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", required_argument
, 0, 'c' },
137 { "apps-sock", required_argument
, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
144 { "consumerd32-path", required_argument
, 0, '\0' },
145 { "consumerd32-libdir", required_argument
, 0, '\0' },
146 { "consumerd64-path", required_argument
, 0, '\0' },
147 { "consumerd64-libdir", required_argument
, 0, '\0' },
148 { "daemonize", no_argument
, 0, 'd' },
149 { "background", no_argument
, 0, 'b' },
150 { "sig-parent", no_argument
, 0, 'S' },
151 { "help", no_argument
, 0, 'h' },
152 { "group", required_argument
, 0, 'g' },
153 { "version", no_argument
, 0, 'V' },
154 { "quiet", no_argument
, 0, 'q' },
155 { "verbose", no_argument
, 0, 'v' },
156 { "verbose-consumer", no_argument
, 0, '\0' },
157 { "no-kernel", no_argument
, 0, '\0' },
158 { "pidfile", required_argument
, 0, 'p' },
159 { "agent-tcp-port", required_argument
, 0, '\0' },
160 { "config", required_argument
, 0, 'f' },
161 { "load", required_argument
, 0, 'l' },
162 { "kmod-probes", required_argument
, 0, '\0' },
163 { "extra-kmod-probes", required_argument
, 0, '\0' },
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("Cleaning up all agent apps");
680 agent_app_ht_clean();
682 DBG("Closing all UST sockets");
683 ust_app_clean_list();
684 buffer_reg_destroy_registries();
686 if (is_root
&& !opt_no_kernel
) {
687 DBG2("Closing kernel fd");
688 if (kernel_tracer_fd
>= 0) {
689 ret
= close(kernel_tracer_fd
);
694 DBG("Unloading kernel modules");
695 modprobe_remove_lttng_all();
699 close_consumer_sockets();
702 load_session_destroy_data(load_info
);
707 * Cleanup lock file by deleting it and finaly closing it which will
708 * release the file system lock.
710 if (lockfile_fd
>= 0) {
711 char lockfile_path
[PATH_MAX
];
713 ret
= generate_lock_file_path(lockfile_path
,
714 sizeof(lockfile_path
));
716 ret
= remove(lockfile_path
);
718 PERROR("remove lock file");
720 ret
= close(lockfile_fd
);
722 PERROR("close lock file");
728 * We do NOT rmdir rundir because there are other processes
729 * using it, for instance lttng-relayd, which can start in
730 * parallel with this teardown.
737 * Cleanup the daemon's option data structures.
739 static void sessiond_cleanup_options(void)
741 DBG("Cleaning up options");
744 * If the override option is set, the pointer points to a *non* const
745 * thus freeing it even though the variable type is set to const.
747 if (tracing_group_name_override
) {
748 free((void *) tracing_group_name
);
750 if (consumerd32_bin_override
) {
751 free((void *) consumerd32_bin
);
753 if (consumerd64_bin_override
) {
754 free((void *) consumerd64_bin
);
756 if (consumerd32_libdir_override
) {
757 free((void *) consumerd32_libdir
);
759 if (consumerd64_libdir_override
) {
760 free((void *) consumerd64_libdir
);
764 free(opt_load_session_path
);
765 free(kmod_probes_list
);
766 free(kmod_extra_probes_list
);
769 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
770 "Matthew, BEET driven development works!%c[%dm",
771 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
776 * Send data on a unix socket using the liblttsessiondcomm API.
778 * Return lttcomm error code.
780 static int send_unix_sock(int sock
, void *buf
, size_t len
)
782 /* Check valid length */
787 return lttcomm_send_unix_sock(sock
, buf
, len
);
791 * Free memory of a command context structure.
793 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
795 DBG("Clean command context structure");
797 if ((*cmd_ctx
)->llm
) {
798 free((*cmd_ctx
)->llm
);
800 if ((*cmd_ctx
)->lsm
) {
801 free((*cmd_ctx
)->lsm
);
809 * Notify UST applications using the shm mmap futex.
811 static int notify_ust_apps(int active
)
815 DBG("Notifying applications of session daemon state: %d", active
);
817 /* See shm.c for this call implying mmap, shm and futex calls */
818 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
819 if (wait_shm_mmap
== NULL
) {
823 /* Wake waiting process */
824 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
826 /* Apps notified successfully */
834 * Setup the outgoing data buffer for the response (llm) by allocating the
835 * right amount of memory and copying the original information from the lsm
838 * Return total size of the buffer pointed by buf.
840 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
846 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
847 if (cmd_ctx
->llm
== NULL
) {
853 /* Copy common data */
854 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
855 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
857 cmd_ctx
->llm
->data_size
= size
;
858 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
867 * Update the kernel poll set of all channel fd available over all tracing
868 * session. Add the wakeup pipe at the end of the set.
870 static int update_kernel_poll(struct lttng_poll_event
*events
)
873 struct ltt_session
*session
;
874 struct ltt_kernel_channel
*channel
;
876 DBG("Updating kernel poll set");
879 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
880 session_lock(session
);
881 if (session
->kernel_session
== NULL
) {
882 session_unlock(session
);
886 cds_list_for_each_entry(channel
,
887 &session
->kernel_session
->channel_list
.head
, list
) {
888 /* Add channel fd to the kernel poll set */
889 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
891 session_unlock(session
);
894 DBG("Channel fd %d added to kernel set", channel
->fd
);
896 session_unlock(session
);
898 session_unlock_list();
903 session_unlock_list();
908 * Find the channel fd from 'fd' over all tracing session. When found, check
909 * for new channel stream and send those stream fds to the kernel consumer.
911 * Useful for CPU hotplug feature.
913 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
916 struct ltt_session
*session
;
917 struct ltt_kernel_session
*ksess
;
918 struct ltt_kernel_channel
*channel
;
920 DBG("Updating kernel streams for channel fd %d", fd
);
923 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
924 session_lock(session
);
925 if (session
->kernel_session
== NULL
) {
926 session_unlock(session
);
929 ksess
= session
->kernel_session
;
931 cds_list_for_each_entry(channel
,
932 &ksess
->channel_list
.head
, list
) {
933 struct lttng_ht_iter iter
;
934 struct consumer_socket
*socket
;
936 if (channel
->fd
!= fd
) {
939 DBG("Channel found, updating kernel streams");
940 ret
= kernel_open_channel_stream(channel
);
944 /* Update the stream global counter */
945 ksess
->stream_count_global
+= ret
;
948 * Have we already sent fds to the consumer? If yes, it
949 * means that tracing is started so it is safe to send
950 * our updated stream fds.
952 if (ksess
->consumer_fds_sent
!= 1
953 || ksess
->consumer
== NULL
) {
959 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
960 &iter
.iter
, socket
, node
.node
) {
961 pthread_mutex_lock(socket
->lock
);
962 ret
= kernel_consumer_send_channel_stream(socket
,
964 session
->output_traces
? 1 : 0);
965 pthread_mutex_unlock(socket
->lock
);
973 session_unlock(session
);
975 session_unlock_list();
979 session_unlock(session
);
980 session_unlock_list();
985 * For each tracing session, update newly registered apps. The session list
986 * lock MUST be acquired before calling this.
988 static void update_ust_app(int app_sock
)
990 struct ltt_session
*sess
, *stmp
;
992 /* Consumer is in an ERROR state. Stop any application update. */
993 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
994 /* Stop the update process since the consumer is dead. */
998 /* For all tracing session(s) */
999 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1000 struct ust_app
*app
;
1003 if (!sess
->ust_session
) {
1004 goto unlock_session
;
1008 assert(app_sock
>= 0);
1009 app
= ust_app_find_by_sock(app_sock
);
1012 * Application can be unregistered before so
1013 * this is possible hence simply stopping the
1016 DBG3("UST app update failed to find app sock %d",
1020 ust_app_global_update(sess
->ust_session
, app
);
1024 session_unlock(sess
);
1029 * This thread manage event coming from the kernel.
1031 * Features supported in this thread:
1034 static void *thread_manage_kernel(void *data
)
1036 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1037 uint32_t revents
, nb_fd
;
1039 struct lttng_poll_event events
;
1041 DBG("[thread] Thread manage kernel started");
1043 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1046 * This first step of the while is to clean this structure which could free
1047 * non NULL pointers so initialize it before the loop.
1049 lttng_poll_init(&events
);
1051 if (testpoint(sessiond_thread_manage_kernel
)) {
1052 goto error_testpoint
;
1055 health_code_update();
1057 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1058 goto error_testpoint
;
1062 health_code_update();
1064 if (update_poll_flag
== 1) {
1065 /* Clean events object. We are about to populate it again. */
1066 lttng_poll_clean(&events
);
1068 ret
= sessiond_set_thread_pollset(&events
, 2);
1070 goto error_poll_create
;
1073 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1078 /* This will add the available kernel channel if any. */
1079 ret
= update_kernel_poll(&events
);
1083 update_poll_flag
= 0;
1086 DBG("Thread kernel polling");
1088 /* Poll infinite value of time */
1090 health_poll_entry();
1091 ret
= lttng_poll_wait(&events
, -1);
1092 DBG("Thread kernel return from poll on %d fds",
1093 LTTNG_POLL_GETNB(&events
));
1097 * Restart interrupted system call.
1099 if (errno
== EINTR
) {
1103 } else if (ret
== 0) {
1104 /* Should not happen since timeout is infinite */
1105 ERR("Return value of poll is 0 with an infinite timeout.\n"
1106 "This should not have happened! Continuing...");
1112 for (i
= 0; i
< nb_fd
; i
++) {
1113 /* Fetch once the poll data */
1114 revents
= LTTNG_POLL_GETEV(&events
, i
);
1115 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1117 health_code_update();
1120 /* No activity for this FD (poll implementation). */
1124 /* Thread quit pipe has been closed. Killing thread. */
1125 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1131 /* Check for data on kernel pipe */
1132 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1133 (void) lttng_read(kernel_poll_pipe
[0],
1136 * Ret value is useless here, if this pipe gets any actions an
1137 * update is required anyway.
1139 update_poll_flag
= 1;
1143 * New CPU detected by the kernel. Adding kernel stream to
1144 * kernel session and updating the kernel consumer
1146 if (revents
& LPOLLIN
) {
1147 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1153 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1154 * and unregister kernel stream at this point.
1163 lttng_poll_clean(&events
);
1166 utils_close_pipe(kernel_poll_pipe
);
1167 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1170 ERR("Health error occurred in %s", __func__
);
1171 WARN("Kernel thread died unexpectedly. "
1172 "Kernel tracing can continue but CPU hotplug is disabled.");
1174 health_unregister(health_sessiond
);
1175 DBG("Kernel thread dying");
1180 * Signal pthread condition of the consumer data that the thread.
1182 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1184 pthread_mutex_lock(&data
->cond_mutex
);
1187 * The state is set before signaling. It can be any value, it's the waiter
1188 * job to correctly interpret this condition variable associated to the
1189 * consumer pthread_cond.
1191 * A value of 0 means that the corresponding thread of the consumer data
1192 * was not started. 1 indicates that the thread has started and is ready
1193 * for action. A negative value means that there was an error during the
1196 data
->consumer_thread_is_ready
= state
;
1197 (void) pthread_cond_signal(&data
->cond
);
1199 pthread_mutex_unlock(&data
->cond_mutex
);
1203 * This thread manage the consumer error sent back to the session daemon.
1205 static void *thread_manage_consumer(void *data
)
1207 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1208 uint32_t revents
, nb_fd
;
1209 enum lttcomm_return_code code
;
1210 struct lttng_poll_event events
;
1211 struct consumer_data
*consumer_data
= data
;
1213 DBG("[thread] Manage consumer started");
1215 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1217 health_code_update();
1220 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1221 * metadata_sock. Nothing more will be added to this poll set.
1223 ret
= sessiond_set_thread_pollset(&events
, 3);
1229 * The error socket here is already in a listening state which was done
1230 * just before spawning this thread to avoid a race between the consumer
1231 * daemon exec trying to connect and the listen() call.
1233 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1238 health_code_update();
1240 /* Infinite blocking call, waiting for transmission */
1242 health_poll_entry();
1244 if (testpoint(sessiond_thread_manage_consumer
)) {
1248 ret
= lttng_poll_wait(&events
, -1);
1252 * Restart interrupted system call.
1254 if (errno
== EINTR
) {
1262 for (i
= 0; i
< nb_fd
; i
++) {
1263 /* Fetch once the poll data */
1264 revents
= LTTNG_POLL_GETEV(&events
, i
);
1265 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1267 health_code_update();
1270 /* No activity for this FD (poll implementation). */
1274 /* Thread quit pipe has been closed. Killing thread. */
1275 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1281 /* Event on the registration socket */
1282 if (pollfd
== consumer_data
->err_sock
) {
1283 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1284 ERR("consumer err socket poll error");
1290 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1296 * Set the CLOEXEC flag. Return code is useless because either way, the
1299 (void) utils_set_fd_cloexec(sock
);
1301 health_code_update();
1303 DBG2("Receiving code from consumer err_sock");
1305 /* Getting status code from kconsumerd */
1306 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1307 sizeof(enum lttcomm_return_code
));
1312 health_code_update();
1313 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1314 /* Connect both socket, command and metadata. */
1315 consumer_data
->cmd_sock
=
1316 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1317 consumer_data
->metadata_fd
=
1318 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1319 if (consumer_data
->cmd_sock
< 0
1320 || consumer_data
->metadata_fd
< 0) {
1321 PERROR("consumer connect cmd socket");
1322 /* On error, signal condition and quit. */
1323 signal_consumer_condition(consumer_data
, -1);
1326 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1327 /* Create metadata socket lock. */
1328 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1329 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1330 PERROR("zmalloc pthread mutex");
1334 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1336 signal_consumer_condition(consumer_data
, 1);
1337 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1338 DBG("Consumer metadata socket ready (fd: %d)",
1339 consumer_data
->metadata_fd
);
1341 ERR("consumer error when waiting for SOCK_READY : %s",
1342 lttcomm_get_readable_code(-code
));
1346 /* Remove the consumerd error sock since we've established a connexion */
1347 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1352 /* Add new accepted error socket. */
1353 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1358 /* Add metadata socket that is successfully connected. */
1359 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1360 LPOLLIN
| LPOLLRDHUP
);
1365 health_code_update();
1367 /* Infinite blocking call, waiting for transmission */
1370 health_code_update();
1372 /* Exit the thread because the thread quit pipe has been triggered. */
1374 /* Not a health error. */
1379 health_poll_entry();
1380 ret
= lttng_poll_wait(&events
, -1);
1384 * Restart interrupted system call.
1386 if (errno
== EINTR
) {
1394 for (i
= 0; i
< nb_fd
; i
++) {
1395 /* Fetch once the poll data */
1396 revents
= LTTNG_POLL_GETEV(&events
, i
);
1397 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1399 health_code_update();
1402 /* No activity for this FD (poll implementation). */
1407 * Thread quit pipe has been triggered, flag that we should stop
1408 * but continue the current loop to handle potential data from
1411 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1413 if (pollfd
== sock
) {
1414 /* Event on the consumerd socket */
1415 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1416 ERR("consumer err socket second poll error");
1419 health_code_update();
1420 /* Wait for any kconsumerd error */
1421 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1422 sizeof(enum lttcomm_return_code
));
1424 ERR("consumer closed the command socket");
1428 ERR("consumer return code : %s",
1429 lttcomm_get_readable_code(-code
));
1432 } else if (pollfd
== consumer_data
->metadata_fd
) {
1433 /* UST metadata requests */
1434 ret
= ust_consumer_metadata_request(
1435 &consumer_data
->metadata_sock
);
1437 ERR("Handling metadata request");
1441 /* No need for an else branch all FDs are tested prior. */
1443 health_code_update();
1449 * We lock here because we are about to close the sockets and some other
1450 * thread might be using them so get exclusive access which will abort all
1451 * other consumer command by other threads.
1453 pthread_mutex_lock(&consumer_data
->lock
);
1455 /* Immediately set the consumerd state to stopped */
1456 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1457 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1458 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1459 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1460 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1462 /* Code flow error... */
1466 if (consumer_data
->err_sock
>= 0) {
1467 ret
= close(consumer_data
->err_sock
);
1471 consumer_data
->err_sock
= -1;
1473 if (consumer_data
->cmd_sock
>= 0) {
1474 ret
= close(consumer_data
->cmd_sock
);
1478 consumer_data
->cmd_sock
= -1;
1480 if (consumer_data
->metadata_sock
.fd_ptr
&&
1481 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1482 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1494 unlink(consumer_data
->err_unix_sock_path
);
1495 unlink(consumer_data
->cmd_unix_sock_path
);
1496 consumer_data
->pid
= 0;
1497 pthread_mutex_unlock(&consumer_data
->lock
);
1499 /* Cleanup metadata socket mutex. */
1500 if (consumer_data
->metadata_sock
.lock
) {
1501 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1502 free(consumer_data
->metadata_sock
.lock
);
1504 lttng_poll_clean(&events
);
1508 ERR("Health error occurred in %s", __func__
);
1510 health_unregister(health_sessiond
);
1511 DBG("consumer thread cleanup completed");
1517 * This thread manage application communication.
1519 static void *thread_manage_apps(void *data
)
1521 int i
, ret
, pollfd
, err
= -1;
1523 uint32_t revents
, nb_fd
;
1524 struct lttng_poll_event events
;
1526 DBG("[thread] Manage application started");
1528 rcu_register_thread();
1529 rcu_thread_online();
1531 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1533 if (testpoint(sessiond_thread_manage_apps
)) {
1534 goto error_testpoint
;
1537 health_code_update();
1539 ret
= sessiond_set_thread_pollset(&events
, 2);
1541 goto error_poll_create
;
1544 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1549 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1553 health_code_update();
1556 DBG("Apps thread polling");
1558 /* Inifinite blocking call, waiting for transmission */
1560 health_poll_entry();
1561 ret
= lttng_poll_wait(&events
, -1);
1562 DBG("Apps thread return from poll on %d fds",
1563 LTTNG_POLL_GETNB(&events
));
1567 * Restart interrupted system call.
1569 if (errno
== EINTR
) {
1577 for (i
= 0; i
< nb_fd
; i
++) {
1578 /* Fetch once the poll data */
1579 revents
= LTTNG_POLL_GETEV(&events
, i
);
1580 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1582 health_code_update();
1585 /* No activity for this FD (poll implementation). */
1589 /* Thread quit pipe has been closed. Killing thread. */
1590 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1596 /* Inspect the apps cmd pipe */
1597 if (pollfd
== apps_cmd_pipe
[0]) {
1598 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1599 ERR("Apps command pipe error");
1601 } else if (revents
& LPOLLIN
) {
1605 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1606 if (size_ret
< sizeof(sock
)) {
1607 PERROR("read apps cmd pipe");
1611 health_code_update();
1614 * We only monitor the error events of the socket. This
1615 * thread does not handle any incoming data from UST
1618 ret
= lttng_poll_add(&events
, sock
,
1619 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1624 DBG("Apps with sock %d added to poll set", sock
);
1628 * At this point, we know that a registered application made
1629 * the event at poll_wait.
1631 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1632 /* Removing from the poll set */
1633 ret
= lttng_poll_del(&events
, pollfd
);
1638 /* Socket closed on remote end. */
1639 ust_app_unregister(pollfd
);
1643 health_code_update();
1649 lttng_poll_clean(&events
);
1652 utils_close_pipe(apps_cmd_pipe
);
1653 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1656 * We don't clean the UST app hash table here since already registered
1657 * applications can still be controlled so let them be until the session
1658 * daemon dies or the applications stop.
1663 ERR("Health error occurred in %s", __func__
);
1665 health_unregister(health_sessiond
);
1666 DBG("Application communication apps thread cleanup complete");
1667 rcu_thread_offline();
1668 rcu_unregister_thread();
1673 * Send a socket to a thread This is called from the dispatch UST registration
1674 * thread once all sockets are set for the application.
1676 * The sock value can be invalid, we don't really care, the thread will handle
1677 * it and make the necessary cleanup if so.
1679 * On success, return 0 else a negative value being the errno message of the
1682 static int send_socket_to_thread(int fd
, int sock
)
1687 * It's possible that the FD is set as invalid with -1 concurrently just
1688 * before calling this function being a shutdown state of the thread.
1695 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1696 if (ret
< sizeof(sock
)) {
1697 PERROR("write apps pipe %d", fd
);
1704 /* All good. Don't send back the write positive ret value. */
1711 * Sanitize the wait queue of the dispatch registration thread meaning removing
1712 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1713 * notify socket is never received.
1715 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1717 int ret
, nb_fd
= 0, i
;
1718 unsigned int fd_added
= 0;
1719 struct lttng_poll_event events
;
1720 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1724 lttng_poll_init(&events
);
1726 /* Just skip everything for an empty queue. */
1727 if (!wait_queue
->count
) {
1731 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1736 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1737 &wait_queue
->head
, head
) {
1738 assert(wait_node
->app
);
1739 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1740 LPOLLHUP
| LPOLLERR
);
1753 * Poll but don't block so we can quickly identify the faulty events and
1754 * clean them afterwards from the wait queue.
1756 ret
= lttng_poll_wait(&events
, 0);
1762 for (i
= 0; i
< nb_fd
; i
++) {
1763 /* Get faulty FD. */
1764 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1765 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1768 /* No activity for this FD (poll implementation). */
1772 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1773 &wait_queue
->head
, head
) {
1774 if (pollfd
== wait_node
->app
->sock
&&
1775 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1776 cds_list_del(&wait_node
->head
);
1777 wait_queue
->count
--;
1778 ust_app_destroy(wait_node
->app
);
1786 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1790 lttng_poll_clean(&events
);
1794 lttng_poll_clean(&events
);
1796 ERR("Unable to sanitize wait queue");
1801 * Dispatch request from the registration threads to the application
1802 * communication thread.
1804 static void *thread_dispatch_ust_registration(void *data
)
1807 struct cds_wfcq_node
*node
;
1808 struct ust_command
*ust_cmd
= NULL
;
1809 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1810 struct ust_reg_wait_queue wait_queue
= {
1814 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1816 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1817 goto error_testpoint
;
1820 health_code_update();
1822 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1824 DBG("[thread] Dispatch UST command started");
1826 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1827 health_code_update();
1829 /* Atomically prepare the queue futex */
1830 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1833 struct ust_app
*app
= NULL
;
1837 * Make sure we don't have node(s) that have hung up before receiving
1838 * the notify socket. This is to clean the list in order to avoid
1839 * memory leaks from notify socket that are never seen.
1841 sanitize_wait_queue(&wait_queue
);
1843 health_code_update();
1844 /* Dequeue command for registration */
1845 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1847 DBG("Woken up but nothing in the UST command queue");
1848 /* Continue thread execution */
1852 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1854 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1855 " gid:%d sock:%d name:%s (version %d.%d)",
1856 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1857 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1858 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1859 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1861 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1862 wait_node
= zmalloc(sizeof(*wait_node
));
1864 PERROR("zmalloc wait_node dispatch");
1865 ret
= close(ust_cmd
->sock
);
1867 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1869 lttng_fd_put(LTTNG_FD_APPS
, 1);
1873 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1875 /* Create application object if socket is CMD. */
1876 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1878 if (!wait_node
->app
) {
1879 ret
= close(ust_cmd
->sock
);
1881 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1883 lttng_fd_put(LTTNG_FD_APPS
, 1);
1889 * Add application to the wait queue so we can set the notify
1890 * socket before putting this object in the global ht.
1892 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1897 * We have to continue here since we don't have the notify
1898 * socket and the application MUST be added to the hash table
1899 * only at that moment.
1904 * Look for the application in the local wait queue and set the
1905 * notify socket if found.
1907 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1908 &wait_queue
.head
, head
) {
1909 health_code_update();
1910 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1911 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1912 cds_list_del(&wait_node
->head
);
1914 app
= wait_node
->app
;
1916 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1922 * With no application at this stage the received socket is
1923 * basically useless so close it before we free the cmd data
1924 * structure for good.
1927 ret
= close(ust_cmd
->sock
);
1929 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1931 lttng_fd_put(LTTNG_FD_APPS
, 1);
1938 * @session_lock_list
1940 * Lock the global session list so from the register up to the
1941 * registration done message, no thread can see the application
1942 * and change its state.
1944 session_lock_list();
1948 * Add application to the global hash table. This needs to be
1949 * done before the update to the UST registry can locate the
1954 /* Set app version. This call will print an error if needed. */
1955 (void) ust_app_version(app
);
1957 /* Send notify socket through the notify pipe. */
1958 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1962 session_unlock_list();
1964 * No notify thread, stop the UST tracing. However, this is
1965 * not an internal error of the this thread thus setting
1966 * the health error code to a normal exit.
1973 * Update newly registered application with the tracing
1974 * registry info already enabled information.
1976 update_ust_app(app
->sock
);
1979 * Don't care about return value. Let the manage apps threads
1980 * handle app unregistration upon socket close.
1982 (void) ust_app_register_done(app
->sock
);
1985 * Even if the application socket has been closed, send the app
1986 * to the thread and unregistration will take place at that
1989 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1992 session_unlock_list();
1994 * No apps. thread, stop the UST tracing. However, this is
1995 * not an internal error of the this thread thus setting
1996 * the health error code to a normal exit.
2003 session_unlock_list();
2005 } while (node
!= NULL
);
2007 health_poll_entry();
2008 /* Futex wait on queue. Blocking call on futex() */
2009 futex_nto1_wait(&ust_cmd_queue
.futex
);
2012 /* Normal exit, no error */
2016 /* Clean up wait queue. */
2017 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2018 &wait_queue
.head
, head
) {
2019 cds_list_del(&wait_node
->head
);
2025 DBG("Dispatch thread dying");
2028 ERR("Health error occurred in %s", __func__
);
2030 health_unregister(health_sessiond
);
2035 * This thread manage application registration.
2037 static void *thread_registration_apps(void *data
)
2039 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2040 uint32_t revents
, nb_fd
;
2041 struct lttng_poll_event events
;
2043 * Get allocated in this thread, enqueued to a global queue, dequeued and
2044 * freed in the manage apps thread.
2046 struct ust_command
*ust_cmd
= NULL
;
2048 DBG("[thread] Manage application registration started");
2050 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2052 if (testpoint(sessiond_thread_registration_apps
)) {
2053 goto error_testpoint
;
2056 ret
= lttcomm_listen_unix_sock(apps_sock
);
2062 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2063 * more will be added to this poll set.
2065 ret
= sessiond_set_thread_pollset(&events
, 2);
2067 goto error_create_poll
;
2070 /* Add the application registration socket */
2071 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2073 goto error_poll_add
;
2076 /* Notify all applications to register */
2077 ret
= notify_ust_apps(1);
2079 ERR("Failed to notify applications or create the wait shared memory.\n"
2080 "Execution continues but there might be problem for already\n"
2081 "running applications that wishes to register.");
2085 DBG("Accepting application registration");
2087 /* Inifinite blocking call, waiting for transmission */
2089 health_poll_entry();
2090 ret
= lttng_poll_wait(&events
, -1);
2094 * Restart interrupted system call.
2096 if (errno
== EINTR
) {
2104 for (i
= 0; i
< nb_fd
; i
++) {
2105 health_code_update();
2107 /* Fetch once the poll data */
2108 revents
= LTTNG_POLL_GETEV(&events
, i
);
2109 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2112 /* No activity for this FD (poll implementation). */
2116 /* Thread quit pipe has been closed. Killing thread. */
2117 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2123 /* Event on the registration socket */
2124 if (pollfd
== apps_sock
) {
2125 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2126 ERR("Register apps socket poll error");
2128 } else if (revents
& LPOLLIN
) {
2129 sock
= lttcomm_accept_unix_sock(apps_sock
);
2135 * Set socket timeout for both receiving and ending.
2136 * app_socket_timeout is in seconds, whereas
2137 * lttcomm_setsockopt_rcv_timeout and
2138 * lttcomm_setsockopt_snd_timeout expect msec as
2141 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2142 app_socket_timeout
* 1000);
2143 (void) lttcomm_setsockopt_snd_timeout(sock
,
2144 app_socket_timeout
* 1000);
2147 * Set the CLOEXEC flag. Return code is useless because
2148 * either way, the show must go on.
2150 (void) utils_set_fd_cloexec(sock
);
2152 /* Create UST registration command for enqueuing */
2153 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2154 if (ust_cmd
== NULL
) {
2155 PERROR("ust command zmalloc");
2160 * Using message-based transmissions to ensure we don't
2161 * have to deal with partially received messages.
2163 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2165 ERR("Exhausted file descriptors allowed for applications.");
2175 health_code_update();
2176 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2179 /* Close socket of the application. */
2184 lttng_fd_put(LTTNG_FD_APPS
, 1);
2188 health_code_update();
2190 ust_cmd
->sock
= sock
;
2193 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2194 " gid:%d sock:%d name:%s (version %d.%d)",
2195 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2196 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2197 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2198 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2201 * Lock free enqueue the registration request. The red pill
2202 * has been taken! This apps will be part of the *system*.
2204 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2207 * Wake the registration queue futex. Implicit memory
2208 * barrier with the exchange in cds_wfcq_enqueue.
2210 futex_nto1_wake(&ust_cmd_queue
.futex
);
2218 /* Notify that the registration thread is gone */
2221 if (apps_sock
>= 0) {
2222 ret
= close(apps_sock
);
2232 lttng_fd_put(LTTNG_FD_APPS
, 1);
2234 unlink(apps_unix_sock_path
);
2237 lttng_poll_clean(&events
);
2241 DBG("UST Registration thread cleanup complete");
2244 ERR("Health error occurred in %s", __func__
);
2246 health_unregister(health_sessiond
);
2252 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2253 * exec or it will fails.
2255 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2258 struct timespec timeout
;
2260 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2261 consumer_data
->consumer_thread_is_ready
= 0;
2263 /* Setup pthread condition */
2264 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2267 PERROR("pthread_condattr_init consumer data");
2272 * Set the monotonic clock in order to make sure we DO NOT jump in time
2273 * between the clock_gettime() call and the timedwait call. See bug #324
2274 * for a more details and how we noticed it.
2276 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2279 PERROR("pthread_condattr_setclock consumer data");
2283 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2286 PERROR("pthread_cond_init consumer data");
2290 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2294 PERROR("pthread_create consumer");
2299 /* We are about to wait on a pthread condition */
2300 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2302 /* Get time for sem_timedwait absolute timeout */
2303 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2305 * Set the timeout for the condition timed wait even if the clock gettime
2306 * call fails since we might loop on that call and we want to avoid to
2307 * increment the timeout too many times.
2309 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2312 * The following loop COULD be skipped in some conditions so this is why we
2313 * set ret to 0 in order to make sure at least one round of the loop is
2319 * Loop until the condition is reached or when a timeout is reached. Note
2320 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2321 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2322 * possible. This loop does not take any chances and works with both of
2325 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2326 if (clock_ret
< 0) {
2327 PERROR("clock_gettime spawn consumer");
2328 /* Infinite wait for the consumerd thread to be ready */
2329 ret
= pthread_cond_wait(&consumer_data
->cond
,
2330 &consumer_data
->cond_mutex
);
2332 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2333 &consumer_data
->cond_mutex
, &timeout
);
2337 /* Release the pthread condition */
2338 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2342 if (ret
== ETIMEDOUT
) {
2346 * Call has timed out so we kill the kconsumerd_thread and return
2349 ERR("Condition timed out. The consumer thread was never ready."
2351 pth_ret
= pthread_cancel(consumer_data
->thread
);
2353 PERROR("pthread_cancel consumer thread");
2356 PERROR("pthread_cond_wait failed consumer thread");
2358 /* Caller is expecting a negative value on failure. */
2363 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2364 if (consumer_data
->pid
== 0) {
2365 ERR("Consumerd did not start");
2366 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2369 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2378 * Join consumer thread
2380 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2384 /* Consumer pid must be a real one. */
2385 if (consumer_data
->pid
> 0) {
2387 ret
= kill(consumer_data
->pid
, SIGTERM
);
2389 PERROR("Error killing consumer daemon");
2392 return pthread_join(consumer_data
->thread
, &status
);
2399 * Fork and exec a consumer daemon (consumerd).
2401 * Return pid if successful else -1.
2403 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2407 const char *consumer_to_use
;
2408 const char *verbosity
;
2411 DBG("Spawning consumerd");
2418 if (opt_verbose_consumer
) {
2419 verbosity
= "--verbose";
2420 } else if (lttng_opt_quiet
) {
2421 verbosity
= "--quiet";
2426 switch (consumer_data
->type
) {
2427 case LTTNG_CONSUMER_KERNEL
:
2429 * Find out which consumerd to execute. We will first try the
2430 * 64-bit path, then the sessiond's installation directory, and
2431 * fallback on the 32-bit one,
2433 DBG3("Looking for a kernel consumer at these locations:");
2434 DBG3(" 1) %s", consumerd64_bin
);
2435 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2436 DBG3(" 3) %s", consumerd32_bin
);
2437 if (stat(consumerd64_bin
, &st
) == 0) {
2438 DBG3("Found location #1");
2439 consumer_to_use
= consumerd64_bin
;
2440 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2441 DBG3("Found location #2");
2442 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2443 } else if (stat(consumerd32_bin
, &st
) == 0) {
2444 DBG3("Found location #3");
2445 consumer_to_use
= consumerd32_bin
;
2447 DBG("Could not find any valid consumerd executable");
2451 DBG("Using kernel consumer at: %s", consumer_to_use
);
2452 ret
= execl(consumer_to_use
,
2453 "lttng-consumerd", verbosity
, "-k",
2454 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2455 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2456 "--group", tracing_group_name
,
2459 case LTTNG_CONSUMER64_UST
:
2461 char *tmpnew
= NULL
;
2463 if (consumerd64_libdir
[0] != '\0') {
2467 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2471 tmplen
= strlen("LD_LIBRARY_PATH=")
2472 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2473 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2478 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2479 strcat(tmpnew
, consumerd64_libdir
);
2480 if (tmp
[0] != '\0') {
2481 strcat(tmpnew
, ":");
2482 strcat(tmpnew
, tmp
);
2484 ret
= putenv(tmpnew
);
2491 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2492 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2493 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2494 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2495 "--group", tracing_group_name
,
2497 if (consumerd64_libdir
[0] != '\0') {
2502 case LTTNG_CONSUMER32_UST
:
2504 char *tmpnew
= NULL
;
2506 if (consumerd32_libdir
[0] != '\0') {
2510 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2514 tmplen
= strlen("LD_LIBRARY_PATH=")
2515 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2516 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2521 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2522 strcat(tmpnew
, consumerd32_libdir
);
2523 if (tmp
[0] != '\0') {
2524 strcat(tmpnew
, ":");
2525 strcat(tmpnew
, tmp
);
2527 ret
= putenv(tmpnew
);
2534 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2535 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2536 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2537 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2538 "--group", tracing_group_name
,
2540 if (consumerd32_libdir
[0] != '\0') {
2546 PERROR("unknown consumer type");
2550 PERROR("Consumer execl()");
2552 /* Reaching this point, we got a failure on our execl(). */
2554 } else if (pid
> 0) {
2557 PERROR("start consumer fork");
2565 * Spawn the consumerd daemon and session daemon thread.
2567 static int start_consumerd(struct consumer_data
*consumer_data
)
2572 * Set the listen() state on the socket since there is a possible race
2573 * between the exec() of the consumer daemon and this call if place in the
2574 * consumer thread. See bug #366 for more details.
2576 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2581 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2582 if (consumer_data
->pid
!= 0) {
2583 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2587 ret
= spawn_consumerd(consumer_data
);
2589 ERR("Spawning consumerd failed");
2590 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2594 /* Setting up the consumer_data pid */
2595 consumer_data
->pid
= ret
;
2596 DBG2("Consumer pid %d", consumer_data
->pid
);
2597 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2599 DBG2("Spawning consumer control thread");
2600 ret
= spawn_consumer_thread(consumer_data
);
2602 ERR("Fatal error spawning consumer control thread");
2610 /* Cleanup already created sockets on error. */
2611 if (consumer_data
->err_sock
>= 0) {
2614 err
= close(consumer_data
->err_sock
);
2616 PERROR("close consumer data error socket");
2623 * Setup necessary data for kernel tracer action.
2625 static int init_kernel_tracer(void)
2629 /* Modprobe lttng kernel modules */
2630 ret
= modprobe_lttng_control();
2635 /* Open debugfs lttng */
2636 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2637 if (kernel_tracer_fd
< 0) {
2638 DBG("Failed to open %s", module_proc_lttng
);
2643 /* Validate kernel version */
2644 ret
= kernel_validate_version(kernel_tracer_fd
);
2649 ret
= modprobe_lttng_data();
2654 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2658 modprobe_remove_lttng_control();
2659 ret
= close(kernel_tracer_fd
);
2663 kernel_tracer_fd
= -1;
2664 return LTTNG_ERR_KERN_VERSION
;
2667 ret
= close(kernel_tracer_fd
);
2673 modprobe_remove_lttng_control();
2676 WARN("No kernel tracer available");
2677 kernel_tracer_fd
= -1;
2679 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2681 return LTTNG_ERR_KERN_NA
;
2687 * Copy consumer output from the tracing session to the domain session. The
2688 * function also applies the right modification on a per domain basis for the
2689 * trace files destination directory.
2691 * Should *NOT* be called with RCU read-side lock held.
2693 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2696 const char *dir_name
;
2697 struct consumer_output
*consumer
;
2700 assert(session
->consumer
);
2703 case LTTNG_DOMAIN_KERNEL
:
2704 DBG3("Copying tracing session consumer output in kernel session");
2706 * XXX: We should audit the session creation and what this function
2707 * does "extra" in order to avoid a destroy since this function is used
2708 * in the domain session creation (kernel and ust) only. Same for UST
2711 if (session
->kernel_session
->consumer
) {
2712 consumer_destroy_output(session
->kernel_session
->consumer
);
2714 session
->kernel_session
->consumer
=
2715 consumer_copy_output(session
->consumer
);
2716 /* Ease our life a bit for the next part */
2717 consumer
= session
->kernel_session
->consumer
;
2718 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2720 case LTTNG_DOMAIN_JUL
:
2721 case LTTNG_DOMAIN_LOG4J
:
2722 case LTTNG_DOMAIN_PYTHON
:
2723 case LTTNG_DOMAIN_UST
:
2724 DBG3("Copying tracing session consumer output in UST session");
2725 if (session
->ust_session
->consumer
) {
2726 consumer_destroy_output(session
->ust_session
->consumer
);
2728 session
->ust_session
->consumer
=
2729 consumer_copy_output(session
->consumer
);
2730 /* Ease our life a bit for the next part */
2731 consumer
= session
->ust_session
->consumer
;
2732 dir_name
= DEFAULT_UST_TRACE_DIR
;
2735 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2739 /* Append correct directory to subdir */
2740 strncat(consumer
->subdir
, dir_name
,
2741 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2742 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2751 * Create an UST session and add it to the session ust list.
2753 * Should *NOT* be called with RCU read-side lock held.
2755 static int create_ust_session(struct ltt_session
*session
,
2756 struct lttng_domain
*domain
)
2759 struct ltt_ust_session
*lus
= NULL
;
2763 assert(session
->consumer
);
2765 switch (domain
->type
) {
2766 case LTTNG_DOMAIN_JUL
:
2767 case LTTNG_DOMAIN_LOG4J
:
2768 case LTTNG_DOMAIN_PYTHON
:
2769 case LTTNG_DOMAIN_UST
:
2772 ERR("Unknown UST domain on create session %d", domain
->type
);
2773 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2777 DBG("Creating UST session");
2779 lus
= trace_ust_create_session(session
->id
);
2781 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2785 lus
->uid
= session
->uid
;
2786 lus
->gid
= session
->gid
;
2787 lus
->output_traces
= session
->output_traces
;
2788 lus
->snapshot_mode
= session
->snapshot_mode
;
2789 lus
->live_timer_interval
= session
->live_timer
;
2790 session
->ust_session
= lus
;
2791 if (session
->shm_path
[0]) {
2792 strncpy(lus
->root_shm_path
, session
->shm_path
,
2793 sizeof(lus
->root_shm_path
));
2794 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2795 strncpy(lus
->shm_path
, session
->shm_path
,
2796 sizeof(lus
->shm_path
));
2797 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2798 strncat(lus
->shm_path
, "/ust",
2799 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2801 /* Copy session output to the newly created UST session */
2802 ret
= copy_session_consumer(domain
->type
, session
);
2803 if (ret
!= LTTNG_OK
) {
2811 session
->ust_session
= NULL
;
2816 * Create a kernel tracer session then create the default channel.
2818 static int create_kernel_session(struct ltt_session
*session
)
2822 DBG("Creating kernel session");
2824 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2826 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2830 /* Code flow safety */
2831 assert(session
->kernel_session
);
2833 /* Copy session output to the newly created Kernel session */
2834 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2835 if (ret
!= LTTNG_OK
) {
2839 /* Create directory(ies) on local filesystem. */
2840 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2841 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2842 ret
= run_as_mkdir_recursive(
2843 session
->kernel_session
->consumer
->dst
.trace_path
,
2844 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2846 if (ret
!= -EEXIST
) {
2847 ERR("Trace directory creation error");
2853 session
->kernel_session
->uid
= session
->uid
;
2854 session
->kernel_session
->gid
= session
->gid
;
2855 session
->kernel_session
->output_traces
= session
->output_traces
;
2856 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2861 trace_kernel_destroy_session(session
->kernel_session
);
2862 session
->kernel_session
= NULL
;
2867 * Count number of session permitted by uid/gid.
2869 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2872 struct ltt_session
*session
;
2874 DBG("Counting number of available session for UID %d GID %d",
2876 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2878 * Only list the sessions the user can control.
2880 if (!session_access_ok(session
, uid
, gid
)) {
2889 * Process the command requested by the lttng client within the command
2890 * context structure. This function make sure that the return structure (llm)
2891 * is set and ready for transmission before returning.
2893 * Return any error encountered or 0 for success.
2895 * "sock" is only used for special-case var. len data.
2897 * Should *NOT* be called with RCU read-side lock held.
2899 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2903 int need_tracing_session
= 1;
2906 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2910 switch (cmd_ctx
->lsm
->cmd_type
) {
2911 case LTTNG_CREATE_SESSION
:
2912 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2913 case LTTNG_CREATE_SESSION_LIVE
:
2914 case LTTNG_DESTROY_SESSION
:
2915 case LTTNG_LIST_SESSIONS
:
2916 case LTTNG_LIST_DOMAINS
:
2917 case LTTNG_START_TRACE
:
2918 case LTTNG_STOP_TRACE
:
2919 case LTTNG_DATA_PENDING
:
2920 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2921 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2922 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2923 case LTTNG_SNAPSHOT_RECORD
:
2924 case LTTNG_SAVE_SESSION
:
2925 case LTTNG_SET_SESSION_SHM_PATH
:
2932 if (opt_no_kernel
&& need_domain
2933 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2935 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2937 ret
= LTTNG_ERR_KERN_NA
;
2942 /* Deny register consumer if we already have a spawned consumer. */
2943 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2944 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2945 if (kconsumer_data
.pid
> 0) {
2946 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2947 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2950 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2954 * Check for command that don't needs to allocate a returned payload. We do
2955 * this here so we don't have to make the call for no payload at each
2958 switch(cmd_ctx
->lsm
->cmd_type
) {
2959 case LTTNG_LIST_SESSIONS
:
2960 case LTTNG_LIST_TRACEPOINTS
:
2961 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2962 case LTTNG_LIST_DOMAINS
:
2963 case LTTNG_LIST_CHANNELS
:
2964 case LTTNG_LIST_EVENTS
:
2965 case LTTNG_LIST_SYSCALLS
:
2966 case LTTNG_LIST_TRACKER_PIDS
:
2969 /* Setup lttng message with no payload */
2970 ret
= setup_lttng_msg(cmd_ctx
, 0);
2972 /* This label does not try to unlock the session */
2973 goto init_setup_error
;
2977 /* Commands that DO NOT need a session. */
2978 switch (cmd_ctx
->lsm
->cmd_type
) {
2979 case LTTNG_CREATE_SESSION
:
2980 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2981 case LTTNG_CREATE_SESSION_LIVE
:
2982 case LTTNG_CALIBRATE
:
2983 case LTTNG_LIST_SESSIONS
:
2984 case LTTNG_LIST_TRACEPOINTS
:
2985 case LTTNG_LIST_SYSCALLS
:
2986 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2987 case LTTNG_SAVE_SESSION
:
2988 need_tracing_session
= 0;
2991 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2993 * We keep the session list lock across _all_ commands
2994 * for now, because the per-session lock does not
2995 * handle teardown properly.
2997 session_lock_list();
2998 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2999 if (cmd_ctx
->session
== NULL
) {
3000 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3003 /* Acquire lock for the session */
3004 session_lock(cmd_ctx
->session
);
3010 * Commands that need a valid session but should NOT create one if none
3011 * exists. Instead of creating one and destroying it when the command is
3012 * handled, process that right before so we save some round trip in useless
3015 switch (cmd_ctx
->lsm
->cmd_type
) {
3016 case LTTNG_DISABLE_CHANNEL
:
3017 case LTTNG_DISABLE_EVENT
:
3018 switch (cmd_ctx
->lsm
->domain
.type
) {
3019 case LTTNG_DOMAIN_KERNEL
:
3020 if (!cmd_ctx
->session
->kernel_session
) {
3021 ret
= LTTNG_ERR_NO_CHANNEL
;
3025 case LTTNG_DOMAIN_JUL
:
3026 case LTTNG_DOMAIN_LOG4J
:
3027 case LTTNG_DOMAIN_PYTHON
:
3028 case LTTNG_DOMAIN_UST
:
3029 if (!cmd_ctx
->session
->ust_session
) {
3030 ret
= LTTNG_ERR_NO_CHANNEL
;
3035 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3047 * Check domain type for specific "pre-action".
3049 switch (cmd_ctx
->lsm
->domain
.type
) {
3050 case LTTNG_DOMAIN_KERNEL
:
3052 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3056 /* Kernel tracer check */
3057 if (kernel_tracer_fd
== -1) {
3058 /* Basically, load kernel tracer modules */
3059 ret
= init_kernel_tracer();
3065 /* Consumer is in an ERROR state. Report back to client */
3066 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3067 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3071 /* Need a session for kernel command */
3072 if (need_tracing_session
) {
3073 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3074 ret
= create_kernel_session(cmd_ctx
->session
);
3076 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3081 /* Start the kernel consumer daemon */
3082 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3083 if (kconsumer_data
.pid
== 0 &&
3084 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3085 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3086 ret
= start_consumerd(&kconsumer_data
);
3088 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3091 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3093 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3097 * The consumer was just spawned so we need to add the socket to
3098 * the consumer output of the session if exist.
3100 ret
= consumer_create_socket(&kconsumer_data
,
3101 cmd_ctx
->session
->kernel_session
->consumer
);
3108 case LTTNG_DOMAIN_JUL
:
3109 case LTTNG_DOMAIN_LOG4J
:
3110 case LTTNG_DOMAIN_PYTHON
:
3111 case LTTNG_DOMAIN_UST
:
3113 if (!ust_app_supported()) {
3114 ret
= LTTNG_ERR_NO_UST
;
3117 /* Consumer is in an ERROR state. Report back to client */
3118 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3119 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3123 if (need_tracing_session
) {
3124 /* Create UST session if none exist. */
3125 if (cmd_ctx
->session
->ust_session
== NULL
) {
3126 ret
= create_ust_session(cmd_ctx
->session
,
3127 &cmd_ctx
->lsm
->domain
);
3128 if (ret
!= LTTNG_OK
) {
3133 /* Start the UST consumer daemons */
3135 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3136 if (consumerd64_bin
[0] != '\0' &&
3137 ustconsumer64_data
.pid
== 0 &&
3138 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3139 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3140 ret
= start_consumerd(&ustconsumer64_data
);
3142 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3143 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3147 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3148 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3150 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3154 * Setup socket for consumer 64 bit. No need for atomic access
3155 * since it was set above and can ONLY be set in this thread.
3157 ret
= consumer_create_socket(&ustconsumer64_data
,
3158 cmd_ctx
->session
->ust_session
->consumer
);
3164 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3165 if (consumerd32_bin
[0] != '\0' &&
3166 ustconsumer32_data
.pid
== 0 &&
3167 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3168 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3169 ret
= start_consumerd(&ustconsumer32_data
);
3171 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3172 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3176 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3177 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3179 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3183 * Setup socket for consumer 64 bit. No need for atomic access
3184 * since it was set above and can ONLY be set in this thread.
3186 ret
= consumer_create_socket(&ustconsumer32_data
,
3187 cmd_ctx
->session
->ust_session
->consumer
);
3199 /* Validate consumer daemon state when start/stop trace command */
3200 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3201 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3202 switch (cmd_ctx
->lsm
->domain
.type
) {
3203 case LTTNG_DOMAIN_JUL
:
3204 case LTTNG_DOMAIN_LOG4J
:
3205 case LTTNG_DOMAIN_PYTHON
:
3206 case LTTNG_DOMAIN_UST
:
3207 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3208 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3212 case LTTNG_DOMAIN_KERNEL
:
3213 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3214 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3222 * Check that the UID or GID match that of the tracing session.
3223 * The root user can interact with all sessions.
3225 if (need_tracing_session
) {
3226 if (!session_access_ok(cmd_ctx
->session
,
3227 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3228 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3229 ret
= LTTNG_ERR_EPERM
;
3235 * Send relayd information to consumer as soon as we have a domain and a
3238 if (cmd_ctx
->session
&& need_domain
) {
3240 * Setup relayd if not done yet. If the relayd information was already
3241 * sent to the consumer, this call will gracefully return.
3243 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3244 if (ret
!= LTTNG_OK
) {
3249 /* Process by command type */
3250 switch (cmd_ctx
->lsm
->cmd_type
) {
3251 case LTTNG_ADD_CONTEXT
:
3253 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3254 cmd_ctx
->lsm
->u
.context
.channel_name
,
3255 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3258 case LTTNG_DISABLE_CHANNEL
:
3260 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3261 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3264 case LTTNG_DISABLE_EVENT
:
3266 /* FIXME: passing packed structure to non-packed pointer */
3267 /* TODO: handle filter */
3268 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3269 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3270 &cmd_ctx
->lsm
->u
.disable
.event
);
3273 case LTTNG_ENABLE_CHANNEL
:
3275 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3276 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3279 case LTTNG_TRACK_PID
:
3281 ret
= cmd_track_pid(cmd_ctx
->session
,
3282 cmd_ctx
->lsm
->domain
.type
,
3283 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3286 case LTTNG_UNTRACK_PID
:
3288 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3289 cmd_ctx
->lsm
->domain
.type
,
3290 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3293 case LTTNG_ENABLE_EVENT
:
3295 struct lttng_event_exclusion
*exclusion
= NULL
;
3296 struct lttng_filter_bytecode
*bytecode
= NULL
;
3297 char *filter_expression
= NULL
;
3299 /* Handle exclusion events and receive it from the client. */
3300 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3301 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3303 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3304 (count
* LTTNG_SYMBOL_NAME_LEN
));
3306 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3310 DBG("Receiving var len exclusion event list from client ...");
3311 exclusion
->count
= count
;
3312 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3313 count
* LTTNG_SYMBOL_NAME_LEN
);
3315 DBG("Nothing recv() from client var len data... continuing");
3318 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3323 /* Get filter expression from client. */
3324 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3325 size_t expression_len
=
3326 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3328 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3329 ret
= LTTNG_ERR_FILTER_INVAL
;
3334 filter_expression
= zmalloc(expression_len
);
3335 if (!filter_expression
) {
3337 ret
= LTTNG_ERR_FILTER_NOMEM
;
3341 /* Receive var. len. data */
3342 DBG("Receiving var len filter's expression from client ...");
3343 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3346 DBG("Nothing recv() from client car len data... continuing");
3348 free(filter_expression
);
3350 ret
= LTTNG_ERR_FILTER_INVAL
;
3355 /* Handle filter and get bytecode from client. */
3356 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3357 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3359 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3360 ret
= LTTNG_ERR_FILTER_INVAL
;
3361 free(filter_expression
);
3366 bytecode
= zmalloc(bytecode_len
);
3368 free(filter_expression
);
3370 ret
= LTTNG_ERR_FILTER_NOMEM
;
3374 /* Receive var. len. data */
3375 DBG("Receiving var len filter's bytecode from client ...");
3376 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3378 DBG("Nothing recv() from client car len data... continuing");
3380 free(filter_expression
);
3383 ret
= LTTNG_ERR_FILTER_INVAL
;
3387 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3388 free(filter_expression
);
3391 ret
= LTTNG_ERR_FILTER_INVAL
;
3396 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3397 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3398 &cmd_ctx
->lsm
->u
.enable
.event
,
3399 filter_expression
, bytecode
, exclusion
,
3400 kernel_poll_pipe
[1]);
3403 case LTTNG_LIST_TRACEPOINTS
:
3405 struct lttng_event
*events
;
3408 session_lock_list();
3409 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3410 session_unlock_list();
3411 if (nb_events
< 0) {
3412 /* Return value is a negative lttng_error_code. */
3418 * Setup lttng message with payload size set to the event list size in
3419 * bytes and then copy list into the llm payload.
3421 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3427 /* Copy event list into message payload */
3428 memcpy(cmd_ctx
->llm
->payload
, events
,
3429 sizeof(struct lttng_event
) * nb_events
);
3436 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3438 struct lttng_event_field
*fields
;
3441 session_lock_list();
3442 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3444 session_unlock_list();
3445 if (nb_fields
< 0) {
3446 /* Return value is a negative lttng_error_code. */
3452 * Setup lttng message with payload size set to the event list size in
3453 * bytes and then copy list into the llm payload.
3455 ret
= setup_lttng_msg(cmd_ctx
,
3456 sizeof(struct lttng_event_field
) * nb_fields
);
3462 /* Copy event list into message payload */
3463 memcpy(cmd_ctx
->llm
->payload
, fields
,
3464 sizeof(struct lttng_event_field
) * nb_fields
);
3471 case LTTNG_LIST_SYSCALLS
:
3473 struct lttng_event
*events
;
3476 nb_events
= cmd_list_syscalls(&events
);
3477 if (nb_events
< 0) {
3478 /* Return value is a negative lttng_error_code. */
3484 * Setup lttng message with payload size set to the event list size in
3485 * bytes and then copy list into the llm payload.
3487 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3493 /* Copy event list into message payload */
3494 memcpy(cmd_ctx
->llm
->payload
, events
,
3495 sizeof(struct lttng_event
) * nb_events
);
3502 case LTTNG_LIST_TRACKER_PIDS
:
3504 int32_t *pids
= NULL
;
3507 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3508 cmd_ctx
->lsm
->domain
.type
, &pids
);
3510 /* Return value is a negative lttng_error_code. */
3516 * Setup lttng message with payload size set to the event list size in
3517 * bytes and then copy list into the llm payload.
3519 ret
= setup_lttng_msg(cmd_ctx
, sizeof(int32_t) * nr_pids
);
3525 /* Copy event list into message payload */
3526 memcpy(cmd_ctx
->llm
->payload
, pids
,
3527 sizeof(int) * nr_pids
);
3534 case LTTNG_SET_CONSUMER_URI
:
3537 struct lttng_uri
*uris
;
3539 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3540 len
= nb_uri
* sizeof(struct lttng_uri
);
3543 ret
= LTTNG_ERR_INVALID
;
3547 uris
= zmalloc(len
);
3549 ret
= LTTNG_ERR_FATAL
;
3553 /* Receive variable len data */
3554 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3555 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3557 DBG("No URIs received from client... continuing");
3559 ret
= LTTNG_ERR_SESSION_FAIL
;
3564 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3566 if (ret
!= LTTNG_OK
) {
3573 case LTTNG_START_TRACE
:
3575 ret
= cmd_start_trace(cmd_ctx
->session
);
3578 case LTTNG_STOP_TRACE
:
3580 ret
= cmd_stop_trace(cmd_ctx
->session
);
3583 case LTTNG_CREATE_SESSION
:
3586 struct lttng_uri
*uris
= NULL
;
3588 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3589 len
= nb_uri
* sizeof(struct lttng_uri
);
3592 uris
= zmalloc(len
);
3594 ret
= LTTNG_ERR_FATAL
;
3598 /* Receive variable len data */
3599 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3600 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3602 DBG("No URIs received from client... continuing");
3604 ret
= LTTNG_ERR_SESSION_FAIL
;
3609 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3610 DBG("Creating session with ONE network URI is a bad call");
3611 ret
= LTTNG_ERR_SESSION_FAIL
;
3617 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3618 &cmd_ctx
->creds
, 0);
3624 case LTTNG_DESTROY_SESSION
:
3626 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3628 /* Set session to NULL so we do not unlock it after free. */
3629 cmd_ctx
->session
= NULL
;
3632 case LTTNG_LIST_DOMAINS
:
3635 struct lttng_domain
*domains
= NULL
;
3637 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3639 /* Return value is a negative lttng_error_code. */
3644 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3650 /* Copy event list into message payload */
3651 memcpy(cmd_ctx
->llm
->payload
, domains
,
3652 nb_dom
* sizeof(struct lttng_domain
));
3659 case LTTNG_LIST_CHANNELS
:
3662 struct lttng_channel
*channels
= NULL
;
3664 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3665 cmd_ctx
->session
, &channels
);
3667 /* Return value is a negative lttng_error_code. */
3672 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3678 /* Copy event list into message payload */
3679 memcpy(cmd_ctx
->llm
->payload
, channels
,
3680 nb_chan
* sizeof(struct lttng_channel
));
3687 case LTTNG_LIST_EVENTS
:
3690 struct lttng_event
*events
= NULL
;
3692 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3693 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3695 /* Return value is a negative lttng_error_code. */
3700 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3706 /* Copy event list into message payload */
3707 memcpy(cmd_ctx
->llm
->payload
, events
,
3708 nb_event
* sizeof(struct lttng_event
));
3715 case LTTNG_LIST_SESSIONS
:
3717 unsigned int nr_sessions
;
3719 session_lock_list();
3720 nr_sessions
= lttng_sessions_count(
3721 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3722 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3724 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3726 session_unlock_list();
3730 /* Filled the session array */
3731 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3732 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3733 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3735 session_unlock_list();
3740 case LTTNG_CALIBRATE
:
3742 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3743 &cmd_ctx
->lsm
->u
.calibrate
);
3746 case LTTNG_REGISTER_CONSUMER
:
3748 struct consumer_data
*cdata
;
3750 switch (cmd_ctx
->lsm
->domain
.type
) {
3751 case LTTNG_DOMAIN_KERNEL
:
3752 cdata
= &kconsumer_data
;
3755 ret
= LTTNG_ERR_UND
;
3759 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3760 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3763 case LTTNG_DATA_PENDING
:
3765 ret
= cmd_data_pending(cmd_ctx
->session
);
3768 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3770 struct lttcomm_lttng_output_id reply
;
3772 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3773 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3774 if (ret
!= LTTNG_OK
) {
3778 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3783 /* Copy output list into message payload */
3784 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3788 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3790 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3791 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3794 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3797 struct lttng_snapshot_output
*outputs
= NULL
;
3799 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3800 if (nb_output
< 0) {
3805 ret
= setup_lttng_msg(cmd_ctx
,
3806 nb_output
* sizeof(struct lttng_snapshot_output
));
3813 /* Copy output list into message payload */
3814 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3815 nb_output
* sizeof(struct lttng_snapshot_output
));
3822 case LTTNG_SNAPSHOT_RECORD
:
3824 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3825 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3826 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3829 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3832 struct lttng_uri
*uris
= NULL
;
3834 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3835 len
= nb_uri
* sizeof(struct lttng_uri
);
3838 uris
= zmalloc(len
);
3840 ret
= LTTNG_ERR_FATAL
;
3844 /* Receive variable len data */
3845 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3846 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3848 DBG("No URIs received from client... continuing");
3850 ret
= LTTNG_ERR_SESSION_FAIL
;
3855 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3856 DBG("Creating session with ONE network URI is a bad call");
3857 ret
= LTTNG_ERR_SESSION_FAIL
;
3863 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3864 nb_uri
, &cmd_ctx
->creds
);
3868 case LTTNG_CREATE_SESSION_LIVE
:
3871 struct lttng_uri
*uris
= NULL
;
3873 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3874 len
= nb_uri
* sizeof(struct lttng_uri
);
3877 uris
= zmalloc(len
);
3879 ret
= LTTNG_ERR_FATAL
;
3883 /* Receive variable len data */
3884 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3885 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3887 DBG("No URIs received from client... continuing");
3889 ret
= LTTNG_ERR_SESSION_FAIL
;
3894 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3895 DBG("Creating session with ONE network URI is a bad call");
3896 ret
= LTTNG_ERR_SESSION_FAIL
;
3902 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3903 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3907 case LTTNG_SAVE_SESSION
:
3909 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
3913 case LTTNG_SET_SESSION_SHM_PATH
:
3915 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
3916 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
3920 ret
= LTTNG_ERR_UND
;
3925 if (cmd_ctx
->llm
== NULL
) {
3926 DBG("Missing llm structure. Allocating one.");
3927 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3931 /* Set return code */
3932 cmd_ctx
->llm
->ret_code
= ret
;
3934 if (cmd_ctx
->session
) {
3935 session_unlock(cmd_ctx
->session
);
3937 if (need_tracing_session
) {
3938 session_unlock_list();
3945 * Thread managing health check socket.
3947 static void *thread_manage_health(void *data
)
3949 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3950 uint32_t revents
, nb_fd
;
3951 struct lttng_poll_event events
;
3952 struct health_comm_msg msg
;
3953 struct health_comm_reply reply
;
3955 DBG("[thread] Manage health check started");
3957 rcu_register_thread();
3959 /* We might hit an error path before this is created. */
3960 lttng_poll_init(&events
);
3962 /* Create unix socket */
3963 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3965 ERR("Unable to create health check Unix socket");
3971 /* lttng health client socket path permissions */
3972 ret
= chown(health_unix_sock_path
, 0,
3973 utils_get_group_id(tracing_group_name
));
3975 ERR("Unable to set group on %s", health_unix_sock_path
);
3981 ret
= chmod(health_unix_sock_path
,
3982 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3984 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3992 * Set the CLOEXEC flag. Return code is useless because either way, the
3995 (void) utils_set_fd_cloexec(sock
);
3997 ret
= lttcomm_listen_unix_sock(sock
);
4003 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4004 * more will be added to this poll set.
4006 ret
= sessiond_set_thread_pollset(&events
, 2);
4011 /* Add the application registration socket */
4012 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4017 sessiond_notify_ready();
4020 DBG("Health check ready");
4022 /* Inifinite blocking call, waiting for transmission */
4024 ret
= lttng_poll_wait(&events
, -1);
4027 * Restart interrupted system call.
4029 if (errno
== EINTR
) {
4037 for (i
= 0; i
< nb_fd
; i
++) {
4038 /* Fetch once the poll data */
4039 revents
= LTTNG_POLL_GETEV(&events
, i
);
4040 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4043 /* No activity for this FD (poll implementation). */
4047 /* Thread quit pipe has been closed. Killing thread. */
4048 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4054 /* Event on the registration socket */
4055 if (pollfd
== sock
) {
4056 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4057 ERR("Health socket poll error");
4063 new_sock
= lttcomm_accept_unix_sock(sock
);
4069 * Set the CLOEXEC flag. Return code is useless because either way, the
4072 (void) utils_set_fd_cloexec(new_sock
);
4074 DBG("Receiving data from client for health...");
4075 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4077 DBG("Nothing recv() from client... continuing");
4078 ret
= close(new_sock
);
4086 rcu_thread_online();
4088 memset(&reply
, 0, sizeof(reply
));
4089 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4091 * health_check_state returns 0 if health is
4094 if (!health_check_state(health_sessiond
, i
)) {
4095 reply
.ret_code
|= 1ULL << i
;
4099 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4101 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4103 ERR("Failed to send health data back to client");
4106 /* End of transmission */
4107 ret
= close(new_sock
);
4117 ERR("Health error occurred in %s", __func__
);
4119 DBG("Health check thread dying");
4120 unlink(health_unix_sock_path
);
4128 lttng_poll_clean(&events
);
4130 rcu_unregister_thread();
4135 * This thread manage all clients request using the unix client socket for
4138 static void *thread_manage_clients(void *data
)
4140 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4142 uint32_t revents
, nb_fd
;
4143 struct command_ctx
*cmd_ctx
= NULL
;
4144 struct lttng_poll_event events
;
4146 DBG("[thread] Manage client started");
4148 rcu_register_thread();
4150 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4152 health_code_update();
4154 ret
= lttcomm_listen_unix_sock(client_sock
);
4160 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4161 * more will be added to this poll set.
4163 ret
= sessiond_set_thread_pollset(&events
, 2);
4165 goto error_create_poll
;
4168 /* Add the application registration socket */
4169 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4174 sessiond_notify_ready();
4175 ret
= sem_post(&load_info
->message_thread_ready
);
4177 PERROR("sem_post message_thread_ready");
4181 /* This testpoint is after we signal readiness to the parent. */
4182 if (testpoint(sessiond_thread_manage_clients
)) {
4186 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4190 health_code_update();
4193 DBG("Accepting client command ...");
4195 /* Inifinite blocking call, waiting for transmission */
4197 health_poll_entry();
4198 ret
= lttng_poll_wait(&events
, -1);
4202 * Restart interrupted system call.
4204 if (errno
== EINTR
) {
4212 for (i
= 0; i
< nb_fd
; i
++) {
4213 /* Fetch once the poll data */
4214 revents
= LTTNG_POLL_GETEV(&events
, i
);
4215 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4217 health_code_update();
4220 /* No activity for this FD (poll implementation). */
4224 /* Thread quit pipe has been closed. Killing thread. */
4225 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4231 /* Event on the registration socket */
4232 if (pollfd
== client_sock
) {
4233 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4234 ERR("Client socket poll error");
4240 DBG("Wait for client response");
4242 health_code_update();
4244 sock
= lttcomm_accept_unix_sock(client_sock
);
4250 * Set the CLOEXEC flag. Return code is useless because either way, the
4253 (void) utils_set_fd_cloexec(sock
);
4255 /* Set socket option for credentials retrieval */
4256 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4261 /* Allocate context command to process the client request */
4262 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4263 if (cmd_ctx
== NULL
) {
4264 PERROR("zmalloc cmd_ctx");
4268 /* Allocate data buffer for reception */
4269 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4270 if (cmd_ctx
->lsm
== NULL
) {
4271 PERROR("zmalloc cmd_ctx->lsm");
4275 cmd_ctx
->llm
= NULL
;
4276 cmd_ctx
->session
= NULL
;
4278 health_code_update();
4281 * Data is received from the lttng client. The struct
4282 * lttcomm_session_msg (lsm) contains the command and data request of
4285 DBG("Receiving data from client ...");
4286 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4287 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4289 DBG("Nothing recv() from client... continuing");
4295 clean_command_ctx(&cmd_ctx
);
4299 health_code_update();
4301 // TODO: Validate cmd_ctx including sanity check for
4302 // security purpose.
4304 rcu_thread_online();
4306 * This function dispatch the work to the kernel or userspace tracer
4307 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4308 * informations for the client. The command context struct contains
4309 * everything this function may needs.
4311 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4312 rcu_thread_offline();
4320 * TODO: Inform client somehow of the fatal error. At
4321 * this point, ret < 0 means that a zmalloc failed
4322 * (ENOMEM). Error detected but still accept
4323 * command, unless a socket error has been
4326 clean_command_ctx(&cmd_ctx
);
4330 health_code_update();
4332 DBG("Sending response (size: %d, retcode: %s)",
4333 cmd_ctx
->lttng_msg_size
,
4334 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4335 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4337 ERR("Failed to send data back to client");
4340 /* End of transmission */
4347 clean_command_ctx(&cmd_ctx
);
4349 health_code_update();
4361 lttng_poll_clean(&events
);
4362 clean_command_ctx(&cmd_ctx
);
4366 unlink(client_unix_sock_path
);
4367 if (client_sock
>= 0) {
4368 ret
= close(client_sock
);
4376 ERR("Health error occurred in %s", __func__
);
4379 health_unregister(health_sessiond
);
4381 DBG("Client thread dying");
4383 rcu_unregister_thread();
4386 * Since we are creating the consumer threads, we own them, so we need
4387 * to join them before our thread exits.
4389 ret
= join_consumer_thread(&kconsumer_data
);
4392 PERROR("join_consumer");
4395 ret
= join_consumer_thread(&ustconsumer32_data
);
4398 PERROR("join_consumer ust32");
4401 ret
= join_consumer_thread(&ustconsumer64_data
);
4404 PERROR("join_consumer ust64");
4411 * usage function on stderr
4413 static void usage(void)
4415 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4416 fprintf(stderr
, " -h, --help Display this usage.\n");
4417 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4418 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4419 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4420 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4421 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4422 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4423 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4424 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4425 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4426 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4427 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4428 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4429 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4430 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4431 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4432 fprintf(stderr
, " -V, --version Show version number.\n");
4433 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4434 fprintf(stderr
, " -q, --quiet No output at all.\n");
4435 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4436 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4437 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4438 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4439 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4440 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4441 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4442 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4443 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4447 * Take an option from the getopt output and set it in the right variable to be
4450 * Return 0 on success else a negative value.
4452 static int set_option(int opt
, const char *arg
, const char *optname
)
4456 if (arg
&& arg
[0] == '\0') {
4458 * This only happens if the value is read from daemon config
4459 * file. This means the option requires an argument and the
4460 * configuration file contains a line such as:
4469 fprintf(stderr
, "option %s", optname
);
4471 fprintf(stderr
, " with arg %s\n", arg
);
4475 if (lttng_is_setuid_setgid()) {
4476 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4477 "-c, --client-sock");
4479 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4483 if (lttng_is_setuid_setgid()) {
4484 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4487 snprintf(apps_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.",
4502 * If the override option is set, the pointer points to a
4503 * *non* const thus freeing it even though the variable type is
4506 if (tracing_group_name_override
) {
4507 free((void *) tracing_group_name
);
4509 tracing_group_name
= strdup(arg
);
4510 if (!tracing_group_name
) {
4514 tracing_group_name_override
= 1;
4521 fprintf(stdout
, "%s\n", VERSION
);
4527 if (lttng_is_setuid_setgid()) {
4528 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4529 "--kconsumerd-err-sock");
4531 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4535 if (lttng_is_setuid_setgid()) {
4536 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4537 "--kconsumerd-cmd-sock");
4539 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4543 if (lttng_is_setuid_setgid()) {
4544 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4545 "--ustconsumerd64-err-sock");
4547 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4551 if (lttng_is_setuid_setgid()) {
4552 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4553 "--ustconsumerd64-cmd-sock");
4555 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4559 if (lttng_is_setuid_setgid()) {
4560 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4561 "--ustconsumerd32-err-sock");
4563 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4567 if (lttng_is_setuid_setgid()) {
4568 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4569 "--ustconsumerd32-cmd-sock");
4571 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4578 lttng_opt_quiet
= 1;
4581 /* Verbose level can increase using multiple -v */
4583 /* Value obtained from config file */
4584 lttng_opt_verbose
= config_parse_value(arg
);
4586 /* -v used on command line */
4587 lttng_opt_verbose
++;
4589 /* Clamp value to [0, 3] */
4590 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4591 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4595 opt_verbose_consumer
= config_parse_value(arg
);
4597 opt_verbose_consumer
+= 1;
4601 if (lttng_is_setuid_setgid()) {
4602 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4603 "--consumerd32-path");
4605 if (consumerd32_bin_override
) {
4606 free((void *) consumerd32_bin
);
4608 consumerd32_bin
= strdup(arg
);
4609 if (!consumerd32_bin
) {
4613 consumerd32_bin_override
= 1;
4617 if (lttng_is_setuid_setgid()) {
4618 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4619 "--consumerd32-libdir");
4621 if (consumerd32_libdir_override
) {
4622 free((void *) consumerd32_libdir
);
4624 consumerd32_libdir
= strdup(arg
);
4625 if (!consumerd32_libdir
) {
4629 consumerd32_libdir_override
= 1;
4633 if (lttng_is_setuid_setgid()) {
4634 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4635 "--consumerd64-path");
4637 if (consumerd64_bin_override
) {
4638 free((void *) consumerd64_bin
);
4640 consumerd64_bin
= strdup(arg
);
4641 if (!consumerd64_bin
) {
4645 consumerd64_bin_override
= 1;
4649 if (lttng_is_setuid_setgid()) {
4650 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4651 "--consumerd64-libdir");
4653 if (consumerd64_libdir_override
) {
4654 free((void *) consumerd64_libdir
);
4656 consumerd64_libdir
= strdup(arg
);
4657 if (!consumerd64_libdir
) {
4661 consumerd64_libdir_override
= 1;
4665 if (lttng_is_setuid_setgid()) {
4666 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4670 opt_pidfile
= strdup(arg
);
4677 case 'J': /* Agent TCP port. */
4679 if (lttng_is_setuid_setgid()) {
4680 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4681 "--agent-tcp-port");
4690 v
= strtoul(arg
, NULL
, 0);
4691 if (errno
!= 0 || !isdigit(arg
[0])) {
4692 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4695 if (v
== 0 || v
>= 65535) {
4696 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4699 agent_tcp_port
= (uint32_t) v
;
4700 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4705 if (lttng_is_setuid_setgid()) {
4706 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4709 free(opt_load_session_path
);
4710 opt_load_session_path
= strdup(arg
);
4711 if (!opt_load_session_path
) {
4717 case 'P': /* probe modules list */
4718 if (lttng_is_setuid_setgid()) {
4719 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4722 free(kmod_probes_list
);
4723 kmod_probes_list
= strdup(arg
);
4724 if (!kmod_probes_list
) {
4731 if (lttng_is_setuid_setgid()) {
4732 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4733 "--extra-kmod-probes");
4735 free(kmod_extra_probes_list
);
4736 kmod_extra_probes_list
= strdup(arg
);
4737 if (!kmod_extra_probes_list
) {
4744 /* This is handled in set_options() thus silent break. */
4747 /* Unknown option or other error.
4748 * Error is printed by getopt, just return */
4753 if (ret
== -EINVAL
) {
4754 const char *opt_name
= "unknown";
4757 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4759 if (opt
== long_options
[i
].val
) {
4760 opt_name
= long_options
[i
].name
;
4765 WARN("Invalid argument provided for option \"%s\", using default value.",
4773 * config_entry_handler_cb used to handle options read from a config file.
4774 * See config_entry_handler_cb comment in common/config/config.h for the
4775 * return value conventions.
4777 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4781 if (!entry
|| !entry
->name
|| !entry
->value
) {
4786 /* Check if the option is to be ignored */
4787 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4788 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4793 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4796 /* Ignore if not fully matched. */
4797 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4802 * If the option takes no argument on the command line, we have to
4803 * check if the value is "true". We support non-zero numeric values,
4806 if (!long_options
[i
].has_arg
) {
4807 ret
= config_parse_value(entry
->value
);
4810 WARN("Invalid configuration value \"%s\" for option %s",
4811 entry
->value
, entry
->name
);
4813 /* False, skip boolean config option. */
4818 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4822 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4829 * daemon configuration loading and argument parsing
4831 static int set_options(int argc
, char **argv
)
4833 int ret
= 0, c
= 0, option_index
= 0;
4834 int orig_optopt
= optopt
, orig_optind
= optind
;
4836 const char *config_path
= NULL
;
4838 optstring
= utils_generate_optstring(long_options
,
4839 sizeof(long_options
) / sizeof(struct option
));
4845 /* Check for the --config option */
4846 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4847 &option_index
)) != -1) {
4851 } else if (c
!= 'f') {
4852 /* if not equal to --config option. */
4856 if (lttng_is_setuid_setgid()) {
4857 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4860 config_path
= utils_expand_path(optarg
);
4862 ERR("Failed to resolve path: %s", optarg
);
4867 ret
= config_get_section_entries(config_path
, config_section_name
,
4868 config_entry_handler
, NULL
);
4871 ERR("Invalid configuration option at line %i", ret
);
4877 /* Reset getopt's global state */
4878 optopt
= orig_optopt
;
4879 optind
= orig_optind
;
4881 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
4886 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
4898 * Creates the two needed socket by the daemon.
4899 * apps_sock - The communication socket for all UST apps.
4900 * client_sock - The communication of the cli tool (lttng).
4902 static int init_daemon_socket(void)
4907 old_umask
= umask(0);
4909 /* Create client tool unix socket */
4910 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4911 if (client_sock
< 0) {
4912 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4917 /* Set the cloexec flag */
4918 ret
= utils_set_fd_cloexec(client_sock
);
4920 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4921 "Continuing but note that the consumer daemon will have a "
4922 "reference to this socket on exec()", client_sock
);
4925 /* File permission MUST be 660 */
4926 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4928 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4933 /* Create the application unix socket */
4934 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4935 if (apps_sock
< 0) {
4936 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4941 /* Set the cloexec flag */
4942 ret
= utils_set_fd_cloexec(apps_sock
);
4944 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4945 "Continuing but note that the consumer daemon will have a "
4946 "reference to this socket on exec()", apps_sock
);
4949 /* File permission MUST be 666 */
4950 ret
= chmod(apps_unix_sock_path
,
4951 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4953 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4958 DBG3("Session daemon client socket %d and application socket %d created",
4959 client_sock
, apps_sock
);
4967 * Check if the global socket is available, and if a daemon is answering at the
4968 * other side. If yes, error is returned.
4970 static int check_existing_daemon(void)
4972 /* Is there anybody out there ? */
4973 if (lttng_session_daemon_alive()) {
4981 * Set the tracing group gid onto the client socket.
4983 * Race window between mkdir and chown is OK because we are going from more
4984 * permissive (root.root) to less permissive (root.tracing).
4986 static int set_permissions(char *rundir
)
4991 gid
= utils_get_group_id(tracing_group_name
);
4993 /* Set lttng run dir */
4994 ret
= chown(rundir
, 0, gid
);
4996 ERR("Unable to set group on %s", rundir
);
5001 * Ensure all applications and tracing group can search the run
5002 * dir. Allow everyone to read the directory, since it does not
5003 * buy us anything to hide its content.
5005 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5007 ERR("Unable to set permissions on %s", rundir
);
5011 /* lttng client socket path */
5012 ret
= chown(client_unix_sock_path
, 0, gid
);
5014 ERR("Unable to set group on %s", client_unix_sock_path
);
5018 /* kconsumer error socket path */
5019 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5021 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5025 /* 64-bit ustconsumer error socket path */
5026 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5028 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5032 /* 32-bit ustconsumer compat32 error socket path */
5033 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5035 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5039 DBG("All permissions are set");
5045 * Create the lttng run directory needed for all global sockets and pipe.
5047 static int create_lttng_rundir(const char *rundir
)
5051 DBG3("Creating LTTng run directory: %s", rundir
);
5053 ret
= mkdir(rundir
, S_IRWXU
);
5055 if (errno
!= EEXIST
) {
5056 ERR("Unable to create %s", rundir
);
5068 * Setup sockets and directory needed by the kconsumerd communication with the
5071 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5075 char path
[PATH_MAX
];
5077 switch (consumer_data
->type
) {
5078 case LTTNG_CONSUMER_KERNEL
:
5079 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5081 case LTTNG_CONSUMER64_UST
:
5082 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5084 case LTTNG_CONSUMER32_UST
:
5085 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5088 ERR("Consumer type unknown");
5093 DBG2("Creating consumer directory: %s", path
);
5095 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5097 if (errno
!= EEXIST
) {
5099 ERR("Failed to create %s", path
);
5105 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5107 ERR("Unable to set group on %s", path
);
5113 /* Create the kconsumerd error unix socket */
5114 consumer_data
->err_sock
=
5115 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5116 if (consumer_data
->err_sock
< 0) {
5117 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5123 * Set the CLOEXEC flag. Return code is useless because either way, the
5126 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5128 PERROR("utils_set_fd_cloexec");
5129 /* continue anyway */
5132 /* File permission MUST be 660 */
5133 ret
= chmod(consumer_data
->err_unix_sock_path
,
5134 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5136 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5146 * Signal handler for the daemon
5148 * Simply stop all worker threads, leaving main() return gracefully after
5149 * joining all threads and calling cleanup().
5151 static void sighandler(int sig
)
5155 DBG("SIGPIPE caught");
5158 DBG("SIGINT caught");
5162 DBG("SIGTERM caught");
5166 CMM_STORE_SHARED(recv_child_signal
, 1);
5174 * Setup signal handler for :
5175 * SIGINT, SIGTERM, SIGPIPE
5177 static int set_signal_handler(void)
5180 struct sigaction sa
;
5183 if ((ret
= sigemptyset(&sigset
)) < 0) {
5184 PERROR("sigemptyset");
5188 sa
.sa_handler
= sighandler
;
5189 sa
.sa_mask
= sigset
;
5191 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5192 PERROR("sigaction");
5196 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5197 PERROR("sigaction");
5201 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5202 PERROR("sigaction");
5206 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5207 PERROR("sigaction");
5211 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5217 * Set open files limit to unlimited. This daemon can open a large number of
5218 * file descriptors in order to consumer multiple kernel traces.
5220 static void set_ulimit(void)
5225 /* The kernel does not allowed an infinite limit for open files */
5226 lim
.rlim_cur
= 65535;
5227 lim
.rlim_max
= 65535;
5229 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5231 PERROR("failed to set open files limit");
5236 * Write pidfile using the rundir and opt_pidfile.
5238 static int write_pidfile(void)
5241 char pidfile_path
[PATH_MAX
];
5246 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
5248 /* Build pidfile path from rundir and opt_pidfile. */
5249 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5250 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5252 PERROR("snprintf pidfile path");
5258 * Create pid file in rundir.
5260 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5266 * Create lockfile using the rundir and return its fd.
5268 static int create_lockfile(void)
5271 char lockfile_path
[PATH_MAX
];
5273 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5278 ret
= utils_create_lock_file(lockfile_path
);
5284 * Write agent TCP port using the rundir.
5286 static int write_agent_port(void)
5289 char path
[PATH_MAX
];
5293 ret
= snprintf(path
, sizeof(path
), "%s/"
5294 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5296 PERROR("snprintf agent port path");
5301 * Create TCP agent port file in rundir.
5303 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5312 int main(int argc
, char **argv
)
5314 int ret
= 0, retval
= 0;
5316 const char *home_path
, *env_app_timeout
;
5318 init_kernel_workarounds();
5320 rcu_register_thread();
5322 if (set_signal_handler()) {
5324 goto exit_set_signal_handler
;
5327 setup_consumerd_path();
5329 page_size
= sysconf(_SC_PAGESIZE
);
5330 if (page_size
< 0) {
5331 PERROR("sysconf _SC_PAGESIZE");
5332 page_size
= LONG_MAX
;
5333 WARN("Fallback page size to %ld", page_size
);
5337 * Parse arguments and load the daemon configuration file.
5339 * We have an exit_options exit path to free memory reserved by
5340 * set_options. This is needed because the rest of sessiond_cleanup()
5341 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5342 * depends on set_options.
5345 if (set_options(argc
, argv
)) {
5351 if (opt_daemon
|| opt_background
) {
5354 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5362 * We are in the child. Make sure all other file descriptors are
5363 * closed, in case we are called with more opened file
5364 * descriptors than the standard ones.
5366 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5372 * Starting from here, we can create threads. This needs to be after
5373 * lttng_daemonize due to RCU.
5377 * Initialize the health check subsystem. This call should set the
5378 * appropriate time values.
5380 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5381 if (!health_sessiond
) {
5382 PERROR("health_app_create error");
5384 goto exit_health_sessiond_cleanup
;
5387 if (init_ht_cleanup_quit_pipe()) {
5389 goto exit_ht_cleanup_quit_pipe
;
5392 /* Setup the thread ht_cleanup communication pipe. */
5393 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5395 goto exit_ht_cleanup_pipe
;
5398 /* Set up max poll set size */
5399 if (lttng_poll_set_max_size()) {
5401 goto exit_set_max_size
;
5404 /* Create thread to clean up RCU hash tables */
5405 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5406 thread_ht_cleanup
, (void *) NULL
);
5409 PERROR("pthread_create ht_cleanup");
5411 goto exit_ht_cleanup
;
5414 /* Create thread quit pipe */
5415 if (init_thread_quit_pipe()) {
5417 goto exit_init_data
;
5420 /* Check if daemon is UID = 0 */
5421 is_root
= !getuid();
5424 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5427 goto exit_init_data
;
5430 /* Create global run dir with root access */
5431 if (create_lttng_rundir(rundir
)) {
5433 goto exit_init_data
;
5436 if (strlen(apps_unix_sock_path
) == 0) {
5437 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5438 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5441 goto exit_init_data
;
5445 if (strlen(client_unix_sock_path
) == 0) {
5446 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5447 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5450 goto exit_init_data
;
5454 /* Set global SHM for ust */
5455 if (strlen(wait_shm_path
) == 0) {
5456 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5457 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5460 goto exit_init_data
;
5464 if (strlen(health_unix_sock_path
) == 0) {
5465 ret
= snprintf(health_unix_sock_path
,
5466 sizeof(health_unix_sock_path
),
5467 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5470 goto exit_init_data
;
5474 /* Setup kernel consumerd path */
5475 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5476 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5479 goto exit_init_data
;
5481 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5482 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5485 goto exit_init_data
;
5488 DBG2("Kernel consumer err path: %s",
5489 kconsumer_data
.err_unix_sock_path
);
5490 DBG2("Kernel consumer cmd path: %s",
5491 kconsumer_data
.cmd_unix_sock_path
);
5493 home_path
= utils_get_home_dir();
5494 if (home_path
== NULL
) {
5495 /* TODO: Add --socket PATH option */
5496 ERR("Can't get HOME directory for sockets creation.");
5498 goto exit_init_data
;
5502 * Create rundir from home path. This will create something like
5505 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5508 goto exit_init_data
;
5511 if (create_lttng_rundir(rundir
)) {
5513 goto exit_init_data
;
5516 if (strlen(apps_unix_sock_path
) == 0) {
5517 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5518 DEFAULT_HOME_APPS_UNIX_SOCK
,
5522 goto exit_init_data
;
5526 /* Set the cli tool unix socket path */
5527 if (strlen(client_unix_sock_path
) == 0) {
5528 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5529 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5533 goto exit_init_data
;
5537 /* Set global SHM for ust */
5538 if (strlen(wait_shm_path
) == 0) {
5539 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5540 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5544 goto exit_init_data
;
5548 /* Set health check Unix path */
5549 if (strlen(health_unix_sock_path
) == 0) {
5550 ret
= snprintf(health_unix_sock_path
,
5551 sizeof(health_unix_sock_path
),
5552 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5556 goto exit_init_data
;
5561 lockfile_fd
= create_lockfile();
5562 if (lockfile_fd
< 0) {
5564 goto exit_init_data
;
5567 /* Set consumer initial state */
5568 kernel_consumerd_state
= CONSUMER_STOPPED
;
5569 ust_consumerd_state
= CONSUMER_STOPPED
;
5571 DBG("Client socket path %s", client_unix_sock_path
);
5572 DBG("Application socket path %s", apps_unix_sock_path
);
5573 DBG("Application wait path %s", wait_shm_path
);
5574 DBG("LTTng run directory path: %s", rundir
);
5576 /* 32 bits consumerd path setup */
5577 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5578 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5580 PERROR("snprintf 32-bit consumer error socket path");
5582 goto exit_init_data
;
5584 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5585 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5587 PERROR("snprintf 32-bit consumer command socket path");
5589 goto exit_init_data
;
5592 DBG2("UST consumer 32 bits err path: %s",
5593 ustconsumer32_data
.err_unix_sock_path
);
5594 DBG2("UST consumer 32 bits cmd path: %s",
5595 ustconsumer32_data
.cmd_unix_sock_path
);
5597 /* 64 bits consumerd path setup */
5598 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5599 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5601 PERROR("snprintf 64-bit consumer error socket path");
5603 goto exit_init_data
;
5605 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5606 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5608 PERROR("snprintf 64-bit consumer command socket path");
5610 goto exit_init_data
;
5613 DBG2("UST consumer 64 bits err path: %s",
5614 ustconsumer64_data
.err_unix_sock_path
);
5615 DBG2("UST consumer 64 bits cmd path: %s",
5616 ustconsumer64_data
.cmd_unix_sock_path
);
5619 * See if daemon already exist.
5621 if (check_existing_daemon()) {
5622 ERR("Already running daemon.\n");
5624 * We do not goto exit because we must not cleanup()
5625 * because a daemon is already running.
5628 goto exit_init_data
;
5632 * Init UST app hash table. Alloc hash table before this point since
5633 * cleanup() can get called after that point.
5635 if (ust_app_ht_alloc()) {
5636 ERR("Failed to allocate UST app hash table");
5638 goto exit_init_data
;
5642 * Initialize agent app hash table. We allocate the hash table here
5643 * since cleanup() can get called after this point.
5645 if (agent_app_ht_alloc()) {
5646 ERR("Failed to allocate Agent app hash table");
5648 goto exit_init_data
;
5652 * These actions must be executed as root. We do that *after* setting up
5653 * the sockets path because we MUST make the check for another daemon using
5654 * those paths *before* trying to set the kernel consumer sockets and init
5658 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5660 goto exit_init_data
;
5663 /* Setup kernel tracer */
5664 if (!opt_no_kernel
) {
5665 init_kernel_tracer();
5666 if (kernel_tracer_fd
>= 0) {
5667 ret
= syscall_init_table();
5669 ERR("Unable to populate syscall table. "
5670 "Syscall tracing won't work "
5671 "for this session daemon.");
5676 /* Set ulimit for open files */
5679 /* init lttng_fd tracking must be done after set_ulimit. */
5682 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5684 goto exit_init_data
;
5687 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5689 goto exit_init_data
;
5692 /* Setup the needed unix socket */
5693 if (init_daemon_socket()) {
5695 goto exit_init_data
;
5698 /* Set credentials to socket */
5699 if (is_root
&& set_permissions(rundir
)) {
5701 goto exit_init_data
;
5704 /* Get parent pid if -S, --sig-parent is specified. */
5705 if (opt_sig_parent
) {
5709 /* Setup the kernel pipe for waking up the kernel thread */
5710 if (is_root
&& !opt_no_kernel
) {
5711 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5713 goto exit_init_data
;
5717 /* Setup the thread apps communication pipe. */
5718 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5720 goto exit_init_data
;
5723 /* Setup the thread apps notify communication pipe. */
5724 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5726 goto exit_init_data
;
5729 /* Initialize global buffer per UID and PID registry. */
5730 buffer_reg_init_uid_registry();
5731 buffer_reg_init_pid_registry();
5733 /* Init UST command queue. */
5734 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5737 * Get session list pointer. This pointer MUST NOT be free'd. This list
5738 * is statically declared in session.c
5740 session_list_ptr
= session_get_list();
5744 /* Check for the application socket timeout env variable. */
5745 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5746 if (env_app_timeout
) {
5747 app_socket_timeout
= atoi(env_app_timeout
);
5749 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5752 ret
= write_pidfile();
5754 ERR("Error in write_pidfile");
5756 goto exit_init_data
;
5758 ret
= write_agent_port();
5760 ERR("Error in write_agent_port");
5762 goto exit_init_data
;
5765 /* Initialize communication library */
5767 /* Initialize TCP timeout values */
5768 lttcomm_inet_init();
5770 if (load_session_init_data(&load_info
) < 0) {
5772 goto exit_init_data
;
5774 load_info
->path
= opt_load_session_path
;
5776 /* Create health-check thread */
5777 ret
= pthread_create(&health_thread
, NULL
,
5778 thread_manage_health
, (void *) NULL
);
5781 PERROR("pthread_create health");
5786 /* Create thread to manage the client socket */
5787 ret
= pthread_create(&client_thread
, NULL
,
5788 thread_manage_clients
, (void *) NULL
);
5791 PERROR("pthread_create clients");
5796 /* Create thread to dispatch registration */
5797 ret
= pthread_create(&dispatch_thread
, NULL
,
5798 thread_dispatch_ust_registration
, (void *) NULL
);
5801 PERROR("pthread_create dispatch");
5806 /* Create thread to manage application registration. */
5807 ret
= pthread_create(®_apps_thread
, NULL
,
5808 thread_registration_apps
, (void *) NULL
);
5811 PERROR("pthread_create registration");
5816 /* Create thread to manage application socket */
5817 ret
= pthread_create(&apps_thread
, NULL
,
5818 thread_manage_apps
, (void *) NULL
);
5821 PERROR("pthread_create apps");
5826 /* Create thread to manage application notify socket */
5827 ret
= pthread_create(&apps_notify_thread
, NULL
,
5828 ust_thread_manage_notify
, (void *) NULL
);
5831 PERROR("pthread_create notify");
5833 goto exit_apps_notify
;
5836 /* Create agent registration thread. */
5837 ret
= pthread_create(&agent_reg_thread
, NULL
,
5838 agent_thread_manage_registration
, (void *) NULL
);
5841 PERROR("pthread_create agent");
5843 goto exit_agent_reg
;
5846 /* Don't start this thread if kernel tracing is not requested nor root */
5847 if (is_root
&& !opt_no_kernel
) {
5848 /* Create kernel thread to manage kernel event */
5849 ret
= pthread_create(&kernel_thread
, NULL
,
5850 thread_manage_kernel
, (void *) NULL
);
5853 PERROR("pthread_create kernel");
5859 /* Create session loading thread. */
5860 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
5864 PERROR("pthread_create load_session_thread");
5866 goto exit_load_session
;
5870 * This is where we start awaiting program completion (e.g. through
5871 * signal that asks threads to teardown).
5874 ret
= pthread_join(load_session_thread
, &status
);
5877 PERROR("pthread_join load_session_thread");
5882 if (is_root
&& !opt_no_kernel
) {
5883 ret
= pthread_join(kernel_thread
, &status
);
5886 PERROR("pthread_join");
5892 ret
= pthread_join(agent_reg_thread
, &status
);
5895 PERROR("pthread_join agent");
5900 ret
= pthread_join(apps_notify_thread
, &status
);
5903 PERROR("pthread_join apps notify");
5908 ret
= pthread_join(apps_thread
, &status
);
5911 PERROR("pthread_join apps");
5916 ret
= pthread_join(reg_apps_thread
, &status
);
5919 PERROR("pthread_join");
5924 ret
= pthread_join(dispatch_thread
, &status
);
5927 PERROR("pthread_join");
5932 ret
= pthread_join(client_thread
, &status
);
5935 PERROR("pthread_join");
5940 ret
= pthread_join(health_thread
, &status
);
5943 PERROR("pthread_join health thread");
5950 * sessiond_cleanup() is called when no other thread is running, except
5951 * the ht_cleanup thread, which is needed to destroy the hash tables.
5953 rcu_thread_online();
5955 rcu_thread_offline();
5956 rcu_unregister_thread();
5958 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
5960 ERR("write error on ht_cleanup quit pipe");
5964 ret
= pthread_join(ht_cleanup_thread
, &status
);
5967 PERROR("pthread_join ht cleanup thread");
5973 utils_close_pipe(ht_cleanup_pipe
);
5974 exit_ht_cleanup_pipe
:
5977 * Close the ht_cleanup quit pipe.
5979 utils_close_pipe(ht_cleanup_quit_pipe
);
5980 exit_ht_cleanup_quit_pipe
:
5982 health_app_destroy(health_sessiond
);
5983 exit_health_sessiond_cleanup
:
5986 sessiond_cleanup_options();
5988 exit_set_signal_handler
: