2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
34 #include <sys/types.h>
36 #include <urcu/uatomic.h>
40 #include <common/common.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
48 #include "lttng-sessiond.h"
49 #include "buffer-registry.h"
56 #include "kernel-consumer.h"
60 #include "ust-consumer.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
67 #define CONSUMERD_FILE "lttng-consumerd"
70 const char default_tracing_group
[] = DEFAULT_TRACING_GROUP
;
73 const char *opt_tracing_group
;
74 static const char *opt_pidfile
;
75 static int opt_sig_parent
;
76 static int opt_verbose_consumer
;
77 static int opt_daemon
;
78 static int opt_no_kernel
;
79 static int is_root
; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid
; /* Parent PID for --sig-parent option */
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
87 static struct consumer_data kconsumer_data
= {
88 .type
= LTTNG_CONSUMER_KERNEL
,
89 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
90 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
93 .metadata_sock
.fd
= -1,
94 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
95 .lock
= PTHREAD_MUTEX_INITIALIZER
,
96 .cond
= PTHREAD_COND_INITIALIZER
,
97 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
99 static struct consumer_data ustconsumer64_data
= {
100 .type
= LTTNG_CONSUMER64_UST
,
101 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
102 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
105 .metadata_sock
.fd
= -1,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer32_data
= {
112 .type
= LTTNG_CONSUMER32_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
117 .metadata_sock
.fd
= -1,
118 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 .lock
= PTHREAD_MUTEX_INITIALIZER
,
120 .cond
= PTHREAD_COND_INITIALIZER
,
121 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
124 /* Shared between threads */
125 static int dispatch_thread_exit
;
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path
[PATH_MAX
];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path
[PATH_MAX
];
131 /* global wait shm path for UST */
132 static char wait_shm_path
[PATH_MAX
];
133 /* Global health check unix path */
134 static char health_unix_sock_path
[PATH_MAX
];
136 /* Sockets and FDs */
137 static int client_sock
= -1;
138 static int apps_sock
= -1;
139 int kernel_tracer_fd
= -1;
140 static int kernel_poll_pipe
[2] = { -1, -1 };
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
146 static int thread_quit_pipe
[2] = { -1, -1 };
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
152 static int apps_cmd_pipe
[2] = { -1, -1 };
154 int apps_cmd_notify_pipe
[2] = { -1, -1 };
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread
;
158 static pthread_t apps_notify_thread
;
159 static pthread_t reg_apps_thread
;
160 static pthread_t client_thread
;
161 static pthread_t kernel_thread
;
162 static pthread_t dispatch_thread
;
163 static pthread_t health_thread
;
166 * UST registration command queue. This queue is tied with a futex and uses a N
167 * wakers / 1 waiter implemented and detailed in futex.c/.h
169 * The thread_manage_apps and thread_dispatch_ust_registration interact with
170 * this queue and the wait/wake scheme.
172 static struct ust_cmd_queue ust_cmd_queue
;
175 * Pointer initialized before thread creation.
177 * This points to the tracing session list containing the session count and a
178 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
179 * MUST NOT be taken if you call a public function in session.c.
181 * The lock is nested inside the structure: session_list_ptr->lock. Please use
182 * session_lock_list and session_unlock_list for lock acquisition.
184 static struct ltt_session_list
*session_list_ptr
;
186 int ust_consumerd64_fd
= -1;
187 int ust_consumerd32_fd
= -1;
189 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
190 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
191 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
192 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
194 static const char *module_proc_lttng
= "/proc/lttng";
197 * Consumer daemon state which is changed when spawning it, killing it or in
198 * case of a fatal error.
200 enum consumerd_state
{
201 CONSUMER_STARTED
= 1,
202 CONSUMER_STOPPED
= 2,
207 * This consumer daemon state is used to validate if a client command will be
208 * able to reach the consumer. If not, the client is informed. For instance,
209 * doing a "lttng start" when the consumer state is set to ERROR will return an
210 * error to the client.
212 * The following example shows a possible race condition of this scheme:
214 * consumer thread error happens
216 * client cmd checks state -> still OK
217 * consumer thread exit, sets error
218 * client cmd try to talk to consumer
221 * However, since the consumer is a different daemon, we have no way of making
222 * sure the command will reach it safely even with this state flag. This is why
223 * we consider that up to the state validation during command processing, the
224 * command is safe. After that, we can not guarantee the correctness of the
225 * client request vis-a-vis the consumer.
227 static enum consumerd_state ust_consumerd_state
;
228 static enum consumerd_state kernel_consumerd_state
;
231 * Socket timeout for receiving and sending in seconds.
233 static int app_socket_timeout
;
235 /* Set in main() with the current page size. */
239 void setup_consumerd_path(void)
241 const char *bin
, *libdir
;
244 * Allow INSTALL_BIN_PATH to be used as a target path for the
245 * native architecture size consumer if CONFIG_CONSUMER*_PATH
246 * has not been defined.
248 #if (CAA_BITS_PER_LONG == 32)
249 if (!consumerd32_bin
[0]) {
250 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
252 if (!consumerd32_libdir
[0]) {
253 consumerd32_libdir
= INSTALL_LIB_PATH
;
255 #elif (CAA_BITS_PER_LONG == 64)
256 if (!consumerd64_bin
[0]) {
257 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
259 if (!consumerd64_libdir
[0]) {
260 consumerd64_libdir
= INSTALL_LIB_PATH
;
263 #error "Unknown bitness"
267 * runtime env. var. overrides the build default.
269 bin
= getenv("LTTNG_CONSUMERD32_BIN");
271 consumerd32_bin
= bin
;
273 bin
= getenv("LTTNG_CONSUMERD64_BIN");
275 consumerd64_bin
= bin
;
277 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
279 consumerd32_libdir
= libdir
;
281 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
283 consumerd64_libdir
= libdir
;
288 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
290 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
296 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
302 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
314 * Check if the thread quit pipe was triggered.
316 * Return 1 if it was triggered else 0;
318 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
320 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
328 * Return group ID of the tracing group or -1 if not found.
330 static gid_t
allowed_group(void)
334 if (opt_tracing_group
) {
335 grp
= getgrnam(opt_tracing_group
);
337 grp
= getgrnam(default_tracing_group
);
347 * Init thread quit pipe.
349 * Return -1 on error or 0 if all pipes are created.
351 static int init_thread_quit_pipe(void)
355 ret
= pipe(thread_quit_pipe
);
357 PERROR("thread quit pipe");
361 for (i
= 0; i
< 2; i
++) {
362 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
374 * Stop all threads by closing the thread quit pipe.
376 static void stop_threads(void)
380 /* Stopping all threads */
381 DBG("Terminating all threads");
382 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
384 ERR("write error on thread quit pipe");
387 /* Dispatch thread */
388 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
389 futex_nto1_wake(&ust_cmd_queue
.futex
);
395 static void cleanup(void)
399 struct ltt_session
*sess
, *stmp
;
403 /* First thing first, stop all threads */
404 utils_close_pipe(thread_quit_pipe
);
407 * If opt_pidfile is undefined, the default file will be wiped when
408 * removing the rundir.
411 ret
= remove(opt_pidfile
);
413 PERROR("remove pidfile %s", opt_pidfile
);
417 DBG("Removing %s directory", rundir
);
418 ret
= asprintf(&cmd
, "rm -rf %s", rundir
);
420 ERR("asprintf failed. Something is really wrong!");
423 /* Remove lttng run directory */
426 ERR("Unable to clean %s", rundir
);
431 DBG("Cleaning up all sessions");
433 /* Destroy session list mutex */
434 if (session_list_ptr
!= NULL
) {
435 pthread_mutex_destroy(&session_list_ptr
->lock
);
437 /* Cleanup ALL session */
438 cds_list_for_each_entry_safe(sess
, stmp
,
439 &session_list_ptr
->head
, list
) {
440 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
444 DBG("Closing all UST sockets");
445 ust_app_clean_list();
446 buffer_reg_destroy_registries();
448 if (is_root
&& !opt_no_kernel
) {
449 DBG2("Closing kernel fd");
450 if (kernel_tracer_fd
>= 0) {
451 ret
= close(kernel_tracer_fd
);
456 DBG("Unloading kernel modules");
457 modprobe_remove_lttng_all();
461 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
462 "Matthew, BEET driven development works!%c[%dm",
463 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
468 * Send data on a unix socket using the liblttsessiondcomm API.
470 * Return lttcomm error code.
472 static int send_unix_sock(int sock
, void *buf
, size_t len
)
474 /* Check valid length */
479 return lttcomm_send_unix_sock(sock
, buf
, len
);
483 * Free memory of a command context structure.
485 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
487 DBG("Clean command context structure");
489 if ((*cmd_ctx
)->llm
) {
490 free((*cmd_ctx
)->llm
);
492 if ((*cmd_ctx
)->lsm
) {
493 free((*cmd_ctx
)->lsm
);
501 * Notify UST applications using the shm mmap futex.
503 static int notify_ust_apps(int active
)
507 DBG("Notifying applications of session daemon state: %d", active
);
509 /* See shm.c for this call implying mmap, shm and futex calls */
510 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
511 if (wait_shm_mmap
== NULL
) {
515 /* Wake waiting process */
516 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
518 /* Apps notified successfully */
526 * Setup the outgoing data buffer for the response (llm) by allocating the
527 * right amount of memory and copying the original information from the lsm
530 * Return total size of the buffer pointed by buf.
532 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
538 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
539 if (cmd_ctx
->llm
== NULL
) {
545 /* Copy common data */
546 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
547 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
549 cmd_ctx
->llm
->data_size
= size
;
550 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
559 * Update the kernel poll set of all channel fd available over all tracing
560 * session. Add the wakeup pipe at the end of the set.
562 static int update_kernel_poll(struct lttng_poll_event
*events
)
565 struct ltt_session
*session
;
566 struct ltt_kernel_channel
*channel
;
568 DBG("Updating kernel poll set");
571 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
572 session_lock(session
);
573 if (session
->kernel_session
== NULL
) {
574 session_unlock(session
);
578 cds_list_for_each_entry(channel
,
579 &session
->kernel_session
->channel_list
.head
, list
) {
580 /* Add channel fd to the kernel poll set */
581 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
583 session_unlock(session
);
586 DBG("Channel fd %d added to kernel set", channel
->fd
);
588 session_unlock(session
);
590 session_unlock_list();
595 session_unlock_list();
600 * Find the channel fd from 'fd' over all tracing session. When found, check
601 * for new channel stream and send those stream fds to the kernel consumer.
603 * Useful for CPU hotplug feature.
605 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
608 struct ltt_session
*session
;
609 struct ltt_kernel_session
*ksess
;
610 struct ltt_kernel_channel
*channel
;
612 DBG("Updating kernel streams for channel fd %d", fd
);
615 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
616 session_lock(session
);
617 if (session
->kernel_session
== NULL
) {
618 session_unlock(session
);
621 ksess
= session
->kernel_session
;
623 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
624 if (channel
->fd
== fd
) {
625 DBG("Channel found, updating kernel streams");
626 ret
= kernel_open_channel_stream(channel
);
632 * Have we already sent fds to the consumer? If yes, it means
633 * that tracing is started so it is safe to send our updated
636 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
637 struct lttng_ht_iter iter
;
638 struct consumer_socket
*socket
;
641 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
642 &iter
.iter
, socket
, node
.node
) {
643 /* Code flow error */
644 assert(socket
->fd
>= 0);
646 pthread_mutex_lock(socket
->lock
);
647 ret
= kernel_consumer_send_channel_stream(socket
,
649 pthread_mutex_unlock(socket
->lock
);
660 session_unlock(session
);
662 session_unlock_list();
666 session_unlock(session
);
667 session_unlock_list();
672 * For each tracing session, update newly registered apps. The session list
673 * lock MUST be acquired before calling this.
675 static void update_ust_app(int app_sock
)
677 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
);
690 * This thread manage event coming from the kernel.
692 * Features supported in this thread:
695 static void *thread_manage_kernel(void *data
)
697 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
698 uint32_t revents
, nb_fd
;
700 struct lttng_poll_event events
;
702 DBG("[thread] Thread manage kernel started");
704 health_register(HEALTH_TYPE_KERNEL
);
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();
718 if (testpoint(thread_manage_kernel_before_loop
)) {
719 goto error_testpoint
;
723 health_code_update();
725 if (update_poll_flag
== 1) {
726 /* Clean events object. We are about to populate it again. */
727 lttng_poll_clean(&events
);
729 ret
= sessiond_set_thread_pollset(&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 */
752 ret
= lttng_poll_wait(&events
, -1);
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();
778 /* Thread quit pipe has been closed. Killing thread. */
779 ret
= sessiond_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;
825 ERR("Health error occurred in %s", __func__
);
826 WARN("Kernel thread died unexpectedly. "
827 "Kernel tracing can continue but CPU hotplug is disabled.");
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");
870 health_register(HEALTH_TYPE_CONSUMER
);
872 health_code_update();
875 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
876 * metadata_sock. Nothing more will be added to this poll set.
878 ret
= sessiond_set_thread_pollset(&events
, 3);
884 * The error socket here is already in a listening state which was done
885 * just before spawning this thread to avoid a race between the consumer
886 * daemon exec trying to connect and the listen() call.
888 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
893 health_code_update();
895 /* Infinite blocking call, waiting for transmission */
899 if (testpoint(thread_manage_consumer
)) {
903 ret
= lttng_poll_wait(&events
, -1);
907 * Restart interrupted system call.
909 if (errno
== EINTR
) {
917 for (i
= 0; i
< nb_fd
; i
++) {
918 /* Fetch once the poll data */
919 revents
= LTTNG_POLL_GETEV(&events
, i
);
920 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
922 health_code_update();
924 /* Thread quit pipe has been closed. Killing thread. */
925 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
931 /* Event on the registration socket */
932 if (pollfd
== consumer_data
->err_sock
) {
933 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
934 ERR("consumer err socket poll error");
940 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
946 * Set the CLOEXEC flag. Return code is useless because either way, the
949 (void) utils_set_fd_cloexec(sock
);
951 health_code_update();
953 DBG2("Receiving code from consumer err_sock");
955 /* Getting status code from kconsumerd */
956 ret
= lttcomm_recv_unix_sock(sock
, &code
,
957 sizeof(enum lttcomm_return_code
));
962 health_code_update();
964 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
965 /* Connect both socket, command and metadata. */
966 consumer_data
->cmd_sock
=
967 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
968 consumer_data
->metadata_sock
.fd
=
969 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
970 if (consumer_data
->cmd_sock
< 0 ||
971 consumer_data
->metadata_sock
.fd
< 0) {
972 PERROR("consumer connect cmd socket");
973 /* On error, signal condition and quit. */
974 signal_consumer_condition(consumer_data
, -1);
977 /* Create metadata socket lock. */
978 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
979 if (consumer_data
->metadata_sock
.lock
== NULL
) {
980 PERROR("zmalloc pthread mutex");
984 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
986 signal_consumer_condition(consumer_data
, 1);
987 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
988 DBG("Consumer metadata socket ready (fd: %d)",
989 consumer_data
->metadata_sock
.fd
);
991 ERR("consumer error when waiting for SOCK_READY : %s",
992 lttcomm_get_readable_code(-code
));
996 /* Remove the consumerd error sock since we've established a connexion */
997 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1002 /* Add new accepted error socket. */
1003 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1008 /* Add metadata socket that is successfully connected. */
1009 ret
= lttng_poll_add(&events
, consumer_data
->metadata_sock
.fd
,
1010 LPOLLIN
| LPOLLRDHUP
);
1015 health_code_update();
1017 /* Infinite blocking call, waiting for transmission */
1020 health_poll_entry();
1021 ret
= lttng_poll_wait(&events
, -1);
1025 * Restart interrupted system call.
1027 if (errno
== EINTR
) {
1035 for (i
= 0; i
< nb_fd
; i
++) {
1036 /* Fetch once the poll data */
1037 revents
= LTTNG_POLL_GETEV(&events
, i
);
1038 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1040 health_code_update();
1042 /* Thread quit pipe has been closed. Killing thread. */
1043 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1049 if (pollfd
== sock
) {
1050 /* Event on the consumerd socket */
1051 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1052 ERR("consumer err socket second poll error");
1055 health_code_update();
1056 /* Wait for any kconsumerd error */
1057 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1058 sizeof(enum lttcomm_return_code
));
1060 ERR("consumer closed the command socket");
1064 ERR("consumer return code : %s",
1065 lttcomm_get_readable_code(-code
));
1068 } else if (pollfd
== consumer_data
->metadata_sock
.fd
) {
1069 /* UST metadata requests */
1070 ret
= ust_consumer_metadata_request(
1071 &consumer_data
->metadata_sock
);
1073 ERR("Handling metadata request");
1078 ERR("Unknown pollfd");
1082 health_code_update();
1087 /* Immediately set the consumerd state to stopped */
1088 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1089 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1090 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1091 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1092 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1094 /* Code flow error... */
1098 if (consumer_data
->err_sock
>= 0) {
1099 ret
= close(consumer_data
->err_sock
);
1104 if (consumer_data
->cmd_sock
>= 0) {
1105 ret
= close(consumer_data
->cmd_sock
);
1110 if (consumer_data
->metadata_sock
.fd
>= 0) {
1111 ret
= close(consumer_data
->metadata_sock
.fd
);
1116 /* Cleanup metadata socket mutex. */
1117 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1118 free(consumer_data
->metadata_sock
.lock
);
1127 unlink(consumer_data
->err_unix_sock_path
);
1128 unlink(consumer_data
->cmd_unix_sock_path
);
1129 consumer_data
->pid
= 0;
1131 lttng_poll_clean(&events
);
1135 ERR("Health error occurred in %s", __func__
);
1137 health_unregister();
1138 DBG("consumer thread cleanup completed");
1144 * This thread manage application communication.
1146 static void *thread_manage_apps(void *data
)
1148 int i
, ret
, pollfd
, err
= -1;
1149 uint32_t revents
, nb_fd
;
1150 struct lttng_poll_event events
;
1152 DBG("[thread] Manage application started");
1154 rcu_register_thread();
1155 rcu_thread_online();
1157 health_register(HEALTH_TYPE_APP_MANAGE
);
1159 if (testpoint(thread_manage_apps
)) {
1160 goto error_testpoint
;
1163 health_code_update();
1165 ret
= sessiond_set_thread_pollset(&events
, 2);
1167 goto error_poll_create
;
1170 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1175 if (testpoint(thread_manage_apps_before_loop
)) {
1179 health_code_update();
1182 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1184 /* Inifinite blocking call, waiting for transmission */
1186 health_poll_entry();
1187 ret
= lttng_poll_wait(&events
, -1);
1191 * Restart interrupted system call.
1193 if (errno
== EINTR
) {
1201 for (i
= 0; i
< nb_fd
; i
++) {
1202 /* Fetch once the poll data */
1203 revents
= LTTNG_POLL_GETEV(&events
, i
);
1204 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1206 health_code_update();
1208 /* Thread quit pipe has been closed. Killing thread. */
1209 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1215 /* Inspect the apps cmd pipe */
1216 if (pollfd
== apps_cmd_pipe
[0]) {
1217 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1218 ERR("Apps command pipe error");
1220 } else if (revents
& LPOLLIN
) {
1225 ret
= read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1226 } while (ret
< 0 && errno
== EINTR
);
1227 if (ret
< 0 || ret
< sizeof(sock
)) {
1228 PERROR("read apps cmd pipe");
1232 health_code_update();
1235 * We only monitor the error events of the socket. This
1236 * thread does not handle any incoming data from UST
1239 ret
= lttng_poll_add(&events
, sock
,
1240 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1245 /* Set socket timeout for both receiving and ending */
1246 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1247 app_socket_timeout
);
1248 (void) lttcomm_setsockopt_snd_timeout(sock
,
1249 app_socket_timeout
);
1251 DBG("Apps with sock %d added to poll set", sock
);
1253 health_code_update();
1259 * At this point, we know that a registered application made
1260 * the event at poll_wait.
1262 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1263 /* Removing from the poll set */
1264 ret
= lttng_poll_del(&events
, pollfd
);
1269 /* Socket closed on remote end. */
1270 ust_app_unregister(pollfd
);
1275 health_code_update();
1281 lttng_poll_clean(&events
);
1284 utils_close_pipe(apps_cmd_pipe
);
1285 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1288 * We don't clean the UST app hash table here since already registered
1289 * applications can still be controlled so let them be until the session
1290 * daemon dies or the applications stop.
1295 ERR("Health error occurred in %s", __func__
);
1297 health_unregister();
1298 DBG("Application communication apps thread cleanup complete");
1299 rcu_thread_offline();
1300 rcu_unregister_thread();
1305 * Send a socket to a thread This is called from the dispatch UST registration
1306 * thread once all sockets are set for the application.
1308 * On success, return 0 else a negative value being the errno message of the
1311 static int send_socket_to_thread(int fd
, int sock
)
1315 /* Sockets MUST be set or else this should not have been called. */
1320 ret
= write(fd
, &sock
, sizeof(sock
));
1321 } while (ret
< 0 && errno
== EINTR
);
1322 if (ret
< 0 || ret
!= sizeof(sock
)) {
1323 PERROR("write apps pipe %d", fd
);
1330 /* All good. Don't send back the write positive ret value. */
1337 * Dispatch request from the registration threads to the application
1338 * communication thread.
1340 static void *thread_dispatch_ust_registration(void *data
)
1343 struct cds_wfq_node
*node
;
1344 struct ust_command
*ust_cmd
= NULL
;
1346 struct ust_app
*app
;
1347 struct cds_list_head head
;
1348 } *wait_node
= NULL
, *tmp_wait_node
;
1350 CDS_LIST_HEAD(wait_queue
);
1352 DBG("[thread] Dispatch UST command started");
1354 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1355 /* Atomically prepare the queue futex */
1356 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1359 struct ust_app
*app
= NULL
;
1362 /* Dequeue command for registration */
1363 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1365 DBG("Woken up but nothing in the UST command queue");
1366 /* Continue thread execution */
1370 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1372 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1373 " gid:%d sock:%d name:%s (version %d.%d)",
1374 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1375 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1376 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1377 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1379 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1380 wait_node
= zmalloc(sizeof(*wait_node
));
1382 PERROR("zmalloc wait_node dispatch");
1386 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1388 /* Create application object if socket is CMD. */
1389 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1391 if (!wait_node
->app
) {
1392 ret
= close(ust_cmd
->sock
);
1394 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1396 lttng_fd_put(1, LTTNG_FD_APPS
);
1402 * Add application to the wait queue so we can set the notify
1403 * socket before putting this object in the global ht.
1405 cds_list_add(&wait_node
->head
, &wait_queue
);
1409 * We have to continue here since we don't have the notify
1410 * socket and the application MUST be added to the hash table
1411 * only at that moment.
1416 * Look for the application in the local wait queue and set the
1417 * notify socket if found.
1419 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1420 &wait_queue
, head
) {
1421 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1422 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1423 cds_list_del(&wait_node
->head
);
1424 app
= wait_node
->app
;
1426 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1435 * @session_lock_list
1437 * Lock the global session list so from the register up to the
1438 * registration done message, no thread can see the application
1439 * and change its state.
1441 session_lock_list();
1445 * Add application to the global hash table. This needs to be
1446 * done before the update to the UST registry can locate the
1451 /* Set app version. This call will print an error if needed. */
1452 (void) ust_app_version(app
);
1454 /* Send notify socket through the notify pipe. */
1455 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1459 session_unlock_list();
1460 /* No notify thread, stop the UST tracing. */
1465 * Update newly registered application with the tracing
1466 * registry info already enabled information.
1468 update_ust_app(app
->sock
);
1471 * Don't care about return value. Let the manage apps threads
1472 * handle app unregistration upon socket close.
1474 (void) ust_app_register_done(app
->sock
);
1477 * Even if the application socket has been closed, send the app
1478 * to the thread and unregistration will take place at that
1481 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1484 session_unlock_list();
1485 /* No apps. thread, stop the UST tracing. */
1490 session_unlock_list();
1492 /* Application manager threads are not available. */
1493 ret
= close(ust_cmd
->sock
);
1495 PERROR("close ust_cmd sock");
1497 lttng_fd_put(1, LTTNG_FD_APPS
);
1499 } while (node
!= NULL
);
1501 /* Futex wait on queue. Blocking call on futex() */
1502 futex_nto1_wait(&ust_cmd_queue
.futex
);
1506 /* Clean up wait queue. */
1507 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1508 &wait_queue
, head
) {
1509 cds_list_del(&wait_node
->head
);
1513 DBG("Dispatch thread dying");
1518 * This thread manage application registration.
1520 static void *thread_registration_apps(void *data
)
1522 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1523 uint32_t revents
, nb_fd
;
1524 struct lttng_poll_event events
;
1526 * Get allocated in this thread, enqueued to a global queue, dequeued and
1527 * freed in the manage apps thread.
1529 struct ust_command
*ust_cmd
= NULL
;
1531 DBG("[thread] Manage application registration started");
1533 health_register(HEALTH_TYPE_APP_REG
);
1535 if (testpoint(thread_registration_apps
)) {
1536 goto error_testpoint
;
1539 ret
= lttcomm_listen_unix_sock(apps_sock
);
1545 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1546 * more will be added to this poll set.
1548 ret
= sessiond_set_thread_pollset(&events
, 2);
1550 goto error_create_poll
;
1553 /* Add the application registration socket */
1554 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1556 goto error_poll_add
;
1559 /* Notify all applications to register */
1560 ret
= notify_ust_apps(1);
1562 ERR("Failed to notify applications or create the wait shared memory.\n"
1563 "Execution continues but there might be problem for already\n"
1564 "running applications that wishes to register.");
1568 DBG("Accepting application registration");
1570 /* Inifinite blocking call, waiting for transmission */
1572 health_poll_entry();
1573 ret
= lttng_poll_wait(&events
, -1);
1577 * Restart interrupted system call.
1579 if (errno
== EINTR
) {
1587 for (i
= 0; i
< nb_fd
; i
++) {
1588 health_code_update();
1590 /* Fetch once the poll data */
1591 revents
= LTTNG_POLL_GETEV(&events
, i
);
1592 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1594 /* Thread quit pipe has been closed. Killing thread. */
1595 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1601 /* Event on the registration socket */
1602 if (pollfd
== apps_sock
) {
1603 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1604 ERR("Register apps socket poll error");
1606 } else if (revents
& LPOLLIN
) {
1607 sock
= lttcomm_accept_unix_sock(apps_sock
);
1613 * Set the CLOEXEC flag. Return code is useless because
1614 * either way, the show must go on.
1616 (void) utils_set_fd_cloexec(sock
);
1618 /* Create UST registration command for enqueuing */
1619 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1620 if (ust_cmd
== NULL
) {
1621 PERROR("ust command zmalloc");
1626 * Using message-based transmissions to ensure we don't
1627 * have to deal with partially received messages.
1629 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1631 ERR("Exhausted file descriptors allowed for applications.");
1641 health_code_update();
1642 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1645 /* Close socket of the application. */
1650 lttng_fd_put(LTTNG_FD_APPS
, 1);
1654 health_code_update();
1656 ust_cmd
->sock
= sock
;
1659 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1660 " gid:%d sock:%d name:%s (version %d.%d)",
1661 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1662 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1663 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1664 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1667 * Lock free enqueue the registration request. The red pill
1668 * has been taken! This apps will be part of the *system*.
1670 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
1673 * Wake the registration queue futex. Implicit memory
1674 * barrier with the exchange in cds_wfq_enqueue.
1676 futex_nto1_wake(&ust_cmd_queue
.futex
);
1686 ERR("Health error occurred in %s", __func__
);
1689 /* Notify that the registration thread is gone */
1692 if (apps_sock
>= 0) {
1693 ret
= close(apps_sock
);
1703 lttng_fd_put(LTTNG_FD_APPS
, 1);
1705 unlink(apps_unix_sock_path
);
1708 lttng_poll_clean(&events
);
1712 DBG("UST Registration thread cleanup complete");
1713 health_unregister();
1719 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1720 * exec or it will fails.
1722 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
1725 struct timespec timeout
;
1727 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1728 consumer_data
->consumer_thread_is_ready
= 0;
1730 /* Setup pthread condition */
1731 ret
= pthread_condattr_init(&consumer_data
->condattr
);
1734 PERROR("pthread_condattr_init consumer data");
1739 * Set the monotonic clock in order to make sure we DO NOT jump in time
1740 * between the clock_gettime() call and the timedwait call. See bug #324
1741 * for a more details and how we noticed it.
1743 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
1746 PERROR("pthread_condattr_setclock consumer data");
1750 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
1753 PERROR("pthread_cond_init consumer data");
1757 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
1760 PERROR("pthread_create consumer");
1765 /* We are about to wait on a pthread condition */
1766 pthread_mutex_lock(&consumer_data
->cond_mutex
);
1768 /* Get time for sem_timedwait absolute timeout */
1769 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
1771 * Set the timeout for the condition timed wait even if the clock gettime
1772 * call fails since we might loop on that call and we want to avoid to
1773 * increment the timeout too many times.
1775 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
1778 * The following loop COULD be skipped in some conditions so this is why we
1779 * set ret to 0 in order to make sure at least one round of the loop is
1785 * Loop until the condition is reached or when a timeout is reached. Note
1786 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1787 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1788 * possible. This loop does not take any chances and works with both of
1791 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
1792 if (clock_ret
< 0) {
1793 PERROR("clock_gettime spawn consumer");
1794 /* Infinite wait for the consumerd thread to be ready */
1795 ret
= pthread_cond_wait(&consumer_data
->cond
,
1796 &consumer_data
->cond_mutex
);
1798 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
1799 &consumer_data
->cond_mutex
, &timeout
);
1803 /* Release the pthread condition */
1804 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
1808 if (ret
== ETIMEDOUT
) {
1810 * Call has timed out so we kill the kconsumerd_thread and return
1813 ERR("Condition timed out. The consumer thread was never ready."
1815 ret
= pthread_cancel(consumer_data
->thread
);
1817 PERROR("pthread_cancel consumer thread");
1820 PERROR("pthread_cond_wait failed consumer thread");
1825 pthread_mutex_lock(&consumer_data
->pid_mutex
);
1826 if (consumer_data
->pid
== 0) {
1827 ERR("Consumerd did not start");
1828 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1831 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
1840 * Join consumer thread
1842 static int join_consumer_thread(struct consumer_data
*consumer_data
)
1846 /* Consumer pid must be a real one. */
1847 if (consumer_data
->pid
> 0) {
1849 ret
= kill(consumer_data
->pid
, SIGTERM
);
1851 ERR("Error killing consumer daemon");
1854 return pthread_join(consumer_data
->thread
, &status
);
1861 * Fork and exec a consumer daemon (consumerd).
1863 * Return pid if successful else -1.
1865 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
1869 const char *consumer_to_use
;
1870 const char *verbosity
;
1873 DBG("Spawning consumerd");
1880 if (opt_verbose_consumer
) {
1881 verbosity
= "--verbose";
1883 verbosity
= "--quiet";
1885 switch (consumer_data
->type
) {
1886 case LTTNG_CONSUMER_KERNEL
:
1888 * Find out which consumerd to execute. We will first try the
1889 * 64-bit path, then the sessiond's installation directory, and
1890 * fallback on the 32-bit one,
1892 DBG3("Looking for a kernel consumer at these locations:");
1893 DBG3(" 1) %s", consumerd64_bin
);
1894 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
1895 DBG3(" 3) %s", consumerd32_bin
);
1896 if (stat(consumerd64_bin
, &st
) == 0) {
1897 DBG3("Found location #1");
1898 consumer_to_use
= consumerd64_bin
;
1899 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
1900 DBG3("Found location #2");
1901 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
1902 } else if (stat(consumerd32_bin
, &st
) == 0) {
1903 DBG3("Found location #3");
1904 consumer_to_use
= consumerd32_bin
;
1906 DBG("Could not find any valid consumerd executable");
1909 DBG("Using kernel consumer at: %s", consumer_to_use
);
1910 execl(consumer_to_use
,
1911 "lttng-consumerd", verbosity
, "-k",
1912 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1913 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1916 case LTTNG_CONSUMER64_UST
:
1918 char *tmpnew
= NULL
;
1920 if (consumerd64_libdir
[0] != '\0') {
1924 tmp
= getenv("LD_LIBRARY_PATH");
1928 tmplen
= strlen("LD_LIBRARY_PATH=")
1929 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
1930 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1935 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1936 strcat(tmpnew
, consumerd64_libdir
);
1937 if (tmp
[0] != '\0') {
1938 strcat(tmpnew
, ":");
1939 strcat(tmpnew
, tmp
);
1941 ret
= putenv(tmpnew
);
1948 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
1949 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
1950 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1951 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1953 if (consumerd64_libdir
[0] != '\0') {
1961 case LTTNG_CONSUMER32_UST
:
1963 char *tmpnew
= NULL
;
1965 if (consumerd32_libdir
[0] != '\0') {
1969 tmp
= getenv("LD_LIBRARY_PATH");
1973 tmplen
= strlen("LD_LIBRARY_PATH=")
1974 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
1975 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
1980 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
1981 strcat(tmpnew
, consumerd32_libdir
);
1982 if (tmp
[0] != '\0') {
1983 strcat(tmpnew
, ":");
1984 strcat(tmpnew
, tmp
);
1986 ret
= putenv(tmpnew
);
1993 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
1994 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
1995 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
1996 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
1998 if (consumerd32_libdir
[0] != '\0') {
2007 PERROR("unknown consumer type");
2011 PERROR("kernel start consumer exec");
2014 } else if (pid
> 0) {
2017 PERROR("start consumer fork");
2025 * Spawn the consumerd daemon and session daemon thread.
2027 static int start_consumerd(struct consumer_data
*consumer_data
)
2032 * Set the listen() state on the socket since there is a possible race
2033 * between the exec() of the consumer daemon and this call if place in the
2034 * consumer thread. See bug #366 for more details.
2036 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2041 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2042 if (consumer_data
->pid
!= 0) {
2043 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2047 ret
= spawn_consumerd(consumer_data
);
2049 ERR("Spawning consumerd failed");
2050 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2054 /* Setting up the consumer_data pid */
2055 consumer_data
->pid
= ret
;
2056 DBG2("Consumer pid %d", consumer_data
->pid
);
2057 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2059 DBG2("Spawning consumer control thread");
2060 ret
= spawn_consumer_thread(consumer_data
);
2062 ERR("Fatal error spawning consumer control thread");
2070 /* Cleanup already created sockets on error. */
2071 if (consumer_data
->err_sock
>= 0) {
2074 err
= close(consumer_data
->err_sock
);
2076 PERROR("close consumer data error socket");
2083 * Compute health status of each consumer. If one of them is zero (bad
2084 * state), we return 0.
2086 static int check_consumer_health(void)
2090 ret
= health_check_state(HEALTH_TYPE_CONSUMER
);
2092 DBG3("Health consumer check %d", ret
);
2098 * Setup necessary data for kernel tracer action.
2100 static int init_kernel_tracer(void)
2104 /* Modprobe lttng kernel modules */
2105 ret
= modprobe_lttng_control();
2110 /* Open debugfs lttng */
2111 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2112 if (kernel_tracer_fd
< 0) {
2113 DBG("Failed to open %s", module_proc_lttng
);
2118 /* Validate kernel version */
2119 ret
= kernel_validate_version(kernel_tracer_fd
);
2124 ret
= modprobe_lttng_data();
2129 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2133 modprobe_remove_lttng_control();
2134 ret
= close(kernel_tracer_fd
);
2138 kernel_tracer_fd
= -1;
2139 return LTTNG_ERR_KERN_VERSION
;
2142 ret
= close(kernel_tracer_fd
);
2148 modprobe_remove_lttng_control();
2151 WARN("No kernel tracer available");
2152 kernel_tracer_fd
= -1;
2154 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2156 return LTTNG_ERR_KERN_NA
;
2162 * Copy consumer output from the tracing session to the domain session. The
2163 * function also applies the right modification on a per domain basis for the
2164 * trace files destination directory.
2166 * Should *NOT* be called with RCU read-side lock held.
2168 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2171 const char *dir_name
;
2172 struct consumer_output
*consumer
;
2175 assert(session
->consumer
);
2178 case LTTNG_DOMAIN_KERNEL
:
2179 DBG3("Copying tracing session consumer output in kernel session");
2181 * XXX: We should audit the session creation and what this function
2182 * does "extra" in order to avoid a destroy since this function is used
2183 * in the domain session creation (kernel and ust) only. Same for UST
2186 if (session
->kernel_session
->consumer
) {
2187 consumer_destroy_output(session
->kernel_session
->consumer
);
2189 session
->kernel_session
->consumer
=
2190 consumer_copy_output(session
->consumer
);
2191 /* Ease our life a bit for the next part */
2192 consumer
= session
->kernel_session
->consumer
;
2193 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2195 case LTTNG_DOMAIN_UST
:
2196 DBG3("Copying tracing session consumer output in UST session");
2197 if (session
->ust_session
->consumer
) {
2198 consumer_destroy_output(session
->ust_session
->consumer
);
2200 session
->ust_session
->consumer
=
2201 consumer_copy_output(session
->consumer
);
2202 /* Ease our life a bit for the next part */
2203 consumer
= session
->ust_session
->consumer
;
2204 dir_name
= DEFAULT_UST_TRACE_DIR
;
2207 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2211 /* Append correct directory to subdir */
2212 strncat(consumer
->subdir
, dir_name
,
2213 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2214 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2223 * Create an UST session and add it to the session ust list.
2225 * Should *NOT* be called with RCU read-side lock held.
2227 static int create_ust_session(struct ltt_session
*session
,
2228 struct lttng_domain
*domain
)
2231 struct ltt_ust_session
*lus
= NULL
;
2235 assert(session
->consumer
);
2237 switch (domain
->type
) {
2238 case LTTNG_DOMAIN_UST
:
2241 ERR("Unknown UST domain on create session %d", domain
->type
);
2242 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2246 DBG("Creating UST session");
2248 lus
= trace_ust_create_session(session
->id
);
2250 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2254 lus
->uid
= session
->uid
;
2255 lus
->gid
= session
->gid
;
2256 session
->ust_session
= lus
;
2258 /* Copy session output to the newly created UST session */
2259 ret
= copy_session_consumer(domain
->type
, session
);
2260 if (ret
!= LTTNG_OK
) {
2268 session
->ust_session
= NULL
;
2273 * Create a kernel tracer session then create the default channel.
2275 static int create_kernel_session(struct ltt_session
*session
)
2279 DBG("Creating kernel session");
2281 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2283 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2287 /* Code flow safety */
2288 assert(session
->kernel_session
);
2290 /* Copy session output to the newly created Kernel session */
2291 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2292 if (ret
!= LTTNG_OK
) {
2296 /* Create directory(ies) on local filesystem. */
2297 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2298 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2299 ret
= run_as_mkdir_recursive(
2300 session
->kernel_session
->consumer
->dst
.trace_path
,
2301 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2303 if (ret
!= -EEXIST
) {
2304 ERR("Trace directory creation error");
2310 session
->kernel_session
->uid
= session
->uid
;
2311 session
->kernel_session
->gid
= session
->gid
;
2316 trace_kernel_destroy_session(session
->kernel_session
);
2317 session
->kernel_session
= NULL
;
2322 * Count number of session permitted by uid/gid.
2324 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2327 struct ltt_session
*session
;
2329 DBG("Counting number of available session for UID %d GID %d",
2331 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2333 * Only list the sessions the user can control.
2335 if (!session_access_ok(session
, uid
, gid
)) {
2344 * Process the command requested by the lttng client within the command
2345 * context structure. This function make sure that the return structure (llm)
2346 * is set and ready for transmission before returning.
2348 * Return any error encountered or 0 for success.
2350 * "sock" is only used for special-case var. len data.
2352 * Should *NOT* be called with RCU read-side lock held.
2354 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2358 int need_tracing_session
= 1;
2361 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2365 switch (cmd_ctx
->lsm
->cmd_type
) {
2366 case LTTNG_CREATE_SESSION
:
2367 case LTTNG_DESTROY_SESSION
:
2368 case LTTNG_LIST_SESSIONS
:
2369 case LTTNG_LIST_DOMAINS
:
2370 case LTTNG_START_TRACE
:
2371 case LTTNG_STOP_TRACE
:
2372 case LTTNG_DATA_PENDING
:
2379 if (opt_no_kernel
&& need_domain
2380 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2382 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2384 ret
= LTTNG_ERR_KERN_NA
;
2389 /* Deny register consumer if we already have a spawned consumer. */
2390 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2391 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2392 if (kconsumer_data
.pid
> 0) {
2393 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2394 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2397 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2401 * Check for command that don't needs to allocate a returned payload. We do
2402 * this here so we don't have to make the call for no payload at each
2405 switch(cmd_ctx
->lsm
->cmd_type
) {
2406 case LTTNG_LIST_SESSIONS
:
2407 case LTTNG_LIST_TRACEPOINTS
:
2408 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2409 case LTTNG_LIST_DOMAINS
:
2410 case LTTNG_LIST_CHANNELS
:
2411 case LTTNG_LIST_EVENTS
:
2414 /* Setup lttng message with no payload */
2415 ret
= setup_lttng_msg(cmd_ctx
, 0);
2417 /* This label does not try to unlock the session */
2418 goto init_setup_error
;
2422 /* Commands that DO NOT need a session. */
2423 switch (cmd_ctx
->lsm
->cmd_type
) {
2424 case LTTNG_CREATE_SESSION
:
2425 case LTTNG_CALIBRATE
:
2426 case LTTNG_LIST_SESSIONS
:
2427 case LTTNG_LIST_TRACEPOINTS
:
2428 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2429 need_tracing_session
= 0;
2432 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2434 * We keep the session list lock across _all_ commands
2435 * for now, because the per-session lock does not
2436 * handle teardown properly.
2438 session_lock_list();
2439 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2440 if (cmd_ctx
->session
== NULL
) {
2441 if (cmd_ctx
->lsm
->session
.name
!= NULL
) {
2442 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2444 /* If no session name specified */
2445 ret
= LTTNG_ERR_SELECT_SESS
;
2449 /* Acquire lock for the session */
2450 session_lock(cmd_ctx
->session
);
2460 * Check domain type for specific "pre-action".
2462 switch (cmd_ctx
->lsm
->domain
.type
) {
2463 case LTTNG_DOMAIN_KERNEL
:
2465 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2469 /* Kernel tracer check */
2470 if (kernel_tracer_fd
== -1) {
2471 /* Basically, load kernel tracer modules */
2472 ret
= init_kernel_tracer();
2478 /* Consumer is in an ERROR state. Report back to client */
2479 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2480 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2484 /* Need a session for kernel command */
2485 if (need_tracing_session
) {
2486 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2487 ret
= create_kernel_session(cmd_ctx
->session
);
2489 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2494 /* Start the kernel consumer daemon */
2495 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2496 if (kconsumer_data
.pid
== 0 &&
2497 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2498 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2499 ret
= start_consumerd(&kconsumer_data
);
2501 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2504 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2506 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2510 * The consumer was just spawned so we need to add the socket to
2511 * the consumer output of the session if exist.
2513 ret
= consumer_create_socket(&kconsumer_data
,
2514 cmd_ctx
->session
->kernel_session
->consumer
);
2521 case LTTNG_DOMAIN_UST
:
2523 /* Consumer is in an ERROR state. Report back to client */
2524 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2525 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2529 if (need_tracing_session
) {
2530 /* Create UST session if none exist. */
2531 if (cmd_ctx
->session
->ust_session
== NULL
) {
2532 ret
= create_ust_session(cmd_ctx
->session
,
2533 &cmd_ctx
->lsm
->domain
);
2534 if (ret
!= LTTNG_OK
) {
2539 /* Start the UST consumer daemons */
2541 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2542 if (consumerd64_bin
[0] != '\0' &&
2543 ustconsumer64_data
.pid
== 0 &&
2544 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2545 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2546 ret
= start_consumerd(&ustconsumer64_data
);
2548 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2549 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2553 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2554 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2556 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2560 * Setup socket for consumer 64 bit. No need for atomic access
2561 * since it was set above and can ONLY be set in this thread.
2563 ret
= consumer_create_socket(&ustconsumer64_data
,
2564 cmd_ctx
->session
->ust_session
->consumer
);
2570 if (consumerd32_bin
[0] != '\0' &&
2571 ustconsumer32_data
.pid
== 0 &&
2572 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2573 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2574 ret
= start_consumerd(&ustconsumer32_data
);
2576 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2577 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2581 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2582 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2584 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2588 * Setup socket for consumer 64 bit. No need for atomic access
2589 * since it was set above and can ONLY be set in this thread.
2591 ret
= consumer_create_socket(&ustconsumer32_data
,
2592 cmd_ctx
->session
->ust_session
->consumer
);
2604 /* Validate consumer daemon state when start/stop trace command */
2605 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2606 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2607 switch (cmd_ctx
->lsm
->domain
.type
) {
2608 case LTTNG_DOMAIN_UST
:
2609 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2610 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2614 case LTTNG_DOMAIN_KERNEL
:
2615 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2616 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2624 * Check that the UID or GID match that of the tracing session.
2625 * The root user can interact with all sessions.
2627 if (need_tracing_session
) {
2628 if (!session_access_ok(cmd_ctx
->session
,
2629 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2630 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2631 ret
= LTTNG_ERR_EPERM
;
2637 * Send relayd information to consumer as soon as we have a domain and a
2640 if (cmd_ctx
->session
&& need_domain
) {
2642 * Setup relayd if not done yet. If the relayd information was already
2643 * sent to the consumer, this call will gracefully return.
2645 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2646 if (ret
!= LTTNG_OK
) {
2651 /* Process by command type */
2652 switch (cmd_ctx
->lsm
->cmd_type
) {
2653 case LTTNG_ADD_CONTEXT
:
2655 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2656 cmd_ctx
->lsm
->u
.context
.channel_name
,
2657 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
2660 case LTTNG_DISABLE_CHANNEL
:
2662 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2663 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2666 case LTTNG_DISABLE_EVENT
:
2668 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2669 cmd_ctx
->lsm
->u
.disable
.channel_name
,
2670 cmd_ctx
->lsm
->u
.disable
.name
);
2673 case LTTNG_DISABLE_ALL_EVENT
:
2675 DBG("Disabling all events");
2677 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
2678 cmd_ctx
->lsm
->u
.disable
.channel_name
);
2681 case LTTNG_ENABLE_CHANNEL
:
2683 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2684 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
2687 case LTTNG_ENABLE_EVENT
:
2689 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2690 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2691 &cmd_ctx
->lsm
->u
.enable
.event
, NULL
, kernel_poll_pipe
[1]);
2694 case LTTNG_ENABLE_ALL_EVENT
:
2696 DBG("Enabling all events");
2698 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
2699 cmd_ctx
->lsm
->u
.enable
.channel_name
,
2700 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
2703 case LTTNG_LIST_TRACEPOINTS
:
2705 struct lttng_event
*events
;
2708 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
2709 if (nb_events
< 0) {
2710 /* Return value is a negative lttng_error_code. */
2716 * Setup lttng message with payload size set to the event list size in
2717 * bytes and then copy list into the llm payload.
2719 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
2725 /* Copy event list into message payload */
2726 memcpy(cmd_ctx
->llm
->payload
, events
,
2727 sizeof(struct lttng_event
) * nb_events
);
2734 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2736 struct lttng_event_field
*fields
;
2739 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
2741 if (nb_fields
< 0) {
2742 /* Return value is a negative lttng_error_code. */
2748 * Setup lttng message with payload size set to the event list size in
2749 * bytes and then copy list into the llm payload.
2751 ret
= setup_lttng_msg(cmd_ctx
,
2752 sizeof(struct lttng_event_field
) * nb_fields
);
2758 /* Copy event list into message payload */
2759 memcpy(cmd_ctx
->llm
->payload
, fields
,
2760 sizeof(struct lttng_event_field
) * nb_fields
);
2767 case LTTNG_SET_CONSUMER_URI
:
2770 struct lttng_uri
*uris
;
2772 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2773 len
= nb_uri
* sizeof(struct lttng_uri
);
2776 ret
= LTTNG_ERR_INVALID
;
2780 uris
= zmalloc(len
);
2782 ret
= LTTNG_ERR_FATAL
;
2786 /* Receive variable len data */
2787 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
2788 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2790 DBG("No URIs received from client... continuing");
2792 ret
= LTTNG_ERR_SESSION_FAIL
;
2797 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2799 if (ret
!= LTTNG_OK
) {
2805 * XXX: 0 means that this URI should be applied on the session. Should
2806 * be a DOMAIN enuam.
2808 if (cmd_ctx
->lsm
->domain
.type
== 0) {
2809 /* Add the URI for the UST session if a consumer is present. */
2810 if (cmd_ctx
->session
->ust_session
&&
2811 cmd_ctx
->session
->ust_session
->consumer
) {
2812 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
2814 } else if (cmd_ctx
->session
->kernel_session
&&
2815 cmd_ctx
->session
->kernel_session
->consumer
) {
2816 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
2817 cmd_ctx
->session
, nb_uri
, uris
);
2825 case LTTNG_START_TRACE
:
2827 ret
= cmd_start_trace(cmd_ctx
->session
);
2830 case LTTNG_STOP_TRACE
:
2832 ret
= cmd_stop_trace(cmd_ctx
->session
);
2835 case LTTNG_CREATE_SESSION
:
2838 struct lttng_uri
*uris
= NULL
;
2840 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
2841 len
= nb_uri
* sizeof(struct lttng_uri
);
2844 uris
= zmalloc(len
);
2846 ret
= LTTNG_ERR_FATAL
;
2850 /* Receive variable len data */
2851 DBG("Waiting for %zu URIs from client ...", nb_uri
);
2852 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
2854 DBG("No URIs received from client... continuing");
2856 ret
= LTTNG_ERR_SESSION_FAIL
;
2861 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
2862 DBG("Creating session with ONE network URI is a bad call");
2863 ret
= LTTNG_ERR_SESSION_FAIL
;
2869 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
2876 case LTTNG_DESTROY_SESSION
:
2878 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
2880 /* Set session to NULL so we do not unlock it after free. */
2881 cmd_ctx
->session
= NULL
;
2884 case LTTNG_LIST_DOMAINS
:
2887 struct lttng_domain
*domains
;
2889 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
2891 /* Return value is a negative lttng_error_code. */
2896 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
2902 /* Copy event list into message payload */
2903 memcpy(cmd_ctx
->llm
->payload
, domains
,
2904 nb_dom
* sizeof(struct lttng_domain
));
2911 case LTTNG_LIST_CHANNELS
:
2914 struct lttng_channel
*channels
;
2916 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
2917 cmd_ctx
->session
, &channels
);
2919 /* Return value is a negative lttng_error_code. */
2924 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
2930 /* Copy event list into message payload */
2931 memcpy(cmd_ctx
->llm
->payload
, channels
,
2932 nb_chan
* sizeof(struct lttng_channel
));
2939 case LTTNG_LIST_EVENTS
:
2942 struct lttng_event
*events
= NULL
;
2944 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
2945 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
2947 /* Return value is a negative lttng_error_code. */
2952 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
2958 /* Copy event list into message payload */
2959 memcpy(cmd_ctx
->llm
->payload
, events
,
2960 nb_event
* sizeof(struct lttng_event
));
2967 case LTTNG_LIST_SESSIONS
:
2969 unsigned int nr_sessions
;
2971 session_lock_list();
2972 nr_sessions
= lttng_sessions_count(
2973 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2974 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2976 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
2978 session_unlock_list();
2982 /* Filled the session array */
2983 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
2984 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2985 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
2987 session_unlock_list();
2992 case LTTNG_CALIBRATE
:
2994 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
2995 &cmd_ctx
->lsm
->u
.calibrate
);
2998 case LTTNG_REGISTER_CONSUMER
:
3000 struct consumer_data
*cdata
;
3002 switch (cmd_ctx
->lsm
->domain
.type
) {
3003 case LTTNG_DOMAIN_KERNEL
:
3004 cdata
= &kconsumer_data
;
3007 ret
= LTTNG_ERR_UND
;
3011 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3012 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3015 case LTTNG_ENABLE_EVENT_WITH_FILTER
:
3017 struct lttng_filter_bytecode
*bytecode
;
3019 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3020 ret
= LTTNG_ERR_FILTER_INVAL
;
3023 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
== 0) {
3024 ret
= LTTNG_ERR_FILTER_INVAL
;
3027 bytecode
= zmalloc(cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3029 ret
= LTTNG_ERR_FILTER_NOMEM
;
3032 /* Receive var. len. data */
3033 DBG("Receiving var len data from client ...");
3034 ret
= lttcomm_recv_unix_sock(sock
, bytecode
,
3035 cmd_ctx
->lsm
->u
.enable
.bytecode_len
);
3037 DBG("Nothing recv() from client var len data... continuing");
3039 ret
= LTTNG_ERR_FILTER_INVAL
;
3043 if (bytecode
->len
+ sizeof(*bytecode
)
3044 != cmd_ctx
->lsm
->u
.enable
.bytecode_len
) {
3046 ret
= LTTNG_ERR_FILTER_INVAL
;
3050 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3051 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3052 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, kernel_poll_pipe
[1]);
3055 case LTTNG_DATA_PENDING
:
3057 ret
= cmd_data_pending(cmd_ctx
->session
);
3061 ret
= LTTNG_ERR_UND
;
3066 if (cmd_ctx
->llm
== NULL
) {
3067 DBG("Missing llm structure. Allocating one.");
3068 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3072 /* Set return code */
3073 cmd_ctx
->llm
->ret_code
= ret
;
3075 if (cmd_ctx
->session
) {
3076 session_unlock(cmd_ctx
->session
);
3078 if (need_tracing_session
) {
3079 session_unlock_list();
3086 * Thread managing health check socket.
3088 static void *thread_manage_health(void *data
)
3090 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3091 uint32_t revents
, nb_fd
;
3092 struct lttng_poll_event events
;
3093 struct lttcomm_health_msg msg
;
3094 struct lttcomm_health_data reply
;
3096 DBG("[thread] Manage health check started");
3098 rcu_register_thread();
3100 /* We might hit an error path before this is set once. */
3101 memset(&events
, 0, sizeof(events
));
3104 /* Create unix socket */
3105 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3107 ERR("Unable to create health check Unix socket");
3113 * Set the CLOEXEC flag. Return code is useless because either way, the
3116 (void) utils_set_fd_cloexec(sock
);
3118 ret
= lttcomm_listen_unix_sock(sock
);
3124 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3125 * more will be added to this poll set.
3127 ret
= sessiond_set_thread_pollset(&events
, 2);
3132 /* Add the application registration socket */
3133 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3139 DBG("Health check ready");
3141 /* Inifinite blocking call, waiting for transmission */
3143 ret
= lttng_poll_wait(&events
, -1);
3146 * Restart interrupted system call.
3148 if (errno
== EINTR
) {
3156 for (i
= 0; i
< nb_fd
; i
++) {
3157 /* Fetch once the poll data */
3158 revents
= LTTNG_POLL_GETEV(&events
, i
);
3159 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3161 /* Thread quit pipe has been closed. Killing thread. */
3162 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3168 /* Event on the registration socket */
3169 if (pollfd
== sock
) {
3170 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3171 ERR("Health socket poll error");
3177 new_sock
= lttcomm_accept_unix_sock(sock
);
3183 * Set the CLOEXEC flag. Return code is useless because either way, the
3186 (void) utils_set_fd_cloexec(new_sock
);
3188 DBG("Receiving data from client for health...");
3189 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3191 DBG("Nothing recv() from client... continuing");
3192 ret
= close(new_sock
);
3200 rcu_thread_online();
3202 switch (msg
.component
) {
3203 case LTTNG_HEALTH_CMD
:
3204 reply
.ret_code
= health_check_state(HEALTH_TYPE_CMD
);
3206 case LTTNG_HEALTH_APP_MANAGE
:
3207 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_MANAGE
);
3209 case LTTNG_HEALTH_APP_REG
:
3210 reply
.ret_code
= health_check_state(HEALTH_TYPE_APP_REG
);
3212 case LTTNG_HEALTH_KERNEL
:
3213 reply
.ret_code
= health_check_state(HEALTH_TYPE_KERNEL
);
3215 case LTTNG_HEALTH_CONSUMER
:
3216 reply
.ret_code
= check_consumer_health();
3218 case LTTNG_HEALTH_ALL
:
3220 health_check_state(HEALTH_TYPE_APP_MANAGE
) &&
3221 health_check_state(HEALTH_TYPE_APP_REG
) &&
3222 health_check_state(HEALTH_TYPE_CMD
) &&
3223 health_check_state(HEALTH_TYPE_KERNEL
) &&
3224 check_consumer_health();
3227 reply
.ret_code
= LTTNG_ERR_UND
;
3232 * Flip ret value since 0 is a success and 1 indicates a bad health for
3233 * the client where in the sessiond it is the opposite. Again, this is
3234 * just to make things easier for us poor developer which enjoy a lot
3237 if (reply
.ret_code
== 0 || reply
.ret_code
== 1) {
3238 reply
.ret_code
= !reply
.ret_code
;
3241 DBG2("Health check return value %d", reply
.ret_code
);
3243 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3245 ERR("Failed to send health data back to client");
3248 /* End of transmission */
3249 ret
= close(new_sock
);
3259 ERR("Health error occurred in %s", __func__
);
3261 DBG("Health check thread dying");
3262 unlink(health_unix_sock_path
);
3269 if (new_sock
>= 0) {
3270 ret
= close(new_sock
);
3276 lttng_poll_clean(&events
);
3278 rcu_unregister_thread();
3283 * This thread manage all clients request using the unix client socket for
3286 static void *thread_manage_clients(void *data
)
3288 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3290 uint32_t revents
, nb_fd
;
3291 struct command_ctx
*cmd_ctx
= NULL
;
3292 struct lttng_poll_event events
;
3294 DBG("[thread] Manage client started");
3296 rcu_register_thread();
3298 health_register(HEALTH_TYPE_CMD
);
3300 if (testpoint(thread_manage_clients
)) {
3301 goto error_testpoint
;
3304 health_code_update();
3306 ret
= lttcomm_listen_unix_sock(client_sock
);
3312 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3313 * more will be added to this poll set.
3315 ret
= sessiond_set_thread_pollset(&events
, 2);
3317 goto error_create_poll
;
3320 /* Add the application registration socket */
3321 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3327 * Notify parent pid that we are ready to accept command for client side.
3329 if (opt_sig_parent
) {
3330 kill(ppid
, SIGUSR1
);
3333 if (testpoint(thread_manage_clients_before_loop
)) {
3337 health_code_update();
3340 DBG("Accepting client command ...");
3342 /* Inifinite blocking call, waiting for transmission */
3344 health_poll_entry();
3345 ret
= lttng_poll_wait(&events
, -1);
3349 * Restart interrupted system call.
3351 if (errno
== EINTR
) {
3359 for (i
= 0; i
< nb_fd
; i
++) {
3360 /* Fetch once the poll data */
3361 revents
= LTTNG_POLL_GETEV(&events
, i
);
3362 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3364 health_code_update();
3366 /* Thread quit pipe has been closed. Killing thread. */
3367 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3373 /* Event on the registration socket */
3374 if (pollfd
== client_sock
) {
3375 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3376 ERR("Client socket poll error");
3382 DBG("Wait for client response");
3384 health_code_update();
3386 sock
= lttcomm_accept_unix_sock(client_sock
);
3392 * Set the CLOEXEC flag. Return code is useless because either way, the
3395 (void) utils_set_fd_cloexec(sock
);
3397 /* Set socket option for credentials retrieval */
3398 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3403 /* Allocate context command to process the client request */
3404 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3405 if (cmd_ctx
== NULL
) {
3406 PERROR("zmalloc cmd_ctx");
3410 /* Allocate data buffer for reception */
3411 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3412 if (cmd_ctx
->lsm
== NULL
) {
3413 PERROR("zmalloc cmd_ctx->lsm");
3417 cmd_ctx
->llm
= NULL
;
3418 cmd_ctx
->session
= NULL
;
3420 health_code_update();
3423 * Data is received from the lttng client. The struct
3424 * lttcomm_session_msg (lsm) contains the command and data request of
3427 DBG("Receiving data from client ...");
3428 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3429 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3431 DBG("Nothing recv() from client... continuing");
3437 clean_command_ctx(&cmd_ctx
);
3441 health_code_update();
3443 // TODO: Validate cmd_ctx including sanity check for
3444 // security purpose.
3446 rcu_thread_online();
3448 * This function dispatch the work to the kernel or userspace tracer
3449 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3450 * informations for the client. The command context struct contains
3451 * everything this function may needs.
3453 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3454 rcu_thread_offline();
3462 * TODO: Inform client somehow of the fatal error. At
3463 * this point, ret < 0 means that a zmalloc failed
3464 * (ENOMEM). Error detected but still accept
3465 * command, unless a socket error has been
3468 clean_command_ctx(&cmd_ctx
);
3472 health_code_update();
3474 DBG("Sending response (size: %d, retcode: %s)",
3475 cmd_ctx
->lttng_msg_size
,
3476 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3477 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3479 ERR("Failed to send data back to client");
3482 /* End of transmission */
3489 clean_command_ctx(&cmd_ctx
);
3491 health_code_update();
3503 lttng_poll_clean(&events
);
3504 clean_command_ctx(&cmd_ctx
);
3509 unlink(client_unix_sock_path
);
3510 if (client_sock
>= 0) {
3511 ret
= close(client_sock
);
3519 ERR("Health error occurred in %s", __func__
);
3522 health_unregister();
3524 DBG("Client thread dying");
3526 rcu_unregister_thread();
3532 * usage function on stderr
3534 static void usage(void)
3536 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
3537 fprintf(stderr
, " -h, --help Display this usage.\n");
3538 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
3539 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3540 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3541 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3542 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3543 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3544 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3545 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3546 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3547 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3548 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3549 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3550 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
3551 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3552 fprintf(stderr
, " -V, --version Show version number.\n");
3553 fprintf(stderr
, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3554 fprintf(stderr
, " -q, --quiet No output at all.\n");
3555 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3556 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3557 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3558 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
3562 * daemon argument parsing
3564 static int parse_args(int argc
, char **argv
)
3568 static struct option long_options
[] = {
3569 { "client-sock", 1, 0, 'c' },
3570 { "apps-sock", 1, 0, 'a' },
3571 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3572 { "kconsumerd-err-sock", 1, 0, 'E' },
3573 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3574 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3575 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3576 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3577 { "consumerd32-path", 1, 0, 'u' },
3578 { "consumerd32-libdir", 1, 0, 'U' },
3579 { "consumerd64-path", 1, 0, 't' },
3580 { "consumerd64-libdir", 1, 0, 'T' },
3581 { "daemonize", 0, 0, 'd' },
3582 { "sig-parent", 0, 0, 'S' },
3583 { "help", 0, 0, 'h' },
3584 { "group", 1, 0, 'g' },
3585 { "version", 0, 0, 'V' },
3586 { "quiet", 0, 0, 'q' },
3587 { "verbose", 0, 0, 'v' },
3588 { "verbose-consumer", 0, 0, 'Z' },
3589 { "no-kernel", 0, 0, 'N' },
3590 { "pidfile", 1, 0, 'p' },
3595 int option_index
= 0;
3596 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3597 long_options
, &option_index
);
3604 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
3606 fprintf(stderr
, " with arg %s\n", optarg
);
3610 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3613 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3619 opt_tracing_group
= optarg
;
3625 fprintf(stdout
, "%s\n", VERSION
);
3631 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3634 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3637 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3640 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3643 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3646 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
3652 lttng_opt_quiet
= 1;
3655 /* Verbose level can increase using multiple -v */
3656 lttng_opt_verbose
+= 1;
3659 opt_verbose_consumer
+= 1;
3662 consumerd32_bin
= optarg
;
3665 consumerd32_libdir
= optarg
;
3668 consumerd64_bin
= optarg
;
3671 consumerd64_libdir
= optarg
;
3674 opt_pidfile
= optarg
;
3677 /* Unknown option or other error.
3678 * Error is printed by getopt, just return */
3687 * Creates the two needed socket by the daemon.
3688 * apps_sock - The communication socket for all UST apps.
3689 * client_sock - The communication of the cli tool (lttng).
3691 static int init_daemon_socket(void)
3696 old_umask
= umask(0);
3698 /* Create client tool unix socket */
3699 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
3700 if (client_sock
< 0) {
3701 ERR("Create unix sock failed: %s", client_unix_sock_path
);
3706 /* Set the cloexec flag */
3707 ret
= utils_set_fd_cloexec(client_sock
);
3709 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3710 "Continuing but note that the consumer daemon will have a "
3711 "reference to this socket on exec()", client_sock
);
3714 /* File permission MUST be 660 */
3715 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3717 ERR("Set file permissions failed: %s", client_unix_sock_path
);
3722 /* Create the application unix socket */
3723 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
3724 if (apps_sock
< 0) {
3725 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
3730 /* Set the cloexec flag */
3731 ret
= utils_set_fd_cloexec(apps_sock
);
3733 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3734 "Continuing but note that the consumer daemon will have a "
3735 "reference to this socket on exec()", apps_sock
);
3738 /* File permission MUST be 666 */
3739 ret
= chmod(apps_unix_sock_path
,
3740 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
3742 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
3747 DBG3("Session daemon client socket %d and application socket %d created",
3748 client_sock
, apps_sock
);
3756 * Check if the global socket is available, and if a daemon is answering at the
3757 * other side. If yes, error is returned.
3759 static int check_existing_daemon(void)
3761 /* Is there anybody out there ? */
3762 if (lttng_session_daemon_alive()) {
3770 * Set the tracing group gid onto the client socket.
3772 * Race window between mkdir and chown is OK because we are going from more
3773 * permissive (root.root) to less permissive (root.tracing).
3775 static int set_permissions(char *rundir
)
3780 ret
= allowed_group();
3782 WARN("No tracing group detected");
3789 /* Set lttng run dir */
3790 ret
= chown(rundir
, 0, gid
);
3792 ERR("Unable to set group on %s", rundir
);
3796 /* Ensure tracing group can search the run dir */
3797 ret
= chmod(rundir
, S_IRWXU
| S_IXGRP
| S_IXOTH
);
3799 ERR("Unable to set permissions on %s", rundir
);
3803 /* lttng client socket path */
3804 ret
= chown(client_unix_sock_path
, 0, gid
);
3806 ERR("Unable to set group on %s", client_unix_sock_path
);
3810 /* kconsumer error socket path */
3811 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, gid
);
3813 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
3817 /* 64-bit ustconsumer error socket path */
3818 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, gid
);
3820 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
3824 /* 32-bit ustconsumer compat32 error socket path */
3825 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, gid
);
3827 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
3831 DBG("All permissions are set");
3838 * Create the lttng run directory needed for all global sockets and pipe.
3840 static int create_lttng_rundir(const char *rundir
)
3844 DBG3("Creating LTTng run directory: %s", rundir
);
3846 ret
= mkdir(rundir
, S_IRWXU
);
3848 if (errno
!= EEXIST
) {
3849 ERR("Unable to create %s", rundir
);
3861 * Setup sockets and directory needed by the kconsumerd communication with the
3864 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
3868 char path
[PATH_MAX
];
3870 switch (consumer_data
->type
) {
3871 case LTTNG_CONSUMER_KERNEL
:
3872 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
3874 case LTTNG_CONSUMER64_UST
:
3875 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
3877 case LTTNG_CONSUMER32_UST
:
3878 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
3881 ERR("Consumer type unknown");
3886 DBG2("Creating consumer directory: %s", path
);
3888 ret
= mkdir(path
, S_IRWXU
);
3890 if (errno
!= EEXIST
) {
3892 ERR("Failed to create %s", path
);
3898 /* Create the kconsumerd error unix socket */
3899 consumer_data
->err_sock
=
3900 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
3901 if (consumer_data
->err_sock
< 0) {
3902 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
3907 /* File permission MUST be 660 */
3908 ret
= chmod(consumer_data
->err_unix_sock_path
,
3909 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3911 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
3921 * Signal handler for the daemon
3923 * Simply stop all worker threads, leaving main() return gracefully after
3924 * joining all threads and calling cleanup().
3926 static void sighandler(int sig
)
3930 DBG("SIGPIPE caught");
3933 DBG("SIGINT caught");
3937 DBG("SIGTERM caught");
3946 * Setup signal handler for :
3947 * SIGINT, SIGTERM, SIGPIPE
3949 static int set_signal_handler(void)
3952 struct sigaction sa
;
3955 if ((ret
= sigemptyset(&sigset
)) < 0) {
3956 PERROR("sigemptyset");
3960 sa
.sa_handler
= sighandler
;
3961 sa
.sa_mask
= sigset
;
3963 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
3964 PERROR("sigaction");
3968 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
3969 PERROR("sigaction");
3973 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
3974 PERROR("sigaction");
3978 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3984 * Set open files limit to unlimited. This daemon can open a large number of
3985 * file descriptors in order to consumer multiple kernel traces.
3987 static void set_ulimit(void)
3992 /* The kernel does not allowed an infinite limit for open files */
3993 lim
.rlim_cur
= 65535;
3994 lim
.rlim_max
= 65535;
3996 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
3998 PERROR("failed to set open files limit");
4003 * Write pidfile using the rundir and opt_pidfile.
4005 static void write_pidfile(void)
4008 char pidfile_path
[PATH_MAX
];
4013 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4015 /* Build pidfile path from rundir and opt_pidfile. */
4016 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4017 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4019 PERROR("snprintf pidfile path");
4025 * Create pid file in rundir. Return value is of no importance. The
4026 * execution will continue even though we are not able to write the file.
4028 (void) utils_create_pid_file(getpid(), pidfile_path
);
4037 int main(int argc
, char **argv
)
4041 const char *home_path
, *env_app_timeout
;
4043 init_kernel_workarounds();
4045 rcu_register_thread();
4047 setup_consumerd_path();
4049 page_size
= sysconf(_SC_PAGESIZE
);
4050 if (page_size
< 0) {
4051 PERROR("sysconf _SC_PAGESIZE");
4052 page_size
= LONG_MAX
;
4053 WARN("Fallback page size to %ld", page_size
);
4056 /* Parse arguments */
4058 if ((ret
= parse_args(argc
, argv
)) < 0) {
4068 * child: setsid, close FD 0, 1, 2, chdir /
4069 * parent: exit (if fork is successful)
4077 * We are in the child. Make sure all other file
4078 * descriptors are closed, in case we are called with
4079 * more opened file descriptors than the standard ones.
4081 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4086 /* Create thread quit pipe */
4087 if ((ret
= init_thread_quit_pipe()) < 0) {
4091 /* Check if daemon is UID = 0 */
4092 is_root
= !getuid();
4095 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4097 /* Create global run dir with root access */
4098 ret
= create_lttng_rundir(rundir
);
4103 if (strlen(apps_unix_sock_path
) == 0) {
4104 snprintf(apps_unix_sock_path
, PATH_MAX
,
4105 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4108 if (strlen(client_unix_sock_path
) == 0) {
4109 snprintf(client_unix_sock_path
, PATH_MAX
,
4110 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4113 /* Set global SHM for ust */
4114 if (strlen(wait_shm_path
) == 0) {
4115 snprintf(wait_shm_path
, PATH_MAX
,
4116 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4119 if (strlen(health_unix_sock_path
) == 0) {
4120 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4121 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4124 /* Setup kernel consumerd path */
4125 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4126 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4127 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4128 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4130 DBG2("Kernel consumer err path: %s",
4131 kconsumer_data
.err_unix_sock_path
);
4132 DBG2("Kernel consumer cmd path: %s",
4133 kconsumer_data
.cmd_unix_sock_path
);
4135 home_path
= get_home_dir();
4136 if (home_path
== NULL
) {
4137 /* TODO: Add --socket PATH option */
4138 ERR("Can't get HOME directory for sockets creation.");
4144 * Create rundir from home path. This will create something like
4147 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4153 ret
= create_lttng_rundir(rundir
);
4158 if (strlen(apps_unix_sock_path
) == 0) {
4159 snprintf(apps_unix_sock_path
, PATH_MAX
,
4160 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4163 /* Set the cli tool unix socket path */
4164 if (strlen(client_unix_sock_path
) == 0) {
4165 snprintf(client_unix_sock_path
, PATH_MAX
,
4166 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4169 /* Set global SHM for ust */
4170 if (strlen(wait_shm_path
) == 0) {
4171 snprintf(wait_shm_path
, PATH_MAX
,
4172 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4175 /* Set health check Unix path */
4176 if (strlen(health_unix_sock_path
) == 0) {
4177 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4178 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4182 /* Set consumer initial state */
4183 kernel_consumerd_state
= CONSUMER_STOPPED
;
4184 ust_consumerd_state
= CONSUMER_STOPPED
;
4186 DBG("Client socket path %s", client_unix_sock_path
);
4187 DBG("Application socket path %s", apps_unix_sock_path
);
4188 DBG("Application wait path %s", wait_shm_path
);
4189 DBG("LTTng run directory path: %s", rundir
);
4191 /* 32 bits consumerd path setup */
4192 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4193 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4194 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4195 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4197 DBG2("UST consumer 32 bits err path: %s",
4198 ustconsumer32_data
.err_unix_sock_path
);
4199 DBG2("UST consumer 32 bits cmd path: %s",
4200 ustconsumer32_data
.cmd_unix_sock_path
);
4202 /* 64 bits consumerd path setup */
4203 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4204 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4205 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4206 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4208 DBG2("UST consumer 64 bits err path: %s",
4209 ustconsumer64_data
.err_unix_sock_path
);
4210 DBG2("UST consumer 64 bits cmd path: %s",
4211 ustconsumer64_data
.cmd_unix_sock_path
);
4214 * See if daemon already exist.
4216 if ((ret
= check_existing_daemon()) < 0) {
4217 ERR("Already running daemon.\n");
4219 * We do not goto exit because we must not cleanup()
4220 * because a daemon is already running.
4226 * Init UST app hash table. Alloc hash table before this point since
4227 * cleanup() can get called after that point.
4231 /* After this point, we can safely call cleanup() with "goto exit" */
4234 * These actions must be executed as root. We do that *after* setting up
4235 * the sockets path because we MUST make the check for another daemon using
4236 * those paths *before* trying to set the kernel consumer sockets and init
4240 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4245 /* Setup kernel tracer */
4246 if (!opt_no_kernel
) {
4247 init_kernel_tracer();
4250 /* Set ulimit for open files */
4253 /* init lttng_fd tracking must be done after set_ulimit. */
4256 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4261 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4266 if ((ret
= set_signal_handler()) < 0) {
4270 /* Setup the needed unix socket */
4271 if ((ret
= init_daemon_socket()) < 0) {
4275 /* Set credentials to socket */
4276 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4280 /* Get parent pid if -S, --sig-parent is specified. */
4281 if (opt_sig_parent
) {
4285 /* Setup the kernel pipe for waking up the kernel thread */
4286 if (is_root
&& !opt_no_kernel
) {
4287 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4292 /* Setup the thread apps communication pipe. */
4293 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4297 /* Setup the thread apps notify communication pipe. */
4298 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4302 /* Initialize global buffer per UID and PID registry. */
4303 buffer_reg_init_uid_registry();
4304 buffer_reg_init_pid_registry();
4306 /* Init UST command queue. */
4307 cds_wfq_init(&ust_cmd_queue
.queue
);
4310 * Get session list pointer. This pointer MUST NOT be free(). This list is
4311 * statically declared in session.c
4313 session_list_ptr
= session_get_list();
4315 /* Set up max poll set size */
4316 lttng_poll_set_max_size();
4320 /* Check for the application socket timeout env variable. */
4321 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4322 if (env_app_timeout
) {
4323 app_socket_timeout
= atoi(env_app_timeout
);
4325 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4330 /* Create thread to manage the client socket */
4331 ret
= pthread_create(&health_thread
, NULL
,
4332 thread_manage_health
, (void *) NULL
);
4334 PERROR("pthread_create health");
4338 /* Create thread to manage the client socket */
4339 ret
= pthread_create(&client_thread
, NULL
,
4340 thread_manage_clients
, (void *) NULL
);
4342 PERROR("pthread_create clients");
4346 /* Create thread to dispatch registration */
4347 ret
= pthread_create(&dispatch_thread
, NULL
,
4348 thread_dispatch_ust_registration
, (void *) NULL
);
4350 PERROR("pthread_create dispatch");
4354 /* Create thread to manage application registration. */
4355 ret
= pthread_create(®_apps_thread
, NULL
,
4356 thread_registration_apps
, (void *) NULL
);
4358 PERROR("pthread_create registration");
4362 /* Create thread to manage application socket */
4363 ret
= pthread_create(&apps_thread
, NULL
,
4364 thread_manage_apps
, (void *) NULL
);
4366 PERROR("pthread_create apps");
4370 /* Create thread to manage application notify socket */
4371 ret
= pthread_create(&apps_notify_thread
, NULL
,
4372 ust_thread_manage_notify
, (void *) NULL
);
4374 PERROR("pthread_create apps");
4378 /* Don't start this thread if kernel tracing is not requested nor root */
4379 if (is_root
&& !opt_no_kernel
) {
4380 /* Create kernel thread to manage kernel event */
4381 ret
= pthread_create(&kernel_thread
, NULL
,
4382 thread_manage_kernel
, (void *) NULL
);
4384 PERROR("pthread_create kernel");
4388 ret
= pthread_join(kernel_thread
, &status
);
4390 PERROR("pthread_join");
4391 goto error
; /* join error, exit without cleanup */
4396 ret
= pthread_join(apps_thread
, &status
);
4398 PERROR("pthread_join");
4399 goto error
; /* join error, exit without cleanup */
4403 ret
= pthread_join(reg_apps_thread
, &status
);
4405 PERROR("pthread_join");
4406 goto error
; /* join error, exit without cleanup */
4410 ret
= pthread_join(dispatch_thread
, &status
);
4412 PERROR("pthread_join");
4413 goto error
; /* join error, exit without cleanup */
4417 ret
= pthread_join(client_thread
, &status
);
4419 PERROR("pthread_join");
4420 goto error
; /* join error, exit without cleanup */
4423 ret
= join_consumer_thread(&kconsumer_data
);
4425 PERROR("join_consumer");
4426 goto error
; /* join error, exit without cleanup */
4429 ret
= join_consumer_thread(&ustconsumer32_data
);
4431 PERROR("join_consumer ust32");
4432 goto error
; /* join error, exit without cleanup */
4435 ret
= join_consumer_thread(&ustconsumer64_data
);
4437 PERROR("join_consumer ust64");
4438 goto error
; /* join error, exit without cleanup */
4442 ret
= pthread_join(health_thread
, &status
);
4444 PERROR("pthread_join health thread");
4445 goto error
; /* join error, exit without cleanup */
4451 * cleanup() is called when no other thread is running.
4453 rcu_thread_online();
4455 rcu_thread_offline();
4456 rcu_unregister_thread();