2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
36 #include <sys/types.h>
38 #include <urcu/uatomic.h>
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/defaults.h>
45 #include <common/kernel-consumer/kernel-consumer.h>
46 #include <common/futex.h>
47 #include <common/relayd/relayd.h>
48 #include <common/utils.h>
49 #include <common/daemonize.h>
50 #include <common/config/config.h>
52 #include "lttng-sessiond.h"
53 #include "buffer-registry.h"
60 #include "kernel-consumer.h"
64 #include "ust-consumer.h"
67 #include "health-sessiond.h"
68 #include "testpoint.h"
69 #include "ust-thread.h"
70 #include "jul-thread.h"
72 #define CONSUMERD_FILE "lttng-consumerd"
75 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
76 static int tracing_group_name_override
;
77 static char *opt_pidfile
;
78 static int opt_sig_parent
;
79 static int opt_verbose_consumer
;
80 static int opt_daemon
;
81 static int opt_no_kernel
;
82 static int is_root
; /* Set to 1 if the daemon is running as root */
83 static pid_t ppid
; /* Parent PID for --sig-parent option */
84 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
87 /* Set to 1 when a SIGUSR1 signal is received. */
88 static int recv_child_signal
;
91 * Consumer daemon specific control data. Every value not initialized here is
92 * set to 0 by the static definition.
94 static struct consumer_data kconsumer_data
= {
95 .type
= LTTNG_CONSUMER_KERNEL
,
96 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
97 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
100 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
101 .lock
= PTHREAD_MUTEX_INITIALIZER
,
102 .cond
= PTHREAD_COND_INITIALIZER
,
103 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
105 static struct consumer_data ustconsumer64_data
= {
106 .type
= LTTNG_CONSUMER64_UST
,
107 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
108 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
111 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
112 .lock
= PTHREAD_MUTEX_INITIALIZER
,
113 .cond
= PTHREAD_COND_INITIALIZER
,
114 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
116 static struct consumer_data ustconsumer32_data
= {
117 .type
= LTTNG_CONSUMER32_UST
,
118 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
119 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
122 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
123 .lock
= PTHREAD_MUTEX_INITIALIZER
,
124 .cond
= PTHREAD_COND_INITIALIZER
,
125 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
128 /* Command line options */
129 static const struct option long_options
[] = {
130 { "client-sock", 1, 0, 'c' },
131 { "apps-sock", 1, 0, 'a' },
132 { "kconsumerd-cmd-sock", 1, 0, 'C' },
133 { "kconsumerd-err-sock", 1, 0, 'E' },
134 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
135 { "ustconsumerd32-err-sock", 1, 0, 'H' },
136 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
137 { "ustconsumerd64-err-sock", 1, 0, 'F' },
138 { "consumerd32-path", 1, 0, 'u' },
139 { "consumerd32-libdir", 1, 0, 'U' },
140 { "consumerd64-path", 1, 0, 't' },
141 { "consumerd64-libdir", 1, 0, 'T' },
142 { "daemonize", 0, 0, 'd' },
143 { "sig-parent", 0, 0, 'S' },
144 { "help", 0, 0, 'h' },
145 { "group", 1, 0, 'g' },
146 { "version", 0, 0, 'V' },
147 { "quiet", 0, 0, 'q' },
148 { "verbose", 0, 0, 'v' },
149 { "verbose-consumer", 0, 0, 'Z' },
150 { "no-kernel", 0, 0, 'N' },
151 { "pidfile", 1, 0, 'p' },
152 { "jul-tcp-port", 1, 0, 'J' },
153 { "config", 1, 0, 'f' },
157 /* Command line options to ignore from configuration file */
158 static const char *config_ignore_options
[] = { "help", "version", "config" };
160 /* Shared between threads */
161 static int dispatch_thread_exit
;
163 /* Global application Unix socket path */
164 static char apps_unix_sock_path
[PATH_MAX
];
165 /* Global client Unix socket path */
166 static char client_unix_sock_path
[PATH_MAX
];
167 /* global wait shm path for UST */
168 static char wait_shm_path
[PATH_MAX
];
169 /* Global health check unix path */
170 static char health_unix_sock_path
[PATH_MAX
];
172 /* Sockets and FDs */
173 static int client_sock
= -1;
174 static int apps_sock
= -1;
175 int kernel_tracer_fd
= -1;
176 static int kernel_poll_pipe
[2] = { -1, -1 };
179 * Quit pipe for all threads. This permits a single cancellation point
180 * for all threads when receiving an event on the pipe.
182 static int thread_quit_pipe
[2] = { -1, -1 };
185 * This pipe is used to inform the thread managing application communication
186 * that a command is queued and ready to be processed.
188 static int apps_cmd_pipe
[2] = { -1, -1 };
190 int apps_cmd_notify_pipe
[2] = { -1, -1 };
192 /* Pthread, Mutexes and Semaphores */
193 static pthread_t apps_thread
;
194 static pthread_t apps_notify_thread
;
195 static pthread_t reg_apps_thread
;
196 static pthread_t client_thread
;
197 static pthread_t kernel_thread
;
198 static pthread_t dispatch_thread
;
199 static pthread_t health_thread
;
200 static pthread_t ht_cleanup_thread
;
201 static pthread_t jul_reg_thread
;
204 * UST registration command queue. This queue is tied with a futex and uses a N
205 * wakers / 1 waiter implemented and detailed in futex.c/.h
207 * The thread_manage_apps and thread_dispatch_ust_registration interact with
208 * this queue and the wait/wake scheme.
210 static struct ust_cmd_queue ust_cmd_queue
;
213 * Pointer initialized before thread creation.
215 * This points to the tracing session list containing the session count and a
216 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
217 * MUST NOT be taken if you call a public function in session.c.
219 * The lock is nested inside the structure: session_list_ptr->lock. Please use
220 * session_lock_list and session_unlock_list for lock acquisition.
222 static struct ltt_session_list
*session_list_ptr
;
224 int ust_consumerd64_fd
= -1;
225 int ust_consumerd32_fd
= -1;
227 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
228 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
229 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
230 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
231 static int consumerd32_bin_override
;
232 static int consumerd64_bin_override
;
233 static int consumerd32_libdir_override
;
234 static int consumerd64_libdir_override
;
236 static const char *module_proc_lttng
= "/proc/lttng";
239 * Consumer daemon state which is changed when spawning it, killing it or in
240 * case of a fatal error.
242 enum consumerd_state
{
243 CONSUMER_STARTED
= 1,
244 CONSUMER_STOPPED
= 2,
249 * This consumer daemon state is used to validate if a client command will be
250 * able to reach the consumer. If not, the client is informed. For instance,
251 * doing a "lttng start" when the consumer state is set to ERROR will return an
252 * error to the client.
254 * The following example shows a possible race condition of this scheme:
256 * consumer thread error happens
258 * client cmd checks state -> still OK
259 * consumer thread exit, sets error
260 * client cmd try to talk to consumer
263 * However, since the consumer is a different daemon, we have no way of making
264 * sure the command will reach it safely even with this state flag. This is why
265 * we consider that up to the state validation during command processing, the
266 * command is safe. After that, we can not guarantee the correctness of the
267 * client request vis-a-vis the consumer.
269 static enum consumerd_state ust_consumerd_state
;
270 static enum consumerd_state kernel_consumerd_state
;
273 * Socket timeout for receiving and sending in seconds.
275 static int app_socket_timeout
;
277 /* Set in main() with the current page size. */
280 /* Application health monitoring */
281 struct health_app
*health_sessiond
;
283 /* JUL TCP port for registration. Used by the JUL thread. */
284 unsigned int jul_tcp_port
= DEFAULT_JUL_TCP_PORT
;
286 const char * const config_section_name
= "sessiond";
289 * Whether sessiond is ready for commands/health check requests.
290 * NR_LTTNG_SESSIOND_READY must match the number of calls to
291 * lttng_sessiond_notify_ready().
293 #define NR_LTTNG_SESSIOND_READY 2
294 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
296 /* Notify parents that we are ready for cmd and health check */
298 void lttng_sessiond_notify_ready(void)
300 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
302 * Notify parent pid that we are ready to accept command
303 * for client side. This ppid is the one from the
304 * external process that spawned us.
306 if (opt_sig_parent
) {
311 * Notify the parent of the fork() process that we are
315 kill(child_ppid
, SIGUSR1
);
321 void setup_consumerd_path(void)
323 const char *bin
, *libdir
;
326 * Allow INSTALL_BIN_PATH to be used as a target path for the
327 * native architecture size consumer if CONFIG_CONSUMER*_PATH
328 * has not been defined.
330 #if (CAA_BITS_PER_LONG == 32)
331 if (!consumerd32_bin
[0]) {
332 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
334 if (!consumerd32_libdir
[0]) {
335 consumerd32_libdir
= INSTALL_LIB_PATH
;
337 #elif (CAA_BITS_PER_LONG == 64)
338 if (!consumerd64_bin
[0]) {
339 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
341 if (!consumerd64_libdir
[0]) {
342 consumerd64_libdir
= INSTALL_LIB_PATH
;
345 #error "Unknown bitness"
349 * runtime env. var. overrides the build default.
351 bin
= getenv("LTTNG_CONSUMERD32_BIN");
353 consumerd32_bin
= bin
;
355 bin
= getenv("LTTNG_CONSUMERD64_BIN");
357 consumerd64_bin
= bin
;
359 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
361 consumerd32_libdir
= libdir
;
363 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
365 consumerd64_libdir
= libdir
;
370 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
372 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
378 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
384 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
396 * Check if the thread quit pipe was triggered.
398 * Return 1 if it was triggered else 0;
400 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
402 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
410 * Init thread quit pipe.
412 * Return -1 on error or 0 if all pipes are created.
414 static int init_thread_quit_pipe(void)
418 ret
= pipe(thread_quit_pipe
);
420 PERROR("thread quit pipe");
424 for (i
= 0; i
< 2; i
++) {
425 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
437 * Stop all threads by closing the thread quit pipe.
439 static void stop_threads(void)
443 /* Stopping all threads */
444 DBG("Terminating all threads");
445 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
447 ERR("write error on thread quit pipe");
450 /* Dispatch thread */
451 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
452 futex_nto1_wake(&ust_cmd_queue
.futex
);
456 * Close every consumer sockets.
458 static void close_consumer_sockets(void)
462 if (kconsumer_data
.err_sock
>= 0) {
463 ret
= close(kconsumer_data
.err_sock
);
465 PERROR("kernel consumer err_sock close");
468 if (ustconsumer32_data
.err_sock
>= 0) {
469 ret
= close(ustconsumer32_data
.err_sock
);
471 PERROR("UST consumerd32 err_sock close");
474 if (ustconsumer64_data
.err_sock
>= 0) {
475 ret
= close(ustconsumer64_data
.err_sock
);
477 PERROR("UST consumerd64 err_sock close");
480 if (kconsumer_data
.cmd_sock
>= 0) {
481 ret
= close(kconsumer_data
.cmd_sock
);
483 PERROR("kernel consumer cmd_sock close");
486 if (ustconsumer32_data
.cmd_sock
>= 0) {
487 ret
= close(ustconsumer32_data
.cmd_sock
);
489 PERROR("UST consumerd32 cmd_sock close");
492 if (ustconsumer64_data
.cmd_sock
>= 0) {
493 ret
= close(ustconsumer64_data
.cmd_sock
);
495 PERROR("UST consumerd64 cmd_sock close");
503 static void cleanup(void)
506 struct ltt_session
*sess
, *stmp
;
512 * Close the thread quit pipe. It has already done its job,
513 * since we are now called.
515 utils_close_pipe(thread_quit_pipe
);
518 * If opt_pidfile is undefined, the default file will be wiped when
519 * removing the rundir.
522 ret
= remove(opt_pidfile
);
524 PERROR("remove pidfile %s", opt_pidfile
);
528 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
531 snprintf(path
, PATH_MAX
,
533 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
534 DBG("Removing %s", path
);
537 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
538 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
);
539 DBG("Removing %s", path
);
543 snprintf(path
, PATH_MAX
,
544 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
546 DBG("Removing %s", path
);
549 snprintf(path
, PATH_MAX
,
550 DEFAULT_KCONSUMERD_PATH
,
552 DBG("Removing directory %s", path
);
555 /* ust consumerd 32 */
556 snprintf(path
, PATH_MAX
,
557 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
559 DBG("Removing %s", path
);
562 snprintf(path
, PATH_MAX
,
563 DEFAULT_USTCONSUMERD32_PATH
,
565 DBG("Removing directory %s", path
);
568 /* ust consumerd 64 */
569 snprintf(path
, PATH_MAX
,
570 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
572 DBG("Removing %s", path
);
575 snprintf(path
, PATH_MAX
,
576 DEFAULT_USTCONSUMERD64_PATH
,
578 DBG("Removing directory %s", path
);
582 * We do NOT rmdir rundir because there are other processes
583 * using it, for instance lttng-relayd, which can start in
584 * parallel with this teardown.
589 DBG("Cleaning up all sessions");
591 /* Destroy session list mutex */
592 if (session_list_ptr
!= NULL
) {
593 pthread_mutex_destroy(&session_list_ptr
->lock
);
595 /* Cleanup ALL session */
596 cds_list_for_each_entry_safe(sess
, stmp
,
597 &session_list_ptr
->head
, list
) {
598 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
602 DBG("Closing all UST sockets");
603 ust_app_clean_list();
604 buffer_reg_destroy_registries();
606 if (is_root
&& !opt_no_kernel
) {
607 DBG2("Closing kernel fd");
608 if (kernel_tracer_fd
>= 0) {
609 ret
= close(kernel_tracer_fd
);
614 DBG("Unloading kernel modules");
615 modprobe_remove_lttng_all();
618 close_consumer_sockets();
621 * If the override option is set, the pointer points to a *non* const thus
622 * freeing it even though the variable type is set to const.
624 if (tracing_group_name_override
) {
625 free((void *) tracing_group_name
);
627 if (consumerd32_bin_override
) {
628 free((void *) consumerd32_bin
);
630 if (consumerd64_bin_override
) {
631 free((void *) consumerd64_bin
);
633 if (consumerd32_libdir_override
) {
634 free((void *) consumerd32_libdir
);
636 if (consumerd64_libdir_override
) {
637 free((void *) consumerd64_libdir
);
645 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
646 "Matthew, BEET driven development works!%c[%dm",
647 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
652 * Send data on a unix socket using the liblttsessiondcomm API.
654 * Return lttcomm error code.
656 static int send_unix_sock(int sock
, void *buf
, size_t len
)
658 /* Check valid length */
663 return lttcomm_send_unix_sock(sock
, buf
, len
);
667 * Free memory of a command context structure.
669 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
671 DBG("Clean command context structure");
673 if ((*cmd_ctx
)->llm
) {
674 free((*cmd_ctx
)->llm
);
676 if ((*cmd_ctx
)->lsm
) {
677 free((*cmd_ctx
)->lsm
);
685 * Notify UST applications using the shm mmap futex.
687 static int notify_ust_apps(int active
)
691 DBG("Notifying applications of session daemon state: %d", active
);
693 /* See shm.c for this call implying mmap, shm and futex calls */
694 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
695 if (wait_shm_mmap
== NULL
) {
699 /* Wake waiting process */
700 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
702 /* Apps notified successfully */
710 * Setup the outgoing data buffer for the response (llm) by allocating the
711 * right amount of memory and copying the original information from the lsm
714 * Return total size of the buffer pointed by buf.
716 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
722 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
723 if (cmd_ctx
->llm
== NULL
) {
729 /* Copy common data */
730 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
731 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
733 cmd_ctx
->llm
->data_size
= size
;
734 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
743 * Update the kernel poll set of all channel fd available over all tracing
744 * session. Add the wakeup pipe at the end of the set.
746 static int update_kernel_poll(struct lttng_poll_event
*events
)
749 struct ltt_session
*session
;
750 struct ltt_kernel_channel
*channel
;
752 DBG("Updating kernel poll set");
755 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
756 session_lock(session
);
757 if (session
->kernel_session
== NULL
) {
758 session_unlock(session
);
762 cds_list_for_each_entry(channel
,
763 &session
->kernel_session
->channel_list
.head
, list
) {
764 /* Add channel fd to the kernel poll set */
765 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
767 session_unlock(session
);
770 DBG("Channel fd %d added to kernel set", channel
->fd
);
772 session_unlock(session
);
774 session_unlock_list();
779 session_unlock_list();
784 * Find the channel fd from 'fd' over all tracing session. When found, check
785 * for new channel stream and send those stream fds to the kernel consumer.
787 * Useful for CPU hotplug feature.
789 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
792 struct ltt_session
*session
;
793 struct ltt_kernel_session
*ksess
;
794 struct ltt_kernel_channel
*channel
;
796 DBG("Updating kernel streams for channel fd %d", fd
);
799 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
800 session_lock(session
);
801 if (session
->kernel_session
== NULL
) {
802 session_unlock(session
);
805 ksess
= session
->kernel_session
;
807 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
808 if (channel
->fd
== fd
) {
809 DBG("Channel found, updating kernel streams");
810 ret
= kernel_open_channel_stream(channel
);
814 /* Update the stream global counter */
815 ksess
->stream_count_global
+= ret
;
818 * Have we already sent fds to the consumer? If yes, it means
819 * that tracing is started so it is safe to send our updated
822 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
823 struct lttng_ht_iter iter
;
824 struct consumer_socket
*socket
;
827 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
828 &iter
.iter
, socket
, node
.node
) {
829 pthread_mutex_lock(socket
->lock
);
830 ret
= kernel_consumer_send_channel_stream(socket
,
832 session
->output_traces
? 1 : 0);
833 pthread_mutex_unlock(socket
->lock
);
844 session_unlock(session
);
846 session_unlock_list();
850 session_unlock(session
);
851 session_unlock_list();
856 * For each tracing session, update newly registered apps. The session list
857 * lock MUST be acquired before calling this.
859 static void update_ust_app(int app_sock
)
861 struct ltt_session
*sess
, *stmp
;
863 /* Consumer is in an ERROR state. Stop any application update. */
864 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
865 /* Stop the update process since the consumer is dead. */
869 /* For all tracing session(s) */
870 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
872 if (sess
->ust_session
) {
873 ust_app_global_update(sess
->ust_session
, app_sock
);
875 session_unlock(sess
);
880 * This thread manage event coming from the kernel.
882 * Features supported in this thread:
885 static void *thread_manage_kernel(void *data
)
887 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
888 uint32_t revents
, nb_fd
;
890 struct lttng_poll_event events
;
892 DBG("[thread] Thread manage kernel started");
894 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
897 * This first step of the while is to clean this structure which could free
898 * non NULL pointers so initialize it before the loop.
900 lttng_poll_init(&events
);
902 if (testpoint(sessiond_thread_manage_kernel
)) {
903 goto error_testpoint
;
906 health_code_update();
908 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
909 goto error_testpoint
;
913 health_code_update();
915 if (update_poll_flag
== 1) {
916 /* Clean events object. We are about to populate it again. */
917 lttng_poll_clean(&events
);
919 ret
= sessiond_set_thread_pollset(&events
, 2);
921 goto error_poll_create
;
924 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
929 /* This will add the available kernel channel if any. */
930 ret
= update_kernel_poll(&events
);
934 update_poll_flag
= 0;
937 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
939 /* Poll infinite value of time */
942 ret
= lttng_poll_wait(&events
, -1);
946 * Restart interrupted system call.
948 if (errno
== EINTR
) {
952 } else if (ret
== 0) {
953 /* Should not happen since timeout is infinite */
954 ERR("Return value of poll is 0 with an infinite timeout.\n"
955 "This should not have happened! Continuing...");
961 for (i
= 0; i
< nb_fd
; i
++) {
962 /* Fetch once the poll data */
963 revents
= LTTNG_POLL_GETEV(&events
, i
);
964 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
966 health_code_update();
968 /* Thread quit pipe has been closed. Killing thread. */
969 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
975 /* Check for data on kernel pipe */
976 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
977 (void) lttng_read(kernel_poll_pipe
[0],
980 * Ret value is useless here, if this pipe gets any actions an
981 * update is required anyway.
983 update_poll_flag
= 1;
987 * New CPU detected by the kernel. Adding kernel stream to
988 * kernel session and updating the kernel consumer
990 if (revents
& LPOLLIN
) {
991 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
997 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
998 * and unregister kernel stream at this point.
1007 lttng_poll_clean(&events
);
1010 utils_close_pipe(kernel_poll_pipe
);
1011 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1014 ERR("Health error occurred in %s", __func__
);
1015 WARN("Kernel thread died unexpectedly. "
1016 "Kernel tracing can continue but CPU hotplug is disabled.");
1018 health_unregister(health_sessiond
);
1019 DBG("Kernel thread dying");
1024 * Signal pthread condition of the consumer data that the thread.
1026 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1028 pthread_mutex_lock(&data
->cond_mutex
);
1031 * The state is set before signaling. It can be any value, it's the waiter
1032 * job to correctly interpret this condition variable associated to the
1033 * consumer pthread_cond.
1035 * A value of 0 means that the corresponding thread of the consumer data
1036 * was not started. 1 indicates that the thread has started and is ready
1037 * for action. A negative value means that there was an error during the
1040 data
->consumer_thread_is_ready
= state
;
1041 (void) pthread_cond_signal(&data
->cond
);
1043 pthread_mutex_unlock(&data
->cond_mutex
);
1047 * This thread manage the consumer error sent back to the session daemon.
1049 static void *thread_manage_consumer(void *data
)
1051 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1052 uint32_t revents
, nb_fd
;
1053 enum lttcomm_return_code code
;
1054 struct lttng_poll_event events
;
1055 struct consumer_data
*consumer_data
= data
;
1057 DBG("[thread] Manage consumer started");
1059 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1061 health_code_update();
1064 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1065 * metadata_sock. Nothing more will be added to this poll set.
1067 ret
= sessiond_set_thread_pollset(&events
, 3);
1073 * The error socket here is already in a listening state which was done
1074 * just before spawning this thread to avoid a race between the consumer
1075 * daemon exec trying to connect and the listen() call.
1077 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1082 health_code_update();
1084 /* Infinite blocking call, waiting for transmission */
1086 health_poll_entry();
1088 if (testpoint(sessiond_thread_manage_consumer
)) {
1092 ret
= lttng_poll_wait(&events
, -1);
1096 * Restart interrupted system call.
1098 if (errno
== EINTR
) {
1106 for (i
= 0; i
< nb_fd
; i
++) {
1107 /* Fetch once the poll data */
1108 revents
= LTTNG_POLL_GETEV(&events
, i
);
1109 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1111 health_code_update();
1113 /* Thread quit pipe has been closed. Killing thread. */
1114 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1120 /* Event on the registration socket */
1121 if (pollfd
== consumer_data
->err_sock
) {
1122 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1123 ERR("consumer err socket poll error");
1129 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1135 * Set the CLOEXEC flag. Return code is useless because either way, the
1138 (void) utils_set_fd_cloexec(sock
);
1140 health_code_update();
1142 DBG2("Receiving code from consumer err_sock");
1144 /* Getting status code from kconsumerd */
1145 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1146 sizeof(enum lttcomm_return_code
));
1151 health_code_update();
1153 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1154 /* Connect both socket, command and metadata. */
1155 consumer_data
->cmd_sock
=
1156 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1157 consumer_data
->metadata_fd
=
1158 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1159 if (consumer_data
->cmd_sock
< 0
1160 || consumer_data
->metadata_fd
< 0) {
1161 PERROR("consumer connect cmd socket");
1162 /* On error, signal condition and quit. */
1163 signal_consumer_condition(consumer_data
, -1);
1166 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1167 /* Create metadata socket lock. */
1168 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1169 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1170 PERROR("zmalloc pthread mutex");
1174 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1176 signal_consumer_condition(consumer_data
, 1);
1177 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1178 DBG("Consumer metadata socket ready (fd: %d)",
1179 consumer_data
->metadata_fd
);
1181 ERR("consumer error when waiting for SOCK_READY : %s",
1182 lttcomm_get_readable_code(-code
));
1186 /* Remove the consumerd error sock since we've established a connexion */
1187 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1192 /* Add new accepted error socket. */
1193 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1198 /* Add metadata socket that is successfully connected. */
1199 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1200 LPOLLIN
| LPOLLRDHUP
);
1205 health_code_update();
1207 /* Infinite blocking call, waiting for transmission */
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();
1232 /* Thread quit pipe has been closed. Killing thread. */
1233 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1239 if (pollfd
== sock
) {
1240 /* Event on the consumerd socket */
1241 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1242 ERR("consumer err socket second poll error");
1245 health_code_update();
1246 /* Wait for any kconsumerd error */
1247 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1248 sizeof(enum lttcomm_return_code
));
1250 ERR("consumer closed the command socket");
1254 ERR("consumer return code : %s",
1255 lttcomm_get_readable_code(-code
));
1258 } else if (pollfd
== consumer_data
->metadata_fd
) {
1259 /* UST metadata requests */
1260 ret
= ust_consumer_metadata_request(
1261 &consumer_data
->metadata_sock
);
1263 ERR("Handling metadata request");
1268 ERR("Unknown pollfd");
1272 health_code_update();
1278 * We lock here because we are about to close the sockets and some other
1279 * thread might be using them so get exclusive access which will abort all
1280 * other consumer command by other threads.
1282 pthread_mutex_lock(&consumer_data
->lock
);
1284 /* Immediately set the consumerd state to stopped */
1285 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1286 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1287 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1288 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1289 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1291 /* Code flow error... */
1295 if (consumer_data
->err_sock
>= 0) {
1296 ret
= close(consumer_data
->err_sock
);
1300 consumer_data
->err_sock
= -1;
1302 if (consumer_data
->cmd_sock
>= 0) {
1303 ret
= close(consumer_data
->cmd_sock
);
1307 consumer_data
->cmd_sock
= -1;
1309 if (*consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1310 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1323 unlink(consumer_data
->err_unix_sock_path
);
1324 unlink(consumer_data
->cmd_unix_sock_path
);
1325 consumer_data
->pid
= 0;
1326 pthread_mutex_unlock(&consumer_data
->lock
);
1328 /* Cleanup metadata socket mutex. */
1329 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1330 free(consumer_data
->metadata_sock
.lock
);
1332 lttng_poll_clean(&events
);
1336 ERR("Health error occurred in %s", __func__
);
1338 health_unregister(health_sessiond
);
1339 DBG("consumer thread cleanup completed");
1345 * This thread manage application communication.
1347 static void *thread_manage_apps(void *data
)
1349 int i
, ret
, pollfd
, err
= -1;
1351 uint32_t revents
, nb_fd
;
1352 struct lttng_poll_event events
;
1354 DBG("[thread] Manage application started");
1356 rcu_register_thread();
1357 rcu_thread_online();
1359 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1361 if (testpoint(sessiond_thread_manage_apps
)) {
1362 goto error_testpoint
;
1365 health_code_update();
1367 ret
= sessiond_set_thread_pollset(&events
, 2);
1369 goto error_poll_create
;
1372 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1377 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1381 health_code_update();
1384 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1386 /* Inifinite blocking call, waiting for transmission */
1388 health_poll_entry();
1389 ret
= lttng_poll_wait(&events
, -1);
1393 * Restart interrupted system call.
1395 if (errno
== EINTR
) {
1403 for (i
= 0; i
< nb_fd
; i
++) {
1404 /* Fetch once the poll data */
1405 revents
= LTTNG_POLL_GETEV(&events
, i
);
1406 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1408 health_code_update();
1410 /* Thread quit pipe has been closed. Killing thread. */
1411 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1417 /* Inspect the apps cmd pipe */
1418 if (pollfd
== apps_cmd_pipe
[0]) {
1419 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1420 ERR("Apps command pipe error");
1422 } else if (revents
& LPOLLIN
) {
1426 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1427 if (size_ret
< sizeof(sock
)) {
1428 PERROR("read apps cmd pipe");
1432 health_code_update();
1435 * We only monitor the error events of the socket. This
1436 * thread does not handle any incoming data from UST
1439 ret
= lttng_poll_add(&events
, sock
,
1440 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1445 DBG("Apps with sock %d added to poll set", sock
);
1449 * At this point, we know that a registered application made
1450 * the event at poll_wait.
1452 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1453 /* Removing from the poll set */
1454 ret
= lttng_poll_del(&events
, pollfd
);
1459 /* Socket closed on remote end. */
1460 ust_app_unregister(pollfd
);
1464 health_code_update();
1470 lttng_poll_clean(&events
);
1473 utils_close_pipe(apps_cmd_pipe
);
1474 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1477 * We don't clean the UST app hash table here since already registered
1478 * applications can still be controlled so let them be until the session
1479 * daemon dies or the applications stop.
1484 ERR("Health error occurred in %s", __func__
);
1486 health_unregister(health_sessiond
);
1487 DBG("Application communication apps thread cleanup complete");
1488 rcu_thread_offline();
1489 rcu_unregister_thread();
1494 * Send a socket to a thread This is called from the dispatch UST registration
1495 * thread once all sockets are set for the application.
1497 * The sock value can be invalid, we don't really care, the thread will handle
1498 * it and make the necessary cleanup if so.
1500 * On success, return 0 else a negative value being the errno message of the
1503 static int send_socket_to_thread(int fd
, int sock
)
1508 * It's possible that the FD is set as invalid with -1 concurrently just
1509 * before calling this function being a shutdown state of the thread.
1516 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1517 if (ret
< sizeof(sock
)) {
1518 PERROR("write apps pipe %d", fd
);
1525 /* All good. Don't send back the write positive ret value. */
1532 * Sanitize the wait queue of the dispatch registration thread meaning removing
1533 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1534 * notify socket is never received.
1536 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1538 int ret
, nb_fd
= 0, i
;
1539 unsigned int fd_added
= 0;
1540 struct lttng_poll_event events
;
1541 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1545 lttng_poll_init(&events
);
1547 /* Just skip everything for an empty queue. */
1548 if (!wait_queue
->count
) {
1552 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1557 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1558 &wait_queue
->head
, head
) {
1559 assert(wait_node
->app
);
1560 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1561 LPOLLHUP
| LPOLLERR
);
1574 * Poll but don't block so we can quickly identify the faulty events and
1575 * clean them afterwards from the wait queue.
1577 ret
= lttng_poll_wait(&events
, 0);
1583 for (i
= 0; i
< nb_fd
; i
++) {
1584 /* Get faulty FD. */
1585 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1586 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1588 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1589 &wait_queue
->head
, head
) {
1590 if (pollfd
== wait_node
->app
->sock
&&
1591 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1592 cds_list_del(&wait_node
->head
);
1593 wait_queue
->count
--;
1594 ust_app_destroy(wait_node
->app
);
1602 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1606 lttng_poll_clean(&events
);
1610 lttng_poll_clean(&events
);
1612 ERR("Unable to sanitize wait queue");
1617 * Dispatch request from the registration threads to the application
1618 * communication thread.
1620 static void *thread_dispatch_ust_registration(void *data
)
1623 struct cds_wfq_node
*node
;
1624 struct ust_command
*ust_cmd
= NULL
;
1625 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1626 struct ust_reg_wait_queue wait_queue
= {
1630 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1632 health_code_update();
1634 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1636 DBG("[thread] Dispatch UST command started");
1638 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1639 health_code_update();
1641 /* Atomically prepare the queue futex */
1642 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1645 struct ust_app
*app
= NULL
;
1649 * Make sure we don't have node(s) that have hung up before receiving
1650 * the notify socket. This is to clean the list in order to avoid
1651 * memory leaks from notify socket that are never seen.
1653 sanitize_wait_queue(&wait_queue
);
1655 health_code_update();
1656 /* Dequeue command for registration */
1657 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1659 DBG("Woken up but nothing in the UST command queue");
1660 /* Continue thread execution */
1664 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1666 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1667 " gid:%d sock:%d name:%s (version %d.%d)",
1668 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1669 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1670 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1671 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1673 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1674 wait_node
= zmalloc(sizeof(*wait_node
));
1676 PERROR("zmalloc wait_node dispatch");
1677 ret
= close(ust_cmd
->sock
);
1679 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1681 lttng_fd_put(LTTNG_FD_APPS
, 1);
1685 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1687 /* Create application object if socket is CMD. */
1688 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1690 if (!wait_node
->app
) {
1691 ret
= close(ust_cmd
->sock
);
1693 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1695 lttng_fd_put(LTTNG_FD_APPS
, 1);
1701 * Add application to the wait queue so we can set the notify
1702 * socket before putting this object in the global ht.
1704 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1709 * We have to continue here since we don't have the notify
1710 * socket and the application MUST be added to the hash table
1711 * only at that moment.
1716 * Look for the application in the local wait queue and set the
1717 * notify socket if found.
1719 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1720 &wait_queue
.head
, head
) {
1721 health_code_update();
1722 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1723 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1724 cds_list_del(&wait_node
->head
);
1726 app
= wait_node
->app
;
1728 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1734 * With no application at this stage the received socket is
1735 * basically useless so close it before we free the cmd data
1736 * structure for good.
1739 ret
= close(ust_cmd
->sock
);
1741 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1743 lttng_fd_put(LTTNG_FD_APPS
, 1);
1750 * @session_lock_list
1752 * Lock the global session list so from the register up to the
1753 * registration done message, no thread can see the application
1754 * and change its state.
1756 session_lock_list();
1760 * Add application to the global hash table. This needs to be
1761 * done before the update to the UST registry can locate the
1766 /* Set app version. This call will print an error if needed. */
1767 (void) ust_app_version(app
);
1769 /* Send notify socket through the notify pipe. */
1770 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1774 session_unlock_list();
1776 * No notify thread, stop the UST tracing. However, this is
1777 * not an internal error of the this thread thus setting
1778 * the health error code to a normal exit.
1785 * Update newly registered application with the tracing
1786 * registry info already enabled information.
1788 update_ust_app(app
->sock
);
1791 * Don't care about return value. Let the manage apps threads
1792 * handle app unregistration upon socket close.
1794 (void) ust_app_register_done(app
->sock
);
1797 * Even if the application socket has been closed, send the app
1798 * to the thread and unregistration will take place at that
1801 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1804 session_unlock_list();
1806 * No apps. thread, stop the UST tracing. However, this is
1807 * not an internal error of the this thread thus setting
1808 * the health error code to a normal exit.
1815 session_unlock_list();
1817 } while (node
!= NULL
);
1819 health_poll_entry();
1820 /* Futex wait on queue. Blocking call on futex() */
1821 futex_nto1_wait(&ust_cmd_queue
.futex
);
1824 /* Normal exit, no error */
1828 /* Clean up wait queue. */
1829 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1830 &wait_queue
.head
, head
) {
1831 cds_list_del(&wait_node
->head
);
1836 DBG("Dispatch thread dying");
1839 ERR("Health error occurred in %s", __func__
);
1841 health_unregister(health_sessiond
);
1846 * This thread manage application registration.
1848 static void *thread_registration_apps(void *data
)
1850 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1851 uint32_t revents
, nb_fd
;
1852 struct lttng_poll_event events
;
1854 * Get allocated in this thread, enqueued to a global queue, dequeued and
1855 * freed in the manage apps thread.
1857 struct ust_command
*ust_cmd
= NULL
;
1859 DBG("[thread] Manage application registration started");
1861 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1863 if (testpoint(sessiond_thread_registration_apps
)) {
1864 goto error_testpoint
;
1867 ret
= lttcomm_listen_unix_sock(apps_sock
);
1873 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1874 * more will be added to this poll set.
1876 ret
= sessiond_set_thread_pollset(&events
, 2);
1878 goto error_create_poll
;
1881 /* Add the application registration socket */
1882 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1884 goto error_poll_add
;
1887 /* Notify all applications to register */
1888 ret
= notify_ust_apps(1);
1890 ERR("Failed to notify applications or create the wait shared memory.\n"
1891 "Execution continues but there might be problem for already\n"
1892 "running applications that wishes to register.");
1896 DBG("Accepting application registration");
1898 /* Inifinite blocking call, waiting for transmission */
1900 health_poll_entry();
1901 ret
= lttng_poll_wait(&events
, -1);
1905 * Restart interrupted system call.
1907 if (errno
== EINTR
) {
1915 for (i
= 0; i
< nb_fd
; i
++) {
1916 health_code_update();
1918 /* Fetch once the poll data */
1919 revents
= LTTNG_POLL_GETEV(&events
, i
);
1920 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1922 /* Thread quit pipe has been closed. Killing thread. */
1923 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1929 /* Event on the registration socket */
1930 if (pollfd
== apps_sock
) {
1931 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1932 ERR("Register apps socket poll error");
1934 } else if (revents
& LPOLLIN
) {
1935 sock
= lttcomm_accept_unix_sock(apps_sock
);
1941 * Set socket timeout for both receiving and ending.
1942 * app_socket_timeout is in seconds, whereas
1943 * lttcomm_setsockopt_rcv_timeout and
1944 * lttcomm_setsockopt_snd_timeout expect msec as
1947 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1948 app_socket_timeout
* 1000);
1949 (void) lttcomm_setsockopt_snd_timeout(sock
,
1950 app_socket_timeout
* 1000);
1953 * Set the CLOEXEC flag. Return code is useless because
1954 * either way, the show must go on.
1956 (void) utils_set_fd_cloexec(sock
);
1958 /* Create UST registration command for enqueuing */
1959 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1960 if (ust_cmd
== NULL
) {
1961 PERROR("ust command zmalloc");
1966 * Using message-based transmissions to ensure we don't
1967 * have to deal with partially received messages.
1969 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1971 ERR("Exhausted file descriptors allowed for applications.");
1981 health_code_update();
1982 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
1985 /* Close socket of the application. */
1990 lttng_fd_put(LTTNG_FD_APPS
, 1);
1994 health_code_update();
1996 ust_cmd
->sock
= sock
;
1999 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2000 " gid:%d sock:%d name:%s (version %d.%d)",
2001 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2002 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2003 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2004 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2007 * Lock free enqueue the registration request. The red pill
2008 * has been taken! This apps will be part of the *system*.
2010 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
2013 * Wake the registration queue futex. Implicit memory
2014 * barrier with the exchange in cds_wfq_enqueue.
2016 futex_nto1_wake(&ust_cmd_queue
.futex
);
2026 ERR("Health error occurred in %s", __func__
);
2029 /* Notify that the registration thread is gone */
2032 if (apps_sock
>= 0) {
2033 ret
= close(apps_sock
);
2043 lttng_fd_put(LTTNG_FD_APPS
, 1);
2045 unlink(apps_unix_sock_path
);
2048 lttng_poll_clean(&events
);
2052 DBG("UST Registration thread cleanup complete");
2053 health_unregister(health_sessiond
);
2059 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2060 * exec or it will fails.
2062 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2065 struct timespec timeout
;
2067 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2068 consumer_data
->consumer_thread_is_ready
= 0;
2070 /* Setup pthread condition */
2071 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2074 PERROR("pthread_condattr_init consumer data");
2079 * Set the monotonic clock in order to make sure we DO NOT jump in time
2080 * between the clock_gettime() call and the timedwait call. See bug #324
2081 * for a more details and how we noticed it.
2083 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2086 PERROR("pthread_condattr_setclock consumer data");
2090 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2093 PERROR("pthread_cond_init consumer data");
2097 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2100 PERROR("pthread_create consumer");
2105 /* We are about to wait on a pthread condition */
2106 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2108 /* Get time for sem_timedwait absolute timeout */
2109 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2111 * Set the timeout for the condition timed wait even if the clock gettime
2112 * call fails since we might loop on that call and we want to avoid to
2113 * increment the timeout too many times.
2115 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2118 * The following loop COULD be skipped in some conditions so this is why we
2119 * set ret to 0 in order to make sure at least one round of the loop is
2125 * Loop until the condition is reached or when a timeout is reached. Note
2126 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2127 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2128 * possible. This loop does not take any chances and works with both of
2131 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2132 if (clock_ret
< 0) {
2133 PERROR("clock_gettime spawn consumer");
2134 /* Infinite wait for the consumerd thread to be ready */
2135 ret
= pthread_cond_wait(&consumer_data
->cond
,
2136 &consumer_data
->cond_mutex
);
2138 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2139 &consumer_data
->cond_mutex
, &timeout
);
2143 /* Release the pthread condition */
2144 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2148 if (ret
== ETIMEDOUT
) {
2150 * Call has timed out so we kill the kconsumerd_thread and return
2153 ERR("Condition timed out. The consumer thread was never ready."
2155 ret
= pthread_cancel(consumer_data
->thread
);
2157 PERROR("pthread_cancel consumer thread");
2160 PERROR("pthread_cond_wait failed consumer thread");
2165 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2166 if (consumer_data
->pid
== 0) {
2167 ERR("Consumerd did not start");
2168 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2171 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2180 * Join consumer thread
2182 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2186 /* Consumer pid must be a real one. */
2187 if (consumer_data
->pid
> 0) {
2189 ret
= kill(consumer_data
->pid
, SIGTERM
);
2191 ERR("Error killing consumer daemon");
2194 return pthread_join(consumer_data
->thread
, &status
);
2201 * Fork and exec a consumer daemon (consumerd).
2203 * Return pid if successful else -1.
2205 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2209 const char *consumer_to_use
;
2210 const char *verbosity
;
2213 DBG("Spawning consumerd");
2220 if (opt_verbose_consumer
) {
2221 verbosity
= "--verbose";
2223 verbosity
= "--quiet";
2225 switch (consumer_data
->type
) {
2226 case LTTNG_CONSUMER_KERNEL
:
2228 * Find out which consumerd to execute. We will first try the
2229 * 64-bit path, then the sessiond's installation directory, and
2230 * fallback on the 32-bit one,
2232 DBG3("Looking for a kernel consumer at these locations:");
2233 DBG3(" 1) %s", consumerd64_bin
);
2234 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2235 DBG3(" 3) %s", consumerd32_bin
);
2236 if (stat(consumerd64_bin
, &st
) == 0) {
2237 DBG3("Found location #1");
2238 consumer_to_use
= consumerd64_bin
;
2239 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2240 DBG3("Found location #2");
2241 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2242 } else if (stat(consumerd32_bin
, &st
) == 0) {
2243 DBG3("Found location #3");
2244 consumer_to_use
= consumerd32_bin
;
2246 DBG("Could not find any valid consumerd executable");
2249 DBG("Using kernel consumer at: %s", consumer_to_use
);
2250 execl(consumer_to_use
,
2251 "lttng-consumerd", verbosity
, "-k",
2252 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2253 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2254 "--group", tracing_group_name
,
2257 case LTTNG_CONSUMER64_UST
:
2259 char *tmpnew
= NULL
;
2261 if (consumerd64_libdir
[0] != '\0') {
2265 tmp
= getenv("LD_LIBRARY_PATH");
2269 tmplen
= strlen("LD_LIBRARY_PATH=")
2270 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2271 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2276 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2277 strcat(tmpnew
, consumerd64_libdir
);
2278 if (tmp
[0] != '\0') {
2279 strcat(tmpnew
, ":");
2280 strcat(tmpnew
, tmp
);
2282 ret
= putenv(tmpnew
);
2289 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2290 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2291 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2292 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2293 "--group", tracing_group_name
,
2295 if (consumerd64_libdir
[0] != '\0') {
2303 case LTTNG_CONSUMER32_UST
:
2305 char *tmpnew
= NULL
;
2307 if (consumerd32_libdir
[0] != '\0') {
2311 tmp
= getenv("LD_LIBRARY_PATH");
2315 tmplen
= strlen("LD_LIBRARY_PATH=")
2316 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2317 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2322 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2323 strcat(tmpnew
, consumerd32_libdir
);
2324 if (tmp
[0] != '\0') {
2325 strcat(tmpnew
, ":");
2326 strcat(tmpnew
, tmp
);
2328 ret
= putenv(tmpnew
);
2335 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2336 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2337 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2338 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2339 "--group", tracing_group_name
,
2341 if (consumerd32_libdir
[0] != '\0') {
2350 PERROR("unknown consumer type");
2354 PERROR("kernel start consumer exec");
2357 } else if (pid
> 0) {
2360 PERROR("start consumer fork");
2368 * Spawn the consumerd daemon and session daemon thread.
2370 static int start_consumerd(struct consumer_data
*consumer_data
)
2375 * Set the listen() state on the socket since there is a possible race
2376 * between the exec() of the consumer daemon and this call if place in the
2377 * consumer thread. See bug #366 for more details.
2379 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2384 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2385 if (consumer_data
->pid
!= 0) {
2386 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2390 ret
= spawn_consumerd(consumer_data
);
2392 ERR("Spawning consumerd failed");
2393 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2397 /* Setting up the consumer_data pid */
2398 consumer_data
->pid
= ret
;
2399 DBG2("Consumer pid %d", consumer_data
->pid
);
2400 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2402 DBG2("Spawning consumer control thread");
2403 ret
= spawn_consumer_thread(consumer_data
);
2405 ERR("Fatal error spawning consumer control thread");
2413 /* Cleanup already created sockets on error. */
2414 if (consumer_data
->err_sock
>= 0) {
2417 err
= close(consumer_data
->err_sock
);
2419 PERROR("close consumer data error socket");
2426 * Setup necessary data for kernel tracer action.
2428 static int init_kernel_tracer(void)
2432 /* Modprobe lttng kernel modules */
2433 ret
= modprobe_lttng_control();
2438 /* Open debugfs lttng */
2439 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2440 if (kernel_tracer_fd
< 0) {
2441 DBG("Failed to open %s", module_proc_lttng
);
2446 /* Validate kernel version */
2447 ret
= kernel_validate_version(kernel_tracer_fd
);
2452 ret
= modprobe_lttng_data();
2457 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2461 modprobe_remove_lttng_control();
2462 ret
= close(kernel_tracer_fd
);
2466 kernel_tracer_fd
= -1;
2467 return LTTNG_ERR_KERN_VERSION
;
2470 ret
= close(kernel_tracer_fd
);
2476 modprobe_remove_lttng_control();
2479 WARN("No kernel tracer available");
2480 kernel_tracer_fd
= -1;
2482 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2484 return LTTNG_ERR_KERN_NA
;
2490 * Copy consumer output from the tracing session to the domain session. The
2491 * function also applies the right modification on a per domain basis for the
2492 * trace files destination directory.
2494 * Should *NOT* be called with RCU read-side lock held.
2496 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2499 const char *dir_name
;
2500 struct consumer_output
*consumer
;
2503 assert(session
->consumer
);
2506 case LTTNG_DOMAIN_KERNEL
:
2507 DBG3("Copying tracing session consumer output in kernel session");
2509 * XXX: We should audit the session creation and what this function
2510 * does "extra" in order to avoid a destroy since this function is used
2511 * in the domain session creation (kernel and ust) only. Same for UST
2514 if (session
->kernel_session
->consumer
) {
2515 consumer_destroy_output(session
->kernel_session
->consumer
);
2517 session
->kernel_session
->consumer
=
2518 consumer_copy_output(session
->consumer
);
2519 /* Ease our life a bit for the next part */
2520 consumer
= session
->kernel_session
->consumer
;
2521 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2523 case LTTNG_DOMAIN_JUL
:
2524 case LTTNG_DOMAIN_UST
:
2525 DBG3("Copying tracing session consumer output in UST session");
2526 if (session
->ust_session
->consumer
) {
2527 consumer_destroy_output(session
->ust_session
->consumer
);
2529 session
->ust_session
->consumer
=
2530 consumer_copy_output(session
->consumer
);
2531 /* Ease our life a bit for the next part */
2532 consumer
= session
->ust_session
->consumer
;
2533 dir_name
= DEFAULT_UST_TRACE_DIR
;
2536 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2540 /* Append correct directory to subdir */
2541 strncat(consumer
->subdir
, dir_name
,
2542 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2543 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2552 * Create an UST session and add it to the session ust list.
2554 * Should *NOT* be called with RCU read-side lock held.
2556 static int create_ust_session(struct ltt_session
*session
,
2557 struct lttng_domain
*domain
)
2560 struct ltt_ust_session
*lus
= NULL
;
2564 assert(session
->consumer
);
2566 switch (domain
->type
) {
2567 case LTTNG_DOMAIN_JUL
:
2568 case LTTNG_DOMAIN_UST
:
2571 ERR("Unknown UST domain on create session %d", domain
->type
);
2572 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2576 DBG("Creating UST session");
2578 lus
= trace_ust_create_session(session
->id
);
2580 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2584 lus
->uid
= session
->uid
;
2585 lus
->gid
= session
->gid
;
2586 lus
->output_traces
= session
->output_traces
;
2587 lus
->snapshot_mode
= session
->snapshot_mode
;
2588 lus
->live_timer_interval
= session
->live_timer
;
2589 session
->ust_session
= lus
;
2591 /* Copy session output to the newly created UST session */
2592 ret
= copy_session_consumer(domain
->type
, session
);
2593 if (ret
!= LTTNG_OK
) {
2601 session
->ust_session
= NULL
;
2606 * Create a kernel tracer session then create the default channel.
2608 static int create_kernel_session(struct ltt_session
*session
)
2612 DBG("Creating kernel session");
2614 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2616 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2620 /* Code flow safety */
2621 assert(session
->kernel_session
);
2623 /* Copy session output to the newly created Kernel session */
2624 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2625 if (ret
!= LTTNG_OK
) {
2629 /* Create directory(ies) on local filesystem. */
2630 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2631 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2632 ret
= run_as_mkdir_recursive(
2633 session
->kernel_session
->consumer
->dst
.trace_path
,
2634 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2636 if (ret
!= -EEXIST
) {
2637 ERR("Trace directory creation error");
2643 session
->kernel_session
->uid
= session
->uid
;
2644 session
->kernel_session
->gid
= session
->gid
;
2645 session
->kernel_session
->output_traces
= session
->output_traces
;
2646 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2651 trace_kernel_destroy_session(session
->kernel_session
);
2652 session
->kernel_session
= NULL
;
2657 * Count number of session permitted by uid/gid.
2659 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2662 struct ltt_session
*session
;
2664 DBG("Counting number of available session for UID %d GID %d",
2666 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2668 * Only list the sessions the user can control.
2670 if (!session_access_ok(session
, uid
, gid
)) {
2679 * Process the command requested by the lttng client within the command
2680 * context structure. This function make sure that the return structure (llm)
2681 * is set and ready for transmission before returning.
2683 * Return any error encountered or 0 for success.
2685 * "sock" is only used for special-case var. len data.
2687 * Should *NOT* be called with RCU read-side lock held.
2689 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2693 int need_tracing_session
= 1;
2696 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2700 switch (cmd_ctx
->lsm
->cmd_type
) {
2701 case LTTNG_CREATE_SESSION
:
2702 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2703 case LTTNG_CREATE_SESSION_LIVE
:
2704 case LTTNG_DESTROY_SESSION
:
2705 case LTTNG_LIST_SESSIONS
:
2706 case LTTNG_LIST_DOMAINS
:
2707 case LTTNG_START_TRACE
:
2708 case LTTNG_STOP_TRACE
:
2709 case LTTNG_DATA_PENDING
:
2710 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2711 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2712 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2713 case LTTNG_SNAPSHOT_RECORD
:
2720 if (opt_no_kernel
&& need_domain
2721 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2723 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2725 ret
= LTTNG_ERR_KERN_NA
;
2730 /* Deny register consumer if we already have a spawned consumer. */
2731 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2732 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2733 if (kconsumer_data
.pid
> 0) {
2734 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2735 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2738 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2742 * Check for command that don't needs to allocate a returned payload. We do
2743 * this here so we don't have to make the call for no payload at each
2746 switch(cmd_ctx
->lsm
->cmd_type
) {
2747 case LTTNG_LIST_SESSIONS
:
2748 case LTTNG_LIST_TRACEPOINTS
:
2749 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2750 case LTTNG_LIST_DOMAINS
:
2751 case LTTNG_LIST_CHANNELS
:
2752 case LTTNG_LIST_EVENTS
:
2755 /* Setup lttng message with no payload */
2756 ret
= setup_lttng_msg(cmd_ctx
, 0);
2758 /* This label does not try to unlock the session */
2759 goto init_setup_error
;
2763 /* Commands that DO NOT need a session. */
2764 switch (cmd_ctx
->lsm
->cmd_type
) {
2765 case LTTNG_CREATE_SESSION
:
2766 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2767 case LTTNG_CREATE_SESSION_LIVE
:
2768 case LTTNG_CALIBRATE
:
2769 case LTTNG_LIST_SESSIONS
:
2770 case LTTNG_LIST_TRACEPOINTS
:
2771 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2772 need_tracing_session
= 0;
2775 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2777 * We keep the session list lock across _all_ commands
2778 * for now, because the per-session lock does not
2779 * handle teardown properly.
2781 session_lock_list();
2782 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2783 if (cmd_ctx
->session
== NULL
) {
2784 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2787 /* Acquire lock for the session */
2788 session_lock(cmd_ctx
->session
);
2798 * Check domain type for specific "pre-action".
2800 switch (cmd_ctx
->lsm
->domain
.type
) {
2801 case LTTNG_DOMAIN_KERNEL
:
2803 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2807 /* Kernel tracer check */
2808 if (kernel_tracer_fd
== -1) {
2809 /* Basically, load kernel tracer modules */
2810 ret
= init_kernel_tracer();
2816 /* Consumer is in an ERROR state. Report back to client */
2817 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2818 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2822 /* Need a session for kernel command */
2823 if (need_tracing_session
) {
2824 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2825 ret
= create_kernel_session(cmd_ctx
->session
);
2827 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2832 /* Start the kernel consumer daemon */
2833 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2834 if (kconsumer_data
.pid
== 0 &&
2835 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2836 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2837 ret
= start_consumerd(&kconsumer_data
);
2839 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2842 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2844 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2848 * The consumer was just spawned so we need to add the socket to
2849 * the consumer output of the session if exist.
2851 ret
= consumer_create_socket(&kconsumer_data
,
2852 cmd_ctx
->session
->kernel_session
->consumer
);
2859 case LTTNG_DOMAIN_JUL
:
2860 case LTTNG_DOMAIN_UST
:
2862 if (!ust_app_supported()) {
2863 ret
= LTTNG_ERR_NO_UST
;
2866 /* Consumer is in an ERROR state. Report back to client */
2867 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2868 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2872 if (need_tracing_session
) {
2873 /* Create UST session if none exist. */
2874 if (cmd_ctx
->session
->ust_session
== NULL
) {
2875 ret
= create_ust_session(cmd_ctx
->session
,
2876 &cmd_ctx
->lsm
->domain
);
2877 if (ret
!= LTTNG_OK
) {
2882 /* Start the UST consumer daemons */
2884 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2885 if (consumerd64_bin
[0] != '\0' &&
2886 ustconsumer64_data
.pid
== 0 &&
2887 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2888 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2889 ret
= start_consumerd(&ustconsumer64_data
);
2891 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2892 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2896 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2897 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2899 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2903 * Setup socket for consumer 64 bit. No need for atomic access
2904 * since it was set above and can ONLY be set in this thread.
2906 ret
= consumer_create_socket(&ustconsumer64_data
,
2907 cmd_ctx
->session
->ust_session
->consumer
);
2913 if (consumerd32_bin
[0] != '\0' &&
2914 ustconsumer32_data
.pid
== 0 &&
2915 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2916 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2917 ret
= start_consumerd(&ustconsumer32_data
);
2919 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2920 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2924 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2925 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2927 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2931 * Setup socket for consumer 64 bit. No need for atomic access
2932 * since it was set above and can ONLY be set in this thread.
2934 ret
= consumer_create_socket(&ustconsumer32_data
,
2935 cmd_ctx
->session
->ust_session
->consumer
);
2947 /* Validate consumer daemon state when start/stop trace command */
2948 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
2949 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
2950 switch (cmd_ctx
->lsm
->domain
.type
) {
2951 case LTTNG_DOMAIN_JUL
:
2952 case LTTNG_DOMAIN_UST
:
2953 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
2954 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2958 case LTTNG_DOMAIN_KERNEL
:
2959 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
2960 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2968 * Check that the UID or GID match that of the tracing session.
2969 * The root user can interact with all sessions.
2971 if (need_tracing_session
) {
2972 if (!session_access_ok(cmd_ctx
->session
,
2973 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
2974 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
2975 ret
= LTTNG_ERR_EPERM
;
2981 * Send relayd information to consumer as soon as we have a domain and a
2984 if (cmd_ctx
->session
&& need_domain
) {
2986 * Setup relayd if not done yet. If the relayd information was already
2987 * sent to the consumer, this call will gracefully return.
2989 ret
= cmd_setup_relayd(cmd_ctx
->session
);
2990 if (ret
!= LTTNG_OK
) {
2995 /* Process by command type */
2996 switch (cmd_ctx
->lsm
->cmd_type
) {
2997 case LTTNG_ADD_CONTEXT
:
2999 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3000 cmd_ctx
->lsm
->u
.context
.channel_name
,
3001 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3004 case LTTNG_DISABLE_CHANNEL
:
3006 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3007 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3010 case LTTNG_DISABLE_EVENT
:
3012 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3013 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3014 cmd_ctx
->lsm
->u
.disable
.name
);
3017 case LTTNG_DISABLE_ALL_EVENT
:
3019 DBG("Disabling all events");
3021 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3022 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3025 case LTTNG_ENABLE_CHANNEL
:
3027 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3028 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3031 case LTTNG_ENABLE_EVENT
:
3033 struct lttng_event_exclusion
*exclusion
= NULL
;
3034 struct lttng_filter_bytecode
*bytecode
= NULL
;
3036 /* Handle exclusion events and receive it from the client. */
3037 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3038 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3040 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3041 (count
* LTTNG_SYMBOL_NAME_LEN
));
3043 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3047 DBG("Receiving var len exclusion event list from client ...");
3048 exclusion
->count
= count
;
3049 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3050 count
* LTTNG_SYMBOL_NAME_LEN
);
3052 DBG("Nothing recv() from client var len data... continuing");
3055 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3060 /* Handle filter and get bytecode from client. */
3061 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3062 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3064 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3065 ret
= LTTNG_ERR_FILTER_INVAL
;
3070 bytecode
= zmalloc(bytecode_len
);
3073 ret
= LTTNG_ERR_FILTER_NOMEM
;
3077 /* Receive var. len. data */
3078 DBG("Receiving var len filter's bytecode from client ...");
3079 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3081 DBG("Nothing recv() from client car len data... continuing");
3085 ret
= LTTNG_ERR_FILTER_INVAL
;
3089 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3092 ret
= LTTNG_ERR_FILTER_INVAL
;
3097 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3098 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3099 &cmd_ctx
->lsm
->u
.enable
.event
, bytecode
, exclusion
,
3100 kernel_poll_pipe
[1]);
3103 case LTTNG_ENABLE_ALL_EVENT
:
3105 DBG("Enabling all events");
3107 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3108 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3109 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, kernel_poll_pipe
[1]);
3112 case LTTNG_LIST_TRACEPOINTS
:
3114 struct lttng_event
*events
;
3117 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3118 if (nb_events
< 0) {
3119 /* Return value is a negative lttng_error_code. */
3125 * Setup lttng message with payload size set to the event list size in
3126 * bytes and then copy list into the llm payload.
3128 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3134 /* Copy event list into message payload */
3135 memcpy(cmd_ctx
->llm
->payload
, events
,
3136 sizeof(struct lttng_event
) * nb_events
);
3143 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3145 struct lttng_event_field
*fields
;
3148 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3150 if (nb_fields
< 0) {
3151 /* Return value is a negative lttng_error_code. */
3157 * Setup lttng message with payload size set to the event list size in
3158 * bytes and then copy list into the llm payload.
3160 ret
= setup_lttng_msg(cmd_ctx
,
3161 sizeof(struct lttng_event_field
) * nb_fields
);
3167 /* Copy event list into message payload */
3168 memcpy(cmd_ctx
->llm
->payload
, fields
,
3169 sizeof(struct lttng_event_field
) * nb_fields
);
3176 case LTTNG_SET_CONSUMER_URI
:
3179 struct lttng_uri
*uris
;
3181 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3182 len
= nb_uri
* sizeof(struct lttng_uri
);
3185 ret
= LTTNG_ERR_INVALID
;
3189 uris
= zmalloc(len
);
3191 ret
= LTTNG_ERR_FATAL
;
3195 /* Receive variable len data */
3196 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3197 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3199 DBG("No URIs received from client... continuing");
3201 ret
= LTTNG_ERR_SESSION_FAIL
;
3206 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3208 if (ret
!= LTTNG_OK
) {
3214 * XXX: 0 means that this URI should be applied on the session. Should
3215 * be a DOMAIN enuam.
3217 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3218 /* Add the URI for the UST session if a consumer is present. */
3219 if (cmd_ctx
->session
->ust_session
&&
3220 cmd_ctx
->session
->ust_session
->consumer
) {
3221 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3223 } else if (cmd_ctx
->session
->kernel_session
&&
3224 cmd_ctx
->session
->kernel_session
->consumer
) {
3225 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3226 cmd_ctx
->session
, nb_uri
, uris
);
3234 case LTTNG_START_TRACE
:
3236 ret
= cmd_start_trace(cmd_ctx
->session
);
3239 case LTTNG_STOP_TRACE
:
3241 ret
= cmd_stop_trace(cmd_ctx
->session
);
3244 case LTTNG_CREATE_SESSION
:
3247 struct lttng_uri
*uris
= NULL
;
3249 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3250 len
= nb_uri
* sizeof(struct lttng_uri
);
3253 uris
= zmalloc(len
);
3255 ret
= LTTNG_ERR_FATAL
;
3259 /* Receive variable len data */
3260 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3261 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3263 DBG("No URIs received from client... continuing");
3265 ret
= LTTNG_ERR_SESSION_FAIL
;
3270 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3271 DBG("Creating session with ONE network URI is a bad call");
3272 ret
= LTTNG_ERR_SESSION_FAIL
;
3278 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3279 &cmd_ctx
->creds
, 0);
3285 case LTTNG_DESTROY_SESSION
:
3287 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3289 /* Set session to NULL so we do not unlock it after free. */
3290 cmd_ctx
->session
= NULL
;
3293 case LTTNG_LIST_DOMAINS
:
3296 struct lttng_domain
*domains
;
3298 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3300 /* Return value is a negative lttng_error_code. */
3305 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3311 /* Copy event list into message payload */
3312 memcpy(cmd_ctx
->llm
->payload
, domains
,
3313 nb_dom
* sizeof(struct lttng_domain
));
3320 case LTTNG_LIST_CHANNELS
:
3323 struct lttng_channel
*channels
;
3325 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3326 cmd_ctx
->session
, &channels
);
3328 /* Return value is a negative lttng_error_code. */
3333 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3339 /* Copy event list into message payload */
3340 memcpy(cmd_ctx
->llm
->payload
, channels
,
3341 nb_chan
* sizeof(struct lttng_channel
));
3348 case LTTNG_LIST_EVENTS
:
3351 struct lttng_event
*events
= NULL
;
3353 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3354 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3356 /* Return value is a negative lttng_error_code. */
3361 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3367 /* Copy event list into message payload */
3368 memcpy(cmd_ctx
->llm
->payload
, events
,
3369 nb_event
* sizeof(struct lttng_event
));
3376 case LTTNG_LIST_SESSIONS
:
3378 unsigned int nr_sessions
;
3380 session_lock_list();
3381 nr_sessions
= lttng_sessions_count(
3382 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3383 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3385 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3387 session_unlock_list();
3391 /* Filled the session array */
3392 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3393 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3394 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3396 session_unlock_list();
3401 case LTTNG_CALIBRATE
:
3403 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3404 &cmd_ctx
->lsm
->u
.calibrate
);
3407 case LTTNG_REGISTER_CONSUMER
:
3409 struct consumer_data
*cdata
;
3411 switch (cmd_ctx
->lsm
->domain
.type
) {
3412 case LTTNG_DOMAIN_KERNEL
:
3413 cdata
= &kconsumer_data
;
3416 ret
= LTTNG_ERR_UND
;
3420 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3421 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3424 case LTTNG_DATA_PENDING
:
3426 ret
= cmd_data_pending(cmd_ctx
->session
);
3429 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3431 struct lttcomm_lttng_output_id reply
;
3433 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3434 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3435 if (ret
!= LTTNG_OK
) {
3439 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3444 /* Copy output list into message payload */
3445 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3449 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3451 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3452 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3455 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3458 struct lttng_snapshot_output
*outputs
= NULL
;
3460 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3461 if (nb_output
< 0) {
3466 ret
= setup_lttng_msg(cmd_ctx
,
3467 nb_output
* sizeof(struct lttng_snapshot_output
));
3474 /* Copy output list into message payload */
3475 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3476 nb_output
* sizeof(struct lttng_snapshot_output
));
3483 case LTTNG_SNAPSHOT_RECORD
:
3485 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3486 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3487 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3490 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3493 struct lttng_uri
*uris
= NULL
;
3495 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3496 len
= nb_uri
* sizeof(struct lttng_uri
);
3499 uris
= zmalloc(len
);
3501 ret
= LTTNG_ERR_FATAL
;
3505 /* Receive variable len data */
3506 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3507 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3509 DBG("No URIs received from client... continuing");
3511 ret
= LTTNG_ERR_SESSION_FAIL
;
3516 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3517 DBG("Creating session with ONE network URI is a bad call");
3518 ret
= LTTNG_ERR_SESSION_FAIL
;
3524 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3525 nb_uri
, &cmd_ctx
->creds
);
3529 case LTTNG_CREATE_SESSION_LIVE
:
3532 struct lttng_uri
*uris
= NULL
;
3534 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3535 len
= nb_uri
* sizeof(struct lttng_uri
);
3538 uris
= zmalloc(len
);
3540 ret
= LTTNG_ERR_FATAL
;
3544 /* Receive variable len data */
3545 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3546 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3548 DBG("No URIs received from client... continuing");
3550 ret
= LTTNG_ERR_SESSION_FAIL
;
3555 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3556 DBG("Creating session with ONE network URI is a bad call");
3557 ret
= LTTNG_ERR_SESSION_FAIL
;
3563 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3564 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3569 ret
= LTTNG_ERR_UND
;
3574 if (cmd_ctx
->llm
== NULL
) {
3575 DBG("Missing llm structure. Allocating one.");
3576 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3580 /* Set return code */
3581 cmd_ctx
->llm
->ret_code
= ret
;
3583 if (cmd_ctx
->session
) {
3584 session_unlock(cmd_ctx
->session
);
3586 if (need_tracing_session
) {
3587 session_unlock_list();
3594 * Thread managing health check socket.
3596 static void *thread_manage_health(void *data
)
3598 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3599 uint32_t revents
, nb_fd
;
3600 struct lttng_poll_event events
;
3601 struct health_comm_msg msg
;
3602 struct health_comm_reply reply
;
3604 DBG("[thread] Manage health check started");
3606 rcu_register_thread();
3608 /* We might hit an error path before this is created. */
3609 lttng_poll_init(&events
);
3611 /* Create unix socket */
3612 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3614 ERR("Unable to create health check Unix socket");
3620 /* lttng health client socket path permissions */
3621 ret
= chown(health_unix_sock_path
, 0,
3622 utils_get_group_id(tracing_group_name
));
3624 ERR("Unable to set group on %s", health_unix_sock_path
);
3630 ret
= chmod(health_unix_sock_path
,
3631 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3633 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3641 * Set the CLOEXEC flag. Return code is useless because either way, the
3644 (void) utils_set_fd_cloexec(sock
);
3646 ret
= lttcomm_listen_unix_sock(sock
);
3652 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3653 * more will be added to this poll set.
3655 ret
= sessiond_set_thread_pollset(&events
, 2);
3660 /* Add the application registration socket */
3661 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3666 lttng_sessiond_notify_ready();
3669 DBG("Health check ready");
3671 /* Inifinite blocking call, waiting for transmission */
3673 ret
= lttng_poll_wait(&events
, -1);
3676 * Restart interrupted system call.
3678 if (errno
== EINTR
) {
3686 for (i
= 0; i
< nb_fd
; i
++) {
3687 /* Fetch once the poll data */
3688 revents
= LTTNG_POLL_GETEV(&events
, i
);
3689 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3691 /* Thread quit pipe has been closed. Killing thread. */
3692 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3698 /* Event on the registration socket */
3699 if (pollfd
== sock
) {
3700 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3701 ERR("Health socket poll error");
3707 new_sock
= lttcomm_accept_unix_sock(sock
);
3713 * Set the CLOEXEC flag. Return code is useless because either way, the
3716 (void) utils_set_fd_cloexec(new_sock
);
3718 DBG("Receiving data from client for health...");
3719 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3721 DBG("Nothing recv() from client... continuing");
3722 ret
= close(new_sock
);
3730 rcu_thread_online();
3733 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3735 * health_check_state returns 0 if health is
3738 if (!health_check_state(health_sessiond
, i
)) {
3739 reply
.ret_code
|= 1ULL << i
;
3743 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3745 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3747 ERR("Failed to send health data back to client");
3750 /* End of transmission */
3751 ret
= close(new_sock
);
3761 ERR("Health error occurred in %s", __func__
);
3763 DBG("Health check thread dying");
3764 unlink(health_unix_sock_path
);
3772 lttng_poll_clean(&events
);
3774 rcu_unregister_thread();
3779 * This thread manage all clients request using the unix client socket for
3782 static void *thread_manage_clients(void *data
)
3784 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3786 uint32_t revents
, nb_fd
;
3787 struct command_ctx
*cmd_ctx
= NULL
;
3788 struct lttng_poll_event events
;
3790 DBG("[thread] Manage client started");
3792 rcu_register_thread();
3794 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3796 health_code_update();
3798 ret
= lttcomm_listen_unix_sock(client_sock
);
3804 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3805 * more will be added to this poll set.
3807 ret
= sessiond_set_thread_pollset(&events
, 2);
3809 goto error_create_poll
;
3812 /* Add the application registration socket */
3813 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3818 lttng_sessiond_notify_ready();
3820 /* This testpoint is after we signal readiness to the parent. */
3821 if (testpoint(sessiond_thread_manage_clients
)) {
3825 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
3829 health_code_update();
3832 DBG("Accepting client command ...");
3834 /* Inifinite blocking call, waiting for transmission */
3836 health_poll_entry();
3837 ret
= lttng_poll_wait(&events
, -1);
3841 * Restart interrupted system call.
3843 if (errno
== EINTR
) {
3851 for (i
= 0; i
< nb_fd
; i
++) {
3852 /* Fetch once the poll data */
3853 revents
= LTTNG_POLL_GETEV(&events
, i
);
3854 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3856 health_code_update();
3858 /* Thread quit pipe has been closed. Killing thread. */
3859 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3865 /* Event on the registration socket */
3866 if (pollfd
== client_sock
) {
3867 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3868 ERR("Client socket poll error");
3874 DBG("Wait for client response");
3876 health_code_update();
3878 sock
= lttcomm_accept_unix_sock(client_sock
);
3884 * Set the CLOEXEC flag. Return code is useless because either way, the
3887 (void) utils_set_fd_cloexec(sock
);
3889 /* Set socket option for credentials retrieval */
3890 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
3895 /* Allocate context command to process the client request */
3896 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
3897 if (cmd_ctx
== NULL
) {
3898 PERROR("zmalloc cmd_ctx");
3902 /* Allocate data buffer for reception */
3903 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
3904 if (cmd_ctx
->lsm
== NULL
) {
3905 PERROR("zmalloc cmd_ctx->lsm");
3909 cmd_ctx
->llm
= NULL
;
3910 cmd_ctx
->session
= NULL
;
3912 health_code_update();
3915 * Data is received from the lttng client. The struct
3916 * lttcomm_session_msg (lsm) contains the command and data request of
3919 DBG("Receiving data from client ...");
3920 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
3921 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
3923 DBG("Nothing recv() from client... continuing");
3929 clean_command_ctx(&cmd_ctx
);
3933 health_code_update();
3935 // TODO: Validate cmd_ctx including sanity check for
3936 // security purpose.
3938 rcu_thread_online();
3940 * This function dispatch the work to the kernel or userspace tracer
3941 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3942 * informations for the client. The command context struct contains
3943 * everything this function may needs.
3945 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
3946 rcu_thread_offline();
3954 * TODO: Inform client somehow of the fatal error. At
3955 * this point, ret < 0 means that a zmalloc failed
3956 * (ENOMEM). Error detected but still accept
3957 * command, unless a socket error has been
3960 clean_command_ctx(&cmd_ctx
);
3964 health_code_update();
3966 DBG("Sending response (size: %d, retcode: %s)",
3967 cmd_ctx
->lttng_msg_size
,
3968 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
3969 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
3971 ERR("Failed to send data back to client");
3974 /* End of transmission */
3981 clean_command_ctx(&cmd_ctx
);
3983 health_code_update();
3995 lttng_poll_clean(&events
);
3996 clean_command_ctx(&cmd_ctx
);
4000 unlink(client_unix_sock_path
);
4001 if (client_sock
>= 0) {
4002 ret
= close(client_sock
);
4010 ERR("Health error occurred in %s", __func__
);
4013 health_unregister(health_sessiond
);
4015 DBG("Client thread dying");
4017 rcu_unregister_thread();
4023 * usage function on stderr
4025 static void usage(void)
4027 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4028 fprintf(stderr
, " -h, --help Display this usage.\n");
4029 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4030 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4031 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4032 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4033 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4034 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4035 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4036 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4037 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4038 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4039 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4040 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4041 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4042 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4043 fprintf(stderr
, " -V, --version Show version number.\n");
4044 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4045 fprintf(stderr
, " -q, --quiet No output at all.\n");
4046 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4047 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4048 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4049 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4050 fprintf(stderr
, " --jul-tcp-port JUL application registration TCP port\n");
4051 fprintf(stderr
, " -f --config Load daemon configuration file\n");
4055 * Take an option from the getopt output and set it in the right variable to be
4058 * Return 0 on success else a negative value.
4060 static int set_option(int opt
, const char *arg
, const char *optname
)
4066 fprintf(stderr
, "option %s", optname
);
4068 fprintf(stderr
, " with arg %s\n", arg
);
4072 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4075 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4081 tracing_group_name
= strdup(arg
);
4087 fprintf(stdout
, "%s\n", VERSION
);
4093 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4096 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4099 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4102 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4105 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4108 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4114 lttng_opt_quiet
= 1;
4117 /* Verbose level can increase using multiple -v */
4119 lttng_opt_verbose
= config_parse_value(arg
);
4121 lttng_opt_verbose
+= 1;
4126 opt_verbose_consumer
= config_parse_value(arg
);
4128 opt_verbose_consumer
+= 1;
4132 consumerd32_bin
= strdup(arg
);
4133 consumerd32_bin_override
= 1;
4136 consumerd32_libdir
= strdup(arg
);
4137 consumerd32_libdir_override
= 1;
4140 consumerd64_bin
= strdup(arg
);
4141 consumerd64_bin_override
= 1;
4144 consumerd64_libdir
= strdup(arg
);
4145 consumerd64_libdir_override
= 1;
4148 opt_pidfile
= strdup(arg
);
4150 case 'J': /* JUL TCP port. */
4155 v
= strtoul(arg
, NULL
, 0);
4156 if (errno
!= 0 || !isdigit(arg
[0])) {
4157 ERR("Wrong value in --jul-tcp-port parameter: %s", arg
);
4160 if (v
== 0 || v
>= 65535) {
4161 ERR("Port overflow in --jul-tcp-port parameter: %s", arg
);
4164 jul_tcp_port
= (uint32_t) v
;
4165 DBG3("JUL TCP port set to non default: %u", jul_tcp_port
);
4169 /* Unknown option or other error.
4170 * Error is printed by getopt, just return */
4178 * config_entry_handler_cb used to handle options read from a config file.
4179 * See config_entry_handler_cb comment in common/config/config.h for the
4180 * return value conventions.
4182 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4186 if (!entry
|| !entry
->name
|| !entry
->value
) {
4191 /* Check if the option is to be ignored */
4192 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4193 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4198 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4201 /* Ignore if not fully matched. */
4202 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4207 * If the option takes no argument on the command line, we have to
4208 * check if the value is "true". We support non-zero numeric values,
4211 if (!long_options
[i
].has_arg
) {
4212 ret
= config_parse_value(entry
->value
);
4215 WARN("Invalid configuration value \"%s\" for option %s",
4216 entry
->value
, entry
->name
);
4218 /* False, skip boolean config option. */
4223 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4227 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4234 * daemon configuration loading and argument parsing
4236 static int set_options(int argc
, char **argv
)
4238 int ret
= 0, c
= 0, option_index
= 0;
4239 int orig_optopt
= optopt
, orig_optind
= optind
;
4241 const char *config_path
= NULL
;
4243 optstring
= utils_generate_optstring(long_options
,
4244 sizeof(long_options
) / sizeof(struct option
));
4250 /* Check for the --config option */
4251 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4252 &option_index
)) != -1) {
4256 } else if (c
!= 'f') {
4257 /* if not equal to --config option. */
4261 config_path
= utils_expand_path(optarg
);
4263 ERR("Failed to resolve path: %s", optarg
);
4267 ret
= config_get_section_entries(config_path
, config_section_name
,
4268 config_entry_handler
, NULL
);
4271 ERR("Invalid configuration option at line %i", ret
);
4277 /* Reset getopt's global state */
4278 optopt
= orig_optopt
;
4279 optind
= orig_optind
;
4281 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
4286 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
4298 * Creates the two needed socket by the daemon.
4299 * apps_sock - The communication socket for all UST apps.
4300 * client_sock - The communication of the cli tool (lttng).
4302 static int init_daemon_socket(void)
4307 old_umask
= umask(0);
4309 /* Create client tool unix socket */
4310 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4311 if (client_sock
< 0) {
4312 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4317 /* Set the cloexec flag */
4318 ret
= utils_set_fd_cloexec(client_sock
);
4320 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4321 "Continuing but note that the consumer daemon will have a "
4322 "reference to this socket on exec()", client_sock
);
4325 /* File permission MUST be 660 */
4326 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4328 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4333 /* Create the application unix socket */
4334 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4335 if (apps_sock
< 0) {
4336 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4341 /* Set the cloexec flag */
4342 ret
= utils_set_fd_cloexec(apps_sock
);
4344 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4345 "Continuing but note that the consumer daemon will have a "
4346 "reference to this socket on exec()", apps_sock
);
4349 /* File permission MUST be 666 */
4350 ret
= chmod(apps_unix_sock_path
,
4351 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4353 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4358 DBG3("Session daemon client socket %d and application socket %d created",
4359 client_sock
, apps_sock
);
4367 * Check if the global socket is available, and if a daemon is answering at the
4368 * other side. If yes, error is returned.
4370 static int check_existing_daemon(void)
4372 /* Is there anybody out there ? */
4373 if (lttng_session_daemon_alive()) {
4381 * Set the tracing group gid onto the client socket.
4383 * Race window between mkdir and chown is OK because we are going from more
4384 * permissive (root.root) to less permissive (root.tracing).
4386 static int set_permissions(char *rundir
)
4391 gid
= utils_get_group_id(tracing_group_name
);
4393 /* Set lttng run dir */
4394 ret
= chown(rundir
, 0, gid
);
4396 ERR("Unable to set group on %s", rundir
);
4401 * Ensure all applications and tracing group can search the run
4402 * dir. Allow everyone to read the directory, since it does not
4403 * buy us anything to hide its content.
4405 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4407 ERR("Unable to set permissions on %s", rundir
);
4411 /* lttng client socket path */
4412 ret
= chown(client_unix_sock_path
, 0, gid
);
4414 ERR("Unable to set group on %s", client_unix_sock_path
);
4418 /* kconsumer error socket path */
4419 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4421 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4425 /* 64-bit ustconsumer error socket path */
4426 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4428 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4432 /* 32-bit ustconsumer compat32 error socket path */
4433 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4435 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4439 DBG("All permissions are set");
4445 * Create the lttng run directory needed for all global sockets and pipe.
4447 static int create_lttng_rundir(const char *rundir
)
4451 DBG3("Creating LTTng run directory: %s", rundir
);
4453 ret
= mkdir(rundir
, S_IRWXU
);
4455 if (errno
!= EEXIST
) {
4456 ERR("Unable to create %s", rundir
);
4468 * Setup sockets and directory needed by the kconsumerd communication with the
4471 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4475 char path
[PATH_MAX
];
4477 switch (consumer_data
->type
) {
4478 case LTTNG_CONSUMER_KERNEL
:
4479 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4481 case LTTNG_CONSUMER64_UST
:
4482 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4484 case LTTNG_CONSUMER32_UST
:
4485 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4488 ERR("Consumer type unknown");
4493 DBG2("Creating consumer directory: %s", path
);
4495 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4497 if (errno
!= EEXIST
) {
4499 ERR("Failed to create %s", path
);
4505 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4507 ERR("Unable to set group on %s", path
);
4513 /* Create the kconsumerd error unix socket */
4514 consumer_data
->err_sock
=
4515 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4516 if (consumer_data
->err_sock
< 0) {
4517 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4523 * Set the CLOEXEC flag. Return code is useless because either way, the
4526 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4528 PERROR("utils_set_fd_cloexec");
4529 /* continue anyway */
4532 /* File permission MUST be 660 */
4533 ret
= chmod(consumer_data
->err_unix_sock_path
,
4534 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4536 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4546 * Signal handler for the daemon
4548 * Simply stop all worker threads, leaving main() return gracefully after
4549 * joining all threads and calling cleanup().
4551 static void sighandler(int sig
)
4555 DBG("SIGPIPE caught");
4558 DBG("SIGINT caught");
4562 DBG("SIGTERM caught");
4566 CMM_STORE_SHARED(recv_child_signal
, 1);
4574 * Setup signal handler for :
4575 * SIGINT, SIGTERM, SIGPIPE
4577 static int set_signal_handler(void)
4580 struct sigaction sa
;
4583 if ((ret
= sigemptyset(&sigset
)) < 0) {
4584 PERROR("sigemptyset");
4588 sa
.sa_handler
= sighandler
;
4589 sa
.sa_mask
= sigset
;
4591 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4592 PERROR("sigaction");
4596 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4597 PERROR("sigaction");
4601 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4602 PERROR("sigaction");
4606 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4607 PERROR("sigaction");
4611 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4617 * Set open files limit to unlimited. This daemon can open a large number of
4618 * file descriptors in order to consumer multiple kernel traces.
4620 static void set_ulimit(void)
4625 /* The kernel does not allowed an infinite limit for open files */
4626 lim
.rlim_cur
= 65535;
4627 lim
.rlim_max
= 65535;
4629 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4631 PERROR("failed to set open files limit");
4636 * Write pidfile using the rundir and opt_pidfile.
4638 static void write_pidfile(void)
4641 char pidfile_path
[PATH_MAX
];
4646 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4648 /* Build pidfile path from rundir and opt_pidfile. */
4649 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4650 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4652 PERROR("snprintf pidfile path");
4658 * Create pid file in rundir. Return value is of no importance. The
4659 * execution will continue even though we are not able to write the file.
4661 (void) utils_create_pid_file(getpid(), pidfile_path
);
4668 * Write JUL TCP port using the rundir.
4670 static void write_julport(void)
4673 char path
[PATH_MAX
];
4677 ret
= snprintf(path
, sizeof(path
), "%s/"
4678 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
, rundir
);
4680 PERROR("snprintf julport path");
4685 * Create TCP JUL port file in rundir. Return value is of no importance.
4686 * The execution will continue even though we are not able to write the
4689 (void) utils_create_pid_file(jul_tcp_port
, path
);
4698 int main(int argc
, char **argv
)
4702 const char *home_path
, *env_app_timeout
;
4704 init_kernel_workarounds();
4706 rcu_register_thread();
4708 if ((ret
= set_signal_handler()) < 0) {
4712 setup_consumerd_path();
4714 page_size
= sysconf(_SC_PAGESIZE
);
4715 if (page_size
< 0) {
4716 PERROR("sysconf _SC_PAGESIZE");
4717 page_size
= LONG_MAX
;
4718 WARN("Fallback page size to %ld", page_size
);
4721 /* Parse arguments and load the daemon configuration file */
4723 if ((ret
= set_options(argc
, argv
)) < 0) {
4731 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
, 1);
4737 * We are in the child. Make sure all other file descriptors are
4738 * closed, in case we are called with more opened file descriptors than
4739 * the standard ones.
4741 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4746 /* Create thread quit pipe */
4747 if ((ret
= init_thread_quit_pipe()) < 0) {
4751 /* Check if daemon is UID = 0 */
4752 is_root
= !getuid();
4755 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4757 /* Create global run dir with root access */
4758 ret
= create_lttng_rundir(rundir
);
4763 if (strlen(apps_unix_sock_path
) == 0) {
4764 snprintf(apps_unix_sock_path
, PATH_MAX
,
4765 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4768 if (strlen(client_unix_sock_path
) == 0) {
4769 snprintf(client_unix_sock_path
, PATH_MAX
,
4770 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4773 /* Set global SHM for ust */
4774 if (strlen(wait_shm_path
) == 0) {
4775 snprintf(wait_shm_path
, PATH_MAX
,
4776 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4779 if (strlen(health_unix_sock_path
) == 0) {
4780 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4781 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4784 /* Setup kernel consumerd path */
4785 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4786 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4787 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4788 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4790 DBG2("Kernel consumer err path: %s",
4791 kconsumer_data
.err_unix_sock_path
);
4792 DBG2("Kernel consumer cmd path: %s",
4793 kconsumer_data
.cmd_unix_sock_path
);
4795 home_path
= utils_get_home_dir();
4796 if (home_path
== NULL
) {
4797 /* TODO: Add --socket PATH option */
4798 ERR("Can't get HOME directory for sockets creation.");
4804 * Create rundir from home path. This will create something like
4807 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4813 ret
= create_lttng_rundir(rundir
);
4818 if (strlen(apps_unix_sock_path
) == 0) {
4819 snprintf(apps_unix_sock_path
, PATH_MAX
,
4820 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4823 /* Set the cli tool unix socket path */
4824 if (strlen(client_unix_sock_path
) == 0) {
4825 snprintf(client_unix_sock_path
, PATH_MAX
,
4826 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4829 /* Set global SHM for ust */
4830 if (strlen(wait_shm_path
) == 0) {
4831 snprintf(wait_shm_path
, PATH_MAX
,
4832 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
4835 /* Set health check Unix path */
4836 if (strlen(health_unix_sock_path
) == 0) {
4837 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4838 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
4842 /* Set consumer initial state */
4843 kernel_consumerd_state
= CONSUMER_STOPPED
;
4844 ust_consumerd_state
= CONSUMER_STOPPED
;
4846 DBG("Client socket path %s", client_unix_sock_path
);
4847 DBG("Application socket path %s", apps_unix_sock_path
);
4848 DBG("Application wait path %s", wait_shm_path
);
4849 DBG("LTTng run directory path: %s", rundir
);
4851 /* 32 bits consumerd path setup */
4852 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
4853 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
4854 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
4855 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
4857 DBG2("UST consumer 32 bits err path: %s",
4858 ustconsumer32_data
.err_unix_sock_path
);
4859 DBG2("UST consumer 32 bits cmd path: %s",
4860 ustconsumer32_data
.cmd_unix_sock_path
);
4862 /* 64 bits consumerd path setup */
4863 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
4864 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
4865 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
4866 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
4868 DBG2("UST consumer 64 bits err path: %s",
4869 ustconsumer64_data
.err_unix_sock_path
);
4870 DBG2("UST consumer 64 bits cmd path: %s",
4871 ustconsumer64_data
.cmd_unix_sock_path
);
4874 * See if daemon already exist.
4876 if ((ret
= check_existing_daemon()) < 0) {
4877 ERR("Already running daemon.\n");
4879 * We do not goto exit because we must not cleanup()
4880 * because a daemon is already running.
4886 * Init UST app hash table. Alloc hash table before this point since
4887 * cleanup() can get called after that point.
4891 /* Initialize JUL domain subsystem. */
4892 if ((ret
= jul_init()) < 0) {
4893 /* ENOMEM at this point. */
4897 /* After this point, we can safely call cleanup() with "goto exit" */
4900 * These actions must be executed as root. We do that *after* setting up
4901 * the sockets path because we MUST make the check for another daemon using
4902 * those paths *before* trying to set the kernel consumer sockets and init
4906 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
4911 /* Setup kernel tracer */
4912 if (!opt_no_kernel
) {
4913 init_kernel_tracer();
4916 /* Set ulimit for open files */
4919 /* init lttng_fd tracking must be done after set_ulimit. */
4922 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
4927 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
4932 /* Setup the needed unix socket */
4933 if ((ret
= init_daemon_socket()) < 0) {
4937 /* Set credentials to socket */
4938 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
4942 /* Get parent pid if -S, --sig-parent is specified. */
4943 if (opt_sig_parent
) {
4947 /* Setup the kernel pipe for waking up the kernel thread */
4948 if (is_root
&& !opt_no_kernel
) {
4949 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
4954 /* Setup the thread ht_cleanup communication pipe. */
4955 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
4959 /* Setup the thread apps communication pipe. */
4960 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
4964 /* Setup the thread apps notify communication pipe. */
4965 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
4969 /* Initialize global buffer per UID and PID registry. */
4970 buffer_reg_init_uid_registry();
4971 buffer_reg_init_pid_registry();
4973 /* Init UST command queue. */
4974 cds_wfq_init(&ust_cmd_queue
.queue
);
4977 * Get session list pointer. This pointer MUST NOT be free(). This list is
4978 * statically declared in session.c
4980 session_list_ptr
= session_get_list();
4982 /* Set up max poll set size */
4983 lttng_poll_set_max_size();
4987 /* Check for the application socket timeout env variable. */
4988 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
4989 if (env_app_timeout
) {
4990 app_socket_timeout
= atoi(env_app_timeout
);
4992 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
4998 /* Initialize communication library */
5000 /* This is to get the TCP timeout value. */
5001 lttcomm_inet_init();
5004 * Initialize the health check subsystem. This call should set the
5005 * appropriate time values.
5007 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5008 if (!health_sessiond
) {
5009 PERROR("health_app_create error");
5010 goto exit_health_sessiond_cleanup
;
5013 /* Create thread to clean up RCU hash tables */
5014 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5015 thread_ht_cleanup
, (void *) NULL
);
5017 PERROR("pthread_create ht_cleanup");
5018 goto exit_ht_cleanup
;
5021 /* Create health-check thread */
5022 ret
= pthread_create(&health_thread
, NULL
,
5023 thread_manage_health
, (void *) NULL
);
5025 PERROR("pthread_create health");
5029 /* Create thread to manage the client socket */
5030 ret
= pthread_create(&client_thread
, NULL
,
5031 thread_manage_clients
, (void *) NULL
);
5033 PERROR("pthread_create clients");
5037 /* Create thread to dispatch registration */
5038 ret
= pthread_create(&dispatch_thread
, NULL
,
5039 thread_dispatch_ust_registration
, (void *) NULL
);
5041 PERROR("pthread_create dispatch");
5045 /* Create thread to manage application registration. */
5046 ret
= pthread_create(®_apps_thread
, NULL
,
5047 thread_registration_apps
, (void *) NULL
);
5049 PERROR("pthread_create registration");
5053 /* Create thread to manage application socket */
5054 ret
= pthread_create(&apps_thread
, NULL
,
5055 thread_manage_apps
, (void *) NULL
);
5057 PERROR("pthread_create apps");
5061 /* Create thread to manage application notify socket */
5062 ret
= pthread_create(&apps_notify_thread
, NULL
,
5063 ust_thread_manage_notify
, (void *) NULL
);
5065 PERROR("pthread_create apps");
5066 goto exit_apps_notify
;
5069 /* Create JUL registration thread. */
5070 ret
= pthread_create(&jul_reg_thread
, NULL
,
5071 jul_thread_manage_registration
, (void *) NULL
);
5073 PERROR("pthread_create apps");
5077 /* Don't start this thread if kernel tracing is not requested nor root */
5078 if (is_root
&& !opt_no_kernel
) {
5079 /* Create kernel thread to manage kernel event */
5080 ret
= pthread_create(&kernel_thread
, NULL
,
5081 thread_manage_kernel
, (void *) NULL
);
5083 PERROR("pthread_create kernel");
5087 ret
= pthread_join(kernel_thread
, &status
);
5089 PERROR("pthread_join");
5090 goto error
; /* join error, exit without cleanup */
5095 ret
= pthread_join(jul_reg_thread
, &status
);
5097 PERROR("pthread_join JUL");
5098 goto error
; /* join error, exit without cleanup */
5102 ret
= pthread_join(apps_notify_thread
, &status
);
5104 PERROR("pthread_join apps notify");
5105 goto error
; /* join error, exit without cleanup */
5109 ret
= pthread_join(apps_thread
, &status
);
5111 PERROR("pthread_join apps");
5112 goto error
; /* join error, exit without cleanup */
5117 ret
= pthread_join(reg_apps_thread
, &status
);
5119 PERROR("pthread_join");
5120 goto error
; /* join error, exit without cleanup */
5124 ret
= pthread_join(dispatch_thread
, &status
);
5126 PERROR("pthread_join");
5127 goto error
; /* join error, exit without cleanup */
5131 ret
= pthread_join(client_thread
, &status
);
5133 PERROR("pthread_join");
5134 goto error
; /* join error, exit without cleanup */
5137 ret
= join_consumer_thread(&kconsumer_data
);
5139 PERROR("join_consumer");
5140 goto error
; /* join error, exit without cleanup */
5143 ret
= join_consumer_thread(&ustconsumer32_data
);
5145 PERROR("join_consumer ust32");
5146 goto error
; /* join error, exit without cleanup */
5149 ret
= join_consumer_thread(&ustconsumer64_data
);
5151 PERROR("join_consumer ust64");
5152 goto error
; /* join error, exit without cleanup */
5156 ret
= pthread_join(health_thread
, &status
);
5158 PERROR("pthread_join health thread");
5159 goto error
; /* join error, exit without cleanup */
5163 ret
= pthread_join(ht_cleanup_thread
, &status
);
5165 PERROR("pthread_join ht cleanup thread");
5166 goto error
; /* join error, exit without cleanup */
5169 health_app_destroy(health_sessiond
);
5170 exit_health_sessiond_cleanup
:
5173 * cleanup() is called when no other thread is running.
5175 rcu_thread_online();
5177 rcu_thread_offline();
5178 rcu_unregister_thread();