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 health_register(HEALTH_TYPE_APP_REG_DISPATCH
);
1353 health_code_update();
1355 CDS_LIST_HEAD(wait_queue
);
1357 DBG("[thread] Dispatch UST command started");
1359 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1360 health_code_update();
1362 /* Atomically prepare the queue futex */
1363 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1366 struct ust_app
*app
= NULL
;
1369 health_code_update();
1370 /* Dequeue command for registration */
1371 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1373 DBG("Woken up but nothing in the UST command queue");
1374 /* Continue thread execution */
1378 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1380 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1381 " gid:%d sock:%d name:%s (version %d.%d)",
1382 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1383 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1384 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1385 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1387 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1388 wait_node
= zmalloc(sizeof(*wait_node
));
1390 PERROR("zmalloc wait_node dispatch");
1391 ret
= close(ust_cmd
->sock
);
1393 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1395 lttng_fd_put(1, LTTNG_FD_APPS
);
1399 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1401 /* Create application object if socket is CMD. */
1402 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1404 if (!wait_node
->app
) {
1405 ret
= close(ust_cmd
->sock
);
1407 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1409 lttng_fd_put(1, LTTNG_FD_APPS
);
1415 * Add application to the wait queue so we can set the notify
1416 * socket before putting this object in the global ht.
1418 cds_list_add(&wait_node
->head
, &wait_queue
);
1422 * We have to continue here since we don't have the notify
1423 * socket and the application MUST be added to the hash table
1424 * only at that moment.
1429 * Look for the application in the local wait queue and set the
1430 * notify socket if found.
1432 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1433 &wait_queue
, head
) {
1434 health_code_update();
1435 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1436 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1437 cds_list_del(&wait_node
->head
);
1438 app
= wait_node
->app
;
1440 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1446 * With no application at this stage the received socket is
1447 * basically useless so close it before we free the cmd data
1448 * structure for good.
1451 ret
= close(ust_cmd
->sock
);
1453 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1455 lttng_fd_put(1, LTTNG_FD_APPS
);
1462 * @session_lock_list
1464 * Lock the global session list so from the register up to the
1465 * registration done message, no thread can see the application
1466 * and change its state.
1468 session_lock_list();
1472 * Add application to the global hash table. This needs to be
1473 * done before the update to the UST registry can locate the
1478 /* Set app version. This call will print an error if needed. */
1479 (void) ust_app_version(app
);
1481 /* Send notify socket through the notify pipe. */
1482 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1486 session_unlock_list();
1487 /* No notify thread, stop the UST tracing. */
1492 * Update newly registered application with the tracing
1493 * registry info already enabled information.
1495 update_ust_app(app
->sock
);
1498 * Don't care about return value. Let the manage apps threads
1499 * handle app unregistration upon socket close.
1501 (void) ust_app_register_done(app
->sock
);
1504 * Even if the application socket has been closed, send the app
1505 * to the thread and unregistration will take place at that
1508 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1511 session_unlock_list();
1512 /* No apps. thread, stop the UST tracing. */
1517 session_unlock_list();
1519 } while (node
!= NULL
);
1521 health_poll_entry();
1522 /* Futex wait on queue. Blocking call on futex() */
1523 futex_nto1_wait(&ust_cmd_queue
.futex
);
1526 /* Normal exit, no error */
1530 /* Clean up wait queue. */
1531 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1532 &wait_queue
, head
) {
1533 cds_list_del(&wait_node
->head
);
1537 DBG("Dispatch thread dying");
1540 ERR("Health error occurred in %s", __func__
);
1542 health_unregister();
1547 * This thread manage application registration.
1549 static void *thread_registration_apps(void *data
)
1551 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1552 uint32_t revents
, nb_fd
;
1553 struct lttng_poll_event events
;
1555 * Get allocated in this thread, enqueued to a global queue, dequeued and
1556 * freed in the manage apps thread.
1558 struct ust_command
*ust_cmd
= NULL
;
1560 DBG("[thread] Manage application registration started");
1562 health_register(HEALTH_TYPE_APP_REG
);
1564 if (testpoint(thread_registration_apps
)) {
1565 goto error_testpoint
;
1568 ret
= lttcomm_listen_unix_sock(apps_sock
);
1574 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1575 * more will be added to this poll set.
1577 ret
= sessiond_set_thread_pollset(&events
, 2);
1579 goto error_create_poll
;
1582 /* Add the application registration socket */
1583 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1585 goto error_poll_add
;
1588 /* Notify all applications to register */
1589 ret
= notify_ust_apps(1);
1591 ERR("Failed to notify applications or create the wait shared memory.\n"
1592 "Execution continues but there might be problem for already\n"
1593 "running applications that wishes to register.");
1597 DBG("Accepting application registration");
1599 /* Inifinite blocking call, waiting for transmission */
1601 health_poll_entry();
1602 ret
= lttng_poll_wait(&events
, -1);
1606 * Restart interrupted system call.
1608 if (errno
== EINTR
) {
1616 for (i
= 0; i
< nb_fd
; i
++) {
1617 health_code_update();
1619 /* Fetch once the poll data */
1620 revents
= LTTNG_POLL_GETEV(&events
, i
);
1621 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1623 /* Thread quit pipe has been closed. Killing thread. */
1624 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1630 /* Event on the registration socket */
1631 if (pollfd
== apps_sock
) {
1632 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1633 ERR("Register apps socket poll error");
1635 } else if (revents
& LPOLLIN
) {
1636 sock
= lttcomm_accept_unix_sock(apps_sock
);
1642 * Set the CLOEXEC flag. Return code is useless because
1643 * either way, the show must go on.
1645 (void) utils_set_fd_cloexec(sock
);
1647 /* Create UST registration command for enqueuing */
1648 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1649 if (ust_cmd
== NULL
) {
1650 PERROR("ust command zmalloc");
1655 * Using message-based transmissions to ensure we don't
1656 * have to deal with partially received messages.
1658 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1660 ERR("Exhausted file descriptors allowed for applications.");
1670 health_code_update();
1671 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1674 /* Close socket of the application. */
1679 lttng_fd_put(LTTNG_FD_APPS
, 1);
1683 health_code_update();
1685 ust_cmd
->sock
= sock
;
1688 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1689 " gid:%d sock:%d name:%s (version %d.%d)",
1690 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1691 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1692 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1693 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1696 * Lock free enqueue the registration request. The red pill
1697 * has been taken! This apps will be part of the *system*.
1699 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1702 * Wake the registration queue futex. Implicit memory
1703 * barrier with the exchange in cds_wfq_enqueue.
1705 futex_nto1_wake(&ust_cmd_queue
.futex
);
1715 ERR("Health error occurred in %s", __func__
);
1718 /* Notify that the registration thread is gone */
1721 if (apps_sock
>= 0) {
1722 ret
= close(apps_sock
);
1732 lttng_fd_put(LTTNG_FD_APPS
, 1);
1734 unlink(apps_unix_sock_path
);
1737 lttng_poll_clean(&events
);
1741 DBG("UST Registration thread cleanup complete");
1742 health_unregister();
1748 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1749 * exec or it will fails.
1751 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1754 struct timespec timeout
;
1756 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1757 consumer_data
->consumer_thread_is_ready
= 0;
1759 /* Setup pthread condition */
1760 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1763 PERROR("pthread_condattr_init consumer data");
1768 * Set the monotonic clock in order to make sure we DO NOT jump in time
1769 * between the clock_gettime() call and the timedwait call. See bug #324
1770 * for a more details and how we noticed it.
1772 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1775 PERROR("pthread_condattr_setclock consumer data");
1779 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1782 PERROR("pthread_cond_init consumer data");
1786 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1789 PERROR("pthread_create consumer");
1794 /* We are about to wait on a pthread condition */
1795 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1797 /* Get time for sem_timedwait absolute timeout */
1798 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1800 * Set the timeout for the condition timed wait even if the clock gettime
1801 * call fails since we might loop on that call and we want to avoid to
1802 * increment the timeout too many times.
1804 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1807 * The following loop COULD be skipped in some conditions so this is why we
1808 * set ret to 0 in order to make sure at least one round of the loop is
1814 * Loop until the condition is reached or when a timeout is reached. Note
1815 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1816 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1817 * possible. This loop does not take any chances and works with both of
1820 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1821 if (clock_ret
< 0) {
1822 PERROR("clock_gettime spawn consumer");
1823 /* Infinite wait for the consumerd thread to be ready */
1824 ret
= pthread_cond_wait(&consumer_data
->cond
,
1825 &consumer_data
->cond_mutex
);
1827 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1828 &consumer_data
->cond_mutex
, &timeout
);
1832 /* Release the pthread condition */
1833 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1837 if (ret
== ETIMEDOUT
) {
1839 * Call has timed out so we kill the kconsumerd_thread and return
1842 ERR("Condition timed out. The consumer thread was never ready."
1844 ret
= pthread_cancel(consumer_data
->thread
);
1846 PERROR("pthread_cancel consumer thread");
1849 PERROR("pthread_cond_wait failed consumer thread");
1854 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1855 if (consumer_data
->pid
== 0) {
1856 ERR("Consumerd did not start");
1857 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1860 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1869 * Join consumer thread
1871 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1875 /* Consumer pid must be a real one. */
1876 if (consumer_data
->pid
> 0) {
1878 ret
= kill(consumer_data
->pid
, SIGTERM
);
1880 ERR("Error killing consumer daemon");
1883 return pthread_join(consumer_data
->thread
, &status
);
1890 * Fork and exec a consumer daemon (consumerd).
1892 * Return pid if successful else -1.
1894 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1898 const char *consumer_to_use
;
1899 const char *verbosity
;
1902 DBG("Spawning consumerd");
1909 if (opt_verbose_consumer
) {
1910 verbosity
= "--verbose";
1912 verbosity
= "--quiet";
1914 switch (consumer_data
->type
) {
1915 case LTTNG_CONSUMER_KERNEL
:
1917 * Find out which consumerd to execute. We will first try the
1918 * 64-bit path, then the sessiond's installation directory, and
1919 * fallback on the 32-bit one,
1921 DBG3("Looking for a kernel consumer at these locations:");
1922 DBG3(" 1) %s", consumerd64_bin
);
1923 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1924 DBG3(" 3) %s", consumerd32_bin
);
1925 if (stat(consumerd64_bin
, &st
) == 0) {
1926 DBG3("Found location #1");
1927 consumer_to_use
= consumerd64_bin
;
1928 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1929 DBG3("Found location #2");
1930 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1931 } else if (stat(consumerd32_bin
, &st
) == 0) {
1932 DBG3("Found location #3");
1933 consumer_to_use
= consumerd32_bin
;
1935 DBG("Could not find any valid consumerd executable");
1938 DBG("Using kernel consumer at: %s", consumer_to_use
);
1939 execl(consumer_to_use
,
1940 "lttng-consumerd", verbosity
, "-k",
1941 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1942 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1945 case LTTNG_CONSUMER64_UST
:
1947 char *tmpnew
= NULL
;
1949 if (consumerd64_libdir
[0] != '\0') {
1953 tmp
= getenv("LD_LIBRARY_PATH");
1957 tmplen
= strlen("LD_LIBRARY_PATH=")
1958 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1959 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1964 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1965 strcat(tmpnew
, consumerd64_libdir
);
1966 if (tmp
[0] != '\0') {
1967 strcat(tmpnew
, ":");
1968 strcat(tmpnew
, tmp
);
1970 ret
= putenv(tmpnew
);
1977 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1978 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1979 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1980 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1982 if (consumerd64_libdir
[0] != '\0') {
1990 case LTTNG_CONSUMER32_UST
:
1992 char *tmpnew
= NULL
;
1994 if (consumerd32_libdir
[0] != '\0') {
1998 tmp
= getenv("LD_LIBRARY_PATH");
2002 tmplen
= strlen("LD_LIBRARY_PATH=")
2003 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2004 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2009 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2010 strcat(tmpnew
, consumerd32_libdir
);
2011 if (tmp
[0] != '\0') {
2012 strcat(tmpnew
, ":");
2013 strcat(tmpnew
, tmp
);
2015 ret
= putenv(tmpnew
);
2022 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2023 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2024 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2025 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2027 if (consumerd32_libdir
[0] != '\0') {
2036 PERROR("unknown consumer type");
2040 PERROR("kernel start consumer exec");
2043 } else if (pid
> 0) {
2046 PERROR("start consumer fork");
2054 * Spawn the consumerd daemon and session daemon thread.
2056 static int start_consumerd(struct consumer_data
*consumer_data
)
2061 * Set the listen() state on the socket since there is a possible race
2062 * between the exec() of the consumer daemon and this call if place in the
2063 * consumer thread. See bug #366 for more details.
2065 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2070 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2071 if (consumer_data
->pid
!= 0) {
2072 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2076 ret
= spawn_consumerd(consumer_data
);
2078 ERR("Spawning consumerd failed");
2079 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2083 /* Setting up the consumer_data pid */
2084 consumer_data
->pid
= ret
;
2085 DBG2("Consumer pid %d", consumer_data
->pid
);
2086 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2088 DBG2("Spawning consumer control thread");
2089 ret
= spawn_consumer_thread(consumer_data
);
2091 ERR("Fatal error spawning consumer control thread");
2099 /* Cleanup already created sockets on error. */
2100 if (consumer_data
->err_sock
>= 0) {
2103 err
= close(consumer_data
->err_sock
);
2105 PERROR("close consumer data error socket");
2112 * Compute health status of each consumer. If one of them is zero (bad
2113 * state), we return 0.
2115 static int check_consumer_health(void)
2119 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2121 DBG3("Health consumer check %d", ret
);
2127 * Setup necessary data for kernel tracer action.
2129 static int init_kernel_tracer(void)
2133 /* Modprobe lttng kernel modules */
2134 ret
= modprobe_lttng_control();
2139 /* Open debugfs lttng */
2140 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2141 if (kernel_tracer_fd
< 0) {
2142 DBG("Failed to open %s", module_proc_lttng
);
2147 /* Validate kernel version */
2148 ret
= kernel_validate_version(kernel_tracer_fd
);
2153 ret
= modprobe_lttng_data();
2158 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2162 modprobe_remove_lttng_control();
2163 ret
= close(kernel_tracer_fd
);
2167 kernel_tracer_fd
= -1;
2168 return LTTNG_ERR_KERN_VERSION
;
2171 ret
= close(kernel_tracer_fd
);
2177 modprobe_remove_lttng_control();
2180 WARN("No kernel tracer available");
2181 kernel_tracer_fd
= -1;
2183 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2185 return LTTNG_ERR_KERN_NA
;
2191 * Copy consumer output from the tracing session to the domain session. The
2192 * function also applies the right modification on a per domain basis for the
2193 * trace files destination directory.
2195 * Should *NOT* be called with RCU read-side lock held.
2197 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2200 const char *dir_name
;
2201 struct consumer_output
*consumer
;
2204 assert(session
->consumer
);
2207 case LTTNG_DOMAIN_KERNEL
:
2208 DBG3("Copying tracing session consumer output in kernel session");
2210 * XXX: We should audit the session creation and what this function
2211 * does "extra" in order to avoid a destroy since this function is used
2212 * in the domain session creation (kernel and ust) only. Same for UST
2215 if (session
->kernel_session
->consumer
) {
2216 consumer_destroy_output(session
->kernel_session
->consumer
);
2218 session
->kernel_session
->consumer
=
2219 consumer_copy_output(session
->consumer
);
2220 /* Ease our life a bit for the next part */
2221 consumer
= session
->kernel_session
->consumer
;
2222 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2224 case LTTNG_DOMAIN_UST
:
2225 DBG3("Copying tracing session consumer output in UST session");
2226 if (session
->ust_session
->consumer
) {
2227 consumer_destroy_output(session
->ust_session
->consumer
);
2229 session
->ust_session
->consumer
=
2230 consumer_copy_output(session
->consumer
);
2231 /* Ease our life a bit for the next part */
2232 consumer
= session
->ust_session
->consumer
;
2233 dir_name
= DEFAULT_UST_TRACE_DIR
;
2236 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2240 /* Append correct directory to subdir */
2241 strncat(consumer
->subdir
, dir_name
,
2242 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2243 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2252 * Create an UST session and add it to the session ust list.
2254 * Should *NOT* be called with RCU read-side lock held.
2256 static int create_ust_session(struct ltt_session
*session
,
2257 struct lttng_domain
*domain
)
2260 struct ltt_ust_session
*lus
= NULL
;
2264 assert(session
->consumer
);
2266 switch (domain
->type
) {
2267 case LTTNG_DOMAIN_UST
:
2270 ERR("Unknown UST domain on create session %d", domain
->type
);
2271 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2275 DBG("Creating UST session");
2277 lus
= trace_ust_create_session(session
->id
);
2279 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2283 lus
->uid
= session
->uid
;
2284 lus
->gid
= session
->gid
;
2285 session
->ust_session
= lus
;
2287 /* Copy session output to the newly created UST session */
2288 ret
= copy_session_consumer(domain
->type
, session
);
2289 if (ret
!= LTTNG_OK
) {
2297 session
->ust_session
= NULL
;
2302 * Create a kernel tracer session then create the default channel.
2304 static int create_kernel_session(struct ltt_session
*session
)
2308 DBG("Creating kernel session");
2310 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2312 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2316 /* Code flow safety */
2317 assert(session
->kernel_session
);
2319 /* Copy session output to the newly created Kernel session */
2320 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2321 if (ret
!= LTTNG_OK
) {
2325 /* Create directory(ies) on local filesystem. */
2326 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2327 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2328 ret
= run_as_mkdir_recursive(
2329 session
->kernel_session
->consumer
->dst
.trace_path
,
2330 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2332 if (ret
!= -EEXIST
) {
2333 ERR("Trace directory creation error");
2339 session
->kernel_session
->uid
= session
->uid
;
2340 session
->kernel_session
->gid
= session
->gid
;
2345 trace_kernel_destroy_session(session
->kernel_session
);
2346 session
->kernel_session
= NULL
;
2351 * Count number of session permitted by uid/gid.
2353 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2356 struct ltt_session
*session
;
2358 DBG("Counting number of available session for UID %d GID %d",
2360 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2362 * Only list the sessions the user can control.
2364 if (!session_access_ok(session
, uid
, gid
)) {
2373 * Process the command requested by the lttng client within the command
2374 * context structure. This function make sure that the return structure (llm)
2375 * is set and ready for transmission before returning.
2377 * Return any error encountered or 0 for success.
2379 * "sock" is only used for special-case var. len data.
2381 * Should *NOT* be called with RCU read-side lock held.
2383 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2387 int need_tracing_session
= 1;
2390 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2394 switch (cmd_ctx
->lsm
->cmd_type
) {
2395 case LTTNG_CREATE_SESSION
:
2396 case LTTNG_DESTROY_SESSION
:
2397 case LTTNG_LIST_SESSIONS
:
2398 case LTTNG_LIST_DOMAINS
:
2399 case LTTNG_START_TRACE
:
2400 case LTTNG_STOP_TRACE
:
2401 case LTTNG_DATA_PENDING
:
2408 if (opt_no_kernel
&& need_domain
2409 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2411 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2413 ret
= LTTNG_ERR_KERN_NA
;
2418 /* Deny register consumer if we already have a spawned consumer. */
2419 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2420 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2421 if (kconsumer_data
.pid
> 0) {
2422 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2423 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2426 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2430 * Check for command that don't needs to allocate a returned payload. We do
2431 * this here so we don't have to make the call for no payload at each
2434 switch(cmd_ctx
->lsm
->cmd_type
) {
2435 case LTTNG_LIST_SESSIONS
:
2436 case LTTNG_LIST_TRACEPOINTS
:
2437 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2438 case LTTNG_LIST_DOMAINS
:
2439 case LTTNG_LIST_CHANNELS
:
2440 case LTTNG_LIST_EVENTS
:
2443 /* Setup lttng message with no payload */
2444 ret
= setup_lttng_msg(cmd_ctx
, 0);
2446 /* This label does not try to unlock the session */
2447 goto init_setup_error
;
2451 /* Commands that DO NOT need a session. */
2452 switch (cmd_ctx
->lsm
->cmd_type
) {
2453 case LTTNG_CREATE_SESSION
:
2454 case LTTNG_CALIBRATE
:
2455 case LTTNG_LIST_SESSIONS
:
2456 case LTTNG_LIST_TRACEPOINTS
:
2457 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2458 need_tracing_session
= 0;
2461 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2463 * We keep the session list lock across _all_ commands
2464 * for now, because the per-session lock does not
2465 * handle teardown properly.
2467 session_lock_list();
2468 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2469 if (cmd_ctx
->session
== NULL
) {
2470 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2471 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2473 /* If no session name specified */
2474 ret
= LTTNG_ERR_SELECT_SESS
;
2478 /* Acquire lock for the session */
2479 session_lock(cmd_ctx
->session
);
2489 * Check domain type for specific "pre-action".
2491 switch (cmd_ctx
->lsm
->domain
.type
) {
2492 case LTTNG_DOMAIN_KERNEL
:
2494 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2498 /* Kernel tracer check */
2499 if (kernel_tracer_fd
== -1) {
2500 /* Basically, load kernel tracer modules */
2501 ret
= init_kernel_tracer();
2507 /* Consumer is in an ERROR state. Report back to client */
2508 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2509 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2513 /* Need a session for kernel command */
2514 if (need_tracing_session
) {
2515 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2516 ret
= create_kernel_session(cmd_ctx
->session
);
2518 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2523 /* Start the kernel consumer daemon */
2524 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2525 if (kconsumer_data
.pid
== 0 &&
2526 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2527 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2528 ret
= start_consumerd(&kconsumer_data
);
2530 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2533 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2535 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2539 * The consumer was just spawned so we need to add the socket to
2540 * the consumer output of the session if exist.
2542 ret
= consumer_create_socket(&kconsumer_data
,
2543 cmd_ctx
->session
->kernel_session
->consumer
);
2550 case LTTNG_DOMAIN_UST
:
2552 /* Consumer is in an ERROR state. Report back to client */
2553 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2554 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2558 if (need_tracing_session
) {
2559 /* Create UST session if none exist. */
2560 if (cmd_ctx
->session
->ust_session
== NULL
) {
2561 ret
= create_ust_session(cmd_ctx
->session
,
2562 &cmd_ctx
->lsm
->domain
);
2563 if (ret
!= LTTNG_OK
) {
2568 /* Start the UST consumer daemons */
2570 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2571 if (consumerd64_bin
[0] != '\0' &&
2572 ustconsumer64_data
.pid
== 0 &&
2573 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2574 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2575 ret
= start_consumerd(&ustconsumer64_data
);
2577 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2578 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2582 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2583 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2585 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2589 * Setup socket for consumer 64 bit. No need for atomic access
2590 * since it was set above and can ONLY be set in this thread.
2592 ret
= consumer_create_socket(&ustconsumer64_data
,
2593 cmd_ctx
->session
->ust_session
->consumer
);
2599 if (consumerd32_bin
[0] != '\0' &&
2600 ustconsumer32_data
.pid
== 0 &&
2601 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2602 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2603 ret
= start_consumerd(&ustconsumer32_data
);
2605 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2606 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2610 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2611 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2613 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2617 * Setup socket for consumer 64 bit. No need for atomic access
2618 * since it was set above and can ONLY be set in this thread.
2620 ret
= consumer_create_socket(&ustconsumer32_data
,
2621 cmd_ctx
->session
->ust_session
->consumer
);
2633 /* Validate consumer daemon state when start/stop trace command */
2634 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2635 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2636 switch (cmd_ctx
->lsm
->domain
.type
) {
2637 case LTTNG_DOMAIN_UST
:
2638 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2639 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2643 case LTTNG_DOMAIN_KERNEL
:
2644 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2645 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2653 * Check that the UID or GID match that of the tracing session.
2654 * The root user can interact with all sessions.
2656 if (need_tracing_session
) {
2657 if (!session_access_ok(cmd_ctx
->session
,
2658 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2659 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2660 ret
= LTTNG_ERR_EPERM
;
2666 * Send relayd information to consumer as soon as we have a domain and a
2669 if (cmd_ctx
->session
&& need_domain
) {
2671 * Setup relayd if not done yet. If the relayd information was already
2672 * sent to the consumer, this call will gracefully return.
2674 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2675 if (ret
!= LTTNG_OK
) {
2680 /* Process by command type */
2681 switch (cmd_ctx
->lsm
->cmd_type
) {
2682 case LTTNG_ADD_CONTEXT
:
2684 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2685 cmd_ctx
->lsm
->u
.context
.channel_name
,
2686 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2689 case LTTNG_DISABLE_CHANNEL
:
2691 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2692 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2695 case LTTNG_DISABLE_EVENT
:
2697 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2698 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2699 cmd_ctx
->lsm
->u
.disable
.name
);
2702 case LTTNG_DISABLE_ALL_EVENT
:
2704 DBG("Disabling all events");
2706 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2707 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2710 case LTTNG_ENABLE_CHANNEL
:
2712 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2713 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2716 case LTTNG_ENABLE_EVENT
:
2718 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2719 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2720 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2723 case LTTNG_ENABLE_ALL_EVENT
:
2725 DBG("Enabling all events");
2727 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2728 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2729 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2732 case LTTNG_LIST_TRACEPOINTS
:
2734 struct lttng_event
*events
;
2737 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2738 if (nb_events
< 0) {
2739 /* Return value is a negative lttng_error_code. */
2745 * Setup lttng message with payload size set to the event list size in
2746 * bytes and then copy list into the llm payload.
2748 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2754 /* Copy event list into message payload */
2755 memcpy(cmd_ctx
->llm
->payload
, events
,
2756 sizeof(struct lttng_event
) * nb_events
);
2763 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2765 struct lttng_event_field
*fields
;
2768 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2770 if (nb_fields
< 0) {
2771 /* Return value is a negative lttng_error_code. */
2777 * Setup lttng message with payload size set to the event list size in
2778 * bytes and then copy list into the llm payload.
2780 ret
= setup_lttng_msg(cmd_ctx
,
2781 sizeof(struct lttng_event_field
) * nb_fields
);
2787 /* Copy event list into message payload */
2788 memcpy(cmd_ctx
->llm
->payload
, fields
,
2789 sizeof(struct lttng_event_field
) * nb_fields
);
2796 case LTTNG_SET_CONSUMER_URI
:
2799 struct lttng_uri
*uris
;
2801 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2802 len
= nb_uri
* sizeof(struct lttng_uri
);
2805 ret
= LTTNG_ERR_INVALID
;
2809 uris
= zmalloc(len
);
2811 ret
= LTTNG_ERR_FATAL
;
2815 /* Receive variable len data */
2816 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2817 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2819 DBG("No URIs received from client... continuing");
2821 ret
= LTTNG_ERR_SESSION_FAIL
;
2826 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2828 if (ret
!= LTTNG_OK
) {
2834 * XXX: 0 means that this URI should be applied on the session. Should
2835 * be a DOMAIN enuam.
2837 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2838 /* Add the URI for the UST session if a consumer is present. */
2839 if (cmd_ctx
->session
->ust_session
&&
2840 cmd_ctx
->session
->ust_session
->consumer
) {
2841 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2843 } else if (cmd_ctx
->session
->kernel_session
&&
2844 cmd_ctx
->session
->kernel_session
->consumer
) {
2845 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2846 cmd_ctx
->session
, nb_uri
, uris
);
2854 case LTTNG_START_TRACE
:
2856 ret
= cmd_start_trace(cmd_ctx
->session
);
2859 case LTTNG_STOP_TRACE
:
2861 ret
= cmd_stop_trace(cmd_ctx
->session
);
2864 case LTTNG_CREATE_SESSION
:
2867 struct lttng_uri
*uris
= NULL
;
2869 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2870 len
= nb_uri
* sizeof(struct lttng_uri
);
2873 uris
= zmalloc(len
);
2875 ret
= LTTNG_ERR_FATAL
;
2879 /* Receive variable len data */
2880 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2881 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2883 DBG("No URIs received from client... continuing");
2885 ret
= LTTNG_ERR_SESSION_FAIL
;
2890 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2891 DBG("Creating session with ONE network URI is a bad call");
2892 ret
= LTTNG_ERR_SESSION_FAIL
;
2898 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2905 case LTTNG_DESTROY_SESSION
:
2907 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2909 /* Set session to NULL so we do not unlock it after free. */
2910 cmd_ctx
->session
= NULL
;
2913 case LTTNG_LIST_DOMAINS
:
2916 struct lttng_domain
*domains
;
2918 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2920 /* Return value is a negative lttng_error_code. */
2925 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2931 /* Copy event list into message payload */
2932 memcpy(cmd_ctx
->llm
->payload
, domains
,
2933 nb_dom
* sizeof(struct lttng_domain
));
2940 case LTTNG_LIST_CHANNELS
:
2943 struct lttng_channel
*channels
;
2945 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2946 cmd_ctx
->session
, &channels
);
2948 /* Return value is a negative lttng_error_code. */
2953 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2959 /* Copy event list into message payload */
2960 memcpy(cmd_ctx
->llm
->payload
, channels
,
2961 nb_chan
* sizeof(struct lttng_channel
));
2968 case LTTNG_LIST_EVENTS
:
2971 struct lttng_event
*events
= NULL
;
2973 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2974 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2976 /* Return value is a negative lttng_error_code. */
2981 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2987 /* Copy event list into message payload */
2988 memcpy(cmd_ctx
->llm
->payload
, events
,
2989 nb_event
* sizeof(struct lttng_event
));
2996 case LTTNG_LIST_SESSIONS
:
2998 unsigned int nr_sessions
;
3000 session_lock_list();
3001 nr_sessions
= lttng_sessions_count(
3002 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3003 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3005 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3007 session_unlock_list();
3011 /* Filled the session array */
3012 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3013 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3014 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3016 session_unlock_list();
3021 case LTTNG_CALIBRATE
:
3023 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3024 &cmd_ctx
->lsm
->u
.calibrate
);
3027 case LTTNG_REGISTER_CONSUMER
:
3029 struct consumer_data
*cdata
;
3031 switch (cmd_ctx
->lsm
->domain
.type
) {
3032 case LTTNG_DOMAIN_KERNEL
:
3033 cdata
= &kconsumer_data
;
3036 ret
= LTTNG_ERR_UND
;
3040 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3041 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3044 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3046 struct lttng_filter_bytecode
*bytecode
;
3048 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3049 ret
= LTTNG_ERR_FILTER_INVAL
;
3052 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3053 ret
= LTTNG_ERR_FILTER_INVAL
;
3056 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3058 ret
= LTTNG_ERR_FILTER_NOMEM
;
3061 /* Receive var. len. data */
3062 DBG("Receiving var len data from client ...");
3063 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3064 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3066 DBG("Nothing recv() from client var len data... continuing");
3068 ret
= LTTNG_ERR_FILTER_INVAL
;
3072 if (bytecode
->len
+ sizeof(*bytecode
)
3073 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3075 ret
= LTTNG_ERR_FILTER_INVAL
;
3079 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3080 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3081 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3084 case LTTNG_DATA_PENDING
:
3086 ret
= cmd_data_pending(cmd_ctx
->session
);
3090 ret
= LTTNG_ERR_UND
;
3095 if (cmd_ctx
->llm
== NULL
) {
3096 DBG("Missing llm structure. Allocating one.");
3097 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3101 /* Set return code */
3102 cmd_ctx
->llm
->ret_code
= ret
;
3104 if (cmd_ctx
->session
) {
3105 session_unlock(cmd_ctx
->session
);
3107 if (need_tracing_session
) {
3108 session_unlock_list();
3115 * Thread managing health check socket.
3117 static void *thread_manage_health(void *data
)
3119 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3120 uint32_t revents
, nb_fd
;
3121 struct lttng_poll_event events
;
3122 struct lttcomm_health_msg msg
;
3123 struct lttcomm_health_data reply
;
3125 DBG("[thread] Manage health check started");
3127 rcu_register_thread();
3129 /* We might hit an error path before this is created. */
3130 lttng_poll_init(&events
);
3132 /* Create unix socket */
3133 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3135 ERR("Unable to create health check Unix socket");
3141 * Set the CLOEXEC flag. Return code is useless because either way, the
3144 (void) utils_set_fd_cloexec(sock
);
3146 ret
= lttcomm_listen_unix_sock(sock
);
3152 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3153 * more will be added to this poll set.
3155 ret
= sessiond_set_thread_pollset(&events
, 2);
3160 /* Add the application registration socket */
3161 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3167 DBG("Health check ready");
3169 /* Inifinite blocking call, waiting for transmission */
3171 ret
= lttng_poll_wait(&events
, -1);
3174 * Restart interrupted system call.
3176 if (errno
== EINTR
) {
3184 for (i
= 0; i
< nb_fd
; i
++) {
3185 /* Fetch once the poll data */
3186 revents
= LTTNG_POLL_GETEV(&events
, i
);
3187 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3189 /* Thread quit pipe has been closed. Killing thread. */
3190 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3196 /* Event on the registration socket */
3197 if (pollfd
== sock
) {
3198 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3199 ERR("Health socket poll error");
3205 new_sock
= lttcomm_accept_unix_sock(sock
);
3211 * Set the CLOEXEC flag. Return code is useless because either way, the
3214 (void) utils_set_fd_cloexec(new_sock
);
3216 DBG("Receiving data from client for health...");
3217 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3219 DBG("Nothing recv() from client... continuing");
3220 ret
= close(new_sock
);
3228 rcu_thread_online();
3230 switch (msg
.component
) {
3231 case LTTNG_HEALTH_CMD
:
3232 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3234 case LTTNG_HEALTH_APP_MANAGE
:
3235 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3237 case LTTNG_HEALTH_APP_REG
:
3238 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3240 case LTTNG_HEALTH_KERNEL
:
3241 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3243 case LTTNG_HEALTH_CONSUMER
:
3244 reply
.ret_code
= check_consumer_health();
3246 case LTTNG_HEALTH_HT_CLEANUP
:
3247 reply
.ret_code
= health_check_state(HEALTH_TYPE_HT_CLEANUP
);
3249 case LTTNG_HEALTH_APP_MANAGE_NOTIFY
:
3250 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
);
3252 case LTTNG_HEALTH_APP_REG_DISPATCH
:
3253 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3255 case LTTNG_HEALTH_ALL
:
3257 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3258 health_check_state(HEALTH_TYPE_APP_REG
) &&
3259 health_check_state(HEALTH_TYPE_CMD
) &&
3260 health_check_state(HEALTH_TYPE_KERNEL
) &&
3261 check_consumer_health() &&
3262 health_check_state(HEALTH_TYPE_HT_CLEANUP
) &&
3263 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY
) &&
3264 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH
);
3267 reply
.ret_code
= LTTNG_ERR_UND
;
3272 * Flip ret value since 0 is a success and 1 indicates a bad health for
3273 * the client where in the sessiond it is the opposite. Again, this is
3274 * just to make things easier for us poor developer which enjoy a lot
3277 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3278 reply
.ret_code
= !reply
.ret_code
;
3281 DBG2("Health check return value %d", reply
.ret_code
);
3283 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3285 ERR("Failed to send health data back to client");
3288 /* End of transmission */
3289 ret
= close(new_sock
);
3299 ERR("Health error occurred in %s", __func__
);
3301 DBG("Health check thread dying");
3302 unlink(health_unix_sock_path
);
3310 lttng_poll_clean(&events
);
3312 rcu_unregister_thread();
3317 * This thread manage all clients request using the unix client socket for
3320 static void *thread_manage_clients(void *data
)
3322 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3324 uint32_t revents
, nb_fd
;
3325 struct command_ctx
*cmd_ctx
= NULL
;
3326 struct lttng_poll_event events
;
3328 DBG("[thread] Manage client started");
3330 rcu_register_thread();
3332 health_register(HEALTH_TYPE_CMD
);
3334 if (testpoint(thread_manage_clients
)) {
3335 goto error_testpoint
;
3338 health_code_update();
3340 ret
= lttcomm_listen_unix_sock(client_sock
);
3346 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3347 * more will be added to this poll set.
3349 ret
= sessiond_set_thread_pollset(&events
, 2);
3351 goto error_create_poll
;
3354 /* Add the application registration socket */
3355 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3361 * Notify parent pid that we are ready to accept command for client side.
3363 if (opt_sig_parent
) {
3364 kill(ppid
, SIGUSR1
);
3367 if (testpoint(thread_manage_clients_before_loop
)) {
3371 health_code_update();
3374 DBG("Accepting client command ...");
3376 /* Inifinite blocking call, waiting for transmission */
3378 health_poll_entry();
3379 ret
= lttng_poll_wait(&events
, -1);
3383 * Restart interrupted system call.
3385 if (errno
== EINTR
) {
3393 for (i
= 0; i
< nb_fd
; i
++) {
3394 /* Fetch once the poll data */
3395 revents
= LTTNG_POLL_GETEV(&events
, i
);
3396 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3398 health_code_update();
3400 /* Thread quit pipe has been closed. Killing thread. */
3401 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3407 /* Event on the registration socket */
3408 if (pollfd
== client_sock
) {
3409 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3410 ERR("Client socket poll error");
3416 DBG("Wait for client response");
3418 health_code_update();
3420 sock
= lttcomm_accept_unix_sock(client_sock
);
3426 * Set the CLOEXEC flag. Return code is useless because either way, the
3429 (void) utils_set_fd_cloexec(sock
);
3431 /* Set socket option for credentials retrieval */
3432 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3437 /* Allocate context command to process the client request */
3438 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3439 if (cmd_ctx
== NULL
) {
3440 PERROR("zmalloc cmd_ctx");
3444 /* Allocate data buffer for reception */
3445 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3446 if (cmd_ctx
->lsm
== NULL
) {
3447 PERROR("zmalloc cmd_ctx->lsm");
3451 cmd_ctx
->llm
= NULL
;
3452 cmd_ctx
->session
= NULL
;
3454 health_code_update();
3457 * Data is received from the lttng client. The struct
3458 * lttcomm_session_msg (lsm) contains the command and data request of
3461 DBG("Receiving data from client ...");
3462 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3463 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3465 DBG("Nothing recv() from client... continuing");
3471 clean_command_ctx(&cmd_ctx
);
3475 health_code_update();
3477 // TODO: Validate cmd_ctx including sanity check for
3478 // security purpose.
3480 rcu_thread_online();
3482 * This function dispatch the work to the kernel or userspace tracer
3483 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3484 * informations for the client. The command context struct contains
3485 * everything this function may needs.
3487 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3488 rcu_thread_offline();
3496 * TODO: Inform client somehow of the fatal error. At
3497 * this point, ret < 0 means that a zmalloc failed
3498 * (ENOMEM). Error detected but still accept
3499 * command, unless a socket error has been
3502 clean_command_ctx(&cmd_ctx
);
3506 health_code_update();
3508 DBG("Sending response (size: %d, retcode: %s)",
3509 cmd_ctx
->lttng_msg_size
,
3510 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3511 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3513 ERR("Failed to send data back to client");
3516 /* End of transmission */
3523 clean_command_ctx(&cmd_ctx
);
3525 health_code_update();
3537 lttng_poll_clean(&events
);
3538 clean_command_ctx(&cmd_ctx
);
3543 unlink(client_unix_sock_path
);
3544 if (client_sock
>= 0) {
3545 ret
= close(client_sock
);
3553 ERR("Health error occurred in %s", __func__
);
3556 health_unregister();
3558 DBG("Client thread dying");
3560 rcu_unregister_thread();
3566 * usage function on stderr
3568 static void usage(void)
3570 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3571 fprintf(stderr
, " -h, --help Display this usage.\n");
3572 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3573 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3574 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3575 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3576 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3577 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3578 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3579 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3580 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3581 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3582 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3583 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3584 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3585 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3586 fprintf(stderr
, " -V, --version Show version number.\n");
3587 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3588 fprintf(stderr
, " -q, --quiet No output at all.\n");
3589 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3590 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3591 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3592 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3596 * daemon argument parsing
3598 static int parse_args(int argc
, char **argv
)
3602 static struct option long_options
[] = {
3603 { "client-sock", 1, 0, 'c' },
3604 { "apps-sock", 1, 0, 'a' },
3605 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3606 { "kconsumerd-err-sock", 1, 0, 'E' },
3607 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3608 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3609 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3610 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3611 { "consumerd32-path", 1, 0, 'u' },
3612 { "consumerd32-libdir", 1, 0, 'U' },
3613 { "consumerd64-path", 1, 0, 't' },
3614 { "consumerd64-libdir", 1, 0, 'T' },
3615 { "daemonize", 0, 0, 'd' },
3616 { "sig-parent", 0, 0, 'S' },
3617 { "help", 0, 0, 'h' },
3618 { "group", 1, 0, 'g' },
3619 { "version", 0, 0, 'V' },
3620 { "quiet", 0, 0, 'q' },
3621 { "verbose", 0, 0, 'v' },
3622 { "verbose-consumer", 0, 0, 'Z' },
3623 { "no-kernel", 0, 0, 'N' },
3624 { "pidfile", 1, 0, 'p' },
3629 int option_index
= 0;
3630 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3631 long_options
, &option_index
);
3638 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3640 fprintf(stderr
, " with arg %s\n", optarg
);
3644 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3647 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3653 opt_tracing_group
= optarg
;
3659 fprintf(stdout
, "%s\n", VERSION
);
3665 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3668 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3671 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3674 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3677 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3680 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3686 lttng_opt_quiet
= 1;
3689 /* Verbose level can increase using multiple -v */
3690 lttng_opt_verbose
+= 1;
3693 opt_verbose_consumer
+= 1;
3696 consumerd32_bin
= optarg
;
3699 consumerd32_libdir
= optarg
;
3702 consumerd64_bin
= optarg
;
3705 consumerd64_libdir
= optarg
;
3708 opt_pidfile
= optarg
;
3711 /* Unknown option or other error.
3712 * Error is printed by getopt, just return */
3721 * Creates the two needed socket by the daemon.
3722 * apps_sock - The communication socket for all UST apps.
3723 * client_sock - The communication of the cli tool (lttng).
3725 static int init_daemon_socket(void)
3730 old_umask
= umask(0);
3732 /* Create client tool unix socket */
3733 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3734 if (client_sock
< 0) {
3735 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3740 /* Set the cloexec flag */
3741 ret
= utils_set_fd_cloexec(client_sock
);
3743 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3744 "Continuing but note that the consumer daemon will have a "
3745 "reference to this socket on exec()", client_sock
);
3748 /* File permission MUST be 660 */
3749 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3751 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3756 /* Create the application unix socket */
3757 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3758 if (apps_sock
< 0) {
3759 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3764 /* Set the cloexec flag */
3765 ret
= utils_set_fd_cloexec(apps_sock
);
3767 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3768 "Continuing but note that the consumer daemon will have a "
3769 "reference to this socket on exec()", apps_sock
);
3772 /* File permission MUST be 666 */
3773 ret
= chmod(apps_unix_sock_path
,
3774 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3776 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3781 DBG3("Session daemon client socket %d and application socket %d created",
3782 client_sock
, apps_sock
);
3790 * Check if the global socket is available, and if a daemon is answering at the
3791 * other side. If yes, error is returned.
3793 static int check_existing_daemon(void)
3795 /* Is there anybody out there ? */
3796 if (lttng_session_daemon_alive()) {
3804 * Set the tracing group gid onto the client socket.
3806 * Race window between mkdir and chown is OK because we are going from more
3807 * permissive (root.root) to less permissive (root.tracing).
3809 static int set_permissions(char *rundir
)
3814 ret
= allowed_group();
3816 WARN("No tracing group detected");
3823 /* Set lttng run dir */
3824 ret
= chown(rundir
, 0, gid
);
3826 ERR("Unable to set group on %s", rundir
);
3830 /* Ensure tracing group can search the run dir */
3831 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3833 ERR("Unable to set permissions on %s", rundir
);
3837 /* lttng client socket path */
3838 ret
= chown(client_unix_sock_path
, 0, gid
);
3840 ERR("Unable to set group on %s", client_unix_sock_path
);
3844 /* kconsumer error socket path */
3845 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3847 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3851 /* 64-bit ustconsumer error socket path */
3852 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3854 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3858 /* 32-bit ustconsumer compat32 error socket path */
3859 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3861 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3865 DBG("All permissions are set");
3872 * Create the lttng run directory needed for all global sockets and pipe.
3874 static int create_lttng_rundir(const char *rundir
)
3878 DBG3("Creating LTTng run directory: %s", rundir
);
3880 ret
= mkdir(rundir
, S_IRWXU
);
3882 if (errno
!= EEXIST
) {
3883 ERR("Unable to create %s", rundir
);
3895 * Setup sockets and directory needed by the kconsumerd communication with the
3898 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3902 char path
[PATH_MAX
];
3904 switch (consumer_data
->type
) {
3905 case LTTNG_CONSUMER_KERNEL
:
3906 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3908 case LTTNG_CONSUMER64_UST
:
3909 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3911 case LTTNG_CONSUMER32_UST
:
3912 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3915 ERR("Consumer type unknown");
3920 DBG2("Creating consumer directory: %s", path
);
3922 ret
= mkdir(path
, S_IRWXU
);
3924 if (errno
!= EEXIST
) {
3926 ERR("Failed to create %s", path
);
3932 /* Create the kconsumerd error unix socket */
3933 consumer_data
->err_sock
=
3934 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3935 if (consumer_data
->err_sock
< 0) {
3936 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3941 /* File permission MUST be 660 */
3942 ret
= chmod(consumer_data
->err_unix_sock_path
,
3943 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3945 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3955 * Signal handler for the daemon
3957 * Simply stop all worker threads, leaving main() return gracefully after
3958 * joining all threads and calling cleanup().
3960 static void sighandler(int sig
)
3964 DBG("SIGPIPE caught");
3967 DBG("SIGINT caught");
3971 DBG("SIGTERM caught");
3980 * Setup signal handler for :
3981 * SIGINT, SIGTERM, SIGPIPE
3983 static int set_signal_handler(void)
3986 struct sigaction sa
;
3989 if ((ret
= sigemptyset(&sigset
)) < 0) {
3990 PERROR("sigemptyset");
3994 sa
.sa_handler
= sighandler
;
3995 sa
.sa_mask
= sigset
;
3997 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3998 PERROR("sigaction");
4002 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4003 PERROR("sigaction");
4007 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4008 PERROR("sigaction");
4012 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4018 * Set open files limit to unlimited. This daemon can open a large number of
4019 * file descriptors in order to consumer multiple kernel traces.
4021 static void set_ulimit(void)
4026 /* The kernel does not allowed an infinite limit for open files */
4027 lim
.rlim_cur
= 65535;
4028 lim
.rlim_max
= 65535;
4030 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4032 PERROR("failed to set open files limit");
4037 * Write pidfile using the rundir and opt_pidfile.
4039 static void write_pidfile(void)
4042 char pidfile_path
[PATH_MAX
];
4047 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4049 /* Build pidfile path from rundir and opt_pidfile. */
4050 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4051 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4053 PERROR("snprintf pidfile path");
4059 * Create pid file in rundir. Return value is of no importance. The
4060 * execution will continue even though we are not able to write the file.
4062 (void) utils_create_pid_file(getpid(), pidfile_path
);
4071 int main(int argc
, char **argv
)
4075 const char *home_path
, *env_app_timeout
;
4077 init_kernel_workarounds();
4079 rcu_register_thread();
4081 setup_consumerd_path();
4083 page_size
= sysconf(_SC_PAGESIZE
);
4084 if (page_size
< 0) {
4085 PERROR("sysconf _SC_PAGESIZE");
4086 page_size
= LONG_MAX
;
4087 WARN("Fallback page size to %ld", page_size
);
4090 /* Parse arguments */
4092 if ((ret
= parse_args(argc
, argv
)) < 0) {
4102 * child: setsid, close FD 0, 1, 2, chdir /
4103 * parent: exit (if fork is successful)
4111 * We are in the child. Make sure all other file
4112 * descriptors are closed, in case we are called with
4113 * more opened file descriptors than the standard ones.
4115 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4120 /* Create thread quit pipe */
4121 if ((ret
= init_thread_quit_pipe()) < 0) {
4125 /* Check if daemon is UID = 0 */
4126 is_root
= !getuid();
4129 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4131 /* Create global run dir with root access */
4132 ret
= create_lttng_rundir(rundir
);
4137 if (strlen(apps_unix_sock_path
) == 0) {
4138 snprintf(apps_unix_sock_path
, PATH_MAX
,
4139 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4142 if (strlen(client_unix_sock_path
) == 0) {
4143 snprintf(client_unix_sock_path
, PATH_MAX
,
4144 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4147 /* Set global SHM for ust */
4148 if (strlen(wait_shm_path
) == 0) {
4149 snprintf(wait_shm_path
, PATH_MAX
,
4150 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4153 if (strlen(health_unix_sock_path
) == 0) {
4154 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4155 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4158 /* Setup kernel consumerd path */
4159 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4160 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4161 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4162 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4164 DBG2("Kernel consumer err path: %s",
4165 kconsumer_data
.err_unix_sock_path
);
4166 DBG2("Kernel consumer cmd path: %s",
4167 kconsumer_data
.cmd_unix_sock_path
);
4169 home_path
= get_home_dir();
4170 if (home_path
== NULL
) {
4171 /* TODO: Add --socket PATH option */
4172 ERR("Can't get HOME directory for sockets creation.");
4178 * Create rundir from home path. This will create something like
4181 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4187 ret
= create_lttng_rundir(rundir
);
4192 if (strlen(apps_unix_sock_path
) == 0) {
4193 snprintf(apps_unix_sock_path
, PATH_MAX
,
4194 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4197 /* Set the cli tool unix socket path */
4198 if (strlen(client_unix_sock_path
) == 0) {
4199 snprintf(client_unix_sock_path
, PATH_MAX
,
4200 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4203 /* Set global SHM for ust */
4204 if (strlen(wait_shm_path
) == 0) {
4205 snprintf(wait_shm_path
, PATH_MAX
,
4206 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4209 /* Set health check Unix path */
4210 if (strlen(health_unix_sock_path
) == 0) {
4211 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4212 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4216 /* Set consumer initial state */
4217 kernel_consumerd_state
= CONSUMER_STOPPED
;
4218 ust_consumerd_state
= CONSUMER_STOPPED
;
4220 DBG("Client socket path %s", client_unix_sock_path
);
4221 DBG("Application socket path %s", apps_unix_sock_path
);
4222 DBG("Application wait path %s", wait_shm_path
);
4223 DBG("LTTng run directory path: %s", rundir
);
4225 /* 32 bits consumerd path setup */
4226 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4227 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4228 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4229 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4231 DBG2("UST consumer 32 bits err path: %s",
4232 ustconsumer32_data
.err_unix_sock_path
);
4233 DBG2("UST consumer 32 bits cmd path: %s",
4234 ustconsumer32_data
.cmd_unix_sock_path
);
4236 /* 64 bits consumerd path setup */
4237 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4238 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4239 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4240 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4242 DBG2("UST consumer 64 bits err path: %s",
4243 ustconsumer64_data
.err_unix_sock_path
);
4244 DBG2("UST consumer 64 bits cmd path: %s",
4245 ustconsumer64_data
.cmd_unix_sock_path
);
4248 * See if daemon already exist.
4250 if ((ret
= check_existing_daemon()) < 0) {
4251 ERR("Already running daemon.\n");
4253 * We do not goto exit because we must not cleanup()
4254 * because a daemon is already running.
4260 * Init UST app hash table. Alloc hash table before this point since
4261 * cleanup() can get called after that point.
4265 /* After this point, we can safely call cleanup() with "goto exit" */
4268 * These actions must be executed as root. We do that *after* setting up
4269 * the sockets path because we MUST make the check for another daemon using
4270 * those paths *before* trying to set the kernel consumer sockets and init
4274 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4279 /* Setup kernel tracer */
4280 if (!opt_no_kernel
) {
4281 init_kernel_tracer();
4284 /* Set ulimit for open files */
4287 /* init lttng_fd tracking must be done after set_ulimit. */
4290 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4295 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4300 if ((ret
= set_signal_handler()) < 0) {
4304 /* Setup the needed unix socket */
4305 if ((ret
= init_daemon_socket()) < 0) {
4309 /* Set credentials to socket */
4310 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4314 /* Get parent pid if -S, --sig-parent is specified. */
4315 if (opt_sig_parent
) {
4319 /* Setup the kernel pipe for waking up the kernel thread */
4320 if (is_root
&& !opt_no_kernel
) {
4321 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4326 /* Setup the thread ht_cleanup communication pipe. */
4327 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4331 /* Setup the thread apps communication pipe. */
4332 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4336 /* Setup the thread apps notify communication pipe. */
4337 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4341 /* Initialize global buffer per UID and PID registry. */
4342 buffer_reg_init_uid_registry();
4343 buffer_reg_init_pid_registry();
4345 /* Init UST command queue. */
4346 cds_wfq_init(&ust_cmd_queue
.queue
);
4349 * Get session list pointer. This pointer MUST NOT be free(). This list is
4350 * statically declared in session.c
4352 session_list_ptr
= session_get_list();
4354 /* Set up max poll set size */
4355 lttng_poll_set_max_size();
4359 /* Check for the application socket timeout env variable. */
4360 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4361 if (env_app_timeout
) {
4362 app_socket_timeout
= atoi(env_app_timeout
);
4364 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4369 /* Create thread to manage the client socket */
4370 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4371 thread_ht_cleanup
, (void *) NULL
);
4373 PERROR("pthread_create ht_cleanup");
4374 goto exit_ht_cleanup
;
4377 /* Create thread to manage the client socket */
4378 ret
= pthread_create(&health_thread
, NULL
,
4379 thread_manage_health
, (void *) NULL
);
4381 PERROR("pthread_create health");
4385 /* Create thread to manage the client socket */
4386 ret
= pthread_create(&client_thread
, NULL
,
4387 thread_manage_clients
, (void *) NULL
);
4389 PERROR("pthread_create clients");
4393 /* Create thread to dispatch registration */
4394 ret
= pthread_create(&dispatch_thread
, NULL
,
4395 thread_dispatch_ust_registration
, (void *) NULL
);
4397 PERROR("pthread_create dispatch");
4401 /* Create thread to manage application registration. */
4402 ret
= pthread_create(®_apps_thread
, NULL
,
4403 thread_registration_apps
, (void *) NULL
);
4405 PERROR("pthread_create registration");
4409 /* Create thread to manage application socket */
4410 ret
= pthread_create(&apps_thread
, NULL
,
4411 thread_manage_apps
, (void *) NULL
);
4413 PERROR("pthread_create apps");
4417 /* Create thread to manage application notify socket */
4418 ret
= pthread_create(&apps_notify_thread
, NULL
,
4419 ust_thread_manage_notify
, (void *) NULL
);
4421 PERROR("pthread_create apps");
4425 /* Don't start this thread if kernel tracing is not requested nor root */
4426 if (is_root
&& !opt_no_kernel
) {
4427 /* Create kernel thread to manage kernel event */
4428 ret
= pthread_create(&kernel_thread
, NULL
,
4429 thread_manage_kernel
, (void *) NULL
);
4431 PERROR("pthread_create kernel");
4435 ret
= pthread_join(kernel_thread
, &status
);
4437 PERROR("pthread_join");
4438 goto error
; /* join error, exit without cleanup */
4443 ret
= pthread_join(apps_thread
, &status
);
4445 PERROR("pthread_join");
4446 goto error
; /* join error, exit without cleanup */
4450 ret
= pthread_join(reg_apps_thread
, &status
);
4452 PERROR("pthread_join");
4453 goto error
; /* join error, exit without cleanup */
4457 ret
= pthread_join(dispatch_thread
, &status
);
4459 PERROR("pthread_join");
4460 goto error
; /* join error, exit without cleanup */
4464 ret
= pthread_join(client_thread
, &status
);
4466 PERROR("pthread_join");
4467 goto error
; /* join error, exit without cleanup */
4470 ret
= join_consumer_thread(&kconsumer_data
);
4472 PERROR("join_consumer");
4473 goto error
; /* join error, exit without cleanup */
4476 ret
= join_consumer_thread(&ustconsumer32_data
);
4478 PERROR("join_consumer ust32");
4479 goto error
; /* join error, exit without cleanup */
4482 ret
= join_consumer_thread(&ustconsumer64_data
);
4484 PERROR("join_consumer ust64");
4485 goto error
; /* join error, exit without cleanup */
4489 ret
= pthread_join(health_thread
, &status
);
4491 PERROR("pthread_join health thread");
4492 goto error
; /* join error, exit without cleanup */
4496 ret
= pthread_join(ht_cleanup_thread
, &status
);
4498 PERROR("pthread_join ht cleanup thread");
4499 goto error
; /* join error, exit without cleanup */
4504 * cleanup() is called when no other thread is running.
4506 rcu_thread_online();
4508 rcu_thread_offline();
4509 rcu_unregister_thread();