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.
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
35 #include <sys/types.h>
37 #include <urcu/uatomic.h>
41 #include <common/common.h>
42 #include <common/compat/socket.h>
43 #include <common/defaults.h>
44 #include <common/kernel-consumer/kernel-consumer.h>
45 #include <common/futex.h>
46 #include <common/relayd/relayd.h>
47 #include <common/utils.h>
48 #include <common/daemonize.h>
50 #include "lttng-sessiond.h"
51 #include "buffer-registry.h"
58 #include "kernel-consumer.h"
62 #include "ust-consumer.h"
65 #include "health-sessiond.h"
66 #include "testpoint.h"
67 #include "ust-thread.h"
68 #include "jul-thread.h"
70 #define CONSUMERD_FILE "lttng-consumerd"
73 static const char *tracing_group_name
= DEFAULT_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
, opt_background
;
78 static int opt_no_kernel
;
79 static pid_t ppid
; /* Parent PID for --sig-parent option */
80 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
82 static int lockfile_fd
= -1;
84 /* Set to 1 when a SIGUSR1 signal is received. */
85 static int recv_child_signal
;
88 * Consumer daemon specific control data. Every value not initialized here is
89 * set to 0 by the static definition.
91 static struct consumer_data kconsumer_data
= {
92 .type
= LTTNG_CONSUMER_KERNEL
,
93 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
94 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
97 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
98 .lock
= PTHREAD_MUTEX_INITIALIZER
,
99 .cond
= PTHREAD_COND_INITIALIZER
,
100 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
102 static struct consumer_data ustconsumer64_data
= {
103 .type
= LTTNG_CONSUMER64_UST
,
104 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
105 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
108 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
109 .lock
= PTHREAD_MUTEX_INITIALIZER
,
110 .cond
= PTHREAD_COND_INITIALIZER
,
111 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
113 static struct consumer_data ustconsumer32_data
= {
114 .type
= LTTNG_CONSUMER32_UST
,
115 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
116 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
119 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
120 .lock
= PTHREAD_MUTEX_INITIALIZER
,
121 .cond
= PTHREAD_COND_INITIALIZER
,
122 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
125 /* Shared between threads */
126 static int dispatch_thread_exit
;
128 /* Global application Unix socket path */
129 static char apps_unix_sock_path
[PATH_MAX
];
130 /* Global client Unix socket path */
131 static char client_unix_sock_path
[PATH_MAX
];
132 /* global wait shm path for UST */
133 static char wait_shm_path
[PATH_MAX
];
134 /* Global health check unix path */
135 static char health_unix_sock_path
[PATH_MAX
];
137 /* Sockets and FDs */
138 static int client_sock
= -1;
139 static int apps_sock
= -1;
140 int kernel_tracer_fd
= -1;
141 static int kernel_poll_pipe
[2] = { -1, -1 };
144 * Quit pipe for all threads. This permits a single cancellation point
145 * for all threads when receiving an event on the pipe.
147 static int thread_quit_pipe
[2] = { -1, -1 };
150 * This pipe is used to inform the thread managing application communication
151 * that a command is queued and ready to be processed.
153 static int apps_cmd_pipe
[2] = { -1, -1 };
155 int apps_cmd_notify_pipe
[2] = { -1, -1 };
157 /* Pthread, Mutexes and Semaphores */
158 static pthread_t apps_thread
;
159 static pthread_t apps_notify_thread
;
160 static pthread_t reg_apps_thread
;
161 static pthread_t client_thread
;
162 static pthread_t kernel_thread
;
163 static pthread_t dispatch_thread
;
164 static pthread_t health_thread
;
165 static pthread_t ht_cleanup_thread
;
166 static pthread_t jul_reg_thread
;
169 * UST registration command queue. This queue is tied with a futex and uses a N
170 * wakers / 1 waiter implemented and detailed in futex.c/.h
172 * The thread_manage_apps and thread_dispatch_ust_registration interact with
173 * this queue and the wait/wake scheme.
175 static struct ust_cmd_queue ust_cmd_queue
;
178 * Pointer initialized before thread creation.
180 * This points to the tracing session list containing the session count and a
181 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
182 * MUST NOT be taken if you call a public function in session.c.
184 * The lock is nested inside the structure: session_list_ptr->lock. Please use
185 * session_lock_list and session_unlock_list for lock acquisition.
187 static struct ltt_session_list
*session_list_ptr
;
189 int ust_consumerd64_fd
= -1;
190 int ust_consumerd32_fd
= -1;
192 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
193 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
194 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
195 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
197 static const char *module_proc_lttng
= "/proc/lttng";
200 * Consumer daemon state which is changed when spawning it, killing it or in
201 * case of a fatal error.
203 enum consumerd_state
{
204 CONSUMER_STARTED
= 1,
205 CONSUMER_STOPPED
= 2,
210 * This consumer daemon state is used to validate if a client command will be
211 * able to reach the consumer. If not, the client is informed. For instance,
212 * doing a "lttng start" when the consumer state is set to ERROR will return an
213 * error to the client.
215 * The following example shows a possible race condition of this scheme:
217 * consumer thread error happens
219 * client cmd checks state -> still OK
220 * consumer thread exit, sets error
221 * client cmd try to talk to consumer
224 * However, since the consumer is a different daemon, we have no way of making
225 * sure the command will reach it safely even with this state flag. This is why
226 * we consider that up to the state validation during command processing, the
227 * command is safe. After that, we can not guarantee the correctness of the
228 * client request vis-a-vis the consumer.
230 static enum consumerd_state ust_consumerd_state
;
231 static enum consumerd_state kernel_consumerd_state
;
234 * Socket timeout for receiving and sending in seconds.
236 static int app_socket_timeout
;
238 /* Set in main() with the current page size. */
241 /* Application health monitoring */
242 struct health_app
*health_sessiond
;
244 /* JUL TCP port for registration. Used by the JUL thread. */
245 unsigned int jul_tcp_port
= DEFAULT_JUL_TCP_PORT
;
247 /* Am I root or not. */
248 int is_root
; /* Set to 1 if the daemon is running as root */
251 * Whether sessiond is ready for commands/health check requests.
252 * NR_LTTNG_SESSIOND_READY must match the number of calls to
253 * lttng_sessiond_notify_ready().
255 #define NR_LTTNG_SESSIOND_READY 2
256 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
258 /* Notify parents that we are ready for cmd and health check */
260 void lttng_sessiond_notify_ready(void)
262 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
264 * Notify parent pid that we are ready to accept command
265 * for client side. This ppid is the one from the
266 * external process that spawned us.
268 if (opt_sig_parent
) {
273 * Notify the parent of the fork() process that we are
276 if (opt_daemon
|| opt_background
) {
277 kill(child_ppid
, SIGUSR1
);
283 void setup_consumerd_path(void)
285 const char *bin
, *libdir
;
288 * Allow INSTALL_BIN_PATH to be used as a target path for the
289 * native architecture size consumer if CONFIG_CONSUMER*_PATH
290 * has not been defined.
292 #if (CAA_BITS_PER_LONG == 32)
293 if (!consumerd32_bin
[0]) {
294 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
296 if (!consumerd32_libdir
[0]) {
297 consumerd32_libdir
= INSTALL_LIB_PATH
;
299 #elif (CAA_BITS_PER_LONG == 64)
300 if (!consumerd64_bin
[0]) {
301 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
303 if (!consumerd64_libdir
[0]) {
304 consumerd64_libdir
= INSTALL_LIB_PATH
;
307 #error "Unknown bitness"
311 * runtime env. var. overrides the build default.
313 bin
= getenv("LTTNG_CONSUMERD32_BIN");
315 consumerd32_bin
= bin
;
317 bin
= getenv("LTTNG_CONSUMERD64_BIN");
319 consumerd64_bin
= bin
;
321 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
323 consumerd32_libdir
= libdir
;
325 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
327 consumerd64_libdir
= libdir
;
332 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
334 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
340 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
346 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
358 * Check if the thread quit pipe was triggered.
360 * Return 1 if it was triggered else 0;
362 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
364 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
372 * Init thread quit pipe.
374 * Return -1 on error or 0 if all pipes are created.
376 static int init_thread_quit_pipe(void)
380 ret
= pipe(thread_quit_pipe
);
382 PERROR("thread quit pipe");
386 for (i
= 0; i
< 2; i
++) {
387 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
399 * Stop all threads by closing the thread quit pipe.
401 static void stop_threads(void)
405 /* Stopping all threads */
406 DBG("Terminating all threads");
407 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
409 ERR("write error on thread quit pipe");
412 /* Dispatch thread */
413 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
414 futex_nto1_wake(&ust_cmd_queue
.futex
);
418 * Close every consumer sockets.
420 static void close_consumer_sockets(void)
424 if (kconsumer_data
.err_sock
>= 0) {
425 ret
= close(kconsumer_data
.err_sock
);
427 PERROR("kernel consumer err_sock close");
430 if (ustconsumer32_data
.err_sock
>= 0) {
431 ret
= close(ustconsumer32_data
.err_sock
);
433 PERROR("UST consumerd32 err_sock close");
436 if (ustconsumer64_data
.err_sock
>= 0) {
437 ret
= close(ustconsumer64_data
.err_sock
);
439 PERROR("UST consumerd64 err_sock close");
442 if (kconsumer_data
.cmd_sock
>= 0) {
443 ret
= close(kconsumer_data
.cmd_sock
);
445 PERROR("kernel consumer cmd_sock close");
448 if (ustconsumer32_data
.cmd_sock
>= 0) {
449 ret
= close(ustconsumer32_data
.cmd_sock
);
451 PERROR("UST consumerd32 cmd_sock close");
454 if (ustconsumer64_data
.cmd_sock
>= 0) {
455 ret
= close(ustconsumer64_data
.cmd_sock
);
457 PERROR("UST consumerd64 cmd_sock close");
463 * Generate the full lock file path using the rundir.
465 * Return the snprintf() return value thus a negative value is an error.
467 static int generate_lock_file_path(char *path
, size_t len
)
474 /* Build lockfile path from rundir. */
475 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
477 PERROR("snprintf lockfile path");
486 static void cleanup(void)
489 struct ltt_session
*sess
, *stmp
;
495 * Close the thread quit pipe. It has already done its job,
496 * since we are now called.
498 utils_close_pipe(thread_quit_pipe
);
501 * If opt_pidfile is undefined, the default file will be wiped when
502 * removing the rundir.
505 ret
= remove(opt_pidfile
);
507 PERROR("remove pidfile %s", opt_pidfile
);
511 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
514 snprintf(path
, PATH_MAX
,
516 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
517 DBG("Removing %s", path
);
520 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
521 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
);
522 DBG("Removing %s", path
);
526 snprintf(path
, PATH_MAX
,
527 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
529 DBG("Removing %s", path
);
532 snprintf(path
, PATH_MAX
,
533 DEFAULT_KCONSUMERD_PATH
,
535 DBG("Removing directory %s", path
);
538 /* ust consumerd 32 */
539 snprintf(path
, PATH_MAX
,
540 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
542 DBG("Removing %s", path
);
545 snprintf(path
, PATH_MAX
,
546 DEFAULT_USTCONSUMERD32_PATH
,
548 DBG("Removing directory %s", path
);
551 /* ust consumerd 64 */
552 snprintf(path
, PATH_MAX
,
553 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
555 DBG("Removing %s", path
);
558 snprintf(path
, PATH_MAX
,
559 DEFAULT_USTCONSUMERD64_PATH
,
561 DBG("Removing directory %s", path
);
564 DBG("Cleaning up all sessions");
566 /* Destroy session list mutex */
567 if (session_list_ptr
!= NULL
) {
568 pthread_mutex_destroy(&session_list_ptr
->lock
);
570 /* Cleanup ALL session */
571 cds_list_for_each_entry_safe(sess
, stmp
,
572 &session_list_ptr
->head
, list
) {
573 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
577 DBG("Closing all UST sockets");
578 ust_app_clean_list();
579 buffer_reg_destroy_registries();
581 if (is_root
&& !opt_no_kernel
) {
582 DBG2("Closing kernel fd");
583 if (kernel_tracer_fd
>= 0) {
584 ret
= close(kernel_tracer_fd
);
589 DBG("Unloading kernel modules");
590 modprobe_remove_lttng_all();
593 close_consumer_sockets();
597 * Cleanup lock file by deleting it and finaly closing it which will
598 * release the file system lock.
600 if (lockfile_fd
>= 0) {
601 char lockfile_path
[PATH_MAX
];
603 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
605 ret
= remove(lockfile_path
);
607 PERROR("remove lock file");
609 ret
= close(lockfile_fd
);
611 PERROR("close lock file");
617 * We do NOT rmdir rundir because there are other processes
618 * using it, for instance lttng-relayd, which can start in
619 * parallel with this teardown.
625 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
626 "Matthew, BEET driven development works!%c[%dm",
627 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
632 * Send data on a unix socket using the liblttsessiondcomm API.
634 * Return lttcomm error code.
636 static int send_unix_sock(int sock
, void *buf
, size_t len
)
638 /* Check valid length */
643 return lttcomm_send_unix_sock(sock
, buf
, len
);
647 * Free memory of a command context structure.
649 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
651 DBG("Clean command context structure");
653 if ((*cmd_ctx
)->llm
) {
654 free((*cmd_ctx
)->llm
);
656 if ((*cmd_ctx
)->lsm
) {
657 free((*cmd_ctx
)->lsm
);
665 * Notify UST applications using the shm mmap futex.
667 static int notify_ust_apps(int active
)
671 DBG("Notifying applications of session daemon state: %d", active
);
673 /* See shm.c for this call implying mmap, shm and futex calls */
674 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
675 if (wait_shm_mmap
== NULL
) {
679 /* Wake waiting process */
680 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
682 /* Apps notified successfully */
690 * Setup the outgoing data buffer for the response (llm) by allocating the
691 * right amount of memory and copying the original information from the lsm
694 * Return total size of the buffer pointed by buf.
696 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
702 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
703 if (cmd_ctx
->llm
== NULL
) {
709 /* Copy common data */
710 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
711 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
713 cmd_ctx
->llm
->data_size
= size
;
714 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
723 * Update the kernel poll set of all channel fd available over all tracing
724 * session. Add the wakeup pipe at the end of the set.
726 static int update_kernel_poll(struct lttng_poll_event
*events
)
729 struct ltt_session
*session
;
730 struct ltt_kernel_channel
*channel
;
732 DBG("Updating kernel poll set");
735 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
736 session_lock(session
);
737 if (session
->kernel_session
== NULL
) {
738 session_unlock(session
);
742 cds_list_for_each_entry(channel
,
743 &session
->kernel_session
->channel_list
.head
, list
) {
744 /* Add channel fd to the kernel poll set */
745 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
747 session_unlock(session
);
750 DBG("Channel fd %d added to kernel set", channel
->fd
);
752 session_unlock(session
);
754 session_unlock_list();
759 session_unlock_list();
764 * Find the channel fd from 'fd' over all tracing session. When found, check
765 * for new channel stream and send those stream fds to the kernel consumer.
767 * Useful for CPU hotplug feature.
769 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
772 struct ltt_session
*session
;
773 struct ltt_kernel_session
*ksess
;
774 struct ltt_kernel_channel
*channel
;
776 DBG("Updating kernel streams for channel fd %d", fd
);
779 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
780 session_lock(session
);
781 if (session
->kernel_session
== NULL
) {
782 session_unlock(session
);
785 ksess
= session
->kernel_session
;
787 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
788 if (channel
->fd
== fd
) {
789 DBG("Channel found, updating kernel streams");
790 ret
= kernel_open_channel_stream(channel
);
794 /* Update the stream global counter */
795 ksess
->stream_count_global
+= ret
;
798 * Have we already sent fds to the consumer? If yes, it means
799 * that tracing is started so it is safe to send our updated
802 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
803 struct lttng_ht_iter iter
;
804 struct consumer_socket
*socket
;
807 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
808 &iter
.iter
, socket
, node
.node
) {
809 pthread_mutex_lock(socket
->lock
);
810 ret
= kernel_consumer_send_channel_stream(socket
,
812 session
->output_traces
? 1 : 0);
813 pthread_mutex_unlock(socket
->lock
);
824 session_unlock(session
);
826 session_unlock_list();
830 session_unlock(session
);
831 session_unlock_list();
836 * For each tracing session, update newly registered apps. The session list
837 * lock MUST be acquired before calling this.
839 static void update_ust_app(int app_sock
)
841 struct ltt_session
*sess
, *stmp
;
843 /* Consumer is in an ERROR state. Stop any application update. */
844 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
845 /* Stop the update process since the consumer is dead. */
849 /* For all tracing session(s) */
850 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
852 if (sess
->ust_session
) {
853 ust_app_global_update(sess
->ust_session
, app_sock
);
855 session_unlock(sess
);
860 * This thread manage event coming from the kernel.
862 * Features supported in this thread:
865 static void *thread_manage_kernel(void *data
)
867 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
868 uint32_t revents
, nb_fd
;
870 struct lttng_poll_event events
;
872 DBG("[thread] Thread manage kernel started");
874 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
877 * This first step of the while is to clean this structure which could free
878 * non NULL pointers so initialize it before the loop.
880 lttng_poll_init(&events
);
882 if (testpoint(sessiond_thread_manage_kernel
)) {
883 goto error_testpoint
;
886 health_code_update();
888 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
889 goto error_testpoint
;
893 health_code_update();
895 if (update_poll_flag
== 1) {
896 /* Clean events object. We are about to populate it again. */
897 lttng_poll_clean(&events
);
899 ret
= sessiond_set_thread_pollset(&events
, 2);
901 goto error_poll_create
;
904 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
909 /* This will add the available kernel channel if any. */
910 ret
= update_kernel_poll(&events
);
914 update_poll_flag
= 0;
917 DBG("Thread kernel polling");
919 /* Poll infinite value of time */
922 ret
= lttng_poll_wait(&events
, -1);
923 DBG("Thread kernel return from poll on %d fds",
924 LTTNG_POLL_GETNB(&events
));
928 * Restart interrupted system call.
930 if (errno
== EINTR
) {
934 } else if (ret
== 0) {
935 /* Should not happen since timeout is infinite */
936 ERR("Return value of poll is 0 with an infinite timeout.\n"
937 "This should not have happened! Continuing...");
943 for (i
= 0; i
< nb_fd
; i
++) {
944 /* Fetch once the poll data */
945 revents
= LTTNG_POLL_GETEV(&events
, i
);
946 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
948 health_code_update();
951 /* No activity for this FD (poll implementation). */
955 /* Thread quit pipe has been closed. Killing thread. */
956 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
962 /* Check for data on kernel pipe */
963 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
964 (void) lttng_read(kernel_poll_pipe
[0],
967 * Ret value is useless here, if this pipe gets any actions an
968 * update is required anyway.
970 update_poll_flag
= 1;
974 * New CPU detected by the kernel. Adding kernel stream to
975 * kernel session and updating the kernel consumer
977 if (revents
& LPOLLIN
) {
978 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
984 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
985 * and unregister kernel stream at this point.
994 lttng_poll_clean(&events
);
997 utils_close_pipe(kernel_poll_pipe
);
998 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1001 ERR("Health error occurred in %s", __func__
);
1002 WARN("Kernel thread died unexpectedly. "
1003 "Kernel tracing can continue but CPU hotplug is disabled.");
1005 health_unregister(health_sessiond
);
1006 DBG("Kernel thread dying");
1011 * Signal pthread condition of the consumer data that the thread.
1013 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1015 pthread_mutex_lock(&data
->cond_mutex
);
1018 * The state is set before signaling. It can be any value, it's the waiter
1019 * job to correctly interpret this condition variable associated to the
1020 * consumer pthread_cond.
1022 * A value of 0 means that the corresponding thread of the consumer data
1023 * was not started. 1 indicates that the thread has started and is ready
1024 * for action. A negative value means that there was an error during the
1027 data
->consumer_thread_is_ready
= state
;
1028 (void) pthread_cond_signal(&data
->cond
);
1030 pthread_mutex_unlock(&data
->cond_mutex
);
1034 * This thread manage the consumer error sent back to the session daemon.
1036 static void *thread_manage_consumer(void *data
)
1038 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1039 uint32_t revents
, nb_fd
;
1040 enum lttcomm_return_code code
;
1041 struct lttng_poll_event events
;
1042 struct consumer_data
*consumer_data
= data
;
1044 DBG("[thread] Manage consumer started");
1046 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1048 health_code_update();
1051 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1052 * metadata_sock. Nothing more will be added to this poll set.
1054 ret
= sessiond_set_thread_pollset(&events
, 3);
1060 * The error socket here is already in a listening state which was done
1061 * just before spawning this thread to avoid a race between the consumer
1062 * daemon exec trying to connect and the listen() call.
1064 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1069 health_code_update();
1071 /* Infinite blocking call, waiting for transmission */
1073 health_poll_entry();
1075 if (testpoint(sessiond_thread_manage_consumer
)) {
1079 ret
= lttng_poll_wait(&events
, -1);
1083 * Restart interrupted system call.
1085 if (errno
== EINTR
) {
1093 for (i
= 0; i
< nb_fd
; i
++) {
1094 /* Fetch once the poll data */
1095 revents
= LTTNG_POLL_GETEV(&events
, i
);
1096 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1098 health_code_update();
1101 /* No activity for this FD (poll implementation). */
1105 /* Thread quit pipe has been closed. Killing thread. */
1106 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1112 /* Event on the registration socket */
1113 if (pollfd
== consumer_data
->err_sock
) {
1114 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1115 ERR("consumer err socket poll error");
1121 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1127 * Set the CLOEXEC flag. Return code is useless because either way, the
1130 (void) utils_set_fd_cloexec(sock
);
1132 health_code_update();
1134 DBG2("Receiving code from consumer err_sock");
1136 /* Getting status code from kconsumerd */
1137 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1138 sizeof(enum lttcomm_return_code
));
1143 health_code_update();
1144 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1145 /* Connect both socket, command and metadata. */
1146 consumer_data
->cmd_sock
=
1147 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1148 consumer_data
->metadata_fd
=
1149 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1150 if (consumer_data
->cmd_sock
< 0
1151 || consumer_data
->metadata_fd
< 0) {
1152 PERROR("consumer connect cmd socket");
1153 /* On error, signal condition and quit. */
1154 signal_consumer_condition(consumer_data
, -1);
1157 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1158 /* Create metadata socket lock. */
1159 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1160 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1161 PERROR("zmalloc pthread mutex");
1165 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1167 signal_consumer_condition(consumer_data
, 1);
1168 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1169 DBG("Consumer metadata socket ready (fd: %d)",
1170 consumer_data
->metadata_fd
);
1172 ERR("consumer error when waiting for SOCK_READY : %s",
1173 lttcomm_get_readable_code(-code
));
1177 /* Remove the consumerd error sock since we've established a connexion */
1178 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1183 /* Add new accepted error socket. */
1184 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1189 /* Add metadata socket that is successfully connected. */
1190 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1191 LPOLLIN
| LPOLLRDHUP
);
1196 health_code_update();
1198 /* Infinite blocking call, waiting for transmission */
1201 health_code_update();
1203 /* Exit the thread because the thread quit pipe has been triggered. */
1205 /* Not a health error. */
1210 health_poll_entry();
1211 ret
= lttng_poll_wait(&events
, -1);
1215 * Restart interrupted system call.
1217 if (errno
== EINTR
) {
1225 for (i
= 0; i
< nb_fd
; i
++) {
1226 /* Fetch once the poll data */
1227 revents
= LTTNG_POLL_GETEV(&events
, i
);
1228 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1230 health_code_update();
1233 /* No activity for this FD (poll implementation). */
1238 * Thread quit pipe has been triggered, flag that we should stop
1239 * but continue the current loop to handle potential data from
1242 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1244 if (pollfd
== sock
) {
1245 /* Event on the consumerd socket */
1246 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1247 ERR("consumer err socket second poll error");
1250 health_code_update();
1251 /* Wait for any kconsumerd error */
1252 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1253 sizeof(enum lttcomm_return_code
));
1255 ERR("consumer closed the command socket");
1259 ERR("consumer return code : %s",
1260 lttcomm_get_readable_code(-code
));
1263 } else if (pollfd
== consumer_data
->metadata_fd
) {
1264 /* UST metadata requests */
1265 ret
= ust_consumer_metadata_request(
1266 &consumer_data
->metadata_sock
);
1268 ERR("Handling metadata request");
1272 /* No need for an else branch all FDs are tested prior. */
1274 health_code_update();
1280 * We lock here because we are about to close the sockets and some other
1281 * thread might be using them so get exclusive access which will abort all
1282 * other consumer command by other threads.
1284 pthread_mutex_lock(&consumer_data
->lock
);
1286 /* Immediately set the consumerd state to stopped */
1287 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1288 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1289 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1290 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1291 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1293 /* Code flow error... */
1297 if (consumer_data
->err_sock
>= 0) {
1298 ret
= close(consumer_data
->err_sock
);
1302 consumer_data
->err_sock
= -1;
1304 if (consumer_data
->cmd_sock
>= 0) {
1305 ret
= close(consumer_data
->cmd_sock
);
1309 consumer_data
->cmd_sock
= -1;
1311 if (consumer_data
->metadata_sock
.fd_ptr
&&
1312 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1313 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1325 unlink(consumer_data
->err_unix_sock_path
);
1326 unlink(consumer_data
->cmd_unix_sock_path
);
1327 consumer_data
->pid
= 0;
1328 pthread_mutex_unlock(&consumer_data
->lock
);
1330 /* Cleanup metadata socket mutex. */
1331 if (consumer_data
->metadata_sock
.lock
) {
1332 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1333 free(consumer_data
->metadata_sock
.lock
);
1335 lttng_poll_clean(&events
);
1339 ERR("Health error occurred in %s", __func__
);
1341 health_unregister(health_sessiond
);
1342 DBG("consumer thread cleanup completed");
1348 * This thread manage application communication.
1350 static void *thread_manage_apps(void *data
)
1352 int i
, ret
, pollfd
, err
= -1;
1354 uint32_t revents
, nb_fd
;
1355 struct lttng_poll_event events
;
1357 DBG("[thread] Manage application started");
1359 rcu_register_thread();
1360 rcu_thread_online();
1362 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1364 if (testpoint(sessiond_thread_manage_apps
)) {
1365 goto error_testpoint
;
1368 health_code_update();
1370 ret
= sessiond_set_thread_pollset(&events
, 2);
1372 goto error_poll_create
;
1375 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1380 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1384 health_code_update();
1387 DBG("Apps thread polling");
1389 /* Inifinite blocking call, waiting for transmission */
1391 health_poll_entry();
1392 ret
= lttng_poll_wait(&events
, -1);
1393 DBG("Apps thread return from poll on %d fds",
1394 LTTNG_POLL_GETNB(&events
));
1398 * Restart interrupted system call.
1400 if (errno
== EINTR
) {
1408 for (i
= 0; i
< nb_fd
; i
++) {
1409 /* Fetch once the poll data */
1410 revents
= LTTNG_POLL_GETEV(&events
, i
);
1411 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1413 health_code_update();
1416 /* No activity for this FD (poll implementation). */
1420 /* Thread quit pipe has been closed. Killing thread. */
1421 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1427 /* Inspect the apps cmd pipe */
1428 if (pollfd
== apps_cmd_pipe
[0]) {
1429 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1430 ERR("Apps command pipe error");
1432 } else if (revents
& LPOLLIN
) {
1436 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1437 if (size_ret
< sizeof(sock
)) {
1438 PERROR("read apps cmd pipe");
1442 health_code_update();
1445 * We only monitor the error events of the socket. This
1446 * thread does not handle any incoming data from UST
1449 ret
= lttng_poll_add(&events
, sock
,
1450 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1455 DBG("Apps with sock %d added to poll set", sock
);
1459 * At this point, we know that a registered application made
1460 * the event at poll_wait.
1462 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1463 /* Removing from the poll set */
1464 ret
= lttng_poll_del(&events
, pollfd
);
1469 /* Socket closed on remote end. */
1470 ust_app_unregister(pollfd
);
1474 health_code_update();
1480 lttng_poll_clean(&events
);
1483 utils_close_pipe(apps_cmd_pipe
);
1484 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1487 * We don't clean the UST app hash table here since already registered
1488 * applications can still be controlled so let them be until the session
1489 * daemon dies or the applications stop.
1494 ERR("Health error occurred in %s", __func__
);
1496 health_unregister(health_sessiond
);
1497 DBG("Application communication apps thread cleanup complete");
1498 rcu_thread_offline();
1499 rcu_unregister_thread();
1504 * Send a socket to a thread This is called from the dispatch UST registration
1505 * thread once all sockets are set for the application.
1507 * The sock value can be invalid, we don't really care, the thread will handle
1508 * it and make the necessary cleanup if so.
1510 * On success, return 0 else a negative value being the errno message of the
1513 static int send_socket_to_thread(int fd
, int sock
)
1518 * It's possible that the FD is set as invalid with -1 concurrently just
1519 * before calling this function being a shutdown state of the thread.
1526 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1527 if (ret
< sizeof(sock
)) {
1528 PERROR("write apps pipe %d", fd
);
1535 /* All good. Don't send back the write positive ret value. */
1542 * Sanitize the wait queue of the dispatch registration thread meaning removing
1543 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1544 * notify socket is never received.
1546 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1548 int ret
, nb_fd
= 0, i
;
1549 unsigned int fd_added
= 0;
1550 struct lttng_poll_event events
;
1551 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1555 lttng_poll_init(&events
);
1557 /* Just skip everything for an empty queue. */
1558 if (!wait_queue
->count
) {
1562 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1567 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1568 &wait_queue
->head
, head
) {
1569 assert(wait_node
->app
);
1570 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1571 LPOLLHUP
| LPOLLERR
);
1584 * Poll but don't block so we can quickly identify the faulty events and
1585 * clean them afterwards from the wait queue.
1587 ret
= lttng_poll_wait(&events
, 0);
1593 for (i
= 0; i
< nb_fd
; i
++) {
1594 /* Get faulty FD. */
1595 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1596 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1599 /* No activity for this FD (poll implementation). */
1603 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1604 &wait_queue
->head
, head
) {
1605 if (pollfd
== wait_node
->app
->sock
&&
1606 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1607 cds_list_del(&wait_node
->head
);
1608 wait_queue
->count
--;
1609 ust_app_destroy(wait_node
->app
);
1617 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1621 lttng_poll_clean(&events
);
1625 lttng_poll_clean(&events
);
1627 ERR("Unable to sanitize wait queue");
1632 * Dispatch request from the registration threads to the application
1633 * communication thread.
1635 static void *thread_dispatch_ust_registration(void *data
)
1638 struct cds_wfq_node
*node
;
1639 struct ust_command
*ust_cmd
= NULL
;
1640 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1641 struct ust_reg_wait_queue wait_queue
= {
1645 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1647 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1648 goto error_testpoint
;
1651 health_code_update();
1653 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1655 DBG("[thread] Dispatch UST command started");
1657 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1658 health_code_update();
1660 /* Atomically prepare the queue futex */
1661 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1664 struct ust_app
*app
= NULL
;
1668 * Make sure we don't have node(s) that have hung up before receiving
1669 * the notify socket. This is to clean the list in order to avoid
1670 * memory leaks from notify socket that are never seen.
1672 sanitize_wait_queue(&wait_queue
);
1674 health_code_update();
1675 /* Dequeue command for registration */
1676 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1678 DBG("Woken up but nothing in the UST command queue");
1679 /* Continue thread execution */
1683 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1685 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1686 " gid:%d sock:%d name:%s (version %d.%d)",
1687 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1688 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1689 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1690 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1692 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1693 wait_node
= zmalloc(sizeof(*wait_node
));
1695 PERROR("zmalloc wait_node dispatch");
1696 ret
= close(ust_cmd
->sock
);
1698 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1700 lttng_fd_put(LTTNG_FD_APPS
, 1);
1704 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1706 /* Create application object if socket is CMD. */
1707 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1709 if (!wait_node
->app
) {
1710 ret
= close(ust_cmd
->sock
);
1712 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1714 lttng_fd_put(LTTNG_FD_APPS
, 1);
1720 * Add application to the wait queue so we can set the notify
1721 * socket before putting this object in the global ht.
1723 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1728 * We have to continue here since we don't have the notify
1729 * socket and the application MUST be added to the hash table
1730 * only at that moment.
1735 * Look for the application in the local wait queue and set the
1736 * notify socket if found.
1738 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1739 &wait_queue
.head
, head
) {
1740 health_code_update();
1741 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1742 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1743 cds_list_del(&wait_node
->head
);
1745 app
= wait_node
->app
;
1747 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1753 * With no application at this stage the received socket is
1754 * basically useless so close it before we free the cmd data
1755 * structure for good.
1758 ret
= close(ust_cmd
->sock
);
1760 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1762 lttng_fd_put(LTTNG_FD_APPS
, 1);
1769 * @session_lock_list
1771 * Lock the global session list so from the register up to the
1772 * registration done message, no thread can see the application
1773 * and change its state.
1775 session_lock_list();
1779 * Add application to the global hash table. This needs to be
1780 * done before the update to the UST registry can locate the
1785 /* Set app version. This call will print an error if needed. */
1786 (void) ust_app_version(app
);
1788 /* Send notify socket through the notify pipe. */
1789 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1793 session_unlock_list();
1795 * No notify thread, stop the UST tracing. However, this is
1796 * not an internal error of the this thread thus setting
1797 * the health error code to a normal exit.
1804 * Update newly registered application with the tracing
1805 * registry info already enabled information.
1807 update_ust_app(app
->sock
);
1810 * Don't care about return value. Let the manage apps threads
1811 * handle app unregistration upon socket close.
1813 (void) ust_app_register_done(app
->sock
);
1816 * Even if the application socket has been closed, send the app
1817 * to the thread and unregistration will take place at that
1820 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1823 session_unlock_list();
1825 * No apps. thread, stop the UST tracing. However, this is
1826 * not an internal error of the this thread thus setting
1827 * the health error code to a normal exit.
1834 session_unlock_list();
1836 } while (node
!= NULL
);
1838 health_poll_entry();
1839 /* Futex wait on queue. Blocking call on futex() */
1840 futex_nto1_wait(&ust_cmd_queue
.futex
);
1843 /* Normal exit, no error */
1847 /* Clean up wait queue. */
1848 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1849 &wait_queue
.head
, head
) {
1850 cds_list_del(&wait_node
->head
);
1856 DBG("Dispatch thread dying");
1859 ERR("Health error occurred in %s", __func__
);
1861 health_unregister(health_sessiond
);
1866 * This thread manage application registration.
1868 static void *thread_registration_apps(void *data
)
1870 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1871 uint32_t revents
, nb_fd
;
1872 struct lttng_poll_event events
;
1874 * Get allocated in this thread, enqueued to a global queue, dequeued and
1875 * freed in the manage apps thread.
1877 struct ust_command
*ust_cmd
= NULL
;
1879 DBG("[thread] Manage application registration started");
1881 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1883 if (testpoint(sessiond_thread_registration_apps
)) {
1884 goto error_testpoint
;
1887 ret
= lttcomm_listen_unix_sock(apps_sock
);
1893 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1894 * more will be added to this poll set.
1896 ret
= sessiond_set_thread_pollset(&events
, 2);
1898 goto error_create_poll
;
1901 /* Add the application registration socket */
1902 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1904 goto error_poll_add
;
1907 /* Notify all applications to register */
1908 ret
= notify_ust_apps(1);
1910 ERR("Failed to notify applications or create the wait shared memory.\n"
1911 "Execution continues but there might be problem for already\n"
1912 "running applications that wishes to register.");
1916 DBG("Accepting application registration");
1918 /* Inifinite blocking call, waiting for transmission */
1920 health_poll_entry();
1921 ret
= lttng_poll_wait(&events
, -1);
1925 * Restart interrupted system call.
1927 if (errno
== EINTR
) {
1935 for (i
= 0; i
< nb_fd
; i
++) {
1936 health_code_update();
1938 /* Fetch once the poll data */
1939 revents
= LTTNG_POLL_GETEV(&events
, i
);
1940 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1943 /* No activity for this FD (poll implementation). */
1947 /* Thread quit pipe has been closed. Killing thread. */
1948 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1954 /* Event on the registration socket */
1955 if (pollfd
== apps_sock
) {
1956 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1957 ERR("Register apps socket poll error");
1959 } else if (revents
& LPOLLIN
) {
1960 sock
= lttcomm_accept_unix_sock(apps_sock
);
1966 * Set socket timeout for both receiving and ending.
1967 * app_socket_timeout is in seconds, whereas
1968 * lttcomm_setsockopt_rcv_timeout and
1969 * lttcomm_setsockopt_snd_timeout expect msec as
1972 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1973 app_socket_timeout
* 1000);
1974 (void) lttcomm_setsockopt_snd_timeout(sock
,
1975 app_socket_timeout
* 1000);
1978 * Set the CLOEXEC flag. Return code is useless because
1979 * either way, the show must go on.
1981 (void) utils_set_fd_cloexec(sock
);
1983 /* Create UST registration command for enqueuing */
1984 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1985 if (ust_cmd
== NULL
) {
1986 PERROR("ust command zmalloc");
1991 * Using message-based transmissions to ensure we don't
1992 * have to deal with partially received messages.
1994 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1996 ERR("Exhausted file descriptors allowed for applications.");
2006 health_code_update();
2007 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2010 /* Close socket of the application. */
2015 lttng_fd_put(LTTNG_FD_APPS
, 1);
2019 health_code_update();
2021 ust_cmd
->sock
= sock
;
2024 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2025 " gid:%d sock:%d name:%s (version %d.%d)",
2026 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2027 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2028 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2029 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2032 * Lock free enqueue the registration request. The red pill
2033 * has been taken! This apps will be part of the *system*.
2035 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
2038 * Wake the registration queue futex. Implicit memory
2039 * barrier with the exchange in cds_wfq_enqueue.
2041 futex_nto1_wake(&ust_cmd_queue
.futex
);
2049 /* Notify that the registration thread is gone */
2052 if (apps_sock
>= 0) {
2053 ret
= close(apps_sock
);
2063 lttng_fd_put(LTTNG_FD_APPS
, 1);
2065 unlink(apps_unix_sock_path
);
2068 lttng_poll_clean(&events
);
2072 DBG("UST Registration thread cleanup complete");
2075 ERR("Health error occurred in %s", __func__
);
2077 health_unregister(health_sessiond
);
2083 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2084 * exec or it will fails.
2086 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2089 struct timespec timeout
;
2091 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2092 consumer_data
->consumer_thread_is_ready
= 0;
2094 /* Setup pthread condition */
2095 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2098 PERROR("pthread_condattr_init consumer data");
2103 * Set the monotonic clock in order to make sure we DO NOT jump in time
2104 * between the clock_gettime() call and the timedwait call. See bug #324
2105 * for a more details and how we noticed it.
2107 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2110 PERROR("pthread_condattr_setclock consumer data");
2114 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2117 PERROR("pthread_cond_init consumer data");
2121 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2124 PERROR("pthread_create consumer");
2129 /* We are about to wait on a pthread condition */
2130 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2132 /* Get time for sem_timedwait absolute timeout */
2133 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2135 * Set the timeout for the condition timed wait even if the clock gettime
2136 * call fails since we might loop on that call and we want to avoid to
2137 * increment the timeout too many times.
2139 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2142 * The following loop COULD be skipped in some conditions so this is why we
2143 * set ret to 0 in order to make sure at least one round of the loop is
2149 * Loop until the condition is reached or when a timeout is reached. Note
2150 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2151 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2152 * possible. This loop does not take any chances and works with both of
2155 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2156 if (clock_ret
< 0) {
2157 PERROR("clock_gettime spawn consumer");
2158 /* Infinite wait for the consumerd thread to be ready */
2159 ret
= pthread_cond_wait(&consumer_data
->cond
,
2160 &consumer_data
->cond_mutex
);
2162 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2163 &consumer_data
->cond_mutex
, &timeout
);
2167 /* Release the pthread condition */
2168 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2172 if (ret
== ETIMEDOUT
) {
2176 * Call has timed out so we kill the kconsumerd_thread and return
2179 ERR("Condition timed out. The consumer thread was never ready."
2181 pth_ret
= pthread_cancel(consumer_data
->thread
);
2183 PERROR("pthread_cancel consumer thread");
2186 PERROR("pthread_cond_wait failed consumer thread");
2188 /* Caller is expecting a negative value on failure. */
2193 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2194 if (consumer_data
->pid
== 0) {
2195 ERR("Consumerd did not start");
2196 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2199 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2208 * Join consumer thread
2210 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2214 /* Consumer pid must be a real one. */
2215 if (consumer_data
->pid
> 0) {
2217 ret
= kill(consumer_data
->pid
, SIGTERM
);
2219 ERR("Error killing consumer daemon");
2222 return pthread_join(consumer_data
->thread
, &status
);
2229 * Fork and exec a consumer daemon (consumerd).
2231 * Return pid if successful else -1.
2233 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2237 const char *consumer_to_use
;
2238 const char *verbosity
;
2241 DBG("Spawning consumerd");
2248 if (opt_verbose_consumer
) {
2249 verbosity
= "--verbose";
2251 verbosity
= "--quiet";
2253 switch (consumer_data
->type
) {
2254 case LTTNG_CONSUMER_KERNEL
:
2256 * Find out which consumerd to execute. We will first try the
2257 * 64-bit path, then the sessiond's installation directory, and
2258 * fallback on the 32-bit one,
2260 DBG3("Looking for a kernel consumer at these locations:");
2261 DBG3(" 1) %s", consumerd64_bin
);
2262 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2263 DBG3(" 3) %s", consumerd32_bin
);
2264 if (stat(consumerd64_bin
, &st
) == 0) {
2265 DBG3("Found location #1");
2266 consumer_to_use
= consumerd64_bin
;
2267 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2268 DBG3("Found location #2");
2269 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2270 } else if (stat(consumerd32_bin
, &st
) == 0) {
2271 DBG3("Found location #3");
2272 consumer_to_use
= consumerd32_bin
;
2274 DBG("Could not find any valid consumerd executable");
2278 DBG("Using kernel consumer at: %s", consumer_to_use
);
2279 ret
= execl(consumer_to_use
,
2280 "lttng-consumerd", verbosity
, "-k",
2281 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2282 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2283 "--group", tracing_group_name
,
2286 case LTTNG_CONSUMER64_UST
:
2288 char *tmpnew
= NULL
;
2290 if (consumerd64_libdir
[0] != '\0') {
2294 tmp
= getenv("LD_LIBRARY_PATH");
2298 tmplen
= strlen("LD_LIBRARY_PATH=")
2299 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2300 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2305 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2306 strcat(tmpnew
, consumerd64_libdir
);
2307 if (tmp
[0] != '\0') {
2308 strcat(tmpnew
, ":");
2309 strcat(tmpnew
, tmp
);
2311 ret
= putenv(tmpnew
);
2318 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2319 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2320 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2321 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2322 "--group", tracing_group_name
,
2324 if (consumerd64_libdir
[0] != '\0') {
2329 case LTTNG_CONSUMER32_UST
:
2331 char *tmpnew
= NULL
;
2333 if (consumerd32_libdir
[0] != '\0') {
2337 tmp
= getenv("LD_LIBRARY_PATH");
2341 tmplen
= strlen("LD_LIBRARY_PATH=")
2342 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2343 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2348 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2349 strcat(tmpnew
, consumerd32_libdir
);
2350 if (tmp
[0] != '\0') {
2351 strcat(tmpnew
, ":");
2352 strcat(tmpnew
, tmp
);
2354 ret
= putenv(tmpnew
);
2361 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2362 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2363 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2364 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2365 "--group", tracing_group_name
,
2367 if (consumerd32_libdir
[0] != '\0') {
2373 PERROR("unknown consumer type");
2377 PERROR("Consumer execl()");
2379 /* Reaching this point, we got a failure on our execl(). */
2381 } else if (pid
> 0) {
2384 PERROR("start consumer fork");
2392 * Spawn the consumerd daemon and session daemon thread.
2394 static int start_consumerd(struct consumer_data
*consumer_data
)
2399 * Set the listen() state on the socket since there is a possible race
2400 * between the exec() of the consumer daemon and this call if place in the
2401 * consumer thread. See bug #366 for more details.
2403 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2408 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2409 if (consumer_data
->pid
!= 0) {
2410 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2414 ret
= spawn_consumerd(consumer_data
);
2416 ERR("Spawning consumerd failed");
2417 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2421 /* Setting up the consumer_data pid */
2422 consumer_data
->pid
= ret
;
2423 DBG2("Consumer pid %d", consumer_data
->pid
);
2424 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2426 DBG2("Spawning consumer control thread");
2427 ret
= spawn_consumer_thread(consumer_data
);
2429 ERR("Fatal error spawning consumer control thread");
2437 /* Cleanup already created sockets on error. */
2438 if (consumer_data
->err_sock
>= 0) {
2441 err
= close(consumer_data
->err_sock
);
2443 PERROR("close consumer data error socket");
2450 * Setup necessary data for kernel tracer action.
2452 static int init_kernel_tracer(void)
2456 /* Modprobe lttng kernel modules */
2457 ret
= modprobe_lttng_control();
2462 /* Open debugfs lttng */
2463 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2464 if (kernel_tracer_fd
< 0) {
2465 DBG("Failed to open %s", module_proc_lttng
);
2470 /* Validate kernel version */
2471 ret
= kernel_validate_version(kernel_tracer_fd
);
2476 ret
= modprobe_lttng_data();
2481 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2485 modprobe_remove_lttng_control();
2486 ret
= close(kernel_tracer_fd
);
2490 kernel_tracer_fd
= -1;
2491 return LTTNG_ERR_KERN_VERSION
;
2494 ret
= close(kernel_tracer_fd
);
2500 modprobe_remove_lttng_control();
2503 WARN("No kernel tracer available");
2504 kernel_tracer_fd
= -1;
2506 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2508 return LTTNG_ERR_KERN_NA
;
2514 * Copy consumer output from the tracing session to the domain session. The
2515 * function also applies the right modification on a per domain basis for the
2516 * trace files destination directory.
2518 * Should *NOT* be called with RCU read-side lock held.
2520 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2523 const char *dir_name
;
2524 struct consumer_output
*consumer
;
2527 assert(session
->consumer
);
2530 case LTTNG_DOMAIN_KERNEL
:
2531 DBG3("Copying tracing session consumer output in kernel session");
2533 * XXX: We should audit the session creation and what this function
2534 * does "extra" in order to avoid a destroy since this function is used
2535 * in the domain session creation (kernel and ust) only. Same for UST
2538 if (session
->kernel_session
->consumer
) {
2539 consumer_destroy_output(session
->kernel_session
->consumer
);
2541 session
->kernel_session
->consumer
=
2542 consumer_copy_output(session
->consumer
);
2543 /* Ease our life a bit for the next part */
2544 consumer
= session
->kernel_session
->consumer
;
2545 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2547 case LTTNG_DOMAIN_JUL
:
2548 case LTTNG_DOMAIN_UST
:
2549 DBG3("Copying tracing session consumer output in UST session");
2550 if (session
->ust_session
->consumer
) {
2551 consumer_destroy_output(session
->ust_session
->consumer
);
2553 session
->ust_session
->consumer
=
2554 consumer_copy_output(session
->consumer
);
2555 /* Ease our life a bit for the next part */
2556 consumer
= session
->ust_session
->consumer
;
2557 dir_name
= DEFAULT_UST_TRACE_DIR
;
2560 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2564 /* Append correct directory to subdir */
2565 strncat(consumer
->subdir
, dir_name
,
2566 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2567 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2576 * Create an UST session and add it to the session ust list.
2578 * Should *NOT* be called with RCU read-side lock held.
2580 static int create_ust_session(struct ltt_session
*session
,
2581 struct lttng_domain
*domain
)
2584 struct ltt_ust_session
*lus
= NULL
;
2588 assert(session
->consumer
);
2590 switch (domain
->type
) {
2591 case LTTNG_DOMAIN_JUL
:
2592 case LTTNG_DOMAIN_UST
:
2595 ERR("Unknown UST domain on create session %d", domain
->type
);
2596 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2600 DBG("Creating UST session");
2602 lus
= trace_ust_create_session(session
->id
);
2604 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2608 lus
->uid
= session
->uid
;
2609 lus
->gid
= session
->gid
;
2610 lus
->output_traces
= session
->output_traces
;
2611 lus
->snapshot_mode
= session
->snapshot_mode
;
2612 lus
->live_timer_interval
= session
->live_timer
;
2613 session
->ust_session
= lus
;
2615 /* Copy session output to the newly created UST session */
2616 ret
= copy_session_consumer(domain
->type
, session
);
2617 if (ret
!= LTTNG_OK
) {
2625 session
->ust_session
= NULL
;
2630 * Create a kernel tracer session then create the default channel.
2632 static int create_kernel_session(struct ltt_session
*session
)
2636 DBG("Creating kernel session");
2638 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2640 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2644 /* Code flow safety */
2645 assert(session
->kernel_session
);
2647 /* Copy session output to the newly created Kernel session */
2648 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2649 if (ret
!= LTTNG_OK
) {
2653 /* Create directory(ies) on local filesystem. */
2654 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2655 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2656 ret
= run_as_mkdir_recursive(
2657 session
->kernel_session
->consumer
->dst
.trace_path
,
2658 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2660 if (ret
!= -EEXIST
) {
2661 ERR("Trace directory creation error");
2667 session
->kernel_session
->uid
= session
->uid
;
2668 session
->kernel_session
->gid
= session
->gid
;
2669 session
->kernel_session
->output_traces
= session
->output_traces
;
2670 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2675 trace_kernel_destroy_session(session
->kernel_session
);
2676 session
->kernel_session
= NULL
;
2681 * Count number of session permitted by uid/gid.
2683 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2686 struct ltt_session
*session
;
2688 DBG("Counting number of available session for UID %d GID %d",
2690 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2692 * Only list the sessions the user can control.
2694 if (!session_access_ok(session
, uid
, gid
)) {
2703 * Process the command requested by the lttng client within the command
2704 * context structure. This function make sure that the return structure (llm)
2705 * is set and ready for transmission before returning.
2707 * Return any error encountered or 0 for success.
2709 * "sock" is only used for special-case var. len data.
2711 * Should *NOT* be called with RCU read-side lock held.
2713 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2717 int need_tracing_session
= 1;
2720 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2724 switch (cmd_ctx
->lsm
->cmd_type
) {
2725 case LTTNG_CREATE_SESSION
:
2726 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2727 case LTTNG_CREATE_SESSION_LIVE
:
2728 case LTTNG_DESTROY_SESSION
:
2729 case LTTNG_LIST_SESSIONS
:
2730 case LTTNG_LIST_DOMAINS
:
2731 case LTTNG_START_TRACE
:
2732 case LTTNG_STOP_TRACE
:
2733 case LTTNG_DATA_PENDING
:
2734 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2735 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2736 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2737 case LTTNG_SNAPSHOT_RECORD
:
2744 if (opt_no_kernel
&& need_domain
2745 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2747 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2749 ret
= LTTNG_ERR_KERN_NA
;
2754 /* Deny register consumer if we already have a spawned consumer. */
2755 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2756 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2757 if (kconsumer_data
.pid
> 0) {
2758 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2759 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2762 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2766 * Check for command that don't needs to allocate a returned payload. We do
2767 * this here so we don't have to make the call for no payload at each
2770 switch(cmd_ctx
->lsm
->cmd_type
) {
2771 case LTTNG_LIST_SESSIONS
:
2772 case LTTNG_LIST_TRACEPOINTS
:
2773 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2774 case LTTNG_LIST_DOMAINS
:
2775 case LTTNG_LIST_CHANNELS
:
2776 case LTTNG_LIST_EVENTS
:
2779 /* Setup lttng message with no payload */
2780 ret
= setup_lttng_msg(cmd_ctx
, 0);
2782 /* This label does not try to unlock the session */
2783 goto init_setup_error
;
2787 /* Commands that DO NOT need a session. */
2788 switch (cmd_ctx
->lsm
->cmd_type
) {
2789 case LTTNG_CREATE_SESSION
:
2790 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2791 case LTTNG_CREATE_SESSION_LIVE
:
2792 case LTTNG_CALIBRATE
:
2793 case LTTNG_LIST_SESSIONS
:
2794 case LTTNG_LIST_TRACEPOINTS
:
2795 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2796 need_tracing_session
= 0;
2799 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2801 * We keep the session list lock across _all_ commands
2802 * for now, because the per-session lock does not
2803 * handle teardown properly.
2805 session_lock_list();
2806 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2807 if (cmd_ctx
->session
== NULL
) {
2808 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2811 /* Acquire lock for the session */
2812 session_lock(cmd_ctx
->session
);
2822 * Check domain type for specific "pre-action".
2824 switch (cmd_ctx
->lsm
->domain
.type
) {
2825 case LTTNG_DOMAIN_KERNEL
:
2827 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2831 /* Kernel tracer check */
2832 if (kernel_tracer_fd
== -1) {
2833 /* Basically, load kernel tracer modules */
2834 ret
= init_kernel_tracer();
2840 /* Consumer is in an ERROR state. Report back to client */
2841 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2842 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2846 /* Need a session for kernel command */
2847 if (need_tracing_session
) {
2848 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2849 ret
= create_kernel_session(cmd_ctx
->session
);
2851 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2856 /* Start the kernel consumer daemon */
2857 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2858 if (kconsumer_data
.pid
== 0 &&
2859 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2860 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2861 ret
= start_consumerd(&kconsumer_data
);
2863 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2866 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2868 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2872 * The consumer was just spawned so we need to add the socket to
2873 * the consumer output of the session if exist.
2875 ret
= consumer_create_socket(&kconsumer_data
,
2876 cmd_ctx
->session
->kernel_session
->consumer
);
2883 case LTTNG_DOMAIN_JUL
:
2884 case LTTNG_DOMAIN_UST
:
2886 if (!ust_app_supported()) {
2887 ret
= LTTNG_ERR_NO_UST
;
2890 /* Consumer is in an ERROR state. Report back to client */
2891 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2892 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2896 if (need_tracing_session
) {
2897 /* Create UST session if none exist. */
2898 if (cmd_ctx
->session
->ust_session
== NULL
) {
2899 ret
= create_ust_session(cmd_ctx
->session
,
2900 &cmd_ctx
->lsm
->domain
);
2901 if (ret
!= LTTNG_OK
) {
2906 /* Start the UST consumer daemons */
2908 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2909 if (consumerd64_bin
[0] != '\0' &&
2910 ustconsumer64_data
.pid
== 0 &&
2911 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2912 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2913 ret
= start_consumerd(&ustconsumer64_data
);
2915 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2916 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2920 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2921 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2923 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2927 * Setup socket for consumer 64 bit. No need for atomic access
2928 * since it was set above and can ONLY be set in this thread.
2930 ret
= consumer_create_socket(&ustconsumer64_data
,
2931 cmd_ctx
->session
->ust_session
->consumer
);
2937 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
2938 if (consumerd32_bin
[0] != '\0' &&
2939 ustconsumer32_data
.pid
== 0 &&
2940 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2941 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2942 ret
= start_consumerd(&ustconsumer32_data
);
2944 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2945 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2949 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2950 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2952 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2956 * Setup socket for consumer 64 bit. No need for atomic access
2957 * since it was set above and can ONLY be set in this thread.
2959 ret
= consumer_create_socket(&ustconsumer32_data
,
2960 cmd_ctx
->session
->ust_session
->consumer
);
2972 /* Validate consumer daemon state when start/stop trace command */
2973 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2974 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2975 switch (cmd_ctx
->lsm
->domain
.type
) {
2976 case LTTNG_DOMAIN_JUL
:
2977 case LTTNG_DOMAIN_UST
:
2978 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2979 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2983 case LTTNG_DOMAIN_KERNEL
:
2984 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2985 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2993 * Check that the UID or GID match that of the tracing session.
2994 * The root user can interact with all sessions.
2996 if (need_tracing_session
) {
2997 if (!session_access_ok(cmd_ctx
->session
,
2998 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2999 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3000 ret
= LTTNG_ERR_EPERM
;
3006 * Send relayd information to consumer as soon as we have a domain and a
3009 if (cmd_ctx
->session
&& need_domain
) {
3011 * Setup relayd if not done yet. If the relayd information was already
3012 * sent to the consumer, this call will gracefully return.
3014 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3015 if (ret
!= LTTNG_OK
) {
3020 /* Process by command type */
3021 switch (cmd_ctx
->lsm
->cmd_type
) {
3022 case LTTNG_ADD_CONTEXT
:
3024 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3025 cmd_ctx
->lsm
->u
.context
.channel_name
,
3026 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3029 case LTTNG_DISABLE_CHANNEL
:
3031 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3032 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3035 case LTTNG_DISABLE_EVENT
:
3037 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3038 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3039 cmd_ctx
->lsm
->u
.disable
.name
);
3042 case LTTNG_DISABLE_ALL_EVENT
:
3044 DBG("Disabling all events");
3046 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3047 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3050 case LTTNG_ENABLE_CHANNEL
:
3052 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3053 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3056 case LTTNG_ENABLE_EVENT
:
3058 struct lttng_event_exclusion
*exclusion
= NULL
;
3059 struct lttng_filter_bytecode
*bytecode
= NULL
;
3061 /* Handle exclusion events and receive it from the client. */
3062 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3063 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3065 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3066 (count
* LTTNG_SYMBOL_NAME_LEN
));
3068 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3072 DBG("Receiving var len exclusion event list from client ...");
3073 exclusion
->count
= count
;
3074 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3075 count
* LTTNG_SYMBOL_NAME_LEN
);
3077 DBG("Nothing recv() from client var len data... continuing");
3080 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3085 /* Handle filter and get bytecode from client. */
3086 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3087 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3089 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3090 ret
= LTTNG_ERR_FILTER_INVAL
;
3095 bytecode
= zmalloc(bytecode_len
);
3098 ret
= LTTNG_ERR_FILTER_NOMEM
;
3102 /* Receive var. len. data */
3103 DBG("Receiving var len filter's bytecode from client ...");
3104 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3106 DBG("Nothing recv() from client car len data... continuing");
3110 ret
= LTTNG_ERR_FILTER_INVAL
;
3114 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3117 ret
= LTTNG_ERR_FILTER_INVAL
;
3122 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3123 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3124 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, exclusion
,
3125 kernel_poll_pipe
[1]);
3128 case LTTNG_ENABLE_ALL_EVENT
:
3130 DBG("Enabling all events");
3132 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3133 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3134 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
3137 case LTTNG_LIST_TRACEPOINTS
:
3139 struct lttng_event
*events
;
3142 session_lock_list();
3143 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3144 session_unlock_list();
3145 if (nb_events
< 0) {
3146 /* Return value is a negative lttng_error_code. */
3152 * Setup lttng message with payload size set to the event list size in
3153 * bytes and then copy list into the llm payload.
3155 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3161 /* Copy event list into message payload */
3162 memcpy(cmd_ctx
->llm
->payload
, events
,
3163 sizeof(struct lttng_event
) * nb_events
);
3170 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3172 struct lttng_event_field
*fields
;
3175 session_lock_list();
3176 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3178 session_unlock_list();
3179 if (nb_fields
< 0) {
3180 /* Return value is a negative lttng_error_code. */
3186 * Setup lttng message with payload size set to the event list size in
3187 * bytes and then copy list into the llm payload.
3189 ret
= setup_lttng_msg(cmd_ctx
,
3190 sizeof(struct lttng_event_field
) * nb_fields
);
3196 /* Copy event list into message payload */
3197 memcpy(cmd_ctx
->llm
->payload
, fields
,
3198 sizeof(struct lttng_event_field
) * nb_fields
);
3205 case LTTNG_SET_CONSUMER_URI
:
3208 struct lttng_uri
*uris
;
3210 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3211 len
= nb_uri
* sizeof(struct lttng_uri
);
3214 ret
= LTTNG_ERR_INVALID
;
3218 uris
= zmalloc(len
);
3220 ret
= LTTNG_ERR_FATAL
;
3224 /* Receive variable len data */
3225 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3226 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3228 DBG("No URIs received from client... continuing");
3230 ret
= LTTNG_ERR_SESSION_FAIL
;
3235 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3237 if (ret
!= LTTNG_OK
) {
3243 * XXX: 0 means that this URI should be applied on the session. Should
3244 * be a DOMAIN enuam.
3246 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3247 /* Add the URI for the UST session if a consumer is present. */
3248 if (cmd_ctx
->session
->ust_session
&&
3249 cmd_ctx
->session
->ust_session
->consumer
) {
3250 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3252 } else if (cmd_ctx
->session
->kernel_session
&&
3253 cmd_ctx
->session
->kernel_session
->consumer
) {
3254 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3255 cmd_ctx
->session
, nb_uri
, uris
);
3263 case LTTNG_START_TRACE
:
3265 ret
= cmd_start_trace(cmd_ctx
->session
);
3268 case LTTNG_STOP_TRACE
:
3270 ret
= cmd_stop_trace(cmd_ctx
->session
);
3273 case LTTNG_CREATE_SESSION
:
3276 struct lttng_uri
*uris
= NULL
;
3278 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3279 len
= nb_uri
* sizeof(struct lttng_uri
);
3282 uris
= zmalloc(len
);
3284 ret
= LTTNG_ERR_FATAL
;
3288 /* Receive variable len data */
3289 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3290 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3292 DBG("No URIs received from client... continuing");
3294 ret
= LTTNG_ERR_SESSION_FAIL
;
3299 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3300 DBG("Creating session with ONE network URI is a bad call");
3301 ret
= LTTNG_ERR_SESSION_FAIL
;
3307 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3308 &cmd_ctx
->creds
, 0);
3314 case LTTNG_DESTROY_SESSION
:
3316 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3318 /* Set session to NULL so we do not unlock it after free. */
3319 cmd_ctx
->session
= NULL
;
3322 case LTTNG_LIST_DOMAINS
:
3325 struct lttng_domain
*domains
;
3327 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3329 /* Return value is a negative lttng_error_code. */
3334 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3340 /* Copy event list into message payload */
3341 memcpy(cmd_ctx
->llm
->payload
, domains
,
3342 nb_dom
* sizeof(struct lttng_domain
));
3349 case LTTNG_LIST_CHANNELS
:
3352 struct lttng_channel
*channels
= NULL
;
3354 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3355 cmd_ctx
->session
, &channels
);
3357 /* Return value is a negative lttng_error_code. */
3362 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3368 /* Copy event list into message payload */
3369 memcpy(cmd_ctx
->llm
->payload
, channels
,
3370 nb_chan
* sizeof(struct lttng_channel
));
3377 case LTTNG_LIST_EVENTS
:
3380 struct lttng_event
*events
= NULL
;
3382 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3383 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3385 /* Return value is a negative lttng_error_code. */
3390 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3396 /* Copy event list into message payload */
3397 memcpy(cmd_ctx
->llm
->payload
, events
,
3398 nb_event
* sizeof(struct lttng_event
));
3405 case LTTNG_LIST_SESSIONS
:
3407 unsigned int nr_sessions
;
3409 session_lock_list();
3410 nr_sessions
= lttng_sessions_count(
3411 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3412 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3414 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3416 session_unlock_list();
3420 /* Filled the session array */
3421 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3422 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3423 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3425 session_unlock_list();
3430 case LTTNG_CALIBRATE
:
3432 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3433 &cmd_ctx
->lsm
->u
.calibrate
);
3436 case LTTNG_REGISTER_CONSUMER
:
3438 struct consumer_data
*cdata
;
3440 switch (cmd_ctx
->lsm
->domain
.type
) {
3441 case LTTNG_DOMAIN_KERNEL
:
3442 cdata
= &kconsumer_data
;
3445 ret
= LTTNG_ERR_UND
;
3449 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3450 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3453 case LTTNG_DATA_PENDING
:
3455 ret
= cmd_data_pending(cmd_ctx
->session
);
3458 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3460 struct lttcomm_lttng_output_id reply
;
3462 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3463 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3464 if (ret
!= LTTNG_OK
) {
3468 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3473 /* Copy output list into message payload */
3474 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3478 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3480 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3481 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3484 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3487 struct lttng_snapshot_output
*outputs
= NULL
;
3489 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3490 if (nb_output
< 0) {
3495 ret
= setup_lttng_msg(cmd_ctx
,
3496 nb_output
* sizeof(struct lttng_snapshot_output
));
3503 /* Copy output list into message payload */
3504 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3505 nb_output
* sizeof(struct lttng_snapshot_output
));
3512 case LTTNG_SNAPSHOT_RECORD
:
3514 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3515 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3516 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3519 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3522 struct lttng_uri
*uris
= NULL
;
3524 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3525 len
= nb_uri
* sizeof(struct lttng_uri
);
3528 uris
= zmalloc(len
);
3530 ret
= LTTNG_ERR_FATAL
;
3534 /* Receive variable len data */
3535 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3536 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3538 DBG("No URIs received from client... continuing");
3540 ret
= LTTNG_ERR_SESSION_FAIL
;
3545 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3546 DBG("Creating session with ONE network URI is a bad call");
3547 ret
= LTTNG_ERR_SESSION_FAIL
;
3553 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3554 nb_uri
, &cmd_ctx
->creds
);
3558 case LTTNG_CREATE_SESSION_LIVE
:
3561 struct lttng_uri
*uris
= NULL
;
3563 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3564 len
= nb_uri
* sizeof(struct lttng_uri
);
3567 uris
= zmalloc(len
);
3569 ret
= LTTNG_ERR_FATAL
;
3573 /* Receive variable len data */
3574 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3575 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3577 DBG("No URIs received from client... continuing");
3579 ret
= LTTNG_ERR_SESSION_FAIL
;
3584 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3585 DBG("Creating session with ONE network URI is a bad call");
3586 ret
= LTTNG_ERR_SESSION_FAIL
;
3592 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3593 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3598 ret
= LTTNG_ERR_UND
;
3603 if (cmd_ctx
->llm
== NULL
) {
3604 DBG("Missing llm structure. Allocating one.");
3605 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3609 /* Set return code */
3610 cmd_ctx
->llm
->ret_code
= ret
;
3612 if (cmd_ctx
->session
) {
3613 session_unlock(cmd_ctx
->session
);
3615 if (need_tracing_session
) {
3616 session_unlock_list();
3623 * Thread managing health check socket.
3625 static void *thread_manage_health(void *data
)
3627 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3628 uint32_t revents
, nb_fd
;
3629 struct lttng_poll_event events
;
3630 struct health_comm_msg msg
;
3631 struct health_comm_reply reply
;
3633 DBG("[thread] Manage health check started");
3635 rcu_register_thread();
3637 /* We might hit an error path before this is created. */
3638 lttng_poll_init(&events
);
3640 /* Create unix socket */
3641 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3643 ERR("Unable to create health check Unix socket");
3649 /* lttng health client socket path permissions */
3650 ret
= chown(health_unix_sock_path
, 0,
3651 utils_get_group_id(tracing_group_name
));
3653 ERR("Unable to set group on %s", health_unix_sock_path
);
3659 ret
= chmod(health_unix_sock_path
,
3660 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3662 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3670 * Set the CLOEXEC flag. Return code is useless because either way, the
3673 (void) utils_set_fd_cloexec(sock
);
3675 ret
= lttcomm_listen_unix_sock(sock
);
3681 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3682 * more will be added to this poll set.
3684 ret
= sessiond_set_thread_pollset(&events
, 2);
3689 /* Add the application registration socket */
3690 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3695 lttng_sessiond_notify_ready();
3698 DBG("Health check ready");
3700 /* Inifinite blocking call, waiting for transmission */
3702 ret
= lttng_poll_wait(&events
, -1);
3705 * Restart interrupted system call.
3707 if (errno
== EINTR
) {
3715 for (i
= 0; i
< nb_fd
; i
++) {
3716 /* Fetch once the poll data */
3717 revents
= LTTNG_POLL_GETEV(&events
, i
);
3718 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3721 /* No activity for this FD (poll implementation). */
3725 /* Thread quit pipe has been closed. Killing thread. */
3726 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3732 /* Event on the registration socket */
3733 if (pollfd
== sock
) {
3734 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3735 ERR("Health socket poll error");
3741 new_sock
= lttcomm_accept_unix_sock(sock
);
3747 * Set the CLOEXEC flag. Return code is useless because either way, the
3750 (void) utils_set_fd_cloexec(new_sock
);
3752 DBG("Receiving data from client for health...");
3753 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3755 DBG("Nothing recv() from client... continuing");
3756 ret
= close(new_sock
);
3764 rcu_thread_online();
3766 memset(&reply
, 0, sizeof(reply
));
3767 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3769 * health_check_state returns 0 if health is
3772 if (!health_check_state(health_sessiond
, i
)) {
3773 reply
.ret_code
|= 1ULL << i
;
3777 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3779 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3781 ERR("Failed to send health data back to client");
3784 /* End of transmission */
3785 ret
= close(new_sock
);
3795 ERR("Health error occurred in %s", __func__
);
3797 DBG("Health check thread dying");
3798 unlink(health_unix_sock_path
);
3806 lttng_poll_clean(&events
);
3808 rcu_unregister_thread();
3813 * This thread manage all clients request using the unix client socket for
3816 static void *thread_manage_clients(void *data
)
3818 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3820 uint32_t revents
, nb_fd
;
3821 struct command_ctx
*cmd_ctx
= NULL
;
3822 struct lttng_poll_event events
;
3824 DBG("[thread] Manage client started");
3826 rcu_register_thread();
3828 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3830 health_code_update();
3832 ret
= lttcomm_listen_unix_sock(client_sock
);
3838 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3839 * more will be added to this poll set.
3841 ret
= sessiond_set_thread_pollset(&events
, 2);
3843 goto error_create_poll
;
3846 /* Add the application registration socket */
3847 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3852 lttng_sessiond_notify_ready();
3854 /* This testpoint is after we signal readiness to the parent. */
3855 if (testpoint(sessiond_thread_manage_clients
)) {
3859 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
3863 health_code_update();
3866 DBG("Accepting client command ...");
3868 /* Inifinite blocking call, waiting for transmission */
3870 health_poll_entry();
3871 ret
= lttng_poll_wait(&events
, -1);
3875 * Restart interrupted system call.
3877 if (errno
== EINTR
) {
3885 for (i
= 0; i
< nb_fd
; i
++) {
3886 /* Fetch once the poll data */
3887 revents
= LTTNG_POLL_GETEV(&events
, i
);
3888 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3890 health_code_update();
3893 /* No activity for this FD (poll implementation). */
3897 /* Thread quit pipe has been closed. Killing thread. */
3898 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3904 /* Event on the registration socket */
3905 if (pollfd
== client_sock
) {
3906 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3907 ERR("Client socket poll error");
3913 DBG("Wait for client response");
3915 health_code_update();
3917 sock
= lttcomm_accept_unix_sock(client_sock
);
3923 * Set the CLOEXEC flag. Return code is useless because either way, the
3926 (void) utils_set_fd_cloexec(sock
);
3928 /* Set socket option for credentials retrieval */
3929 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3934 /* Allocate context command to process the client request */
3935 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3936 if (cmd_ctx
== NULL
) {
3937 PERROR("zmalloc cmd_ctx");
3941 /* Allocate data buffer for reception */
3942 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3943 if (cmd_ctx
->lsm
== NULL
) {
3944 PERROR("zmalloc cmd_ctx->lsm");
3948 cmd_ctx
->llm
= NULL
;
3949 cmd_ctx
->session
= NULL
;
3951 health_code_update();
3954 * Data is received from the lttng client. The struct
3955 * lttcomm_session_msg (lsm) contains the command and data request of
3958 DBG("Receiving data from client ...");
3959 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3960 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3962 DBG("Nothing recv() from client... continuing");
3968 clean_command_ctx(&cmd_ctx
);
3972 health_code_update();
3974 // TODO: Validate cmd_ctx including sanity check for
3975 // security purpose.
3977 rcu_thread_online();
3979 * This function dispatch the work to the kernel or userspace tracer
3980 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3981 * informations for the client. The command context struct contains
3982 * everything this function may needs.
3984 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3985 rcu_thread_offline();
3993 * TODO: Inform client somehow of the fatal error. At
3994 * this point, ret < 0 means that a zmalloc failed
3995 * (ENOMEM). Error detected but still accept
3996 * command, unless a socket error has been
3999 clean_command_ctx(&cmd_ctx
);
4003 health_code_update();
4005 DBG("Sending response (size: %d, retcode: %s)",
4006 cmd_ctx
->lttng_msg_size
,
4007 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4008 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4010 ERR("Failed to send data back to client");
4013 /* End of transmission */
4020 clean_command_ctx(&cmd_ctx
);
4022 health_code_update();
4034 lttng_poll_clean(&events
);
4035 clean_command_ctx(&cmd_ctx
);
4039 unlink(client_unix_sock_path
);
4040 if (client_sock
>= 0) {
4041 ret
= close(client_sock
);
4049 ERR("Health error occurred in %s", __func__
);
4052 health_unregister(health_sessiond
);
4054 DBG("Client thread dying");
4056 rcu_unregister_thread();
4062 * usage function on stderr
4064 static void usage(void)
4066 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4067 fprintf(stderr
, " -h, --help Display this usage.\n");
4068 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4069 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4070 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4071 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4072 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4073 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4074 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4075 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4076 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4077 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4078 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4079 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4080 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4081 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4082 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4083 fprintf(stderr
, " -V, --version Show version number.\n");
4084 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4085 fprintf(stderr
, " -q, --quiet No output at all.\n");
4086 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4087 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4088 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4089 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4090 fprintf(stderr
, " --jul-tcp-port JUL application registration TCP port\n");
4094 * daemon argument parsing
4096 static int parse_args(int argc
, char **argv
)
4100 static struct option long_options
[] = {
4101 { "client-sock", 1, 0, 'c' },
4102 { "apps-sock", 1, 0, 'a' },
4103 { "kconsumerd-cmd-sock", 1, 0, 'C' },
4104 { "kconsumerd-err-sock", 1, 0, 'E' },
4105 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
4106 { "ustconsumerd32-err-sock", 1, 0, 'H' },
4107 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
4108 { "ustconsumerd64-err-sock", 1, 0, 'F' },
4109 { "consumerd32-path", 1, 0, 'u' },
4110 { "consumerd32-libdir", 1, 0, 'U' },
4111 { "consumerd64-path", 1, 0, 't' },
4112 { "consumerd64-libdir", 1, 0, 'T' },
4113 { "daemonize", 0, 0, 'd' },
4114 { "sig-parent", 0, 0, 'S' },
4115 { "help", 0, 0, 'h' },
4116 { "group", 1, 0, 'g' },
4117 { "version", 0, 0, 'V' },
4118 { "quiet", 0, 0, 'q' },
4119 { "verbose", 0, 0, 'v' },
4120 { "verbose-consumer", 0, 0, 'Z' },
4121 { "no-kernel", 0, 0, 'N' },
4122 { "pidfile", 1, 0, 'p' },
4123 { "jul-tcp-port", 1, 0, 'J' },
4124 { "background", 0, 0, 'b' },
4129 int option_index
= 0;
4130 c
= getopt_long(argc
, argv
, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:b",
4131 long_options
, &option_index
);
4138 fprintf(stderr
, "option %s", long_options
[option_index
].name
);
4140 fprintf(stderr
, " with arg %s\n", optarg
);
4144 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4147 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4156 tracing_group_name
= optarg
;
4162 fprintf(stdout
, "%s\n", VERSION
);
4168 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4171 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4174 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4177 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4180 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4183 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", optarg
);
4189 lttng_opt_quiet
= 1;
4192 /* Verbose level can increase using multiple -v */
4193 lttng_opt_verbose
+= 1;
4196 opt_verbose_consumer
+= 1;
4199 consumerd32_bin
= optarg
;
4202 consumerd32_libdir
= optarg
;
4205 consumerd64_bin
= optarg
;
4208 consumerd64_libdir
= optarg
;
4211 opt_pidfile
= optarg
;
4213 case 'J': /* JUL TCP port. */
4218 v
= strtoul(optarg
, NULL
, 0);
4219 if (errno
!= 0 || !isdigit(optarg
[0])) {
4220 ERR("Wrong value in --jul-tcp-port parameter: %s", optarg
);
4223 if (v
== 0 || v
>= 65535) {
4224 ERR("Port overflow in --jul-tcp-port parameter: %s", optarg
);
4227 jul_tcp_port
= (uint32_t) v
;
4228 DBG3("JUL TCP port set to non default: %u", jul_tcp_port
);
4232 /* Unknown option or other error.
4233 * Error is printed by getopt, just return */
4242 * Creates the two needed socket by the daemon.
4243 * apps_sock - The communication socket for all UST apps.
4244 * client_sock - The communication of the cli tool (lttng).
4246 static int init_daemon_socket(void)
4251 old_umask
= umask(0);
4253 /* Create client tool unix socket */
4254 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4255 if (client_sock
< 0) {
4256 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4261 /* Set the cloexec flag */
4262 ret
= utils_set_fd_cloexec(client_sock
);
4264 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4265 "Continuing but note that the consumer daemon will have a "
4266 "reference to this socket on exec()", client_sock
);
4269 /* File permission MUST be 660 */
4270 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4272 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4277 /* Create the application unix socket */
4278 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4279 if (apps_sock
< 0) {
4280 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4285 /* Set the cloexec flag */
4286 ret
= utils_set_fd_cloexec(apps_sock
);
4288 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4289 "Continuing but note that the consumer daemon will have a "
4290 "reference to this socket on exec()", apps_sock
);
4293 /* File permission MUST be 666 */
4294 ret
= chmod(apps_unix_sock_path
,
4295 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4297 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4302 DBG3("Session daemon client socket %d and application socket %d created",
4303 client_sock
, apps_sock
);
4311 * Check if the global socket is available, and if a daemon is answering at the
4312 * other side. If yes, error is returned.
4314 static int check_existing_daemon(void)
4316 /* Is there anybody out there ? */
4317 if (lttng_session_daemon_alive()) {
4325 * Set the tracing group gid onto the client socket.
4327 * Race window between mkdir and chown is OK because we are going from more
4328 * permissive (root.root) to less permissive (root.tracing).
4330 static int set_permissions(char *rundir
)
4335 gid
= utils_get_group_id(tracing_group_name
);
4337 /* Set lttng run dir */
4338 ret
= chown(rundir
, 0, gid
);
4340 ERR("Unable to set group on %s", rundir
);
4345 * Ensure all applications and tracing group can search the run
4346 * dir. Allow everyone to read the directory, since it does not
4347 * buy us anything to hide its content.
4349 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4351 ERR("Unable to set permissions on %s", rundir
);
4355 /* lttng client socket path */
4356 ret
= chown(client_unix_sock_path
, 0, gid
);
4358 ERR("Unable to set group on %s", client_unix_sock_path
);
4362 /* kconsumer error socket path */
4363 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4365 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4369 /* 64-bit ustconsumer error socket path */
4370 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4372 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4376 /* 32-bit ustconsumer compat32 error socket path */
4377 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4379 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4383 DBG("All permissions are set");
4389 * Create the lttng run directory needed for all global sockets and pipe.
4391 static int create_lttng_rundir(const char *rundir
)
4395 DBG3("Creating LTTng run directory: %s", rundir
);
4397 ret
= mkdir(rundir
, S_IRWXU
);
4399 if (errno
!= EEXIST
) {
4400 ERR("Unable to create %s", rundir
);
4412 * Setup sockets and directory needed by the kconsumerd communication with the
4415 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4419 char path
[PATH_MAX
];
4421 switch (consumer_data
->type
) {
4422 case LTTNG_CONSUMER_KERNEL
:
4423 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4425 case LTTNG_CONSUMER64_UST
:
4426 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4428 case LTTNG_CONSUMER32_UST
:
4429 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4432 ERR("Consumer type unknown");
4437 DBG2("Creating consumer directory: %s", path
);
4439 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4441 if (errno
!= EEXIST
) {
4443 ERR("Failed to create %s", path
);
4449 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4451 ERR("Unable to set group on %s", path
);
4457 /* Create the kconsumerd error unix socket */
4458 consumer_data
->err_sock
=
4459 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4460 if (consumer_data
->err_sock
< 0) {
4461 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4467 * Set the CLOEXEC flag. Return code is useless because either way, the
4470 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4472 PERROR("utils_set_fd_cloexec");
4473 /* continue anyway */
4476 /* File permission MUST be 660 */
4477 ret
= chmod(consumer_data
->err_unix_sock_path
,
4478 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4480 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4490 * Signal handler for the daemon
4492 * Simply stop all worker threads, leaving main() return gracefully after
4493 * joining all threads and calling cleanup().
4495 static void sighandler(int sig
)
4499 DBG("SIGPIPE caught");
4502 DBG("SIGINT caught");
4506 DBG("SIGTERM caught");
4510 CMM_STORE_SHARED(recv_child_signal
, 1);
4518 * Setup signal handler for :
4519 * SIGINT, SIGTERM, SIGPIPE
4521 static int set_signal_handler(void)
4524 struct sigaction sa
;
4527 if ((ret
= sigemptyset(&sigset
)) < 0) {
4528 PERROR("sigemptyset");
4532 sa
.sa_handler
= sighandler
;
4533 sa
.sa_mask
= sigset
;
4535 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4536 PERROR("sigaction");
4540 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4541 PERROR("sigaction");
4545 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4546 PERROR("sigaction");
4550 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4551 PERROR("sigaction");
4555 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4561 * Set open files limit to unlimited. This daemon can open a large number of
4562 * file descriptors in order to consumer multiple kernel traces.
4564 static void set_ulimit(void)
4569 /* The kernel does not allowed an infinite limit for open files */
4570 lim
.rlim_cur
= 65535;
4571 lim
.rlim_max
= 65535;
4573 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4575 PERROR("failed to set open files limit");
4580 * Write pidfile using the rundir and opt_pidfile.
4582 static void write_pidfile(void)
4585 char pidfile_path
[PATH_MAX
];
4590 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4592 /* Build pidfile path from rundir and opt_pidfile. */
4593 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4594 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4596 PERROR("snprintf pidfile path");
4602 * Create pid file in rundir. Return value is of no importance. The
4603 * execution will continue even though we are not able to write the file.
4605 (void) utils_create_pid_file(getpid(), pidfile_path
);
4612 * Create lockfile using the rundir and return its fd.
4614 static int create_lockfile(void)
4617 char lockfile_path
[PATH_MAX
];
4619 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
4624 ret
= utils_create_lock_file(lockfile_path
);
4630 * Write JUL TCP port using the rundir.
4632 static void write_julport(void)
4635 char path
[PATH_MAX
];
4639 ret
= snprintf(path
, sizeof(path
), "%s/"
4640 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
, rundir
);
4642 PERROR("snprintf julport path");
4647 * Create TCP JUL port file in rundir. Return value is of no importance.
4648 * The execution will continue even though we are not able to write the
4651 (void) utils_create_pid_file(jul_tcp_port
, path
);
4660 int main(int argc
, char **argv
)
4664 const char *home_path
, *env_app_timeout
;
4666 init_kernel_workarounds();
4668 rcu_register_thread();
4670 if ((ret
= set_signal_handler()) < 0) {
4674 setup_consumerd_path();
4676 page_size
= sysconf(_SC_PAGESIZE
);
4677 if (page_size
< 0) {
4678 PERROR("sysconf _SC_PAGESIZE");
4679 page_size
= LONG_MAX
;
4680 WARN("Fallback page size to %ld", page_size
);
4683 /* Parse arguments */
4685 if ((ret
= parse_args(argc
, argv
)) < 0) {
4690 if (opt_daemon
|| opt_background
) {
4693 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4700 * We are in the child. Make sure all other file descriptors are
4701 * closed, in case we are called with more opened file descriptors than
4702 * the standard ones.
4704 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4709 /* Create thread quit pipe */
4710 if ((ret
= init_thread_quit_pipe()) < 0) {
4714 /* Check if daemon is UID = 0 */
4715 is_root
= !getuid();
4718 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4724 /* Create global run dir with root access */
4725 ret
= create_lttng_rundir(rundir
);
4730 if (strlen(apps_unix_sock_path
) == 0) {
4731 snprintf(apps_unix_sock_path
, PATH_MAX
,
4732 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4735 if (strlen(client_unix_sock_path
) == 0) {
4736 snprintf(client_unix_sock_path
, PATH_MAX
,
4737 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4740 /* Set global SHM for ust */
4741 if (strlen(wait_shm_path
) == 0) {
4742 snprintf(wait_shm_path
, PATH_MAX
,
4743 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4746 if (strlen(health_unix_sock_path
) == 0) {
4747 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4748 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4751 /* Setup kernel consumerd path */
4752 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4753 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4754 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4755 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4757 DBG2("Kernel consumer err path: %s",
4758 kconsumer_data
.err_unix_sock_path
);
4759 DBG2("Kernel consumer cmd path: %s",
4760 kconsumer_data
.cmd_unix_sock_path
);
4762 home_path
= utils_get_home_dir();
4763 if (home_path
== NULL
) {
4764 /* TODO: Add --socket PATH option */
4765 ERR("Can't get HOME directory for sockets creation.");
4771 * Create rundir from home path. This will create something like
4774 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4780 ret
= create_lttng_rundir(rundir
);
4785 if (strlen(apps_unix_sock_path
) == 0) {
4786 snprintf(apps_unix_sock_path
, PATH_MAX
,
4787 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4790 /* Set the cli tool unix socket path */
4791 if (strlen(client_unix_sock_path
) == 0) {
4792 snprintf(client_unix_sock_path
, PATH_MAX
,
4793 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4796 /* Set global SHM for ust */
4797 if (strlen(wait_shm_path
) == 0) {
4798 snprintf(wait_shm_path
, PATH_MAX
,
4799 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4802 /* Set health check Unix path */
4803 if (strlen(health_unix_sock_path
) == 0) {
4804 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4805 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4809 lockfile_fd
= create_lockfile();
4810 if (lockfile_fd
< 0) {
4814 /* Set consumer initial state */
4815 kernel_consumerd_state
= CONSUMER_STOPPED
;
4816 ust_consumerd_state
= CONSUMER_STOPPED
;
4818 DBG("Client socket path %s", client_unix_sock_path
);
4819 DBG("Application socket path %s", apps_unix_sock_path
);
4820 DBG("Application wait path %s", wait_shm_path
);
4821 DBG("LTTng run directory path: %s", rundir
);
4823 /* 32 bits consumerd path setup */
4824 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4825 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4826 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4827 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4829 DBG2("UST consumer 32 bits err path: %s",
4830 ustconsumer32_data
.err_unix_sock_path
);
4831 DBG2("UST consumer 32 bits cmd path: %s",
4832 ustconsumer32_data
.cmd_unix_sock_path
);
4834 /* 64 bits consumerd path setup */
4835 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4836 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4837 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4838 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4840 DBG2("UST consumer 64 bits err path: %s",
4841 ustconsumer64_data
.err_unix_sock_path
);
4842 DBG2("UST consumer 64 bits cmd path: %s",
4843 ustconsumer64_data
.cmd_unix_sock_path
);
4846 * See if daemon already exist.
4848 if ((ret
= check_existing_daemon()) < 0) {
4849 ERR("Already running daemon.\n");
4851 * We do not goto exit because we must not cleanup()
4852 * because a daemon is already running.
4858 * Init UST app hash table. Alloc hash table before this point since
4859 * cleanup() can get called after that point.
4863 /* Initialize JUL domain subsystem. */
4864 if ((ret
= jul_init()) < 0) {
4865 /* ENOMEM at this point. */
4869 /* After this point, we can safely call cleanup() with "goto exit" */
4872 * These actions must be executed as root. We do that *after* setting up
4873 * the sockets path because we MUST make the check for another daemon using
4874 * those paths *before* trying to set the kernel consumer sockets and init
4878 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4883 /* Setup kernel tracer */
4884 if (!opt_no_kernel
) {
4885 init_kernel_tracer();
4888 /* Set ulimit for open files */
4891 /* init lttng_fd tracking must be done after set_ulimit. */
4894 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4899 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4904 /* Setup the needed unix socket */
4905 if ((ret
= init_daemon_socket()) < 0) {
4909 /* Set credentials to socket */
4910 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4914 /* Get parent pid if -S, --sig-parent is specified. */
4915 if (opt_sig_parent
) {
4919 /* Setup the kernel pipe for waking up the kernel thread */
4920 if (is_root
&& !opt_no_kernel
) {
4921 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4926 /* Setup the thread ht_cleanup communication pipe. */
4927 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4931 /* Setup the thread apps communication pipe. */
4932 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4936 /* Setup the thread apps notify communication pipe. */
4937 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4941 /* Initialize global buffer per UID and PID registry. */
4942 buffer_reg_init_uid_registry();
4943 buffer_reg_init_pid_registry();
4945 /* Init UST command queue. */
4946 cds_wfq_init(&ust_cmd_queue
.queue
);
4949 * Get session list pointer. This pointer MUST NOT be free(). This list is
4950 * statically declared in session.c
4952 session_list_ptr
= session_get_list();
4954 /* Set up max poll set size */
4955 lttng_poll_set_max_size();
4959 /* Check for the application socket timeout env variable. */
4960 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4961 if (env_app_timeout
) {
4962 app_socket_timeout
= atoi(env_app_timeout
);
4964 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4970 /* Initialize communication library */
4972 /* This is to get the TCP timeout value. */
4973 lttcomm_inet_init();
4976 * Initialize the health check subsystem. This call should set the
4977 * appropriate time values.
4979 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
4980 if (!health_sessiond
) {
4981 PERROR("health_app_create error");
4982 goto exit_health_sessiond_cleanup
;
4985 /* Create thread to clean up RCU hash tables */
4986 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
4987 thread_ht_cleanup
, (void *) NULL
);
4989 PERROR("pthread_create ht_cleanup");
4990 goto exit_ht_cleanup
;
4993 /* Create health-check thread */
4994 ret
= pthread_create(&health_thread
, NULL
,
4995 thread_manage_health
, (void *) NULL
);
4997 PERROR("pthread_create health");
5001 /* Create thread to manage the client socket */
5002 ret
= pthread_create(&client_thread
, NULL
,
5003 thread_manage_clients
, (void *) NULL
);
5005 PERROR("pthread_create clients");
5009 /* Create thread to dispatch registration */
5010 ret
= pthread_create(&dispatch_thread
, NULL
,
5011 thread_dispatch_ust_registration
, (void *) NULL
);
5013 PERROR("pthread_create dispatch");
5017 /* Create thread to manage application registration. */
5018 ret
= pthread_create(®_apps_thread
, NULL
,
5019 thread_registration_apps
, (void *) NULL
);
5021 PERROR("pthread_create registration");
5025 /* Create thread to manage application socket */
5026 ret
= pthread_create(&apps_thread
, NULL
,
5027 thread_manage_apps
, (void *) NULL
);
5029 PERROR("pthread_create apps");
5033 /* Create thread to manage application notify socket */
5034 ret
= pthread_create(&apps_notify_thread
, NULL
,
5035 ust_thread_manage_notify
, (void *) NULL
);
5037 PERROR("pthread_create notify");
5038 goto exit_apps_notify
;
5041 /* Create JUL registration thread. */
5042 ret
= pthread_create(&jul_reg_thread
, NULL
,
5043 jul_thread_manage_registration
, (void *) NULL
);
5045 PERROR("pthread_create JUL");
5049 /* Don't start this thread if kernel tracing is not requested nor root */
5050 if (is_root
&& !opt_no_kernel
) {
5051 /* Create kernel thread to manage kernel event */
5052 ret
= pthread_create(&kernel_thread
, NULL
,
5053 thread_manage_kernel
, (void *) NULL
);
5055 PERROR("pthread_create kernel");
5059 ret
= pthread_join(kernel_thread
, &status
);
5061 PERROR("pthread_join");
5062 goto error
; /* join error, exit without cleanup */
5067 ret
= pthread_join(jul_reg_thread
, &status
);
5069 PERROR("pthread_join JUL");
5070 goto error
; /* join error, exit without cleanup */
5074 ret
= pthread_join(apps_notify_thread
, &status
);
5076 PERROR("pthread_join apps notify");
5077 goto error
; /* join error, exit without cleanup */
5081 ret
= pthread_join(apps_thread
, &status
);
5083 PERROR("pthread_join apps");
5084 goto error
; /* join error, exit without cleanup */
5089 ret
= pthread_join(reg_apps_thread
, &status
);
5091 PERROR("pthread_join");
5092 goto error
; /* join error, exit without cleanup */
5096 ret
= pthread_join(dispatch_thread
, &status
);
5098 PERROR("pthread_join");
5099 goto error
; /* join error, exit without cleanup */
5103 ret
= pthread_join(client_thread
, &status
);
5105 PERROR("pthread_join");
5106 goto error
; /* join error, exit without cleanup */
5109 ret
= join_consumer_thread(&kconsumer_data
);
5111 PERROR("join_consumer");
5112 goto error
; /* join error, exit without cleanup */
5115 ret
= join_consumer_thread(&ustconsumer32_data
);
5117 PERROR("join_consumer ust32");
5118 goto error
; /* join error, exit without cleanup */
5121 ret
= join_consumer_thread(&ustconsumer64_data
);
5123 PERROR("join_consumer ust64");
5124 goto error
; /* join error, exit without cleanup */
5128 ret
= pthread_join(health_thread
, &status
);
5130 PERROR("pthread_join health thread");
5131 goto error
; /* join error, exit without cleanup */
5135 ret
= pthread_join(ht_cleanup_thread
, &status
);
5137 PERROR("pthread_join ht cleanup thread");
5138 goto error
; /* join error, exit without cleanup */
5141 health_app_destroy(health_sessiond
);
5142 exit_health_sessiond_cleanup
:
5145 * cleanup() is called when no other thread is running.
5147 rcu_thread_online();
5149 rcu_thread_offline();
5150 rcu_unregister_thread();