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.
33 #include <sys/mount.h>
34 #include <sys/resource.h>
35 #include <sys/socket.h>
37 #include <sys/types.h>
39 #include <urcu/uatomic.h>
43 #include <common/common.h>
44 #include <common/compat/socket.h>
45 #include <common/compat/getenv.h>
46 #include <common/defaults.h>
47 #include <common/kernel-consumer/kernel-consumer.h>
48 #include <common/futex.h>
49 #include <common/relayd/relayd.h>
50 #include <common/utils.h>
51 #include <common/daemonize.h>
52 #include <common/config/config.h>
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
62 #include "kernel-consumer.h"
66 #include "ust-consumer.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "ust-thread.h"
72 #include "agent-thread.h"
74 #include "load-session-thread.h"
77 #define CONSUMERD_FILE "lttng-consumerd"
80 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
81 static int tracing_group_name_override
;
82 static char *opt_pidfile
;
83 static int opt_sig_parent
;
84 static int opt_verbose_consumer
;
85 static int opt_daemon
, opt_background
;
86 static int opt_no_kernel
;
87 static char *opt_load_session_path
;
88 static pid_t ppid
; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
91 static int lockfile_fd
= -1;
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal
;
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
100 static struct consumer_data kconsumer_data
= {
101 .type
= LTTNG_CONSUMER_KERNEL
,
102 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
103 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer64_data
= {
112 .type
= LTTNG_CONSUMER64_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
117 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
118 .lock
= PTHREAD_MUTEX_INITIALIZER
,
119 .cond
= PTHREAD_COND_INITIALIZER
,
120 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
122 static struct consumer_data ustconsumer32_data
= {
123 .type
= LTTNG_CONSUMER32_UST
,
124 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
125 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
128 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
129 .lock
= PTHREAD_MUTEX_INITIALIZER
,
130 .cond
= PTHREAD_COND_INITIALIZER
,
131 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
134 /* Command line options */
135 static const struct option long_options
[] = {
136 { "client-sock", 1, 0, 'c' },
137 { "apps-sock", 1, 0, 'a' },
138 { "kconsumerd-cmd-sock", 1, 0, 'C' },
139 { "kconsumerd-err-sock", 1, 0, 'E' },
140 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
141 { "ustconsumerd32-err-sock", 1, 0, 'H' },
142 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
143 { "ustconsumerd64-err-sock", 1, 0, 'F' },
144 { "consumerd32-path", 1, 0, 'u' },
145 { "consumerd32-libdir", 1, 0, 'U' },
146 { "consumerd64-path", 1, 0, 't' },
147 { "consumerd64-libdir", 1, 0, 'T' },
148 { "daemonize", 0, 0, 'd' },
149 { "background", 0, 0, 'b' },
150 { "sig-parent", 0, 0, 'S' },
151 { "help", 0, 0, 'h' },
152 { "group", 1, 0, 'g' },
153 { "version", 0, 0, 'V' },
154 { "quiet", 0, 0, 'q' },
155 { "verbose", 0, 0, 'v' },
156 { "verbose-consumer", 0, 0, 'Z' },
157 { "no-kernel", 0, 0, 'N' },
158 { "pidfile", 1, 0, 'p' },
159 { "agent-tcp-port", 1, 0, 'J' },
160 { "config", 1, 0, 'f' },
161 { "load", 1, 0, 'l' },
162 { "kmod-probes", 1, 0, 'P' },
163 { "extra-kmod-probes", 1, 0, 'e' },
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options
[] = { "help", "version", "config" };
170 /* Shared between threads */
171 static int dispatch_thread_exit
;
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path
[PATH_MAX
];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path
[PATH_MAX
];
177 /* global wait shm path for UST */
178 static char wait_shm_path
[PATH_MAX
];
179 /* Global health check unix path */
180 static char health_unix_sock_path
[PATH_MAX
];
182 /* Sockets and FDs */
183 static int client_sock
= -1;
184 static int apps_sock
= -1;
185 int kernel_tracer_fd
= -1;
186 static int kernel_poll_pipe
[2] = { -1, -1 };
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
192 static int thread_quit_pipe
[2] = { -1, -1 };
193 static int ht_cleanup_quit_pipe
[2] = { -1, -1 };
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
199 static int apps_cmd_pipe
[2] = { -1, -1 };
201 int apps_cmd_notify_pipe
[2] = { -1, -1 };
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread
;
205 static pthread_t apps_notify_thread
;
206 static pthread_t reg_apps_thread
;
207 static pthread_t client_thread
;
208 static pthread_t kernel_thread
;
209 static pthread_t dispatch_thread
;
210 static pthread_t health_thread
;
211 static pthread_t ht_cleanup_thread
;
212 static pthread_t agent_reg_thread
;
213 static pthread_t load_session_thread
;
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
224 static struct ust_cmd_queue ust_cmd_queue
;
227 * Pointer initialized before thread creation.
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
236 static struct ltt_session_list
*session_list_ptr
;
238 int ust_consumerd64_fd
= -1;
239 int ust_consumerd32_fd
= -1;
241 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
242 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
243 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
244 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
245 static int consumerd32_bin_override
;
246 static int consumerd64_bin_override
;
247 static int consumerd32_libdir_override
;
248 static int consumerd64_libdir_override
;
250 static const char *module_proc_lttng
= "/proc/lttng";
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
256 enum consumerd_state
{
257 CONSUMER_STARTED
= 1,
258 CONSUMER_STOPPED
= 2,
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
268 * The following example shows a possible race condition of this scheme:
270 * consumer thread error happens
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
283 static enum consumerd_state ust_consumerd_state
;
284 static enum consumerd_state kernel_consumerd_state
;
287 * Socket timeout for receiving and sending in seconds.
289 static int app_socket_timeout
;
291 /* Set in main() with the current page size. */
294 /* Application health monitoring */
295 struct health_app
*health_sessiond
;
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
300 /* Am I root or not. */
301 int is_root
; /* Set to 1 if the daemon is running as root */
303 const char * const config_section_name
= "sessiond";
305 /* Load session thread information to operate. */
306 struct load_session_thread_data
*load_info
;
309 * Whether sessiond is ready for commands/health check requests.
310 * NR_LTTNG_SESSIOND_READY must match the number of calls to
311 * sessiond_notify_ready().
313 #define NR_LTTNG_SESSIOND_READY 3
314 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
316 /* Notify parents that we are ready for cmd and health check */
318 void sessiond_notify_ready(void)
320 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
322 * Notify parent pid that we are ready to accept command
323 * for client side. This ppid is the one from the
324 * external process that spawned us.
326 if (opt_sig_parent
) {
331 * Notify the parent of the fork() process that we are
334 if (opt_daemon
|| opt_background
) {
335 kill(child_ppid
, SIGUSR1
);
341 void setup_consumerd_path(void)
343 const char *bin
, *libdir
;
346 * Allow INSTALL_BIN_PATH to be used as a target path for the
347 * native architecture size consumer if CONFIG_CONSUMER*_PATH
348 * has not been defined.
350 #if (CAA_BITS_PER_LONG == 32)
351 if (!consumerd32_bin
[0]) {
352 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
354 if (!consumerd32_libdir
[0]) {
355 consumerd32_libdir
= INSTALL_LIB_PATH
;
357 #elif (CAA_BITS_PER_LONG == 64)
358 if (!consumerd64_bin
[0]) {
359 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
361 if (!consumerd64_libdir
[0]) {
362 consumerd64_libdir
= INSTALL_LIB_PATH
;
365 #error "Unknown bitness"
369 * runtime env. var. overrides the build default.
371 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
373 consumerd32_bin
= bin
;
375 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
377 consumerd64_bin
= bin
;
379 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
381 consumerd32_libdir
= libdir
;
383 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
385 consumerd64_libdir
= libdir
;
390 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
397 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
403 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
415 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
417 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
419 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
425 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event
*events
,
428 return __sessiond_set_thread_pollset(events
, size
,
429 ht_cleanup_quit_pipe
);
433 int __sessiond_check_thread_quit_pipe(int fd
, uint32_t events
, int a_pipe
)
435 if (fd
== a_pipe
&& (events
& LPOLLIN
)) {
442 * Check if the thread quit pipe was triggered.
444 * Return 1 if it was triggered else 0;
446 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
448 return __sessiond_check_thread_quit_pipe(fd
, events
,
449 thread_quit_pipe
[0]);
453 * Check if the ht_cleanup thread quit pipe was triggered.
455 * Return 1 if it was triggered else 0;
457 int sessiond_check_ht_cleanup_quit(int fd
, uint32_t events
)
459 return __sessiond_check_thread_quit_pipe(fd
, events
,
460 ht_cleanup_quit_pipe
[0]);
464 * Init thread quit pipe.
466 * Return -1 on error or 0 if all pipes are created.
468 static int __init_thread_quit_pipe(int *a_pipe
)
474 PERROR("thread quit pipe");
478 for (i
= 0; i
< 2; i
++) {
479 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
490 static int init_thread_quit_pipe(void)
492 return __init_thread_quit_pipe(thread_quit_pipe
);
495 static int init_ht_cleanup_quit_pipe(void)
497 return __init_thread_quit_pipe(ht_cleanup_quit_pipe
);
501 * Stop all threads by closing the thread quit pipe.
503 static void stop_threads(void)
507 /* Stopping all threads */
508 DBG("Terminating all threads");
509 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
511 ERR("write error on thread quit pipe");
514 /* Dispatch thread */
515 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
516 futex_nto1_wake(&ust_cmd_queue
.futex
);
520 * Close every consumer sockets.
522 static void close_consumer_sockets(void)
526 if (kconsumer_data
.err_sock
>= 0) {
527 ret
= close(kconsumer_data
.err_sock
);
529 PERROR("kernel consumer err_sock close");
532 if (ustconsumer32_data
.err_sock
>= 0) {
533 ret
= close(ustconsumer32_data
.err_sock
);
535 PERROR("UST consumerd32 err_sock close");
538 if (ustconsumer64_data
.err_sock
>= 0) {
539 ret
= close(ustconsumer64_data
.err_sock
);
541 PERROR("UST consumerd64 err_sock close");
544 if (kconsumer_data
.cmd_sock
>= 0) {
545 ret
= close(kconsumer_data
.cmd_sock
);
547 PERROR("kernel consumer cmd_sock close");
550 if (ustconsumer32_data
.cmd_sock
>= 0) {
551 ret
= close(ustconsumer32_data
.cmd_sock
);
553 PERROR("UST consumerd32 cmd_sock close");
556 if (ustconsumer64_data
.cmd_sock
>= 0) {
557 ret
= close(ustconsumer64_data
.cmd_sock
);
559 PERROR("UST consumerd64 cmd_sock close");
565 * Generate the full lock file path using the rundir.
567 * Return the snprintf() return value thus a negative value is an error.
569 static int generate_lock_file_path(char *path
, size_t len
)
576 /* Build lockfile path from rundir. */
577 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
579 PERROR("snprintf lockfile path");
586 * Cleanup the session daemon's data structures.
588 static void sessiond_cleanup(void)
591 struct ltt_session
*sess
, *stmp
;
594 DBG("Cleanup sessiond");
597 * Close the thread quit pipe. It has already done its job,
598 * since we are now called.
600 utils_close_pipe(thread_quit_pipe
);
603 * If opt_pidfile is undefined, the default file will be wiped when
604 * removing the rundir.
607 ret
= remove(opt_pidfile
);
609 PERROR("remove pidfile %s", opt_pidfile
);
613 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
616 snprintf(path
, PATH_MAX
,
618 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
619 DBG("Removing %s", path
);
622 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
623 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
624 DBG("Removing %s", path
);
628 snprintf(path
, PATH_MAX
,
629 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
631 DBG("Removing %s", path
);
634 snprintf(path
, PATH_MAX
,
635 DEFAULT_KCONSUMERD_PATH
,
637 DBG("Removing directory %s", path
);
640 /* ust consumerd 32 */
641 snprintf(path
, PATH_MAX
,
642 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
644 DBG("Removing %s", path
);
647 snprintf(path
, PATH_MAX
,
648 DEFAULT_USTCONSUMERD32_PATH
,
650 DBG("Removing directory %s", path
);
653 /* ust consumerd 64 */
654 snprintf(path
, PATH_MAX
,
655 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
657 DBG("Removing %s", path
);
660 snprintf(path
, PATH_MAX
,
661 DEFAULT_USTCONSUMERD64_PATH
,
663 DBG("Removing directory %s", path
);
666 DBG("Cleaning up all sessions");
668 /* Destroy session list mutex */
669 if (session_list_ptr
!= NULL
) {
670 pthread_mutex_destroy(&session_list_ptr
->lock
);
672 /* Cleanup ALL session */
673 cds_list_for_each_entry_safe(sess
, stmp
,
674 &session_list_ptr
->head
, list
) {
675 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
679 DBG("Closing all UST sockets");
680 ust_app_clean_list();
681 buffer_reg_destroy_registries();
683 if (is_root
&& !opt_no_kernel
) {
684 DBG2("Closing kernel fd");
685 if (kernel_tracer_fd
>= 0) {
686 ret
= close(kernel_tracer_fd
);
691 DBG("Unloading kernel modules");
692 modprobe_remove_lttng_all();
696 close_consumer_sockets();
699 load_session_destroy_data(load_info
);
704 * Cleanup lock file by deleting it and finaly closing it which will
705 * release the file system lock.
707 if (lockfile_fd
>= 0) {
708 char lockfile_path
[PATH_MAX
];
710 ret
= generate_lock_file_path(lockfile_path
,
711 sizeof(lockfile_path
));
713 ret
= remove(lockfile_path
);
715 PERROR("remove lock file");
717 ret
= close(lockfile_fd
);
719 PERROR("close lock file");
725 * We do NOT rmdir rundir because there are other processes
726 * using it, for instance lttng-relayd, which can start in
727 * parallel with this teardown.
734 * Cleanup the daemon's option data structures.
736 static void sessiond_cleanup_options(void)
738 DBG("Cleaning up options");
741 * If the override option is set, the pointer points to a *non* const
742 * thus freeing it even though the variable type is set to const.
744 if (tracing_group_name_override
) {
745 free((void *) tracing_group_name
);
747 if (consumerd32_bin_override
) {
748 free((void *) consumerd32_bin
);
750 if (consumerd64_bin_override
) {
751 free((void *) consumerd64_bin
);
753 if (consumerd32_libdir_override
) {
754 free((void *) consumerd32_libdir
);
756 if (consumerd64_libdir_override
) {
757 free((void *) consumerd64_libdir
);
761 free(opt_load_session_path
);
762 free(kmod_probes_list
);
763 free(kmod_extra_probes_list
);
766 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
767 "Matthew, BEET driven development works!%c[%dm",
768 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
773 * Send data on a unix socket using the liblttsessiondcomm API.
775 * Return lttcomm error code.
777 static int send_unix_sock(int sock
, void *buf
, size_t len
)
779 /* Check valid length */
784 return lttcomm_send_unix_sock(sock
, buf
, len
);
788 * Free memory of a command context structure.
790 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
792 DBG("Clean command context structure");
794 if ((*cmd_ctx
)->llm
) {
795 free((*cmd_ctx
)->llm
);
797 if ((*cmd_ctx
)->lsm
) {
798 free((*cmd_ctx
)->lsm
);
806 * Notify UST applications using the shm mmap futex.
808 static int notify_ust_apps(int active
)
812 DBG("Notifying applications of session daemon state: %d", active
);
814 /* See shm.c for this call implying mmap, shm and futex calls */
815 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
816 if (wait_shm_mmap
== NULL
) {
820 /* Wake waiting process */
821 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
823 /* Apps notified successfully */
831 * Setup the outgoing data buffer for the response (llm) by allocating the
832 * right amount of memory and copying the original information from the lsm
835 * Return total size of the buffer pointed by buf.
837 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
843 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
844 if (cmd_ctx
->llm
== NULL
) {
850 /* Copy common data */
851 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
852 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
854 cmd_ctx
->llm
->data_size
= size
;
855 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
864 * Update the kernel poll set of all channel fd available over all tracing
865 * session. Add the wakeup pipe at the end of the set.
867 static int update_kernel_poll(struct lttng_poll_event
*events
)
870 struct ltt_session
*session
;
871 struct ltt_kernel_channel
*channel
;
873 DBG("Updating kernel poll set");
876 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
877 session_lock(session
);
878 if (session
->kernel_session
== NULL
) {
879 session_unlock(session
);
883 cds_list_for_each_entry(channel
,
884 &session
->kernel_session
->channel_list
.head
, list
) {
885 /* Add channel fd to the kernel poll set */
886 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
888 session_unlock(session
);
891 DBG("Channel fd %d added to kernel set", channel
->fd
);
893 session_unlock(session
);
895 session_unlock_list();
900 session_unlock_list();
905 * Find the channel fd from 'fd' over all tracing session. When found, check
906 * for new channel stream and send those stream fds to the kernel consumer.
908 * Useful for CPU hotplug feature.
910 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
913 struct ltt_session
*session
;
914 struct ltt_kernel_session
*ksess
;
915 struct ltt_kernel_channel
*channel
;
917 DBG("Updating kernel streams for channel fd %d", fd
);
920 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
921 session_lock(session
);
922 if (session
->kernel_session
== NULL
) {
923 session_unlock(session
);
926 ksess
= session
->kernel_session
;
928 cds_list_for_each_entry(channel
,
929 &ksess
->channel_list
.head
, list
) {
930 struct lttng_ht_iter iter
;
931 struct consumer_socket
*socket
;
933 if (channel
->fd
!= fd
) {
936 DBG("Channel found, updating kernel streams");
937 ret
= kernel_open_channel_stream(channel
);
941 /* Update the stream global counter */
942 ksess
->stream_count_global
+= ret
;
945 * Have we already sent fds to the consumer? If yes, it
946 * means that tracing is started so it is safe to send
947 * our updated stream fds.
949 if (ksess
->consumer_fds_sent
!= 1
950 || ksess
->consumer
== NULL
) {
956 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
957 &iter
.iter
, socket
, node
.node
) {
958 pthread_mutex_lock(socket
->lock
);
959 ret
= kernel_consumer_send_channel_stream(socket
,
961 session
->output_traces
? 1 : 0);
962 pthread_mutex_unlock(socket
->lock
);
970 session_unlock(session
);
972 session_unlock_list();
976 session_unlock(session
);
977 session_unlock_list();
982 * For each tracing session, update newly registered apps. The session list
983 * lock MUST be acquired before calling this.
985 static void update_ust_app(int app_sock
)
987 struct ltt_session
*sess
, *stmp
;
989 /* Consumer is in an ERROR state. Stop any application update. */
990 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
991 /* Stop the update process since the consumer is dead. */
995 /* For all tracing session(s) */
996 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
998 if (sess
->ust_session
) {
999 ust_app_global_update(sess
->ust_session
, app_sock
);
1001 session_unlock(sess
);
1006 * This thread manage event coming from the kernel.
1008 * Features supported in this thread:
1011 static void *thread_manage_kernel(void *data
)
1013 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1014 uint32_t revents
, nb_fd
;
1016 struct lttng_poll_event events
;
1018 DBG("[thread] Thread manage kernel started");
1020 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1023 * This first step of the while is to clean this structure which could free
1024 * non NULL pointers so initialize it before the loop.
1026 lttng_poll_init(&events
);
1028 if (testpoint(sessiond_thread_manage_kernel
)) {
1029 goto error_testpoint
;
1032 health_code_update();
1034 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1035 goto error_testpoint
;
1039 health_code_update();
1041 if (update_poll_flag
== 1) {
1042 /* Clean events object. We are about to populate it again. */
1043 lttng_poll_clean(&events
);
1045 ret
= sessiond_set_thread_pollset(&events
, 2);
1047 goto error_poll_create
;
1050 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1055 /* This will add the available kernel channel if any. */
1056 ret
= update_kernel_poll(&events
);
1060 update_poll_flag
= 0;
1063 DBG("Thread kernel polling");
1065 /* Poll infinite value of time */
1067 health_poll_entry();
1068 ret
= lttng_poll_wait(&events
, -1);
1069 DBG("Thread kernel return from poll on %d fds",
1070 LTTNG_POLL_GETNB(&events
));
1074 * Restart interrupted system call.
1076 if (errno
== EINTR
) {
1080 } else if (ret
== 0) {
1081 /* Should not happen since timeout is infinite */
1082 ERR("Return value of poll is 0 with an infinite timeout.\n"
1083 "This should not have happened! Continuing...");
1089 for (i
= 0; i
< nb_fd
; i
++) {
1090 /* Fetch once the poll data */
1091 revents
= LTTNG_POLL_GETEV(&events
, i
);
1092 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1094 health_code_update();
1097 /* No activity for this FD (poll implementation). */
1101 /* Thread quit pipe has been closed. Killing thread. */
1102 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1108 /* Check for data on kernel pipe */
1109 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1110 (void) lttng_read(kernel_poll_pipe
[0],
1113 * Ret value is useless here, if this pipe gets any actions an
1114 * update is required anyway.
1116 update_poll_flag
= 1;
1120 * New CPU detected by the kernel. Adding kernel stream to
1121 * kernel session and updating the kernel consumer
1123 if (revents
& LPOLLIN
) {
1124 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1130 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1131 * and unregister kernel stream at this point.
1140 lttng_poll_clean(&events
);
1143 utils_close_pipe(kernel_poll_pipe
);
1144 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1147 ERR("Health error occurred in %s", __func__
);
1148 WARN("Kernel thread died unexpectedly. "
1149 "Kernel tracing can continue but CPU hotplug is disabled.");
1151 health_unregister(health_sessiond
);
1152 DBG("Kernel thread dying");
1157 * Signal pthread condition of the consumer data that the thread.
1159 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1161 pthread_mutex_lock(&data
->cond_mutex
);
1164 * The state is set before signaling. It can be any value, it's the waiter
1165 * job to correctly interpret this condition variable associated to the
1166 * consumer pthread_cond.
1168 * A value of 0 means that the corresponding thread of the consumer data
1169 * was not started. 1 indicates that the thread has started and is ready
1170 * for action. A negative value means that there was an error during the
1173 data
->consumer_thread_is_ready
= state
;
1174 (void) pthread_cond_signal(&data
->cond
);
1176 pthread_mutex_unlock(&data
->cond_mutex
);
1180 * This thread manage the consumer error sent back to the session daemon.
1182 static void *thread_manage_consumer(void *data
)
1184 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1185 uint32_t revents
, nb_fd
;
1186 enum lttcomm_return_code code
;
1187 struct lttng_poll_event events
;
1188 struct consumer_data
*consumer_data
= data
;
1190 DBG("[thread] Manage consumer started");
1192 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1194 health_code_update();
1197 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1198 * metadata_sock. Nothing more will be added to this poll set.
1200 ret
= sessiond_set_thread_pollset(&events
, 3);
1206 * The error socket here is already in a listening state which was done
1207 * just before spawning this thread to avoid a race between the consumer
1208 * daemon exec trying to connect and the listen() call.
1210 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1215 health_code_update();
1217 /* Infinite blocking call, waiting for transmission */
1219 health_poll_entry();
1221 if (testpoint(sessiond_thread_manage_consumer
)) {
1225 ret
= lttng_poll_wait(&events
, -1);
1229 * Restart interrupted system call.
1231 if (errno
== EINTR
) {
1239 for (i
= 0; i
< nb_fd
; i
++) {
1240 /* Fetch once the poll data */
1241 revents
= LTTNG_POLL_GETEV(&events
, i
);
1242 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1244 health_code_update();
1247 /* No activity for this FD (poll implementation). */
1251 /* Thread quit pipe has been closed. Killing thread. */
1252 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1258 /* Event on the registration socket */
1259 if (pollfd
== consumer_data
->err_sock
) {
1260 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1261 ERR("consumer err socket poll error");
1267 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1273 * Set the CLOEXEC flag. Return code is useless because either way, the
1276 (void) utils_set_fd_cloexec(sock
);
1278 health_code_update();
1280 DBG2("Receiving code from consumer err_sock");
1282 /* Getting status code from kconsumerd */
1283 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1284 sizeof(enum lttcomm_return_code
));
1289 health_code_update();
1290 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1291 /* Connect both socket, command and metadata. */
1292 consumer_data
->cmd_sock
=
1293 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1294 consumer_data
->metadata_fd
=
1295 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1296 if (consumer_data
->cmd_sock
< 0
1297 || consumer_data
->metadata_fd
< 0) {
1298 PERROR("consumer connect cmd socket");
1299 /* On error, signal condition and quit. */
1300 signal_consumer_condition(consumer_data
, -1);
1303 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1304 /* Create metadata socket lock. */
1305 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1306 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1307 PERROR("zmalloc pthread mutex");
1311 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1313 signal_consumer_condition(consumer_data
, 1);
1314 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1315 DBG("Consumer metadata socket ready (fd: %d)",
1316 consumer_data
->metadata_fd
);
1318 ERR("consumer error when waiting for SOCK_READY : %s",
1319 lttcomm_get_readable_code(-code
));
1323 /* Remove the consumerd error sock since we've established a connexion */
1324 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1329 /* Add new accepted error socket. */
1330 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1335 /* Add metadata socket that is successfully connected. */
1336 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1337 LPOLLIN
| LPOLLRDHUP
);
1342 health_code_update();
1344 /* Infinite blocking call, waiting for transmission */
1347 health_code_update();
1349 /* Exit the thread because the thread quit pipe has been triggered. */
1351 /* Not a health error. */
1356 health_poll_entry();
1357 ret
= lttng_poll_wait(&events
, -1);
1361 * Restart interrupted system call.
1363 if (errno
== EINTR
) {
1371 for (i
= 0; i
< nb_fd
; i
++) {
1372 /* Fetch once the poll data */
1373 revents
= LTTNG_POLL_GETEV(&events
, i
);
1374 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1376 health_code_update();
1379 /* No activity for this FD (poll implementation). */
1384 * Thread quit pipe has been triggered, flag that we should stop
1385 * but continue the current loop to handle potential data from
1388 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1390 if (pollfd
== sock
) {
1391 /* Event on the consumerd socket */
1392 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1393 ERR("consumer err socket second poll error");
1396 health_code_update();
1397 /* Wait for any kconsumerd error */
1398 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1399 sizeof(enum lttcomm_return_code
));
1401 ERR("consumer closed the command socket");
1405 ERR("consumer return code : %s",
1406 lttcomm_get_readable_code(-code
));
1409 } else if (pollfd
== consumer_data
->metadata_fd
) {
1410 /* UST metadata requests */
1411 ret
= ust_consumer_metadata_request(
1412 &consumer_data
->metadata_sock
);
1414 ERR("Handling metadata request");
1418 /* No need for an else branch all FDs are tested prior. */
1420 health_code_update();
1426 * We lock here because we are about to close the sockets and some other
1427 * thread might be using them so get exclusive access which will abort all
1428 * other consumer command by other threads.
1430 pthread_mutex_lock(&consumer_data
->lock
);
1432 /* Immediately set the consumerd state to stopped */
1433 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1434 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1435 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1436 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1437 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1439 /* Code flow error... */
1443 if (consumer_data
->err_sock
>= 0) {
1444 ret
= close(consumer_data
->err_sock
);
1448 consumer_data
->err_sock
= -1;
1450 if (consumer_data
->cmd_sock
>= 0) {
1451 ret
= close(consumer_data
->cmd_sock
);
1455 consumer_data
->cmd_sock
= -1;
1457 if (consumer_data
->metadata_sock
.fd_ptr
&&
1458 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1459 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1471 unlink(consumer_data
->err_unix_sock_path
);
1472 unlink(consumer_data
->cmd_unix_sock_path
);
1473 consumer_data
->pid
= 0;
1474 pthread_mutex_unlock(&consumer_data
->lock
);
1476 /* Cleanup metadata socket mutex. */
1477 if (consumer_data
->metadata_sock
.lock
) {
1478 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1479 free(consumer_data
->metadata_sock
.lock
);
1481 lttng_poll_clean(&events
);
1485 ERR("Health error occurred in %s", __func__
);
1487 health_unregister(health_sessiond
);
1488 DBG("consumer thread cleanup completed");
1494 * This thread manage application communication.
1496 static void *thread_manage_apps(void *data
)
1498 int i
, ret
, pollfd
, err
= -1;
1500 uint32_t revents
, nb_fd
;
1501 struct lttng_poll_event events
;
1503 DBG("[thread] Manage application started");
1505 rcu_register_thread();
1506 rcu_thread_online();
1508 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1510 if (testpoint(sessiond_thread_manage_apps
)) {
1511 goto error_testpoint
;
1514 health_code_update();
1516 ret
= sessiond_set_thread_pollset(&events
, 2);
1518 goto error_poll_create
;
1521 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1526 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1530 health_code_update();
1533 DBG("Apps thread polling");
1535 /* Inifinite blocking call, waiting for transmission */
1537 health_poll_entry();
1538 ret
= lttng_poll_wait(&events
, -1);
1539 DBG("Apps thread return from poll on %d fds",
1540 LTTNG_POLL_GETNB(&events
));
1544 * Restart interrupted system call.
1546 if (errno
== EINTR
) {
1554 for (i
= 0; i
< nb_fd
; i
++) {
1555 /* Fetch once the poll data */
1556 revents
= LTTNG_POLL_GETEV(&events
, i
);
1557 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1559 health_code_update();
1562 /* No activity for this FD (poll implementation). */
1566 /* Thread quit pipe has been closed. Killing thread. */
1567 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1573 /* Inspect the apps cmd pipe */
1574 if (pollfd
== apps_cmd_pipe
[0]) {
1575 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1576 ERR("Apps command pipe error");
1578 } else if (revents
& LPOLLIN
) {
1582 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1583 if (size_ret
< sizeof(sock
)) {
1584 PERROR("read apps cmd pipe");
1588 health_code_update();
1591 * We only monitor the error events of the socket. This
1592 * thread does not handle any incoming data from UST
1595 ret
= lttng_poll_add(&events
, sock
,
1596 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1601 DBG("Apps with sock %d added to poll set", sock
);
1605 * At this point, we know that a registered application made
1606 * the event at poll_wait.
1608 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1609 /* Removing from the poll set */
1610 ret
= lttng_poll_del(&events
, pollfd
);
1615 /* Socket closed on remote end. */
1616 ust_app_unregister(pollfd
);
1620 health_code_update();
1626 lttng_poll_clean(&events
);
1629 utils_close_pipe(apps_cmd_pipe
);
1630 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1633 * We don't clean the UST app hash table here since already registered
1634 * applications can still be controlled so let them be until the session
1635 * daemon dies or the applications stop.
1640 ERR("Health error occurred in %s", __func__
);
1642 health_unregister(health_sessiond
);
1643 DBG("Application communication apps thread cleanup complete");
1644 rcu_thread_offline();
1645 rcu_unregister_thread();
1650 * Send a socket to a thread This is called from the dispatch UST registration
1651 * thread once all sockets are set for the application.
1653 * The sock value can be invalid, we don't really care, the thread will handle
1654 * it and make the necessary cleanup if so.
1656 * On success, return 0 else a negative value being the errno message of the
1659 static int send_socket_to_thread(int fd
, int sock
)
1664 * It's possible that the FD is set as invalid with -1 concurrently just
1665 * before calling this function being a shutdown state of the thread.
1672 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1673 if (ret
< sizeof(sock
)) {
1674 PERROR("write apps pipe %d", fd
);
1681 /* All good. Don't send back the write positive ret value. */
1688 * Sanitize the wait queue of the dispatch registration thread meaning removing
1689 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1690 * notify socket is never received.
1692 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1694 int ret
, nb_fd
= 0, i
;
1695 unsigned int fd_added
= 0;
1696 struct lttng_poll_event events
;
1697 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1701 lttng_poll_init(&events
);
1703 /* Just skip everything for an empty queue. */
1704 if (!wait_queue
->count
) {
1708 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1713 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1714 &wait_queue
->head
, head
) {
1715 assert(wait_node
->app
);
1716 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1717 LPOLLHUP
| LPOLLERR
);
1730 * Poll but don't block so we can quickly identify the faulty events and
1731 * clean them afterwards from the wait queue.
1733 ret
= lttng_poll_wait(&events
, 0);
1739 for (i
= 0; i
< nb_fd
; i
++) {
1740 /* Get faulty FD. */
1741 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1742 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1745 /* No activity for this FD (poll implementation). */
1749 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1750 &wait_queue
->head
, head
) {
1751 if (pollfd
== wait_node
->app
->sock
&&
1752 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1753 cds_list_del(&wait_node
->head
);
1754 wait_queue
->count
--;
1755 ust_app_destroy(wait_node
->app
);
1763 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1767 lttng_poll_clean(&events
);
1771 lttng_poll_clean(&events
);
1773 ERR("Unable to sanitize wait queue");
1778 * Dispatch request from the registration threads to the application
1779 * communication thread.
1781 static void *thread_dispatch_ust_registration(void *data
)
1784 struct cds_wfcq_node
*node
;
1785 struct ust_command
*ust_cmd
= NULL
;
1786 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1787 struct ust_reg_wait_queue wait_queue
= {
1791 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1793 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1794 goto error_testpoint
;
1797 health_code_update();
1799 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1801 DBG("[thread] Dispatch UST command started");
1803 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1804 health_code_update();
1806 /* Atomically prepare the queue futex */
1807 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1810 struct ust_app
*app
= NULL
;
1814 * Make sure we don't have node(s) that have hung up before receiving
1815 * the notify socket. This is to clean the list in order to avoid
1816 * memory leaks from notify socket that are never seen.
1818 sanitize_wait_queue(&wait_queue
);
1820 health_code_update();
1821 /* Dequeue command for registration */
1822 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1824 DBG("Woken up but nothing in the UST command queue");
1825 /* Continue thread execution */
1829 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1831 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1832 " gid:%d sock:%d name:%s (version %d.%d)",
1833 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1834 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1835 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1836 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1838 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1839 wait_node
= zmalloc(sizeof(*wait_node
));
1841 PERROR("zmalloc wait_node dispatch");
1842 ret
= close(ust_cmd
->sock
);
1844 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1846 lttng_fd_put(LTTNG_FD_APPS
, 1);
1850 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1852 /* Create application object if socket is CMD. */
1853 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1855 if (!wait_node
->app
) {
1856 ret
= close(ust_cmd
->sock
);
1858 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1860 lttng_fd_put(LTTNG_FD_APPS
, 1);
1866 * Add application to the wait queue so we can set the notify
1867 * socket before putting this object in the global ht.
1869 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1874 * We have to continue here since we don't have the notify
1875 * socket and the application MUST be added to the hash table
1876 * only at that moment.
1881 * Look for the application in the local wait queue and set the
1882 * notify socket if found.
1884 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1885 &wait_queue
.head
, head
) {
1886 health_code_update();
1887 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1888 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1889 cds_list_del(&wait_node
->head
);
1891 app
= wait_node
->app
;
1893 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1899 * With no application at this stage the received socket is
1900 * basically useless so close it before we free the cmd data
1901 * structure for good.
1904 ret
= close(ust_cmd
->sock
);
1906 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1908 lttng_fd_put(LTTNG_FD_APPS
, 1);
1915 * @session_lock_list
1917 * Lock the global session list so from the register up to the
1918 * registration done message, no thread can see the application
1919 * and change its state.
1921 session_lock_list();
1925 * Add application to the global hash table. This needs to be
1926 * done before the update to the UST registry can locate the
1931 /* Set app version. This call will print an error if needed. */
1932 (void) ust_app_version(app
);
1934 /* Send notify socket through the notify pipe. */
1935 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1939 session_unlock_list();
1941 * No notify thread, stop the UST tracing. However, this is
1942 * not an internal error of the this thread thus setting
1943 * the health error code to a normal exit.
1950 * Update newly registered application with the tracing
1951 * registry info already enabled information.
1953 update_ust_app(app
->sock
);
1956 * Don't care about return value. Let the manage apps threads
1957 * handle app unregistration upon socket close.
1959 (void) ust_app_register_done(app
->sock
);
1962 * Even if the application socket has been closed, send the app
1963 * to the thread and unregistration will take place at that
1966 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1969 session_unlock_list();
1971 * No apps. thread, stop the UST tracing. However, this is
1972 * not an internal error of the this thread thus setting
1973 * the health error code to a normal exit.
1980 session_unlock_list();
1982 } while (node
!= NULL
);
1984 health_poll_entry();
1985 /* Futex wait on queue. Blocking call on futex() */
1986 futex_nto1_wait(&ust_cmd_queue
.futex
);
1989 /* Normal exit, no error */
1993 /* Clean up wait queue. */
1994 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1995 &wait_queue
.head
, head
) {
1996 cds_list_del(&wait_node
->head
);
2002 DBG("Dispatch thread dying");
2005 ERR("Health error occurred in %s", __func__
);
2007 health_unregister(health_sessiond
);
2012 * This thread manage application registration.
2014 static void *thread_registration_apps(void *data
)
2016 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2017 uint32_t revents
, nb_fd
;
2018 struct lttng_poll_event events
;
2020 * Get allocated in this thread, enqueued to a global queue, dequeued and
2021 * freed in the manage apps thread.
2023 struct ust_command
*ust_cmd
= NULL
;
2025 DBG("[thread] Manage application registration started");
2027 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2029 if (testpoint(sessiond_thread_registration_apps
)) {
2030 goto error_testpoint
;
2033 ret
= lttcomm_listen_unix_sock(apps_sock
);
2039 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2040 * more will be added to this poll set.
2042 ret
= sessiond_set_thread_pollset(&events
, 2);
2044 goto error_create_poll
;
2047 /* Add the application registration socket */
2048 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2050 goto error_poll_add
;
2053 /* Notify all applications to register */
2054 ret
= notify_ust_apps(1);
2056 ERR("Failed to notify applications or create the wait shared memory.\n"
2057 "Execution continues but there might be problem for already\n"
2058 "running applications that wishes to register.");
2062 DBG("Accepting application registration");
2064 /* Inifinite blocking call, waiting for transmission */
2066 health_poll_entry();
2067 ret
= lttng_poll_wait(&events
, -1);
2071 * Restart interrupted system call.
2073 if (errno
== EINTR
) {
2081 for (i
= 0; i
< nb_fd
; i
++) {
2082 health_code_update();
2084 /* Fetch once the poll data */
2085 revents
= LTTNG_POLL_GETEV(&events
, i
);
2086 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2089 /* No activity for this FD (poll implementation). */
2093 /* Thread quit pipe has been closed. Killing thread. */
2094 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2100 /* Event on the registration socket */
2101 if (pollfd
== apps_sock
) {
2102 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2103 ERR("Register apps socket poll error");
2105 } else if (revents
& LPOLLIN
) {
2106 sock
= lttcomm_accept_unix_sock(apps_sock
);
2112 * Set socket timeout for both receiving and ending.
2113 * app_socket_timeout is in seconds, whereas
2114 * lttcomm_setsockopt_rcv_timeout and
2115 * lttcomm_setsockopt_snd_timeout expect msec as
2118 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2119 app_socket_timeout
* 1000);
2120 (void) lttcomm_setsockopt_snd_timeout(sock
,
2121 app_socket_timeout
* 1000);
2124 * Set the CLOEXEC flag. Return code is useless because
2125 * either way, the show must go on.
2127 (void) utils_set_fd_cloexec(sock
);
2129 /* Create UST registration command for enqueuing */
2130 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2131 if (ust_cmd
== NULL
) {
2132 PERROR("ust command zmalloc");
2137 * Using message-based transmissions to ensure we don't
2138 * have to deal with partially received messages.
2140 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2142 ERR("Exhausted file descriptors allowed for applications.");
2152 health_code_update();
2153 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2156 /* Close socket of the application. */
2161 lttng_fd_put(LTTNG_FD_APPS
, 1);
2165 health_code_update();
2167 ust_cmd
->sock
= sock
;
2170 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2171 " gid:%d sock:%d name:%s (version %d.%d)",
2172 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2173 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2174 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2175 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2178 * Lock free enqueue the registration request. The red pill
2179 * has been taken! This apps will be part of the *system*.
2181 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2184 * Wake the registration queue futex. Implicit memory
2185 * barrier with the exchange in cds_wfcq_enqueue.
2187 futex_nto1_wake(&ust_cmd_queue
.futex
);
2195 /* Notify that the registration thread is gone */
2198 if (apps_sock
>= 0) {
2199 ret
= close(apps_sock
);
2209 lttng_fd_put(LTTNG_FD_APPS
, 1);
2211 unlink(apps_unix_sock_path
);
2214 lttng_poll_clean(&events
);
2218 DBG("UST Registration thread cleanup complete");
2221 ERR("Health error occurred in %s", __func__
);
2223 health_unregister(health_sessiond
);
2229 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2230 * exec or it will fails.
2232 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2235 struct timespec timeout
;
2237 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2238 consumer_data
->consumer_thread_is_ready
= 0;
2240 /* Setup pthread condition */
2241 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2244 PERROR("pthread_condattr_init consumer data");
2249 * Set the monotonic clock in order to make sure we DO NOT jump in time
2250 * between the clock_gettime() call and the timedwait call. See bug #324
2251 * for a more details and how we noticed it.
2253 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2256 PERROR("pthread_condattr_setclock consumer data");
2260 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2263 PERROR("pthread_cond_init consumer data");
2267 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2271 PERROR("pthread_create consumer");
2276 /* We are about to wait on a pthread condition */
2277 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2279 /* Get time for sem_timedwait absolute timeout */
2280 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2282 * Set the timeout for the condition timed wait even if the clock gettime
2283 * call fails since we might loop on that call and we want to avoid to
2284 * increment the timeout too many times.
2286 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2289 * The following loop COULD be skipped in some conditions so this is why we
2290 * set ret to 0 in order to make sure at least one round of the loop is
2296 * Loop until the condition is reached or when a timeout is reached. Note
2297 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2298 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2299 * possible. This loop does not take any chances and works with both of
2302 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2303 if (clock_ret
< 0) {
2304 PERROR("clock_gettime spawn consumer");
2305 /* Infinite wait for the consumerd thread to be ready */
2306 ret
= pthread_cond_wait(&consumer_data
->cond
,
2307 &consumer_data
->cond_mutex
);
2309 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2310 &consumer_data
->cond_mutex
, &timeout
);
2314 /* Release the pthread condition */
2315 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2319 if (ret
== ETIMEDOUT
) {
2323 * Call has timed out so we kill the kconsumerd_thread and return
2326 ERR("Condition timed out. The consumer thread was never ready."
2328 pth_ret
= pthread_cancel(consumer_data
->thread
);
2330 PERROR("pthread_cancel consumer thread");
2333 PERROR("pthread_cond_wait failed consumer thread");
2335 /* Caller is expecting a negative value on failure. */
2340 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2341 if (consumer_data
->pid
== 0) {
2342 ERR("Consumerd did not start");
2343 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2346 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2355 * Join consumer thread
2357 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2361 /* Consumer pid must be a real one. */
2362 if (consumer_data
->pid
> 0) {
2364 ret
= kill(consumer_data
->pid
, SIGTERM
);
2366 PERROR("Error killing consumer daemon");
2369 return pthread_join(consumer_data
->thread
, &status
);
2376 * Fork and exec a consumer daemon (consumerd).
2378 * Return pid if successful else -1.
2380 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2384 const char *consumer_to_use
;
2385 const char *verbosity
;
2388 DBG("Spawning consumerd");
2395 if (opt_verbose_consumer
) {
2396 verbosity
= "--verbose";
2397 } else if (lttng_opt_quiet
) {
2398 verbosity
= "--quiet";
2403 switch (consumer_data
->type
) {
2404 case LTTNG_CONSUMER_KERNEL
:
2406 * Find out which consumerd to execute. We will first try the
2407 * 64-bit path, then the sessiond's installation directory, and
2408 * fallback on the 32-bit one,
2410 DBG3("Looking for a kernel consumer at these locations:");
2411 DBG3(" 1) %s", consumerd64_bin
);
2412 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2413 DBG3(" 3) %s", consumerd32_bin
);
2414 if (stat(consumerd64_bin
, &st
) == 0) {
2415 DBG3("Found location #1");
2416 consumer_to_use
= consumerd64_bin
;
2417 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2418 DBG3("Found location #2");
2419 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2420 } else if (stat(consumerd32_bin
, &st
) == 0) {
2421 DBG3("Found location #3");
2422 consumer_to_use
= consumerd32_bin
;
2424 DBG("Could not find any valid consumerd executable");
2428 DBG("Using kernel consumer at: %s", consumer_to_use
);
2429 ret
= execl(consumer_to_use
,
2430 "lttng-consumerd", verbosity
, "-k",
2431 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2432 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2433 "--group", tracing_group_name
,
2436 case LTTNG_CONSUMER64_UST
:
2438 char *tmpnew
= NULL
;
2440 if (consumerd64_libdir
[0] != '\0') {
2444 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2448 tmplen
= strlen("LD_LIBRARY_PATH=")
2449 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2450 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2455 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2456 strcat(tmpnew
, consumerd64_libdir
);
2457 if (tmp
[0] != '\0') {
2458 strcat(tmpnew
, ":");
2459 strcat(tmpnew
, tmp
);
2461 ret
= putenv(tmpnew
);
2468 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2469 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2470 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2471 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2472 "--group", tracing_group_name
,
2474 if (consumerd64_libdir
[0] != '\0') {
2479 case LTTNG_CONSUMER32_UST
:
2481 char *tmpnew
= NULL
;
2483 if (consumerd32_libdir
[0] != '\0') {
2487 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2491 tmplen
= strlen("LD_LIBRARY_PATH=")
2492 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2493 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2498 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2499 strcat(tmpnew
, consumerd32_libdir
);
2500 if (tmp
[0] != '\0') {
2501 strcat(tmpnew
, ":");
2502 strcat(tmpnew
, tmp
);
2504 ret
= putenv(tmpnew
);
2511 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2512 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2513 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2514 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2515 "--group", tracing_group_name
,
2517 if (consumerd32_libdir
[0] != '\0') {
2523 PERROR("unknown consumer type");
2527 PERROR("Consumer execl()");
2529 /* Reaching this point, we got a failure on our execl(). */
2531 } else if (pid
> 0) {
2534 PERROR("start consumer fork");
2542 * Spawn the consumerd daemon and session daemon thread.
2544 static int start_consumerd(struct consumer_data
*consumer_data
)
2549 * Set the listen() state on the socket since there is a possible race
2550 * between the exec() of the consumer daemon and this call if place in the
2551 * consumer thread. See bug #366 for more details.
2553 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2558 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2559 if (consumer_data
->pid
!= 0) {
2560 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2564 ret
= spawn_consumerd(consumer_data
);
2566 ERR("Spawning consumerd failed");
2567 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2571 /* Setting up the consumer_data pid */
2572 consumer_data
->pid
= ret
;
2573 DBG2("Consumer pid %d", consumer_data
->pid
);
2574 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2576 DBG2("Spawning consumer control thread");
2577 ret
= spawn_consumer_thread(consumer_data
);
2579 ERR("Fatal error spawning consumer control thread");
2587 /* Cleanup already created sockets on error. */
2588 if (consumer_data
->err_sock
>= 0) {
2591 err
= close(consumer_data
->err_sock
);
2593 PERROR("close consumer data error socket");
2600 * Setup necessary data for kernel tracer action.
2602 static int init_kernel_tracer(void)
2606 /* Modprobe lttng kernel modules */
2607 ret
= modprobe_lttng_control();
2612 /* Open debugfs lttng */
2613 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2614 if (kernel_tracer_fd
< 0) {
2615 DBG("Failed to open %s", module_proc_lttng
);
2620 /* Validate kernel version */
2621 ret
= kernel_validate_version(kernel_tracer_fd
);
2626 ret
= modprobe_lttng_data();
2631 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2635 modprobe_remove_lttng_control();
2636 ret
= close(kernel_tracer_fd
);
2640 kernel_tracer_fd
= -1;
2641 return LTTNG_ERR_KERN_VERSION
;
2644 ret
= close(kernel_tracer_fd
);
2650 modprobe_remove_lttng_control();
2653 WARN("No kernel tracer available");
2654 kernel_tracer_fd
= -1;
2656 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2658 return LTTNG_ERR_KERN_NA
;
2664 * Copy consumer output from the tracing session to the domain session. The
2665 * function also applies the right modification on a per domain basis for the
2666 * trace files destination directory.
2668 * Should *NOT* be called with RCU read-side lock held.
2670 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2673 const char *dir_name
;
2674 struct consumer_output
*consumer
;
2677 assert(session
->consumer
);
2680 case LTTNG_DOMAIN_KERNEL
:
2681 DBG3("Copying tracing session consumer output in kernel session");
2683 * XXX: We should audit the session creation and what this function
2684 * does "extra" in order to avoid a destroy since this function is used
2685 * in the domain session creation (kernel and ust) only. Same for UST
2688 if (session
->kernel_session
->consumer
) {
2689 consumer_destroy_output(session
->kernel_session
->consumer
);
2691 session
->kernel_session
->consumer
=
2692 consumer_copy_output(session
->consumer
);
2693 /* Ease our life a bit for the next part */
2694 consumer
= session
->kernel_session
->consumer
;
2695 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2697 case LTTNG_DOMAIN_JUL
:
2698 case LTTNG_DOMAIN_LOG4J
:
2699 case LTTNG_DOMAIN_PYTHON
:
2700 case LTTNG_DOMAIN_UST
:
2701 DBG3("Copying tracing session consumer output in UST session");
2702 if (session
->ust_session
->consumer
) {
2703 consumer_destroy_output(session
->ust_session
->consumer
);
2705 session
->ust_session
->consumer
=
2706 consumer_copy_output(session
->consumer
);
2707 /* Ease our life a bit for the next part */
2708 consumer
= session
->ust_session
->consumer
;
2709 dir_name
= DEFAULT_UST_TRACE_DIR
;
2712 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2716 /* Append correct directory to subdir */
2717 strncat(consumer
->subdir
, dir_name
,
2718 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2719 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2728 * Create an UST session and add it to the session ust list.
2730 * Should *NOT* be called with RCU read-side lock held.
2732 static int create_ust_session(struct ltt_session
*session
,
2733 struct lttng_domain
*domain
)
2736 struct ltt_ust_session
*lus
= NULL
;
2740 assert(session
->consumer
);
2742 switch (domain
->type
) {
2743 case LTTNG_DOMAIN_JUL
:
2744 case LTTNG_DOMAIN_LOG4J
:
2745 case LTTNG_DOMAIN_PYTHON
:
2746 case LTTNG_DOMAIN_UST
:
2749 ERR("Unknown UST domain on create session %d", domain
->type
);
2750 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2754 DBG("Creating UST session");
2756 lus
= trace_ust_create_session(session
->id
);
2758 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2762 lus
->uid
= session
->uid
;
2763 lus
->gid
= session
->gid
;
2764 lus
->output_traces
= session
->output_traces
;
2765 lus
->snapshot_mode
= session
->snapshot_mode
;
2766 lus
->live_timer_interval
= session
->live_timer
;
2767 session
->ust_session
= lus
;
2768 if (session
->shm_path
[0]) {
2769 strncpy(lus
->root_shm_path
, session
->shm_path
,
2770 sizeof(lus
->root_shm_path
));
2771 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2772 strncpy(lus
->shm_path
, session
->shm_path
,
2773 sizeof(lus
->shm_path
));
2774 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2775 strncat(lus
->shm_path
, "/ust",
2776 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2778 /* Copy session output to the newly created UST session */
2779 ret
= copy_session_consumer(domain
->type
, session
);
2780 if (ret
!= LTTNG_OK
) {
2788 session
->ust_session
= NULL
;
2793 * Create a kernel tracer session then create the default channel.
2795 static int create_kernel_session(struct ltt_session
*session
)
2799 DBG("Creating kernel session");
2801 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2803 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2807 /* Code flow safety */
2808 assert(session
->kernel_session
);
2810 /* Copy session output to the newly created Kernel session */
2811 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2812 if (ret
!= LTTNG_OK
) {
2816 /* Create directory(ies) on local filesystem. */
2817 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2818 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2819 ret
= run_as_mkdir_recursive(
2820 session
->kernel_session
->consumer
->dst
.trace_path
,
2821 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2823 if (ret
!= -EEXIST
) {
2824 ERR("Trace directory creation error");
2830 session
->kernel_session
->uid
= session
->uid
;
2831 session
->kernel_session
->gid
= session
->gid
;
2832 session
->kernel_session
->output_traces
= session
->output_traces
;
2833 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2838 trace_kernel_destroy_session(session
->kernel_session
);
2839 session
->kernel_session
= NULL
;
2844 * Count number of session permitted by uid/gid.
2846 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2849 struct ltt_session
*session
;
2851 DBG("Counting number of available session for UID %d GID %d",
2853 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2855 * Only list the sessions the user can control.
2857 if (!session_access_ok(session
, uid
, gid
)) {
2866 * Process the command requested by the lttng client within the command
2867 * context structure. This function make sure that the return structure (llm)
2868 * is set and ready for transmission before returning.
2870 * Return any error encountered or 0 for success.
2872 * "sock" is only used for special-case var. len data.
2874 * Should *NOT* be called with RCU read-side lock held.
2876 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2880 int need_tracing_session
= 1;
2883 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2887 switch (cmd_ctx
->lsm
->cmd_type
) {
2888 case LTTNG_CREATE_SESSION
:
2889 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2890 case LTTNG_CREATE_SESSION_LIVE
:
2891 case LTTNG_DESTROY_SESSION
:
2892 case LTTNG_LIST_SESSIONS
:
2893 case LTTNG_LIST_DOMAINS
:
2894 case LTTNG_START_TRACE
:
2895 case LTTNG_STOP_TRACE
:
2896 case LTTNG_DATA_PENDING
:
2897 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2898 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2899 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2900 case LTTNG_SNAPSHOT_RECORD
:
2901 case LTTNG_SAVE_SESSION
:
2902 case LTTNG_SET_SESSION_SHM_PATH
:
2909 if (opt_no_kernel
&& need_domain
2910 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2912 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2914 ret
= LTTNG_ERR_KERN_NA
;
2919 /* Deny register consumer if we already have a spawned consumer. */
2920 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2921 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2922 if (kconsumer_data
.pid
> 0) {
2923 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2924 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2927 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2931 * Check for command that don't needs to allocate a returned payload. We do
2932 * this here so we don't have to make the call for no payload at each
2935 switch(cmd_ctx
->lsm
->cmd_type
) {
2936 case LTTNG_LIST_SESSIONS
:
2937 case LTTNG_LIST_TRACEPOINTS
:
2938 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2939 case LTTNG_LIST_DOMAINS
:
2940 case LTTNG_LIST_CHANNELS
:
2941 case LTTNG_LIST_EVENTS
:
2942 case LTTNG_LIST_SYSCALLS
:
2945 /* Setup lttng message with no payload */
2946 ret
= setup_lttng_msg(cmd_ctx
, 0);
2948 /* This label does not try to unlock the session */
2949 goto init_setup_error
;
2953 /* Commands that DO NOT need a session. */
2954 switch (cmd_ctx
->lsm
->cmd_type
) {
2955 case LTTNG_CREATE_SESSION
:
2956 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2957 case LTTNG_CREATE_SESSION_LIVE
:
2958 case LTTNG_CALIBRATE
:
2959 case LTTNG_LIST_SESSIONS
:
2960 case LTTNG_LIST_TRACEPOINTS
:
2961 case LTTNG_LIST_SYSCALLS
:
2962 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2963 case LTTNG_SAVE_SESSION
:
2964 need_tracing_session
= 0;
2967 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2969 * We keep the session list lock across _all_ commands
2970 * for now, because the per-session lock does not
2971 * handle teardown properly.
2973 session_lock_list();
2974 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2975 if (cmd_ctx
->session
== NULL
) {
2976 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2979 /* Acquire lock for the session */
2980 session_lock(cmd_ctx
->session
);
2986 * Commands that need a valid session but should NOT create one if none
2987 * exists. Instead of creating one and destroying it when the command is
2988 * handled, process that right before so we save some round trip in useless
2991 switch (cmd_ctx
->lsm
->cmd_type
) {
2992 case LTTNG_DISABLE_CHANNEL
:
2993 case LTTNG_DISABLE_EVENT
:
2994 switch (cmd_ctx
->lsm
->domain
.type
) {
2995 case LTTNG_DOMAIN_KERNEL
:
2996 if (!cmd_ctx
->session
->kernel_session
) {
2997 ret
= LTTNG_ERR_NO_CHANNEL
;
3001 case LTTNG_DOMAIN_JUL
:
3002 case LTTNG_DOMAIN_LOG4J
:
3003 case LTTNG_DOMAIN_PYTHON
:
3004 case LTTNG_DOMAIN_UST
:
3005 if (!cmd_ctx
->session
->ust_session
) {
3006 ret
= LTTNG_ERR_NO_CHANNEL
;
3011 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3023 * Check domain type for specific "pre-action".
3025 switch (cmd_ctx
->lsm
->domain
.type
) {
3026 case LTTNG_DOMAIN_KERNEL
:
3028 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3032 /* Kernel tracer check */
3033 if (kernel_tracer_fd
== -1) {
3034 /* Basically, load kernel tracer modules */
3035 ret
= init_kernel_tracer();
3041 /* Consumer is in an ERROR state. Report back to client */
3042 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3043 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3047 /* Need a session for kernel command */
3048 if (need_tracing_session
) {
3049 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3050 ret
= create_kernel_session(cmd_ctx
->session
);
3052 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3057 /* Start the kernel consumer daemon */
3058 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3059 if (kconsumer_data
.pid
== 0 &&
3060 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3061 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3062 ret
= start_consumerd(&kconsumer_data
);
3064 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3067 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3069 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3073 * The consumer was just spawned so we need to add the socket to
3074 * the consumer output of the session if exist.
3076 ret
= consumer_create_socket(&kconsumer_data
,
3077 cmd_ctx
->session
->kernel_session
->consumer
);
3084 case LTTNG_DOMAIN_JUL
:
3085 case LTTNG_DOMAIN_LOG4J
:
3086 case LTTNG_DOMAIN_PYTHON
:
3087 case LTTNG_DOMAIN_UST
:
3089 if (!ust_app_supported()) {
3090 ret
= LTTNG_ERR_NO_UST
;
3093 /* Consumer is in an ERROR state. Report back to client */
3094 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3095 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3099 if (need_tracing_session
) {
3100 /* Create UST session if none exist. */
3101 if (cmd_ctx
->session
->ust_session
== NULL
) {
3102 ret
= create_ust_session(cmd_ctx
->session
,
3103 &cmd_ctx
->lsm
->domain
);
3104 if (ret
!= LTTNG_OK
) {
3109 /* Start the UST consumer daemons */
3111 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3112 if (consumerd64_bin
[0] != '\0' &&
3113 ustconsumer64_data
.pid
== 0 &&
3114 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3115 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3116 ret
= start_consumerd(&ustconsumer64_data
);
3118 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3119 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3123 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3124 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3126 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3130 * Setup socket for consumer 64 bit. No need for atomic access
3131 * since it was set above and can ONLY be set in this thread.
3133 ret
= consumer_create_socket(&ustconsumer64_data
,
3134 cmd_ctx
->session
->ust_session
->consumer
);
3140 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3141 if (consumerd32_bin
[0] != '\0' &&
3142 ustconsumer32_data
.pid
== 0 &&
3143 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3144 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3145 ret
= start_consumerd(&ustconsumer32_data
);
3147 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3148 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3152 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3153 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3155 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3159 * Setup socket for consumer 64 bit. No need for atomic access
3160 * since it was set above and can ONLY be set in this thread.
3162 ret
= consumer_create_socket(&ustconsumer32_data
,
3163 cmd_ctx
->session
->ust_session
->consumer
);
3175 /* Validate consumer daemon state when start/stop trace command */
3176 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3177 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3178 switch (cmd_ctx
->lsm
->domain
.type
) {
3179 case LTTNG_DOMAIN_JUL
:
3180 case LTTNG_DOMAIN_LOG4J
:
3181 case LTTNG_DOMAIN_PYTHON
:
3182 case LTTNG_DOMAIN_UST
:
3183 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3184 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3188 case LTTNG_DOMAIN_KERNEL
:
3189 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3190 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3198 * Check that the UID or GID match that of the tracing session.
3199 * The root user can interact with all sessions.
3201 if (need_tracing_session
) {
3202 if (!session_access_ok(cmd_ctx
->session
,
3203 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3204 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3205 ret
= LTTNG_ERR_EPERM
;
3211 * Send relayd information to consumer as soon as we have a domain and a
3214 if (cmd_ctx
->session
&& need_domain
) {
3216 * Setup relayd if not done yet. If the relayd information was already
3217 * sent to the consumer, this call will gracefully return.
3219 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3220 if (ret
!= LTTNG_OK
) {
3225 /* Process by command type */
3226 switch (cmd_ctx
->lsm
->cmd_type
) {
3227 case LTTNG_ADD_CONTEXT
:
3229 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3230 cmd_ctx
->lsm
->u
.context
.channel_name
,
3231 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3234 case LTTNG_DISABLE_CHANNEL
:
3236 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3237 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3240 case LTTNG_DISABLE_EVENT
:
3242 /* FIXME: passing packed structure to non-packed pointer */
3243 /* TODO: handle filter */
3244 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3245 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3246 &cmd_ctx
->lsm
->u
.disable
.event
);
3249 case LTTNG_ENABLE_CHANNEL
:
3251 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3252 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3255 case LTTNG_ENABLE_EVENT
:
3257 struct lttng_event_exclusion
*exclusion
= NULL
;
3258 struct lttng_filter_bytecode
*bytecode
= NULL
;
3259 char *filter_expression
= NULL
;
3261 /* Handle exclusion events and receive it from the client. */
3262 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3263 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3265 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3266 (count
* LTTNG_SYMBOL_NAME_LEN
));
3268 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3272 DBG("Receiving var len exclusion event list from client ...");
3273 exclusion
->count
= count
;
3274 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3275 count
* LTTNG_SYMBOL_NAME_LEN
);
3277 DBG("Nothing recv() from client var len data... continuing");
3280 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3285 /* Get filter expression from client. */
3286 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3287 size_t expression_len
=
3288 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3290 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3291 ret
= LTTNG_ERR_FILTER_INVAL
;
3296 filter_expression
= zmalloc(expression_len
);
3297 if (!filter_expression
) {
3299 ret
= LTTNG_ERR_FILTER_NOMEM
;
3303 /* Receive var. len. data */
3304 DBG("Receiving var len filter's expression from client ...");
3305 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3308 DBG("Nothing recv() from client car len data... continuing");
3310 free(filter_expression
);
3312 ret
= LTTNG_ERR_FILTER_INVAL
;
3317 /* Handle filter and get bytecode from client. */
3318 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3319 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3321 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3322 ret
= LTTNG_ERR_FILTER_INVAL
;
3323 free(filter_expression
);
3328 bytecode
= zmalloc(bytecode_len
);
3330 free(filter_expression
);
3332 ret
= LTTNG_ERR_FILTER_NOMEM
;
3336 /* Receive var. len. data */
3337 DBG("Receiving var len filter's bytecode from client ...");
3338 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3340 DBG("Nothing recv() from client car len data... continuing");
3342 free(filter_expression
);
3345 ret
= LTTNG_ERR_FILTER_INVAL
;
3349 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3350 free(filter_expression
);
3353 ret
= LTTNG_ERR_FILTER_INVAL
;
3358 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3359 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3360 &cmd_ctx
->lsm
->u
.enable
.event
,
3361 filter_expression
, bytecode
, exclusion
,
3362 kernel_poll_pipe
[1]);
3365 case LTTNG_LIST_TRACEPOINTS
:
3367 struct lttng_event
*events
;
3370 session_lock_list();
3371 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3372 session_unlock_list();
3373 if (nb_events
< 0) {
3374 /* Return value is a negative lttng_error_code. */
3380 * Setup lttng message with payload size set to the event list size in
3381 * bytes and then copy list into the llm payload.
3383 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3389 /* Copy event list into message payload */
3390 memcpy(cmd_ctx
->llm
->payload
, events
,
3391 sizeof(struct lttng_event
) * nb_events
);
3398 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3400 struct lttng_event_field
*fields
;
3403 session_lock_list();
3404 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3406 session_unlock_list();
3407 if (nb_fields
< 0) {
3408 /* Return value is a negative lttng_error_code. */
3414 * Setup lttng message with payload size set to the event list size in
3415 * bytes and then copy list into the llm payload.
3417 ret
= setup_lttng_msg(cmd_ctx
,
3418 sizeof(struct lttng_event_field
) * nb_fields
);
3424 /* Copy event list into message payload */
3425 memcpy(cmd_ctx
->llm
->payload
, fields
,
3426 sizeof(struct lttng_event_field
) * nb_fields
);
3433 case LTTNG_LIST_SYSCALLS
:
3435 struct lttng_event
*events
;
3438 nb_events
= cmd_list_syscalls(&events
);
3439 if (nb_events
< 0) {
3440 /* Return value is a negative lttng_error_code. */
3446 * Setup lttng message with payload size set to the event list size in
3447 * bytes and then copy list into the llm payload.
3449 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3455 /* Copy event list into message payload */
3456 memcpy(cmd_ctx
->llm
->payload
, events
,
3457 sizeof(struct lttng_event
) * nb_events
);
3464 case LTTNG_SET_CONSUMER_URI
:
3467 struct lttng_uri
*uris
;
3469 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3470 len
= nb_uri
* sizeof(struct lttng_uri
);
3473 ret
= LTTNG_ERR_INVALID
;
3477 uris
= zmalloc(len
);
3479 ret
= LTTNG_ERR_FATAL
;
3483 /* Receive variable len data */
3484 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3485 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3487 DBG("No URIs received from client... continuing");
3489 ret
= LTTNG_ERR_SESSION_FAIL
;
3494 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3496 if (ret
!= LTTNG_OK
) {
3503 case LTTNG_START_TRACE
:
3505 ret
= cmd_start_trace(cmd_ctx
->session
);
3508 case LTTNG_STOP_TRACE
:
3510 ret
= cmd_stop_trace(cmd_ctx
->session
);
3513 case LTTNG_CREATE_SESSION
:
3516 struct lttng_uri
*uris
= NULL
;
3518 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3519 len
= nb_uri
* sizeof(struct lttng_uri
);
3522 uris
= zmalloc(len
);
3524 ret
= LTTNG_ERR_FATAL
;
3528 /* Receive variable len data */
3529 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3530 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3532 DBG("No URIs received from client... continuing");
3534 ret
= LTTNG_ERR_SESSION_FAIL
;
3539 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3540 DBG("Creating session with ONE network URI is a bad call");
3541 ret
= LTTNG_ERR_SESSION_FAIL
;
3547 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3548 &cmd_ctx
->creds
, 0);
3554 case LTTNG_DESTROY_SESSION
:
3556 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3558 /* Set session to NULL so we do not unlock it after free. */
3559 cmd_ctx
->session
= NULL
;
3562 case LTTNG_LIST_DOMAINS
:
3565 struct lttng_domain
*domains
= NULL
;
3567 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3569 /* Return value is a negative lttng_error_code. */
3574 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3580 /* Copy event list into message payload */
3581 memcpy(cmd_ctx
->llm
->payload
, domains
,
3582 nb_dom
* sizeof(struct lttng_domain
));
3589 case LTTNG_LIST_CHANNELS
:
3592 struct lttng_channel
*channels
= NULL
;
3594 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3595 cmd_ctx
->session
, &channels
);
3597 /* Return value is a negative lttng_error_code. */
3602 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3608 /* Copy event list into message payload */
3609 memcpy(cmd_ctx
->llm
->payload
, channels
,
3610 nb_chan
* sizeof(struct lttng_channel
));
3617 case LTTNG_LIST_EVENTS
:
3620 struct lttng_event
*events
= NULL
;
3622 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3623 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3625 /* Return value is a negative lttng_error_code. */
3630 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3636 /* Copy event list into message payload */
3637 memcpy(cmd_ctx
->llm
->payload
, events
,
3638 nb_event
* sizeof(struct lttng_event
));
3645 case LTTNG_LIST_SESSIONS
:
3647 unsigned int nr_sessions
;
3649 session_lock_list();
3650 nr_sessions
= lttng_sessions_count(
3651 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3652 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3654 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3656 session_unlock_list();
3660 /* Filled the session array */
3661 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3662 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3663 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3665 session_unlock_list();
3670 case LTTNG_CALIBRATE
:
3672 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3673 &cmd_ctx
->lsm
->u
.calibrate
);
3676 case LTTNG_REGISTER_CONSUMER
:
3678 struct consumer_data
*cdata
;
3680 switch (cmd_ctx
->lsm
->domain
.type
) {
3681 case LTTNG_DOMAIN_KERNEL
:
3682 cdata
= &kconsumer_data
;
3685 ret
= LTTNG_ERR_UND
;
3689 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3690 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3693 case LTTNG_DATA_PENDING
:
3695 ret
= cmd_data_pending(cmd_ctx
->session
);
3698 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3700 struct lttcomm_lttng_output_id reply
;
3702 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3703 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3704 if (ret
!= LTTNG_OK
) {
3708 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3713 /* Copy output list into message payload */
3714 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3718 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3720 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3721 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3724 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3727 struct lttng_snapshot_output
*outputs
= NULL
;
3729 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3730 if (nb_output
< 0) {
3735 ret
= setup_lttng_msg(cmd_ctx
,
3736 nb_output
* sizeof(struct lttng_snapshot_output
));
3743 /* Copy output list into message payload */
3744 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3745 nb_output
* sizeof(struct lttng_snapshot_output
));
3752 case LTTNG_SNAPSHOT_RECORD
:
3754 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3755 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3756 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3759 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3762 struct lttng_uri
*uris
= NULL
;
3764 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3765 len
= nb_uri
* sizeof(struct lttng_uri
);
3768 uris
= zmalloc(len
);
3770 ret
= LTTNG_ERR_FATAL
;
3774 /* Receive variable len data */
3775 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3776 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3778 DBG("No URIs received from client... continuing");
3780 ret
= LTTNG_ERR_SESSION_FAIL
;
3785 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3786 DBG("Creating session with ONE network URI is a bad call");
3787 ret
= LTTNG_ERR_SESSION_FAIL
;
3793 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3794 nb_uri
, &cmd_ctx
->creds
);
3798 case LTTNG_CREATE_SESSION_LIVE
:
3801 struct lttng_uri
*uris
= NULL
;
3803 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3804 len
= nb_uri
* sizeof(struct lttng_uri
);
3807 uris
= zmalloc(len
);
3809 ret
= LTTNG_ERR_FATAL
;
3813 /* Receive variable len data */
3814 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3815 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3817 DBG("No URIs received from client... continuing");
3819 ret
= LTTNG_ERR_SESSION_FAIL
;
3824 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3825 DBG("Creating session with ONE network URI is a bad call");
3826 ret
= LTTNG_ERR_SESSION_FAIL
;
3832 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3833 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3837 case LTTNG_SAVE_SESSION
:
3839 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
3843 case LTTNG_SET_SESSION_SHM_PATH
:
3845 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
3846 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
3850 ret
= LTTNG_ERR_UND
;
3855 if (cmd_ctx
->llm
== NULL
) {
3856 DBG("Missing llm structure. Allocating one.");
3857 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3861 /* Set return code */
3862 cmd_ctx
->llm
->ret_code
= ret
;
3864 if (cmd_ctx
->session
) {
3865 session_unlock(cmd_ctx
->session
);
3867 if (need_tracing_session
) {
3868 session_unlock_list();
3875 * Thread managing health check socket.
3877 static void *thread_manage_health(void *data
)
3879 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3880 uint32_t revents
, nb_fd
;
3881 struct lttng_poll_event events
;
3882 struct health_comm_msg msg
;
3883 struct health_comm_reply reply
;
3885 DBG("[thread] Manage health check started");
3887 rcu_register_thread();
3889 /* We might hit an error path before this is created. */
3890 lttng_poll_init(&events
);
3892 /* Create unix socket */
3893 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3895 ERR("Unable to create health check Unix socket");
3901 /* lttng health client socket path permissions */
3902 ret
= chown(health_unix_sock_path
, 0,
3903 utils_get_group_id(tracing_group_name
));
3905 ERR("Unable to set group on %s", health_unix_sock_path
);
3911 ret
= chmod(health_unix_sock_path
,
3912 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3914 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3922 * Set the CLOEXEC flag. Return code is useless because either way, the
3925 (void) utils_set_fd_cloexec(sock
);
3927 ret
= lttcomm_listen_unix_sock(sock
);
3933 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3934 * more will be added to this poll set.
3936 ret
= sessiond_set_thread_pollset(&events
, 2);
3941 /* Add the application registration socket */
3942 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3947 sessiond_notify_ready();
3950 DBG("Health check ready");
3952 /* Inifinite blocking call, waiting for transmission */
3954 ret
= lttng_poll_wait(&events
, -1);
3957 * Restart interrupted system call.
3959 if (errno
== EINTR
) {
3967 for (i
= 0; i
< nb_fd
; i
++) {
3968 /* Fetch once the poll data */
3969 revents
= LTTNG_POLL_GETEV(&events
, i
);
3970 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3973 /* No activity for this FD (poll implementation). */
3977 /* Thread quit pipe has been closed. Killing thread. */
3978 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3984 /* Event on the registration socket */
3985 if (pollfd
== sock
) {
3986 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3987 ERR("Health socket poll error");
3993 new_sock
= lttcomm_accept_unix_sock(sock
);
3999 * Set the CLOEXEC flag. Return code is useless because either way, the
4002 (void) utils_set_fd_cloexec(new_sock
);
4004 DBG("Receiving data from client for health...");
4005 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4007 DBG("Nothing recv() from client... continuing");
4008 ret
= close(new_sock
);
4016 rcu_thread_online();
4018 memset(&reply
, 0, sizeof(reply
));
4019 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4021 * health_check_state returns 0 if health is
4024 if (!health_check_state(health_sessiond
, i
)) {
4025 reply
.ret_code
|= 1ULL << i
;
4029 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4031 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4033 ERR("Failed to send health data back to client");
4036 /* End of transmission */
4037 ret
= close(new_sock
);
4047 ERR("Health error occurred in %s", __func__
);
4049 DBG("Health check thread dying");
4050 unlink(health_unix_sock_path
);
4058 lttng_poll_clean(&events
);
4060 rcu_unregister_thread();
4065 * This thread manage all clients request using the unix client socket for
4068 static void *thread_manage_clients(void *data
)
4070 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4072 uint32_t revents
, nb_fd
;
4073 struct command_ctx
*cmd_ctx
= NULL
;
4074 struct lttng_poll_event events
;
4076 DBG("[thread] Manage client started");
4078 rcu_register_thread();
4080 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4082 health_code_update();
4084 ret
= lttcomm_listen_unix_sock(client_sock
);
4090 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4091 * more will be added to this poll set.
4093 ret
= sessiond_set_thread_pollset(&events
, 2);
4095 goto error_create_poll
;
4098 /* Add the application registration socket */
4099 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4104 sessiond_notify_ready();
4105 ret
= sem_post(&load_info
->message_thread_ready
);
4107 PERROR("sem_post message_thread_ready");
4111 /* This testpoint is after we signal readiness to the parent. */
4112 if (testpoint(sessiond_thread_manage_clients
)) {
4116 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4120 health_code_update();
4123 DBG("Accepting client command ...");
4125 /* Inifinite blocking call, waiting for transmission */
4127 health_poll_entry();
4128 ret
= lttng_poll_wait(&events
, -1);
4132 * Restart interrupted system call.
4134 if (errno
== EINTR
) {
4142 for (i
= 0; i
< nb_fd
; i
++) {
4143 /* Fetch once the poll data */
4144 revents
= LTTNG_POLL_GETEV(&events
, i
);
4145 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4147 health_code_update();
4150 /* No activity for this FD (poll implementation). */
4154 /* Thread quit pipe has been closed. Killing thread. */
4155 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4161 /* Event on the registration socket */
4162 if (pollfd
== client_sock
) {
4163 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4164 ERR("Client socket poll error");
4170 DBG("Wait for client response");
4172 health_code_update();
4174 sock
= lttcomm_accept_unix_sock(client_sock
);
4180 * Set the CLOEXEC flag. Return code is useless because either way, the
4183 (void) utils_set_fd_cloexec(sock
);
4185 /* Set socket option for credentials retrieval */
4186 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4191 /* Allocate context command to process the client request */
4192 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4193 if (cmd_ctx
== NULL
) {
4194 PERROR("zmalloc cmd_ctx");
4198 /* Allocate data buffer for reception */
4199 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4200 if (cmd_ctx
->lsm
== NULL
) {
4201 PERROR("zmalloc cmd_ctx->lsm");
4205 cmd_ctx
->llm
= NULL
;
4206 cmd_ctx
->session
= NULL
;
4208 health_code_update();
4211 * Data is received from the lttng client. The struct
4212 * lttcomm_session_msg (lsm) contains the command and data request of
4215 DBG("Receiving data from client ...");
4216 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4217 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4219 DBG("Nothing recv() from client... continuing");
4225 clean_command_ctx(&cmd_ctx
);
4229 health_code_update();
4231 // TODO: Validate cmd_ctx including sanity check for
4232 // security purpose.
4234 rcu_thread_online();
4236 * This function dispatch the work to the kernel or userspace tracer
4237 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4238 * informations for the client. The command context struct contains
4239 * everything this function may needs.
4241 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4242 rcu_thread_offline();
4250 * TODO: Inform client somehow of the fatal error. At
4251 * this point, ret < 0 means that a zmalloc failed
4252 * (ENOMEM). Error detected but still accept
4253 * command, unless a socket error has been
4256 clean_command_ctx(&cmd_ctx
);
4260 health_code_update();
4262 DBG("Sending response (size: %d, retcode: %s)",
4263 cmd_ctx
->lttng_msg_size
,
4264 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4265 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4267 ERR("Failed to send data back to client");
4270 /* End of transmission */
4277 clean_command_ctx(&cmd_ctx
);
4279 health_code_update();
4291 lttng_poll_clean(&events
);
4292 clean_command_ctx(&cmd_ctx
);
4296 unlink(client_unix_sock_path
);
4297 if (client_sock
>= 0) {
4298 ret
= close(client_sock
);
4306 ERR("Health error occurred in %s", __func__
);
4309 health_unregister(health_sessiond
);
4311 DBG("Client thread dying");
4313 rcu_unregister_thread();
4316 * Since we are creating the consumer threads, we own them, so we need
4317 * to join them before our thread exits.
4319 ret
= join_consumer_thread(&kconsumer_data
);
4322 PERROR("join_consumer");
4325 ret
= join_consumer_thread(&ustconsumer32_data
);
4328 PERROR("join_consumer ust32");
4331 ret
= join_consumer_thread(&ustconsumer64_data
);
4334 PERROR("join_consumer ust64");
4341 * usage function on stderr
4343 static void usage(void)
4345 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4346 fprintf(stderr
, " -h, --help Display this usage.\n");
4347 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4348 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4349 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4350 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4351 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4352 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4353 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4354 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4355 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4356 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4357 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4358 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4359 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4360 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4361 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4362 fprintf(stderr
, " -V, --version Show version number.\n");
4363 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4364 fprintf(stderr
, " -q, --quiet No output at all.\n");
4365 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4366 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4367 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4368 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4369 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4370 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4371 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4372 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4373 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4377 * Take an option from the getopt output and set it in the right variable to be
4380 * Return 0 on success else a negative value.
4382 static int set_option(int opt
, const char *arg
, const char *optname
)
4386 if (arg
&& arg
[0] == '\0') {
4388 * This only happens if the value is read from daemon config
4389 * file. This means the option requires an argument and the
4390 * configuration file contains a line such as:
4399 fprintf(stderr
, "option %s", optname
);
4401 fprintf(stderr
, " with arg %s\n", arg
);
4405 if (lttng_is_setuid_setgid()) {
4406 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4407 "-c, --client-sock");
4409 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4413 if (lttng_is_setuid_setgid()) {
4414 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4417 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4427 if (lttng_is_setuid_setgid()) {
4428 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4432 * If the override option is set, the pointer points to a
4433 * *non* const thus freeing it even though the variable type is
4436 if (tracing_group_name_override
) {
4437 free((void *) tracing_group_name
);
4439 tracing_group_name
= strdup(arg
);
4440 if (!tracing_group_name
) {
4444 tracing_group_name_override
= 1;
4451 fprintf(stdout
, "%s\n", VERSION
);
4457 if (lttng_is_setuid_setgid()) {
4458 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4459 "--kconsumerd-err-sock");
4461 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4465 if (lttng_is_setuid_setgid()) {
4466 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4467 "--kconsumerd-cmd-sock");
4469 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4473 if (lttng_is_setuid_setgid()) {
4474 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4475 "--ustconsumerd64-err-sock");
4477 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4481 if (lttng_is_setuid_setgid()) {
4482 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4483 "--ustconsumerd64-cmd-sock");
4485 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4489 if (lttng_is_setuid_setgid()) {
4490 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4491 "--ustconsumerd32-err-sock");
4493 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4497 if (lttng_is_setuid_setgid()) {
4498 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4499 "--ustconsumerd32-cmd-sock");
4501 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4508 lttng_opt_quiet
= 1;
4511 /* Verbose level can increase using multiple -v */
4513 /* Value obtained from config file */
4514 lttng_opt_verbose
= config_parse_value(arg
);
4516 /* -v used on command line */
4517 lttng_opt_verbose
++;
4519 /* Clamp value to [0, 3] */
4520 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4521 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4525 opt_verbose_consumer
= config_parse_value(arg
);
4527 opt_verbose_consumer
+= 1;
4531 if (lttng_is_setuid_setgid()) {
4532 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4533 "--consumerd32-path");
4535 if (consumerd32_bin_override
) {
4536 free((void *) consumerd32_bin
);
4538 consumerd32_bin
= strdup(arg
);
4539 if (!consumerd32_bin
) {
4543 consumerd32_bin_override
= 1;
4547 if (lttng_is_setuid_setgid()) {
4548 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4549 "--consumerd32-libdir");
4551 if (consumerd32_libdir_override
) {
4552 free((void *) consumerd32_libdir
);
4554 consumerd32_libdir
= strdup(arg
);
4555 if (!consumerd32_libdir
) {
4559 consumerd32_libdir_override
= 1;
4563 if (lttng_is_setuid_setgid()) {
4564 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4565 "--consumerd64-path");
4567 if (consumerd64_bin_override
) {
4568 free((void *) consumerd64_bin
);
4570 consumerd64_bin
= strdup(arg
);
4571 if (!consumerd64_bin
) {
4575 consumerd64_bin_override
= 1;
4579 if (lttng_is_setuid_setgid()) {
4580 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4581 "--consumerd64-libdir");
4583 if (consumerd64_libdir_override
) {
4584 free((void *) consumerd64_libdir
);
4586 consumerd64_libdir
= strdup(arg
);
4587 if (!consumerd64_libdir
) {
4591 consumerd64_libdir_override
= 1;
4595 if (lttng_is_setuid_setgid()) {
4596 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4600 opt_pidfile
= strdup(arg
);
4607 case 'J': /* Agent TCP port. */
4609 if (lttng_is_setuid_setgid()) {
4610 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4611 "--agent-tcp-port");
4620 v
= strtoul(arg
, NULL
, 0);
4621 if (errno
!= 0 || !isdigit(arg
[0])) {
4622 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4625 if (v
== 0 || v
>= 65535) {
4626 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4629 agent_tcp_port
= (uint32_t) v
;
4630 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4635 if (lttng_is_setuid_setgid()) {
4636 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4639 free(opt_load_session_path
);
4640 opt_load_session_path
= strdup(arg
);
4641 if (!opt_load_session_path
) {
4647 case 'P': /* probe modules list */
4648 if (lttng_is_setuid_setgid()) {
4649 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4652 free(kmod_probes_list
);
4653 kmod_probes_list
= strdup(arg
);
4654 if (!kmod_probes_list
) {
4661 if (lttng_is_setuid_setgid()) {
4662 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4663 "--extra-kmod-probes");
4665 free(kmod_extra_probes_list
);
4666 kmod_extra_probes_list
= strdup(arg
);
4667 if (!kmod_extra_probes_list
) {
4674 /* This is handled in set_options() thus silent break. */
4677 /* Unknown option or other error.
4678 * Error is printed by getopt, just return */
4683 if (ret
== -EINVAL
) {
4684 const char *opt_name
= "unknown";
4687 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4689 if (opt
== long_options
[i
].val
) {
4690 opt_name
= long_options
[i
].name
;
4695 WARN("Invalid argument provided for option \"%s\", using default value.",
4703 * config_entry_handler_cb used to handle options read from a config file.
4704 * See config_entry_handler_cb comment in common/config/config.h for the
4705 * return value conventions.
4707 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4711 if (!entry
|| !entry
->name
|| !entry
->value
) {
4716 /* Check if the option is to be ignored */
4717 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4718 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4723 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4726 /* Ignore if not fully matched. */
4727 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4732 * If the option takes no argument on the command line, we have to
4733 * check if the value is "true". We support non-zero numeric values,
4736 if (!long_options
[i
].has_arg
) {
4737 ret
= config_parse_value(entry
->value
);
4740 WARN("Invalid configuration value \"%s\" for option %s",
4741 entry
->value
, entry
->name
);
4743 /* False, skip boolean config option. */
4748 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4752 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4759 * daemon configuration loading and argument parsing
4761 static int set_options(int argc
, char **argv
)
4763 int ret
= 0, c
= 0, option_index
= 0;
4764 int orig_optopt
= optopt
, orig_optind
= optind
;
4766 const char *config_path
= NULL
;
4768 optstring
= utils_generate_optstring(long_options
,
4769 sizeof(long_options
) / sizeof(struct option
));
4775 /* Check for the --config option */
4776 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4777 &option_index
)) != -1) {
4781 } else if (c
!= 'f') {
4782 /* if not equal to --config option. */
4786 if (lttng_is_setuid_setgid()) {
4787 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4790 config_path
= utils_expand_path(optarg
);
4792 ERR("Failed to resolve path: %s", optarg
);
4797 ret
= config_get_section_entries(config_path
, config_section_name
,
4798 config_entry_handler
, NULL
);
4801 ERR("Invalid configuration option at line %i", ret
);
4807 /* Reset getopt's global state */
4808 optopt
= orig_optopt
;
4809 optind
= orig_optind
;
4811 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
4816 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
4828 * Creates the two needed socket by the daemon.
4829 * apps_sock - The communication socket for all UST apps.
4830 * client_sock - The communication of the cli tool (lttng).
4832 static int init_daemon_socket(void)
4837 old_umask
= umask(0);
4839 /* Create client tool unix socket */
4840 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4841 if (client_sock
< 0) {
4842 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4847 /* Set the cloexec flag */
4848 ret
= utils_set_fd_cloexec(client_sock
);
4850 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4851 "Continuing but note that the consumer daemon will have a "
4852 "reference to this socket on exec()", client_sock
);
4855 /* File permission MUST be 660 */
4856 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4858 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4863 /* Create the application unix socket */
4864 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4865 if (apps_sock
< 0) {
4866 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4871 /* Set the cloexec flag */
4872 ret
= utils_set_fd_cloexec(apps_sock
);
4874 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4875 "Continuing but note that the consumer daemon will have a "
4876 "reference to this socket on exec()", apps_sock
);
4879 /* File permission MUST be 666 */
4880 ret
= chmod(apps_unix_sock_path
,
4881 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4883 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4888 DBG3("Session daemon client socket %d and application socket %d created",
4889 client_sock
, apps_sock
);
4897 * Check if the global socket is available, and if a daemon is answering at the
4898 * other side. If yes, error is returned.
4900 static int check_existing_daemon(void)
4902 /* Is there anybody out there ? */
4903 if (lttng_session_daemon_alive()) {
4911 * Set the tracing group gid onto the client socket.
4913 * Race window between mkdir and chown is OK because we are going from more
4914 * permissive (root.root) to less permissive (root.tracing).
4916 static int set_permissions(char *rundir
)
4921 gid
= utils_get_group_id(tracing_group_name
);
4923 /* Set lttng run dir */
4924 ret
= chown(rundir
, 0, gid
);
4926 ERR("Unable to set group on %s", rundir
);
4931 * Ensure all applications and tracing group can search the run
4932 * dir. Allow everyone to read the directory, since it does not
4933 * buy us anything to hide its content.
4935 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4937 ERR("Unable to set permissions on %s", rundir
);
4941 /* lttng client socket path */
4942 ret
= chown(client_unix_sock_path
, 0, gid
);
4944 ERR("Unable to set group on %s", client_unix_sock_path
);
4948 /* kconsumer error socket path */
4949 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4951 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4955 /* 64-bit ustconsumer error socket path */
4956 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4958 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4962 /* 32-bit ustconsumer compat32 error socket path */
4963 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4965 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4969 DBG("All permissions are set");
4975 * Create the lttng run directory needed for all global sockets and pipe.
4977 static int create_lttng_rundir(const char *rundir
)
4981 DBG3("Creating LTTng run directory: %s", rundir
);
4983 ret
= mkdir(rundir
, S_IRWXU
);
4985 if (errno
!= EEXIST
) {
4986 ERR("Unable to create %s", rundir
);
4998 * Setup sockets and directory needed by the kconsumerd communication with the
5001 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5005 char path
[PATH_MAX
];
5007 switch (consumer_data
->type
) {
5008 case LTTNG_CONSUMER_KERNEL
:
5009 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5011 case LTTNG_CONSUMER64_UST
:
5012 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5014 case LTTNG_CONSUMER32_UST
:
5015 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5018 ERR("Consumer type unknown");
5023 DBG2("Creating consumer directory: %s", path
);
5025 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5027 if (errno
!= EEXIST
) {
5029 ERR("Failed to create %s", path
);
5035 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5037 ERR("Unable to set group on %s", path
);
5043 /* Create the kconsumerd error unix socket */
5044 consumer_data
->err_sock
=
5045 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5046 if (consumer_data
->err_sock
< 0) {
5047 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5053 * Set the CLOEXEC flag. Return code is useless because either way, the
5056 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5058 PERROR("utils_set_fd_cloexec");
5059 /* continue anyway */
5062 /* File permission MUST be 660 */
5063 ret
= chmod(consumer_data
->err_unix_sock_path
,
5064 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5066 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5076 * Signal handler for the daemon
5078 * Simply stop all worker threads, leaving main() return gracefully after
5079 * joining all threads and calling cleanup().
5081 static void sighandler(int sig
)
5085 DBG("SIGPIPE caught");
5088 DBG("SIGINT caught");
5092 DBG("SIGTERM caught");
5096 CMM_STORE_SHARED(recv_child_signal
, 1);
5104 * Setup signal handler for :
5105 * SIGINT, SIGTERM, SIGPIPE
5107 static int set_signal_handler(void)
5110 struct sigaction sa
;
5113 if ((ret
= sigemptyset(&sigset
)) < 0) {
5114 PERROR("sigemptyset");
5118 sa
.sa_handler
= sighandler
;
5119 sa
.sa_mask
= sigset
;
5121 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5122 PERROR("sigaction");
5126 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5127 PERROR("sigaction");
5131 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5132 PERROR("sigaction");
5136 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5137 PERROR("sigaction");
5141 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5147 * Set open files limit to unlimited. This daemon can open a large number of
5148 * file descriptors in order to consumer multiple kernel traces.
5150 static void set_ulimit(void)
5155 /* The kernel does not allowed an infinite limit for open files */
5156 lim
.rlim_cur
= 65535;
5157 lim
.rlim_max
= 65535;
5159 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5161 PERROR("failed to set open files limit");
5166 * Write pidfile using the rundir and opt_pidfile.
5168 static int write_pidfile(void)
5171 char pidfile_path
[PATH_MAX
];
5176 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
5178 /* Build pidfile path from rundir and opt_pidfile. */
5179 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5180 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5182 PERROR("snprintf pidfile path");
5188 * Create pid file in rundir.
5190 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5196 * Create lockfile using the rundir and return its fd.
5198 static int create_lockfile(void)
5201 char lockfile_path
[PATH_MAX
];
5203 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5208 ret
= utils_create_lock_file(lockfile_path
);
5214 * Write agent TCP port using the rundir.
5216 static int write_agent_port(void)
5219 char path
[PATH_MAX
];
5223 ret
= snprintf(path
, sizeof(path
), "%s/"
5224 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5226 PERROR("snprintf agent port path");
5231 * Create TCP agent port file in rundir.
5233 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5242 int main(int argc
, char **argv
)
5244 int ret
= 0, retval
= 0;
5246 const char *home_path
, *env_app_timeout
;
5248 init_kernel_workarounds();
5250 rcu_register_thread();
5252 if (set_signal_handler()) {
5254 goto exit_set_signal_handler
;
5257 setup_consumerd_path();
5259 page_size
= sysconf(_SC_PAGESIZE
);
5260 if (page_size
< 0) {
5261 PERROR("sysconf _SC_PAGESIZE");
5262 page_size
= LONG_MAX
;
5263 WARN("Fallback page size to %ld", page_size
);
5267 * Parse arguments and load the daemon configuration file.
5269 * We have an exit_options exit path to free memory reserved by
5270 * set_options. This is needed because the rest of sessiond_cleanup()
5271 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5272 * depends on set_options.
5275 if (set_options(argc
, argv
)) {
5281 if (opt_daemon
|| opt_background
) {
5284 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5292 * We are in the child. Make sure all other file descriptors are
5293 * closed, in case we are called with more opened file
5294 * descriptors than the standard ones.
5296 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5302 * Starting from here, we can create threads. This needs to be after
5303 * lttng_daemonize due to RCU.
5307 * Initialize the health check subsystem. This call should set the
5308 * appropriate time values.
5310 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5311 if (!health_sessiond
) {
5312 PERROR("health_app_create error");
5314 goto exit_health_sessiond_cleanup
;
5317 if (init_ht_cleanup_quit_pipe()) {
5319 goto exit_ht_cleanup_quit_pipe
;
5322 /* Setup the thread ht_cleanup communication pipe. */
5323 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5325 goto exit_ht_cleanup_pipe
;
5328 /* Set up max poll set size */
5329 if (lttng_poll_set_max_size()) {
5331 goto exit_set_max_size
;
5334 /* Create thread to clean up RCU hash tables */
5335 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5336 thread_ht_cleanup
, (void *) NULL
);
5339 PERROR("pthread_create ht_cleanup");
5341 goto exit_ht_cleanup
;
5344 /* Create thread quit pipe */
5345 if (init_thread_quit_pipe()) {
5347 goto exit_init_data
;
5350 /* Check if daemon is UID = 0 */
5351 is_root
= !getuid();
5354 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5357 goto exit_init_data
;
5360 /* Create global run dir with root access */
5361 if (create_lttng_rundir(rundir
)) {
5363 goto exit_init_data
;
5366 if (strlen(apps_unix_sock_path
) == 0) {
5367 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5368 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5371 goto exit_init_data
;
5375 if (strlen(client_unix_sock_path
) == 0) {
5376 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5377 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5380 goto exit_init_data
;
5384 /* Set global SHM for ust */
5385 if (strlen(wait_shm_path
) == 0) {
5386 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5387 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5390 goto exit_init_data
;
5394 if (strlen(health_unix_sock_path
) == 0) {
5395 ret
= snprintf(health_unix_sock_path
,
5396 sizeof(health_unix_sock_path
),
5397 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5400 goto exit_init_data
;
5404 /* Setup kernel consumerd path */
5405 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5406 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5409 goto exit_init_data
;
5411 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5412 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5415 goto exit_init_data
;
5418 DBG2("Kernel consumer err path: %s",
5419 kconsumer_data
.err_unix_sock_path
);
5420 DBG2("Kernel consumer cmd path: %s",
5421 kconsumer_data
.cmd_unix_sock_path
);
5423 home_path
= utils_get_home_dir();
5424 if (home_path
== NULL
) {
5425 /* TODO: Add --socket PATH option */
5426 ERR("Can't get HOME directory for sockets creation.");
5428 goto exit_init_data
;
5432 * Create rundir from home path. This will create something like
5435 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5438 goto exit_init_data
;
5441 if (create_lttng_rundir(rundir
)) {
5443 goto exit_init_data
;
5446 if (strlen(apps_unix_sock_path
) == 0) {
5447 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5448 DEFAULT_HOME_APPS_UNIX_SOCK
,
5452 goto exit_init_data
;
5456 /* Set the cli tool unix socket path */
5457 if (strlen(client_unix_sock_path
) == 0) {
5458 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5459 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5463 goto exit_init_data
;
5467 /* Set global SHM for ust */
5468 if (strlen(wait_shm_path
) == 0) {
5469 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5470 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5474 goto exit_init_data
;
5478 /* Set health check Unix path */
5479 if (strlen(health_unix_sock_path
) == 0) {
5480 ret
= snprintf(health_unix_sock_path
,
5481 sizeof(health_unix_sock_path
),
5482 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5486 goto exit_init_data
;
5491 lockfile_fd
= create_lockfile();
5492 if (lockfile_fd
< 0) {
5494 goto exit_init_data
;
5497 /* Set consumer initial state */
5498 kernel_consumerd_state
= CONSUMER_STOPPED
;
5499 ust_consumerd_state
= CONSUMER_STOPPED
;
5501 DBG("Client socket path %s", client_unix_sock_path
);
5502 DBG("Application socket path %s", apps_unix_sock_path
);
5503 DBG("Application wait path %s", wait_shm_path
);
5504 DBG("LTTng run directory path: %s", rundir
);
5506 /* 32 bits consumerd path setup */
5507 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5508 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5510 PERROR("snprintf 32-bit consumer error socket path");
5512 goto exit_init_data
;
5514 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5515 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5517 PERROR("snprintf 32-bit consumer command socket path");
5519 goto exit_init_data
;
5522 DBG2("UST consumer 32 bits err path: %s",
5523 ustconsumer32_data
.err_unix_sock_path
);
5524 DBG2("UST consumer 32 bits cmd path: %s",
5525 ustconsumer32_data
.cmd_unix_sock_path
);
5527 /* 64 bits consumerd path setup */
5528 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5529 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5531 PERROR("snprintf 64-bit consumer error socket path");
5533 goto exit_init_data
;
5535 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5536 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5538 PERROR("snprintf 64-bit consumer command socket path");
5540 goto exit_init_data
;
5543 DBG2("UST consumer 64 bits err path: %s",
5544 ustconsumer64_data
.err_unix_sock_path
);
5545 DBG2("UST consumer 64 bits cmd path: %s",
5546 ustconsumer64_data
.cmd_unix_sock_path
);
5549 * See if daemon already exist.
5551 if (check_existing_daemon()) {
5552 ERR("Already running daemon.\n");
5554 * We do not goto exit because we must not cleanup()
5555 * because a daemon is already running.
5558 goto exit_init_data
;
5562 * Init UST app hash table. Alloc hash table before this point since
5563 * cleanup() can get called after that point.
5565 if (ust_app_ht_alloc()) {
5567 goto exit_init_data
;
5570 /* Initialize agent domain subsystem. */
5571 if (agent_setup()) {
5572 /* ENOMEM at this point. */
5574 goto exit_init_data
;
5578 * These actions must be executed as root. We do that *after* setting up
5579 * the sockets path because we MUST make the check for another daemon using
5580 * those paths *before* trying to set the kernel consumer sockets and init
5584 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5586 goto exit_init_data
;
5589 /* Setup kernel tracer */
5590 if (!opt_no_kernel
) {
5591 init_kernel_tracer();
5592 if (kernel_tracer_fd
>= 0) {
5593 ret
= syscall_init_table();
5595 ERR("Unable to populate syscall table. "
5596 "Syscall tracing won't work "
5597 "for this session daemon.");
5602 /* Set ulimit for open files */
5605 /* init lttng_fd tracking must be done after set_ulimit. */
5608 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5610 goto exit_init_data
;
5613 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5615 goto exit_init_data
;
5618 /* Setup the needed unix socket */
5619 if (init_daemon_socket()) {
5621 goto exit_init_data
;
5624 /* Set credentials to socket */
5625 if (is_root
&& set_permissions(rundir
)) {
5627 goto exit_init_data
;
5630 /* Get parent pid if -S, --sig-parent is specified. */
5631 if (opt_sig_parent
) {
5635 /* Setup the kernel pipe for waking up the kernel thread */
5636 if (is_root
&& !opt_no_kernel
) {
5637 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5639 goto exit_init_data
;
5643 /* Setup the thread apps communication pipe. */
5644 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5646 goto exit_init_data
;
5649 /* Setup the thread apps notify communication pipe. */
5650 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5652 goto exit_init_data
;
5655 /* Initialize global buffer per UID and PID registry. */
5656 buffer_reg_init_uid_registry();
5657 buffer_reg_init_pid_registry();
5659 /* Init UST command queue. */
5660 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5663 * Get session list pointer. This pointer MUST NOT be free'd. This list
5664 * is statically declared in session.c
5666 session_list_ptr
= session_get_list();
5670 /* Check for the application socket timeout env variable. */
5671 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5672 if (env_app_timeout
) {
5673 app_socket_timeout
= atoi(env_app_timeout
);
5675 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5678 ret
= write_pidfile();
5680 ERR("Error in write_pidfile");
5682 goto exit_init_data
;
5684 ret
= write_agent_port();
5686 ERR("Error in write_agent_port");
5688 goto exit_init_data
;
5691 /* Initialize communication library */
5693 /* Initialize TCP timeout values */
5694 lttcomm_inet_init();
5696 if (load_session_init_data(&load_info
) < 0) {
5698 goto exit_init_data
;
5700 load_info
->path
= opt_load_session_path
;
5702 /* Create health-check thread */
5703 ret
= pthread_create(&health_thread
, NULL
,
5704 thread_manage_health
, (void *) NULL
);
5707 PERROR("pthread_create health");
5712 /* Create thread to manage the client socket */
5713 ret
= pthread_create(&client_thread
, NULL
,
5714 thread_manage_clients
, (void *) NULL
);
5717 PERROR("pthread_create clients");
5722 /* Create thread to dispatch registration */
5723 ret
= pthread_create(&dispatch_thread
, NULL
,
5724 thread_dispatch_ust_registration
, (void *) NULL
);
5727 PERROR("pthread_create dispatch");
5732 /* Create thread to manage application registration. */
5733 ret
= pthread_create(®_apps_thread
, NULL
,
5734 thread_registration_apps
, (void *) NULL
);
5737 PERROR("pthread_create registration");
5742 /* Create thread to manage application socket */
5743 ret
= pthread_create(&apps_thread
, NULL
,
5744 thread_manage_apps
, (void *) NULL
);
5747 PERROR("pthread_create apps");
5752 /* Create thread to manage application notify socket */
5753 ret
= pthread_create(&apps_notify_thread
, NULL
,
5754 ust_thread_manage_notify
, (void *) NULL
);
5757 PERROR("pthread_create notify");
5759 goto exit_apps_notify
;
5762 /* Create agent registration thread. */
5763 ret
= pthread_create(&agent_reg_thread
, NULL
,
5764 agent_thread_manage_registration
, (void *) NULL
);
5767 PERROR("pthread_create agent");
5769 goto exit_agent_reg
;
5772 /* Don't start this thread if kernel tracing is not requested nor root */
5773 if (is_root
&& !opt_no_kernel
) {
5774 /* Create kernel thread to manage kernel event */
5775 ret
= pthread_create(&kernel_thread
, NULL
,
5776 thread_manage_kernel
, (void *) NULL
);
5779 PERROR("pthread_create kernel");
5785 /* Create session loading thread. */
5786 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
5790 PERROR("pthread_create load_session_thread");
5792 goto exit_load_session
;
5796 * This is where we start awaiting program completion (e.g. through
5797 * signal that asks threads to teardown).
5800 ret
= pthread_join(load_session_thread
, &status
);
5803 PERROR("pthread_join load_session_thread");
5808 if (is_root
&& !opt_no_kernel
) {
5809 ret
= pthread_join(kernel_thread
, &status
);
5812 PERROR("pthread_join");
5818 ret
= pthread_join(agent_reg_thread
, &status
);
5821 PERROR("pthread_join agent");
5826 ret
= pthread_join(apps_notify_thread
, &status
);
5829 PERROR("pthread_join apps notify");
5834 ret
= pthread_join(apps_thread
, &status
);
5837 PERROR("pthread_join apps");
5842 ret
= pthread_join(reg_apps_thread
, &status
);
5845 PERROR("pthread_join");
5850 ret
= pthread_join(dispatch_thread
, &status
);
5853 PERROR("pthread_join");
5858 ret
= pthread_join(client_thread
, &status
);
5861 PERROR("pthread_join");
5866 ret
= pthread_join(health_thread
, &status
);
5869 PERROR("pthread_join health thread");
5876 * sessiond_cleanup() is called when no other thread is running, except
5877 * the ht_cleanup thread, which is needed to destroy the hash tables.
5879 rcu_thread_online();
5881 rcu_thread_offline();
5882 rcu_unregister_thread();
5884 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
5886 ERR("write error on ht_cleanup quit pipe");
5890 ret
= pthread_join(ht_cleanup_thread
, &status
);
5893 PERROR("pthread_join ht cleanup thread");
5899 utils_close_pipe(ht_cleanup_pipe
);
5900 exit_ht_cleanup_pipe
:
5903 * Close the ht_cleanup quit pipe.
5905 utils_close_pipe(ht_cleanup_quit_pipe
);
5906 exit_ht_cleanup_quit_pipe
:
5908 health_app_destroy(health_sessiond
);
5909 exit_health_sessiond_cleanup
:
5912 sessiond_cleanup_options();
5914 exit_set_signal_handler
: