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 #include "load-session-thread.h"
74 #define CONSUMERD_FILE "lttng-consumerd"
77 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
78 static int tracing_group_name_override
;
79 static char *opt_pidfile
;
80 static int opt_sig_parent
;
81 static int opt_verbose_consumer
;
82 static int opt_daemon
, opt_background
;
83 static int opt_no_kernel
;
84 static char *opt_load_session_path
;
85 static pid_t ppid
; /* Parent PID for --sig-parent option */
86 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
89 /* Set to 1 when a SIGUSR1 signal is received. */
90 static int recv_child_signal
;
93 * Consumer daemon specific control data. Every value not initialized here is
94 * set to 0 by the static definition.
96 static struct consumer_data kconsumer_data
= {
97 .type
= LTTNG_CONSUMER_KERNEL
,
98 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
99 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
102 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
103 .lock
= PTHREAD_MUTEX_INITIALIZER
,
104 .cond
= PTHREAD_COND_INITIALIZER
,
105 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 static struct consumer_data ustconsumer64_data
= {
108 .type
= LTTNG_CONSUMER64_UST
,
109 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
110 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
113 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
114 .lock
= PTHREAD_MUTEX_INITIALIZER
,
115 .cond
= PTHREAD_COND_INITIALIZER
,
116 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
118 static struct consumer_data ustconsumer32_data
= {
119 .type
= LTTNG_CONSUMER32_UST
,
120 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
121 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
124 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
125 .lock
= PTHREAD_MUTEX_INITIALIZER
,
126 .cond
= PTHREAD_COND_INITIALIZER
,
127 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
130 /* Command line options */
131 static const struct option long_options
[] = {
132 { "client-sock", 1, 0, 'c' },
133 { "apps-sock", 1, 0, 'a' },
134 { "kconsumerd-cmd-sock", 1, 0, 'C' },
135 { "kconsumerd-err-sock", 1, 0, 'E' },
136 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
137 { "ustconsumerd32-err-sock", 1, 0, 'H' },
138 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
139 { "ustconsumerd64-err-sock", 1, 0, 'F' },
140 { "consumerd32-path", 1, 0, 'u' },
141 { "consumerd32-libdir", 1, 0, 'U' },
142 { "consumerd64-path", 1, 0, 't' },
143 { "consumerd64-libdir", 1, 0, 'T' },
144 { "daemonize", 0, 0, 'd' },
145 { "background", 0, 0, 'b' },
146 { "sig-parent", 0, 0, 'S' },
147 { "help", 0, 0, 'h' },
148 { "group", 1, 0, 'g' },
149 { "version", 0, 0, 'V' },
150 { "quiet", 0, 0, 'q' },
151 { "verbose", 0, 0, 'v' },
152 { "verbose-consumer", 0, 0, 'Z' },
153 { "no-kernel", 0, 0, 'N' },
154 { "pidfile", 1, 0, 'p' },
155 { "jul-tcp-port", 1, 0, 'J' },
156 { "config", 1, 0, 'f' },
157 { "load", 1, 0, 'l' },
158 { "kmod-probes", 1, 0, 'P' },
162 /* Command line options to ignore from configuration file */
163 static const char *config_ignore_options
[] = { "help", "version", "config" };
165 /* Shared between threads */
166 static int dispatch_thread_exit
;
168 /* Global application Unix socket path */
169 static char apps_unix_sock_path
[PATH_MAX
];
170 /* Global client Unix socket path */
171 static char client_unix_sock_path
[PATH_MAX
];
172 /* global wait shm path for UST */
173 static char wait_shm_path
[PATH_MAX
];
174 /* Global health check unix path */
175 static char health_unix_sock_path
[PATH_MAX
];
177 /* Sockets and FDs */
178 static int client_sock
= -1;
179 static int apps_sock
= -1;
180 int kernel_tracer_fd
= -1;
181 static int kernel_poll_pipe
[2] = { -1, -1 };
184 * Quit pipe for all threads. This permits a single cancellation point
185 * for all threads when receiving an event on the pipe.
187 static int thread_quit_pipe
[2] = { -1, -1 };
190 * This pipe is used to inform the thread managing application communication
191 * that a command is queued and ready to be processed.
193 static int apps_cmd_pipe
[2] = { -1, -1 };
195 int apps_cmd_notify_pipe
[2] = { -1, -1 };
197 /* Pthread, Mutexes and Semaphores */
198 static pthread_t apps_thread
;
199 static pthread_t apps_notify_thread
;
200 static pthread_t reg_apps_thread
;
201 static pthread_t client_thread
;
202 static pthread_t kernel_thread
;
203 static pthread_t dispatch_thread
;
204 static pthread_t health_thread
;
205 static pthread_t ht_cleanup_thread
;
206 static pthread_t jul_reg_thread
;
207 static pthread_t load_session_thread
;
210 * UST registration command queue. This queue is tied with a futex and uses a N
211 * wakers / 1 waiter implemented and detailed in futex.c/.h
213 * The thread_registration_apps and thread_dispatch_ust_registration uses this
214 * queue along with the wait/wake scheme. The thread_manage_apps receives down
215 * the line new application socket and monitors it for any I/O error or clean
216 * close that triggers an unregistration of the application.
218 static struct ust_cmd_queue ust_cmd_queue
;
221 * Pointer initialized before thread creation.
223 * This points to the tracing session list containing the session count and a
224 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
225 * MUST NOT be taken if you call a public function in session.c.
227 * The lock is nested inside the structure: session_list_ptr->lock. Please use
228 * session_lock_list and session_unlock_list for lock acquisition.
230 static struct ltt_session_list
*session_list_ptr
;
232 int ust_consumerd64_fd
= -1;
233 int ust_consumerd32_fd
= -1;
235 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
236 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
237 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
238 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
239 static int consumerd32_bin_override
;
240 static int consumerd64_bin_override
;
241 static int consumerd32_libdir_override
;
242 static int consumerd64_libdir_override
;
244 static const char *module_proc_lttng
= "/proc/lttng";
247 * Consumer daemon state which is changed when spawning it, killing it or in
248 * case of a fatal error.
250 enum consumerd_state
{
251 CONSUMER_STARTED
= 1,
252 CONSUMER_STOPPED
= 2,
257 * This consumer daemon state is used to validate if a client command will be
258 * able to reach the consumer. If not, the client is informed. For instance,
259 * doing a "lttng start" when the consumer state is set to ERROR will return an
260 * error to the client.
262 * The following example shows a possible race condition of this scheme:
264 * consumer thread error happens
266 * client cmd checks state -> still OK
267 * consumer thread exit, sets error
268 * client cmd try to talk to consumer
271 * However, since the consumer is a different daemon, we have no way of making
272 * sure the command will reach it safely even with this state flag. This is why
273 * we consider that up to the state validation during command processing, the
274 * command is safe. After that, we can not guarantee the correctness of the
275 * client request vis-a-vis the consumer.
277 static enum consumerd_state ust_consumerd_state
;
278 static enum consumerd_state kernel_consumerd_state
;
281 * Socket timeout for receiving and sending in seconds.
283 static int app_socket_timeout
;
285 /* Set in main() with the current page size. */
288 /* Application health monitoring */
289 struct health_app
*health_sessiond
;
291 /* JUL TCP port for registration. Used by the JUL thread. */
292 unsigned int jul_tcp_port
= DEFAULT_JUL_TCP_PORT
;
294 /* Am I root or not. */
295 int is_root
; /* Set to 1 if the daemon is running as root */
297 const char * const config_section_name
= "sessiond";
299 /* Load session thread information to operate. */
300 struct load_session_thread_data
*load_info
;
303 * Whether sessiond is ready for commands/health check requests.
304 * NR_LTTNG_SESSIOND_READY must match the number of calls to
305 * sessiond_notify_ready().
307 #define NR_LTTNG_SESSIOND_READY 3
308 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
310 /* Notify parents that we are ready for cmd and health check */
312 void sessiond_notify_ready(void)
314 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
316 * Notify parent pid that we are ready to accept command
317 * for client side. This ppid is the one from the
318 * external process that spawned us.
320 if (opt_sig_parent
) {
325 * Notify the parent of the fork() process that we are
328 if (opt_daemon
|| opt_background
) {
329 kill(child_ppid
, SIGUSR1
);
335 void setup_consumerd_path(void)
337 const char *bin
, *libdir
;
340 * Allow INSTALL_BIN_PATH to be used as a target path for the
341 * native architecture size consumer if CONFIG_CONSUMER*_PATH
342 * has not been defined.
344 #if (CAA_BITS_PER_LONG == 32)
345 if (!consumerd32_bin
[0]) {
346 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
348 if (!consumerd32_libdir
[0]) {
349 consumerd32_libdir
= INSTALL_LIB_PATH
;
351 #elif (CAA_BITS_PER_LONG == 64)
352 if (!consumerd64_bin
[0]) {
353 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
355 if (!consumerd64_libdir
[0]) {
356 consumerd64_libdir
= INSTALL_LIB_PATH
;
359 #error "Unknown bitness"
363 * runtime env. var. overrides the build default.
365 bin
= getenv("LTTNG_CONSUMERD32_BIN");
367 consumerd32_bin
= bin
;
369 bin
= getenv("LTTNG_CONSUMERD64_BIN");
371 consumerd64_bin
= bin
;
373 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
375 consumerd32_libdir
= libdir
;
377 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
379 consumerd64_libdir
= libdir
;
384 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
386 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
392 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
398 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
410 * Check if the thread quit pipe was triggered.
412 * Return 1 if it was triggered else 0;
414 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
416 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
424 * Init thread quit pipe.
426 * Return -1 on error or 0 if all pipes are created.
428 static int init_thread_quit_pipe(void)
432 ret
= pipe(thread_quit_pipe
);
434 PERROR("thread quit pipe");
438 for (i
= 0; i
< 2; i
++) {
439 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
451 * Stop all threads by closing the thread quit pipe.
453 static void stop_threads(void)
457 /* Stopping all threads */
458 DBG("Terminating all threads");
459 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
461 ERR("write error on thread quit pipe");
464 /* Dispatch thread */
465 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
466 futex_nto1_wake(&ust_cmd_queue
.futex
);
470 * Close every consumer sockets.
472 static void close_consumer_sockets(void)
476 if (kconsumer_data
.err_sock
>= 0) {
477 ret
= close(kconsumer_data
.err_sock
);
479 PERROR("kernel consumer err_sock close");
482 if (ustconsumer32_data
.err_sock
>= 0) {
483 ret
= close(ustconsumer32_data
.err_sock
);
485 PERROR("UST consumerd32 err_sock close");
488 if (ustconsumer64_data
.err_sock
>= 0) {
489 ret
= close(ustconsumer64_data
.err_sock
);
491 PERROR("UST consumerd64 err_sock close");
494 if (kconsumer_data
.cmd_sock
>= 0) {
495 ret
= close(kconsumer_data
.cmd_sock
);
497 PERROR("kernel consumer cmd_sock close");
500 if (ustconsumer32_data
.cmd_sock
>= 0) {
501 ret
= close(ustconsumer32_data
.cmd_sock
);
503 PERROR("UST consumerd32 cmd_sock close");
506 if (ustconsumer64_data
.cmd_sock
>= 0) {
507 ret
= close(ustconsumer64_data
.cmd_sock
);
509 PERROR("UST consumerd64 cmd_sock close");
517 static void cleanup(void)
520 struct ltt_session
*sess
, *stmp
;
526 * Close the thread quit pipe. It has already done its job,
527 * since we are now called.
529 utils_close_pipe(thread_quit_pipe
);
532 * If opt_pidfile is undefined, the default file will be wiped when
533 * removing the rundir.
536 ret
= remove(opt_pidfile
);
538 PERROR("remove pidfile %s", opt_pidfile
);
542 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
545 snprintf(path
, PATH_MAX
,
547 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
548 DBG("Removing %s", path
);
551 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
552 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
);
553 DBG("Removing %s", path
);
557 snprintf(path
, PATH_MAX
,
558 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
560 DBG("Removing %s", path
);
563 snprintf(path
, PATH_MAX
,
564 DEFAULT_KCONSUMERD_PATH
,
566 DBG("Removing directory %s", path
);
569 /* ust consumerd 32 */
570 snprintf(path
, PATH_MAX
,
571 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
573 DBG("Removing %s", path
);
576 snprintf(path
, PATH_MAX
,
577 DEFAULT_USTCONSUMERD32_PATH
,
579 DBG("Removing directory %s", path
);
582 /* ust consumerd 64 */
583 snprintf(path
, PATH_MAX
,
584 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
586 DBG("Removing %s", path
);
589 snprintf(path
, PATH_MAX
,
590 DEFAULT_USTCONSUMERD64_PATH
,
592 DBG("Removing directory %s", path
);
596 * We do NOT rmdir rundir because there are other processes
597 * using it, for instance lttng-relayd, which can start in
598 * parallel with this teardown.
603 DBG("Cleaning up all sessions");
605 /* Destroy session list mutex */
606 if (session_list_ptr
!= NULL
) {
607 pthread_mutex_destroy(&session_list_ptr
->lock
);
609 /* Cleanup ALL session */
610 cds_list_for_each_entry_safe(sess
, stmp
,
611 &session_list_ptr
->head
, list
) {
612 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
616 DBG("Closing all UST sockets");
617 ust_app_clean_list();
618 buffer_reg_destroy_registries();
620 if (is_root
&& !opt_no_kernel
) {
621 DBG2("Closing kernel fd");
622 if (kernel_tracer_fd
>= 0) {
623 ret
= close(kernel_tracer_fd
);
628 DBG("Unloading kernel modules");
629 modprobe_remove_lttng_all();
632 close_consumer_sockets();
635 * If the override option is set, the pointer points to a *non* const thus
636 * freeing it even though the variable type is set to const.
638 if (tracing_group_name_override
) {
639 free((void *) tracing_group_name
);
641 if (consumerd32_bin_override
) {
642 free((void *) consumerd32_bin
);
644 if (consumerd64_bin_override
) {
645 free((void *) consumerd64_bin
);
647 if (consumerd32_libdir_override
) {
648 free((void *) consumerd32_libdir
);
650 if (consumerd64_libdir_override
) {
651 free((void *) consumerd64_libdir
);
658 if (opt_load_session_path
) {
659 free(opt_load_session_path
);
663 load_session_destroy_data(load_info
);
668 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
669 "Matthew, BEET driven development works!%c[%dm",
670 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
675 * Send data on a unix socket using the liblttsessiondcomm API.
677 * Return lttcomm error code.
679 static int send_unix_sock(int sock
, void *buf
, size_t len
)
681 /* Check valid length */
686 return lttcomm_send_unix_sock(sock
, buf
, len
);
690 * Free memory of a command context structure.
692 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
694 DBG("Clean command context structure");
696 if ((*cmd_ctx
)->llm
) {
697 free((*cmd_ctx
)->llm
);
699 if ((*cmd_ctx
)->lsm
) {
700 free((*cmd_ctx
)->lsm
);
708 * Notify UST applications using the shm mmap futex.
710 static int notify_ust_apps(int active
)
714 DBG("Notifying applications of session daemon state: %d", active
);
716 /* See shm.c for this call implying mmap, shm and futex calls */
717 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
718 if (wait_shm_mmap
== NULL
) {
722 /* Wake waiting process */
723 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
725 /* Apps notified successfully */
733 * Setup the outgoing data buffer for the response (llm) by allocating the
734 * right amount of memory and copying the original information from the lsm
737 * Return total size of the buffer pointed by buf.
739 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
745 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
746 if (cmd_ctx
->llm
== NULL
) {
752 /* Copy common data */
753 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
754 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
756 cmd_ctx
->llm
->data_size
= size
;
757 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
766 * Update the kernel poll set of all channel fd available over all tracing
767 * session. Add the wakeup pipe at the end of the set.
769 static int update_kernel_poll(struct lttng_poll_event
*events
)
772 struct ltt_session
*session
;
773 struct ltt_kernel_channel
*channel
;
775 DBG("Updating kernel poll set");
778 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
779 session_lock(session
);
780 if (session
->kernel_session
== NULL
) {
781 session_unlock(session
);
785 cds_list_for_each_entry(channel
,
786 &session
->kernel_session
->channel_list
.head
, list
) {
787 /* Add channel fd to the kernel poll set */
788 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
790 session_unlock(session
);
793 DBG("Channel fd %d added to kernel set", channel
->fd
);
795 session_unlock(session
);
797 session_unlock_list();
802 session_unlock_list();
807 * Find the channel fd from 'fd' over all tracing session. When found, check
808 * for new channel stream and send those stream fds to the kernel consumer.
810 * Useful for CPU hotplug feature.
812 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
815 struct ltt_session
*session
;
816 struct ltt_kernel_session
*ksess
;
817 struct ltt_kernel_channel
*channel
;
819 DBG("Updating kernel streams for channel fd %d", fd
);
822 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
823 session_lock(session
);
824 if (session
->kernel_session
== NULL
) {
825 session_unlock(session
);
828 ksess
= session
->kernel_session
;
830 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
831 if (channel
->fd
== fd
) {
832 DBG("Channel found, updating kernel streams");
833 ret
= kernel_open_channel_stream(channel
);
837 /* Update the stream global counter */
838 ksess
->stream_count_global
+= ret
;
841 * Have we already sent fds to the consumer? If yes, it means
842 * that tracing is started so it is safe to send our updated
845 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
846 struct lttng_ht_iter iter
;
847 struct consumer_socket
*socket
;
850 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
851 &iter
.iter
, socket
, node
.node
) {
852 pthread_mutex_lock(socket
->lock
);
853 ret
= kernel_consumer_send_channel_stream(socket
,
855 session
->output_traces
? 1 : 0);
856 pthread_mutex_unlock(socket
->lock
);
867 session_unlock(session
);
869 session_unlock_list();
873 session_unlock(session
);
874 session_unlock_list();
879 * For each tracing session, update newly registered apps. The session list
880 * lock MUST be acquired before calling this.
882 static void update_ust_app(int app_sock
)
884 struct ltt_session
*sess
, *stmp
;
886 /* Consumer is in an ERROR state. Stop any application update. */
887 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
888 /* Stop the update process since the consumer is dead. */
892 /* For all tracing session(s) */
893 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
895 if (sess
->ust_session
) {
896 ust_app_global_update(sess
->ust_session
, app_sock
);
898 session_unlock(sess
);
903 * This thread manage event coming from the kernel.
905 * Features supported in this thread:
908 static void *thread_manage_kernel(void *data
)
910 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
911 uint32_t revents
, nb_fd
;
913 struct lttng_poll_event events
;
915 DBG("[thread] Thread manage kernel started");
917 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
920 * This first step of the while is to clean this structure which could free
921 * non NULL pointers so initialize it before the loop.
923 lttng_poll_init(&events
);
925 if (testpoint(sessiond_thread_manage_kernel
)) {
926 goto error_testpoint
;
929 health_code_update();
931 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
932 goto error_testpoint
;
936 health_code_update();
938 if (update_poll_flag
== 1) {
939 /* Clean events object. We are about to populate it again. */
940 lttng_poll_clean(&events
);
942 ret
= sessiond_set_thread_pollset(&events
, 2);
944 goto error_poll_create
;
947 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
952 /* This will add the available kernel channel if any. */
953 ret
= update_kernel_poll(&events
);
957 update_poll_flag
= 0;
960 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
962 /* Poll infinite value of time */
965 ret
= lttng_poll_wait(&events
, -1);
969 * Restart interrupted system call.
971 if (errno
== EINTR
) {
975 } else if (ret
== 0) {
976 /* Should not happen since timeout is infinite */
977 ERR("Return value of poll is 0 with an infinite timeout.\n"
978 "This should not have happened! Continuing...");
984 for (i
= 0; i
< nb_fd
; i
++) {
985 /* Fetch once the poll data */
986 revents
= LTTNG_POLL_GETEV(&events
, i
);
987 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
989 health_code_update();
991 /* Thread quit pipe has been closed. Killing thread. */
992 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
998 /* Check for data on kernel pipe */
999 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1000 (void) lttng_read(kernel_poll_pipe
[0],
1003 * Ret value is useless here, if this pipe gets any actions an
1004 * update is required anyway.
1006 update_poll_flag
= 1;
1010 * New CPU detected by the kernel. Adding kernel stream to
1011 * kernel session and updating the kernel consumer
1013 if (revents
& LPOLLIN
) {
1014 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1020 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1021 * and unregister kernel stream at this point.
1030 lttng_poll_clean(&events
);
1033 utils_close_pipe(kernel_poll_pipe
);
1034 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1037 ERR("Health error occurred in %s", __func__
);
1038 WARN("Kernel thread died unexpectedly. "
1039 "Kernel tracing can continue but CPU hotplug is disabled.");
1041 health_unregister(health_sessiond
);
1042 DBG("Kernel thread dying");
1047 * Signal pthread condition of the consumer data that the thread.
1049 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1051 pthread_mutex_lock(&data
->cond_mutex
);
1054 * The state is set before signaling. It can be any value, it's the waiter
1055 * job to correctly interpret this condition variable associated to the
1056 * consumer pthread_cond.
1058 * A value of 0 means that the corresponding thread of the consumer data
1059 * was not started. 1 indicates that the thread has started and is ready
1060 * for action. A negative value means that there was an error during the
1063 data
->consumer_thread_is_ready
= state
;
1064 (void) pthread_cond_signal(&data
->cond
);
1066 pthread_mutex_unlock(&data
->cond_mutex
);
1070 * This thread manage the consumer error sent back to the session daemon.
1072 static void *thread_manage_consumer(void *data
)
1074 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1075 uint32_t revents
, nb_fd
;
1076 enum lttcomm_return_code code
;
1077 struct lttng_poll_event events
;
1078 struct consumer_data
*consumer_data
= data
;
1080 DBG("[thread] Manage consumer started");
1082 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1084 health_code_update();
1087 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1088 * metadata_sock. Nothing more will be added to this poll set.
1090 ret
= sessiond_set_thread_pollset(&events
, 3);
1096 * The error socket here is already in a listening state which was done
1097 * just before spawning this thread to avoid a race between the consumer
1098 * daemon exec trying to connect and the listen() call.
1100 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1105 health_code_update();
1107 /* Infinite blocking call, waiting for transmission */
1109 health_poll_entry();
1111 if (testpoint(sessiond_thread_manage_consumer
)) {
1115 ret
= lttng_poll_wait(&events
, -1);
1119 * Restart interrupted system call.
1121 if (errno
== EINTR
) {
1129 for (i
= 0; i
< nb_fd
; i
++) {
1130 /* Fetch once the poll data */
1131 revents
= LTTNG_POLL_GETEV(&events
, i
);
1132 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1134 health_code_update();
1136 /* Thread quit pipe has been closed. Killing thread. */
1137 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1143 /* Event on the registration socket */
1144 if (pollfd
== consumer_data
->err_sock
) {
1145 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1146 ERR("consumer err socket poll error");
1152 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1158 * Set the CLOEXEC flag. Return code is useless because either way, the
1161 (void) utils_set_fd_cloexec(sock
);
1163 health_code_update();
1165 DBG2("Receiving code from consumer err_sock");
1167 /* Getting status code from kconsumerd */
1168 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1169 sizeof(enum lttcomm_return_code
));
1174 health_code_update();
1175 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1176 /* Connect both socket, command and metadata. */
1177 consumer_data
->cmd_sock
=
1178 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1179 consumer_data
->metadata_fd
=
1180 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1181 if (consumer_data
->cmd_sock
< 0
1182 || consumer_data
->metadata_fd
< 0) {
1183 PERROR("consumer connect cmd socket");
1184 /* On error, signal condition and quit. */
1185 signal_consumer_condition(consumer_data
, -1);
1188 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1189 /* Create metadata socket lock. */
1190 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1191 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1192 PERROR("zmalloc pthread mutex");
1196 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1198 signal_consumer_condition(consumer_data
, 1);
1199 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1200 DBG("Consumer metadata socket ready (fd: %d)",
1201 consumer_data
->metadata_fd
);
1203 ERR("consumer error when waiting for SOCK_READY : %s",
1204 lttcomm_get_readable_code(-code
));
1208 /* Remove the consumerd error sock since we've established a connexion */
1209 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1214 /* Add new accepted error socket. */
1215 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1220 /* Add metadata socket that is successfully connected. */
1221 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1222 LPOLLIN
| LPOLLRDHUP
);
1227 health_code_update();
1229 /* Infinite blocking call, waiting for transmission */
1232 health_poll_entry();
1233 ret
= lttng_poll_wait(&events
, -1);
1237 * Restart interrupted system call.
1239 if (errno
== EINTR
) {
1247 for (i
= 0; i
< nb_fd
; i
++) {
1248 /* Fetch once the poll data */
1249 revents
= LTTNG_POLL_GETEV(&events
, i
);
1250 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1252 health_code_update();
1254 /* Thread quit pipe has been closed. Killing thread. */
1255 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1261 if (pollfd
== sock
) {
1262 /* Event on the consumerd socket */
1263 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1264 ERR("consumer err socket second poll error");
1267 health_code_update();
1268 /* Wait for any kconsumerd error */
1269 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1270 sizeof(enum lttcomm_return_code
));
1272 ERR("consumer closed the command socket");
1276 ERR("consumer return code : %s",
1277 lttcomm_get_readable_code(-code
));
1280 } else if (pollfd
== consumer_data
->metadata_fd
) {
1281 /* UST metadata requests */
1282 ret
= ust_consumer_metadata_request(
1283 &consumer_data
->metadata_sock
);
1285 ERR("Handling metadata request");
1290 ERR("Unknown pollfd");
1294 health_code_update();
1300 * We lock here because we are about to close the sockets and some other
1301 * thread might be using them so get exclusive access which will abort all
1302 * other consumer command by other threads.
1304 pthread_mutex_lock(&consumer_data
->lock
);
1306 /* Immediately set the consumerd state to stopped */
1307 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1308 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1309 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1310 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1311 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1313 /* Code flow error... */
1317 if (consumer_data
->err_sock
>= 0) {
1318 ret
= close(consumer_data
->err_sock
);
1322 consumer_data
->err_sock
= -1;
1324 if (consumer_data
->cmd_sock
>= 0) {
1325 ret
= close(consumer_data
->cmd_sock
);
1329 consumer_data
->cmd_sock
= -1;
1331 if (consumer_data
->metadata_sock
.fd_ptr
&&
1332 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1333 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1345 unlink(consumer_data
->err_unix_sock_path
);
1346 unlink(consumer_data
->cmd_unix_sock_path
);
1347 consumer_data
->pid
= 0;
1348 pthread_mutex_unlock(&consumer_data
->lock
);
1350 /* Cleanup metadata socket mutex. */
1351 if (consumer_data
->metadata_sock
.lock
) {
1352 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1353 free(consumer_data
->metadata_sock
.lock
);
1355 lttng_poll_clean(&events
);
1359 ERR("Health error occurred in %s", __func__
);
1361 health_unregister(health_sessiond
);
1362 DBG("consumer thread cleanup completed");
1368 * This thread manage application communication.
1370 static void *thread_manage_apps(void *data
)
1372 int i
, ret
, pollfd
, err
= -1;
1374 uint32_t revents
, nb_fd
;
1375 struct lttng_poll_event events
;
1377 DBG("[thread] Manage application started");
1379 rcu_register_thread();
1380 rcu_thread_online();
1382 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1384 if (testpoint(sessiond_thread_manage_apps
)) {
1385 goto error_testpoint
;
1388 health_code_update();
1390 ret
= sessiond_set_thread_pollset(&events
, 2);
1392 goto error_poll_create
;
1395 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1400 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1404 health_code_update();
1407 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1409 /* Inifinite blocking call, waiting for transmission */
1411 health_poll_entry();
1412 ret
= lttng_poll_wait(&events
, -1);
1416 * Restart interrupted system call.
1418 if (errno
== EINTR
) {
1426 for (i
= 0; i
< nb_fd
; i
++) {
1427 /* Fetch once the poll data */
1428 revents
= LTTNG_POLL_GETEV(&events
, i
);
1429 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1431 health_code_update();
1433 /* Thread quit pipe has been closed. Killing thread. */
1434 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1440 /* Inspect the apps cmd pipe */
1441 if (pollfd
== apps_cmd_pipe
[0]) {
1442 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1443 ERR("Apps command pipe error");
1445 } else if (revents
& LPOLLIN
) {
1449 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1450 if (size_ret
< sizeof(sock
)) {
1451 PERROR("read apps cmd pipe");
1455 health_code_update();
1458 * We only monitor the error events of the socket. This
1459 * thread does not handle any incoming data from UST
1462 ret
= lttng_poll_add(&events
, sock
,
1463 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1468 DBG("Apps with sock %d added to poll set", sock
);
1472 * At this point, we know that a registered application made
1473 * the event at poll_wait.
1475 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1476 /* Removing from the poll set */
1477 ret
= lttng_poll_del(&events
, pollfd
);
1482 /* Socket closed on remote end. */
1483 ust_app_unregister(pollfd
);
1487 health_code_update();
1493 lttng_poll_clean(&events
);
1496 utils_close_pipe(apps_cmd_pipe
);
1497 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1500 * We don't clean the UST app hash table here since already registered
1501 * applications can still be controlled so let them be until the session
1502 * daemon dies or the applications stop.
1507 ERR("Health error occurred in %s", __func__
);
1509 health_unregister(health_sessiond
);
1510 DBG("Application communication apps thread cleanup complete");
1511 rcu_thread_offline();
1512 rcu_unregister_thread();
1517 * Send a socket to a thread This is called from the dispatch UST registration
1518 * thread once all sockets are set for the application.
1520 * The sock value can be invalid, we don't really care, the thread will handle
1521 * it and make the necessary cleanup if so.
1523 * On success, return 0 else a negative value being the errno message of the
1526 static int send_socket_to_thread(int fd
, int sock
)
1531 * It's possible that the FD is set as invalid with -1 concurrently just
1532 * before calling this function being a shutdown state of the thread.
1539 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1540 if (ret
< sizeof(sock
)) {
1541 PERROR("write apps pipe %d", fd
);
1548 /* All good. Don't send back the write positive ret value. */
1555 * Sanitize the wait queue of the dispatch registration thread meaning removing
1556 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1557 * notify socket is never received.
1559 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1561 int ret
, nb_fd
= 0, i
;
1562 unsigned int fd_added
= 0;
1563 struct lttng_poll_event events
;
1564 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1568 lttng_poll_init(&events
);
1570 /* Just skip everything for an empty queue. */
1571 if (!wait_queue
->count
) {
1575 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1580 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1581 &wait_queue
->head
, head
) {
1582 assert(wait_node
->app
);
1583 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1584 LPOLLHUP
| LPOLLERR
);
1597 * Poll but don't block so we can quickly identify the faulty events and
1598 * clean them afterwards from the wait queue.
1600 ret
= lttng_poll_wait(&events
, 0);
1606 for (i
= 0; i
< nb_fd
; i
++) {
1607 /* Get faulty FD. */
1608 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1609 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1611 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1612 &wait_queue
->head
, head
) {
1613 if (pollfd
== wait_node
->app
->sock
&&
1614 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1615 cds_list_del(&wait_node
->head
);
1616 wait_queue
->count
--;
1617 ust_app_destroy(wait_node
->app
);
1625 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1629 lttng_poll_clean(&events
);
1633 lttng_poll_clean(&events
);
1635 ERR("Unable to sanitize wait queue");
1640 * Dispatch request from the registration threads to the application
1641 * communication thread.
1643 static void *thread_dispatch_ust_registration(void *data
)
1646 struct cds_wfq_node
*node
;
1647 struct ust_command
*ust_cmd
= NULL
;
1648 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1649 struct ust_reg_wait_queue wait_queue
= {
1653 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1655 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1656 goto error_testpoint
;
1659 health_code_update();
1661 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1663 DBG("[thread] Dispatch UST command started");
1665 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1666 health_code_update();
1668 /* Atomically prepare the queue futex */
1669 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1672 struct ust_app
*app
= NULL
;
1676 * Make sure we don't have node(s) that have hung up before receiving
1677 * the notify socket. This is to clean the list in order to avoid
1678 * memory leaks from notify socket that are never seen.
1680 sanitize_wait_queue(&wait_queue
);
1682 health_code_update();
1683 /* Dequeue command for registration */
1684 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1686 DBG("Woken up but nothing in the UST command queue");
1687 /* Continue thread execution */
1691 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1693 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1694 " gid:%d sock:%d name:%s (version %d.%d)",
1695 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1696 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1697 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1698 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1700 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1701 wait_node
= zmalloc(sizeof(*wait_node
));
1703 PERROR("zmalloc wait_node dispatch");
1704 ret
= close(ust_cmd
->sock
);
1706 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1708 lttng_fd_put(LTTNG_FD_APPS
, 1);
1712 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1714 /* Create application object if socket is CMD. */
1715 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1717 if (!wait_node
->app
) {
1718 ret
= close(ust_cmd
->sock
);
1720 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1722 lttng_fd_put(LTTNG_FD_APPS
, 1);
1728 * Add application to the wait queue so we can set the notify
1729 * socket before putting this object in the global ht.
1731 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1736 * We have to continue here since we don't have the notify
1737 * socket and the application MUST be added to the hash table
1738 * only at that moment.
1743 * Look for the application in the local wait queue and set the
1744 * notify socket if found.
1746 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1747 &wait_queue
.head
, head
) {
1748 health_code_update();
1749 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1750 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1751 cds_list_del(&wait_node
->head
);
1753 app
= wait_node
->app
;
1755 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1761 * With no application at this stage the received socket is
1762 * basically useless so close it before we free the cmd data
1763 * structure for good.
1766 ret
= close(ust_cmd
->sock
);
1768 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1770 lttng_fd_put(LTTNG_FD_APPS
, 1);
1777 * @session_lock_list
1779 * Lock the global session list so from the register up to the
1780 * registration done message, no thread can see the application
1781 * and change its state.
1783 session_lock_list();
1787 * Add application to the global hash table. This needs to be
1788 * done before the update to the UST registry can locate the
1793 /* Set app version. This call will print an error if needed. */
1794 (void) ust_app_version(app
);
1796 /* Send notify socket through the notify pipe. */
1797 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1801 session_unlock_list();
1803 * No notify thread, stop the UST tracing. However, this is
1804 * not an internal error of the this thread thus setting
1805 * the health error code to a normal exit.
1812 * Update newly registered application with the tracing
1813 * registry info already enabled information.
1815 update_ust_app(app
->sock
);
1818 * Don't care about return value. Let the manage apps threads
1819 * handle app unregistration upon socket close.
1821 (void) ust_app_register_done(app
->sock
);
1824 * Even if the application socket has been closed, send the app
1825 * to the thread and unregistration will take place at that
1828 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1831 session_unlock_list();
1833 * No apps. thread, stop the UST tracing. However, this is
1834 * not an internal error of the this thread thus setting
1835 * the health error code to a normal exit.
1842 session_unlock_list();
1844 } while (node
!= NULL
);
1846 health_poll_entry();
1847 /* Futex wait on queue. Blocking call on futex() */
1848 futex_nto1_wait(&ust_cmd_queue
.futex
);
1851 /* Normal exit, no error */
1855 /* Clean up wait queue. */
1856 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1857 &wait_queue
.head
, head
) {
1858 cds_list_del(&wait_node
->head
);
1864 DBG("Dispatch thread dying");
1867 ERR("Health error occurred in %s", __func__
);
1869 health_unregister(health_sessiond
);
1874 * This thread manage application registration.
1876 static void *thread_registration_apps(void *data
)
1878 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1879 uint32_t revents
, nb_fd
;
1880 struct lttng_poll_event events
;
1882 * Get allocated in this thread, enqueued to a global queue, dequeued and
1883 * freed in the manage apps thread.
1885 struct ust_command
*ust_cmd
= NULL
;
1887 DBG("[thread] Manage application registration started");
1889 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1891 if (testpoint(sessiond_thread_registration_apps
)) {
1892 goto error_testpoint
;
1895 ret
= lttcomm_listen_unix_sock(apps_sock
);
1901 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1902 * more will be added to this poll set.
1904 ret
= sessiond_set_thread_pollset(&events
, 2);
1906 goto error_create_poll
;
1909 /* Add the application registration socket */
1910 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1912 goto error_poll_add
;
1915 /* Notify all applications to register */
1916 ret
= notify_ust_apps(1);
1918 ERR("Failed to notify applications or create the wait shared memory.\n"
1919 "Execution continues but there might be problem for already\n"
1920 "running applications that wishes to register.");
1924 DBG("Accepting application registration");
1926 /* Inifinite blocking call, waiting for transmission */
1928 health_poll_entry();
1929 ret
= lttng_poll_wait(&events
, -1);
1933 * Restart interrupted system call.
1935 if (errno
== EINTR
) {
1943 for (i
= 0; i
< nb_fd
; i
++) {
1944 health_code_update();
1946 /* Fetch once the poll data */
1947 revents
= LTTNG_POLL_GETEV(&events
, i
);
1948 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1950 /* Thread quit pipe has been closed. Killing thread. */
1951 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1957 /* Event on the registration socket */
1958 if (pollfd
== apps_sock
) {
1959 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1960 ERR("Register apps socket poll error");
1962 } else if (revents
& LPOLLIN
) {
1963 sock
= lttcomm_accept_unix_sock(apps_sock
);
1969 * Set socket timeout for both receiving and ending.
1970 * app_socket_timeout is in seconds, whereas
1971 * lttcomm_setsockopt_rcv_timeout and
1972 * lttcomm_setsockopt_snd_timeout expect msec as
1975 (void) lttcomm_setsockopt_rcv_timeout(sock
,
1976 app_socket_timeout
* 1000);
1977 (void) lttcomm_setsockopt_snd_timeout(sock
,
1978 app_socket_timeout
* 1000);
1981 * Set the CLOEXEC flag. Return code is useless because
1982 * either way, the show must go on.
1984 (void) utils_set_fd_cloexec(sock
);
1986 /* Create UST registration command for enqueuing */
1987 ust_cmd
= zmalloc(sizeof(struct ust_command
));
1988 if (ust_cmd
== NULL
) {
1989 PERROR("ust command zmalloc");
1994 * Using message-based transmissions to ensure we don't
1995 * have to deal with partially received messages.
1997 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
1999 ERR("Exhausted file descriptors allowed for applications.");
2009 health_code_update();
2010 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2013 /* Close socket of the application. */
2018 lttng_fd_put(LTTNG_FD_APPS
, 1);
2022 health_code_update();
2024 ust_cmd
->sock
= sock
;
2027 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2028 " gid:%d sock:%d name:%s (version %d.%d)",
2029 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2030 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2031 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2032 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2035 * Lock free enqueue the registration request. The red pill
2036 * has been taken! This apps will be part of the *system*.
2038 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
2041 * Wake the registration queue futex. Implicit memory
2042 * barrier with the exchange in cds_wfq_enqueue.
2044 futex_nto1_wake(&ust_cmd_queue
.futex
);
2052 /* Notify that the registration thread is gone */
2055 if (apps_sock
>= 0) {
2056 ret
= close(apps_sock
);
2066 lttng_fd_put(LTTNG_FD_APPS
, 1);
2068 unlink(apps_unix_sock_path
);
2071 lttng_poll_clean(&events
);
2075 DBG("UST Registration thread cleanup complete");
2078 ERR("Health error occurred in %s", __func__
);
2080 health_unregister(health_sessiond
);
2086 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2087 * exec or it will fails.
2089 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2092 struct timespec timeout
;
2094 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2095 consumer_data
->consumer_thread_is_ready
= 0;
2097 /* Setup pthread condition */
2098 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2101 PERROR("pthread_condattr_init consumer data");
2106 * Set the monotonic clock in order to make sure we DO NOT jump in time
2107 * between the clock_gettime() call and the timedwait call. See bug #324
2108 * for a more details and how we noticed it.
2110 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2113 PERROR("pthread_condattr_setclock consumer data");
2117 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2120 PERROR("pthread_cond_init consumer data");
2124 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2127 PERROR("pthread_create consumer");
2132 /* We are about to wait on a pthread condition */
2133 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2135 /* Get time for sem_timedwait absolute timeout */
2136 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2138 * Set the timeout for the condition timed wait even if the clock gettime
2139 * call fails since we might loop on that call and we want to avoid to
2140 * increment the timeout too many times.
2142 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2145 * The following loop COULD be skipped in some conditions so this is why we
2146 * set ret to 0 in order to make sure at least one round of the loop is
2152 * Loop until the condition is reached or when a timeout is reached. Note
2153 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2154 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2155 * possible. This loop does not take any chances and works with both of
2158 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2159 if (clock_ret
< 0) {
2160 PERROR("clock_gettime spawn consumer");
2161 /* Infinite wait for the consumerd thread to be ready */
2162 ret
= pthread_cond_wait(&consumer_data
->cond
,
2163 &consumer_data
->cond_mutex
);
2165 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2166 &consumer_data
->cond_mutex
, &timeout
);
2170 /* Release the pthread condition */
2171 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2175 if (ret
== ETIMEDOUT
) {
2179 * Call has timed out so we kill the kconsumerd_thread and return
2182 ERR("Condition timed out. The consumer thread was never ready."
2184 pth_ret
= pthread_cancel(consumer_data
->thread
);
2186 PERROR("pthread_cancel consumer thread");
2189 PERROR("pthread_cond_wait failed consumer thread");
2191 /* Caller is expecting a negative value on failure. */
2196 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2197 if (consumer_data
->pid
== 0) {
2198 ERR("Consumerd did not start");
2199 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2202 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2211 * Join consumer thread
2213 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2217 /* Consumer pid must be a real one. */
2218 if (consumer_data
->pid
> 0) {
2220 ret
= kill(consumer_data
->pid
, SIGTERM
);
2222 ERR("Error killing consumer daemon");
2225 return pthread_join(consumer_data
->thread
, &status
);
2232 * Fork and exec a consumer daemon (consumerd).
2234 * Return pid if successful else -1.
2236 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2240 const char *consumer_to_use
;
2241 const char *verbosity
;
2244 DBG("Spawning consumerd");
2251 if (opt_verbose_consumer
) {
2252 verbosity
= "--verbose";
2253 } else if (lttng_opt_quiet
) {
2254 verbosity
= "--quiet";
2259 switch (consumer_data
->type
) {
2260 case LTTNG_CONSUMER_KERNEL
:
2262 * Find out which consumerd to execute. We will first try the
2263 * 64-bit path, then the sessiond's installation directory, and
2264 * fallback on the 32-bit one,
2266 DBG3("Looking for a kernel consumer at these locations:");
2267 DBG3(" 1) %s", consumerd64_bin
);
2268 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2269 DBG3(" 3) %s", consumerd32_bin
);
2270 if (stat(consumerd64_bin
, &st
) == 0) {
2271 DBG3("Found location #1");
2272 consumer_to_use
= consumerd64_bin
;
2273 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2274 DBG3("Found location #2");
2275 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2276 } else if (stat(consumerd32_bin
, &st
) == 0) {
2277 DBG3("Found location #3");
2278 consumer_to_use
= consumerd32_bin
;
2280 DBG("Could not find any valid consumerd executable");
2284 DBG("Using kernel consumer at: %s", consumer_to_use
);
2285 ret
= execl(consumer_to_use
,
2286 "lttng-consumerd", verbosity
, "-k",
2287 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2288 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2289 "--group", tracing_group_name
,
2292 case LTTNG_CONSUMER64_UST
:
2294 char *tmpnew
= NULL
;
2296 if (consumerd64_libdir
[0] != '\0') {
2300 tmp
= getenv("LD_LIBRARY_PATH");
2304 tmplen
= strlen("LD_LIBRARY_PATH=")
2305 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2306 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2311 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2312 strcat(tmpnew
, consumerd64_libdir
);
2313 if (tmp
[0] != '\0') {
2314 strcat(tmpnew
, ":");
2315 strcat(tmpnew
, tmp
);
2317 ret
= putenv(tmpnew
);
2324 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2325 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2326 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2327 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2328 "--group", tracing_group_name
,
2330 if (consumerd64_libdir
[0] != '\0') {
2335 case LTTNG_CONSUMER32_UST
:
2337 char *tmpnew
= NULL
;
2339 if (consumerd32_libdir
[0] != '\0') {
2343 tmp
= getenv("LD_LIBRARY_PATH");
2347 tmplen
= strlen("LD_LIBRARY_PATH=")
2348 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2349 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2354 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2355 strcat(tmpnew
, consumerd32_libdir
);
2356 if (tmp
[0] != '\0') {
2357 strcat(tmpnew
, ":");
2358 strcat(tmpnew
, tmp
);
2360 ret
= putenv(tmpnew
);
2367 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2368 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2369 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2370 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2371 "--group", tracing_group_name
,
2373 if (consumerd32_libdir
[0] != '\0') {
2379 PERROR("unknown consumer type");
2383 PERROR("Consumer execl()");
2385 /* Reaching this point, we got a failure on our execl(). */
2387 } else if (pid
> 0) {
2390 PERROR("start consumer fork");
2398 * Spawn the consumerd daemon and session daemon thread.
2400 static int start_consumerd(struct consumer_data
*consumer_data
)
2405 * Set the listen() state on the socket since there is a possible race
2406 * between the exec() of the consumer daemon and this call if place in the
2407 * consumer thread. See bug #366 for more details.
2409 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2414 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2415 if (consumer_data
->pid
!= 0) {
2416 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2420 ret
= spawn_consumerd(consumer_data
);
2422 ERR("Spawning consumerd failed");
2423 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2427 /* Setting up the consumer_data pid */
2428 consumer_data
->pid
= ret
;
2429 DBG2("Consumer pid %d", consumer_data
->pid
);
2430 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2432 DBG2("Spawning consumer control thread");
2433 ret
= spawn_consumer_thread(consumer_data
);
2435 ERR("Fatal error spawning consumer control thread");
2443 /* Cleanup already created sockets on error. */
2444 if (consumer_data
->err_sock
>= 0) {
2447 err
= close(consumer_data
->err_sock
);
2449 PERROR("close consumer data error socket");
2456 * Setup necessary data for kernel tracer action.
2458 static int init_kernel_tracer(void)
2462 /* Modprobe lttng kernel modules */
2463 ret
= modprobe_lttng_control();
2468 /* Open debugfs lttng */
2469 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2470 if (kernel_tracer_fd
< 0) {
2471 DBG("Failed to open %s", module_proc_lttng
);
2476 /* Validate kernel version */
2477 ret
= kernel_validate_version(kernel_tracer_fd
);
2482 ret
= modprobe_lttng_data();
2487 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2491 modprobe_remove_lttng_control();
2492 ret
= close(kernel_tracer_fd
);
2496 kernel_tracer_fd
= -1;
2497 return LTTNG_ERR_KERN_VERSION
;
2500 ret
= close(kernel_tracer_fd
);
2506 modprobe_remove_lttng_control();
2509 WARN("No kernel tracer available");
2510 kernel_tracer_fd
= -1;
2512 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2514 return LTTNG_ERR_KERN_NA
;
2520 * Copy consumer output from the tracing session to the domain session. The
2521 * function also applies the right modification on a per domain basis for the
2522 * trace files destination directory.
2524 * Should *NOT* be called with RCU read-side lock held.
2526 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2529 const char *dir_name
;
2530 struct consumer_output
*consumer
;
2533 assert(session
->consumer
);
2536 case LTTNG_DOMAIN_KERNEL
:
2537 DBG3("Copying tracing session consumer output in kernel session");
2539 * XXX: We should audit the session creation and what this function
2540 * does "extra" in order to avoid a destroy since this function is used
2541 * in the domain session creation (kernel and ust) only. Same for UST
2544 if (session
->kernel_session
->consumer
) {
2545 consumer_destroy_output(session
->kernel_session
->consumer
);
2547 session
->kernel_session
->consumer
=
2548 consumer_copy_output(session
->consumer
);
2549 /* Ease our life a bit for the next part */
2550 consumer
= session
->kernel_session
->consumer
;
2551 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2553 case LTTNG_DOMAIN_JUL
:
2554 case LTTNG_DOMAIN_UST
:
2555 DBG3("Copying tracing session consumer output in UST session");
2556 if (session
->ust_session
->consumer
) {
2557 consumer_destroy_output(session
->ust_session
->consumer
);
2559 session
->ust_session
->consumer
=
2560 consumer_copy_output(session
->consumer
);
2561 /* Ease our life a bit for the next part */
2562 consumer
= session
->ust_session
->consumer
;
2563 dir_name
= DEFAULT_UST_TRACE_DIR
;
2566 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2570 /* Append correct directory to subdir */
2571 strncat(consumer
->subdir
, dir_name
,
2572 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2573 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2582 * Create an UST session and add it to the session ust list.
2584 * Should *NOT* be called with RCU read-side lock held.
2586 static int create_ust_session(struct ltt_session
*session
,
2587 struct lttng_domain
*domain
)
2590 struct ltt_ust_session
*lus
= NULL
;
2594 assert(session
->consumer
);
2596 switch (domain
->type
) {
2597 case LTTNG_DOMAIN_JUL
:
2598 case LTTNG_DOMAIN_UST
:
2601 ERR("Unknown UST domain on create session %d", domain
->type
);
2602 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2606 DBG("Creating UST session");
2608 lus
= trace_ust_create_session(session
->id
);
2610 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2614 lus
->uid
= session
->uid
;
2615 lus
->gid
= session
->gid
;
2616 lus
->output_traces
= session
->output_traces
;
2617 lus
->snapshot_mode
= session
->snapshot_mode
;
2618 lus
->live_timer_interval
= session
->live_timer
;
2619 session
->ust_session
= lus
;
2621 /* Copy session output to the newly created UST session */
2622 ret
= copy_session_consumer(domain
->type
, session
);
2623 if (ret
!= LTTNG_OK
) {
2631 session
->ust_session
= NULL
;
2636 * Create a kernel tracer session then create the default channel.
2638 static int create_kernel_session(struct ltt_session
*session
)
2642 DBG("Creating kernel session");
2644 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2646 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2650 /* Code flow safety */
2651 assert(session
->kernel_session
);
2653 /* Copy session output to the newly created Kernel session */
2654 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2655 if (ret
!= LTTNG_OK
) {
2659 /* Create directory(ies) on local filesystem. */
2660 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2661 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2662 ret
= run_as_mkdir_recursive(
2663 session
->kernel_session
->consumer
->dst
.trace_path
,
2664 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2666 if (ret
!= -EEXIST
) {
2667 ERR("Trace directory creation error");
2673 session
->kernel_session
->uid
= session
->uid
;
2674 session
->kernel_session
->gid
= session
->gid
;
2675 session
->kernel_session
->output_traces
= session
->output_traces
;
2676 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2681 trace_kernel_destroy_session(session
->kernel_session
);
2682 session
->kernel_session
= NULL
;
2687 * Count number of session permitted by uid/gid.
2689 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2692 struct ltt_session
*session
;
2694 DBG("Counting number of available session for UID %d GID %d",
2696 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2698 * Only list the sessions the user can control.
2700 if (!session_access_ok(session
, uid
, gid
)) {
2709 * Process the command requested by the lttng client within the command
2710 * context structure. This function make sure that the return structure (llm)
2711 * is set and ready for transmission before returning.
2713 * Return any error encountered or 0 for success.
2715 * "sock" is only used for special-case var. len data.
2717 * Should *NOT* be called with RCU read-side lock held.
2719 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2723 int need_tracing_session
= 1;
2726 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2730 switch (cmd_ctx
->lsm
->cmd_type
) {
2731 case LTTNG_CREATE_SESSION
:
2732 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2733 case LTTNG_CREATE_SESSION_LIVE
:
2734 case LTTNG_DESTROY_SESSION
:
2735 case LTTNG_LIST_SESSIONS
:
2736 case LTTNG_LIST_DOMAINS
:
2737 case LTTNG_START_TRACE
:
2738 case LTTNG_STOP_TRACE
:
2739 case LTTNG_DATA_PENDING
:
2740 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2741 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2742 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2743 case LTTNG_SNAPSHOT_RECORD
:
2744 case LTTNG_SAVE_SESSION
:
2751 if (opt_no_kernel
&& need_domain
2752 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2754 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2756 ret
= LTTNG_ERR_KERN_NA
;
2761 /* Deny register consumer if we already have a spawned consumer. */
2762 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2763 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2764 if (kconsumer_data
.pid
> 0) {
2765 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2766 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2769 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2773 * Check for command that don't needs to allocate a returned payload. We do
2774 * this here so we don't have to make the call for no payload at each
2777 switch(cmd_ctx
->lsm
->cmd_type
) {
2778 case LTTNG_LIST_SESSIONS
:
2779 case LTTNG_LIST_TRACEPOINTS
:
2780 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2781 case LTTNG_LIST_DOMAINS
:
2782 case LTTNG_LIST_CHANNELS
:
2783 case LTTNG_LIST_EVENTS
:
2786 /* Setup lttng message with no payload */
2787 ret
= setup_lttng_msg(cmd_ctx
, 0);
2789 /* This label does not try to unlock the session */
2790 goto init_setup_error
;
2794 /* Commands that DO NOT need a session. */
2795 switch (cmd_ctx
->lsm
->cmd_type
) {
2796 case LTTNG_CREATE_SESSION
:
2797 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2798 case LTTNG_CREATE_SESSION_LIVE
:
2799 case LTTNG_CALIBRATE
:
2800 case LTTNG_LIST_SESSIONS
:
2801 case LTTNG_LIST_TRACEPOINTS
:
2802 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2803 case LTTNG_SAVE_SESSION
:
2804 need_tracing_session
= 0;
2807 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2809 * We keep the session list lock across _all_ commands
2810 * for now, because the per-session lock does not
2811 * handle teardown properly.
2813 session_lock_list();
2814 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2815 if (cmd_ctx
->session
== NULL
) {
2816 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2819 /* Acquire lock for the session */
2820 session_lock(cmd_ctx
->session
);
2826 * Commands that need a valid session but should NOT create one if none
2827 * exists. Instead of creating one and destroying it when the command is
2828 * handled, process that right before so we save some round trip in useless
2831 switch (cmd_ctx
->lsm
->cmd_type
) {
2832 case LTTNG_DISABLE_CHANNEL
:
2833 case LTTNG_DISABLE_EVENT
:
2834 case LTTNG_DISABLE_ALL_EVENT
:
2835 switch (cmd_ctx
->lsm
->domain
.type
) {
2836 case LTTNG_DOMAIN_KERNEL
:
2837 if (!cmd_ctx
->session
->kernel_session
) {
2838 ret
= LTTNG_ERR_NO_CHANNEL
;
2842 case LTTNG_DOMAIN_JUL
:
2843 case LTTNG_DOMAIN_UST
:
2844 if (!cmd_ctx
->session
->ust_session
) {
2845 ret
= LTTNG_ERR_NO_CHANNEL
;
2850 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2862 * Check domain type for specific "pre-action".
2864 switch (cmd_ctx
->lsm
->domain
.type
) {
2865 case LTTNG_DOMAIN_KERNEL
:
2867 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2871 /* Kernel tracer check */
2872 if (kernel_tracer_fd
== -1) {
2873 /* Basically, load kernel tracer modules */
2874 ret
= init_kernel_tracer();
2880 /* Consumer is in an ERROR state. Report back to client */
2881 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2882 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2886 /* Need a session for kernel command */
2887 if (need_tracing_session
) {
2888 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2889 ret
= create_kernel_session(cmd_ctx
->session
);
2891 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2896 /* Start the kernel consumer daemon */
2897 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2898 if (kconsumer_data
.pid
== 0 &&
2899 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2900 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2901 ret
= start_consumerd(&kconsumer_data
);
2903 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2906 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2908 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2912 * The consumer was just spawned so we need to add the socket to
2913 * the consumer output of the session if exist.
2915 ret
= consumer_create_socket(&kconsumer_data
,
2916 cmd_ctx
->session
->kernel_session
->consumer
);
2923 case LTTNG_DOMAIN_JUL
:
2924 case LTTNG_DOMAIN_UST
:
2926 if (!ust_app_supported()) {
2927 ret
= LTTNG_ERR_NO_UST
;
2930 /* Consumer is in an ERROR state. Report back to client */
2931 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2932 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2936 if (need_tracing_session
) {
2937 /* Create UST session if none exist. */
2938 if (cmd_ctx
->session
->ust_session
== NULL
) {
2939 ret
= create_ust_session(cmd_ctx
->session
,
2940 &cmd_ctx
->lsm
->domain
);
2941 if (ret
!= LTTNG_OK
) {
2946 /* Start the UST consumer daemons */
2948 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2949 if (consumerd64_bin
[0] != '\0' &&
2950 ustconsumer64_data
.pid
== 0 &&
2951 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2952 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2953 ret
= start_consumerd(&ustconsumer64_data
);
2955 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2956 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
2960 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
2961 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2963 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2967 * Setup socket for consumer 64 bit. No need for atomic access
2968 * since it was set above and can ONLY be set in this thread.
2970 ret
= consumer_create_socket(&ustconsumer64_data
,
2971 cmd_ctx
->session
->ust_session
->consumer
);
2977 if (consumerd32_bin
[0] != '\0' &&
2978 ustconsumer32_data
.pid
== 0 &&
2979 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2980 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2981 ret
= start_consumerd(&ustconsumer32_data
);
2983 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
2984 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
2988 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
2989 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
2991 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
2995 * Setup socket for consumer 64 bit. No need for atomic access
2996 * since it was set above and can ONLY be set in this thread.
2998 ret
= consumer_create_socket(&ustconsumer32_data
,
2999 cmd_ctx
->session
->ust_session
->consumer
);
3011 /* Validate consumer daemon state when start/stop trace command */
3012 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3013 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3014 switch (cmd_ctx
->lsm
->domain
.type
) {
3015 case LTTNG_DOMAIN_JUL
:
3016 case LTTNG_DOMAIN_UST
:
3017 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3018 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3022 case LTTNG_DOMAIN_KERNEL
:
3023 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3024 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3032 * Check that the UID or GID match that of the tracing session.
3033 * The root user can interact with all sessions.
3035 if (need_tracing_session
) {
3036 if (!session_access_ok(cmd_ctx
->session
,
3037 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3038 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3039 ret
= LTTNG_ERR_EPERM
;
3045 * Send relayd information to consumer as soon as we have a domain and a
3048 if (cmd_ctx
->session
&& need_domain
) {
3050 * Setup relayd if not done yet. If the relayd information was already
3051 * sent to the consumer, this call will gracefully return.
3053 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3054 if (ret
!= LTTNG_OK
) {
3059 /* Process by command type */
3060 switch (cmd_ctx
->lsm
->cmd_type
) {
3061 case LTTNG_ADD_CONTEXT
:
3063 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3064 cmd_ctx
->lsm
->u
.context
.channel_name
,
3065 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3068 case LTTNG_DISABLE_CHANNEL
:
3070 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3071 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3074 case LTTNG_DISABLE_EVENT
:
3076 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3077 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3078 cmd_ctx
->lsm
->u
.disable
.name
);
3081 case LTTNG_DISABLE_ALL_EVENT
:
3083 DBG("Disabling all events");
3085 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3086 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3089 case LTTNG_ENABLE_CHANNEL
:
3091 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3092 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3095 case LTTNG_ENABLE_EVENT
:
3097 struct lttng_event_exclusion
*exclusion
= NULL
;
3098 struct lttng_filter_bytecode
*bytecode
= NULL
;
3099 char *filter_expression
= NULL
;
3101 /* Handle exclusion events and receive it from the client. */
3102 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3103 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3105 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3106 (count
* LTTNG_SYMBOL_NAME_LEN
));
3108 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3112 DBG("Receiving var len exclusion event list from client ...");
3113 exclusion
->count
= count
;
3114 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3115 count
* LTTNG_SYMBOL_NAME_LEN
);
3117 DBG("Nothing recv() from client var len data... continuing");
3120 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3125 /* Get filter expression from client. */
3126 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3127 size_t expression_len
=
3128 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3130 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3131 ret
= LTTNG_ERR_FILTER_INVAL
;
3136 filter_expression
= zmalloc(expression_len
);
3137 if (!filter_expression
) {
3139 ret
= LTTNG_ERR_FILTER_NOMEM
;
3143 /* Receive var. len. data */
3144 DBG("Receiving var len filter's expression from client ...");
3145 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3148 DBG("Nothing recv() from client car len data... continuing");
3150 free(filter_expression
);
3152 ret
= LTTNG_ERR_FILTER_INVAL
;
3157 /* Handle filter and get bytecode from client. */
3158 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3159 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3161 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3162 ret
= LTTNG_ERR_FILTER_INVAL
;
3167 bytecode
= zmalloc(bytecode_len
);
3170 ret
= LTTNG_ERR_FILTER_NOMEM
;
3174 /* Receive var. len. data */
3175 DBG("Receiving var len filter's bytecode from client ...");
3176 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3178 DBG("Nothing recv() from client car len data... continuing");
3182 ret
= LTTNG_ERR_FILTER_INVAL
;
3186 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3189 ret
= LTTNG_ERR_FILTER_INVAL
;
3194 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3195 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3196 &cmd_ctx
->lsm
->u
.enable
.event
,
3197 filter_expression
, bytecode
, exclusion
,
3198 kernel_poll_pipe
[1]);
3201 case LTTNG_ENABLE_ALL_EVENT
:
3203 DBG("Enabling all events");
3205 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3206 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3207 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, NULL
,
3208 kernel_poll_pipe
[1]);
3211 case LTTNG_LIST_TRACEPOINTS
:
3213 struct lttng_event
*events
;
3216 session_lock_list();
3217 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3218 session_unlock_list();
3219 if (nb_events
< 0) {
3220 /* Return value is a negative lttng_error_code. */
3226 * Setup lttng message with payload size set to the event list size in
3227 * bytes and then copy list into the llm payload.
3229 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3235 /* Copy event list into message payload */
3236 memcpy(cmd_ctx
->llm
->payload
, events
,
3237 sizeof(struct lttng_event
) * nb_events
);
3244 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3246 struct lttng_event_field
*fields
;
3249 session_lock_list();
3250 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3252 session_unlock_list();
3253 if (nb_fields
< 0) {
3254 /* Return value is a negative lttng_error_code. */
3260 * Setup lttng message with payload size set to the event list size in
3261 * bytes and then copy list into the llm payload.
3263 ret
= setup_lttng_msg(cmd_ctx
,
3264 sizeof(struct lttng_event_field
) * nb_fields
);
3270 /* Copy event list into message payload */
3271 memcpy(cmd_ctx
->llm
->payload
, fields
,
3272 sizeof(struct lttng_event_field
) * nb_fields
);
3279 case LTTNG_SET_CONSUMER_URI
:
3282 struct lttng_uri
*uris
;
3284 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3285 len
= nb_uri
* sizeof(struct lttng_uri
);
3288 ret
= LTTNG_ERR_INVALID
;
3292 uris
= zmalloc(len
);
3294 ret
= LTTNG_ERR_FATAL
;
3298 /* Receive variable len data */
3299 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3300 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3302 DBG("No URIs received from client... continuing");
3304 ret
= LTTNG_ERR_SESSION_FAIL
;
3309 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3311 if (ret
!= LTTNG_OK
) {
3317 * XXX: 0 means that this URI should be applied on the session. Should
3318 * be a DOMAIN enuam.
3320 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3321 /* Add the URI for the UST session if a consumer is present. */
3322 if (cmd_ctx
->session
->ust_session
&&
3323 cmd_ctx
->session
->ust_session
->consumer
) {
3324 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3326 } else if (cmd_ctx
->session
->kernel_session
&&
3327 cmd_ctx
->session
->kernel_session
->consumer
) {
3328 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3329 cmd_ctx
->session
, nb_uri
, uris
);
3337 case LTTNG_START_TRACE
:
3339 ret
= cmd_start_trace(cmd_ctx
->session
);
3342 case LTTNG_STOP_TRACE
:
3344 ret
= cmd_stop_trace(cmd_ctx
->session
);
3347 case LTTNG_CREATE_SESSION
:
3350 struct lttng_uri
*uris
= NULL
;
3352 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3353 len
= nb_uri
* sizeof(struct lttng_uri
);
3356 uris
= zmalloc(len
);
3358 ret
= LTTNG_ERR_FATAL
;
3362 /* Receive variable len data */
3363 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3364 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3366 DBG("No URIs received from client... continuing");
3368 ret
= LTTNG_ERR_SESSION_FAIL
;
3373 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3374 DBG("Creating session with ONE network URI is a bad call");
3375 ret
= LTTNG_ERR_SESSION_FAIL
;
3381 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3382 &cmd_ctx
->creds
, 0);
3388 case LTTNG_DESTROY_SESSION
:
3390 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3392 /* Set session to NULL so we do not unlock it after free. */
3393 cmd_ctx
->session
= NULL
;
3396 case LTTNG_LIST_DOMAINS
:
3399 struct lttng_domain
*domains
;
3401 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3403 /* Return value is a negative lttng_error_code. */
3408 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3414 /* Copy event list into message payload */
3415 memcpy(cmd_ctx
->llm
->payload
, domains
,
3416 nb_dom
* sizeof(struct lttng_domain
));
3423 case LTTNG_LIST_CHANNELS
:
3426 struct lttng_channel
*channels
;
3428 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3429 cmd_ctx
->session
, &channels
);
3431 /* Return value is a negative lttng_error_code. */
3436 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3442 /* Copy event list into message payload */
3443 memcpy(cmd_ctx
->llm
->payload
, channels
,
3444 nb_chan
* sizeof(struct lttng_channel
));
3451 case LTTNG_LIST_EVENTS
:
3454 struct lttng_event
*events
= NULL
;
3456 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3457 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3459 /* Return value is a negative lttng_error_code. */
3464 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3470 /* Copy event list into message payload */
3471 memcpy(cmd_ctx
->llm
->payload
, events
,
3472 nb_event
* sizeof(struct lttng_event
));
3479 case LTTNG_LIST_SESSIONS
:
3481 unsigned int nr_sessions
;
3483 session_lock_list();
3484 nr_sessions
= lttng_sessions_count(
3485 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3486 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3488 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3490 session_unlock_list();
3494 /* Filled the session array */
3495 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3496 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3497 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3499 session_unlock_list();
3504 case LTTNG_CALIBRATE
:
3506 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3507 &cmd_ctx
->lsm
->u
.calibrate
);
3510 case LTTNG_REGISTER_CONSUMER
:
3512 struct consumer_data
*cdata
;
3514 switch (cmd_ctx
->lsm
->domain
.type
) {
3515 case LTTNG_DOMAIN_KERNEL
:
3516 cdata
= &kconsumer_data
;
3519 ret
= LTTNG_ERR_UND
;
3523 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3524 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3527 case LTTNG_DATA_PENDING
:
3529 ret
= cmd_data_pending(cmd_ctx
->session
);
3532 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3534 struct lttcomm_lttng_output_id reply
;
3536 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3537 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3538 if (ret
!= LTTNG_OK
) {
3542 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3547 /* Copy output list into message payload */
3548 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3552 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3554 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3555 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3558 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3561 struct lttng_snapshot_output
*outputs
= NULL
;
3563 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3564 if (nb_output
< 0) {
3569 ret
= setup_lttng_msg(cmd_ctx
,
3570 nb_output
* sizeof(struct lttng_snapshot_output
));
3577 /* Copy output list into message payload */
3578 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3579 nb_output
* sizeof(struct lttng_snapshot_output
));
3586 case LTTNG_SNAPSHOT_RECORD
:
3588 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3589 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3590 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3593 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3596 struct lttng_uri
*uris
= NULL
;
3598 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3599 len
= nb_uri
* sizeof(struct lttng_uri
);
3602 uris
= zmalloc(len
);
3604 ret
= LTTNG_ERR_FATAL
;
3608 /* Receive variable len data */
3609 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3610 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3612 DBG("No URIs received from client... continuing");
3614 ret
= LTTNG_ERR_SESSION_FAIL
;
3619 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3620 DBG("Creating session with ONE network URI is a bad call");
3621 ret
= LTTNG_ERR_SESSION_FAIL
;
3627 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3628 nb_uri
, &cmd_ctx
->creds
);
3632 case LTTNG_CREATE_SESSION_LIVE
:
3635 struct lttng_uri
*uris
= NULL
;
3637 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3638 len
= nb_uri
* sizeof(struct lttng_uri
);
3641 uris
= zmalloc(len
);
3643 ret
= LTTNG_ERR_FATAL
;
3647 /* Receive variable len data */
3648 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3649 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3651 DBG("No URIs received from client... continuing");
3653 ret
= LTTNG_ERR_SESSION_FAIL
;
3658 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3659 DBG("Creating session with ONE network URI is a bad call");
3660 ret
= LTTNG_ERR_SESSION_FAIL
;
3666 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3667 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3671 case LTTNG_SAVE_SESSION
:
3673 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
3678 ret
= LTTNG_ERR_UND
;
3683 if (cmd_ctx
->llm
== NULL
) {
3684 DBG("Missing llm structure. Allocating one.");
3685 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3689 /* Set return code */
3690 cmd_ctx
->llm
->ret_code
= ret
;
3692 if (cmd_ctx
->session
) {
3693 session_unlock(cmd_ctx
->session
);
3695 if (need_tracing_session
) {
3696 session_unlock_list();
3703 * Thread managing health check socket.
3705 static void *thread_manage_health(void *data
)
3707 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3708 uint32_t revents
, nb_fd
;
3709 struct lttng_poll_event events
;
3710 struct health_comm_msg msg
;
3711 struct health_comm_reply reply
;
3713 DBG("[thread] Manage health check started");
3715 rcu_register_thread();
3717 /* We might hit an error path before this is created. */
3718 lttng_poll_init(&events
);
3720 /* Create unix socket */
3721 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3723 ERR("Unable to create health check Unix socket");
3729 /* lttng health client socket path permissions */
3730 ret
= chown(health_unix_sock_path
, 0,
3731 utils_get_group_id(tracing_group_name
));
3733 ERR("Unable to set group on %s", health_unix_sock_path
);
3739 ret
= chmod(health_unix_sock_path
,
3740 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3742 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3750 * Set the CLOEXEC flag. Return code is useless because either way, the
3753 (void) utils_set_fd_cloexec(sock
);
3755 ret
= lttcomm_listen_unix_sock(sock
);
3761 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3762 * more will be added to this poll set.
3764 ret
= sessiond_set_thread_pollset(&events
, 2);
3769 /* Add the application registration socket */
3770 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3775 sessiond_notify_ready();
3778 DBG("Health check ready");
3780 /* Inifinite blocking call, waiting for transmission */
3782 ret
= lttng_poll_wait(&events
, -1);
3785 * Restart interrupted system call.
3787 if (errno
== EINTR
) {
3795 for (i
= 0; i
< nb_fd
; i
++) {
3796 /* Fetch once the poll data */
3797 revents
= LTTNG_POLL_GETEV(&events
, i
);
3798 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3800 /* Thread quit pipe has been closed. Killing thread. */
3801 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3807 /* Event on the registration socket */
3808 if (pollfd
== sock
) {
3809 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3810 ERR("Health socket poll error");
3816 new_sock
= lttcomm_accept_unix_sock(sock
);
3822 * Set the CLOEXEC flag. Return code is useless because either way, the
3825 (void) utils_set_fd_cloexec(new_sock
);
3827 DBG("Receiving data from client for health...");
3828 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3830 DBG("Nothing recv() from client... continuing");
3831 ret
= close(new_sock
);
3839 rcu_thread_online();
3841 memset(&reply
, 0, sizeof(reply
));
3842 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3844 * health_check_state returns 0 if health is
3847 if (!health_check_state(health_sessiond
, i
)) {
3848 reply
.ret_code
|= 1ULL << i
;
3852 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3854 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3856 ERR("Failed to send health data back to client");
3859 /* End of transmission */
3860 ret
= close(new_sock
);
3870 ERR("Health error occurred in %s", __func__
);
3872 DBG("Health check thread dying");
3873 unlink(health_unix_sock_path
);
3881 lttng_poll_clean(&events
);
3883 rcu_unregister_thread();
3888 * This thread manage all clients request using the unix client socket for
3891 static void *thread_manage_clients(void *data
)
3893 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3895 uint32_t revents
, nb_fd
;
3896 struct command_ctx
*cmd_ctx
= NULL
;
3897 struct lttng_poll_event events
;
3899 DBG("[thread] Manage client started");
3901 rcu_register_thread();
3903 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3905 health_code_update();
3907 ret
= lttcomm_listen_unix_sock(client_sock
);
3913 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3914 * more will be added to this poll set.
3916 ret
= sessiond_set_thread_pollset(&events
, 2);
3918 goto error_create_poll
;
3921 /* Add the application registration socket */
3922 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3927 sessiond_notify_ready();
3928 ret
= sem_post(&load_info
->message_thread_ready
);
3930 PERROR("sem_post message_thread_ready");
3934 /* This testpoint is after we signal readiness to the parent. */
3935 if (testpoint(sessiond_thread_manage_clients
)) {
3939 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
3943 health_code_update();
3946 DBG("Accepting client command ...");
3948 /* Inifinite blocking call, waiting for transmission */
3950 health_poll_entry();
3951 ret
= lttng_poll_wait(&events
, -1);
3955 * Restart interrupted system call.
3957 if (errno
== EINTR
) {
3965 for (i
= 0; i
< nb_fd
; i
++) {
3966 /* Fetch once the poll data */
3967 revents
= LTTNG_POLL_GETEV(&events
, i
);
3968 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3970 health_code_update();
3972 /* Thread quit pipe has been closed. Killing thread. */
3973 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3979 /* Event on the registration socket */
3980 if (pollfd
== client_sock
) {
3981 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3982 ERR("Client socket poll error");
3988 DBG("Wait for client response");
3990 health_code_update();
3992 sock
= lttcomm_accept_unix_sock(client_sock
);
3998 * Set the CLOEXEC flag. Return code is useless because either way, the
4001 (void) utils_set_fd_cloexec(sock
);
4003 /* Set socket option for credentials retrieval */
4004 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4009 /* Allocate context command to process the client request */
4010 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4011 if (cmd_ctx
== NULL
) {
4012 PERROR("zmalloc cmd_ctx");
4016 /* Allocate data buffer for reception */
4017 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4018 if (cmd_ctx
->lsm
== NULL
) {
4019 PERROR("zmalloc cmd_ctx->lsm");
4023 cmd_ctx
->llm
= NULL
;
4024 cmd_ctx
->session
= NULL
;
4026 health_code_update();
4029 * Data is received from the lttng client. The struct
4030 * lttcomm_session_msg (lsm) contains the command and data request of
4033 DBG("Receiving data from client ...");
4034 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4035 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4037 DBG("Nothing recv() from client... continuing");
4043 clean_command_ctx(&cmd_ctx
);
4047 health_code_update();
4049 // TODO: Validate cmd_ctx including sanity check for
4050 // security purpose.
4052 rcu_thread_online();
4054 * This function dispatch the work to the kernel or userspace tracer
4055 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4056 * informations for the client. The command context struct contains
4057 * everything this function may needs.
4059 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4060 rcu_thread_offline();
4068 * TODO: Inform client somehow of the fatal error. At
4069 * this point, ret < 0 means that a zmalloc failed
4070 * (ENOMEM). Error detected but still accept
4071 * command, unless a socket error has been
4074 clean_command_ctx(&cmd_ctx
);
4078 health_code_update();
4080 DBG("Sending response (size: %d, retcode: %s)",
4081 cmd_ctx
->lttng_msg_size
,
4082 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4083 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4085 ERR("Failed to send data back to client");
4088 /* End of transmission */
4095 clean_command_ctx(&cmd_ctx
);
4097 health_code_update();
4109 lttng_poll_clean(&events
);
4110 clean_command_ctx(&cmd_ctx
);
4114 unlink(client_unix_sock_path
);
4115 if (client_sock
>= 0) {
4116 ret
= close(client_sock
);
4124 ERR("Health error occurred in %s", __func__
);
4127 health_unregister(health_sessiond
);
4129 DBG("Client thread dying");
4131 rcu_unregister_thread();
4137 * usage function on stderr
4139 static void usage(void)
4141 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4142 fprintf(stderr
, " -h, --help Display this usage.\n");
4143 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4144 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4145 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4146 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4147 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4148 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4149 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4150 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4151 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4152 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4153 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4154 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4155 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4156 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4157 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4158 fprintf(stderr
, " -V, --version Show version number.\n");
4159 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4160 fprintf(stderr
, " -q, --quiet No output at all.\n");
4161 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4162 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4163 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4164 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4165 fprintf(stderr
, " --jul-tcp-port JUL application registration TCP port\n");
4166 fprintf(stderr
, " -f --config Load daemon configuration file\n");
4167 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4168 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4172 * Take an option from the getopt output and set it in the right variable to be
4175 * Return 0 on success else a negative value.
4177 static int set_option(int opt
, const char *arg
, const char *optname
)
4183 fprintf(stderr
, "option %s", optname
);
4185 fprintf(stderr
, " with arg %s\n", arg
);
4189 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4192 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4201 tracing_group_name
= strdup(arg
);
4207 fprintf(stdout
, "%s\n", VERSION
);
4213 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4216 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4219 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4222 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4225 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4228 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4234 lttng_opt_quiet
= 1;
4237 /* Verbose level can increase using multiple -v */
4239 lttng_opt_verbose
= config_parse_value(arg
);
4241 /* Only 3 level of verbosity (-vvv). */
4242 if (lttng_opt_verbose
< 3) {
4243 lttng_opt_verbose
+= 1;
4249 opt_verbose_consumer
= config_parse_value(arg
);
4251 opt_verbose_consumer
+= 1;
4255 consumerd32_bin
= strdup(arg
);
4256 consumerd32_bin_override
= 1;
4259 consumerd32_libdir
= strdup(arg
);
4260 consumerd32_libdir_override
= 1;
4263 consumerd64_bin
= strdup(arg
);
4264 consumerd64_bin_override
= 1;
4267 consumerd64_libdir
= strdup(arg
);
4268 consumerd64_libdir_override
= 1;
4271 opt_pidfile
= strdup(arg
);
4273 case 'J': /* JUL TCP port. */
4278 v
= strtoul(arg
, NULL
, 0);
4279 if (errno
!= 0 || !isdigit(arg
[0])) {
4280 ERR("Wrong value in --jul-tcp-port parameter: %s", arg
);
4283 if (v
== 0 || v
>= 65535) {
4284 ERR("Port overflow in --jul-tcp-port parameter: %s", arg
);
4287 jul_tcp_port
= (uint32_t) v
;
4288 DBG3("JUL TCP port set to non default: %u", jul_tcp_port
);
4292 opt_load_session_path
= strdup(arg
);
4293 if (!opt_load_session_path
) {
4298 case 'P': /* probe modules list */
4299 kmod_probes_list
= strdup(arg
);
4302 /* This is handled in set_options() thus silent break. */
4305 /* Unknown option or other error.
4306 * Error is printed by getopt, just return */
4314 * config_entry_handler_cb used to handle options read from a config file.
4315 * See config_entry_handler_cb comment in common/config/config.h for the
4316 * return value conventions.
4318 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4322 if (!entry
|| !entry
->name
|| !entry
->value
) {
4327 /* Check if the option is to be ignored */
4328 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4329 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4334 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4337 /* Ignore if not fully matched. */
4338 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4343 * If the option takes no argument on the command line, we have to
4344 * check if the value is "true". We support non-zero numeric values,
4347 if (!long_options
[i
].has_arg
) {
4348 ret
= config_parse_value(entry
->value
);
4351 WARN("Invalid configuration value \"%s\" for option %s",
4352 entry
->value
, entry
->name
);
4354 /* False, skip boolean config option. */
4359 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4363 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4370 * daemon configuration loading and argument parsing
4372 static int set_options(int argc
, char **argv
)
4374 int ret
= 0, c
= 0, option_index
= 0;
4375 int orig_optopt
= optopt
, orig_optind
= optind
;
4377 const char *config_path
= NULL
;
4379 optstring
= utils_generate_optstring(long_options
,
4380 sizeof(long_options
) / sizeof(struct option
));
4386 /* Check for the --config option */
4387 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4388 &option_index
)) != -1) {
4392 } else if (c
!= 'f') {
4393 /* if not equal to --config option. */
4397 config_path
= utils_expand_path(optarg
);
4399 ERR("Failed to resolve path: %s", optarg
);
4403 ret
= config_get_section_entries(config_path
, config_section_name
,
4404 config_entry_handler
, NULL
);
4407 ERR("Invalid configuration option at line %i", ret
);
4413 /* Reset getopt's global state */
4414 optopt
= orig_optopt
;
4415 optind
= orig_optind
;
4417 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
4422 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
4434 * Creates the two needed socket by the daemon.
4435 * apps_sock - The communication socket for all UST apps.
4436 * client_sock - The communication of the cli tool (lttng).
4438 static int init_daemon_socket(void)
4443 old_umask
= umask(0);
4445 /* Create client tool unix socket */
4446 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4447 if (client_sock
< 0) {
4448 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4453 /* Set the cloexec flag */
4454 ret
= utils_set_fd_cloexec(client_sock
);
4456 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4457 "Continuing but note that the consumer daemon will have a "
4458 "reference to this socket on exec()", client_sock
);
4461 /* File permission MUST be 660 */
4462 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4464 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4469 /* Create the application unix socket */
4470 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4471 if (apps_sock
< 0) {
4472 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4477 /* Set the cloexec flag */
4478 ret
= utils_set_fd_cloexec(apps_sock
);
4480 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4481 "Continuing but note that the consumer daemon will have a "
4482 "reference to this socket on exec()", apps_sock
);
4485 /* File permission MUST be 666 */
4486 ret
= chmod(apps_unix_sock_path
,
4487 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4489 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4494 DBG3("Session daemon client socket %d and application socket %d created",
4495 client_sock
, apps_sock
);
4503 * Check if the global socket is available, and if a daemon is answering at the
4504 * other side. If yes, error is returned.
4506 static int check_existing_daemon(void)
4508 /* Is there anybody out there ? */
4509 if (lttng_session_daemon_alive()) {
4517 * Set the tracing group gid onto the client socket.
4519 * Race window between mkdir and chown is OK because we are going from more
4520 * permissive (root.root) to less permissive (root.tracing).
4522 static int set_permissions(char *rundir
)
4527 gid
= utils_get_group_id(tracing_group_name
);
4529 /* Set lttng run dir */
4530 ret
= chown(rundir
, 0, gid
);
4532 ERR("Unable to set group on %s", rundir
);
4537 * Ensure all applications and tracing group can search the run
4538 * dir. Allow everyone to read the directory, since it does not
4539 * buy us anything to hide its content.
4541 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4543 ERR("Unable to set permissions on %s", rundir
);
4547 /* lttng client socket path */
4548 ret
= chown(client_unix_sock_path
, 0, gid
);
4550 ERR("Unable to set group on %s", client_unix_sock_path
);
4554 /* kconsumer error socket path */
4555 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4557 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4561 /* 64-bit ustconsumer error socket path */
4562 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4564 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4568 /* 32-bit ustconsumer compat32 error socket path */
4569 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4571 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4575 DBG("All permissions are set");
4581 * Create the lttng run directory needed for all global sockets and pipe.
4583 static int create_lttng_rundir(const char *rundir
)
4587 DBG3("Creating LTTng run directory: %s", rundir
);
4589 ret
= mkdir(rundir
, S_IRWXU
);
4591 if (errno
!= EEXIST
) {
4592 ERR("Unable to create %s", rundir
);
4604 * Setup sockets and directory needed by the kconsumerd communication with the
4607 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4611 char path
[PATH_MAX
];
4613 switch (consumer_data
->type
) {
4614 case LTTNG_CONSUMER_KERNEL
:
4615 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4617 case LTTNG_CONSUMER64_UST
:
4618 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4620 case LTTNG_CONSUMER32_UST
:
4621 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4624 ERR("Consumer type unknown");
4629 DBG2("Creating consumer directory: %s", path
);
4631 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4633 if (errno
!= EEXIST
) {
4635 ERR("Failed to create %s", path
);
4641 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4643 ERR("Unable to set group on %s", path
);
4649 /* Create the kconsumerd error unix socket */
4650 consumer_data
->err_sock
=
4651 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4652 if (consumer_data
->err_sock
< 0) {
4653 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4659 * Set the CLOEXEC flag. Return code is useless because either way, the
4662 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4664 PERROR("utils_set_fd_cloexec");
4665 /* continue anyway */
4668 /* File permission MUST be 660 */
4669 ret
= chmod(consumer_data
->err_unix_sock_path
,
4670 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4672 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4682 * Signal handler for the daemon
4684 * Simply stop all worker threads, leaving main() return gracefully after
4685 * joining all threads and calling cleanup().
4687 static void sighandler(int sig
)
4691 DBG("SIGPIPE caught");
4694 DBG("SIGINT caught");
4698 DBG("SIGTERM caught");
4702 CMM_STORE_SHARED(recv_child_signal
, 1);
4710 * Setup signal handler for :
4711 * SIGINT, SIGTERM, SIGPIPE
4713 static int set_signal_handler(void)
4716 struct sigaction sa
;
4719 if ((ret
= sigemptyset(&sigset
)) < 0) {
4720 PERROR("sigemptyset");
4724 sa
.sa_handler
= sighandler
;
4725 sa
.sa_mask
= sigset
;
4727 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4728 PERROR("sigaction");
4732 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4733 PERROR("sigaction");
4737 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4738 PERROR("sigaction");
4742 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4743 PERROR("sigaction");
4747 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4753 * Set open files limit to unlimited. This daemon can open a large number of
4754 * file descriptors in order to consumer multiple kernel traces.
4756 static void set_ulimit(void)
4761 /* The kernel does not allowed an infinite limit for open files */
4762 lim
.rlim_cur
= 65535;
4763 lim
.rlim_max
= 65535;
4765 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4767 PERROR("failed to set open files limit");
4772 * Write pidfile using the rundir and opt_pidfile.
4774 static void write_pidfile(void)
4777 char pidfile_path
[PATH_MAX
];
4782 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4784 /* Build pidfile path from rundir and opt_pidfile. */
4785 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4786 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4788 PERROR("snprintf pidfile path");
4794 * Create pid file in rundir. Return value is of no importance. The
4795 * execution will continue even though we are not able to write the file.
4797 (void) utils_create_pid_file(getpid(), pidfile_path
);
4804 * Write JUL TCP port using the rundir.
4806 static void write_julport(void)
4809 char path
[PATH_MAX
];
4813 ret
= snprintf(path
, sizeof(path
), "%s/"
4814 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
, rundir
);
4816 PERROR("snprintf julport path");
4821 * Create TCP JUL port file in rundir. Return value is of no importance.
4822 * The execution will continue even though we are not able to write the
4825 (void) utils_create_pid_file(jul_tcp_port
, path
);
4832 * Start the load session thread and dettach from it so the main thread can
4833 * continue. This does not return a value since whatever the outcome, the main
4834 * thread will continue.
4836 static void start_load_session_thread(void)
4840 /* Create session loading thread. */
4841 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
4844 PERROR("pthread_create load_session_thread");
4848 ret
= pthread_detach(load_session_thread
);
4850 PERROR("pthread_detach load_session_thread");
4853 /* Everything went well so don't cleanup anything. */
4856 /* The cleanup() function will destroy the load_info data. */
4863 int main(int argc
, char **argv
)
4867 const char *home_path
, *env_app_timeout
;
4869 init_kernel_workarounds();
4871 rcu_register_thread();
4873 if ((ret
= set_signal_handler()) < 0) {
4877 setup_consumerd_path();
4879 page_size
= sysconf(_SC_PAGESIZE
);
4880 if (page_size
< 0) {
4881 PERROR("sysconf _SC_PAGESIZE");
4882 page_size
= LONG_MAX
;
4883 WARN("Fallback page size to %ld", page_size
);
4886 /* Parse arguments and load the daemon configuration file */
4888 if ((ret
= set_options(argc
, argv
)) < 0) {
4893 if (opt_daemon
|| opt_background
) {
4896 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4903 * We are in the child. Make sure all other file descriptors are
4904 * closed, in case we are called with more opened file descriptors than
4905 * the standard ones.
4907 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4912 /* Create thread quit pipe */
4913 if ((ret
= init_thread_quit_pipe()) < 0) {
4917 /* Check if daemon is UID = 0 */
4918 is_root
= !getuid();
4921 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
4923 /* Create global run dir with root access */
4924 ret
= create_lttng_rundir(rundir
);
4929 if (strlen(apps_unix_sock_path
) == 0) {
4930 snprintf(apps_unix_sock_path
, PATH_MAX
,
4931 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
4934 if (strlen(client_unix_sock_path
) == 0) {
4935 snprintf(client_unix_sock_path
, PATH_MAX
,
4936 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
4939 /* Set global SHM for ust */
4940 if (strlen(wait_shm_path
) == 0) {
4941 snprintf(wait_shm_path
, PATH_MAX
,
4942 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
4945 if (strlen(health_unix_sock_path
) == 0) {
4946 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
4947 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
4950 /* Setup kernel consumerd path */
4951 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
4952 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
4953 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
4954 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
4956 DBG2("Kernel consumer err path: %s",
4957 kconsumer_data
.err_unix_sock_path
);
4958 DBG2("Kernel consumer cmd path: %s",
4959 kconsumer_data
.cmd_unix_sock_path
);
4961 home_path
= utils_get_home_dir();
4962 if (home_path
== NULL
) {
4963 /* TODO: Add --socket PATH option */
4964 ERR("Can't get HOME directory for sockets creation.");
4970 * Create rundir from home path. This will create something like
4973 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
4979 ret
= create_lttng_rundir(rundir
);
4984 if (strlen(apps_unix_sock_path
) == 0) {
4985 snprintf(apps_unix_sock_path
, PATH_MAX
,
4986 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
4989 /* Set the cli tool unix socket path */
4990 if (strlen(client_unix_sock_path
) == 0) {
4991 snprintf(client_unix_sock_path
, PATH_MAX
,
4992 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
4995 /* Set global SHM for ust */
4996 if (strlen(wait_shm_path
) == 0) {
4997 snprintf(wait_shm_path
, PATH_MAX
,
4998 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
5001 /* Set health check Unix path */
5002 if (strlen(health_unix_sock_path
) == 0) {
5003 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
5004 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
5008 /* Set consumer initial state */
5009 kernel_consumerd_state
= CONSUMER_STOPPED
;
5010 ust_consumerd_state
= CONSUMER_STOPPED
;
5012 DBG("Client socket path %s", client_unix_sock_path
);
5013 DBG("Application socket path %s", apps_unix_sock_path
);
5014 DBG("Application wait path %s", wait_shm_path
);
5015 DBG("LTTng run directory path: %s", rundir
);
5017 /* 32 bits consumerd path setup */
5018 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5019 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5020 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5021 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5023 DBG2("UST consumer 32 bits err path: %s",
5024 ustconsumer32_data
.err_unix_sock_path
);
5025 DBG2("UST consumer 32 bits cmd path: %s",
5026 ustconsumer32_data
.cmd_unix_sock_path
);
5028 /* 64 bits consumerd path setup */
5029 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5030 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5031 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5032 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5034 DBG2("UST consumer 64 bits err path: %s",
5035 ustconsumer64_data
.err_unix_sock_path
);
5036 DBG2("UST consumer 64 bits cmd path: %s",
5037 ustconsumer64_data
.cmd_unix_sock_path
);
5040 * See if daemon already exist.
5042 if ((ret
= check_existing_daemon()) < 0) {
5043 ERR("Already running daemon.\n");
5045 * We do not goto exit because we must not cleanup()
5046 * because a daemon is already running.
5052 * Init UST app hash table. Alloc hash table before this point since
5053 * cleanup() can get called after that point.
5057 /* Initialize JUL domain subsystem. */
5058 if ((ret
= jul_init()) < 0) {
5059 /* ENOMEM at this point. */
5063 /* After this point, we can safely call cleanup() with "goto exit" */
5066 * These actions must be executed as root. We do that *after* setting up
5067 * the sockets path because we MUST make the check for another daemon using
5068 * those paths *before* trying to set the kernel consumer sockets and init
5072 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
5077 /* Setup kernel tracer */
5078 if (!opt_no_kernel
) {
5079 init_kernel_tracer();
5082 /* Set ulimit for open files */
5085 /* init lttng_fd tracking must be done after set_ulimit. */
5088 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
5093 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
5098 /* Setup the needed unix socket */
5099 if ((ret
= init_daemon_socket()) < 0) {
5103 /* Set credentials to socket */
5104 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
5108 /* Get parent pid if -S, --sig-parent is specified. */
5109 if (opt_sig_parent
) {
5113 /* Setup the kernel pipe for waking up the kernel thread */
5114 if (is_root
&& !opt_no_kernel
) {
5115 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
5120 /* Setup the thread ht_cleanup communication pipe. */
5121 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
5125 /* Setup the thread apps communication pipe. */
5126 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
5130 /* Setup the thread apps notify communication pipe. */
5131 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
5135 /* Initialize global buffer per UID and PID registry. */
5136 buffer_reg_init_uid_registry();
5137 buffer_reg_init_pid_registry();
5139 /* Init UST command queue. */
5140 cds_wfq_init(&ust_cmd_queue
.queue
);
5143 * Get session list pointer. This pointer MUST NOT be free(). This list is
5144 * statically declared in session.c
5146 session_list_ptr
= session_get_list();
5148 /* Set up max poll set size */
5149 lttng_poll_set_max_size();
5153 /* Check for the application socket timeout env variable. */
5154 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5155 if (env_app_timeout
) {
5156 app_socket_timeout
= atoi(env_app_timeout
);
5158 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5164 /* Initialize communication library */
5166 /* This is to get the TCP timeout value. */
5167 lttcomm_inet_init();
5169 if (load_session_init_data(&load_info
) < 0) {
5172 load_info
->path
= opt_load_session_path
;
5175 * Initialize the health check subsystem. This call should set the
5176 * appropriate time values.
5178 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5179 if (!health_sessiond
) {
5180 PERROR("health_app_create error");
5181 goto exit_health_sessiond_cleanup
;
5184 /* Create thread to clean up RCU hash tables */
5185 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5186 thread_ht_cleanup
, (void *) NULL
);
5188 PERROR("pthread_create ht_cleanup");
5189 goto exit_ht_cleanup
;
5192 /* Create health-check thread */
5193 ret
= pthread_create(&health_thread
, NULL
,
5194 thread_manage_health
, (void *) NULL
);
5196 PERROR("pthread_create health");
5200 /* Create thread to manage the client socket */
5201 ret
= pthread_create(&client_thread
, NULL
,
5202 thread_manage_clients
, (void *) NULL
);
5204 PERROR("pthread_create clients");
5208 /* Create thread to dispatch registration */
5209 ret
= pthread_create(&dispatch_thread
, NULL
,
5210 thread_dispatch_ust_registration
, (void *) NULL
);
5212 PERROR("pthread_create dispatch");
5216 /* Create thread to manage application registration. */
5217 ret
= pthread_create(®_apps_thread
, NULL
,
5218 thread_registration_apps
, (void *) NULL
);
5220 PERROR("pthread_create registration");
5224 /* Create thread to manage application socket */
5225 ret
= pthread_create(&apps_thread
, NULL
,
5226 thread_manage_apps
, (void *) NULL
);
5228 PERROR("pthread_create apps");
5232 /* Create thread to manage application notify socket */
5233 ret
= pthread_create(&apps_notify_thread
, NULL
,
5234 ust_thread_manage_notify
, (void *) NULL
);
5236 PERROR("pthread_create notify");
5237 goto exit_apps_notify
;
5240 /* Create JUL registration thread. */
5241 ret
= pthread_create(&jul_reg_thread
, NULL
,
5242 jul_thread_manage_registration
, (void *) NULL
);
5244 PERROR("pthread_create JUL");
5248 /* Don't start this thread if kernel tracing is not requested nor root */
5249 if (is_root
&& !opt_no_kernel
) {
5250 /* Create kernel thread to manage kernel event */
5251 ret
= pthread_create(&kernel_thread
, NULL
,
5252 thread_manage_kernel
, (void *) NULL
);
5254 PERROR("pthread_create kernel");
5259 /* Load possible session(s). */
5260 start_load_session_thread();
5262 if (is_root
&& !opt_no_kernel
) {
5263 ret
= pthread_join(kernel_thread
, &status
);
5265 PERROR("pthread_join");
5266 goto error
; /* join error, exit without cleanup */
5271 ret
= pthread_join(jul_reg_thread
, &status
);
5273 PERROR("pthread_join JUL");
5274 goto error
; /* join error, exit without cleanup */
5278 ret
= pthread_join(apps_notify_thread
, &status
);
5280 PERROR("pthread_join apps notify");
5281 goto error
; /* join error, exit without cleanup */
5285 ret
= pthread_join(apps_thread
, &status
);
5287 PERROR("pthread_join apps");
5288 goto error
; /* join error, exit without cleanup */
5293 ret
= pthread_join(reg_apps_thread
, &status
);
5295 PERROR("pthread_join");
5296 goto error
; /* join error, exit without cleanup */
5300 ret
= pthread_join(dispatch_thread
, &status
);
5302 PERROR("pthread_join");
5303 goto error
; /* join error, exit without cleanup */
5307 ret
= pthread_join(client_thread
, &status
);
5309 PERROR("pthread_join");
5310 goto error
; /* join error, exit without cleanup */
5313 ret
= join_consumer_thread(&kconsumer_data
);
5315 PERROR("join_consumer");
5316 goto error
; /* join error, exit without cleanup */
5319 ret
= join_consumer_thread(&ustconsumer32_data
);
5321 PERROR("join_consumer ust32");
5322 goto error
; /* join error, exit without cleanup */
5325 ret
= join_consumer_thread(&ustconsumer64_data
);
5327 PERROR("join_consumer ust64");
5328 goto error
; /* join error, exit without cleanup */
5332 ret
= pthread_join(health_thread
, &status
);
5334 PERROR("pthread_join health thread");
5335 goto error
; /* join error, exit without cleanup */
5339 ret
= pthread_join(ht_cleanup_thread
, &status
);
5341 PERROR("pthread_join ht cleanup thread");
5342 goto error
; /* join error, exit without cleanup */
5345 health_app_destroy(health_sessiond
);
5346 exit_health_sessiond_cleanup
:
5349 * cleanup() is called when no other thread is running.
5351 rcu_thread_online();
5353 rcu_thread_offline();
5354 rcu_unregister_thread();