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.
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
33 #include <sys/types.h>
35 #include <urcu/uatomic.h>
39 #include <common/common.h>
40 #include <common/compat/poll.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"
55 #include "kernel-consumer.h"
59 #include "ust-consumer.h"
63 #include "testpoint.h"
65 #define CONSUMERD_FILE "lttng-consumerd"
68 const char default_home_dir
[] = DEFAULT_HOME_DIR
;
69 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
70 const char default_ust_sock_dir
[] = DEFAULT_UST_SOCK_DIR
;
71 const char default_global_apps_pipe
[] = DEFAULT_GLOBAL_APPS_PIPE
;
74 const char *opt_tracing_group
;
75 static const char *opt_pidfile
;
76 static int opt_sig_parent
;
77 static int opt_verbose_consumer
;
78 static int opt_daemon
;
79 static int opt_no_kernel
;
80 static int is_root
; /* Set to 1 if the daemon is running as root */
81 static pid_t ppid
; /* Parent PID for --sig-parent option */
85 * Consumer daemon specific control data. Every value not initialized here is
86 * set to 0 by the static definition.
88 static struct consumer_data kconsumer_data
= {
89 .type
= LTTNG_CONSUMER_KERNEL
,
90 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
91 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
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 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
106 .lock
= PTHREAD_MUTEX_INITIALIZER
,
107 .cond
= PTHREAD_COND_INITIALIZER
,
108 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
110 static struct consumer_data ustconsumer32_data
= {
111 .type
= LTTNG_CONSUMER32_UST
,
112 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
113 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
116 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
117 .lock
= PTHREAD_MUTEX_INITIALIZER
,
118 .cond
= PTHREAD_COND_INITIALIZER
,
119 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
122 /* Shared between threads */
123 static int dispatch_thread_exit
;
125 /* Global application Unix socket path */
126 static char apps_unix_sock_path
[PATH_MAX
];
127 /* Global client Unix socket path */
128 static char client_unix_sock_path
[PATH_MAX
];
129 /* global wait shm path for UST */
130 static char wait_shm_path
[PATH_MAX
];
131 /* Global health check unix path */
132 static char health_unix_sock_path
[PATH_MAX
];
134 /* Sockets and FDs */
135 static int client_sock
= -1;
136 static int apps_sock
= -1;
137 int kernel_tracer_fd
= -1;
138 static int kernel_poll_pipe
[2] = { -1, -1 };
141 * Quit pipe for all threads. This permits a single cancellation point
142 * for all threads when receiving an event on the pipe.
144 static int thread_quit_pipe
[2] = { -1, -1 };
147 * This pipe is used to inform the thread managing application communication
148 * that a command is queued and ready to be processed.
150 static int apps_cmd_pipe
[2] = { -1, -1 };
152 /* Pthread, Mutexes and Semaphores */
153 static pthread_t apps_thread
;
154 static pthread_t reg_apps_thread
;
155 static pthread_t client_thread
;
156 static pthread_t kernel_thread
;
157 static pthread_t dispatch_thread
;
158 static pthread_t health_thread
;
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
167 static struct ust_cmd_queue ust_cmd_queue
;
170 * Pointer initialized before thread creation.
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
177 * session_lock_list and session_unlock_list for lock acquisition.
179 static struct ltt_session_list
*session_list_ptr
;
181 int ust_consumerd64_fd
= -1;
182 int ust_consumerd32_fd
= -1;
184 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
185 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
186 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
187 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
189 static const char *module_proc_lttng
= "/proc/lttng";
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
195 enum consumerd_state
{
196 CONSUMER_STARTED
= 1,
197 CONSUMER_STOPPED
= 2,
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
207 * The following example shows a possible race condition of this scheme:
209 * consumer thread error happens
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
222 static enum consumerd_state ust_consumerd_state
;
223 static enum consumerd_state kernel_consumerd_state
;
225 /* Used for the health monitoring of the session daemon. See health.h */
226 struct health_state health_thread_cmd
;
227 struct health_state health_thread_app_manage
;
228 struct health_state health_thread_app_reg
;
229 struct health_state health_thread_kernel
;
232 * Socket timeout for receiving and sending in seconds.
234 static int app_socket_timeout
;
237 void setup_consumerd_path(void)
239 const char *bin
, *libdir
;
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin
[0]) {
248 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
250 if (!consumerd32_libdir
[0]) {
251 consumerd32_libdir
= INSTALL_LIB_PATH
;
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin
[0]) {
255 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
257 if (!consumerd64_libdir
[0]) {
258 consumerd64_libdir
= INSTALL_LIB_PATH
;
261 #error "Unknown bitness"
265 * runtime env. var. overrides the build default.
267 bin
= getenv("LTTNG_CONSUMERD32_BIN");
269 consumerd32_bin
= bin
;
271 bin
= getenv("LTTNG_CONSUMERD64_BIN");
273 consumerd64_bin
= bin
;
275 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
277 consumerd32_libdir
= libdir
;
279 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
281 consumerd64_libdir
= libdir
;
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
288 static int create_thread_poll_set(struct lttng_poll_event
*events
,
293 if (events
== NULL
|| size
== 0) {
298 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
304 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
);
316 * Check if the thread quit pipe was triggered.
318 * Return 1 if it was triggered else 0;
320 static int check_thread_quit_pipe(int fd
, uint32_t events
)
322 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
330 * Return group ID of the tracing group or -1 if not found.
332 static gid_t
allowed_group(void)
336 if (opt_tracing_group
) {
337 grp
= getgrnam(opt_tracing_group
);
339 grp
= getgrnam(default_tracing_group
);
349 * Init thread quit pipe.
351 * Return -1 on error or 0 if all pipes are created.
353 static int init_thread_quit_pipe(void)
357 ret
= pipe(thread_quit_pipe
);
359 PERROR("thread quit pipe");
363 for (i
= 0; i
< 2; i
++) {
364 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
376 * Stop all threads by closing the thread quit pipe.
378 static void stop_threads(void)
382 /* Stopping all threads */
383 DBG("Terminating all threads");
384 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
386 ERR("write error on thread quit pipe");
389 /* Dispatch thread */
390 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
391 futex_nto1_wake(&ust_cmd_queue
.futex
);
397 static void cleanup(void)
401 struct ltt_session
*sess
, *stmp
;
405 /* First thing first, stop all threads */
406 utils_close_pipe(thread_quit_pipe
);
409 * If opt_pidfile is undefined, the default file will be wiped when
410 * removing the rundir.
413 ret
= remove(opt_pidfile
);
415 PERROR("remove pidfile %s", opt_pidfile
);
419 DBG("Removing %s directory", rundir
);
420 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
422 ERR("asprintf failed. Something is really wrong!");
425 /* Remove lttng run directory */
428 ERR("Unable to clean %s", rundir
);
433 DBG("Cleaning up all sessions");
435 /* Destroy session list mutex */
436 if (session_list_ptr
!= NULL
) {
437 pthread_mutex_destroy(&session_list_ptr
->lock
);
439 /* Cleanup ALL session */
440 cds_list_for_each_entry_safe(sess
, stmp
,
441 &session_list_ptr
->head
, list
) {
442 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
446 DBG("Closing all UST sockets");
447 ust_app_clean_list();
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
);
659 session_unlock(session
);
661 session_unlock_list();
665 session_unlock(session
);
666 session_unlock_list();
671 * For each tracing session, update newly registered apps.
673 static void update_ust_app(int app_sock
)
675 struct ltt_session
*sess
, *stmp
;
679 /* For all tracing session(s) */
680 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
682 if (sess
->ust_session
) {
683 ust_app_global_update(sess
->ust_session
, app_sock
);
685 session_unlock(sess
);
688 session_unlock_list();
692 * This thread manage event coming from the kernel.
694 * Features supported in this thread:
697 static void *thread_manage_kernel(void *data
)
699 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
700 uint32_t revents
, nb_fd
;
702 struct lttng_poll_event events
;
704 DBG("[thread] Thread manage kernel started");
707 * This first step of the while is to clean this structure which could free
708 * non NULL pointers so zero it before the loop.
710 memset(&events
, 0, sizeof(events
));
712 if (testpoint(thread_manage_kernel
)) {
713 goto error_testpoint
;
716 health_code_update(&health_thread_kernel
);
718 if (testpoint(thread_manage_kernel_before_loop
)) {
719 goto error_testpoint
;
723 health_code_update(&health_thread_kernel
);
725 if (update_poll_flag
== 1) {
726 /* Clean events object. We are about to populate it again. */
727 lttng_poll_clean(&events
);
729 ret
= create_thread_poll_set(&events
, 2);
731 goto error_poll_create
;
734 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
739 /* This will add the available kernel channel if any. */
740 ret
= update_kernel_poll(&events
);
744 update_poll_flag
= 0;
747 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
749 /* Poll infinite value of time */
751 health_poll_update(&health_thread_kernel
);
752 ret
= lttng_poll_wait(&events
, -1);
753 health_poll_update(&health_thread_kernel
);
756 * Restart interrupted system call.
758 if (errno
== EINTR
) {
762 } else if (ret
== 0) {
763 /* Should not happen since timeout is infinite */
764 ERR("Return value of poll is 0 with an infinite timeout.\n"
765 "This should not have happened! Continuing...");
771 for (i
= 0; i
< nb_fd
; i
++) {
772 /* Fetch once the poll data */
773 revents
= LTTNG_POLL_GETEV(&events
, i
);
774 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
776 health_code_update(&health_thread_kernel
);
778 /* Thread quit pipe has been closed. Killing thread. */
779 ret
= check_thread_quit_pipe(pollfd
, revents
);
785 /* Check for data on kernel pipe */
786 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
788 ret
= read(kernel_poll_pipe
[0], &tmp
, 1);
789 } while (ret
< 0 && errno
== EINTR
);
791 * Ret value is useless here, if this pipe gets any actions an
792 * update is required anyway.
794 update_poll_flag
= 1;
798 * New CPU detected by the kernel. Adding kernel stream to
799 * kernel session and updating the kernel consumer
801 if (revents
& LPOLLIN
) {
802 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
808 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
809 * and unregister kernel stream at this point.
818 lttng_poll_clean(&events
);
821 utils_close_pipe(kernel_poll_pipe
);
822 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
824 health_error(&health_thread_kernel
);
825 ERR("Health error occurred in %s", __func__
);
826 WARN("Kernel thread died unexpectedly. "
827 "Kernel tracing can continue but CPU hotplug is disabled.");
829 health_exit(&health_thread_kernel
);
830 DBG("Kernel thread dying");
835 * Signal pthread condition of the consumer data that the thread.
837 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
839 pthread_mutex_lock(&data
->cond_mutex
);
842 * The state is set before signaling. It can be any value, it's the waiter
843 * job to correctly interpret this condition variable associated to the
844 * consumer pthread_cond.
846 * A value of 0 means that the corresponding thread of the consumer data
847 * was not started. 1 indicates that the thread has started and is ready
848 * for action. A negative value means that there was an error during the
851 data
->consumer_thread_is_ready
= state
;
852 (void) pthread_cond_signal(&data
->cond
);
854 pthread_mutex_unlock(&data
->cond_mutex
);
858 * This thread manage the consumer error sent back to the session daemon.
860 static void *thread_manage_consumer(void *data
)
862 int sock
= -1, i
, ret
, pollfd
, err
= -1;
863 uint32_t revents
, nb_fd
;
864 enum lttcomm_return_code code
;
865 struct lttng_poll_event events
;
866 struct consumer_data
*consumer_data
= data
;
868 DBG("[thread] Manage consumer started");
871 * Since the consumer thread can be spawned at any moment in time, we init
872 * the health to a poll status (1, which is a valid health over time).
873 * When the thread starts, we update here the health to a "code" path being
874 * an even value so this thread, when reaching a poll wait, does not
875 * trigger an error with an even value.
877 * Here is the use case we avoid.
879 * +1: the first poll update during initialization (main())
880 * +2 * x: multiple code update once in this thread.
881 * +1: poll wait in this thread (being a good health state).
882 * == even number which after the wait period shows as a bad health.
884 * In a nutshell, the following poll update to the health state brings back
885 * the state to an even value meaning a code path.
887 health_poll_update(&consumer_data
->health
);
890 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
891 * Nothing more will be added to this poll set.
893 ret
= create_thread_poll_set(&events
, 2);
899 * The error socket here is already in a listening state which was done
900 * just before spawning this thread to avoid a race between the consumer
901 * daemon exec trying to connect and the listen() call.
903 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
908 health_code_update(&consumer_data
->health
);
910 /* Inifinite blocking call, waiting for transmission */
912 health_poll_update(&consumer_data
->health
);
914 if (testpoint(thread_manage_consumer
)) {
918 ret
= lttng_poll_wait(&events
, -1);
919 health_poll_update(&consumer_data
->health
);
922 * Restart interrupted system call.
924 if (errno
== EINTR
) {
932 for (i
= 0; i
< nb_fd
; i
++) {
933 /* Fetch once the poll data */
934 revents
= LTTNG_POLL_GETEV(&events
, i
);
935 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
937 health_code_update(&consumer_data
->health
);
939 /* Thread quit pipe has been closed. Killing thread. */
940 ret
= check_thread_quit_pipe(pollfd
, revents
);
946 /* Event on the registration socket */
947 if (pollfd
== consumer_data
->err_sock
) {
948 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
949 ERR("consumer err socket poll error");
955 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
961 * Set the CLOEXEC flag. Return code is useless because either way, the
964 (void) utils_set_fd_cloexec(sock
);
966 health_code_update(&consumer_data
->health
);
968 DBG2("Receiving code from consumer err_sock");
970 /* Getting status code from kconsumerd */
971 ret
= lttcomm_recv_unix_sock(sock
, &code
,
972 sizeof(enum lttcomm_return_code
));
977 health_code_update(&consumer_data
->health
);
979 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
980 consumer_data
->cmd_sock
=
981 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
982 if (consumer_data
->cmd_sock
< 0) {
983 /* On error, signal condition and quit. */
984 signal_consumer_condition(consumer_data
, -1);
985 PERROR("consumer connect");
988 signal_consumer_condition(consumer_data
, 1);
989 DBG("Consumer command socket ready");
991 ERR("consumer error when waiting for SOCK_READY : %s",
992 lttcomm_get_readable_code(-code
));
996 /* Remove the kconsumerd error sock since we've established a connexion */
997 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1002 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1007 health_code_update(&consumer_data
->health
);
1009 /* Inifinite blocking call, waiting for transmission */
1011 health_poll_update(&consumer_data
->health
);
1012 ret
= lttng_poll_wait(&events
, -1);
1013 health_poll_update(&consumer_data
->health
);
1016 * Restart interrupted system call.
1018 if (errno
== EINTR
) {
1026 for (i
= 0; i
< nb_fd
; i
++) {
1027 /* Fetch once the poll data */
1028 revents
= LTTNG_POLL_GETEV(&events
, i
);
1029 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1031 health_code_update(&consumer_data
->health
);
1033 /* Thread quit pipe has been closed. Killing thread. */
1034 ret
= check_thread_quit_pipe(pollfd
, revents
);
1040 /* Event on the kconsumerd socket */
1041 if (pollfd
== sock
) {
1042 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1043 ERR("consumer err socket second poll error");
1049 health_code_update(&consumer_data
->health
);
1051 /* Wait for any kconsumerd error */
1052 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1053 sizeof(enum lttcomm_return_code
));
1055 ERR("consumer closed the command socket");
1059 ERR("consumer return code : %s", lttcomm_get_readable_code(-code
));
1063 /* Immediately set the consumerd state to stopped */
1064 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1065 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1066 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1067 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1068 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1070 /* Code flow error... */
1074 if (consumer_data
->err_sock
>= 0) {
1075 ret
= close(consumer_data
->err_sock
);
1080 if (consumer_data
->cmd_sock
>= 0) {
1081 ret
= close(consumer_data
->cmd_sock
);
1093 unlink(consumer_data
->err_unix_sock_path
);
1094 unlink(consumer_data
->cmd_unix_sock_path
);
1095 consumer_data
->pid
= 0;
1097 lttng_poll_clean(&events
);
1100 health_error(&consumer_data
->health
);
1101 ERR("Health error occurred in %s", __func__
);
1103 health_exit(&consumer_data
->health
);
1104 DBG("consumer thread cleanup completed");
1110 * This thread manage application communication.
1112 static void *thread_manage_apps(void *data
)
1114 int i
, ret
, pollfd
, err
= -1;
1115 uint32_t revents
, nb_fd
;
1116 struct ust_command ust_cmd
;
1117 struct lttng_poll_event events
;
1119 DBG("[thread] Manage application started");
1121 rcu_register_thread();
1122 rcu_thread_online();
1124 if (testpoint(thread_manage_apps
)) {
1125 goto error_testpoint
;
1128 health_code_update(&health_thread_app_manage
);
1130 ret
= create_thread_poll_set(&events
, 2);
1132 goto error_poll_create
;
1135 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1140 if (testpoint(thread_manage_apps_before_loop
)) {
1144 health_code_update(&health_thread_app_manage
);
1147 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1149 /* Inifinite blocking call, waiting for transmission */
1151 health_poll_update(&health_thread_app_manage
);
1152 ret
= lttng_poll_wait(&events
, -1);
1153 health_poll_update(&health_thread_app_manage
);
1156 * Restart interrupted system call.
1158 if (errno
== EINTR
) {
1166 for (i
= 0; i
< nb_fd
; i
++) {
1167 /* Fetch once the poll data */
1168 revents
= LTTNG_POLL_GETEV(&events
, i
);
1169 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1171 health_code_update(&health_thread_app_manage
);
1173 /* Thread quit pipe has been closed. Killing thread. */
1174 ret
= check_thread_quit_pipe(pollfd
, revents
);
1180 /* Inspect the apps cmd pipe */
1181 if (pollfd
== apps_cmd_pipe
[0]) {
1182 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1183 ERR("Apps command pipe error");
1185 } else if (revents
& LPOLLIN
) {
1188 ret
= read(apps_cmd_pipe
[0], &ust_cmd
, sizeof(ust_cmd
));
1189 } while (ret
< 0 && errno
== EINTR
);
1190 if (ret
< 0 || ret
< sizeof(ust_cmd
)) {
1191 PERROR("read apps cmd pipe");
1195 health_code_update(&health_thread_app_manage
);
1197 /* Register applicaton to the session daemon */
1198 ret
= ust_app_register(&ust_cmd
.reg_msg
,
1200 if (ret
== -ENOMEM
) {
1202 } else if (ret
< 0) {
1206 health_code_update(&health_thread_app_manage
);
1209 * Validate UST version compatibility.
1211 ret
= ust_app_validate_version(ust_cmd
.sock
);
1214 * Add channel(s) and event(s) to newly registered apps
1215 * from lttng global UST domain.
1217 update_ust_app(ust_cmd
.sock
);
1220 health_code_update(&health_thread_app_manage
);
1222 ret
= ust_app_register_done(ust_cmd
.sock
);
1225 * If the registration is not possible, we simply
1226 * unregister the apps and continue
1228 ust_app_unregister(ust_cmd
.sock
);
1231 * We only monitor the error events of the socket. This
1232 * thread does not handle any incoming data from UST
1235 ret
= lttng_poll_add(&events
, ust_cmd
.sock
,
1236 LPOLLERR
& LPOLLHUP
& LPOLLRDHUP
);
1241 /* Set socket timeout for both receiving and ending */
1242 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd
.sock
,
1243 app_socket_timeout
);
1244 (void) lttcomm_setsockopt_snd_timeout(ust_cmd
.sock
,
1245 app_socket_timeout
);
1247 DBG("Apps with sock %d added to poll set",
1251 health_code_update(&health_thread_app_manage
);
1257 * At this point, we know that a registered application made
1258 * the event at poll_wait.
1260 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1261 /* Removing from the poll set */
1262 ret
= lttng_poll_del(&events
, pollfd
);
1267 /* Socket closed on remote end. */
1268 ust_app_unregister(pollfd
);
1273 health_code_update(&health_thread_app_manage
);
1279 lttng_poll_clean(&events
);
1282 utils_close_pipe(apps_cmd_pipe
);
1283 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1286 * We don't clean the UST app hash table here since already registered
1287 * applications can still be controlled so let them be until the session
1288 * daemon dies or the applications stop.
1292 health_error(&health_thread_app_manage
);
1293 ERR("Health error occurred in %s", __func__
);
1295 health_exit(&health_thread_app_manage
);
1296 DBG("Application communication apps thread cleanup complete");
1297 rcu_thread_offline();
1298 rcu_unregister_thread();
1303 * Dispatch request from the registration threads to the application
1304 * communication thread.
1306 static void *thread_dispatch_ust_registration(void *data
)
1309 struct cds_wfq_node
*node
;
1310 struct ust_command
*ust_cmd
= NULL
;
1312 DBG("[thread] Dispatch UST command started");
1314 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1315 /* Atomically prepare the queue futex */
1316 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1319 /* Dequeue command for registration */
1320 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1322 DBG("Woken up but nothing in the UST command queue");
1323 /* Continue thread execution */
1327 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1329 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1330 " gid:%d sock:%d name:%s (version %d.%d)",
1331 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1332 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1333 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1334 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1336 * Inform apps thread of the new application registration. This
1337 * call is blocking so we can be assured that the data will be read
1338 * at some point in time or wait to the end of the world :)
1340 if (apps_cmd_pipe
[1] >= 0) {
1342 ret
= write(apps_cmd_pipe
[1], ust_cmd
,
1343 sizeof(struct ust_command
));
1344 } while (ret
< 0 && errno
== EINTR
);
1345 if (ret
< 0 || ret
!= sizeof(struct ust_command
)) {
1346 PERROR("write apps cmd pipe");
1347 if (errno
== EBADF
) {
1349 * We can't inform the application thread to process
1350 * registration. We will exit or else application
1351 * registration will not occur and tracing will never
1358 /* Application manager thread is not available. */
1359 ret
= close(ust_cmd
->sock
);
1361 PERROR("close ust_cmd sock");
1365 } while (node
!= NULL
);
1367 /* Futex wait on queue. Blocking call on futex() */
1368 futex_nto1_wait(&ust_cmd_queue
.futex
);
1372 DBG("Dispatch thread dying");
1377 * This thread manage application registration.
1379 static void *thread_registration_apps(void *data
)
1381 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1382 uint32_t revents
, nb_fd
;
1383 struct lttng_poll_event events
;
1385 * Get allocated in this thread, enqueued to a global queue, dequeued and
1386 * freed in the manage apps thread.
1388 struct ust_command
*ust_cmd
= NULL
;
1390 DBG("[thread] Manage application registration started");
1392 if (testpoint(thread_registration_apps
)) {
1393 goto error_testpoint
;
1396 ret
= lttcomm_listen_unix_sock(apps_sock
);
1402 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1403 * more will be added to this poll set.
1405 ret
= create_thread_poll_set(&events
, 2);
1407 goto error_create_poll
;
1410 /* Add the application registration socket */
1411 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1413 goto error_poll_add
;
1416 /* Notify all applications to register */
1417 ret
= notify_ust_apps(1);
1419 ERR("Failed to notify applications or create the wait shared memory.\n"
1420 "Execution continues but there might be problem for already\n"
1421 "running applications that wishes to register.");
1425 DBG("Accepting application registration");
1427 /* Inifinite blocking call, waiting for transmission */
1429 health_poll_update(&health_thread_app_reg
);
1430 ret
= lttng_poll_wait(&events
, -1);
1431 health_poll_update(&health_thread_app_reg
);
1434 * Restart interrupted system call.
1436 if (errno
== EINTR
) {
1444 for (i
= 0; i
< nb_fd
; i
++) {
1445 health_code_update(&health_thread_app_reg
);
1447 /* Fetch once the poll data */
1448 revents
= LTTNG_POLL_GETEV(&events
, i
);
1449 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1451 /* Thread quit pipe has been closed. Killing thread. */
1452 ret
= check_thread_quit_pipe(pollfd
, revents
);
1458 /* Event on the registration socket */
1459 if (pollfd
== apps_sock
) {
1460 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1461 ERR("Register apps socket poll error");
1463 } else if (revents
& LPOLLIN
) {
1464 sock
= lttcomm_accept_unix_sock(apps_sock
);
1470 * Set the CLOEXEC flag. Return code is useless because
1471 * either way, the show must go on.
1473 (void) utils_set_fd_cloexec(sock
);
1475 /* Create UST registration command for enqueuing */
1476 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1477 if (ust_cmd
== NULL
) {
1478 PERROR("ust command zmalloc");
1483 * Using message-based transmissions to ensure we don't
1484 * have to deal with partially received messages.
1486 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1488 ERR("Exhausted file descriptors allowed for applications.");
1497 health_code_update(&health_thread_app_reg
);
1498 ret
= lttcomm_recv_unix_sock(sock
, &ust_cmd
->reg_msg
,
1499 sizeof(struct ust_register_msg
));
1500 if (ret
< 0 || ret
< sizeof(struct ust_register_msg
)) {
1502 PERROR("lttcomm_recv_unix_sock register apps");
1504 ERR("Wrong size received on apps register");
1511 lttng_fd_put(LTTNG_FD_APPS
, 1);
1515 health_code_update(&health_thread_app_reg
);
1517 ust_cmd
->sock
= sock
;
1520 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1521 " gid:%d sock:%d name:%s (version %d.%d)",
1522 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1523 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1524 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1525 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1528 * Lock free enqueue the registration request. The red pill
1529 * has been taken! This apps will be part of the *system*.
1531 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1534 * Wake the registration queue futex. Implicit memory
1535 * barrier with the exchange in cds_wfq_enqueue.
1537 futex_nto1_wake(&ust_cmd_queue
.futex
);
1546 health_error(&health_thread_app_reg
);
1547 ERR("Health error occurred in %s", __func__
);
1550 /* Notify that the registration thread is gone */
1553 if (apps_sock
>= 0) {
1554 ret
= close(apps_sock
);
1564 lttng_fd_put(LTTNG_FD_APPS
, 1);
1566 unlink(apps_unix_sock_path
);
1569 lttng_poll_clean(&events
);
1573 DBG("UST Registration thread cleanup complete");
1574 health_exit(&health_thread_app_reg
);
1580 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1581 * exec or it will fails.
1583 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1586 struct timespec timeout
;
1588 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1589 consumer_data
->consumer_thread_is_ready
= 0;
1591 /* Setup pthread condition */
1592 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1595 PERROR("pthread_condattr_init consumer data");
1600 * Set the monotonic clock in order to make sure we DO NOT jump in time
1601 * between the clock_gettime() call and the timedwait call. See bug #324
1602 * for a more details and how we noticed it.
1604 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1607 PERROR("pthread_condattr_setclock consumer data");
1611 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1614 PERROR("pthread_cond_init consumer data");
1618 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1621 PERROR("pthread_create consumer");
1626 /* We are about to wait on a pthread condition */
1627 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1629 /* Get time for sem_timedwait absolute timeout */
1630 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1632 * Set the timeout for the condition timed wait even if the clock gettime
1633 * call fails since we might loop on that call and we want to avoid to
1634 * increment the timeout too many times.
1636 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1639 * The following loop COULD be skipped in some conditions so this is why we
1640 * set ret to 0 in order to make sure at least one round of the loop is
1646 * Loop until the condition is reached or when a timeout is reached. Note
1647 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1648 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1649 * possible. This loop does not take any chances and works with both of
1652 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1653 if (clock_ret
< 0) {
1654 PERROR("clock_gettime spawn consumer");
1655 /* Infinite wait for the consumerd thread to be ready */
1656 ret
= pthread_cond_wait(&consumer_data
->cond
,
1657 &consumer_data
->cond_mutex
);
1659 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1660 &consumer_data
->cond_mutex
, &timeout
);
1664 /* Release the pthread condition */
1665 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1669 if (ret
== ETIMEDOUT
) {
1671 * Call has timed out so we kill the kconsumerd_thread and return
1674 ERR("Condition timed out. The consumer thread was never ready."
1676 ret
= pthread_cancel(consumer_data
->thread
);
1678 PERROR("pthread_cancel consumer thread");
1681 PERROR("pthread_cond_wait failed consumer thread");
1686 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1687 if (consumer_data
->pid
== 0) {
1688 ERR("Consumerd did not start");
1689 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1692 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1701 * Join consumer thread
1703 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1707 /* Consumer pid must be a real one. */
1708 if (consumer_data
->pid
> 0) {
1710 ret
= kill(consumer_data
->pid
, SIGTERM
);
1712 ERR("Error killing consumer daemon");
1715 return pthread_join(consumer_data
->thread
, &status
);
1722 * Fork and exec a consumer daemon (consumerd).
1724 * Return pid if successful else -1.
1726 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1730 const char *consumer_to_use
;
1731 const char *verbosity
;
1734 DBG("Spawning consumerd");
1741 if (opt_verbose_consumer
) {
1742 verbosity
= "--verbose";
1744 verbosity
= "--quiet";
1746 switch (consumer_data
->type
) {
1747 case LTTNG_CONSUMER_KERNEL
:
1749 * Find out which consumerd to execute. We will first try the
1750 * 64-bit path, then the sessiond's installation directory, and
1751 * fallback on the 32-bit one,
1753 DBG3("Looking for a kernel consumer at these locations:");
1754 DBG3(" 1) %s", consumerd64_bin
);
1755 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1756 DBG3(" 3) %s", consumerd32_bin
);
1757 if (stat(consumerd64_bin
, &st
) == 0) {
1758 DBG3("Found location #1");
1759 consumer_to_use
= consumerd64_bin
;
1760 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1761 DBG3("Found location #2");
1762 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1763 } else if (stat(consumerd32_bin
, &st
) == 0) {
1764 DBG3("Found location #3");
1765 consumer_to_use
= consumerd32_bin
;
1767 DBG("Could not find any valid consumerd executable");
1770 DBG("Using kernel consumer at: %s", consumer_to_use
);
1771 execl(consumer_to_use
,
1772 "lttng-consumerd", verbosity
, "-k",
1773 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1774 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1777 case LTTNG_CONSUMER64_UST
:
1779 char *tmpnew
= NULL
;
1781 if (consumerd64_libdir
[0] != '\0') {
1785 tmp
= getenv("LD_LIBRARY_PATH");
1789 tmplen
= strlen("LD_LIBRARY_PATH=")
1790 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1791 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1796 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1797 strcat(tmpnew
, consumerd64_libdir
);
1798 if (tmp
[0] != '\0') {
1799 strcat(tmpnew
, ":");
1800 strcat(tmpnew
, tmp
);
1802 ret
= putenv(tmpnew
);
1808 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1809 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1810 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1811 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1813 if (consumerd64_libdir
[0] != '\0') {
1821 case LTTNG_CONSUMER32_UST
:
1823 char *tmpnew
= NULL
;
1825 if (consumerd32_libdir
[0] != '\0') {
1829 tmp
= getenv("LD_LIBRARY_PATH");
1833 tmplen
= strlen("LD_LIBRARY_PATH=")
1834 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1835 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1840 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1841 strcat(tmpnew
, consumerd32_libdir
);
1842 if (tmp
[0] != '\0') {
1843 strcat(tmpnew
, ":");
1844 strcat(tmpnew
, tmp
);
1846 ret
= putenv(tmpnew
);
1852 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
1853 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
1854 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1855 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1857 if (consumerd32_libdir
[0] != '\0') {
1866 PERROR("unknown consumer type");
1870 PERROR("kernel start consumer exec");
1873 } else if (pid
> 0) {
1876 PERROR("start consumer fork");
1884 * Spawn the consumerd daemon and session daemon thread.
1886 static int start_consumerd(struct consumer_data
*consumer_data
)
1891 * Set the listen() state on the socket since there is a possible race
1892 * between the exec() of the consumer daemon and this call if place in the
1893 * consumer thread. See bug #366 for more details.
1895 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
1900 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1901 if (consumer_data
->pid
!= 0) {
1902 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1906 ret
= spawn_consumerd(consumer_data
);
1908 ERR("Spawning consumerd failed");
1909 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1913 /* Setting up the consumer_data pid */
1914 consumer_data
->pid
= ret
;
1915 DBG2("Consumer pid %d", consumer_data
->pid
);
1916 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1918 DBG2("Spawning consumer control thread");
1919 ret
= spawn_consumer_thread(consumer_data
);
1921 ERR("Fatal error spawning consumer control thread");
1929 /* Cleanup already created socket on error. */
1930 if (consumer_data
->err_sock
>= 0) {
1933 err
= close(consumer_data
->err_sock
);
1935 PERROR("close consumer data error socket");
1942 * Compute health status of each consumer. If one of them is zero (bad
1943 * state), we return 0.
1945 static int check_consumer_health(void)
1949 ret
= health_check_state(&kconsumer_data
.health
) &&
1950 health_check_state(&ustconsumer32_data
.health
) &&
1951 health_check_state(&ustconsumer64_data
.health
);
1953 DBG3("Health consumer check %d", ret
);
1959 * Setup necessary data for kernel tracer action.
1961 static int init_kernel_tracer(void)
1965 /* Modprobe lttng kernel modules */
1966 ret
= modprobe_lttng_control();
1971 /* Open debugfs lttng */
1972 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1973 if (kernel_tracer_fd
< 0) {
1974 DBG("Failed to open %s", module_proc_lttng
);
1979 /* Validate kernel version */
1980 ret
= kernel_validate_version(kernel_tracer_fd
);
1985 ret
= modprobe_lttng_data();
1990 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1994 modprobe_remove_lttng_control();
1995 ret
= close(kernel_tracer_fd
);
1999 kernel_tracer_fd
= -1;
2000 return LTTNG_ERR_KERN_VERSION
;
2003 ret
= close(kernel_tracer_fd
);
2009 modprobe_remove_lttng_control();
2012 WARN("No kernel tracer available");
2013 kernel_tracer_fd
= -1;
2015 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2017 return LTTNG_ERR_KERN_NA
;
2023 * Copy consumer output from the tracing session to the domain session. The
2024 * function also applies the right modification on a per domain basis for the
2025 * trace files destination directory.
2027 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2030 const char *dir_name
;
2031 struct consumer_output
*consumer
;
2034 assert(session
->consumer
);
2037 case LTTNG_DOMAIN_KERNEL
:
2038 DBG3("Copying tracing session consumer output in kernel session");
2040 * XXX: We should audit the session creation and what this function
2041 * does "extra" in order to avoid a destroy since this function is used
2042 * in the domain session creation (kernel and ust) only. Same for UST
2045 if (session
->kernel_session
->consumer
) {
2046 consumer_destroy_output(session
->kernel_session
->consumer
);
2048 session
->kernel_session
->consumer
=
2049 consumer_copy_output(session
->consumer
);
2050 /* Ease our life a bit for the next part */
2051 consumer
= session
->kernel_session
->consumer
;
2052 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2054 case LTTNG_DOMAIN_UST
:
2055 DBG3("Copying tracing session consumer output in UST session");
2056 if (session
->ust_session
->consumer
) {
2057 consumer_destroy_output(session
->ust_session
->consumer
);
2059 session
->ust_session
->consumer
=
2060 consumer_copy_output(session
->consumer
);
2061 /* Ease our life a bit for the next part */
2062 consumer
= session
->ust_session
->consumer
;
2063 dir_name
= DEFAULT_UST_TRACE_DIR
;
2066 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2070 /* Append correct directory to subdir */
2071 strncat(consumer
->subdir
, dir_name
,
2072 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2073 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2082 * Create an UST session and add it to the session ust list.
2084 static int create_ust_session(struct ltt_session
*session
,
2085 struct lttng_domain
*domain
)
2088 struct ltt_ust_session
*lus
= NULL
;
2092 assert(session
->consumer
);
2094 switch (domain
->type
) {
2095 case LTTNG_DOMAIN_UST
:
2098 ERR("Unknown UST domain on create session %d", domain
->type
);
2099 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2103 DBG("Creating UST session");
2105 lus
= trace_ust_create_session(session
->path
, session
->id
, domain
);
2107 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2111 lus
->uid
= session
->uid
;
2112 lus
->gid
= session
->gid
;
2113 session
->ust_session
= lus
;
2115 /* Copy session output to the newly created UST session */
2116 ret
= copy_session_consumer(domain
->type
, session
);
2117 if (ret
!= LTTNG_OK
) {
2125 session
->ust_session
= NULL
;
2130 * Create a kernel tracer session then create the default channel.
2132 static int create_kernel_session(struct ltt_session
*session
)
2136 DBG("Creating kernel session");
2138 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2140 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2144 /* Code flow safety */
2145 assert(session
->kernel_session
);
2147 /* Copy session output to the newly created Kernel session */
2148 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2149 if (ret
!= LTTNG_OK
) {
2153 /* Create directory(ies) on local filesystem. */
2154 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2155 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2156 ret
= run_as_mkdir_recursive(
2157 session
->kernel_session
->consumer
->dst
.trace_path
,
2158 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2160 if (ret
!= -EEXIST
) {
2161 ERR("Trace directory creation error");
2167 session
->kernel_session
->uid
= session
->uid
;
2168 session
->kernel_session
->gid
= session
->gid
;
2173 trace_kernel_destroy_session(session
->kernel_session
);
2174 session
->kernel_session
= NULL
;
2179 * Count number of session permitted by uid/gid.
2181 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2184 struct ltt_session
*session
;
2186 DBG("Counting number of available session for UID %d GID %d",
2188 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2190 * Only list the sessions the user can control.
2192 if (!session_access_ok(session
, uid
, gid
)) {
2201 * Process the command requested by the lttng client within the command
2202 * context structure. This function make sure that the return structure (llm)
2203 * is set and ready for transmission before returning.
2205 * Return any error encountered or 0 for success.
2207 * "sock" is only used for special-case var. len data.
2209 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2213 int need_tracing_session
= 1;
2216 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2220 switch (cmd_ctx
->lsm
->cmd_type
) {
2221 case LTTNG_CREATE_SESSION
:
2222 case LTTNG_DESTROY_SESSION
:
2223 case LTTNG_LIST_SESSIONS
:
2224 case LTTNG_LIST_DOMAINS
:
2225 case LTTNG_START_TRACE
:
2226 case LTTNG_STOP_TRACE
:
2227 case LTTNG_DATA_PENDING
:
2234 if (opt_no_kernel
&& need_domain
2235 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2237 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2239 ret
= LTTNG_ERR_KERN_NA
;
2244 /* Deny register consumer if we already have a spawned consumer. */
2245 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2246 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2247 if (kconsumer_data
.pid
> 0) {
2248 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2249 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2252 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2256 * Check for command that don't needs to allocate a returned payload. We do
2257 * this here so we don't have to make the call for no payload at each
2260 switch(cmd_ctx
->lsm
->cmd_type
) {
2261 case LTTNG_LIST_SESSIONS
:
2262 case LTTNG_LIST_TRACEPOINTS
:
2263 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2264 case LTTNG_LIST_DOMAINS
:
2265 case LTTNG_LIST_CHANNELS
:
2266 case LTTNG_LIST_EVENTS
:
2269 /* Setup lttng message with no payload */
2270 ret
= setup_lttng_msg(cmd_ctx
, 0);
2272 /* This label does not try to unlock the session */
2273 goto init_setup_error
;
2277 /* Commands that DO NOT need a session. */
2278 switch (cmd_ctx
->lsm
->cmd_type
) {
2279 case LTTNG_CREATE_SESSION
:
2280 case LTTNG_CALIBRATE
:
2281 case LTTNG_LIST_SESSIONS
:
2282 case LTTNG_LIST_TRACEPOINTS
:
2283 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2284 need_tracing_session
= 0;
2287 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2289 * We keep the session list lock across _all_ commands
2290 * for now, because the per-session lock does not
2291 * handle teardown properly.
2293 session_lock_list();
2294 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2295 if (cmd_ctx
->session
== NULL
) {
2296 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2297 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2299 /* If no session name specified */
2300 ret
= LTTNG_ERR_SELECT_SESS
;
2304 /* Acquire lock for the session */
2305 session_lock(cmd_ctx
->session
);
2315 * Check domain type for specific "pre-action".
2317 switch (cmd_ctx
->lsm
->domain
.type
) {
2318 case LTTNG_DOMAIN_KERNEL
:
2320 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2324 /* Kernel tracer check */
2325 if (kernel_tracer_fd
== -1) {
2326 /* Basically, load kernel tracer modules */
2327 ret
= init_kernel_tracer();
2333 /* Consumer is in an ERROR state. Report back to client */
2334 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2335 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2339 /* Need a session for kernel command */
2340 if (need_tracing_session
) {
2341 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2342 ret
= create_kernel_session(cmd_ctx
->session
);
2344 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2349 /* Start the kernel consumer daemon */
2350 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2351 if (kconsumer_data
.pid
== 0 &&
2352 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2353 cmd_ctx
->session
->start_consumer
) {
2354 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2355 ret
= start_consumerd(&kconsumer_data
);
2357 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2360 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2362 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2366 * The consumer was just spawned so we need to add the socket to
2367 * the consumer output of the session if exist.
2369 ret
= consumer_create_socket(&kconsumer_data
,
2370 cmd_ctx
->session
->kernel_session
->consumer
);
2377 case LTTNG_DOMAIN_UST
:
2379 /* Consumer is in an ERROR state. Report back to client */
2380 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2381 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2385 if (need_tracing_session
) {
2386 /* Create UST session if none exist. */
2387 if (cmd_ctx
->session
->ust_session
== NULL
) {
2388 ret
= create_ust_session(cmd_ctx
->session
,
2389 &cmd_ctx
->lsm
->domain
);
2390 if (ret
!= LTTNG_OK
) {
2395 /* Start the UST consumer daemons */
2397 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2398 if (consumerd64_bin
[0] != '\0' &&
2399 ustconsumer64_data
.pid
== 0 &&
2400 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2401 cmd_ctx
->session
->start_consumer
) {
2402 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2403 ret
= start_consumerd(&ustconsumer64_data
);
2405 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2406 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2410 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2411 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2413 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2417 * Setup socket for consumer 64 bit. No need for atomic access
2418 * since it was set above and can ONLY be set in this thread.
2420 ret
= consumer_create_socket(&ustconsumer64_data
,
2421 cmd_ctx
->session
->ust_session
->consumer
);
2427 if (consumerd32_bin
[0] != '\0' &&
2428 ustconsumer32_data
.pid
== 0 &&
2429 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
&&
2430 cmd_ctx
->session
->start_consumer
) {
2431 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2432 ret
= start_consumerd(&ustconsumer32_data
);
2434 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2435 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2439 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2440 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2442 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2446 * Setup socket for consumer 64 bit. No need for atomic access
2447 * since it was set above and can ONLY be set in this thread.
2449 ret
= consumer_create_socket(&ustconsumer32_data
,
2450 cmd_ctx
->session
->ust_session
->consumer
);
2462 /* Validate consumer daemon state when start/stop trace command */
2463 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2464 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2465 switch (cmd_ctx
->lsm
->domain
.type
) {
2466 case LTTNG_DOMAIN_UST
:
2467 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2468 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2472 case LTTNG_DOMAIN_KERNEL
:
2473 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2474 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2482 * Check that the UID or GID match that of the tracing session.
2483 * The root user can interact with all sessions.
2485 if (need_tracing_session
) {
2486 if (!session_access_ok(cmd_ctx
->session
,
2487 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2488 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2489 ret
= LTTNG_ERR_EPERM
;
2494 /* Process by command type */
2495 switch (cmd_ctx
->lsm
->cmd_type
) {
2496 case LTTNG_ADD_CONTEXT
:
2498 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2499 cmd_ctx
->lsm
->u
.context
.channel_name
,
2500 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2503 case LTTNG_DISABLE_CHANNEL
:
2505 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2506 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2509 case LTTNG_DISABLE_EVENT
:
2511 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2512 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2513 cmd_ctx
->lsm
->u
.disable
.name
);
2516 case LTTNG_DISABLE_ALL_EVENT
:
2518 DBG("Disabling all events");
2520 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2521 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2524 case LTTNG_DISABLE_CONSUMER
:
2526 ret
= cmd_disable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2529 case LTTNG_ENABLE_CHANNEL
:
2531 ret
= cmd_enable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2532 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2535 case LTTNG_ENABLE_CONSUMER
:
2538 * XXX: 0 means that this URI should be applied on the session. Should
2539 * be a DOMAIN enuam.
2541 ret
= cmd_enable_consumer(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
);
2542 if (ret
!= LTTNG_OK
) {
2546 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2547 /* Add the URI for the UST session if a consumer is present. */
2548 if (cmd_ctx
->session
->ust_session
&&
2549 cmd_ctx
->session
->ust_session
->consumer
) {
2550 ret
= cmd_enable_consumer(LTTNG_DOMAIN_UST
, cmd_ctx
->session
);
2551 } else if (cmd_ctx
->session
->kernel_session
&&
2552 cmd_ctx
->session
->kernel_session
->consumer
) {
2553 ret
= cmd_enable_consumer(LTTNG_DOMAIN_KERNEL
,
2559 case LTTNG_ENABLE_EVENT
:
2561 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2562 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2563 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2566 case LTTNG_ENABLE_ALL_EVENT
:
2568 DBG("Enabling all events");
2570 ret
= cmd_enable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2571 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2572 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2575 case LTTNG_LIST_TRACEPOINTS
:
2577 struct lttng_event
*events
;
2580 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2581 if (nb_events
< 0) {
2582 /* Return value is a negative lttng_error_code. */
2588 * Setup lttng message with payload size set to the event list size in
2589 * bytes and then copy list into the llm payload.
2591 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2597 /* Copy event list into message payload */
2598 memcpy(cmd_ctx
->llm
->payload
, events
,
2599 sizeof(struct lttng_event
) * nb_events
);
2606 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2608 struct lttng_event_field
*fields
;
2611 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2613 if (nb_fields
< 0) {
2614 /* Return value is a negative lttng_error_code. */
2620 * Setup lttng message with payload size set to the event list size in
2621 * bytes and then copy list into the llm payload.
2623 ret
= setup_lttng_msg(cmd_ctx
,
2624 sizeof(struct lttng_event_field
) * nb_fields
);
2630 /* Copy event list into message payload */
2631 memcpy(cmd_ctx
->llm
->payload
, fields
,
2632 sizeof(struct lttng_event_field
) * nb_fields
);
2639 case LTTNG_SET_CONSUMER_URI
:
2642 struct lttng_uri
*uris
;
2644 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2645 len
= nb_uri
* sizeof(struct lttng_uri
);
2648 ret
= LTTNG_ERR_INVALID
;
2652 uris
= zmalloc(len
);
2654 ret
= LTTNG_ERR_FATAL
;
2658 /* Receive variable len data */
2659 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2660 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2662 DBG("No URIs received from client... continuing");
2664 ret
= LTTNG_ERR_SESSION_FAIL
;
2669 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2671 if (ret
!= LTTNG_OK
) {
2677 * XXX: 0 means that this URI should be applied on the session. Should
2678 * be a DOMAIN enuam.
2680 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2681 /* Add the URI for the UST session if a consumer is present. */
2682 if (cmd_ctx
->session
->ust_session
&&
2683 cmd_ctx
->session
->ust_session
->consumer
) {
2684 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2686 } else if (cmd_ctx
->session
->kernel_session
&&
2687 cmd_ctx
->session
->kernel_session
->consumer
) {
2688 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2689 cmd_ctx
->session
, nb_uri
, uris
);
2697 case LTTNG_START_TRACE
:
2699 ret
= cmd_start_trace(cmd_ctx
->session
);
2702 case LTTNG_STOP_TRACE
:
2704 ret
= cmd_stop_trace(cmd_ctx
->session
);
2707 case LTTNG_CREATE_SESSION
:
2710 struct lttng_uri
*uris
= NULL
;
2712 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2713 len
= nb_uri
* sizeof(struct lttng_uri
);
2716 uris
= zmalloc(len
);
2718 ret
= LTTNG_ERR_FATAL
;
2722 /* Receive variable len data */
2723 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2724 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2726 DBG("No URIs received from client... continuing");
2728 ret
= LTTNG_ERR_SESSION_FAIL
;
2733 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2734 DBG("Creating session with ONE network URI is a bad call");
2735 ret
= LTTNG_ERR_SESSION_FAIL
;
2741 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2748 case LTTNG_DESTROY_SESSION
:
2750 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2752 /* Set session to NULL so we do not unlock it after free. */
2753 cmd_ctx
->session
= NULL
;
2756 case LTTNG_LIST_DOMAINS
:
2759 struct lttng_domain
*domains
;
2761 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2763 /* Return value is a negative lttng_error_code. */
2768 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2773 /* Copy event list into message payload */
2774 memcpy(cmd_ctx
->llm
->payload
, domains
,
2775 nb_dom
* sizeof(struct lttng_domain
));
2782 case LTTNG_LIST_CHANNELS
:
2785 struct lttng_channel
*channels
;
2787 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2788 cmd_ctx
->session
, &channels
);
2790 /* Return value is a negative lttng_error_code. */
2795 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2800 /* Copy event list into message payload */
2801 memcpy(cmd_ctx
->llm
->payload
, channels
,
2802 nb_chan
* sizeof(struct lttng_channel
));
2809 case LTTNG_LIST_EVENTS
:
2812 struct lttng_event
*events
= NULL
;
2814 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2815 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2817 /* Return value is a negative lttng_error_code. */
2822 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2827 /* Copy event list into message payload */
2828 memcpy(cmd_ctx
->llm
->payload
, events
,
2829 nb_event
* sizeof(struct lttng_event
));
2836 case LTTNG_LIST_SESSIONS
:
2838 unsigned int nr_sessions
;
2840 session_lock_list();
2841 nr_sessions
= lttng_sessions_count(
2842 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2843 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2845 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2847 session_unlock_list();
2851 /* Filled the session array */
2852 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2853 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2854 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2856 session_unlock_list();
2861 case LTTNG_CALIBRATE
:
2863 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
2864 &cmd_ctx
->lsm
->u
.calibrate
);
2867 case LTTNG_REGISTER_CONSUMER
:
2869 struct consumer_data
*cdata
;
2871 switch (cmd_ctx
->lsm
->domain
.type
) {
2872 case LTTNG_DOMAIN_KERNEL
:
2873 cdata
= &kconsumer_data
;
2876 ret
= LTTNG_ERR_UND
;
2880 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2881 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
2884 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
2886 struct lttng_filter_bytecode
*bytecode
;
2888 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
2889 ret
= LTTNG_ERR_FILTER_INVAL
;
2892 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
2893 ret
= LTTNG_ERR_FILTER_INVAL
;
2896 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2898 ret
= LTTNG_ERR_FILTER_NOMEM
;
2901 /* Receive var. len. data */
2902 DBG("Receiving var len data from client ...");
2903 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
2904 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
2906 DBG("Nothing recv() from client var len data... continuing");
2908 ret
= LTTNG_ERR_FILTER_INVAL
;
2912 if (bytecode
->len
+ sizeof(*bytecode
)
2913 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
2915 ret
= LTTNG_ERR_FILTER_INVAL
;
2919 ret
= cmd_enable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2920 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2921 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
2924 case LTTNG_DATA_PENDING
:
2926 ret
= cmd_data_pending(cmd_ctx
->session
);
2930 ret
= LTTNG_ERR_UND
;
2935 if (cmd_ctx
->llm
== NULL
) {
2936 DBG("Missing llm structure. Allocating one.");
2937 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
2941 /* Set return code */
2942 cmd_ctx
->llm
->ret_code
= ret
;
2944 if (cmd_ctx
->session
) {
2945 session_unlock(cmd_ctx
->session
);
2947 if (need_tracing_session
) {
2948 session_unlock_list();
2955 * Thread managing health check socket.
2957 static void *thread_manage_health(void *data
)
2959 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
2960 uint32_t revents
, nb_fd
;
2961 struct lttng_poll_event events
;
2962 struct lttcomm_health_msg msg
;
2963 struct lttcomm_health_data reply
;
2965 DBG("[thread] Manage health check started");
2967 rcu_register_thread();
2969 /* Create unix socket */
2970 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
2972 ERR("Unable to create health check Unix socket");
2978 * Set the CLOEXEC flag. Return code is useless because either way, the
2981 (void) utils_set_fd_cloexec(sock
);
2983 ret
= lttcomm_listen_unix_sock(sock
);
2989 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2990 * more will be added to this poll set.
2992 ret
= create_thread_poll_set(&events
, 2);
2997 /* Add the application registration socket */
2998 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3004 DBG("Health check ready");
3006 /* Inifinite blocking call, waiting for transmission */
3008 ret
= lttng_poll_wait(&events
, -1);
3011 * Restart interrupted system call.
3013 if (errno
== EINTR
) {
3021 for (i
= 0; i
< nb_fd
; i
++) {
3022 /* Fetch once the poll data */
3023 revents
= LTTNG_POLL_GETEV(&events
, i
);
3024 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3026 /* Thread quit pipe has been closed. Killing thread. */
3027 ret
= check_thread_quit_pipe(pollfd
, revents
);
3033 /* Event on the registration socket */
3034 if (pollfd
== sock
) {
3035 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3036 ERR("Health socket poll error");
3042 new_sock
= lttcomm_accept_unix_sock(sock
);
3048 * Set the CLOEXEC flag. Return code is useless because either way, the
3051 (void) utils_set_fd_cloexec(new_sock
);
3053 DBG("Receiving data from client for health...");
3054 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3056 DBG("Nothing recv() from client... continuing");
3057 ret
= close(new_sock
);
3065 rcu_thread_online();
3067 switch (msg
.component
) {
3068 case LTTNG_HEALTH_CMD
:
3069 reply
.ret_code
= health_check_state(&health_thread_cmd
);
3071 case LTTNG_HEALTH_APP_MANAGE
:
3072 reply
.ret_code
= health_check_state(&health_thread_app_manage
);
3074 case LTTNG_HEALTH_APP_REG
:
3075 reply
.ret_code
= health_check_state(&health_thread_app_reg
);
3077 case LTTNG_HEALTH_KERNEL
:
3078 reply
.ret_code
= health_check_state(&health_thread_kernel
);
3080 case LTTNG_HEALTH_CONSUMER
:
3081 reply
.ret_code
= check_consumer_health();
3083 case LTTNG_HEALTH_ALL
:
3085 health_check_state(&health_thread_app_manage
) &&
3086 health_check_state(&health_thread_app_reg
) &&
3087 health_check_state(&health_thread_cmd
) &&
3088 health_check_state(&health_thread_kernel
) &&
3089 check_consumer_health();
3092 reply
.ret_code
= LTTNG_ERR_UND
;
3097 * Flip ret value since 0 is a success and 1 indicates a bad health for
3098 * the client where in the sessiond it is the opposite. Again, this is
3099 * just to make things easier for us poor developer which enjoy a lot
3102 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3103 reply
.ret_code
= !reply
.ret_code
;
3106 DBG2("Health check return value %d", reply
.ret_code
);
3108 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3110 ERR("Failed to send health data back to client");
3113 /* End of transmission */
3114 ret
= close(new_sock
);
3124 ERR("Health error occurred in %s", __func__
);
3126 DBG("Health check thread dying");
3127 unlink(health_unix_sock_path
);
3134 if (new_sock
>= 0) {
3135 ret
= close(new_sock
);
3141 lttng_poll_clean(&events
);
3143 rcu_unregister_thread();
3148 * This thread manage all clients request using the unix client socket for
3151 static void *thread_manage_clients(void *data
)
3153 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3155 uint32_t revents
, nb_fd
;
3156 struct command_ctx
*cmd_ctx
= NULL
;
3157 struct lttng_poll_event events
;
3159 DBG("[thread] Manage client started");
3161 rcu_register_thread();
3163 if (testpoint(thread_manage_clients
)) {
3164 goto error_testpoint
;
3167 health_code_update(&health_thread_cmd
);
3169 ret
= lttcomm_listen_unix_sock(client_sock
);
3175 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3176 * more will be added to this poll set.
3178 ret
= create_thread_poll_set(&events
, 2);
3180 goto error_create_poll
;
3183 /* Add the application registration socket */
3184 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3190 * Notify parent pid that we are ready to accept command for client side.
3192 if (opt_sig_parent
) {
3193 kill(ppid
, SIGUSR1
);
3196 if (testpoint(thread_manage_clients_before_loop
)) {
3200 health_code_update(&health_thread_cmd
);
3203 DBG("Accepting client command ...");
3205 /* Inifinite blocking call, waiting for transmission */
3207 health_poll_update(&health_thread_cmd
);
3208 ret
= lttng_poll_wait(&events
, -1);
3209 health_poll_update(&health_thread_cmd
);
3212 * Restart interrupted system call.
3214 if (errno
== EINTR
) {
3222 for (i
= 0; i
< nb_fd
; i
++) {
3223 /* Fetch once the poll data */
3224 revents
= LTTNG_POLL_GETEV(&events
, i
);
3225 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3227 health_code_update(&health_thread_cmd
);
3229 /* Thread quit pipe has been closed. Killing thread. */
3230 ret
= check_thread_quit_pipe(pollfd
, revents
);
3236 /* Event on the registration socket */
3237 if (pollfd
== client_sock
) {
3238 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3239 ERR("Client socket poll error");
3245 DBG("Wait for client response");
3247 health_code_update(&health_thread_cmd
);
3249 sock
= lttcomm_accept_unix_sock(client_sock
);
3255 * Set the CLOEXEC flag. Return code is useless because either way, the
3258 (void) utils_set_fd_cloexec(sock
);
3260 /* Set socket option for credentials retrieval */
3261 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3266 /* Allocate context command to process the client request */
3267 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3268 if (cmd_ctx
== NULL
) {
3269 PERROR("zmalloc cmd_ctx");
3273 /* Allocate data buffer for reception */
3274 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3275 if (cmd_ctx
->lsm
== NULL
) {
3276 PERROR("zmalloc cmd_ctx->lsm");
3280 cmd_ctx
->llm
= NULL
;
3281 cmd_ctx
->session
= NULL
;
3283 health_code_update(&health_thread_cmd
);
3286 * Data is received from the lttng client. The struct
3287 * lttcomm_session_msg (lsm) contains the command and data request of
3290 DBG("Receiving data from client ...");
3291 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3292 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3294 DBG("Nothing recv() from client... continuing");
3300 clean_command_ctx(&cmd_ctx
);
3304 health_code_update(&health_thread_cmd
);
3306 // TODO: Validate cmd_ctx including sanity check for
3307 // security purpose.
3309 rcu_thread_online();
3311 * This function dispatch the work to the kernel or userspace tracer
3312 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3313 * informations for the client. The command context struct contains
3314 * everything this function may needs.
3316 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3317 rcu_thread_offline();
3327 * TODO: Inform client somehow of the fatal error. At
3328 * this point, ret < 0 means that a zmalloc failed
3329 * (ENOMEM). Error detected but still accept
3330 * command, unless a socket error has been
3333 clean_command_ctx(&cmd_ctx
);
3337 health_code_update(&health_thread_cmd
);
3339 DBG("Sending response (size: %d, retcode: %s)",
3340 cmd_ctx
->lttng_msg_size
,
3341 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3342 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3344 ERR("Failed to send data back to client");
3347 /* End of transmission */
3354 clean_command_ctx(&cmd_ctx
);
3356 health_code_update(&health_thread_cmd
);
3368 lttng_poll_clean(&events
);
3369 clean_command_ctx(&cmd_ctx
);
3374 unlink(client_unix_sock_path
);
3375 if (client_sock
>= 0) {
3376 ret
= close(client_sock
);
3383 health_error(&health_thread_cmd
);
3384 ERR("Health error occurred in %s", __func__
);
3387 health_exit(&health_thread_cmd
);
3389 DBG("Client thread dying");
3391 rcu_unregister_thread();
3397 * usage function on stderr
3399 static void usage(void)
3401 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3402 fprintf(stderr
, " -h, --help Display this usage.\n");
3403 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3404 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3405 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3406 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3407 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3408 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3409 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3410 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3411 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3412 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3413 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3414 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3415 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3416 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3417 fprintf(stderr
, " -V, --version Show version number.\n");
3418 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3419 fprintf(stderr
, " -q, --quiet No output at all.\n");
3420 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3421 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3422 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3423 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3427 * daemon argument parsing
3429 static int parse_args(int argc
, char **argv
)
3433 static struct option long_options
[] = {
3434 { "client-sock", 1, 0, 'c' },
3435 { "apps-sock", 1, 0, 'a' },
3436 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3437 { "kconsumerd-err-sock", 1, 0, 'E' },
3438 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3439 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3440 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3441 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3442 { "consumerd32-path", 1, 0, 'u' },
3443 { "consumerd32-libdir", 1, 0, 'U' },
3444 { "consumerd64-path", 1, 0, 't' },
3445 { "consumerd64-libdir", 1, 0, 'T' },
3446 { "daemonize", 0, 0, 'd' },
3447 { "sig-parent", 0, 0, 'S' },
3448 { "help", 0, 0, 'h' },
3449 { "group", 1, 0, 'g' },
3450 { "version", 0, 0, 'V' },
3451 { "quiet", 0, 0, 'q' },
3452 { "verbose", 0, 0, 'v' },
3453 { "verbose-consumer", 0, 0, 'Z' },
3454 { "no-kernel", 0, 0, 'N' },
3455 { "pidfile", 1, 0, 'p' },
3460 int option_index
= 0;
3461 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3462 long_options
, &option_index
);
3469 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3471 fprintf(stderr
, " with arg %s\n", optarg
);
3475 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3478 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3484 opt_tracing_group
= optarg
;
3490 fprintf(stdout
, "%s\n", VERSION
);
3496 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3499 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3502 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3505 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3508 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3511 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3517 lttng_opt_quiet
= 1;
3520 /* Verbose level can increase using multiple -v */
3521 lttng_opt_verbose
+= 1;
3524 opt_verbose_consumer
+= 1;
3527 consumerd32_bin
= optarg
;
3530 consumerd32_libdir
= optarg
;
3533 consumerd64_bin
= optarg
;
3536 consumerd64_libdir
= optarg
;
3539 opt_pidfile
= optarg
;
3542 /* Unknown option or other error.
3543 * Error is printed by getopt, just return */
3552 * Creates the two needed socket by the daemon.
3553 * apps_sock - The communication socket for all UST apps.
3554 * client_sock - The communication of the cli tool (lttng).
3556 static int init_daemon_socket(void)
3561 old_umask
= umask(0);
3563 /* Create client tool unix socket */
3564 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3565 if (client_sock
< 0) {
3566 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3571 /* Set the cloexec flag */
3572 ret
= utils_set_fd_cloexec(client_sock
);
3574 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3575 "Continuing but note that the consumer daemon will have a "
3576 "reference to this socket on exec()", client_sock
);
3579 /* File permission MUST be 660 */
3580 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3582 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3587 /* Create the application unix socket */
3588 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3589 if (apps_sock
< 0) {
3590 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3595 /* Set the cloexec flag */
3596 ret
= utils_set_fd_cloexec(apps_sock
);
3598 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3599 "Continuing but note that the consumer daemon will have a "
3600 "reference to this socket on exec()", apps_sock
);
3603 /* File permission MUST be 666 */
3604 ret
= chmod(apps_unix_sock_path
,
3605 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3607 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3612 DBG3("Session daemon client socket %d and application socket %d created",
3613 client_sock
, apps_sock
);
3621 * Check if the global socket is available, and if a daemon is answering at the
3622 * other side. If yes, error is returned.
3624 static int check_existing_daemon(void)
3626 /* Is there anybody out there ? */
3627 if (lttng_session_daemon_alive()) {
3635 * Set the tracing group gid onto the client socket.
3637 * Race window between mkdir and chown is OK because we are going from more
3638 * permissive (root.root) to less permissive (root.tracing).
3640 static int set_permissions(char *rundir
)
3645 ret
= allowed_group();
3647 WARN("No tracing group detected");
3654 /* Set lttng run dir */
3655 ret
= chown(rundir
, 0, gid
);
3657 ERR("Unable to set group on %s", rundir
);
3661 /* Ensure tracing group can search the run dir */
3662 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3664 ERR("Unable to set permissions on %s", rundir
);
3668 /* lttng client socket path */
3669 ret
= chown(client_unix_sock_path
, 0, gid
);
3671 ERR("Unable to set group on %s", client_unix_sock_path
);
3675 /* kconsumer error socket path */
3676 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3678 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3682 /* 64-bit ustconsumer error socket path */
3683 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3685 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3689 /* 32-bit ustconsumer compat32 error socket path */
3690 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3692 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3696 DBG("All permissions are set");
3703 * Create the lttng run directory needed for all global sockets and pipe.
3705 static int create_lttng_rundir(const char *rundir
)
3709 DBG3("Creating LTTng run directory: %s", rundir
);
3711 ret
= mkdir(rundir
, S_IRWXU
);
3713 if (errno
!= EEXIST
) {
3714 ERR("Unable to create %s", rundir
);
3726 * Setup sockets and directory needed by the kconsumerd communication with the
3729 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3733 char path
[PATH_MAX
];
3735 switch (consumer_data
->type
) {
3736 case LTTNG_CONSUMER_KERNEL
:
3737 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3739 case LTTNG_CONSUMER64_UST
:
3740 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3742 case LTTNG_CONSUMER32_UST
:
3743 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3746 ERR("Consumer type unknown");
3751 DBG2("Creating consumer directory: %s", path
);
3753 ret
= mkdir(path
, S_IRWXU
);
3755 if (errno
!= EEXIST
) {
3757 ERR("Failed to create %s", path
);
3763 /* Create the kconsumerd error unix socket */
3764 consumer_data
->err_sock
=
3765 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3766 if (consumer_data
->err_sock
< 0) {
3767 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3772 /* File permission MUST be 660 */
3773 ret
= chmod(consumer_data
->err_unix_sock_path
,
3774 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3776 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3786 * Signal handler for the daemon
3788 * Simply stop all worker threads, leaving main() return gracefully after
3789 * joining all threads and calling cleanup().
3791 static void sighandler(int sig
)
3795 DBG("SIGPIPE caught");
3798 DBG("SIGINT caught");
3802 DBG("SIGTERM caught");
3811 * Setup signal handler for :
3812 * SIGINT, SIGTERM, SIGPIPE
3814 static int set_signal_handler(void)
3817 struct sigaction sa
;
3820 if ((ret
= sigemptyset(&sigset
)) < 0) {
3821 PERROR("sigemptyset");
3825 sa
.sa_handler
= sighandler
;
3826 sa
.sa_mask
= sigset
;
3828 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3829 PERROR("sigaction");
3833 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3834 PERROR("sigaction");
3838 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3839 PERROR("sigaction");
3843 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3849 * Set open files limit to unlimited. This daemon can open a large number of
3850 * file descriptors in order to consumer multiple kernel traces.
3852 static void set_ulimit(void)
3857 /* The kernel does not allowed an infinite limit for open files */
3858 lim
.rlim_cur
= 65535;
3859 lim
.rlim_max
= 65535;
3861 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
3863 PERROR("failed to set open files limit");
3868 * Write pidfile using the rundir and opt_pidfile.
3870 static void write_pidfile(void)
3873 char pidfile_path
[PATH_MAX
];
3878 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
3880 /* Build pidfile path from rundir and opt_pidfile. */
3881 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
3882 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
3884 PERROR("snprintf pidfile path");
3890 * Create pid file in rundir. Return value is of no importance. The
3891 * execution will continue even though we are not able to write the file.
3893 (void) utils_create_pid_file(getpid(), pidfile_path
);
3902 int main(int argc
, char **argv
)
3906 const char *home_path
, *env_app_timeout
;
3908 init_kernel_workarounds();
3910 rcu_register_thread();
3912 setup_consumerd_path();
3914 /* Parse arguments */
3916 if ((ret
= parse_args(argc
, argv
)) < 0) {
3926 * child: setsid, close FD 0, 1, 2, chdir /
3927 * parent: exit (if fork is successful)
3935 * We are in the child. Make sure all other file
3936 * descriptors are closed, in case we are called with
3937 * more opened file descriptors than the standard ones.
3939 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
3944 /* Create thread quit pipe */
3945 if ((ret
= init_thread_quit_pipe()) < 0) {
3949 /* Check if daemon is UID = 0 */
3950 is_root
= !getuid();
3953 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
3955 /* Create global run dir with root access */
3956 ret
= create_lttng_rundir(rundir
);
3961 if (strlen(apps_unix_sock_path
) == 0) {
3962 snprintf(apps_unix_sock_path
, PATH_MAX
,
3963 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
3966 if (strlen(client_unix_sock_path
) == 0) {
3967 snprintf(client_unix_sock_path
, PATH_MAX
,
3968 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
3971 /* Set global SHM for ust */
3972 if (strlen(wait_shm_path
) == 0) {
3973 snprintf(wait_shm_path
, PATH_MAX
,
3974 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
3977 if (strlen(health_unix_sock_path
) == 0) {
3978 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
3979 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
3982 /* Setup kernel consumerd path */
3983 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
3984 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
3985 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
3986 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
3988 DBG2("Kernel consumer err path: %s",
3989 kconsumer_data
.err_unix_sock_path
);
3990 DBG2("Kernel consumer cmd path: %s",
3991 kconsumer_data
.cmd_unix_sock_path
);
3993 home_path
= get_home_dir();
3994 if (home_path
== NULL
) {
3995 /* TODO: Add --socket PATH option */
3996 ERR("Can't get HOME directory for sockets creation.");
4002 * Create rundir from home path. This will create something like
4005 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4011 ret
= create_lttng_rundir(rundir
);
4016 if (strlen(apps_unix_sock_path
) == 0) {
4017 snprintf(apps_unix_sock_path
, PATH_MAX
,
4018 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4021 /* Set the cli tool unix socket path */
4022 if (strlen(client_unix_sock_path
) == 0) {
4023 snprintf(client_unix_sock_path
, PATH_MAX
,
4024 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4027 /* Set global SHM for ust */
4028 if (strlen(wait_shm_path
) == 0) {
4029 snprintf(wait_shm_path
, PATH_MAX
,
4030 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, geteuid());
4033 /* Set health check Unix path */
4034 if (strlen(health_unix_sock_path
) == 0) {
4035 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4036 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4040 /* Set consumer initial state */
4041 kernel_consumerd_state
= CONSUMER_STOPPED
;
4042 ust_consumerd_state
= CONSUMER_STOPPED
;
4044 DBG("Client socket path %s", client_unix_sock_path
);
4045 DBG("Application socket path %s", apps_unix_sock_path
);
4046 DBG("LTTng run directory path: %s", rundir
);
4048 /* 32 bits consumerd path setup */
4049 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4050 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4051 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4052 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4054 DBG2("UST consumer 32 bits err path: %s",
4055 ustconsumer32_data
.err_unix_sock_path
);
4056 DBG2("UST consumer 32 bits cmd path: %s",
4057 ustconsumer32_data
.cmd_unix_sock_path
);
4059 /* 64 bits consumerd path setup */
4060 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4061 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4062 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4063 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4065 DBG2("UST consumer 64 bits err path: %s",
4066 ustconsumer64_data
.err_unix_sock_path
);
4067 DBG2("UST consumer 64 bits cmd path: %s",
4068 ustconsumer64_data
.cmd_unix_sock_path
);
4071 * See if daemon already exist.
4073 if ((ret
= check_existing_daemon()) < 0) {
4074 ERR("Already running daemon.\n");
4076 * We do not goto exit because we must not cleanup()
4077 * because a daemon is already running.
4083 * Init UST app hash table. Alloc hash table before this point since
4084 * cleanup() can get called after that point.
4088 /* After this point, we can safely call cleanup() with "goto exit" */
4091 * These actions must be executed as root. We do that *after* setting up
4092 * the sockets path because we MUST make the check for another daemon using
4093 * those paths *before* trying to set the kernel consumer sockets and init
4097 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4102 /* Setup kernel tracer */
4103 if (!opt_no_kernel
) {
4104 init_kernel_tracer();
4107 /* Set ulimit for open files */
4110 /* init lttng_fd tracking must be done after set_ulimit. */
4113 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4118 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4123 if ((ret
= set_signal_handler()) < 0) {
4127 /* Setup the needed unix socket */
4128 if ((ret
= init_daemon_socket()) < 0) {
4132 /* Set credentials to socket */
4133 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4137 /* Get parent pid if -S, --sig-parent is specified. */
4138 if (opt_sig_parent
) {
4142 /* Setup the kernel pipe for waking up the kernel thread */
4143 if (is_root
&& !opt_no_kernel
) {
4144 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4149 /* Setup the thread apps communication pipe. */
4150 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4154 /* Init UST command queue. */
4155 cds_wfq_init(&ust_cmd_queue
.queue
);
4158 * Get session list pointer. This pointer MUST NOT be free(). This list is
4159 * statically declared in session.c
4161 session_list_ptr
= session_get_list();
4163 /* Set up max poll set size */
4164 lttng_poll_set_max_size();
4168 /* Init all health thread counters. */
4169 health_init(&health_thread_cmd
);
4170 health_init(&health_thread_kernel
);
4171 health_init(&health_thread_app_manage
);
4172 health_init(&health_thread_app_reg
);
4175 * Init health counters of the consumer thread. We do a quick hack here to
4176 * the state of the consumer health is fine even if the thread is not
4177 * started. Once the thread starts, the health state is updated with a poll
4178 * value to set a health code path. This is simply to ease our life and has
4179 * no cost what so ever.
4181 health_init(&kconsumer_data
.health
);
4182 health_poll_update(&kconsumer_data
.health
);
4183 health_init(&ustconsumer32_data
.health
);
4184 health_poll_update(&ustconsumer32_data
.health
);
4185 health_init(&ustconsumer64_data
.health
);
4186 health_poll_update(&ustconsumer64_data
.health
);
4188 /* Check for the application socket timeout env variable. */
4189 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4190 if (env_app_timeout
) {
4191 app_socket_timeout
= atoi(env_app_timeout
);
4193 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4198 /* Create thread to manage the client socket */
4199 ret
= pthread_create(&health_thread
, NULL
,
4200 thread_manage_health
, (void *) NULL
);
4202 PERROR("pthread_create health");
4206 /* Create thread to manage the client socket */
4207 ret
= pthread_create(&client_thread
, NULL
,
4208 thread_manage_clients
, (void *) NULL
);
4210 PERROR("pthread_create clients");
4214 /* Create thread to dispatch registration */
4215 ret
= pthread_create(&dispatch_thread
, NULL
,
4216 thread_dispatch_ust_registration
, (void *) NULL
);
4218 PERROR("pthread_create dispatch");
4222 /* Create thread to manage application registration. */
4223 ret
= pthread_create(®_apps_thread
, NULL
,
4224 thread_registration_apps
, (void *) NULL
);
4226 PERROR("pthread_create registration");
4230 /* Create thread to manage application socket */
4231 ret
= pthread_create(&apps_thread
, NULL
,
4232 thread_manage_apps
, (void *) NULL
);
4234 PERROR("pthread_create apps");
4238 /* Don't start this thread if kernel tracing is not requested nor root */
4239 if (is_root
&& !opt_no_kernel
) {
4240 /* Create kernel thread to manage kernel event */
4241 ret
= pthread_create(&kernel_thread
, NULL
,
4242 thread_manage_kernel
, (void *) NULL
);
4244 PERROR("pthread_create kernel");
4248 ret
= pthread_join(kernel_thread
, &status
);
4250 PERROR("pthread_join");
4251 goto error
; /* join error, exit without cleanup */
4256 ret
= pthread_join(apps_thread
, &status
);
4258 PERROR("pthread_join");
4259 goto error
; /* join error, exit without cleanup */
4263 ret
= pthread_join(reg_apps_thread
, &status
);
4265 PERROR("pthread_join");
4266 goto error
; /* join error, exit without cleanup */
4270 ret
= pthread_join(dispatch_thread
, &status
);
4272 PERROR("pthread_join");
4273 goto error
; /* join error, exit without cleanup */
4277 ret
= pthread_join(client_thread
, &status
);
4279 PERROR("pthread_join");
4280 goto error
; /* join error, exit without cleanup */
4283 ret
= join_consumer_thread(&kconsumer_data
);
4285 PERROR("join_consumer");
4286 goto error
; /* join error, exit without cleanup */
4289 ret
= join_consumer_thread(&ustconsumer32_data
);
4291 PERROR("join_consumer ust32");
4292 goto error
; /* join error, exit without cleanup */
4295 ret
= join_consumer_thread(&ustconsumer64_data
);
4297 PERROR("join_consumer ust64");
4298 goto error
; /* join error, exit without cleanup */
4302 ret
= pthread_join(health_thread
, &status
);
4304 PERROR("pthread_join health thread");
4305 goto error
; /* join error, exit without cleanup */
4311 * cleanup() is called when no other thread is running.
4313 rcu_thread_online();
4315 rcu_thread_offline();
4316 rcu_unregister_thread();