_create ht_cleanup");
4353 goto exit_ht_cleanup
;
4356 /* Create thread to manage the client socket */
4357 ret
= pthread_create(&health_thread
, NULL
,
4358 thread_manage_health
, (void *) NULL
);
4360 PERROR("pthread_create health");
4364 /* Create thread to manage the client socket */
4365 ret
= pthread_create(&client_thread
, NULL
,
4366 thread_manage_clients
, (void *) NULL
);
4368 PERROR("pthread_create clients");
4372 /* Create thread to dispatch registration */
4373 ret
= pthread_create(&dispatch_thread
, NULL
,
4374 thread_dispatch_ust_registration
, (void *) NULL
);
4376 PERROR("pthread_create dispatch");
4380 /* Create thread to manage application registration. */
4381 ret
= pthread_create(®_apps_thread
, NULL
,
4382 thread_registration_apps
, (void *) NULL
);
4384 PERROR("pthread_create registration");
4388 /* Create thread to manage application socket */
4389 ret
= pthread_create(&apps_thread
, NULL
,
4390 thread_manage_apps
, (void *) NULL
);
4392 PERROR("pthread_create apps");
4396 /* Create thread to manage application notify socket */
4397 ret
= pthread_create(&apps_notify_thread
, NULL
,
4398 ust_thread_manage_notify
, (void *) NULL
);
4400 PERROR("pthread_create apps");
4404 /* Don't start this thread if kernel tracing is not requested nor root */
4405 if (is_root
&& !opt_no_kernel
) {
4406 /* Create kernel thread to manage kernel event */
4407 ret
= pthread_create(&kernel_thread
, NULL
,
4408 thread_manage_kernel
, (void *) NULL
);
4410 PERROR("pthread_create kernel");
4414 ret
= pthread_join(kernel_thread
, &status
);
4416 PERROR("pthread_join");
4417 goto error
; /* join error, exit without cleanup */
4422 ret
= pthread_join(apps_thread
, &status
);
4424 PERROR("pthread_join");
4425 goto error
; /* join error, exit without cleanup */
4429 ret
= pthread_join(reg_apps_thread
, &status
);
4431 PERROR("pthread_join");
4432 goto error
; /* join error, exit without cleanup */
4436 ret
= pthread_join(dispatch_thread
, &status
);
4438 PERROR("pthread_join");
4439 goto error
; /* join error, exit without cleanup */
4443 ret
= pthread_join(client_thread
, &status
);
4445 PERROR("pthread_join");
4446 goto error
; /* join error, exit without cleanup */
4449 ret
= join_consumer_thread(&kconsumer_data
);
4451 PERROR("join_consumer");
4452 goto error
; /* join error, exit without cleanup */
4455 ret
= join_consumer_thread(&ustconsumer32_data
);
4457 PERROR("join_consumer ust32");
4458 goto error
; /* join error, exit without cleanup */
4461 ret
= join_consumer_thread(&ustconsumer64_data
);
4463 PERROR("join_consumer ust64");
4464 goto error
; /* join error, exit without cleanup */
4468 ret
= pthread_join(health_thread
, &status
);
4470 PERROR("pthread_join health thread");
4471 goto error
; /* join error, exit without cleanup */
4475 ret
= pthread_join(ht_cleanup_thread
, &status
);
4477 PERROR("pthread_join ht cleanup thread");
4478 goto error
; /* join error, exit without cleanup */
4483 * cleanup() is called when no other thread is running.
4485 rcu_thread_online();
4487 rcu_thread_offline();
4488 rcu_unregister_thread();